fs/xfs/xfs_bmap.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.  All Rights Reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or modify it
   5  * under the terms of version 2 of the GNU General Public License as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it would be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
  11  *
  12  * Further, this software is distributed without any warranty that it is
  13  * free of the rightful claim of any third person regarding infringement
  14  * or the like.  Any license provided herein, whether implied or
  15  * otherwise, applies only to this software file.  Patent licenses, if
  16  * any, provided herein do not apply to combinations of this program with
  17  * other software, or any other product whatsoever.
  18  *
  19  * You should have received a copy of the GNU General Public License along
  20  * with this program; if not, write the Free Software Foundation, Inc., 59
  21  * Temple Place - Suite 330, Boston MA 02111-1307, USA.
  22  *
  23  * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
  24  * Mountain View, CA  94043, or:
  25  *
  26  * http://www.sgi.com
  27  *
  28  * For further information regarding this notice, see:
  29  *
  30  * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
  31  */
  32
  33 #include "xfs.h"
  34
  35 #include "xfs_macros.h"
  36 #include "xfs_types.h"
  37 #include "xfs_inum.h"
  38 #include "xfs_log.h"
  39 #include "xfs_trans.h"
  40 #include "xfs_sb.h"
  41 #include "xfs_ag.h"
  42 #include "xfs_dir.h"
  43 #include "xfs_dir2.h"
  44 #include "xfs_dmapi.h"
  45 #include "xfs_mount.h"
  46 #include "xfs_alloc_btree.h"
  47 #include "xfs_bmap_btree.h"
  48 #include "xfs_ialloc_btree.h"
  49 #include "xfs_btree.h"
  50 #include "xfs_ialloc.h"
  51 #include "xfs_attr_sf.h"
  52 #include "xfs_dir_sf.h"
  53 #include "xfs_dir2_sf.h"
  54 #include "xfs_dinode.h"
  55 #include "xfs_inode_item.h"
  56 #include "xfs_inode.h"
  57 #include "xfs_itable.h"
  58 #include "xfs_extfree_item.h"
  59 #include "xfs_alloc.h"
  60 #include "xfs_bmap.h"
  61 #include "xfs_rtalloc.h"
  62 #include "xfs_error.h"
  63 #include "xfs_da_btree.h"
  64 #include "xfs_dir_leaf.h"
  65 #include "xfs_attr_leaf.h"
  66 #include "xfs_bit.h"
  67 #include "xfs_rw.h"
  68 #include "xfs_quota.h"
  69 #include "xfs_trans_space.h"
  70 #include "xfs_buf_item.h"
  71
  72
  73 #ifdef DEBUG
  74 STATIC void
  75 xfs_bmap_check_leaf_extents(xfs_btree_cur_t *cur, xfs_inode_t *ip, int whichfork);
  76 #endif
  77
  78 kmem_zone_t             *xfs_bmap_free_item_zone;
  79
  80 /*
  81  * Prototypes for internal bmap routines.
  82  */
  83
  84
  85 /*
  86  * Called from xfs_bmap_add_attrfork to handle extents format files.
  87  */
  88 STATIC int                                      /* error */
  89 xfs_bmap_add_attrfork_extents(
  90         xfs_trans_t             *tp,            /* transaction pointer */
  91         xfs_inode_t             *ip,            /* incore inode pointer */
  92         xfs_fsblock_t           *firstblock,    /* first block allocated */
  93         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
  94         int                     *flags);        /* inode logging flags */
  95
  96 /*
  97  * Called from xfs_bmap_add_attrfork to handle local format files.
  98  */
  99 STATIC int                                      /* error */
 100 xfs_bmap_add_attrfork_local(
 101         xfs_trans_t             *tp,            /* transaction pointer */
 102         xfs_inode_t             *ip,            /* incore inode pointer */
 103         xfs_fsblock_t           *firstblock,    /* first block allocated */
 104         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 105         int                     *flags);        /* inode logging flags */
 106
 107 /*
 108  * Called by xfs_bmapi to update extent list structure and the btree
 109  * after allocating space (or doing a delayed allocation).
 110  */
 111 STATIC int                              /* error */
 112 xfs_bmap_add_extent(
 113         xfs_inode_t             *ip,    /* incore inode pointer */
 114         xfs_extnum_t            idx,    /* extent number to update/insert */
 115         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 116         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
 117         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 118         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 119         int                     *logflagsp, /* inode logging flags */
 120         int                     whichfork, /* data or attr fork */
 121         int                     rsvd);  /* OK to allocate reserved blocks */
 122
 123 /*
 124  * Called by xfs_bmap_add_extent to handle cases converting a delayed
 125  * allocation to a real allocation.
 126  */
 127 STATIC int                              /* error */
 128 xfs_bmap_add_extent_delay_real(
 129         xfs_inode_t             *ip,    /* incore inode pointer */
 130         xfs_extnum_t            idx,    /* extent number to update/insert */
 131         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 132         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
 133         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
 134         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 135         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 136         int                     *logflagsp, /* inode logging flags */
 137         int                     rsvd);  /* OK to allocate reserved blocks */
 138
 139 /*
 140  * Called by xfs_bmap_add_extent to handle cases converting a hole
 141  * to a delayed allocation.
 142  */
 143 STATIC int                              /* error */
 144 xfs_bmap_add_extent_hole_delay(
 145         xfs_inode_t             *ip,    /* incore inode pointer */
 146         xfs_extnum_t            idx,    /* extent number to update/insert */
 147         xfs_btree_cur_t         *cur,   /* if null, not a btree */
 148         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
 149         int                     *logflagsp,/* inode logging flags */
 150         int                     rsvd);  /* OK to allocate reserved blocks */
 151
 152 /*
 153  * Called by xfs_bmap_add_extent to handle cases converting a hole
 154  * to a real allocation.
 155  */
 156 STATIC int                              /* error */
 157 xfs_bmap_add_extent_hole_real(
 158         xfs_inode_t             *ip,    /* incore inode pointer */
 159         xfs_extnum_t            idx,    /* extent number to update/insert */
 160         xfs_btree_cur_t         *cur,   /* if null, not a btree */
 161         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
 162         int                     *logflagsp, /* inode logging flags */
 163         int                     whichfork); /* data or attr fork */
 164
 165 /*
 166  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
 167  * allocation to a real allocation or vice versa.
 168  */
 169 STATIC int                              /* error */
 170 xfs_bmap_add_extent_unwritten_real(
 171         xfs_inode_t             *ip,    /* incore inode pointer */
 172         xfs_extnum_t            idx,    /* extent number to update/insert */
 173         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 174         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
 175         int                     *logflagsp); /* inode logging flags */
 176
 177 /*
 178  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
 179  * It figures out where to ask the underlying allocator to put the new extent.
 180  */
 181 STATIC int                              /* error */
 182 xfs_bmap_alloc(
 183         xfs_bmalloca_t          *ap);   /* bmap alloc argument struct */
 184
 185 /*
 186  * Transform a btree format file with only one leaf node, where the
 187  * extents list will fit in the inode, into an extents format file.
 188  * Since the extent list is already in-core, all we have to do is
 189  * give up the space for the btree root and pitch the leaf block.
 190  */
 191 STATIC int                              /* error */
 192 xfs_bmap_btree_to_extents(
 193         xfs_trans_t             *tp,    /* transaction pointer */
 194         xfs_inode_t             *ip,    /* incore inode pointer */
 195         xfs_btree_cur_t         *cur,   /* btree cursor */
 196         int                     *logflagsp, /* inode logging flags */
 197         int                     whichfork); /* data or attr fork */
 198
 199 #ifdef DEBUG
 200 /*
 201  * Check that the extents list for the inode ip is in the right order.
 202  */
 203 STATIC void
 204 xfs_bmap_check_extents(
 205         xfs_inode_t             *ip,            /* incore inode pointer */
 206         int                     whichfork);     /* data or attr fork */
 207 #endif
 208
 209 /*
 210  * Called by xfs_bmapi to update extent list structure and the btree
 211  * after removing space (or undoing a delayed allocation).
 212  */
 213 STATIC int                              /* error */
 214 xfs_bmap_del_extent(
 215         xfs_inode_t             *ip,    /* incore inode pointer */
 216         xfs_trans_t             *tp,    /* current trans pointer */
 217         xfs_extnum_t            idx,    /* extent number to update/insert */
 218         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 219         xfs_btree_cur_t         *cur,   /* if null, not a btree */
 220         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
 221         int                     *logflagsp,/* inode logging flags */
 222         int                     whichfork, /* data or attr fork */
 223         int                     rsvd);   /* OK to allocate reserved blocks */
 224
 225 /*
 226  * Remove the entry "free" from the free item list.  Prev points to the
 227  * previous entry, unless "free" is the head of the list.
 228  */
 229 STATIC void
 230 xfs_bmap_del_free(
 231         xfs_bmap_free_t         *flist, /* free item list header */
 232         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
 233         xfs_bmap_free_item_t    *free); /* list item to be freed */
 234
 235 /*
 236  * Remove count entries from the extents array for inode "ip", starting
 237  * at index "idx".  Copies the remaining items down over the deleted ones,
 238  * and gives back the excess memory.
 239  */
 240 STATIC void
 241 xfs_bmap_delete_exlist(
 242         xfs_inode_t     *ip,            /* incode inode pointer */
 243         xfs_extnum_t    idx,            /* starting delete index */
 244         xfs_extnum_t    count,          /* count of items to delete */
 245         int             whichfork);     /* data or attr fork */
 246
 247 /*
 248  * Convert an extents-format file into a btree-format file.
 249  * The new file will have a root block (in the inode) and a single child block.
 250  */
 251 STATIC int                                      /* error */
 252 xfs_bmap_extents_to_btree(
 253         xfs_trans_t             *tp,            /* transaction pointer */
 254         xfs_inode_t             *ip,            /* incore inode pointer */
 255         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
 256         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
 257         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
 258         int                     wasdel,         /* converting a delayed alloc */
 259         int                     *logflagsp,     /* inode logging flags */
 260         int                     whichfork);     /* data or attr fork */
 261
 262 /*
 263  * Insert new item(s) in the extent list for inode "ip".
 264  * Count new items are inserted at offset idx.
 265  */
 266 STATIC void
 267 xfs_bmap_insert_exlist(
 268         xfs_inode_t     *ip,            /* incore inode pointer */
 269         xfs_extnum_t    idx,            /* starting index of new items */
 270         xfs_extnum_t    count,          /* number of inserted items */
 271         xfs_bmbt_irec_t *new,           /* items to insert */
 272         int             whichfork);     /* data or attr fork */
 273
 274 /*
 275  * Convert a local file to an extents file.
 276  * This code is sort of bogus, since the file data needs to get
 277  * logged so it won't be lost.  The bmap-level manipulations are ok, though.
 278  */
 279 STATIC int                              /* error */
 280 xfs_bmap_local_to_extents(
 281         xfs_trans_t     *tp,            /* transaction pointer */
 282         xfs_inode_t     *ip,            /* incore inode pointer */
 283         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
 284         xfs_extlen_t    total,          /* total blocks needed by transaction */
 285         int             *logflagsp,     /* inode logging flags */
 286         int             whichfork);     /* data or attr fork */
 287
 288 /*
 289  * Search the extents list for the inode, for the extent containing bno.
 290  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
 291  * *eofp will be set, and *prevp will contain the last entry (null if none).
 292  * Else, *lastxp will be set to the index of the found
 293  * entry; *gotp will contain the entry.
 294  */
 295 STATIC xfs_bmbt_rec_t *                 /* pointer to found extent entry */
 296 xfs_bmap_search_extents(
 297         xfs_inode_t     *ip,            /* incore inode pointer */
 298         xfs_fileoff_t   bno,            /* block number searched for */
 299         int             whichfork,      /* data or attr fork */
 300         int             *eofp,          /* out: end of file found */
 301         xfs_extnum_t    *lastxp,        /* out: last extent index */
 302         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
 303         xfs_bmbt_irec_t *prevp);        /* out: previous extent entry found */
 304
 305 /*
 306  * Check the last inode extent to determine whether this allocation will result
 307  * in blocks being allocated at the end of the file. When we allocate new data
 308  * blocks at the end of the file which do not start at the previous data block,
 309  * we will try to align the new blocks at stripe unit boundaries.
 310  */
 311 STATIC int                              /* error */
 312 xfs_bmap_isaeof(
 313         xfs_inode_t     *ip,            /* incore inode pointer */
 314         xfs_fileoff_t   off,            /* file offset in fsblocks */
 315         int             whichfork,      /* data or attribute fork */
 316         char            *aeof);         /* return value */
 317
 318 #ifdef XFS_BMAP_TRACE
 319 /*
 320  * Add a bmap trace buffer entry.  Base routine for the others.
 321  */
 322 STATIC void
 323 xfs_bmap_trace_addentry(
 324         int             opcode,         /* operation */
 325         char            *fname,         /* function name */
 326         char            *desc,          /* operation description */
 327         xfs_inode_t     *ip,            /* incore inode pointer */
 328         xfs_extnum_t    idx,            /* index of entry(ies) */
 329         xfs_extnum_t    cnt,            /* count of entries, 1 or 2 */
 330         xfs_bmbt_rec_t  *r1,            /* first record */
 331         xfs_bmbt_rec_t  *r2,            /* second record or null */
 332         int             whichfork);     /* data or attr fork */
 333
 334 /*
 335  * Add bmap trace entry prior to a call to xfs_bmap_delete_exlist.
 336  */
 337 STATIC void
 338 xfs_bmap_trace_delete(
 339         char            *fname,         /* function name */
 340         char            *desc,          /* operation description */
 341         xfs_inode_t     *ip,            /* incore inode pointer */
 342         xfs_extnum_t    idx,            /* index of entry(entries) deleted */
 343         xfs_extnum_t    cnt,            /* count of entries deleted, 1 or 2 */
 344         int             whichfork);     /* data or attr fork */
 345
 346 /*
 347  * Add bmap trace entry prior to a call to xfs_bmap_insert_exlist, or
 348  * reading in the extents list from the disk (in the btree).
 349  */
 350 STATIC void
 351 xfs_bmap_trace_insert(
 352         char            *fname,         /* function name */
 353         char            *desc,          /* operation description */
 354         xfs_inode_t     *ip,            /* incore inode pointer */
 355         xfs_extnum_t    idx,            /* index of entry(entries) inserted */
 356         xfs_extnum_t    cnt,            /* count of entries inserted, 1 or 2 */
 357         xfs_bmbt_irec_t *r1,            /* inserted record 1 */
 358         xfs_bmbt_irec_t *r2,            /* inserted record 2 or null */
 359         int             whichfork);     /* data or attr fork */
 360
 361 /*
 362  * Add bmap trace entry after updating an extent list entry in place.
 363  */
 364 STATIC void
 365 xfs_bmap_trace_post_update(
 366         char            *fname,         /* function name */
 367         char            *desc,          /* operation description */
 368         xfs_inode_t     *ip,            /* incore inode pointer */
 369         xfs_extnum_t    idx,            /* index of entry updated */
 370         int             whichfork);     /* data or attr fork */
 371
 372 /*
 373  * Add bmap trace entry prior to updating an extent list entry in place.
 374  */
 375 STATIC void
 376 xfs_bmap_trace_pre_update(
 377         char            *fname,         /* function name */
 378         char            *desc,          /* operation description */
 379         xfs_inode_t     *ip,            /* incore inode pointer */
 380         xfs_extnum_t    idx,            /* index of entry to be updated */
 381         int             whichfork);     /* data or attr fork */
 382
 383 #else
 384 #define xfs_bmap_trace_delete(f,d,ip,i,c,w)
 385 #define xfs_bmap_trace_insert(f,d,ip,i,c,r1,r2,w)
 386 #define xfs_bmap_trace_post_update(f,d,ip,i,w)
 387 #define xfs_bmap_trace_pre_update(f,d,ip,i,w)
 388 #endif  /* XFS_BMAP_TRACE */
 389
 390 /*
 391  * Compute the worst-case number of indirect blocks that will be used
 392  * for ip's delayed extent of length "len".
 393  */
 394 STATIC xfs_filblks_t
 395 xfs_bmap_worst_indlen(
 396         xfs_inode_t             *ip,    /* incore inode pointer */
 397         xfs_filblks_t           len);   /* delayed extent length */
 398
 399 #ifdef DEBUG
 400 /*
 401  * Perform various validation checks on the values being returned
 402  * from xfs_bmapi().
 403  */
 404 STATIC void
 405 xfs_bmap_validate_ret(
 406         xfs_fileoff_t           bno,
 407         xfs_filblks_t           len,
 408         int                     flags,
 409         xfs_bmbt_irec_t         *mval,
 410         int                     nmap,
 411         int                     ret_nmap);
 412 #else
 413 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
 414 #endif /* DEBUG */
 415
 416 #if defined(XFS_RW_TRACE)
 417 STATIC void
 418 xfs_bunmap_trace(
 419         xfs_inode_t             *ip,
 420         xfs_fileoff_t           bno,
 421         xfs_filblks_t           len,
 422         int                     flags,
 423         inst_t                  *ra);
 424 #else
 425 #define xfs_bunmap_trace(ip, bno, len, flags, ra)
 426 #endif  /* XFS_RW_TRACE */
 427
 428 STATIC int
 429 xfs_bmap_count_tree(
 430         xfs_mount_t     *mp,
 431         xfs_trans_t     *tp,
 432         xfs_fsblock_t   blockno,
 433         int             levelin,
 434         int             *count);
 435
 436 STATIC int
 437 xfs_bmap_count_leaves(
 438         xfs_bmbt_rec_t          *frp,
 439         int                     numrecs,
 440         int                     *count);
 441
 442 /*
 443  * Bmap internal routines.
 444  */
 445
 446 /*
 447  * Called from xfs_bmap_add_attrfork to handle btree format files.
 448  */
 449 STATIC int                                      /* error */
 450 xfs_bmap_add_attrfork_btree(
 451         xfs_trans_t             *tp,            /* transaction pointer */
 452         xfs_inode_t             *ip,            /* incore inode pointer */
 453         xfs_fsblock_t           *firstblock,    /* first block allocated */
 454         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 455         int                     *flags)         /* inode logging flags */
 456 {
 457         xfs_btree_cur_t         *cur;           /* btree cursor */
 458         int                     error;          /* error return value */
 459         xfs_mount_t             *mp;            /* file system mount struct */
 460         int                     stat;           /* newroot status */
 461
 462         mp = ip->i_mount;
 463         if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
 464                 *flags |= XFS_ILOG_DBROOT;
 465         else {
 466                 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
 467                         XFS_DATA_FORK);
 468                 cur->bc_private.b.flist = flist;
 469                 cur->bc_private.b.firstblock = *firstblock;
 470                 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
 471                         goto error0;
 472                 ASSERT(stat == 1);      /* must be at least one entry */
 473                 if ((error = xfs_bmbt_newroot(cur, flags, &stat)))
 474                         goto error0;
 475                 if (stat == 0) {
 476                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 477                         return XFS_ERROR(ENOSPC);
 478                 }
 479                 *firstblock = cur->bc_private.b.firstblock;
 480                 cur->bc_private.b.allocated = 0;
 481                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 482         }
 483         return 0;
 484 error0:
 485         xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 486         return error;
 487 }
 488
 489 /*
 490  * Called from xfs_bmap_add_attrfork to handle extents format files.
 491  */
 492 STATIC int                                      /* error */
 493 xfs_bmap_add_attrfork_extents(
 494         xfs_trans_t             *tp,            /* transaction pointer */
 495         xfs_inode_t             *ip,            /* incore inode pointer */
 496         xfs_fsblock_t           *firstblock,    /* first block allocated */
 497         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 498         int                     *flags)         /* inode logging flags */
 499 {
 500         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
 501         int                     error;          /* error return value */
 502
 503         if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
 504                 return 0;
 505         cur = NULL;
 506         error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
 507                 flags, XFS_DATA_FORK);
 508         if (cur) {
 509                 cur->bc_private.b.allocated = 0;
 510                 xfs_btree_del_cursor(cur,
 511                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
 512         }
 513         return error;
 514 }
 515
 516 /*
 517  * Called from xfs_bmap_add_attrfork to handle local format files.
 518  */
 519 STATIC int                                      /* error */
 520 xfs_bmap_add_attrfork_local(
 521         xfs_trans_t             *tp,            /* transaction pointer */
 522         xfs_inode_t             *ip,            /* incore inode pointer */
 523         xfs_fsblock_t           *firstblock,    /* first block allocated */
 524         xfs_bmap_free_t         *flist,         /* blocks to free at commit */
 525         int                     *flags)         /* inode logging flags */
 526 {
 527         xfs_da_args_t           dargs;          /* args for dir/attr code */
 528         int                     error;          /* error return value */
 529         xfs_mount_t             *mp;            /* mount structure pointer */
 530
 531         if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
 532                 return 0;
 533         if ((ip->i_d.di_mode & S_IFMT) == S_IFDIR) {
 534                 mp = ip->i_mount;
 535                 memset(&dargs, 0, sizeof(dargs));
 536                 dargs.dp = ip;
 537                 dargs.firstblock = firstblock;
 538                 dargs.flist = flist;
 539                 dargs.total = mp->m_dirblkfsbs;
 540                 dargs.whichfork = XFS_DATA_FORK;
 541                 dargs.trans = tp;
 542                 error = XFS_DIR_SHORTFORM_TO_SINGLE(mp, &dargs);
 543         } else
 544                 error = xfs_bmap_local_to_extents(tp, ip, firstblock, 1, flags,
 545                         XFS_DATA_FORK);
 546         return error;
 547 }
 548
 549 /*
 550  * Called by xfs_bmapi to update extent list structure and the btree
 551  * after allocating space (or doing a delayed allocation).
 552  */
 553 STATIC int                              /* error */
 554 xfs_bmap_add_extent(
 555         xfs_inode_t             *ip,    /* incore inode pointer */
 556         xfs_extnum_t            idx,    /* extent number to update/insert */
 557         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 558         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
 559         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 560         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 561         int                     *logflagsp, /* inode logging flags */
 562         int                     whichfork, /* data or attr fork */
 563         int                     rsvd)   /* OK to use reserved data blocks */
 564 {
 565         xfs_btree_cur_t         *cur;   /* btree cursor or null */
 566         xfs_filblks_t           da_new; /* new count del alloc blocks used */
 567         xfs_filblks_t           da_old; /* old count del alloc blocks used */
 568         int                     error;  /* error return value */
 569 #ifdef XFS_BMAP_TRACE
 570         static char             fname[] = "xfs_bmap_add_extent";
 571 #endif
 572         xfs_ifork_t             *ifp;   /* inode fork ptr */
 573         int                     logflags; /* returned value */
 574         xfs_extnum_t            nextents; /* number of extents in file now */
 575
 576         XFS_STATS_INC(xs_add_exlist);
 577         cur = *curp;
 578         ifp = XFS_IFORK_PTR(ip, whichfork);
 579         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
 580         ASSERT(idx <= nextents);
 581         da_old = da_new = 0;
 582         error = 0;
 583         /*
 584          * This is the first extent added to a new/empty file.
 585          * Special case this one, so other routines get to assume there are
 586          * already extents in the list.
 587          */
 588         if (nextents == 0) {
 589                 xfs_bmap_trace_insert(fname, "insert empty", ip, 0, 1, new,
 590                         NULL, whichfork);
 591                 xfs_bmap_insert_exlist(ip, 0, 1, new, whichfork);
 592                 ASSERT(cur == NULL);
 593                 ifp->if_lastex = 0;
 594                 if (!ISNULLSTARTBLOCK(new->br_startblock)) {
 595                         XFS_IFORK_NEXT_SET(ip, whichfork, 1);
 596                         logflags = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
 597                 } else
 598                         logflags = 0;
 599         }
 600         /*
 601          * Any kind of new delayed allocation goes here.
 602          */
 603         else if (ISNULLSTARTBLOCK(new->br_startblock)) {
 604                 if (cur)
 605                         ASSERT((cur->bc_private.b.flags &
 606                                 XFS_BTCUR_BPRV_WASDEL) == 0);
 607                 if ((error = xfs_bmap_add_extent_hole_delay(ip, idx, cur, new,
 608                                 &logflags, rsvd)))
 609                         goto done;
 610         }
 611         /*
 612          * Real allocation off the end of the file.
 613          */
 614         else if (idx == nextents) {
 615                 if (cur)
 616                         ASSERT((cur->bc_private.b.flags &
 617                                 XFS_BTCUR_BPRV_WASDEL) == 0);
 618                 if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur, new,
 619                                 &logflags, whichfork)))
 620                         goto done;
 621         } else {
 622                 xfs_bmbt_irec_t prev;   /* old extent at offset idx */
 623
 624                 /*
 625                  * Get the record referred to by idx.
 626                  */
 627                 xfs_bmbt_get_all(&ifp->if_u1.if_extents[idx], &prev);
 628                 /*
 629                  * If it's a real allocation record, and the new allocation ends
 630                  * after the start of the referred to record, then we're filling
 631                  * in a delayed or unwritten allocation with a real one, or
 632                  * converting real back to unwritten.
 633                  */
 634                 if (!ISNULLSTARTBLOCK(new->br_startblock) &&
 635                     new->br_startoff + new->br_blockcount > prev.br_startoff) {
 636                         if (prev.br_state != XFS_EXT_UNWRITTEN &&
 637                             ISNULLSTARTBLOCK(prev.br_startblock)) {
 638                                 da_old = STARTBLOCKVAL(prev.br_startblock);
 639                                 if (cur)
 640                                         ASSERT(cur->bc_private.b.flags &
 641                                                 XFS_BTCUR_BPRV_WASDEL);
 642                                 if ((error = xfs_bmap_add_extent_delay_real(ip,
 643                                         idx, &cur, new, &da_new, first, flist,
 644                                         &logflags, rsvd)))
 645                                         goto done;
 646                         } else if (new->br_state == XFS_EXT_NORM) {
 647                                 ASSERT(new->br_state == XFS_EXT_NORM);
 648                                 if ((error = xfs_bmap_add_extent_unwritten_real(
 649                                         ip, idx, &cur, new, &logflags)))
 650                                         goto done;
 651                         } else {
 652                                 ASSERT(new->br_state == XFS_EXT_UNWRITTEN);
 653                                 if ((error = xfs_bmap_add_extent_unwritten_real(
 654                                         ip, idx, &cur, new, &logflags)))
 655                                         goto done;
 656                         }
 657                         ASSERT(*curp == cur || *curp == NULL);
 658                 }
 659                 /*
 660                  * Otherwise we're filling in a hole with an allocation.
 661                  */
 662                 else {
 663                         if (cur)
 664                                 ASSERT((cur->bc_private.b.flags &
 665                                         XFS_BTCUR_BPRV_WASDEL) == 0);
 666                         if ((error = xfs_bmap_add_extent_hole_real(ip, idx, cur,
 667                                         new, &logflags, whichfork)))
 668                                 goto done;
 669                 }
 670         }
 671
 672         ASSERT(*curp == cur || *curp == NULL);
 673         /*
 674          * Convert to a btree if necessary.
 675          */
 676         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
 677             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
 678                 int     tmp_logflags;   /* partial log flag return val */
 679
 680                 ASSERT(cur == NULL);
 681                 error = xfs_bmap_extents_to_btree(ip->i_transp, ip, first,
 682                         flist, &cur, da_old > 0, &tmp_logflags, whichfork);
 683                 logflags |= tmp_logflags;
 684                 if (error)
 685                         goto done;
 686         }
 687         /*
 688          * Adjust for changes in reserved delayed indirect blocks.
 689          * Nothing to do for disk quotas here.
 690          */
 691         if (da_old || da_new) {
 692                 xfs_filblks_t   nblks;
 693
 694                 nblks = da_new;
 695                 if (cur)
 696                         nblks += cur->bc_private.b.allocated;
 697                 ASSERT(nblks <= da_old);
 698                 if (nblks < da_old)
 699                         xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
 700                                 (int)(da_old - nblks), rsvd);
 701         }
 702         /*
 703          * Clear out the allocated field, done with it now in any case.
 704          */
 705         if (cur) {
 706                 cur->bc_private.b.allocated = 0;
 707                 *curp = cur;
 708         }
 709 done:
 710 #ifdef DEBUG
 711         if (!error)
 712                 xfs_bmap_check_leaf_extents(*curp, ip, whichfork);
 713 #endif
 714         *logflagsp = logflags;
 715         return error;
 716 }
 717
 718 /*
 719  * Called by xfs_bmap_add_extent to handle cases converting a delayed
 720  * allocation to a real allocation.
 721  */
 722 STATIC int                              /* error */
 723 xfs_bmap_add_extent_delay_real(
 724         xfs_inode_t             *ip,    /* incore inode pointer */
 725         xfs_extnum_t            idx,    /* extent number to update/insert */
 726         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
 727         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
 728         xfs_filblks_t           *dnew,  /* new delayed-alloc indirect blocks */
 729         xfs_fsblock_t           *first, /* pointer to firstblock variable */
 730         xfs_bmap_free_t         *flist, /* list of extents to be freed */
 731         int                     *logflagsp, /* inode logging flags */
 732         int                     rsvd)   /* OK to use reserved data block allocation */
 733 {
 734         xfs_bmbt_rec_t          *base;  /* base of extent entry list */
 735         xfs_btree_cur_t         *cur;   /* btree cursor */
 736         int                     diff;   /* temp value */
 737         xfs_bmbt_rec_t          *ep;    /* extent entry for idx */
 738         int                     error;  /* error return value */
 739 #ifdef XFS_BMAP_TRACE
 740         static char             fname[] = "xfs_bmap_add_extent_delay_real";
 741 #endif
 742         int                     i;      /* temp state */
 743         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
 744         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
 745                                         /* left is 0, right is 1, prev is 2 */
 746         int                     rval=0; /* return value (logging flags) */
 747         int                     state = 0;/* state bits, accessed thru macros */
 748         xfs_filblks_t           temp;   /* value for dnew calculations */
 749         xfs_filblks_t           temp2;  /* value for dnew calculations */
 750         int                     tmp_rval;       /* partial logging flags */
 751         enum {                          /* bit number definitions for state */
 752                 LEFT_CONTIG,    RIGHT_CONTIG,
 753                 LEFT_FILLING,   RIGHT_FILLING,
 754                 LEFT_DELAY,     RIGHT_DELAY,
 755                 LEFT_VALID,     RIGHT_VALID
 756         };
 757
 758 #define LEFT            r[0]
 759 #define RIGHT           r[1]
 760 #define PREV            r[2]
 761 #define MASK(b)         (1 << (b))
 762 #define MASK2(a,b)      (MASK(a) | MASK(b))
 763 #define MASK3(a,b,c)    (MASK2(a,b) | MASK(c))
 764 #define MASK4(a,b,c,d)  (MASK3(a,b,c) | MASK(d))
 765 #define STATE_SET(b,v)  ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
 766 #define STATE_TEST(b)   (state & MASK(b))
 767 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
 768                                        ((state &= ~MASK(b)), 0))
 769 #define SWITCH_STATE            \
 770         (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
 771
 772         /*
 773          * Set up a bunch of variables to make the tests simpler.
 774          */
 775         cur = *curp;
 776         base = ip->i_df.if_u1.if_extents;
 777         ep = &base[idx];
 778         xfs_bmbt_get_all(ep, &PREV);
 779         new_endoff = new->br_startoff + new->br_blockcount;
 780         ASSERT(PREV.br_startoff <= new->br_startoff);
 781         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
 782         /*
 783          * Set flags determining what part of the previous delayed allocation
 784          * extent is being replaced by a real allocation.
 785          */
 786         STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
 787         STATE_SET(RIGHT_FILLING,
 788                 PREV.br_startoff + PREV.br_blockcount == new_endoff);
 789         /*
 790          * Check and set flags if this segment has a left neighbor.
 791          * Don't set contiguous if the combined extent would be too large.
 792          */
 793         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
 794                 xfs_bmbt_get_all(ep - 1, &LEFT);
 795                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
 796         }
 797         STATE_SET(LEFT_CONTIG,
 798                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
 799                 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
 800                 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
 801                 LEFT.br_state == new->br_state &&
 802                 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
 803         /*
 804          * Check and set flags if this segment has a right neighbor.
 805          * Don't set contiguous if the combined extent would be too large.
 806          * Also check for all-three-contiguous being too large.
 807          */
 808         if (STATE_SET_TEST(RIGHT_VALID,
 809                         idx <
 810                         ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
 811                 xfs_bmbt_get_all(ep + 1, &RIGHT);
 812                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
 813         }
 814         STATE_SET(RIGHT_CONTIG,
 815                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
 816                 new_endoff == RIGHT.br_startoff &&
 817                 new->br_startblock + new->br_blockcount ==
 818                     RIGHT.br_startblock &&
 819                 new->br_state == RIGHT.br_state &&
 820                 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
 821                 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
 822                   MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
 823                  LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
 824                      <= MAXEXTLEN));
 825         error = 0;
 826         /*
 827          * Switch out based on the FILLING and CONTIG state bits.
 828          */
 829         switch (SWITCH_STATE) {
 830
 831         case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
 832                 /*
 833                  * Filling in all of a previously delayed allocation extent.
 834                  * The left and right neighbors are both contiguous with new.
 835                  */
 836                 xfs_bmap_trace_pre_update(fname, "LF|RF|LC|RC", ip, idx - 1,
 837                         XFS_DATA_FORK);
 838                 xfs_bmbt_set_blockcount(ep - 1,
 839                         LEFT.br_blockcount + PREV.br_blockcount +
 840                         RIGHT.br_blockcount);
 841                 xfs_bmap_trace_post_update(fname, "LF|RF|LC|RC", ip, idx - 1,
 842                         XFS_DATA_FORK);
 843                 xfs_bmap_trace_delete(fname, "LF|RF|LC|RC", ip, idx, 2,
 844                         XFS_DATA_FORK);
 845                 xfs_bmap_delete_exlist(ip, idx, 2, XFS_DATA_FORK);
 846                 ip->i_df.if_lastex = idx - 1;
 847                 ip->i_d.di_nextents--;
 848                 if (cur == NULL)
 849                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 850                 else {
 851                         rval = XFS_ILOG_CORE;
 852                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 853                                         RIGHT.br_startblock,
 854                                         RIGHT.br_blockcount, &i)))
 855                                 goto done;
 856                         ASSERT(i == 1);
 857                         if ((error = xfs_bmbt_delete(cur, &i)))
 858                                 goto done;
 859                         ASSERT(i == 1);
 860                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
 861                                 goto done;
 862                         ASSERT(i == 1);
 863                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 864                                         LEFT.br_startblock,
 865                                         LEFT.br_blockcount +
 866                                         PREV.br_blockcount +
 867                                         RIGHT.br_blockcount, LEFT.br_state)))
 868                                 goto done;
 869                 }
 870                 *dnew = 0;
 871                 break;
 872
 873         case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
 874                 /*
 875                  * Filling in all of a previously delayed allocation extent.
 876                  * The left neighbor is contiguous, the right is not.
 877                  */
 878                 xfs_bmap_trace_pre_update(fname, "LF|RF|LC", ip, idx - 1,
 879                         XFS_DATA_FORK);
 880                 xfs_bmbt_set_blockcount(ep - 1,
 881                         LEFT.br_blockcount + PREV.br_blockcount);
 882                 xfs_bmap_trace_post_update(fname, "LF|RF|LC", ip, idx - 1,
 883                         XFS_DATA_FORK);
 884                 ip->i_df.if_lastex = idx - 1;
 885                 xfs_bmap_trace_delete(fname, "LF|RF|LC", ip, idx, 1,
 886                         XFS_DATA_FORK);
 887                 xfs_bmap_delete_exlist(ip, idx, 1, XFS_DATA_FORK);
 888                 if (cur == NULL)
 889                         rval = XFS_ILOG_DEXT;
 890                 else {
 891                         rval = 0;
 892                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
 893                                         LEFT.br_startblock, LEFT.br_blockcount,
 894                                         &i)))
 895                                 goto done;
 896                         ASSERT(i == 1);
 897                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 898                                         LEFT.br_startblock,
 899                                         LEFT.br_blockcount +
 900                                         PREV.br_blockcount, LEFT.br_state)))
 901                                 goto done;
 902                 }
 903                 *dnew = 0;
 904                 break;
 905
 906         case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
 907                 /*
 908                  * Filling in all of a previously delayed allocation extent.
 909                  * The right neighbor is contiguous, the left is not.
 910                  */
 911                 xfs_bmap_trace_pre_update(fname, "LF|RF|RC", ip, idx,
 912                         XFS_DATA_FORK);
 913                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 914                 xfs_bmbt_set_blockcount(ep,
 915                         PREV.br_blockcount + RIGHT.br_blockcount);
 916                 xfs_bmap_trace_post_update(fname, "LF|RF|RC", ip, idx,
 917                         XFS_DATA_FORK);
 918                 ip->i_df.if_lastex = idx;
 919                 xfs_bmap_trace_delete(fname, "LF|RF|RC", ip, idx + 1, 1,
 920                         XFS_DATA_FORK);
 921                 xfs_bmap_delete_exlist(ip, idx + 1, 1, XFS_DATA_FORK);
 922                 if (cur == NULL)
 923                         rval = XFS_ILOG_DEXT;
 924                 else {
 925                         rval = 0;
 926                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
 927                                         RIGHT.br_startblock,
 928                                         RIGHT.br_blockcount, &i)))
 929                                 goto done;
 930                         ASSERT(i == 1);
 931                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
 932                                         new->br_startblock,
 933                                         PREV.br_blockcount +
 934                                         RIGHT.br_blockcount, PREV.br_state)))
 935                                 goto done;
 936                 }
 937                 *dnew = 0;
 938                 break;
 939
 940         case MASK2(LEFT_FILLING, RIGHT_FILLING):
 941                 /*
 942                  * Filling in all of a previously delayed allocation extent.
 943                  * Neither the left nor right neighbors are contiguous with
 944                  * the new one.
 945                  */
 946                 xfs_bmap_trace_pre_update(fname, "LF|RF", ip, idx,
 947                         XFS_DATA_FORK);
 948                 xfs_bmbt_set_startblock(ep, new->br_startblock);
 949                 xfs_bmap_trace_post_update(fname, "LF|RF", ip, idx,
 950                         XFS_DATA_FORK);
 951                 ip->i_df.if_lastex = idx;
 952                 ip->i_d.di_nextents++;
 953                 if (cur == NULL)
 954                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
 955                 else {
 956                         rval = XFS_ILOG_CORE;
 957                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
 958                                         new->br_startblock, new->br_blockcount,
 959                                         &i)))
 960                                 goto done;
 961                         ASSERT(i == 0);
 962                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
 963                         if ((error = xfs_bmbt_insert(cur, &i)))
 964                                 goto done;
 965                         ASSERT(i == 1);
 966                 }
 967                 *dnew = 0;
 968                 break;
 969
 970         case MASK2(LEFT_FILLING, LEFT_CONTIG):
 971                 /*
 972                  * Filling in the first part of a previous delayed allocation.
 973                  * The left neighbor is contiguous.
 974                  */
 975                 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx - 1,
 976                         XFS_DATA_FORK);
 977                 xfs_bmbt_set_blockcount(ep - 1,
 978                         LEFT.br_blockcount + new->br_blockcount);
 979                 xfs_bmbt_set_startoff(ep,
 980                         PREV.br_startoff + new->br_blockcount);
 981                 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx - 1,
 982                         XFS_DATA_FORK);
 983                 temp = PREV.br_blockcount - new->br_blockcount;
 984                 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx,
 985                         XFS_DATA_FORK);
 986                 xfs_bmbt_set_blockcount(ep, temp);
 987                 ip->i_df.if_lastex = idx - 1;
 988                 if (cur == NULL)
 989                         rval = XFS_ILOG_DEXT;
 990                 else {
 991                         rval = 0;
 992                         if ((error = xfs_bmbt_lookup_eq(cur, LEFT.br_startoff,
 993                                         LEFT.br_startblock, LEFT.br_blockcount,
 994                                         &i)))
 995                                 goto done;
 996                         ASSERT(i == 1);
 997                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
 998                                         LEFT.br_startblock,
 999                                         LEFT.br_blockcount +
1000                                         new->br_blockcount,
1001                                         LEFT.br_state)))
1002                                 goto done;
1003                 }
1004                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1005                         STARTBLOCKVAL(PREV.br_startblock));
1006                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1007                 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx,
1008                         XFS_DATA_FORK);
1009                 *dnew = temp;
1010                 break;
1011
1012         case MASK(LEFT_FILLING):
1013                 /*
1014                  * Filling in the first part of a previous delayed allocation.
1015                  * The left neighbor is not contiguous.
1016                  */
1017                 xfs_bmap_trace_pre_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1018                 xfs_bmbt_set_startoff(ep, new_endoff);
1019                 temp = PREV.br_blockcount - new->br_blockcount;
1020                 xfs_bmbt_set_blockcount(ep, temp);
1021                 xfs_bmap_trace_insert(fname, "LF", ip, idx, 1, new, NULL,
1022                         XFS_DATA_FORK);
1023                 xfs_bmap_insert_exlist(ip, idx, 1, new, XFS_DATA_FORK);
1024                 ip->i_df.if_lastex = idx;
1025                 ip->i_d.di_nextents++;
1026                 if (cur == NULL)
1027                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1028                 else {
1029                         rval = XFS_ILOG_CORE;
1030                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1031                                         new->br_startblock, new->br_blockcount,
1032                                         &i)))
1033                                 goto done;
1034                         ASSERT(i == 0);
1035                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1036                         if ((error = xfs_bmbt_insert(cur, &i)))
1037                                 goto done;
1038                         ASSERT(i == 1);
1039                 }
1040                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1041                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1042                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1043                                         first, flist, &cur, 1, &tmp_rval,
1044                                         XFS_DATA_FORK);
1045                         rval |= tmp_rval;
1046                         if (error)
1047                                 goto done;
1048                 }
1049                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1050                         STARTBLOCKVAL(PREV.br_startblock) -
1051                         (cur ? cur->bc_private.b.allocated : 0));
1052                 base = ip->i_df.if_u1.if_extents;
1053                 ep = &base[idx + 1];
1054                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1055                 xfs_bmap_trace_post_update(fname, "LF", ip, idx + 1,
1056                         XFS_DATA_FORK);
1057                 *dnew = temp;
1058                 break;
1059
1060         case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1061                 /*
1062                  * Filling in the last part of a previous delayed allocation.
1063                  * The right neighbor is contiguous with the new allocation.
1064                  */
1065                 temp = PREV.br_blockcount - new->br_blockcount;
1066                 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx,
1067                         XFS_DATA_FORK);
1068                 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx + 1,
1069                         XFS_DATA_FORK);
1070                 xfs_bmbt_set_blockcount(ep, temp);
1071                 xfs_bmbt_set_allf(ep + 1, new->br_startoff, new->br_startblock,
1072                         new->br_blockcount + RIGHT.br_blockcount,
1073                         RIGHT.br_state);
1074                 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx + 1,
1075                         XFS_DATA_FORK);
1076                 ip->i_df.if_lastex = idx + 1;
1077                 if (cur == NULL)
1078                         rval = XFS_ILOG_DEXT;
1079                 else {
1080                         rval = 0;
1081                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1082                                         RIGHT.br_startblock,
1083                                         RIGHT.br_blockcount, &i)))
1084                                 goto done;
1085                         ASSERT(i == 1);
1086                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1087                                         new->br_startblock,
1088                                         new->br_blockcount +
1089                                         RIGHT.br_blockcount,
1090                                         RIGHT.br_state)))
1091                                 goto done;
1092                 }
1093                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1094                         STARTBLOCKVAL(PREV.br_startblock));
1095                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1096                 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx,
1097                         XFS_DATA_FORK);
1098                 *dnew = temp;
1099                 break;
1100
1101         case MASK(RIGHT_FILLING):
1102                 /*
1103                  * Filling in the last part of a previous delayed allocation.
1104                  * The right neighbor is not contiguous.
1105                  */
1106                 temp = PREV.br_blockcount - new->br_blockcount;
1107                 xfs_bmap_trace_pre_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1108                 xfs_bmbt_set_blockcount(ep, temp);
1109                 xfs_bmap_trace_insert(fname, "RF", ip, idx + 1, 1,
1110                         new, NULL, XFS_DATA_FORK);
1111                 xfs_bmap_insert_exlist(ip, idx + 1, 1, new, XFS_DATA_FORK);
1112                 ip->i_df.if_lastex = idx + 1;
1113                 ip->i_d.di_nextents++;
1114                 if (cur == NULL)
1115                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1116                 else {
1117                         rval = XFS_ILOG_CORE;
1118                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1119                                         new->br_startblock, new->br_blockcount,
1120                                         &i)))
1121                                 goto done;
1122                         ASSERT(i == 0);
1123                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1124                         if ((error = xfs_bmbt_insert(cur, &i)))
1125                                 goto done;
1126                         ASSERT(i == 1);
1127                 }
1128                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1129                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1130                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1131                                 first, flist, &cur, 1, &tmp_rval,
1132                                 XFS_DATA_FORK);
1133                         rval |= tmp_rval;
1134                         if (error)
1135                                 goto done;
1136                 }
1137                 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
1138                         STARTBLOCKVAL(PREV.br_startblock) -
1139                         (cur ? cur->bc_private.b.allocated : 0));
1140                 base = ip->i_df.if_u1.if_extents;
1141                 ep = &base[idx];
1142                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1143                 xfs_bmap_trace_post_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1144                 *dnew = temp;
1145                 break;
1146
1147         case 0:
1148                 /*
1149                  * Filling in the middle part of a previous delayed allocation.
1150                  * Contiguity is impossible here.
1151                  * This case is avoided almost all the time.
1152                  */
1153                 temp = new->br_startoff - PREV.br_startoff;
1154                 xfs_bmap_trace_pre_update(fname, "0", ip, idx, XFS_DATA_FORK);
1155                 xfs_bmbt_set_blockcount(ep, temp);
1156                 r[0] = *new;
1157                 r[1].br_startoff = new_endoff;
1158                 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
1159                 r[1].br_blockcount = temp2;
1160                 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 2, &r[0], &r[1],
1161                         XFS_DATA_FORK);
1162                 xfs_bmap_insert_exlist(ip, idx + 1, 2, &r[0], XFS_DATA_FORK);
1163                 ip->i_df.if_lastex = idx + 1;
1164                 ip->i_d.di_nextents++;
1165                 if (cur == NULL)
1166                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1167                 else {
1168                         rval = XFS_ILOG_CORE;
1169                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1170                                         new->br_startblock, new->br_blockcount,
1171                                         &i)))
1172                                 goto done;
1173                         ASSERT(i == 0);
1174                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1175                         if ((error = xfs_bmbt_insert(cur, &i)))
1176                                 goto done;
1177                         ASSERT(i == 1);
1178                 }
1179                 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1180                     ip->i_d.di_nextents > ip->i_df.if_ext_max) {
1181                         error = xfs_bmap_extents_to_btree(ip->i_transp, ip,
1182                                         first, flist, &cur, 1, &tmp_rval,
1183                                         XFS_DATA_FORK);
1184                         rval |= tmp_rval;
1185                         if (error)
1186                                 goto done;
1187                 }
1188                 temp = xfs_bmap_worst_indlen(ip, temp);
1189                 temp2 = xfs_bmap_worst_indlen(ip, temp2);
1190                 diff = (int)(temp + temp2 - STARTBLOCKVAL(PREV.br_startblock) -
1191                         (cur ? cur->bc_private.b.allocated : 0));
1192                 if (diff > 0 &&
1193                     xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS, -diff, rsvd)) {
1194                         /*
1195                          * Ick gross gag me with a spoon.
1196                          */
1197                         ASSERT(0);      /* want to see if this ever happens! */
1198                         while (diff > 0) {
1199                                 if (temp) {
1200                                         temp--;
1201                                         diff--;
1202                                         if (!diff ||
1203                                             !xfs_mod_incore_sb(ip->i_mount,
1204                                                     XFS_SBS_FDBLOCKS, -diff, rsvd))
1205                                                 break;
1206                                 }
1207                                 if (temp2) {
1208                                         temp2--;
1209                                         diff--;
1210                                         if (!diff ||
1211                                             !xfs_mod_incore_sb(ip->i_mount,
1212                                                     XFS_SBS_FDBLOCKS, -diff, rsvd))
1213                                                 break;
1214                                 }
1215                         }
1216                 }
1217                 base = ip->i_df.if_u1.if_extents;
1218                 ep = &base[idx];
1219                 xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
1220                 xfs_bmap_trace_post_update(fname, "0", ip, idx, XFS_DATA_FORK);
1221                 xfs_bmap_trace_pre_update(fname, "0", ip, idx + 2,
1222                         XFS_DATA_FORK);
1223                 xfs_bmbt_set_startblock(ep + 2, NULLSTARTBLOCK((int)temp2));
1224                 xfs_bmap_trace_post_update(fname, "0", ip, idx + 2,
1225                         XFS_DATA_FORK);
1226                 *dnew = temp + temp2;
1227                 break;
1228
1229         case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1230         case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1231         case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1232         case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1233         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1234         case MASK(LEFT_CONTIG):
1235         case MASK(RIGHT_CONTIG):
1236                 /*
1237                  * These cases are all impossible.
1238                  */
1239                 ASSERT(0);
1240         }
1241         *curp = cur;
1242 done:
1243         *logflagsp = rval;
1244         return error;
1245 #undef  LEFT
1246 #undef  RIGHT
1247 #undef  PREV
1248 #undef  MASK
1249 #undef  MASK2
1250 #undef  MASK3
1251 #undef  MASK4
1252 #undef  STATE_SET
1253 #undef  STATE_TEST
1254 #undef  STATE_SET_TEST
1255 #undef  SWITCH_STATE
1256 }
1257
1258 /*
1259  * Called by xfs_bmap_add_extent to handle cases converting an unwritten
1260  * allocation to a real allocation or vice versa.
1261  */
1262 STATIC int                              /* error */
1263 xfs_bmap_add_extent_unwritten_real(
1264         xfs_inode_t             *ip,    /* incore inode pointer */
1265         xfs_extnum_t            idx,    /* extent number to update/insert */
1266         xfs_btree_cur_t         **curp, /* if *curp is null, not a btree */
1267         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
1268         int                     *logflagsp) /* inode logging flags */
1269 {
1270         xfs_bmbt_rec_t          *base;  /* base of extent entry list */
1271         xfs_btree_cur_t         *cur;   /* btree cursor */
1272         xfs_bmbt_rec_t          *ep;    /* extent entry for idx */
1273         int                     error;  /* error return value */
1274 #ifdef XFS_BMAP_TRACE
1275         static char             fname[] = "xfs_bmap_add_extent_unwritten_real";
1276 #endif
1277         int                     i;      /* temp state */
1278         xfs_fileoff_t           new_endoff;     /* end offset of new entry */
1279         xfs_exntst_t            newext; /* new extent state */
1280         xfs_exntst_t            oldext; /* old extent state */
1281         xfs_bmbt_irec_t         r[3];   /* neighbor extent entries */
1282                                         /* left is 0, right is 1, prev is 2 */
1283         int                     rval=0; /* return value (logging flags) */
1284         int                     state = 0;/* state bits, accessed thru macros */
1285         enum {                          /* bit number definitions for state */
1286                 LEFT_CONTIG,    RIGHT_CONTIG,
1287                 LEFT_FILLING,   RIGHT_FILLING,
1288                 LEFT_DELAY,     RIGHT_DELAY,
1289                 LEFT_VALID,     RIGHT_VALID
1290         };
1291
1292 #define LEFT            r[0]
1293 #define RIGHT           r[1]
1294 #define PREV            r[2]
1295 #define MASK(b)         (1 << (b))
1296 #define MASK2(a,b)      (MASK(a) | MASK(b))
1297 #define MASK3(a,b,c)    (MASK2(a,b) | MASK(c))
1298 #define MASK4(a,b,c,d)  (MASK3(a,b,c) | MASK(d))
1299 #define STATE_SET(b,v)  ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1300 #define STATE_TEST(b)   (state & MASK(b))
1301 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
1302                                        ((state &= ~MASK(b)), 0))
1303 #define SWITCH_STATE            \
1304         (state & MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG))
1305
1306         /*
1307          * Set up a bunch of variables to make the tests simpler.
1308          */
1309         error = 0;
1310         cur = *curp;
1311         base = ip->i_df.if_u1.if_extents;
1312         ep = &base[idx];
1313         xfs_bmbt_get_all(ep, &PREV);
1314         newext = new->br_state;
1315         oldext = (newext == XFS_EXT_UNWRITTEN) ?
1316                 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
1317         ASSERT(PREV.br_state == oldext);
1318         new_endoff = new->br_startoff + new->br_blockcount;
1319         ASSERT(PREV.br_startoff <= new->br_startoff);
1320         ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1321         /*
1322          * Set flags determining what part of the previous oldext allocation
1323          * extent is being replaced by a newext allocation.
1324          */
1325         STATE_SET(LEFT_FILLING, PREV.br_startoff == new->br_startoff);
1326         STATE_SET(RIGHT_FILLING,
1327                 PREV.br_startoff + PREV.br_blockcount == new_endoff);
1328         /*
1329          * Check and set flags if this segment has a left neighbor.
1330          * Don't set contiguous if the combined extent would be too large.
1331          */
1332         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1333                 xfs_bmbt_get_all(ep - 1, &LEFT);
1334                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(LEFT.br_startblock));
1335         }
1336         STATE_SET(LEFT_CONTIG,
1337                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
1338                 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1339                 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1340                 LEFT.br_state == newext &&
1341                 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1342         /*
1343          * Check and set flags if this segment has a right neighbor.
1344          * Don't set contiguous if the combined extent would be too large.
1345          * Also check for all-three-contiguous being too large.
1346          */
1347         if (STATE_SET_TEST(RIGHT_VALID,
1348                         idx <
1349                         ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1)) {
1350                 xfs_bmbt_get_all(ep + 1, &RIGHT);
1351                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(RIGHT.br_startblock));
1352         }
1353         STATE_SET(RIGHT_CONTIG,
1354                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
1355                 new_endoff == RIGHT.br_startoff &&
1356                 new->br_startblock + new->br_blockcount ==
1357                     RIGHT.br_startblock &&
1358                 newext == RIGHT.br_state &&
1359                 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1360                 ((state & MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING)) !=
1361                   MASK3(LEFT_CONTIG, LEFT_FILLING, RIGHT_FILLING) ||
1362                  LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1363                      <= MAXEXTLEN));
1364         /*
1365          * Switch out based on the FILLING and CONTIG state bits.
1366          */
1367         switch (SWITCH_STATE) {
1368
1369         case MASK4(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1370                 /*
1371                  * Setting all of a previous oldext extent to newext.
1372                  * The left and right neighbors are both contiguous with new.
1373                  */
1374                 xfs_bmap_trace_pre_update(fname, "LF|RF|LC|RC", ip, idx - 1,
1375                         XFS_DATA_FORK);
1376                 xfs_bmbt_set_blockcount(ep - 1,
1377                         LEFT.br_blockcount + PREV.br_blockcount +
1378                         RIGHT.br_blockcount);
1379                 xfs_bmap_trace_post_update(fname, "LF|RF|LC|RC", ip, idx - 1,
1380                         XFS_DATA_FORK);
1381                 xfs_bmap_trace_delete(fname, "LF|RF|LC|RC", ip, idx, 2,
1382                         XFS_DATA_FORK);
1383                 xfs_bmap_delete_exlist(ip, idx, 2, XFS_DATA_FORK);
1384                 ip->i_df.if_lastex = idx - 1;
1385                 ip->i_d.di_nextents -= 2;
1386                 if (cur == NULL)
1387                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1388                 else {
1389                         rval = XFS_ILOG_CORE;
1390                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1391                                         RIGHT.br_startblock,
1392                                         RIGHT.br_blockcount, &i)))
1393                                 goto done;
1394                         ASSERT(i == 1);
1395                         if ((error = xfs_bmbt_delete(cur, &i)))
1396                                 goto done;
1397                         ASSERT(i == 1);
1398                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1399                                 goto done;
1400                         ASSERT(i == 1);
1401                         if ((error = xfs_bmbt_delete(cur, &i)))
1402                                 goto done;
1403                         ASSERT(i == 1);
1404                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1405                                 goto done;
1406                         ASSERT(i == 1);
1407                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1408                                 LEFT.br_startblock,
1409                                 LEFT.br_blockcount + PREV.br_blockcount +
1410                                 RIGHT.br_blockcount, LEFT.br_state)))
1411                                 goto done;
1412                 }
1413                 break;
1414
1415         case MASK3(LEFT_FILLING, RIGHT_FILLING, LEFT_CONTIG):
1416                 /*
1417                  * Setting all of a previous oldext extent to newext.
1418                  * The left neighbor is contiguous, the right is not.
1419                  */
1420                 xfs_bmap_trace_pre_update(fname, "LF|RF|LC", ip, idx - 1,
1421                         XFS_DATA_FORK);
1422                 xfs_bmbt_set_blockcount(ep - 1,
1423                         LEFT.br_blockcount + PREV.br_blockcount);
1424                 xfs_bmap_trace_post_update(fname, "LF|RF|LC", ip, idx - 1,
1425                         XFS_DATA_FORK);
1426                 ip->i_df.if_lastex = idx - 1;
1427                 xfs_bmap_trace_delete(fname, "LF|RF|LC", ip, idx, 1,
1428                         XFS_DATA_FORK);
1429                 xfs_bmap_delete_exlist(ip, idx, 1, XFS_DATA_FORK);
1430                 ip->i_d.di_nextents--;
1431                 if (cur == NULL)
1432                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1433                 else {
1434                         rval = XFS_ILOG_CORE;
1435                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1436                                         PREV.br_startblock, PREV.br_blockcount,
1437                                         &i)))
1438                                 goto done;
1439                         ASSERT(i == 1);
1440                         if ((error = xfs_bmbt_delete(cur, &i)))
1441                                 goto done;
1442                         ASSERT(i == 1);
1443                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1444                                 goto done;
1445                         ASSERT(i == 1);
1446                         if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
1447                                 LEFT.br_startblock,
1448                                 LEFT.br_blockcount + PREV.br_blockcount,
1449                                 LEFT.br_state)))
1450                                 goto done;
1451                 }
1452                 break;
1453
1454         case MASK3(LEFT_FILLING, RIGHT_FILLING, RIGHT_CONTIG):
1455                 /*
1456                  * Setting all of a previous oldext extent to newext.
1457                  * The right neighbor is contiguous, the left is not.
1458                  */
1459                 xfs_bmap_trace_pre_update(fname, "LF|RF|RC", ip, idx,
1460                         XFS_DATA_FORK);
1461                 xfs_bmbt_set_blockcount(ep,
1462                         PREV.br_blockcount + RIGHT.br_blockcount);
1463                 xfs_bmbt_set_state(ep, newext);
1464                 xfs_bmap_trace_post_update(fname, "LF|RF|RC", ip, idx,
1465                         XFS_DATA_FORK);
1466                 ip->i_df.if_lastex = idx;
1467                 xfs_bmap_trace_delete(fname, "LF|RF|RC", ip, idx + 1, 1,
1468                         XFS_DATA_FORK);
1469                 xfs_bmap_delete_exlist(ip, idx + 1, 1, XFS_DATA_FORK);
1470                 ip->i_d.di_nextents--;
1471                 if (cur == NULL)
1472                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1473                 else {
1474                         rval = XFS_ILOG_CORE;
1475                         if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
1476                                         RIGHT.br_startblock,
1477                                         RIGHT.br_blockcount, &i)))
1478                                 goto done;
1479                         ASSERT(i == 1);
1480                         if ((error = xfs_bmbt_delete(cur, &i)))
1481                                 goto done;
1482                         ASSERT(i == 1);
1483                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1484                                 goto done;
1485                         ASSERT(i == 1);
1486                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1487                                 new->br_startblock,
1488                                 new->br_blockcount + RIGHT.br_blockcount,
1489                                 newext)))
1490                                 goto done;
1491                 }
1492                 break;
1493
1494         case MASK2(LEFT_FILLING, RIGHT_FILLING):
1495                 /*
1496                  * Setting all of a previous oldext extent to newext.
1497                  * Neither the left nor right neighbors are contiguous with
1498                  * the new one.
1499                  */
1500                 xfs_bmap_trace_pre_update(fname, "LF|RF", ip, idx,
1501                         XFS_DATA_FORK);
1502                 xfs_bmbt_set_state(ep, newext);
1503                 xfs_bmap_trace_post_update(fname, "LF|RF", ip, idx,
1504                         XFS_DATA_FORK);
1505                 ip->i_df.if_lastex = idx;
1506                 if (cur == NULL)
1507                         rval = XFS_ILOG_DEXT;
1508                 else {
1509                         rval = 0;
1510                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1511                                         new->br_startblock, new->br_blockcount,
1512                                         &i)))
1513                                 goto done;
1514                         ASSERT(i == 1);
1515                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1516                                 new->br_startblock, new->br_blockcount,
1517                                 newext)))
1518                                 goto done;
1519                 }
1520                 break;
1521
1522         case MASK2(LEFT_FILLING, LEFT_CONTIG):
1523                 /*
1524                  * Setting the first part of a previous oldext extent to newext.
1525                  * The left neighbor is contiguous.
1526                  */
1527                 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx - 1,
1528                         XFS_DATA_FORK);
1529                 xfs_bmbt_set_blockcount(ep - 1,
1530                         LEFT.br_blockcount + new->br_blockcount);
1531                 xfs_bmbt_set_startoff(ep,
1532                         PREV.br_startoff + new->br_blockcount);
1533                 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx - 1,
1534                         XFS_DATA_FORK);
1535                 xfs_bmap_trace_pre_update(fname, "LF|LC", ip, idx,
1536                         XFS_DATA_FORK);
1537                 xfs_bmbt_set_startblock(ep,
1538                         new->br_startblock + new->br_blockcount);
1539                 xfs_bmbt_set_blockcount(ep,
1540                         PREV.br_blockcount - new->br_blockcount);
1541                 xfs_bmap_trace_post_update(fname, "LF|LC", ip, idx,
1542                         XFS_DATA_FORK);
1543                 ip->i_df.if_lastex = idx - 1;
1544                 if (cur == NULL)
1545                         rval = XFS_ILOG_DEXT;
1546                 else {
1547                         rval = 0;
1548                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1549                                         PREV.br_startblock, PREV.br_blockcount,
1550                                         &i)))
1551                                 goto done;
1552                         ASSERT(i == 1);
1553                         if ((error = xfs_bmbt_update(cur,
1554                                 PREV.br_startoff + new->br_blockcount,
1555                                 PREV.br_startblock + new->br_blockcount,
1556                                 PREV.br_blockcount - new->br_blockcount,
1557                                 oldext)))
1558                                 goto done;
1559                         if ((error = xfs_bmbt_decrement(cur, 0, &i)))
1560                                 goto done;
1561                         if (xfs_bmbt_update(cur, LEFT.br_startoff,
1562                                 LEFT.br_startblock,
1563                                 LEFT.br_blockcount + new->br_blockcount,
1564                                 LEFT.br_state))
1565                                 goto done;
1566                 }
1567                 break;
1568
1569         case MASK(LEFT_FILLING):
1570                 /*
1571                  * Setting the first part of a previous oldext extent to newext.
1572                  * The left neighbor is not contiguous.
1573                  */
1574                 xfs_bmap_trace_pre_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1575                 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
1576                 xfs_bmbt_set_startoff(ep, new_endoff);
1577                 xfs_bmbt_set_blockcount(ep,
1578                         PREV.br_blockcount - new->br_blockcount);
1579                 xfs_bmbt_set_startblock(ep,
1580                         new->br_startblock + new->br_blockcount);
1581                 xfs_bmap_trace_post_update(fname, "LF", ip, idx, XFS_DATA_FORK);
1582                 xfs_bmap_trace_insert(fname, "LF", ip, idx, 1, new, NULL,
1583                         XFS_DATA_FORK);
1584                 xfs_bmap_insert_exlist(ip, idx, 1, new, XFS_DATA_FORK);
1585                 ip->i_df.if_lastex = idx;
1586                 ip->i_d.di_nextents++;
1587                 if (cur == NULL)
1588                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1589                 else {
1590                         rval = XFS_ILOG_CORE;
1591                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1592                                         PREV.br_startblock, PREV.br_blockcount,
1593                                         &i)))
1594                                 goto done;
1595                         ASSERT(i == 1);
1596                         if ((error = xfs_bmbt_update(cur,
1597                                 PREV.br_startoff + new->br_blockcount,
1598                                 PREV.br_startblock + new->br_blockcount,
1599                                 PREV.br_blockcount - new->br_blockcount,
1600                                 oldext)))
1601                                 goto done;
1602                         cur->bc_rec.b = *new;
1603                         if ((error = xfs_bmbt_insert(cur, &i)))
1604                                 goto done;
1605                         ASSERT(i == 1);
1606                 }
1607                 break;
1608
1609         case MASK2(RIGHT_FILLING, RIGHT_CONTIG):
1610                 /*
1611                  * Setting the last part of a previous oldext extent to newext.
1612                  * The right neighbor is contiguous with the new allocation.
1613                  */
1614                 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx,
1615                         XFS_DATA_FORK);
1616                 xfs_bmap_trace_pre_update(fname, "RF|RC", ip, idx + 1,
1617                         XFS_DATA_FORK);
1618                 xfs_bmbt_set_blockcount(ep,
1619                         PREV.br_blockcount - new->br_blockcount);
1620                 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx,
1621                         XFS_DATA_FORK);
1622                 xfs_bmbt_set_allf(ep + 1, new->br_startoff, new->br_startblock,
1623                         new->br_blockcount + RIGHT.br_blockcount, newext);
1624                 xfs_bmap_trace_post_update(fname, "RF|RC", ip, idx + 1,
1625                         XFS_DATA_FORK);
1626                 ip->i_df.if_lastex = idx + 1;
1627                 if (cur == NULL)
1628                         rval = XFS_ILOG_DEXT;
1629                 else {
1630                         rval = 0;
1631                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1632                                         PREV.br_startblock,
1633                                         PREV.br_blockcount, &i)))
1634                                 goto done;
1635                         ASSERT(i == 1);
1636                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1637                                 PREV.br_startblock,
1638                                 PREV.br_blockcount - new->br_blockcount,
1639                                 oldext)))
1640                                 goto done;
1641                         if ((error = xfs_bmbt_increment(cur, 0, &i)))
1642                                 goto done;
1643                         if ((error = xfs_bmbt_update(cur, new->br_startoff,
1644                                 new->br_startblock,
1645                                 new->br_blockcount + RIGHT.br_blockcount,
1646                                 newext)))
1647                                 goto done;
1648                 }
1649                 break;
1650
1651         case MASK(RIGHT_FILLING):
1652                 /*
1653                  * Setting the last part of a previous oldext extent to newext.
1654                  * The right neighbor is not contiguous.
1655                  */
1656                 xfs_bmap_trace_pre_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1657                 xfs_bmbt_set_blockcount(ep,
1658                         PREV.br_blockcount - new->br_blockcount);
1659                 xfs_bmap_trace_post_update(fname, "RF", ip, idx, XFS_DATA_FORK);
1660                 xfs_bmap_trace_insert(fname, "RF", ip, idx + 1, 1,
1661                         new, NULL, XFS_DATA_FORK);
1662                 xfs_bmap_insert_exlist(ip, idx + 1, 1, new, XFS_DATA_FORK);
1663                 ip->i_df.if_lastex = idx + 1;
1664                 ip->i_d.di_nextents++;
1665                 if (cur == NULL)
1666                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1667                 else {
1668                         rval = XFS_ILOG_CORE;
1669                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1670                                         PREV.br_startblock, PREV.br_blockcount,
1671                                         &i)))
1672                                 goto done;
1673                         ASSERT(i == 1);
1674                         if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
1675                                 PREV.br_startblock,
1676                                 PREV.br_blockcount - new->br_blockcount,
1677                                 oldext)))
1678                                 goto done;
1679                         if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
1680                                         new->br_startblock, new->br_blockcount,
1681                                         &i)))
1682                                 goto done;
1683                         ASSERT(i == 0);
1684                         cur->bc_rec.b.br_state = XFS_EXT_NORM;
1685                         if ((error = xfs_bmbt_insert(cur, &i)))
1686                                 goto done;
1687                         ASSERT(i == 1);
1688                 }
1689                 break;
1690
1691         case 0:
1692                 /*
1693                  * Setting the middle part of a previous oldext extent to
1694                  * newext.  Contiguity is impossible here.
1695                  * One extent becomes three extents.
1696                  */
1697                 xfs_bmap_trace_pre_update(fname, "0", ip, idx, XFS_DATA_FORK);
1698                 xfs_bmbt_set_blockcount(ep,
1699                         new->br_startoff - PREV.br_startoff);
1700                 xfs_bmap_trace_post_update(fname, "0", ip, idx, XFS_DATA_FORK);
1701                 r[0] = *new;
1702                 r[1].br_startoff = new_endoff;
1703                 r[1].br_blockcount =
1704                         PREV.br_startoff + PREV.br_blockcount - new_endoff;
1705                 r[1].br_startblock = new->br_startblock + new->br_blockcount;
1706                 r[1].br_state = oldext;
1707                 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 2, &r[0], &r[1],
1708                         XFS_DATA_FORK);
1709                 xfs_bmap_insert_exlist(ip, idx + 1, 2, &r[0], XFS_DATA_FORK);
1710                 ip->i_df.if_lastex = idx + 1;
1711                 ip->i_d.di_nextents += 2;
1712                 if (cur == NULL)
1713                         rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1714                 else {
1715                         rval = XFS_ILOG_CORE;
1716                         if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
1717                                         PREV.br_startblock, PREV.br_blockcount,
1718                                         &i)))
1719                                 goto done;
1720                         ASSERT(i == 1);
1721                         /* new right extent - oldext */
1722                         if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
1723                                 r[1].br_startblock, r[1].br_blockcount,
1724                                 r[1].br_state)))
1725                                 goto done;
1726                         /* new left extent - oldext */
1727                         PREV.br_blockcount =
1728                                 new->br_startoff - PREV.br_startoff;
1729                         cur->bc_rec.b = PREV;
1730                         if ((error = xfs_bmbt_insert(cur, &i)))
1731                                 goto done;
1732                         ASSERT(i == 1);
1733                         if ((error = xfs_bmbt_increment(cur, 0, &i)))
1734                                 goto done;
1735                         ASSERT(i == 1);
1736                         /* new middle extent - newext */
1737                         cur->bc_rec.b = *new;
1738                         if ((error = xfs_bmbt_insert(cur, &i)))
1739                                 goto done;
1740                         ASSERT(i == 1);
1741                 }
1742                 break;
1743
1744         case MASK3(LEFT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1745         case MASK3(RIGHT_FILLING, LEFT_CONTIG, RIGHT_CONTIG):
1746         case MASK2(LEFT_FILLING, RIGHT_CONTIG):
1747         case MASK2(RIGHT_FILLING, LEFT_CONTIG):
1748         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1749         case MASK(LEFT_CONTIG):
1750         case MASK(RIGHT_CONTIG):
1751                 /*
1752                  * These cases are all impossible.
1753                  */
1754                 ASSERT(0);
1755         }
1756         *curp = cur;
1757 done:
1758         *logflagsp = rval;
1759         return error;
1760 #undef  LEFT
1761 #undef  RIGHT
1762 #undef  PREV
1763 #undef  MASK
1764 #undef  MASK2
1765 #undef  MASK3
1766 #undef  MASK4
1767 #undef  STATE_SET
1768 #undef  STATE_TEST
1769 #undef  STATE_SET_TEST
1770 #undef  SWITCH_STATE
1771 }
1772
1773 /*
1774  * Called by xfs_bmap_add_extent to handle cases converting a hole
1775  * to a delayed allocation.
1776  */
1777 /*ARGSUSED*/
1778 STATIC int                              /* error */
1779 xfs_bmap_add_extent_hole_delay(
1780         xfs_inode_t             *ip,    /* incore inode pointer */
1781         xfs_extnum_t            idx,    /* extent number to update/insert */
1782         xfs_btree_cur_t         *cur,   /* if null, not a btree */
1783         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
1784         int                     *logflagsp, /* inode logging flags */
1785         int                     rsvd)           /* OK to allocate reserved blocks */
1786 {
1787         xfs_bmbt_rec_t          *base;  /* base of extent entry list */
1788         xfs_bmbt_rec_t          *ep;    /* extent list entry for idx */
1789 #ifdef XFS_BMAP_TRACE
1790         static char             fname[] = "xfs_bmap_add_extent_hole_delay";
1791 #endif
1792         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1793         xfs_filblks_t           newlen=0;       /* new indirect size */
1794         xfs_filblks_t           oldlen=0;       /* old indirect size */
1795         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1796         int                     state;  /* state bits, accessed thru macros */
1797         xfs_filblks_t           temp;   /* temp for indirect calculations */
1798         enum {                          /* bit number definitions for state */
1799                 LEFT_CONTIG,    RIGHT_CONTIG,
1800                 LEFT_DELAY,     RIGHT_DELAY,
1801                 LEFT_VALID,     RIGHT_VALID
1802         };
1803
1804 #define MASK(b)                 (1 << (b))
1805 #define MASK2(a,b)              (MASK(a) | MASK(b))
1806 #define STATE_SET(b,v)          ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1807 #define STATE_TEST(b)           (state & MASK(b))
1808 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
1809                                        ((state &= ~MASK(b)), 0))
1810 #define SWITCH_STATE            (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
1811
1812         base = ip->i_df.if_u1.if_extents;
1813         ep = &base[idx];
1814         state = 0;
1815         ASSERT(ISNULLSTARTBLOCK(new->br_startblock));
1816         /*
1817          * Check and set flags if this segment has a left neighbor
1818          */
1819         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1820                 xfs_bmbt_get_all(ep - 1, &left);
1821                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
1822         }
1823         /*
1824          * Check and set flags if the current (right) segment exists.
1825          * If it doesn't exist, we're converting the hole at end-of-file.
1826          */
1827         if (STATE_SET_TEST(RIGHT_VALID,
1828                            idx <
1829                            ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1830                 xfs_bmbt_get_all(ep, &right);
1831                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
1832         }
1833         /*
1834          * Set contiguity flags on the left and right neighbors.
1835          * Don't let extents get too large, even if the pieces are contiguous.
1836          */
1837         STATE_SET(LEFT_CONTIG,
1838                 STATE_TEST(LEFT_VALID) && STATE_TEST(LEFT_DELAY) &&
1839                 left.br_startoff + left.br_blockcount == new->br_startoff &&
1840                 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
1841         STATE_SET(RIGHT_CONTIG,
1842                 STATE_TEST(RIGHT_VALID) && STATE_TEST(RIGHT_DELAY) &&
1843                 new->br_startoff + new->br_blockcount == right.br_startoff &&
1844                 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
1845                 (!STATE_TEST(LEFT_CONTIG) ||
1846                  (left.br_blockcount + new->br_blockcount +
1847                      right.br_blockcount <= MAXEXTLEN)));
1848         /*
1849          * Switch out based on the contiguity flags.
1850          */
1851         switch (SWITCH_STATE) {
1852
1853         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
1854                 /*
1855                  * New allocation is contiguous with delayed allocations
1856                  * on the left and on the right.
1857                  * Merge all three into a single extent list entry.
1858                  */
1859                 temp = left.br_blockcount + new->br_blockcount +
1860                         right.br_blockcount;
1861                 xfs_bmap_trace_pre_update(fname, "LC|RC", ip, idx - 1,
1862                         XFS_DATA_FORK);
1863                 xfs_bmbt_set_blockcount(ep - 1, temp);
1864                 oldlen = STARTBLOCKVAL(left.br_startblock) +
1865                         STARTBLOCKVAL(new->br_startblock) +
1866                         STARTBLOCKVAL(right.br_startblock);
1867                 newlen = xfs_bmap_worst_indlen(ip, temp);
1868                 xfs_bmbt_set_startblock(ep - 1, NULLSTARTBLOCK((int)newlen));
1869                 xfs_bmap_trace_post_update(fname, "LC|RC", ip, idx - 1,
1870                         XFS_DATA_FORK);
1871                 xfs_bmap_trace_delete(fname, "LC|RC", ip, idx, 1,
1872                         XFS_DATA_FORK);
1873                 xfs_bmap_delete_exlist(ip, idx, 1, XFS_DATA_FORK);
1874                 ip->i_df.if_lastex = idx - 1;
1875                 break;
1876
1877         case MASK(LEFT_CONTIG):
1878                 /*
1879                  * New allocation is contiguous with a delayed allocation
1880                  * on the left.
1881                  * Merge the new allocation with the left neighbor.
1882                  */
1883                 temp = left.br_blockcount + new->br_blockcount;
1884                 xfs_bmap_trace_pre_update(fname, "LC", ip, idx - 1,
1885                         XFS_DATA_FORK);
1886                 xfs_bmbt_set_blockcount(ep - 1, temp);
1887                 oldlen = STARTBLOCKVAL(left.br_startblock) +
1888                         STARTBLOCKVAL(new->br_startblock);
1889                 newlen = xfs_bmap_worst_indlen(ip, temp);
1890                 xfs_bmbt_set_startblock(ep - 1, NULLSTARTBLOCK((int)newlen));
1891                 xfs_bmap_trace_post_update(fname, "LC", ip, idx - 1,
1892                         XFS_DATA_FORK);
1893                 ip->i_df.if_lastex = idx - 1;
1894                 break;
1895
1896         case MASK(RIGHT_CONTIG):
1897                 /*
1898                  * New allocation is contiguous with a delayed allocation
1899                  * on the right.
1900                  * Merge the new allocation with the right neighbor.
1901                  */
1902                 xfs_bmap_trace_pre_update(fname, "RC", ip, idx, XFS_DATA_FORK);
1903                 temp = new->br_blockcount + right.br_blockcount;
1904                 oldlen = STARTBLOCKVAL(new->br_startblock) +
1905                         STARTBLOCKVAL(right.br_startblock);
1906                 newlen = xfs_bmap_worst_indlen(ip, temp);
1907                 xfs_bmbt_set_allf(ep, new->br_startoff,
1908                         NULLSTARTBLOCK((int)newlen), temp, right.br_state);
1909                 xfs_bmap_trace_post_update(fname, "RC", ip, idx, XFS_DATA_FORK);
1910                 ip->i_df.if_lastex = idx;
1911                 break;
1912
1913         case 0:
1914                 /*
1915                  * New allocation is not contiguous with another
1916                  * delayed allocation.
1917                  * Insert a new entry.
1918                  */
1919                 oldlen = newlen = 0;
1920                 xfs_bmap_trace_insert(fname, "0", ip, idx, 1, new, NULL,
1921                         XFS_DATA_FORK);
1922                 xfs_bmap_insert_exlist(ip, idx, 1, new, XFS_DATA_FORK);
1923                 ip->i_df.if_lastex = idx;
1924                 break;
1925         }
1926         if (oldlen != newlen) {
1927                 ASSERT(oldlen > newlen);
1928                 xfs_mod_incore_sb(ip->i_mount, XFS_SBS_FDBLOCKS,
1929                         (int)(oldlen - newlen), rsvd);
1930                 /*
1931                  * Nothing to do for disk quota accounting here.
1932                  */
1933         }
1934         *logflagsp = 0;
1935         return 0;
1936 #undef  MASK
1937 #undef  MASK2
1938 #undef  STATE_SET
1939 #undef  STATE_TEST
1940 #undef  STATE_SET_TEST
1941 #undef  SWITCH_STATE
1942 }
1943
1944 /*
1945  * Called by xfs_bmap_add_extent to handle cases converting a hole
1946  * to a real allocation.
1947  */
1948 STATIC int                              /* error */
1949 xfs_bmap_add_extent_hole_real(
1950         xfs_inode_t             *ip,    /* incore inode pointer */
1951         xfs_extnum_t            idx,    /* extent number to update/insert */
1952         xfs_btree_cur_t         *cur,   /* if null, not a btree */
1953         xfs_bmbt_irec_t         *new,   /* new data to put in extent list */
1954         int                     *logflagsp, /* inode logging flags */
1955         int                     whichfork) /* data or attr fork */
1956 {
1957         xfs_bmbt_rec_t          *ep;    /* pointer to extent entry ins. point */
1958         int                     error;  /* error return value */
1959 #ifdef XFS_BMAP_TRACE
1960         static char             fname[] = "xfs_bmap_add_extent_hole_real";
1961 #endif
1962         int                     i;      /* temp state */
1963         xfs_ifork_t             *ifp;   /* inode fork pointer */
1964         xfs_bmbt_irec_t         left;   /* left neighbor extent entry */
1965         xfs_bmbt_irec_t         right;  /* right neighbor extent entry */
1966         int                     state;  /* state bits, accessed thru macros */
1967         enum {                          /* bit number definitions for state */
1968                 LEFT_CONTIG,    RIGHT_CONTIG,
1969                 LEFT_DELAY,     RIGHT_DELAY,
1970                 LEFT_VALID,     RIGHT_VALID
1971         };
1972
1973 #define MASK(b)                 (1 << (b))
1974 #define MASK2(a,b)              (MASK(a) | MASK(b))
1975 #define STATE_SET(b,v)          ((v) ? (state |= MASK(b)) : (state &= ~MASK(b)))
1976 #define STATE_TEST(b)           (state & MASK(b))
1977 #define STATE_SET_TEST(b,v)     ((v) ? ((state |= MASK(b)), 1) : \
1978                                        ((state &= ~MASK(b)), 0))
1979 #define SWITCH_STATE            (state & MASK2(LEFT_CONTIG, RIGHT_CONTIG))
1980
1981         ifp = XFS_IFORK_PTR(ip, whichfork);
1982         ASSERT(idx <= ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t));
1983         ep = &ifp->if_u1.if_extents[idx];
1984         state = 0;
1985         /*
1986          * Check and set flags if this segment has a left neighbor.
1987          */
1988         if (STATE_SET_TEST(LEFT_VALID, idx > 0)) {
1989                 xfs_bmbt_get_all(ep - 1, &left);
1990                 STATE_SET(LEFT_DELAY, ISNULLSTARTBLOCK(left.br_startblock));
1991         }
1992         /*
1993          * Check and set flags if this segment has a current value.
1994          * Not true if we're inserting into the "hole" at eof.
1995          */
1996         if (STATE_SET_TEST(RIGHT_VALID,
1997                            idx <
1998                            ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1999                 xfs_bmbt_get_all(ep, &right);
2000                 STATE_SET(RIGHT_DELAY, ISNULLSTARTBLOCK(right.br_startblock));
2001         }
2002         /*
2003          * We're inserting a real allocation between "left" and "right".
2004          * Set the contiguity flags.  Don't let extents get too large.
2005          */
2006         STATE_SET(LEFT_CONTIG,
2007                 STATE_TEST(LEFT_VALID) && !STATE_TEST(LEFT_DELAY) &&
2008                 left.br_startoff + left.br_blockcount == new->br_startoff &&
2009                 left.br_startblock + left.br_blockcount == new->br_startblock &&
2010                 left.br_state == new->br_state &&
2011                 left.br_blockcount + new->br_blockcount <= MAXEXTLEN);
2012         STATE_SET(RIGHT_CONTIG,
2013                 STATE_TEST(RIGHT_VALID) && !STATE_TEST(RIGHT_DELAY) &&
2014                 new->br_startoff + new->br_blockcount == right.br_startoff &&
2015                 new->br_startblock + new->br_blockcount ==
2016                     right.br_startblock &&
2017                 new->br_state == right.br_state &&
2018                 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2019                 (!STATE_TEST(LEFT_CONTIG) ||
2020                  left.br_blockcount + new->br_blockcount +
2021                      right.br_blockcount <= MAXEXTLEN));
2022
2023         /*
2024          * Select which case we're in here, and implement it.
2025          */
2026         switch (SWITCH_STATE) {
2027
2028         case MASK2(LEFT_CONTIG, RIGHT_CONTIG):
2029                 /*
2030                  * New allocation is contiguous with real allocations on the
2031                  * left and on the right.
2032                  * Merge all three into a single extent list entry.
2033                  */
2034                 xfs_bmap_trace_pre_update(fname, "LC|RC", ip, idx - 1,
2035                         whichfork);
2036                 xfs_bmbt_set_blockcount(ep - 1,
2037                         left.br_blockcount + new->br_blockcount +
2038                         right.br_blockcount);
2039                 xfs_bmap_trace_post_update(fname, "LC|RC", ip, idx - 1,
2040                         whichfork);
2041                 xfs_bmap_trace_delete(fname, "LC|RC", ip,
2042                         idx, 1, whichfork);
2043                 xfs_bmap_delete_exlist(ip, idx, 1, whichfork);
2044                 ifp->if_lastex = idx - 1;
2045                 XFS_IFORK_NEXT_SET(ip, whichfork,
2046                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2047                 if (cur == NULL) {
2048                         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2049                         return 0;
2050                 }
2051                 *logflagsp = XFS_ILOG_CORE;
2052                 if ((error = xfs_bmbt_lookup_eq(cur, right.br_startoff,
2053                                 right.br_startblock, right.br_blockcount, &i)))
2054                         return error;
2055                 ASSERT(i == 1);
2056                 if ((error = xfs_bmbt_delete(cur, &i)))
2057                         return error;
2058                 ASSERT(i == 1);
2059                 if ((error = xfs_bmbt_decrement(cur, 0, &i)))
2060                         return error;
2061                 ASSERT(i == 1);
2062                 error = xfs_bmbt_update(cur, left.br_startoff,
2063                                 left.br_startblock,
2064                                 left.br_blockcount + new->br_blockcount +
2065                                 right.br_blockcount, left.br_state);
2066                 return error;
2067
2068         case MASK(LEFT_CONTIG):
2069                 /*
2070                  * New allocation is contiguous with a real allocation
2071                  * on the left.
2072                  * Merge the new allocation with the left neighbor.
2073                  */
2074                 xfs_bmap_trace_pre_update(fname, "LC", ip, idx - 1, whichfork);
2075                 xfs_bmbt_set_blockcount(ep - 1,
2076                         left.br_blockcount + new->br_blockcount);
2077                 xfs_bmap_trace_post_update(fname, "LC", ip, idx - 1, whichfork);
2078                 ifp->if_lastex = idx - 1;
2079                 if (cur == NULL) {
2080                         *logflagsp = XFS_ILOG_FEXT(whichfork);
2081                         return 0;
2082                 }
2083                 *logflagsp = 0;
2084                 if ((error = xfs_bmbt_lookup_eq(cur, left.br_startoff,
2085                                 left.br_startblock, left.br_blockcount, &i)))
2086                         return error;
2087                 ASSERT(i == 1);
2088                 error = xfs_bmbt_update(cur, left.br_startoff,
2089                                 left.br_startblock,
2090                                 left.br_blockcount + new->br_blockcount,
2091                                 left.br_state);
2092                 return error;
2093
2094         case MASK(RIGHT_CONTIG):
2095                 /*
2096                  * New allocation is contiguous with a real allocation
2097                  * on the right.
2098                  * Merge the new allocation with the right neighbor.
2099                  */
2100                 xfs_bmap_trace_pre_update(fname, "RC", ip, idx, whichfork);
2101                 xfs_bmbt_set_allf(ep, new->br_startoff, new->br_startblock,
2102                         new->br_blockcount + right.br_blockcount,
2103                         right.br_state);
2104                 xfs_bmap_trace_post_update(fname, "RC", ip, idx, whichfork);
2105                 ifp->if_lastex = idx;
2106                 if (cur == NULL) {
2107                         *logflagsp = XFS_ILOG_FEXT(whichfork);
2108                         return 0;
2109                 }
2110                 *logflagsp = 0;
2111                 if ((error = xfs_bmbt_lookup_eq(cur, right.br_startoff,
2112                                 right.br_startblock, right.br_blockcount, &i)))
2113                         return error;
2114                 ASSERT(i == 1);
2115                 error = xfs_bmbt_update(cur, new->br_startoff,
2116                                 new->br_startblock,
2117                                 new->br_blockcount + right.br_blockcount,
2118                                 right.br_state);
2119                 return error;
2120
2121         case 0:
2122                 /*
2123                  * New allocation is not contiguous with another
2124                  * real allocation.
2125                  * Insert a new entry.
2126                  */
2127                 xfs_bmap_trace_insert(fname, "0", ip, idx, 1, new, NULL,
2128                         whichfork);
2129                 xfs_bmap_insert_exlist(ip, idx, 1, new, whichfork);
2130                 ifp->if_lastex = idx;
2131                 XFS_IFORK_NEXT_SET(ip, whichfork,
2132                         XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
2133                 if (cur == NULL) {
2134                         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2135                         return 0;
2136                 }
2137                 *logflagsp = XFS_ILOG_CORE;
2138                 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2139                                 new->br_startblock, new->br_blockcount, &i)))
2140                         return error;
2141                 ASSERT(i == 0);
2142                 cur->bc_rec.b.br_state = new->br_state;
2143                 if ((error = xfs_bmbt_insert(cur, &i)))
2144                         return error;
2145                 ASSERT(i == 1);
2146                 return 0;
2147         }
2148 #undef  MASK
2149 #undef  MASK2
2150 #undef  STATE_SET
2151 #undef  STATE_TEST
2152 #undef  STATE_SET_TEST
2153 #undef  SWITCH_STATE
2154         /* NOTREACHED */
2155         ASSERT(0);
2156         return 0; /* keep gcc quite */
2157 }
2158
2159 #define XFS_ALLOC_GAP_UNITS     4
2160
2161 /*
2162  * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
2163  * It figures out where to ask the underlying allocator to put the new extent.
2164  */
2165 STATIC int                              /* error */
2166 xfs_bmap_alloc(
2167         xfs_bmalloca_t  *ap)            /* bmap alloc argument struct */
2168 {
2169         xfs_fsblock_t   adjust;         /* adjustment to block numbers */
2170         xfs_alloctype_t atype=0;        /* type for allocation routines */
2171         int             error;          /* error return value */
2172         xfs_agnumber_t  fb_agno;        /* ag number of ap->firstblock */
2173         xfs_mount_t     *mp;            /* mount point structure */
2174         int             nullfb;         /* true if ap->firstblock isn't set */
2175         int             rt;             /* true if inode is realtime */
2176 #ifdef __KERNEL__
2177         xfs_extlen_t    prod=0;         /* product factor for allocators */
2178         xfs_extlen_t    ralen=0;        /* realtime allocation length */
2179 #endif
2180
2181 #define ISVALID(x,y)    \
2182         (rt ? \
2183                 (x) < mp->m_sb.sb_rblocks : \
2184                 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
2185                 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
2186                 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
2187
2188         /*
2189          * Set up variables.
2190          */
2191         mp = ap->ip->i_mount;
2192         nullfb = ap->firstblock == NULLFSBLOCK;
2193         rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
2194         fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, ap->firstblock);
2195 #ifdef __KERNEL__
2196         if (rt) {
2197                 xfs_extlen_t    extsz;          /* file extent size for rt */
2198                 xfs_fileoff_t   nexto;          /* next file offset */
2199                 xfs_extlen_t    orig_alen;      /* original ap->alen */
2200                 xfs_fileoff_t   orig_end;       /* original off+len */
2201                 xfs_fileoff_t   orig_off;       /* original ap->off */
2202                 xfs_extlen_t    mod_off;        /* modulus calculations */
2203                 xfs_fileoff_t   prevo;          /* previous file offset */
2204                 xfs_rtblock_t   rtx;            /* realtime extent number */
2205                 xfs_extlen_t    temp;           /* temp for rt calculations */
2206
2207                 /*
2208                  * Set prod to match the realtime extent size.
2209                  */
2210                 if (!(extsz = ap->ip->i_d.di_extsize))
2211                         extsz = mp->m_sb.sb_rextsize;
2212                 prod = extsz / mp->m_sb.sb_rextsize;
2213                 orig_off = ap->off;
2214                 orig_alen = ap->alen;
2215                 orig_end = orig_off + orig_alen;
2216                 /*
2217                  * If the file offset is unaligned vs. the extent size
2218                  * we need to align it.  This will be possible unless
2219                  * the file was previously written with a kernel that didn't
2220                  * perform this alignment.
2221                  */
2222                 mod_off = do_mod(orig_off, extsz);
2223                 if (mod_off) {
2224                         ap->alen += mod_off;
2225                         ap->off -= mod_off;
2226                 }
2227                 /*
2228                  * Same adjustment for the end of the requested area.
2229                  */
2230                 if ((temp = (ap->alen % extsz)))
2231                         ap->alen += extsz - temp;
2232                 /*
2233                  * If the previous block overlaps with this proposed allocation
2234                  * then move the start forward without adjusting the length.
2235                  */
2236                 prevo =
2237                         ap->prevp->br_startoff == NULLFILEOFF ?
2238                                 0 :
2239                                 (ap->prevp->br_startoff +
2240                                  ap->prevp->br_blockcount);
2241                 if (ap->off != orig_off && ap->off < prevo)
2242                         ap->off = prevo;
2243                 /*
2244                  * If the next block overlaps with this proposed allocation
2245                  * then move the start back without adjusting the length,
2246                  * but not before offset 0.
2247                  * This may of course make the start overlap previous block,
2248                  * and if we hit the offset 0 limit then the next block
2249                  * can still overlap too.
2250                  */
2251                 nexto = (ap->eof || ap->gotp->br_startoff == NULLFILEOFF) ?
2252                         NULLFILEOFF : ap->gotp->br_startoff;
2253                 if (!ap->eof &&
2254                     ap->off + ap->alen != orig_end &&
2255                     ap->off + ap->alen > nexto)
2256                         ap->off = nexto > ap->alen ? nexto - ap->alen : 0;
2257                 /*
2258                  * If we're now overlapping the next or previous extent that
2259                  * means we can't fit an extsz piece in this hole.  Just move
2260                  * the start forward to the first valid spot and set
2261                  * the length so we hit the end.
2262                  */
2263                 if ((ap->off != orig_off && ap->off < prevo) ||
2264                     (ap->off + ap->alen != orig_end &&
2265                      ap->off + ap->alen > nexto)) {
2266                         ap->off = prevo;
2267                         ap->alen = nexto - prevo;
2268                 }
2269                 /*
2270                  * If the result isn't a multiple of rtextents we need to
2271                  * remove blocks until it is.
2272                  */
2273                 if ((temp = (ap->alen % mp->m_sb.sb_rextsize))) {
2274                         /*
2275                          * We're not covering the original request, or
2276                          * we won't be able to once we fix the length.
2277                          */
2278                         if (orig_off < ap->off ||
2279                             orig_end > ap->off + ap->alen ||
2280                             ap->alen - temp < orig_alen)
2281                                 return XFS_ERROR(EINVAL);
2282                         /*
2283                          * Try to fix it by moving the start up.
2284                          */
2285                         if (ap->off + temp <= orig_off) {
2286                                 ap->alen -= temp;
2287                                 ap->off += temp;
2288                         }
2289                         /*
2290                          * Try to fix it by moving the end in.
2291                          */
2292                         else if (ap->off + ap->alen - temp >= orig_end)
2293                                 ap->alen -= temp;
2294                         /*
2295                          * Set the start to the minimum then trim the length.
2296                          */
2297                         else {
2298                                 ap->alen -= orig_off - ap->off;
2299                                 ap->off = orig_off;
2300                                 ap->alen -= ap->alen % mp->m_sb.sb_rextsize;
2301                         }
2302                         /*
2303                          * Result doesn't cover the request, fail it.
2304                          */
2305                         if (orig_off < ap->off || orig_end > ap->off + ap->alen)
2306                                 return XFS_ERROR(EINVAL);
2307                 }
2308                 ASSERT(ap->alen % mp->m_sb.sb_rextsize == 0);
2309                 /*
2310                  * If the offset & length are not perfectly aligned
2311                  * then kill prod, it will just get us in trouble.
2312                  */
2313                 if (do_mod(ap->off, extsz) || ap->alen % extsz)
2314                         prod = 1;
2315                 /*
2316                  * Set ralen to be the actual requested length in rtextents.
2317                  */
2318                 ralen = ap->alen / mp->m_sb.sb_rextsize;
2319                 /*
2320                  * If the old value was close enough to MAXEXTLEN that
2321                  * we rounded up to it, cut it back so it's valid again.
2322                  * Note that if it's a really large request (bigger than
2323                  * MAXEXTLEN), we don't hear about that number, and can't
2324                  * adjust the starting point to match it.
2325                  */
2326                 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
2327                         ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
2328                 /*
2329                  * If it's an allocation to an empty file at offset 0,
2330                  * pick an extent that will space things out in the rt area.
2331                  */
2332                 if (ap->eof && ap->off == 0) {
2333                         error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
2334                         if (error)
2335                                 return error;
2336                         ap->rval = rtx * mp->m_sb.sb_rextsize;
2337                 } else
2338                         ap->rval = 0;
2339         }
2340 #else
2341         if (rt)
2342                 ap->rval = 0;
2343 #endif  /* __KERNEL__ */
2344         else if (nullfb)
2345                 ap->rval = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
2346         else
2347                 ap->rval = ap->firstblock;
2348         /*
2349          * If allocating at eof, and there's a previous real block,
2350          * try to use it's last block as our starting point.
2351          */
2352         if (ap->eof && ap->prevp->br_startoff != NULLFILEOFF &&
2353             !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2354             ISVALID(ap->prevp->br_startblock + ap->prevp->br_blockcount,
2355                     ap->prevp->br_startblock)) {
2356                 ap->rval = ap->prevp->br_startblock + ap->prevp->br_blockcount;
2357                 /*
2358                  * Adjust for the gap between prevp and us.
2359                  */
2360                 adjust = ap->off -
2361                         (ap->prevp->br_startoff + ap->prevp->br_blockcount);
2362                 if (adjust &&
2363                     ISVALID(ap->rval + adjust, ap->prevp->br_startblock))
2364                         ap->rval += adjust;
2365         }
2366         /*
2367          * If not at eof, then compare the two neighbor blocks.
2368          * Figure out whether either one gives us a good starting point,
2369          * and pick the better one.
2370          */
2371         else if (!ap->eof) {
2372                 xfs_fsblock_t   gotbno;         /* right side block number */
2373                 xfs_fsblock_t   gotdiff=0;      /* right side difference */
2374                 xfs_fsblock_t   prevbno;        /* left side block number */
2375                 xfs_fsblock_t   prevdiff=0;     /* left side difference */
2376
2377                 /*
2378                  * If there's a previous (left) block, select a requested
2379                  * start block based on it.
2380                  */
2381                 if (ap->prevp->br_startoff != NULLFILEOFF &&
2382                     !ISNULLSTARTBLOCK(ap->prevp->br_startblock) &&
2383                     (prevbno = ap->prevp->br_startblock +
2384                                ap->prevp->br_blockcount) &&
2385                     ISVALID(prevbno, ap->prevp->br_startblock)) {
2386                         /*
2387                          * Calculate gap to end of previous block.
2388                          */
2389                         adjust = prevdiff = ap->off -
2390                                 (ap->prevp->br_startoff +
2391                                  ap->prevp->br_blockcount);
2392                         /*
2393                          * Figure the startblock based on the previous block's
2394                          * end and the gap size.
2395                          * Heuristic!
2396                          * If the gap is large relative to the piece we're
2397                          * allocating, or using it gives us an invalid block
2398                          * number, then just use the end of the previous block.
2399                          */
2400                         if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2401                             ISVALID(prevbno + prevdiff,
2402                                     ap->prevp->br_startblock))
2403                                 prevbno += adjust;
2404                         else
2405                                 prevdiff += adjust;
2406                         /*
2407                          * If the firstblock forbids it, can't use it,
2408                          * must use default.
2409                          */
2410                         if (!rt && !nullfb &&
2411                             XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
2412                                 prevbno = NULLFSBLOCK;
2413                 }
2414                 /*
2415                  * No previous block or can't follow it, just default.
2416                  */
2417                 else
2418                         prevbno = NULLFSBLOCK;
2419                 /*
2420                  * If there's a following (right) block, select a requested
2421                  * start block based on it.
2422                  */
2423                 if (!ISNULLSTARTBLOCK(ap->gotp->br_startblock)) {
2424                         /*
2425                          * Calculate gap to start of next block.
2426                          */
2427                         adjust = gotdiff = ap->gotp->br_startoff - ap->off;
2428                         /*
2429                          * Figure the startblock based on the next block's
2430                          * start and the gap size.
2431                          */
2432                         gotbno = ap->gotp->br_startblock;
2433                         /*
2434                          * Heuristic!
2435                          * If the gap is large relative to the piece we're
2436                          * allocating, or using it gives us an invalid block
2437                          * number, then just use the start of the next block
2438                          * offset by our length.
2439                          */
2440                         if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->alen &&
2441                             ISVALID(gotbno - gotdiff, gotbno))
2442                                 gotbno -= adjust;
2443                         else if (ISVALID(gotbno - ap->alen, gotbno)) {
2444                                 gotbno -= ap->alen;
2445                                 gotdiff += adjust - ap->alen;
2446                         } else
2447                                 gotdiff += adjust;
2448                         /*
2449                          * If the firstblock forbids it, can't use it,
2450                          * must use default.
2451                          */
2452                         if (!rt && !nullfb &&
2453                             XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
2454                                 gotbno = NULLFSBLOCK;
2455                 }
2456                 /*
2457                  * No next block, just default.
2458                  */
2459                 else
2460                         gotbno = NULLFSBLOCK;
2461                 /*
2462                  * If both valid, pick the better one, else the only good
2463                  * one, else ap->rval is already set (to 0 or the inode block).
2464                  */
2465                 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
2466                         ap->rval = prevdiff <= gotdiff ? prevbno : gotbno;
2467                 else if (prevbno != NULLFSBLOCK)
2468                         ap->rval = prevbno;
2469                 else if (gotbno != NULLFSBLOCK)
2470                         ap->rval = gotbno;
2471         }
2472         /*
2473          * If allowed, use ap->rval; otherwise must use firstblock since
2474          * it's in the right allocation group.
2475          */
2476         if (nullfb || rt || XFS_FSB_TO_AGNO(mp, ap->rval) == fb_agno)
2477                 ;
2478         else
2479                 ap->rval = ap->firstblock;
2480         /*
2481          * Realtime allocation, done through xfs_rtallocate_extent.
2482          */
2483         if (rt) {
2484 #ifndef __KERNEL__
2485                 ASSERT(0);
2486 #else
2487                 xfs_rtblock_t   rtb;
2488
2489                 atype = ap->rval == 0 ?
2490                         XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
2491                 do_div(ap->rval, mp->m_sb.sb_rextsize);
2492                 rtb = ap->rval;
2493                 ap->alen = ralen;
2494                 if ((error = xfs_rtallocate_extent(ap->tp, ap->rval, 1, ap->alen,
2495                                 &ralen, atype, ap->wasdel, prod, &rtb)))
2496                         return error;
2497                 if (rtb == NULLFSBLOCK && prod > 1 &&
2498                     (error = xfs_rtallocate_extent(ap->tp, ap->rval, 1,
2499                                                    ap->alen, &ralen, atype,
2500                                                    ap->wasdel, 1, &rtb)))
2501                         return error;
2502                 ap->rval = rtb;
2503                 if (ap->rval != NULLFSBLOCK) {
2504                         ap->rval *= mp->m_sb.sb_rextsize;
2505                         ralen *= mp->m_sb.sb_rextsize;
2506                         ap->alen = ralen;
2507                         ap->ip->i_d.di_nblocks += ralen;
2508                         xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2509                         if (ap->wasdel)
2510                                 ap->ip->i_delayed_blks -= ralen;
2511                         /*
2512                          * Adjust the disk quota also. This was reserved
2513                          * earlier.
2514                          */
2515                         XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2516                                 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
2517                                                 XFS_TRANS_DQ_RTBCOUNT,
2518                                 (long) ralen);
2519                 } else
2520                         ap->alen = 0;
2521 #endif  /* __KERNEL__ */
2522         }
2523         /*
2524          * Normal allocation, done through xfs_alloc_vextent.
2525          */
2526         else {
2527                 xfs_agnumber_t  ag;
2528                 xfs_alloc_arg_t args;
2529                 xfs_extlen_t    blen;
2530                 xfs_extlen_t    delta;
2531                 int             isaligned;
2532                 xfs_extlen_t    longest;
2533                 xfs_extlen_t    need;
2534                 xfs_extlen_t    nextminlen=0;
2535                 int             notinit;
2536                 xfs_perag_t     *pag;
2537                 xfs_agnumber_t  startag;
2538                 int             tryagain;
2539
2540                 tryagain = isaligned = 0;
2541                 args.tp = ap->tp;
2542                 args.mp = mp;
2543                 args.fsbno = ap->rval;
2544                 args.maxlen = MIN(ap->alen, mp->m_sb.sb_agblocks);
2545                 blen = 0;
2546                 if (nullfb) {
2547                         args.type = XFS_ALLOCTYPE_START_BNO;
2548                         args.total = ap->total;
2549                         /*
2550                          * Find the longest available space.
2551                          * We're going to try for the whole allocation at once.
2552                          */
2553                         startag = ag = XFS_FSB_TO_AGNO(mp, args.fsbno);
2554                         notinit = 0;
2555                         down_read(&mp->m_peraglock);
2556                         while (blen < ap->alen) {
2557                                 pag = &mp->m_perag[ag];
2558                                 if (!pag->pagf_init &&
2559                                     (error = xfs_alloc_pagf_init(mp, args.tp,
2560                                             ag, XFS_ALLOC_FLAG_TRYLOCK))) {
2561                                         up_read(&mp->m_peraglock);
2562                                         return error;
2563                                 }
2564                                 /*
2565                                  * See xfs_alloc_fix_freelist...
2566                                  */
2567                                 if (pag->pagf_init) {
2568                                         need = XFS_MIN_FREELIST_PAG(pag, mp);
2569                                         delta = need > pag->pagf_flcount ?
2570                                                 need - pag->pagf_flcount : 0;
2571                                         longest = (pag->pagf_longest > delta) ?
2572                                                 (pag->pagf_longest - delta) :
2573                                                 (pag->pagf_flcount > 0 ||
2574                                                  pag->pagf_longest > 0);
2575                                         if (blen < longest)
2576                                                 blen = longest;
2577                                 } else
2578                                         notinit = 1;
2579                                 if (++ag == mp->m_sb.sb_agcount)
2580                                         ag = 0;
2581                                 if (ag == startag)
2582                                         break;
2583                         }
2584                         up_read(&mp->m_peraglock);
2585                         /*
2586                          * Since the above loop did a BUF_TRYLOCK, it is
2587                          * possible that there is space for this request.
2588                          */
2589                         if (notinit || blen < ap->minlen)
2590                                 args.minlen = ap->minlen;
2591                         /*
2592                          * If the best seen length is less than the request
2593                          * length, use the best as the minimum.
2594                          */
2595                         else if (blen < ap->alen)
2596                                 args.minlen = blen;
2597                         /*
2598                          * Otherwise we've seen an extent as big as alen,
2599                          * use that as the minimum.
2600                          */
2601                         else
2602                                 args.minlen = ap->alen;
2603                 } else if (ap->low) {
2604                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2605                         args.total = args.minlen = ap->minlen;
2606                 } else {
2607                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
2608                         args.total = ap->total;
2609                         args.minlen = ap->minlen;
2610                 }
2611                 if (ap->ip->i_d.di_extsize) {
2612                         args.prod = ap->ip->i_d.di_extsize;
2613                         if ((args.mod = (xfs_extlen_t)do_mod(ap->off, args.prod)))
2614                                 args.mod = (xfs_extlen_t)(args.prod - args.mod);
2615                 } else if (mp->m_sb.sb_blocksize >= NBPP) {
2616                         args.prod = 1;
2617                         args.mod = 0;
2618                 } else {
2619                         args.prod = NBPP >> mp->m_sb.sb_blocklog;
2620                         if ((args.mod = (xfs_extlen_t)(do_mod(ap->off, args.prod))))
2621                                 args.mod = (xfs_extlen_t)(args.prod - args.mod);
2622                 }
2623                 /*
2624                  * If we are not low on available data blocks, and the
2625                  * underlying logical volume manager is a stripe, and
2626                  * the file offset is zero then try to allocate data
2627                  * blocks on stripe unit boundary.
2628                  * NOTE: ap->aeof is only set if the allocation length
2629                  * is >= the stripe unit and the allocation offset is
2630                  * at the end of file.
2631                  */
2632                 if (!ap->low && ap->aeof) {
2633                         if (!ap->off) {
2634                                 args.alignment = mp->m_dalign;
2635                                 atype = args.type;
2636                                 isaligned = 1;
2637                                 /*
2638                                  * Adjust for alignment
2639                                  */
2640                                 if (blen > args.alignment && blen <= ap->alen)
2641                                         args.minlen = blen - args.alignment;
2642                                 args.minalignslop = 0;
2643                         } else {
2644                                 /*
2645                                  * First try an exact bno allocation.
2646                                  * If it fails then do a near or start bno
2647                                  * allocation with alignment turned on.
2648                                  */
2649                                 atype = args.type;
2650                                 tryagain = 1;
2651                                 args.type = XFS_ALLOCTYPE_THIS_BNO;
2652                                 args.alignment = 1;
2653                                 /*
2654                                  * Compute the minlen+alignment for the
2655                                  * next case.  Set slop so that the value
2656                                  * of minlen+alignment+slop doesn't go up
2657                                  * between the calls.
2658                                  */
2659                                 if (blen > mp->m_dalign && blen <= ap->alen)
2660                                         nextminlen = blen - mp->m_dalign;
2661                                 else
2662                                         nextminlen = args.minlen;
2663                                 if (nextminlen + mp->m_dalign > args.minlen + 1)
2664                                         args.minalignslop =
2665                                                 nextminlen + mp->m_dalign -
2666                                                 args.minlen - 1;
2667                                 else
2668                                         args.minalignslop = 0;
2669                         }
2670                 } else {
2671                         args.alignment = 1;
2672                         args.minalignslop = 0;
2673                 }
2674                 args.minleft = ap->minleft;
2675                 args.wasdel = ap->wasdel;
2676                 args.isfl = 0;
2677                 args.userdata = ap->userdata;
2678                 if ((error = xfs_alloc_vextent(&args)))
2679                         return error;
2680                 if (tryagain && args.fsbno == NULLFSBLOCK) {
2681                         /*
2682                          * Exact allocation failed. Now try with alignment
2683                          * turned on.
2684                          */
2685                         args.type = atype;
2686                         args.fsbno = ap->rval;
2687                         args.alignment = mp->m_dalign;
2688                         args.minlen = nextminlen;
2689                         args.minalignslop = 0;
2690                         isaligned = 1;
2691                         if ((error = xfs_alloc_vextent(&args)))
2692                                 return error;
2693                 }
2694                 if (isaligned && args.fsbno == NULLFSBLOCK) {
2695                         /*
2696                          * allocation failed, so turn off alignment and
2697                          * try again.
2698                          */
2699                         args.type = atype;
2700                         args.fsbno = ap->rval;
2701                         args.alignment = 0;
2702                         if ((error = xfs_alloc_vextent(&args)))
2703                                 return error;
2704                 }
2705                 if (args.fsbno == NULLFSBLOCK && nullfb &&
2706                     args.minlen > ap->minlen) {
2707                         args.minlen = ap->minlen;
2708                         args.type = XFS_ALLOCTYPE_START_BNO;
2709                         args.fsbno = ap->rval;
2710                         if ((error = xfs_alloc_vextent(&args)))
2711                                 return error;
2712                 }
2713                 if (args.fsbno == NULLFSBLOCK && nullfb) {
2714                         args.fsbno = 0;
2715                         args.type = XFS_ALLOCTYPE_FIRST_AG;
2716                         args.total = ap->minlen;
2717                         args.minleft = 0;
2718                         if ((error = xfs_alloc_vextent(&args)))
2719                                 return error;
2720                         ap->low = 1;
2721                 }
2722                 if (args.fsbno != NULLFSBLOCK) {
2723                         ap->firstblock = ap->rval = args.fsbno;
2724                         ASSERT(nullfb || fb_agno == args.agno ||
2725                                (ap->low && fb_agno < args.agno));
2726                         ap->alen = args.len;
2727                         ap->ip->i_d.di_nblocks += args.len;
2728                         xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
2729                         if (ap->wasdel)
2730                                 ap->ip->i_delayed_blks -= args.len;
2731                         /*
2732                          * Adjust the disk quota also. This was reserved
2733                          * earlier.
2734                          */
2735                         XFS_TRANS_MOD_DQUOT_BYINO(mp, ap->tp, ap->ip,
2736                                 ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
2737                                                 XFS_TRANS_DQ_BCOUNT,
2738                                 (long) args.len);
2739                 } else {
2740                         ap->rval = NULLFSBLOCK;
2741                         ap->alen = 0;
2742                 }
2743         }
2744         return 0;
2745 #undef  ISVALID
2746 }
2747
2748 /*
2749  * Transform a btree format file with only one leaf node, where the
2750  * extents list will fit in the inode, into an extents format file.
2751  * Since the extent list is already in-core, all we have to do is
2752  * give up the space for the btree root and pitch the leaf block.
2753  */
2754 STATIC int                              /* error */
2755 xfs_bmap_btree_to_extents(
2756         xfs_trans_t             *tp,    /* transaction pointer */
2757         xfs_inode_t             *ip,    /* incore inode pointer */
2758         xfs_btree_cur_t         *cur,   /* btree cursor */
2759         int                     *logflagsp, /* inode logging flags */
2760         int                     whichfork)  /* data or attr fork */
2761 {
2762         /* REFERENCED */
2763         xfs_bmbt_block_t        *cblock;/* child btree block */
2764         xfs_fsblock_t           cbno;   /* child block number */
2765         xfs_buf_t               *cbp;   /* child block's buffer */
2766         int                     error;  /* error return value */
2767         xfs_ifork_t             *ifp;   /* inode fork data */
2768         xfs_mount_t             *mp;    /* mount point structure */
2769         xfs_bmbt_ptr_t          *pp;    /* ptr to block address */
2770         xfs_bmbt_block_t        *rblock;/* root btree block */
2771
2772         ifp = XFS_IFORK_PTR(ip, whichfork);
2773         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
2774         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
2775         rblock = ifp->if_broot;
2776         ASSERT(INT_GET(rblock->bb_level, ARCH_CONVERT) == 1);
2777         ASSERT(INT_GET(rblock->bb_numrecs, ARCH_CONVERT) == 1);
2778         ASSERT(XFS_BMAP_BROOT_MAXRECS(ifp->if_broot_bytes) == 1);
2779         mp = ip->i_mount;
2780         pp = XFS_BMAP_BROOT_PTR_ADDR(rblock, 1, ifp->if_broot_bytes);
2781         *logflagsp = 0;
2782 #ifdef DEBUG
2783         if ((error = xfs_btree_check_lptr(cur, INT_GET(*pp, ARCH_CONVERT), 1)))
2784                 return error;
2785 #endif
2786         cbno = INT_GET(*pp, ARCH_CONVERT);
2787         if ((error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp,
2788                         XFS_BMAP_BTREE_REF)))
2789                 return error;
2790         cblock = XFS_BUF_TO_BMBT_BLOCK(cbp);
2791         if ((error = xfs_btree_check_lblock(cur, cblock, 0, cbp)))
2792                 return error;
2793         xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
2794         ip->i_d.di_nblocks--;
2795         XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
2796         xfs_trans_binval(tp, cbp);
2797         if (cur->bc_bufs[0] == cbp)
2798                 cur->bc_bufs[0] = NULL;
2799         xfs_iroot_realloc(ip, -1, whichfork);
2800         ASSERT(ifp->if_broot == NULL);
2801         ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
2802         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
2803         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FEXT(whichfork);
2804         return 0;
2805 }
2806
2807 /*
2808  * Called by xfs_bmapi to update extent list structure and the btree
2809  * after removing space (or undoing a delayed allocation).
2810  */
2811 STATIC int                              /* error */
2812 xfs_bmap_del_extent(
2813         xfs_inode_t             *ip,    /* incore inode pointer */
2814         xfs_trans_t             *tp,    /* current transaction pointer */
2815         xfs_extnum_t            idx,    /* extent number to update/delete */
2816         xfs_bmap_free_t         *flist, /* list of extents to be freed */
2817         xfs_btree_cur_t         *cur,   /* if null, not a btree */
2818         xfs_bmbt_irec_t         *del,   /* data to remove from extent list */
2819         int                     *logflagsp, /* inode logging flags */
2820         int                     whichfork, /* data or attr fork */
2821         int                     rsvd)   /* OK to allocate reserved blocks */
2822 {
2823         xfs_filblks_t           da_new; /* new delay-alloc indirect blocks */
2824         xfs_filblks_t           da_old; /* old delay-alloc indirect blocks */
2825         xfs_fsblock_t           del_endblock=0; /* first block past del */
2826         xfs_fileoff_t           del_endoff;     /* first offset past del */
2827         int                     delay;  /* current block is delayed allocated */
2828         int                     do_fx;  /* free extent at end of routine */
2829         xfs_bmbt_rec_t          *ep;    /* current extent entry pointer */
2830         int                     error;  /* error return value */
2831         int                     flags;  /* inode logging flags */
2832 #ifdef XFS_BMAP_TRACE
2833         static char             fname[] = "xfs_bmap_del_extent";
2834 #endif
2835         xfs_bmbt_irec_t         got;    /* current extent entry */
2836         xfs_fileoff_t           got_endoff;     /* first offset past got */
2837         int                     i;      /* temp state */
2838         xfs_ifork_t             *ifp;   /* inode fork pointer */
2839         xfs_mount_t             *mp;    /* mount structure */
2840         xfs_filblks_t           nblks;  /* quota/sb block count */
2841         xfs_bmbt_irec_t         new;    /* new record to be inserted */
2842         /* REFERENCED */
2843         xfs_extnum_t            nextents;       /* number of extents in list */
2844         uint                    qfield; /* quota field to update */
2845         xfs_filblks_t           temp;   /* for indirect length calculations */
2846         xfs_filblks_t           temp2;  /* for indirect length calculations */
2847
2848         XFS_STATS_INC(xs_del_exlist);
2849         mp = ip->i_mount;
2850         ifp = XFS_IFORK_PTR(ip, whichfork);
2851         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
2852         ASSERT(idx >= 0 && idx < nextents);
2853         ASSERT(del->br_blockcount > 0);
2854         ep = &ifp->if_u1.if_extents[idx];
2855         xfs_bmbt_get_all(ep, &got);
2856         ASSERT(got.br_startoff <= del->br_startoff);
2857         del_endoff = del->br_startoff + del->br_blockcount;
2858         got_endoff = got.br_startoff + got.br_blockcount;
2859         ASSERT(got_endoff >= del_endoff);
2860         delay = ISNULLSTARTBLOCK(got.br_startblock);
2861         ASSERT(ISNULLSTARTBLOCK(del->br_startblock) == delay);
2862         flags = 0;
2863         qfield = 0;
2864         error = 0;
2865         /*
2866          * If deleting a real allocation, must free up the disk space.
2867          */
2868         if (!delay) {
2869                 flags = XFS_ILOG_CORE;
2870                 /*
2871                  * Realtime allocation.  Free it and record di_nblocks update.
2872                  */
2873                 if (whichfork == XFS_DATA_FORK &&
2874                     (ip->i_d.di_flags & XFS_DIFLAG_REALTIME)) {
2875                         xfs_fsblock_t   bno;
2876                         xfs_filblks_t   len;
2877
2878                         ASSERT(do_mod(del->br_blockcount,
2879                                       mp->m_sb.sb_rextsize) == 0);
2880                         ASSERT(do_mod(del->br_startblock,
2881                                       mp->m_sb.sb_rextsize) == 0);
2882                         bno = del->br_startblock;
2883                         len = del->br_blockcount;
2884                         do_div(bno, mp->m_sb.sb_rextsize);
2885                         do_div(len, mp->m_sb.sb_rextsize);
2886                         if ((error = xfs_rtfree_extent(ip->i_transp, bno,
2887                                         (xfs_extlen_t)len)))
2888                                 goto done;
2889                         do_fx = 0;
2890                         nblks = len * mp->m_sb.sb_rextsize;
2891                         qfield = XFS_TRANS_DQ_RTBCOUNT;
2892                 }
2893                 /*
2894                  * Ordinary allocation.
2895                  */
2896                 else {
2897                         do_fx = 1;
2898                         nblks = del->br_blockcount;
2899                         qfield = XFS_TRANS_DQ_BCOUNT;
2900                 }
2901                 /*
2902                  * Set up del_endblock and cur for later.
2903                  */
2904                 del_endblock = del->br_startblock + del->br_blockcount;
2905                 if (cur) {
2906                         if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
2907                                         got.br_startblock, got.br_blockcount,
2908                                         &i)))
2909                                 goto done;
2910                         ASSERT(i == 1);
2911                 }
2912                 da_old = da_new = 0;
2913         } else {
2914                 da_old = STARTBLOCKVAL(got.br_startblock);
2915                 da_new = 0;
2916                 nblks = 0;
2917                 do_fx = 0;
2918         }
2919         /*
2920          * Set flag value to use in switch statement.
2921          * Left-contig is 2, right-contig is 1.
2922          */
2923         switch (((got.br_startoff == del->br_startoff) << 1) |
2924                 (got_endoff == del_endoff)) {
2925         case 3:
2926                 /*
2927                  * Matches the whole extent.  Delete the entry.
2928                  */
2929                 xfs_bmap_trace_delete(fname, "3", ip, idx, 1, whichfork);
2930                 xfs_bmap_delete_exlist(ip, idx, 1, whichfork);
2931                 ifp->if_lastex = idx;
2932                 if (delay)
2933                         break;
2934                 XFS_IFORK_NEXT_SET(ip, whichfork,
2935                         XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
2936                 flags |= XFS_ILOG_CORE;
2937                 if (!cur) {
2938                         flags |= XFS_ILOG_FEXT(whichfork);
2939                         break;
2940                 }
2941                 if ((error = xfs_bmbt_delete(cur, &i)))
2942                         goto done;
2943                 ASSERT(i == 1);
2944                 break;
2945
2946         case 2:
2947                 /*
2948                  * Deleting the first part of the extent.
2949                  */
2950                 xfs_bmap_trace_pre_update(fname, "2", ip, idx, whichfork);
2951                 xfs_bmbt_set_startoff(ep, del_endoff);
2952                 temp = got.br_blockcount - del->br_blockcount;
2953                 xfs_bmbt_set_blockcount(ep, temp);
2954                 ifp->if_lastex = idx;
2955                 if (delay) {
2956                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2957                                 da_old);
2958                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
2959                         xfs_bmap_trace_post_update(fname, "2", ip, idx,
2960                                 whichfork);
2961                         da_new = temp;
2962                         break;
2963                 }
2964                 xfs_bmbt_set_startblock(ep, del_endblock);
2965                 xfs_bmap_trace_post_update(fname, "2", ip, idx, whichfork);
2966                 if (!cur) {
2967                         flags |= XFS_ILOG_FEXT(whichfork);
2968                         break;
2969                 }
2970                 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
2971                                 got.br_blockcount - del->br_blockcount,
2972                                 got.br_state)))
2973                         goto done;
2974                 break;
2975
2976         case 1:
2977                 /*
2978                  * Deleting the last part of the extent.
2979                  */
2980                 temp = got.br_blockcount - del->br_blockcount;
2981                 xfs_bmap_trace_pre_update(fname, "1", ip, idx, whichfork);
2982                 xfs_bmbt_set_blockcount(ep, temp);
2983                 ifp->if_lastex = idx;
2984                 if (delay) {
2985                         temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
2986                                 da_old);
2987                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
2988                         xfs_bmap_trace_post_update(fname, "1", ip, idx,
2989                                 whichfork);
2990                         da_new = temp;
2991                         break;
2992                 }
2993                 xfs_bmap_trace_post_update(fname, "1", ip, idx, whichfork);
2994                 if (!cur) {
2995                         flags |= XFS_ILOG_FEXT(whichfork);
2996                         break;
2997                 }
2998                 if ((error = xfs_bmbt_update(cur, got.br_startoff,
2999                                 got.br_startblock,
3000                                 got.br_blockcount - del->br_blockcount,
3001                                 got.br_state)))
3002                         goto done;
3003                 break;
3004
3005         case 0:
3006                 /*
3007                  * Deleting the middle of the extent.
3008                  */
3009                 temp = del->br_startoff - got.br_startoff;
3010                 xfs_bmap_trace_pre_update(fname, "0", ip, idx, whichfork);
3011                 xfs_bmbt_set_blockcount(ep, temp);
3012                 new.br_startoff = del_endoff;
3013                 temp2 = got_endoff - del_endoff;
3014                 new.br_blockcount = temp2;
3015                 new.br_state = got.br_state;
3016                 if (!delay) {
3017                         new.br_startblock = del_endblock;
3018                         flags |= XFS_ILOG_CORE;
3019                         if (cur) {
3020                                 if ((error = xfs_bmbt_update(cur,
3021                                                 got.br_startoff,
3022                                                 got.br_startblock, temp,
3023                                                 got.br_state)))
3024                                         goto done;
3025                                 if ((error = xfs_bmbt_increment(cur, 0, &i)))
3026                                         goto done;
3027                                 cur->bc_rec.b = new;
3028                                 error = xfs_bmbt_insert(cur, &i);
3029                                 if (error && error != ENOSPC)
3030                                         goto done;
3031                                 /*
3032                                  * If get no-space back from btree insert,
3033                                  * it tried a split, and we have a zero
3034                                  * block reservation.
3035                                  * Fix up our state and return the error.
3036                                  */
3037                                 if (error == ENOSPC) {
3038                                         /*
3039                                          * Reset the cursor, don't trust
3040                                          * it after any insert operation.
3041                                          */
3042                                         if ((error = xfs_bmbt_lookup_eq(cur,
3043                                                         got.br_startoff,
3044                                                         got.br_startblock,
3045                                                         temp, &i)))
3046                                                 goto done;
3047                                         ASSERT(i == 1);
3048                                         /*
3049                                          * Update the btree record back
3050                                          * to the original value.
3051                                          */
3052                                         if ((error = xfs_bmbt_update(cur,
3053                                                         got.br_startoff,
3054                                                         got.br_startblock,
3055                                                         got.br_blockcount,
3056                                                         got.br_state)))
3057                                                 goto done;
3058                                         /*
3059                                          * Reset the extent record back
3060                                          * to the original value.
3061                                          */
3062                                         xfs_bmbt_set_blockcount(ep,
3063                                                 got.br_blockcount);
3064                                         flags = 0;
3065                                         error = XFS_ERROR(ENOSPC);
3066                                         goto done;
3067                                 }
3068                                 ASSERT(i == 1);
3069                         } else
3070                                 flags |= XFS_ILOG_FEXT(whichfork);
3071                         XFS_IFORK_NEXT_SET(ip, whichfork,
3072                                 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
3073                 } else {
3074                         ASSERT(whichfork == XFS_DATA_FORK);
3075                         temp = xfs_bmap_worst_indlen(ip, temp);
3076                         xfs_bmbt_set_startblock(ep, NULLSTARTBLOCK((int)temp));
3077                         temp2 = xfs_bmap_worst_indlen(ip, temp2);
3078                         new.br_startblock = NULLSTARTBLOCK((int)temp2);
3079                         da_new = temp + temp2;
3080                         while (da_new > da_old) {
3081                                 if (temp) {
3082                                         temp--;
3083                                         da_new--;
3084                                         xfs_bmbt_set_startblock(ep,
3085                                                 NULLSTARTBLOCK((int)temp));
3086                                 }
3087                                 if (da_new == da_old)
3088                                         break;
3089                                 if (temp2) {
3090                                         temp2--;
3091                                         da_new--;
3092                                         new.br_startblock =
3093                                                 NULLSTARTBLOCK((int)temp2);
3094                                 }
3095                         }
3096                 }
3097                 xfs_bmap_trace_post_update(fname, "0", ip, idx, whichfork);
3098                 xfs_bmap_trace_insert(fname, "0", ip, idx + 1, 1, &new, NULL,
3099                         whichfork);
3100                 xfs_bmap_insert_exlist(ip, idx + 1, 1, &new, whichfork);
3101                 ifp->if_lastex = idx + 1;
3102                 break;
3103         }
3104         /*
3105          * If we need to, add to list of extents to delete.
3106          */
3107         if (do_fx)
3108                 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
3109                         mp);
3110         /*
3111          * Adjust inode # blocks in the file.
3112          */
3113         if (nblks)
3114                 ip->i_d.di_nblocks -= nblks;
3115         /*
3116          * Adjust quota data.
3117          */
3118         if (qfield)
3119                 XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, qfield, (long)-nblks);
3120
3121         /*
3122          * Account for change in delayed indirect blocks.
3123          * Nothing to do for disk quota accounting here.
3124          */
3125         ASSERT(da_old >= da_new);
3126         if (da_old > da_new)
3127                 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS, (int)(da_old - da_new),
3128                         rsvd);
3129 done:
3130         *logflagsp = flags;
3131         return error;
3132 }
3133
3134 /*
3135  * Remove the entry "free" from the free item list.  Prev points to the
3136  * previous entry, unless "free" is the head of the list.
3137  */
3138 STATIC void
3139 xfs_bmap_del_free(
3140         xfs_bmap_free_t         *flist, /* free item list header */
3141         xfs_bmap_free_item_t    *prev,  /* previous item on list, if any */
3142         xfs_bmap_free_item_t    *free)  /* list item to be freed */
3143 {
3144         if (prev)
3145                 prev->xbfi_next = free->xbfi_next;
3146         else
3147                 flist->xbf_first = free->xbfi_next;
3148         flist->xbf_count--;
3149         kmem_zone_free(xfs_bmap_free_item_zone, free);
3150 }
3151
3152 /*
3153  * Remove count entries from the extents array for inode "ip", starting
3154  * at index "idx".  Copies the remaining items down over the deleted ones,
3155  * and gives back the excess memory.
3156  */
3157 STATIC void
3158 xfs_bmap_delete_exlist(
3159         xfs_inode_t     *ip,            /* incore inode pointer */
3160         xfs_extnum_t    idx,            /* starting delete index */
3161         xfs_extnum_t    count,          /* count of items to delete */
3162         int             whichfork)      /* data or attr fork */
3163 {
3164         xfs_bmbt_rec_t  *base;          /* base of extent list */
3165         xfs_ifork_t     *ifp;           /* inode fork pointer */
3166         xfs_extnum_t    nextents;       /* number of extents in list after */
3167
3168         ifp = XFS_IFORK_PTR(ip, whichfork);
3169         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3170         base = ifp->if_u1.if_extents;
3171         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - count;
3172         memmove(&base[idx], &base[idx + count],
3173                 (nextents - idx) * sizeof(*base));
3174         xfs_iext_realloc(ip, -count, whichfork);
3175 }
3176
3177 /*
3178  * Convert an extents-format file into a btree-format file.
3179  * The new file will have a root block (in the inode) and a single child block.
3180  */
3181 STATIC int                                      /* error */
3182 xfs_bmap_extents_to_btree(
3183         xfs_trans_t             *tp,            /* transaction pointer */
3184         xfs_inode_t             *ip,            /* incore inode pointer */
3185         xfs_fsblock_t           *firstblock,    /* first-block-allocated */
3186         xfs_bmap_free_t         *flist,         /* blocks freed in xaction */
3187         xfs_btree_cur_t         **curp,         /* cursor returned to caller */
3188         int                     wasdel,         /* converting a delayed alloc */
3189         int                     *logflagsp,     /* inode logging flags */
3190         int                     whichfork)      /* data or attr fork */
3191 {
3192         xfs_bmbt_block_t        *ablock;        /* allocated (child) bt block */
3193         xfs_buf_t               *abp;           /* buffer for ablock */
3194         xfs_alloc_arg_t         args;           /* allocation arguments */
3195         xfs_bmbt_rec_t          *arp;           /* child record pointer */
3196         xfs_bmbt_block_t        *block;         /* btree root block */
3197         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
3198         xfs_bmbt_rec_t          *ep;            /* extent list pointer */
3199         int                     error;          /* error return value */
3200         xfs_extnum_t            i, cnt;         /* extent list index */
3201         xfs_ifork_t             *ifp;           /* inode fork pointer */
3202         xfs_bmbt_key_t          *kp;            /* root block key pointer */
3203         xfs_mount_t             *mp;            /* mount structure */
3204         xfs_extnum_t            nextents;       /* extent list size */
3205         xfs_bmbt_ptr_t          *pp;            /* root block address pointer */
3206
3207         ifp = XFS_IFORK_PTR(ip, whichfork);
3208         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
3209         ASSERT(ifp->if_ext_max ==
3210                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
3211         /*
3212          * Make space in the inode incore.
3213          */
3214         xfs_iroot_realloc(ip, 1, whichfork);
3215         ifp->if_flags |= XFS_IFBROOT;
3216         /*
3217          * Fill in the root.
3218          */
3219         block = ifp->if_broot;
3220         INT_SET(block->bb_magic, ARCH_CONVERT, XFS_BMAP_MAGIC);
3221         INT_SET(block->bb_level, ARCH_CONVERT, 1);
3222         INT_SET(block->bb_numrecs, ARCH_CONVERT, 1);
3223         INT_SET(block->bb_leftsib, ARCH_CONVERT, NULLDFSBNO);
3224         INT_SET(block->bb_rightsib, ARCH_CONVERT, NULLDFSBNO);
3225         /*
3226          * Need a cursor.  Can't allocate until bb_level is filled in.
3227          */
3228         mp = ip->i_mount;
3229         cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
3230                 whichfork);
3231         cur->bc_private.b.firstblock = *firstblock;
3232         cur->bc_private.b.flist = flist;
3233         cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
3234         /*
3235          * Convert to a btree with two levels, one record in root.
3236          */
3237         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
3238         args.tp = tp;
3239         args.mp = mp;
3240         if (*firstblock == NULLFSBLOCK) {
3241                 args.type = XFS_ALLOCTYPE_START_BNO;
3242                 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
3243         } else if (flist->xbf_low) {
3244                 args.type = XFS_ALLOCTYPE_START_BNO;
3245                 args.fsbno = *firstblock;
3246         } else {
3247                 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3248                 args.fsbno = *firstblock;
3249         }
3250         args.minlen = args.maxlen = args.prod = 1;
3251         args.total = args.minleft = args.alignment = args.mod = args.isfl =
3252                 args.minalignslop = 0;
3253         args.wasdel = wasdel;
3254         *logflagsp = 0;
3255         if ((error = xfs_alloc_vextent(&args))) {
3256                 xfs_iroot_realloc(ip, -1, whichfork);
3257                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
3258                 return error;
3259         }
3260         /*
3261          * Allocation can't fail, the space was reserved.
3262          */
3263         ASSERT(args.fsbno != NULLFSBLOCK);
3264         ASSERT(*firstblock == NULLFSBLOCK ||
3265                args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
3266                (flist->xbf_low &&
3267                 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
3268         *firstblock = cur->bc_private.b.firstblock = args.fsbno;
3269         cur->bc_private.b.allocated++;
3270         ip->i_d.di_nblocks++;
3271         XFS_TRANS_MOD_DQUOT_BYINO(mp, tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
3272         abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
3273         /*
3274          * Fill in the child block.
3275          */
3276         ablock = XFS_BUF_TO_BMBT_BLOCK(abp);
3277         INT_SET(ablock->bb_magic, ARCH_CONVERT, XFS_BMAP_MAGIC);
3278         ablock->bb_level = 0;
3279         INT_SET(ablock->bb_leftsib, ARCH_CONVERT, NULLDFSBNO);
3280         INT_SET(ablock->bb_rightsib, ARCH_CONVERT, NULLDFSBNO);
3281         arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3282         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3283         for (ep = ifp->if_u1.if_extents, cnt = i = 0; i < nextents; i++, ep++) {
3284                 if (!ISNULLSTARTBLOCK(xfs_bmbt_get_startblock(ep))) {
3285                         arp->l0 = INT_GET(ep->l0, ARCH_CONVERT);
3286                         arp->l1 = INT_GET(ep->l1, ARCH_CONVERT);
3287                         arp++; cnt++;
3288                 }
3289         }
3290         INT_SET(ablock->bb_numrecs, ARCH_CONVERT, cnt);
3291         ASSERT(INT_GET(ablock->bb_numrecs, ARCH_CONVERT) == XFS_IFORK_NEXTENTS(ip, whichfork));
3292         /*
3293          * Fill in the root key and pointer.
3294          */
3295         kp = XFS_BMAP_KEY_IADDR(block, 1, cur);
3296         arp = XFS_BMAP_REC_IADDR(ablock, 1, cur);
3297         INT_SET(kp->br_startoff, ARCH_CONVERT, xfs_bmbt_disk_get_startoff(arp));
3298         pp = XFS_BMAP_PTR_IADDR(block, 1, cur);
3299         INT_SET(*pp, ARCH_CONVERT, args.fsbno);
3300         /*
3301          * Do all this logging at the end so that
3302          * the root is at the right level.
3303          */
3304         xfs_bmbt_log_block(cur, abp, XFS_BB_ALL_BITS);
3305         xfs_bmbt_log_recs(cur, abp, 1, INT_GET(ablock->bb_numrecs, ARCH_CONVERT));
3306         ASSERT(*curp == NULL);
3307         *curp = cur;
3308         *logflagsp = XFS_ILOG_CORE | XFS_ILOG_FBROOT(whichfork);
3309         return 0;
3310 }
3311
3312 /*
3313  * Insert new item(s) in the extent list for inode "ip".
3314  * Count new items are inserted at offset idx.
3315  */
3316 STATIC void
3317 xfs_bmap_insert_exlist(
3318         xfs_inode_t     *ip,            /* incore inode pointer */
3319         xfs_extnum_t    idx,            /* starting index of new items */
3320         xfs_extnum_t    count,          /* number of inserted items */
3321         xfs_bmbt_irec_t *new,           /* items to insert */
3322         int             whichfork)      /* data or attr fork */
3323 {
3324         xfs_bmbt_rec_t  *base;          /* extent list base */
3325         xfs_ifork_t     *ifp;           /* inode fork pointer */
3326         xfs_extnum_t    nextents;       /* extent list size */
3327         xfs_extnum_t    to;             /* extent list index */
3328
3329         ifp = XFS_IFORK_PTR(ip, whichfork);
3330         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
3331         xfs_iext_realloc(ip, count, whichfork);
3332         base = ifp->if_u1.if_extents;
3333         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3334         memmove(&base[idx + count], &base[idx],
3335                 (nextents - (idx + count)) * sizeof(*base));
3336         for (to = idx; to < idx + count; to++, new++)
3337                 xfs_bmbt_set_all(&base[to], new);
3338 }
3339
3340 /*
3341  * Helper routine to reset inode di_forkoff field when switching
3342  * attribute fork from local to extent format - we reset it where
3343  * possible to make space available for inline data fork extents.
3344  */
3345 STATIC void
3346 xfs_bmap_forkoff_reset(
3347         xfs_mount_t     *mp,
3348         xfs_inode_t     *ip,
3349         int             whichfork)
3350 {
3351         if (whichfork == XFS_ATTR_FORK &&
3352             (ip->i_d.di_format != XFS_DINODE_FMT_DEV) &&
3353             (ip->i_d.di_format != XFS_DINODE_FMT_UUID) &&
3354             ((mp->m_attroffset >> 3) > ip->i_d.di_forkoff)) {
3355                 ip->i_d.di_forkoff = mp->m_attroffset >> 3;
3356                 ip->i_df.if_ext_max = XFS_IFORK_DSIZE(ip) /
3357                                         (uint)sizeof(xfs_bmbt_rec_t);
3358                 ip->i_afp->if_ext_max = XFS_IFORK_ASIZE(ip) /
3359                                         (uint)sizeof(xfs_bmbt_rec_t);
3360         }
3361 }
3362
3363 /*
3364  * Convert a local file to an extents file.
3365  * This code is out of bounds for data forks of regular files,
3366  * since the file data needs to get logged so things will stay consistent.
3367  * (The bmap-level manipulations are ok, though).
3368  */
3369 STATIC int                              /* error */
3370 xfs_bmap_local_to_extents(
3371         xfs_trans_t     *tp,            /* transaction pointer */
3372         xfs_inode_t     *ip,            /* incore inode pointer */
3373         xfs_fsblock_t   *firstblock,    /* first block allocated in xaction */
3374         xfs_extlen_t    total,          /* total blocks needed by transaction */
3375         int             *logflagsp,     /* inode logging flags */
3376         int             whichfork)      /* data or attr fork */
3377 {
3378         int             error;          /* error return value */
3379         int             flags;          /* logging flags returned */
3380 #ifdef XFS_BMAP_TRACE
3381         static char     fname[] = "xfs_bmap_local_to_extents";
3382 #endif
3383         xfs_ifork_t     *ifp;           /* inode fork pointer */
3384
3385         /*
3386          * We don't want to deal with the case of keeping inode data inline yet.
3387          * So sending the data fork of a regular inode is invalid.
3388          */
3389         ASSERT(!((ip->i_d.di_mode & S_IFMT) == S_IFREG &&
3390                  whichfork == XFS_DATA_FORK));
3391         ifp = XFS_IFORK_PTR(ip, whichfork);
3392         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
3393         flags = 0;
3394         error = 0;
3395         if (ifp->if_bytes) {
3396                 xfs_alloc_arg_t args;   /* allocation arguments */
3397                 xfs_buf_t       *bp;    /* buffer for extent list block */
3398                 xfs_bmbt_rec_t  *ep;    /* extent list pointer */
3399
3400                 args.tp = tp;
3401                 args.mp = ip->i_mount;
3402                 ASSERT(ifp->if_flags & XFS_IFINLINE);
3403                 /*
3404                  * Allocate a block.  We know we need only one, since the
3405                  * file currently fits in an inode.
3406                  */
3407                 if (*firstblock == NULLFSBLOCK) {
3408                         args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
3409                         args.type = XFS_ALLOCTYPE_START_BNO;
3410                 } else {
3411                         args.fsbno = *firstblock;
3412                         args.type = XFS_ALLOCTYPE_NEAR_BNO;
3413                 }
3414                 args.total = total;
3415                 args.mod = args.minleft = args.alignment = args.wasdel =
3416                         args.isfl = args.minalignslop = 0;
3417                 args.minlen = args.maxlen = args.prod = 1;
3418                 if ((error = xfs_alloc_vextent(&args)))
3419                         goto done;
3420                 /*
3421                  * Can't fail, the space was reserved.
3422                  */
3423                 ASSERT(args.fsbno != NULLFSBLOCK);
3424                 ASSERT(args.len == 1);
3425                 *firstblock = args.fsbno;
3426                 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
3427                 memcpy((char *)XFS_BUF_PTR(bp), ifp->if_u1.if_data,
3428                         ifp->if_bytes);
3429                 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
3430                 xfs_bmap_forkoff_reset(args.mp, ip, whichfork);
3431                 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
3432                 xfs_iext_realloc(ip, 1, whichfork);
3433                 ep = ifp->if_u1.if_extents;
3434                 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
3435                 xfs_bmap_trace_post_update(fname, "new", ip, 0, whichfork);
3436                 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
3437                 ip->i_d.di_nblocks = 1;
3438                 XFS_TRANS_MOD_DQUOT_BYINO(args.mp, tp, ip,
3439                         XFS_TRANS_DQ_BCOUNT, 1L);
3440                 flags |= XFS_ILOG_FEXT(whichfork);
3441         } else {
3442                 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
3443                 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
3444         }
3445         ifp->if_flags &= ~XFS_IFINLINE;
3446         ifp->if_flags |= XFS_IFEXTENTS;
3447         XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
3448         flags |= XFS_ILOG_CORE;
3449 done:
3450         *logflagsp = flags;
3451         return error;
3452 }
3453
3454 xfs_bmbt_rec_t *                        /* pointer to found extent entry */
3455 xfs_bmap_do_search_extents(
3456         xfs_bmbt_rec_t  *base,          /* base of extent list */
3457         xfs_extnum_t    lastx,          /* last extent index used */
3458         xfs_extnum_t    nextents,       /* extent list size */
3459         xfs_fileoff_t   bno,            /* block number searched for */
3460         int             *eofp,          /* out: end of file found */
3461         xfs_extnum_t    *lastxp,        /* out: last extent index */
3462         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3463         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3464 {
3465         xfs_bmbt_rec_t  *ep;            /* extent list entry pointer */
3466         xfs_bmbt_irec_t got;            /* extent list entry, decoded */
3467         int             high;           /* high index of binary search */
3468         int             low;            /* low index of binary search */
3469
3470         /*
3471          * Initialize the extent entry structure to catch access to
3472          * uninitialized br_startblock field.
3473          */
3474         got.br_startoff = 0xffa5a5a5a5a5a5a5LL;
3475         got.br_blockcount = 0xa55a5a5a5a5a5a5aLL;
3476         got.br_state = XFS_EXT_INVALID;
3477
3478 #if XFS_BIG_BLKNOS
3479         got.br_startblock = 0xffffa5a5a5a5a5a5LL;
3480 #else
3481         got.br_startblock = 0xffffa5a5;
3482 #endif
3483
3484         if (lastx != NULLEXTNUM && lastx < nextents)
3485                 ep = base + lastx;
3486         else
3487                 ep = NULL;
3488         prevp->br_startoff = NULLFILEOFF;
3489         if (ep && bno >= (got.br_startoff = xfs_bmbt_get_startoff(ep)) &&
3490             bno < got.br_startoff +
3491                   (got.br_blockcount = xfs_bmbt_get_blockcount(ep)))
3492                 *eofp = 0;
3493         else if (ep && lastx < nextents - 1 &&
3494                  bno >= (got.br_startoff = xfs_bmbt_get_startoff(ep + 1)) &&
3495                  bno < got.br_startoff +
3496                        (got.br_blockcount = xfs_bmbt_get_blockcount(ep + 1))) {
3497                 lastx++;
3498                 ep++;
3499                 *eofp = 0;
3500         } else if (nextents == 0)
3501                 *eofp = 1;
3502         else if (bno == 0 &&
3503                  (got.br_startoff = xfs_bmbt_get_startoff(base)) == 0) {
3504                 ep = base;
3505                 lastx = 0;
3506                 got.br_blockcount = xfs_bmbt_get_blockcount(ep);
3507                 *eofp = 0;
3508         } else {
3509                 /* binary search the extents array */
3510                 low = 0;
3511                 high = nextents - 1;
3512                 while (low <= high) {
3513                         XFS_STATS_INC(xs_cmp_exlist);
3514                         lastx = (low + high) >> 1;
3515                         ep = base + lastx;
3516                         got.br_startoff = xfs_bmbt_get_startoff(ep);
3517                         got.br_blockcount = xfs_bmbt_get_blockcount(ep);
3518                         if (bno < got.br_startoff)
3519                                 high = lastx - 1;
3520                         else if (bno >= got.br_startoff + got.br_blockcount)
3521                                 low = lastx + 1;
3522                         else {
3523                                 got.br_startblock = xfs_bmbt_get_startblock(ep);
3524                                 got.br_state = xfs_bmbt_get_state(ep);
3525                                 *eofp = 0;
3526                                 *lastxp = lastx;
3527                                 *gotp = got;
3528                                 return ep;
3529                         }
3530                 }
3531                 if (bno >= got.br_startoff + got.br_blockcount) {
3532                         lastx++;
3533                         if (lastx == nextents) {
3534                                 *eofp = 1;
3535                                 got.br_startblock = xfs_bmbt_get_startblock(ep);
3536                                 got.br_state = xfs_bmbt_get_state(ep);
3537                                 *prevp = got;
3538                                 ep = NULL;
3539                         } else {
3540                                 *eofp = 0;
3541                                 xfs_bmbt_get_all(ep, prevp);
3542                                 ep++;
3543                                 got.br_startoff = xfs_bmbt_get_startoff(ep);
3544                                 got.br_blockcount = xfs_bmbt_get_blockcount(ep);
3545                         }
3546                 } else {
3547                         *eofp = 0;
3548                         if (ep > base)
3549                                 xfs_bmbt_get_all(ep - 1, prevp);
3550                 }
3551         }
3552         if (ep) {
3553                 got.br_startblock = xfs_bmbt_get_startblock(ep);
3554                 got.br_state = xfs_bmbt_get_state(ep);
3555         }
3556         *lastxp = lastx;
3557         *gotp = got;
3558         return ep;
3559 }
3560
3561 /*
3562  * Search the extents list for the inode, for the extent containing bno.
3563  * If bno lies in a hole, point to the next entry.  If bno lies past eof,
3564  * *eofp will be set, and *prevp will contain the last entry (null if none).
3565  * Else, *lastxp will be set to the index of the found
3566  * entry; *gotp will contain the entry.
3567  */
3568 STATIC xfs_bmbt_rec_t *                 /* pointer to found extent entry */
3569 xfs_bmap_search_extents(
3570         xfs_inode_t     *ip,            /* incore inode pointer */
3571         xfs_fileoff_t   bno,            /* block number searched for */
3572         int             whichfork,      /* data or attr fork */
3573         int             *eofp,          /* out: end of file found */
3574         xfs_extnum_t    *lastxp,        /* out: last extent index */
3575         xfs_bmbt_irec_t *gotp,          /* out: extent entry found */
3576         xfs_bmbt_irec_t *prevp)         /* out: previous extent entry found */
3577 {
3578         xfs_ifork_t     *ifp;           /* inode fork pointer */
3579         xfs_bmbt_rec_t  *base;          /* base of extent list */
3580         xfs_extnum_t    lastx;          /* last extent index used */
3581         xfs_extnum_t    nextents;       /* extent list size */
3582         xfs_bmbt_rec_t  *ep;            /* extent list entry pointer */
3583         int             rt;             /* realtime flag    */
3584
3585         XFS_STATS_INC(xs_look_exlist);
3586         ifp = XFS_IFORK_PTR(ip, whichfork);
3587         lastx = ifp->if_lastex;
3588         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
3589         base = &ifp->if_u1.if_extents[0];
3590
3591         ep = xfs_bmap_do_search_extents(base, lastx, nextents, bno, eofp,
3592                                           lastxp, gotp, prevp);
3593         rt = ip->i_d.di_flags & XFS_DIFLAG_REALTIME;
3594         if(!rt && !gotp->br_startblock && (*lastxp != NULLEXTNUM)) {
3595                 cmn_err(CE_PANIC,"Access to block zero: fs: <%s> inode: %lld "
3596                         "start_block : %llx start_off : %llx blkcnt : %llx "
3597                         "extent-state : %x \n",
3598                         (ip->i_mount)->m_fsname,(long long)ip->i_ino,
3599                         gotp->br_startblock, gotp->br_startoff,
3600                         gotp->br_blockcount,gotp->br_state);
3601         }
3602         return ep;
3603 }
3604
3605
3606 #ifdef XFS_BMAP_TRACE
3607 ktrace_t        *xfs_bmap_trace_buf;
3608
3609 /*
3610  * Add a bmap trace buffer entry.  Base routine for the others.
3611  */
3612 STATIC void
3613 xfs_bmap_trace_addentry(
3614         int             opcode,         /* operation */
3615         char            *fname,         /* function name */
3616         char            *desc,          /* operation description */
3617         xfs_inode_t     *ip,            /* incore inode pointer */
3618         xfs_extnum_t    idx,            /* index of entry(ies) */
3619         xfs_extnum_t    cnt,            /* count of entries, 1 or 2 */
3620         xfs_bmbt_rec_t  *r1,            /* first record */
3621         xfs_bmbt_rec_t  *r2,            /* second record or null */
3622         int             whichfork)      /* data or attr fork */
3623 {
3624         xfs_bmbt_rec_t  tr2;
3625
3626         ASSERT(cnt == 1 || cnt == 2);
3627         ASSERT(r1 != NULL);
3628         if (cnt == 1) {
3629                 ASSERT(r2 == NULL);
3630                 r2 = &tr2;
3631                 memset(&tr2, 0, sizeof(tr2));
3632         } else
3633                 ASSERT(r2 != NULL);
3634         ktrace_enter(xfs_bmap_trace_buf,
3635                 (void *)(__psint_t)(opcode | (whichfork << 16)),
3636                 (void *)fname, (void *)desc, (void *)ip,
3637                 (void *)(__psint_t)idx,
3638                 (void *)(__psint_t)cnt,
3639                 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3640                 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3641                 (void *)(__psunsigned_t)(r1->l0 >> 32),
3642                 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3643                 (void *)(__psunsigned_t)(r1->l1 >> 32),
3644                 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3645                 (void *)(__psunsigned_t)(r2->l0 >> 32),
3646                 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3647                 (void *)(__psunsigned_t)(r2->l1 >> 32),
3648                 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3649                 );
3650         ASSERT(ip->i_xtrace);
3651         ktrace_enter(ip->i_xtrace,
3652                 (void *)(__psint_t)(opcode | (whichfork << 16)),
3653                 (void *)fname, (void *)desc, (void *)ip,
3654                 (void *)(__psint_t)idx,
3655                 (void *)(__psint_t)cnt,
3656                 (void *)(__psunsigned_t)(ip->i_ino >> 32),
3657                 (void *)(__psunsigned_t)(unsigned)ip->i_ino,
3658                 (void *)(__psunsigned_t)(r1->l0 >> 32),
3659                 (void *)(__psunsigned_t)(unsigned)(r1->l0),
3660                 (void *)(__psunsigned_t)(r1->l1 >> 32),
3661                 (void *)(__psunsigned_t)(unsigned)(r1->l1),
3662                 (void *)(__psunsigned_t)(r2->l0 >> 32),
3663                 (void *)(__psunsigned_t)(unsigned)(r2->l0),
3664                 (void *)(__psunsigned_t)(r2->l1 >> 32),
3665                 (void *)(__psunsigned_t)(unsigned)(r2->l1)
3666                 );
3667 }
3668
3669 /*
3670  * Add bmap trace entry prior to a call to xfs_bmap_delete_exlist.
3671  */
3672 STATIC void
3673 xfs_bmap_trace_delete(
3674         char            *fname,         /* function name */
3675         char            *desc,          /* operation description */
3676         xfs_inode_t     *ip,            /* incore inode pointer */
3677         xfs_extnum_t    idx,            /* index of entry(entries) deleted */
3678         xfs_extnum_t    cnt,            /* count of entries deleted, 1 or 2 */
3679         int             whichfork)      /* data or attr fork */
3680 {
3681         xfs_ifork_t     *ifp;           /* inode fork pointer */
3682
3683         ifp = XFS_IFORK_PTR(ip, whichfork);
3684         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_DELETE, fname, desc, ip, idx,
3685                 cnt, &ifp->if_u1.if_extents[idx],
3686                 cnt == 2 ? &ifp->if_u1.if_extents[idx + 1] : NULL,
3687                 whichfork);
3688 }
3689
3690 /*
3691  * Add bmap trace entry prior to a call to xfs_bmap_insert_exlist, or
3692  * reading in the extents list from the disk (in the btree).
3693  */
3694 STATIC void
3695 xfs_bmap_trace_insert(
3696         char            *fname,         /* function name */
3697         char            *desc,          /* operation description */
3698         xfs_inode_t     *ip,            /* incore inode pointer */
3699         xfs_extnum_t    idx,            /* index of entry(entries) inserted */
3700         xfs_extnum_t    cnt,            /* count of entries inserted, 1 or 2 */
3701         xfs_bmbt_irec_t *r1,            /* inserted record 1 */
3702         xfs_bmbt_irec_t *r2,            /* inserted record 2 or null */
3703         int             whichfork)      /* data or attr fork */
3704 {
3705         xfs_bmbt_rec_t  tr1;            /* compressed record 1 */
3706         xfs_bmbt_rec_t  tr2;            /* compressed record 2 if needed */
3707
3708         xfs_bmbt_set_all(&tr1, r1);
3709         if (cnt == 2) {
3710                 ASSERT(r2 != NULL);
3711                 xfs_bmbt_set_all(&tr2, r2);
3712         } else {
3713                 ASSERT(cnt == 1);
3714                 ASSERT(r2 == NULL);
3715         }
3716         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_INSERT, fname, desc, ip, idx,
3717                 cnt, &tr1, cnt == 2 ? &tr2 : NULL, whichfork);
3718 }
3719
3720 /*
3721  * Add bmap trace entry after updating an extent list entry in place.
3722  */
3723 STATIC void
3724 xfs_bmap_trace_post_update(
3725         char            *fname,         /* function name */
3726         char            *desc,          /* operation description */
3727         xfs_inode_t     *ip,            /* incore inode pointer */
3728         xfs_extnum_t    idx,            /* index of entry updated */
3729         int             whichfork)      /* data or attr fork */
3730 {
3731         xfs_ifork_t     *ifp;           /* inode fork pointer */
3732
3733         ifp = XFS_IFORK_PTR(ip, whichfork);
3734         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_POST_UP, fname, desc, ip, idx,
3735                 1, &ifp->if_u1.if_extents[idx], NULL, whichfork);
3736 }
3737
3738 /*
3739  * Add bmap trace entry prior to updating an extent list entry in place.
3740  */
3741 STATIC void
3742 xfs_bmap_trace_pre_update(
3743         char            *fname,         /* function name */
3744         char            *desc,          /* operation description */
3745         xfs_inode_t     *ip,            /* incore inode pointer */
3746         xfs_extnum_t    idx,            /* index of entry to be updated */
3747         int             whichfork)      /* data or attr fork */
3748 {
3749         xfs_ifork_t     *ifp;           /* inode fork pointer */
3750
3751         ifp = XFS_IFORK_PTR(ip, whichfork);
3752         xfs_bmap_trace_addentry(XFS_BMAP_KTRACE_PRE_UP, fname, desc, ip, idx, 1,
3753                 &ifp->if_u1.if_extents[idx], NULL, whichfork);
3754 }
3755 #endif  /* XFS_BMAP_TRACE */
3756
3757 /*
3758  * Compute the worst-case number of indirect blocks that will be used
3759  * for ip's delayed extent of length "len".
3760  */
3761 STATIC xfs_filblks_t
3762 xfs_bmap_worst_indlen(
3763         xfs_inode_t     *ip,            /* incore inode pointer */
3764         xfs_filblks_t   len)            /* delayed extent length */
3765 {
3766         int             level;          /* btree level number */
3767         int             maxrecs;        /* maximum record count at this level */
3768         xfs_mount_t     *mp;            /* mount structure */
3769         xfs_filblks_t   rval;           /* return value */
3770
3771         mp = ip->i_mount;
3772         maxrecs = mp->m_bmap_dmxr[0];
3773         for (level = 0, rval = 0;
3774              level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
3775              level++) {
3776                 len += maxrecs - 1;
3777                 do_div(len, maxrecs);
3778                 rval += len;
3779                 if (len == 1)
3780                         return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
3781                                 level - 1;
3782                 if (level == 0)
3783                         maxrecs = mp->m_bmap_dmxr[1];
3784         }
3785         return rval;
3786 }
3787
3788 #if defined(XFS_RW_TRACE)
3789 STATIC void
3790 xfs_bunmap_trace(
3791         xfs_inode_t             *ip,
3792         xfs_fileoff_t           bno,
3793         xfs_filblks_t           len,
3794         int                     flags,
3795         inst_t                  *ra)
3796 {
3797         if (ip->i_rwtrace == NULL)
3798                 return;
3799         ktrace_enter(ip->i_rwtrace,
3800                 (void *)(__psint_t)XFS_BUNMAPI,
3801                 (void *)ip,
3802                 (void *)(__psint_t)((ip->i_d.di_size >> 32) & 0xffffffff),
3803                 (void *)(__psint_t)(ip->i_d.di_size & 0xffffffff),
3804                 (void *)(__psint_t)(((xfs_dfiloff_t)bno >> 32) & 0xffffffff),
3805                 (void *)(__psint_t)((xfs_dfiloff_t)bno & 0xffffffff),
3806                 (void *)(__psint_t)len,
3807                 (void *)(__psint_t)flags,
3808                 (void *)(unsigned long)current_cpu(),
3809                 (void *)ra,
3810                 (void *)0,
3811                 (void *)0,
3812                 (void *)0,
3813                 (void *)0,
3814                 (void *)0,
3815                 (void *)0);
3816 }
3817 #endif
3818
3819 /*
3820  * Convert inode from non-attributed to attributed.
3821  * Must not be in a transaction, ip must not be locked.
3822  */
3823 int                                             /* error code */
3824 xfs_bmap_add_attrfork(
3825         xfs_inode_t             *ip,            /* incore inode pointer */
3826         int                     size,           /* space new attribute needs */
3827         int                     rsvd)           /* xact may use reserved blks */
3828 {
3829         xfs_fsblock_t           firstblock;     /* 1st block/ag allocated */
3830         xfs_bmap_free_t         flist;          /* freed extent list */
3831         xfs_mount_t             *mp;            /* mount structure */
3832         xfs_trans_t             *tp;            /* transaction pointer */
3833         unsigned long           s;              /* spinlock spl value */
3834         int                     blks;           /* space reservation */
3835         int                     version = 1;    /* superblock attr version */
3836         int                     committed;      /* xaction was committed */
3837         int                     logflags;       /* logging flags */
3838         int                     error;          /* error return value */
3839
3840         ASSERT(XFS_IFORK_Q(ip) == 0);
3841         ASSERT(ip->i_df.if_ext_max ==
3842                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3843
3844         mp = ip->i_mount;
3845         ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
3846         tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
3847         blks = XFS_ADDAFORK_SPACE_RES(mp);
3848         if (rsvd)
3849                 tp->t_flags |= XFS_TRANS_RESERVE;
3850         if ((error = xfs_trans_reserve(tp, blks, XFS_ADDAFORK_LOG_RES(mp), 0,
3851                         XFS_TRANS_PERM_LOG_RES, XFS_ADDAFORK_LOG_COUNT)))
3852                 goto error0;
3853         xfs_ilock(ip, XFS_ILOCK_EXCL);
3854         error = XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, tp, ip, blks, 0, rsvd ?
3855                         XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
3856                         XFS_QMOPT_RES_REGBLKS);
3857         if (error) {
3858                 xfs_iunlock(ip, XFS_ILOCK_EXCL);
3859                 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
3860                 return error;
3861         }
3862         if (XFS_IFORK_Q(ip))
3863                 goto error1;
3864         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
3865                 /*
3866                  * For inodes coming from pre-6.2 filesystems.
3867                  */
3868                 ASSERT(ip->i_d.di_aformat == 0);
3869                 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
3870         }
3871         ASSERT(ip->i_d.di_anextents == 0);
3872         VN_HOLD(XFS_ITOV(ip));
3873         xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
3874         xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
3875         switch (ip->i_d.di_format) {
3876         case XFS_DINODE_FMT_DEV:
3877                 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
3878                 break;
3879         case XFS_DINODE_FMT_UUID:
3880                 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
3881                 break;
3882         case XFS_DINODE_FMT_LOCAL:
3883         case XFS_DINODE_FMT_EXTENTS:
3884         case XFS_DINODE_FMT_BTREE:
3885                 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
3886                 if (!ip->i_d.di_forkoff)
3887                         ip->i_d.di_forkoff = mp->m_attroffset >> 3;
3888                 else if (!(mp->m_flags & XFS_MOUNT_COMPAT_ATTR))
3889                         version = 2;
3890                 break;
3891         default:
3892                 ASSERT(0);
3893                 error = XFS_ERROR(EINVAL);
3894                 goto error1;
3895         }
3896         ip->i_df.if_ext_max =
3897                 XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3898         ASSERT(ip->i_afp == NULL);
3899         ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
3900         ip->i_afp->if_ext_max =
3901                 XFS_IFORK_ASIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t);
3902         ip->i_afp->if_flags = XFS_IFEXTENTS;
3903         logflags = 0;
3904         XFS_BMAP_INIT(&flist, &firstblock);
3905         switch (ip->i_d.di_format) {
3906         case XFS_DINODE_FMT_LOCAL:
3907                 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
3908                         &logflags);
3909                 break;
3910         case XFS_DINODE_FMT_EXTENTS:
3911                 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
3912                         &flist, &logflags);
3913                 break;
3914         case XFS_DINODE_FMT_BTREE:
3915                 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
3916                         &logflags);
3917                 break;
3918         default:
3919                 error = 0;
3920                 break;
3921         }
3922         if (logflags)
3923                 xfs_trans_log_inode(tp, ip, logflags);
3924         if (error)
3925                 goto error2;
3926         if (!XFS_SB_VERSION_HASATTR(&mp->m_sb) ||
3927            (!XFS_SB_VERSION_HASATTR2(&mp->m_sb) && version == 2)) {
3928                 logflags = 0;
3929                 s = XFS_SB_LOCK(mp);
3930                 if (!XFS_SB_VERSION_HASATTR(&mp->m_sb)) {
3931                         XFS_SB_VERSION_ADDATTR(&mp->m_sb);
3932                         logflags |= XFS_SB_VERSIONNUM;
3933                 }
3934                 if (!XFS_SB_VERSION_HASATTR2(&mp->m_sb) && version == 2) {
3935                         XFS_SB_VERSION_ADDATTR2(&mp->m_sb);
3936                         logflags |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
3937                 }
3938                 if (logflags) {
3939                         XFS_SB_UNLOCK(mp, s);
3940                         xfs_mod_sb(tp, logflags);
3941                 } else
3942                         XFS_SB_UNLOCK(mp, s);
3943         }
3944         if ((error = xfs_bmap_finish(&tp, &flist, firstblock, &committed)))
3945                 goto error2;
3946         error = xfs_trans_commit(tp, XFS_TRANS_PERM_LOG_RES, NULL);
3947         ASSERT(ip->i_df.if_ext_max ==
3948                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3949         return error;
3950 error2:
3951         xfs_bmap_cancel(&flist);
3952 error1:
3953         ASSERT(ismrlocked(&ip->i_lock,MR_UPDATE));
3954         xfs_iunlock(ip, XFS_ILOCK_EXCL);
3955 error0:
3956         xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
3957         ASSERT(ip->i_df.if_ext_max ==
3958                XFS_IFORK_DSIZE(ip) / (uint)sizeof(xfs_bmbt_rec_t));
3959         return error;
3960 }
3961
3962 /*
3963  * Add the extent to the list of extents to be free at transaction end.
3964  * The list is maintained sorted (by block number).
3965  */
3966 /* ARGSUSED */
3967 void
3968 xfs_bmap_add_free(
3969         xfs_fsblock_t           bno,            /* fs block number of extent */
3970         xfs_filblks_t           len,            /* length of extent */
3971         xfs_bmap_free_t         *flist,         /* list of extents */
3972         xfs_mount_t             *mp)            /* mount point structure */
3973 {
3974         xfs_bmap_free_item_t    *cur;           /* current (next) element */
3975         xfs_bmap_free_item_t    *new;           /* new element */
3976         xfs_bmap_free_item_t    *prev;          /* previous element */
3977 #ifdef DEBUG
3978         xfs_agnumber_t          agno;
3979         xfs_agblock_t           agbno;
3980
3981         ASSERT(bno != NULLFSBLOCK);
3982         ASSERT(len > 0);
3983         ASSERT(len <= MAXEXTLEN);
3984         ASSERT(!ISNULLSTARTBLOCK(bno));
3985         agno = XFS_FSB_TO_AGNO(mp, bno);
3986         agbno = XFS_FSB_TO_AGBNO(mp, bno);
3987         ASSERT(agno < mp->m_sb.sb_agcount);
3988         ASSERT(agbno < mp->m_sb.sb_agblocks);
3989         ASSERT(len < mp->m_sb.sb_agblocks);
3990         ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
3991 #endif
3992         ASSERT(xfs_bmap_free_item_zone != NULL);
3993         new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
3994         new->xbfi_startblock = bno;
3995         new->xbfi_blockcount = (xfs_extlen_t)len;
3996         for (prev = NULL, cur = flist->xbf_first;
3997              cur != NULL;
3998              prev = cur, cur = cur->xbfi_next) {
3999                 if (cur->xbfi_startblock >= bno)
4000                         break;
4001         }
4002         if (prev)
4003                 prev->xbfi_next = new;
4004         else
4005                 flist->xbf_first = new;
4006         new->xbfi_next = cur;
4007         flist->xbf_count++;
4008 }
4009
4010 /*
4011  * Compute and fill in the value of the maximum depth of a bmap btree
4012  * in this filesystem.  Done once, during mount.
4013  */
4014 void
4015 xfs_bmap_compute_maxlevels(
4016         xfs_mount_t     *mp,            /* file system mount structure */
4017         int             whichfork)      /* data or attr fork */
4018 {
4019         int             level;          /* btree level */
4020         uint            maxblocks;      /* max blocks at this level */
4021         uint            maxleafents;    /* max leaf entries possible */
4022         int             maxrootrecs;    /* max records in root block */
4023         int             minleafrecs;    /* min records in leaf block */
4024         int             minnoderecs;    /* min records in node block */
4025         int             sz;             /* root block size */
4026
4027         /*
4028          * The maximum number of extents in a file, hence the maximum
4029          * number of leaf entries, is controlled by the type of di_nextents
4030          * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
4031          * (a signed 16-bit number, xfs_aextnum_t).
4032          */
4033         if (whichfork == XFS_DATA_FORK) {
4034                 maxleafents = MAXEXTNUM;
4035                 sz = (mp->m_flags & XFS_MOUNT_COMPAT_ATTR) ?
4036                         mp->m_attroffset : XFS_BMDR_SPACE_CALC(MINDBTPTRS);
4037         } else {
4038                 maxleafents = MAXAEXTNUM;
4039                 sz = (mp->m_flags & XFS_MOUNT_COMPAT_ATTR) ?
4040                         mp->m_sb.sb_inodesize - mp->m_attroffset :
4041                         XFS_BMDR_SPACE_CALC(MINABTPTRS);
4042         }
4043         maxrootrecs = (int)XFS_BTREE_BLOCK_MAXRECS(sz, xfs_bmdr, 0);
4044         minleafrecs = mp->m_bmap_dmnr[0];
4045         minnoderecs = mp->m_bmap_dmnr[1];
4046         maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
4047         for (level = 1; maxblocks > 1; level++) {
4048                 if (maxblocks <= maxrootrecs)
4049                         maxblocks = 1;
4050                 else
4051                         maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
4052         }
4053         mp->m_bm_maxlevels[whichfork] = level;
4054 }
4055
4056 /*
4057  * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
4058  * caller.  Frees all the extents that need freeing, which must be done
4059  * last due to locking considerations.  We never free any extents in
4060  * the first transaction.  This is to allow the caller to make the first
4061  * transaction a synchronous one so that the pointers to the data being
4062  * broken in this transaction will be permanent before the data is actually
4063  * freed.  This is necessary to prevent blocks from being reallocated
4064  * and written to before the free and reallocation are actually permanent.
4065  * We do not just make the first transaction synchronous here, because
4066  * there are more efficient ways to gain the same protection in some cases
4067  * (see the file truncation code).
4068  *
4069  * Return 1 if the given transaction was committed and a new one
4070  * started, and 0 otherwise in the committed parameter.
4071  */
4072 /*ARGSUSED*/
4073 int                                             /* error */
4074 xfs_bmap_finish(
4075         xfs_trans_t             **tp,           /* transaction pointer addr */
4076         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
4077         xfs_fsblock_t           firstblock,     /* controlled ag for allocs */
4078         int                     *committed)     /* xact committed or not */
4079 {
4080         xfs_efd_log_item_t      *efd;           /* extent free data */
4081         xfs_efi_log_item_t      *efi;           /* extent free intention */
4082         int                     error;          /* error return value */
4083         xfs_bmap_free_item_t    *free;          /* free extent list item */
4084         unsigned int            logres;         /* new log reservation */
4085         unsigned int            logcount;       /* new log count */
4086         xfs_mount_t             *mp;            /* filesystem mount structure */
4087         xfs_bmap_free_item_t    *next;          /* next item on free list */
4088         xfs_trans_t             *ntp;           /* new transaction pointer */
4089
4090         ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
4091         if (flist->xbf_count == 0) {
4092                 *committed = 0;
4093                 return 0;
4094         }
4095         ntp = *tp;
4096         efi = xfs_trans_get_efi(ntp, flist->xbf_count);
4097         for (free = flist->xbf_first; free; free = free->xbfi_next)
4098                 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
4099                         free->xbfi_blockcount);
4100         logres = ntp->t_log_res;
4101         logcount = ntp->t_log_count;
4102         ntp = xfs_trans_dup(*tp);
4103         error = xfs_trans_commit(*tp, 0, NULL);
4104         *tp = ntp;
4105         *committed = 1;
4106         /*
4107          * We have a new transaction, so we should return committed=1,
4108          * even though we're returning an error.
4109          */
4110         if (error) {
4111                 return error;
4112         }
4113         if ((error = xfs_trans_reserve(ntp, 0, logres, 0, XFS_TRANS_PERM_LOG_RES,
4114                         logcount)))
4115                 return error;
4116         efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
4117         for (free = flist->xbf_first; free != NULL; free = next) {
4118                 next = free->xbfi_next;
4119                 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
4120                                 free->xbfi_blockcount))) {
4121                         /*
4122                          * The bmap free list will be cleaned up at a
4123                          * higher level.  The EFI will be canceled when
4124                          * this transaction is aborted.
4125                          * Need to force shutdown here to make sure it
4126                          * happens, since this transaction may not be
4127                          * dirty yet.
4128                          */
4129                         mp = ntp->t_mountp;
4130                         if (!XFS_FORCED_SHUTDOWN(mp))
4131                                 xfs_force_shutdown(mp,
4132                                                    (error == EFSCORRUPTED) ?
4133                                                    XFS_CORRUPT_INCORE :
4134                                                    XFS_METADATA_IO_ERROR);
4135                         return error;
4136                 }
4137                 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
4138                         free->xbfi_blockcount);
4139                 xfs_bmap_del_free(flist, NULL, free);
4140         }
4141         return 0;
4142 }
4143
4144 /*
4145  * Free up any items left in the list.
4146  */
4147 void
4148 xfs_bmap_cancel(
4149         xfs_bmap_free_t         *flist) /* list of bmap_free_items */
4150 {
4151         xfs_bmap_free_item_t    *free;  /* free list item */
4152         xfs_bmap_free_item_t    *next;
4153
4154         if (flist->xbf_count == 0)
4155                 return;
4156         ASSERT(flist->xbf_first != NULL);
4157         for (free = flist->xbf_first; free; free = next) {
4158                 next = free->xbfi_next;
4159                 xfs_bmap_del_free(flist, NULL, free);
4160         }
4161         ASSERT(flist->xbf_count == 0);
4162 }
4163
4164 /*
4165  * Returns the file-relative block number of the first unused block(s)
4166  * in the file with at least "len" logically contiguous blocks free.
4167  * This is the lowest-address hole if the file has holes, else the first block
4168  * past the end of file.
4169  * Return 0 if the file is currently local (in-inode).
4170  */
4171 int                                             /* error */
4172 xfs_bmap_first_unused(
4173         xfs_trans_t     *tp,                    /* transaction pointer */
4174         xfs_inode_t     *ip,                    /* incore inode */
4175         xfs_extlen_t    len,                    /* size of hole to find */
4176         xfs_fileoff_t   *first_unused,          /* unused block */
4177         int             whichfork)              /* data or attr fork */
4178 {
4179         xfs_bmbt_rec_t  *base;                  /* base of extent array */
4180         xfs_bmbt_rec_t  *ep;                    /* pointer to an extent entry */
4181         int             error;                  /* error return value */
4182         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4183         xfs_fileoff_t   lastaddr;               /* last block number seen */
4184         xfs_fileoff_t   lowest;                 /* lowest useful block */
4185         xfs_fileoff_t   max;                    /* starting useful block */
4186         xfs_fileoff_t   off;                    /* offset for this block */
4187         xfs_extnum_t    nextents;               /* number of extent entries */
4188
4189         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
4190                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
4191                XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
4192         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4193                 *first_unused = 0;
4194                 return 0;
4195         }
4196         ifp = XFS_IFORK_PTR(ip, whichfork);
4197         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4198             (error = xfs_iread_extents(tp, ip, whichfork)))
4199                 return error;
4200         lowest = *first_unused;
4201         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4202         base = &ifp->if_u1.if_extents[0];
4203         for (lastaddr = 0, max = lowest, ep = base;
4204              ep < &base[nextents];
4205              ep++) {
4206                 off = xfs_bmbt_get_startoff(ep);
4207                 /*
4208                  * See if the hole before this extent will work.
4209                  */
4210                 if (off >= lowest + len && off - max >= len) {
4211                         *first_unused = max;
4212                         return 0;
4213                 }
4214                 lastaddr = off + xfs_bmbt_get_blockcount(ep);
4215                 max = XFS_FILEOFF_MAX(lastaddr, lowest);
4216         }
4217         *first_unused = max;
4218         return 0;
4219 }
4220
4221 /*
4222  * Returns the file-relative block number of the last block + 1 before
4223  * last_block (input value) in the file.
4224  * This is not based on i_size, it is based on the extent list.
4225  * Returns 0 for local files, as they do not have an extent list.
4226  */
4227 int                                             /* error */
4228 xfs_bmap_last_before(
4229         xfs_trans_t     *tp,                    /* transaction pointer */
4230         xfs_inode_t     *ip,                    /* incore inode */
4231         xfs_fileoff_t   *last_block,            /* last block */
4232         int             whichfork)              /* data or attr fork */
4233 {
4234         xfs_fileoff_t   bno;                    /* input file offset */
4235         int             eof;                    /* hit end of file */
4236         xfs_bmbt_rec_t  *ep;                    /* pointer to last extent */
4237         int             error;                  /* error return value */
4238         xfs_bmbt_irec_t got;                    /* current extent value */
4239         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4240         xfs_extnum_t    lastx;                  /* last extent used */
4241         xfs_bmbt_irec_t prev;                   /* previous extent value */
4242
4243         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4244             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4245             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4246                return XFS_ERROR(EIO);
4247         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4248                 *last_block = 0;
4249                 return 0;
4250         }
4251         ifp = XFS_IFORK_PTR(ip, whichfork);
4252         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4253             (error = xfs_iread_extents(tp, ip, whichfork)))
4254                 return error;
4255         bno = *last_block - 1;
4256         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4257                 &prev);
4258         if (eof || xfs_bmbt_get_startoff(ep) > bno) {
4259                 if (prev.br_startoff == NULLFILEOFF)
4260                         *last_block = 0;
4261                 else
4262                         *last_block = prev.br_startoff + prev.br_blockcount;
4263         }
4264         /*
4265          * Otherwise *last_block is already the right answer.
4266          */
4267         return 0;
4268 }
4269
4270 /*
4271  * Returns the file-relative block number of the first block past eof in
4272  * the file.  This is not based on i_size, it is based on the extent list.
4273  * Returns 0 for local files, as they do not have an extent list.
4274  */
4275 int                                             /* error */
4276 xfs_bmap_last_offset(
4277         xfs_trans_t     *tp,                    /* transaction pointer */
4278         xfs_inode_t     *ip,                    /* incore inode */
4279         xfs_fileoff_t   *last_block,            /* last block */
4280         int             whichfork)              /* data or attr fork */
4281 {
4282         xfs_bmbt_rec_t  *base;                  /* base of extent array */
4283         xfs_bmbt_rec_t  *ep;                    /* pointer to last extent */
4284         int             error;                  /* error return value */
4285         xfs_ifork_t     *ifp;                   /* inode fork pointer */
4286         xfs_extnum_t    nextents;               /* number of extent entries */
4287
4288         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4289             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4290             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
4291                return XFS_ERROR(EIO);
4292         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4293                 *last_block = 0;
4294                 return 0;
4295         }
4296         ifp = XFS_IFORK_PTR(ip, whichfork);
4297         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4298             (error = xfs_iread_extents(tp, ip, whichfork)))
4299                 return error;
4300         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4301         if (!nextents) {
4302                 *last_block = 0;
4303                 return 0;
4304         }
4305         base = &ifp->if_u1.if_extents[0];
4306         ASSERT(base != NULL);
4307         ep = &base[nextents - 1];
4308         *last_block = xfs_bmbt_get_startoff(ep) + xfs_bmbt_get_blockcount(ep);
4309         return 0;
4310 }
4311
4312 /*
4313  * Returns whether the selected fork of the inode has exactly one
4314  * block or not.  For the data fork we check this matches di_size,
4315  * implying the file's range is 0..bsize-1.
4316  */
4317 int                                     /* 1=>1 block, 0=>otherwise */
4318 xfs_bmap_one_block(
4319         xfs_inode_t     *ip,            /* incore inode */
4320         int             whichfork)      /* data or attr fork */
4321 {
4322         xfs_bmbt_rec_t  *ep;            /* ptr to fork's extent */
4323         xfs_ifork_t     *ifp;           /* inode fork pointer */
4324         int             rval;           /* return value */
4325         xfs_bmbt_irec_t s;              /* internal version of extent */
4326
4327 #ifndef DEBUG
4328         if (whichfork == XFS_DATA_FORK)
4329                 return ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize;
4330 #endif  /* !DEBUG */
4331         if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
4332                 return 0;
4333         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4334                 return 0;
4335         ifp = XFS_IFORK_PTR(ip, whichfork);
4336         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
4337         ep = ifp->if_u1.if_extents;
4338         xfs_bmbt_get_all(ep, &s);
4339         rval = s.br_startoff == 0 && s.br_blockcount == 1;
4340         if (rval && whichfork == XFS_DATA_FORK)
4341                 ASSERT(ip->i_d.di_size == ip->i_mount->m_sb.sb_blocksize);
4342         return rval;
4343 }
4344
4345 /*
4346  * Read in the extents to if_extents.
4347  * All inode fields are set up by caller, we just traverse the btree
4348  * and copy the records in. If the file system cannot contain unwritten
4349  * extents, the records are checked for no "state" flags.
4350  */
4351 int                                     /* error */
4352 xfs_bmap_read_extents(
4353         xfs_trans_t             *tp,    /* transaction pointer */
4354         xfs_inode_t             *ip,    /* incore inode */
4355         int                     whichfork) /* data or attr fork */
4356 {
4357         xfs_bmbt_block_t        *block; /* current btree block */
4358         xfs_fsblock_t           bno;    /* block # of "block" */
4359         xfs_buf_t               *bp;    /* buffer for "block" */
4360         int                     error;  /* error return value */
4361         xfs_exntfmt_t           exntf;  /* XFS_EXTFMT_NOSTATE, if checking */
4362 #ifdef XFS_BMAP_TRACE
4363         static char             fname[] = "xfs_bmap_read_extents";
4364 #endif
4365         xfs_extnum_t            i, j;   /* index into the extents list */
4366         xfs_ifork_t             *ifp;   /* fork structure */
4367         int                     level;  /* btree level, for checking */
4368         xfs_mount_t             *mp;    /* file system mount structure */
4369         xfs_bmbt_ptr_t          *pp;    /* pointer to block address */
4370         /* REFERENCED */
4371         xfs_extnum_t            room;   /* number of entries there's room for */
4372         xfs_bmbt_rec_t          *trp;   /* target record pointer */
4373
4374         bno = NULLFSBLOCK;
4375         mp = ip->i_mount;
4376         ifp = XFS_IFORK_PTR(ip, whichfork);
4377         exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
4378                                         XFS_EXTFMT_INODE(ip);
4379         block = ifp->if_broot;
4380         /*
4381          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
4382          */
4383         ASSERT(INT_GET(block->bb_level, ARCH_CONVERT) > 0);
4384         level = INT_GET(block->bb_level, ARCH_CONVERT);
4385         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
4386         ASSERT(INT_GET(*pp, ARCH_CONVERT) != NULLDFSBNO);
4387         ASSERT(XFS_FSB_TO_AGNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agcount);
4388         ASSERT(XFS_FSB_TO_AGBNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agblocks);
4389         bno = INT_GET(*pp, ARCH_CONVERT);
4390         /*
4391          * Go down the tree until leaf level is reached, following the first
4392          * pointer (leftmost) at each level.
4393          */
4394         while (level-- > 0) {
4395                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4396                                 XFS_BMAP_BTREE_REF)))
4397                         return error;
4398                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4399                 XFS_WANT_CORRUPTED_GOTO(
4400                         XFS_BMAP_SANITY_CHECK(mp, block, level),
4401                         error0);
4402                 if (level == 0)
4403                         break;
4404                 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt, block,
4405                         1, mp->m_bmap_dmxr[1]);
4406                 XFS_WANT_CORRUPTED_GOTO(
4407                         XFS_FSB_SANITY_CHECK(mp, INT_GET(*pp, ARCH_CONVERT)),
4408                         error0);
4409                 bno = INT_GET(*pp, ARCH_CONVERT);
4410                 xfs_trans_brelse(tp, bp);
4411         }
4412         /*
4413          * Here with bp and block set to the leftmost leaf node in the tree.
4414          */
4415         room = ifp->if_bytes / (uint)sizeof(*trp);
4416         trp = ifp->if_u1.if_extents;
4417         i = 0;
4418         /*
4419          * Loop over all leaf nodes.  Copy information to the extent list.
4420          */
4421         for (;;) {
4422                 xfs_bmbt_rec_t  *frp, *temp;
4423                 xfs_fsblock_t   nextbno;
4424                 xfs_extnum_t    num_recs;
4425
4426
4427                 num_recs = INT_GET(block->bb_numrecs, ARCH_CONVERT);
4428                 if (unlikely(i + num_recs > room)) {
4429                         ASSERT(i + num_recs <= room);
4430                         xfs_fs_cmn_err(CE_WARN, ip->i_mount,
4431                                 "corrupt dinode %Lu, (btree extents).  Unmount and run xfs_repair.",
4432                                 (unsigned long long) ip->i_ino);
4433                         XFS_ERROR_REPORT("xfs_bmap_read_extents(1)",
4434                                          XFS_ERRLEVEL_LOW,
4435                                         ip->i_mount);
4436                         goto error0;
4437                 }
4438                 XFS_WANT_CORRUPTED_GOTO(
4439                         XFS_BMAP_SANITY_CHECK(mp, block, 0),
4440                         error0);
4441                 /*
4442                  * Read-ahead the next leaf block, if any.
4443                  */
4444                 nextbno = INT_GET(block->bb_rightsib, ARCH_CONVERT);
4445                 if (nextbno != NULLFSBLOCK)
4446                         xfs_btree_reada_bufl(mp, nextbno, 1);
4447                 /*
4448                  * Copy records into the extent list.
4449                  */
4450                 frp = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt,
4451                         block, 1, mp->m_bmap_dmxr[0]);
4452                 temp = trp;
4453                 for (j = 0; j < num_recs; j++, frp++, trp++) {
4454                         trp->l0 = INT_GET(frp->l0, ARCH_CONVERT);
4455                         trp->l1 = INT_GET(frp->l1, ARCH_CONVERT);
4456                 }
4457                 if (exntf == XFS_EXTFMT_NOSTATE) {
4458                         /*
4459                          * Check all attribute bmap btree records and
4460                          * any "older" data bmap btree records for a
4461                          * set bit in the "extent flag" position.
4462                          */
4463                         if (unlikely(xfs_check_nostate_extents(temp, num_recs))) {
4464                                 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
4465                                                  XFS_ERRLEVEL_LOW,
4466                                                  ip->i_mount);
4467                                 goto error0;
4468                         }
4469                 }
4470                 i += num_recs;
4471                 xfs_trans_brelse(tp, bp);
4472                 bno = nextbno;
4473                 /*
4474                  * If we've reached the end, stop.
4475                  */
4476                 if (bno == NULLFSBLOCK)
4477                         break;
4478                 if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
4479                                 XFS_BMAP_BTREE_REF)))
4480                         return error;
4481                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
4482         }
4483         ASSERT(i == ifp->if_bytes / (uint)sizeof(*trp));
4484         ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
4485         xfs_bmap_trace_exlist(fname, ip, i, whichfork);
4486         return 0;
4487 error0:
4488         xfs_trans_brelse(tp, bp);
4489         return XFS_ERROR(EFSCORRUPTED);
4490 }
4491
4492 #ifdef XFS_BMAP_TRACE
4493 /*
4494  * Add bmap trace insert entries for all the contents of the extent list.
4495  */
4496 void
4497 xfs_bmap_trace_exlist(
4498         char            *fname,         /* function name */
4499         xfs_inode_t     *ip,            /* incore inode pointer */
4500         xfs_extnum_t    cnt,            /* count of entries in the list */
4501         int             whichfork)      /* data or attr fork */
4502 {
4503         xfs_bmbt_rec_t  *base;          /* base of extent list */
4504         xfs_bmbt_rec_t  *ep;            /* current entry in extent list */
4505         xfs_extnum_t    idx;            /* extent list entry number */
4506         xfs_ifork_t     *ifp;           /* inode fork pointer */
4507         xfs_bmbt_irec_t s;              /* extent list record */
4508
4509         ifp = XFS_IFORK_PTR(ip, whichfork);
4510         ASSERT(cnt == ifp->if_bytes / (uint)sizeof(*base));
4511         base = ifp->if_u1.if_extents;
4512         for (idx = 0, ep = base; idx < cnt; idx++, ep++) {
4513                 xfs_bmbt_get_all(ep, &s);
4514                 xfs_bmap_trace_insert(fname, "exlist", ip, idx, 1, &s, NULL,
4515                         whichfork);
4516         }
4517 }
4518 #endif
4519
4520 #ifdef DEBUG
4521 /*
4522  * Validate that the bmbt_irecs being returned from bmapi are valid
4523  * given the callers original parameters.  Specifically check the
4524  * ranges of the returned irecs to ensure that they only extent beyond
4525  * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
4526  */
4527 STATIC void
4528 xfs_bmap_validate_ret(
4529         xfs_fileoff_t           bno,
4530         xfs_filblks_t           len,
4531         int                     flags,
4532         xfs_bmbt_irec_t         *mval,
4533         int                     nmap,
4534         int                     ret_nmap)
4535 {
4536         int                     i;              /* index to map values */
4537
4538         ASSERT(ret_nmap <= nmap);
4539
4540         for (i = 0; i < ret_nmap; i++) {
4541                 ASSERT(mval[i].br_blockcount > 0);
4542                 if (!(flags & XFS_BMAPI_ENTIRE)) {
4543                         ASSERT(mval[i].br_startoff >= bno);
4544                         ASSERT(mval[i].br_blockcount <= len);
4545                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
4546                                bno + len);
4547                 } else {
4548                         ASSERT(mval[i].br_startoff < bno + len);
4549                         ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
4550                                bno);
4551                 }
4552                 ASSERT(i == 0 ||
4553                        mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
4554                        mval[i].br_startoff);
4555                 if ((flags & XFS_BMAPI_WRITE) && !(flags & XFS_BMAPI_DELAY))
4556                         ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
4557                                mval[i].br_startblock != HOLESTARTBLOCK);
4558                 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
4559                        mval[i].br_state == XFS_EXT_UNWRITTEN);
4560         }
4561 }
4562 #endif /* DEBUG */
4563
4564
4565 /*
4566  * Map file blocks to filesystem blocks.
4567  * File range is given by the bno/len pair.
4568  * Adds blocks to file if a write ("flags & XFS_BMAPI_WRITE" set)
4569  * into a hole or past eof.
4570  * Only allocates blocks from a single allocation group,
4571  * to avoid locking problems.
4572  * The returned value in "firstblock" from the first call in a transaction
4573  * must be remembered and presented to subsequent calls in "firstblock".
4574  * An upper bound for the number of blocks to be allocated is supplied to
4575  * the first call in "total"; if no allocation group has that many free
4576  * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4577  */
4578 int                                     /* error */
4579 xfs_bmapi(
4580         xfs_trans_t     *tp,            /* transaction pointer */
4581         xfs_inode_t     *ip,            /* incore inode */
4582         xfs_fileoff_t   bno,            /* starting file offs. mapped */
4583         xfs_filblks_t   len,            /* length to map in file */
4584         int             flags,          /* XFS_BMAPI_... */
4585         xfs_fsblock_t   *firstblock,    /* first allocated block
4586                                            controls a.g. for allocs */
4587         xfs_extlen_t    total,          /* total blocks needed */
4588         xfs_bmbt_irec_t *mval,          /* output: map values */
4589         int             *nmap,          /* i/o: mval size/count */
4590         xfs_bmap_free_t *flist)         /* i/o: list extents to free */
4591 {
4592         xfs_fsblock_t   abno;           /* allocated block number */
4593         xfs_extlen_t    alen;           /* allocated extent length */
4594         xfs_fileoff_t   aoff;           /* allocated file offset */
4595         xfs_bmalloca_t  bma;            /* args for xfs_bmap_alloc */
4596         xfs_btree_cur_t *cur;           /* bmap btree cursor */
4597         xfs_fileoff_t   end;            /* end of mapped file region */
4598         int             eof;            /* we've hit the end of extent list */
4599         char            contig;         /* allocation must be one extent */
4600         char            delay;          /* this request is for delayed alloc */
4601         char            exact;          /* don't do all of wasdelayed extent */
4602         xfs_bmbt_rec_t  *ep;            /* extent list entry pointer */
4603         int             error;          /* error return */
4604         xfs_bmbt_irec_t got;            /* current extent list record */
4605         xfs_ifork_t     *ifp;           /* inode fork pointer */
4606         xfs_extlen_t    indlen;         /* indirect blocks length */
4607         xfs_extnum_t    lastx;          /* last useful extent number */
4608         int             logflags;       /* flags for transaction logging */
4609         xfs_extlen_t    minleft;        /* min blocks left after allocation */
4610         xfs_extlen_t    minlen;         /* min allocation size */
4611         xfs_mount_t     *mp;            /* xfs mount structure */
4612         int             n;              /* current extent index */
4613         int             nallocs;        /* number of extents alloc\'d */
4614         xfs_extnum_t    nextents;       /* number of extents in file */
4615         xfs_fileoff_t   obno;           /* old block number (offset) */
4616         xfs_bmbt_irec_t prev;           /* previous extent list record */
4617         int             tmp_logflags;   /* temp flags holder */
4618         int             whichfork;      /* data or attr fork */
4619         char            inhole;         /* current location is hole in file */
4620         char            stateless;      /* ignore state flag set */
4621         char            trim;           /* output trimmed to match range */
4622         char            userdata;       /* allocating non-metadata */
4623         char            wasdelay;       /* old extent was delayed */
4624         char            wr;             /* this is a write request */
4625         char            rt;             /* this is a realtime file */
4626         char            rsvd;           /* OK to allocate reserved blocks */
4627 #ifdef DEBUG
4628         xfs_fileoff_t   orig_bno;       /* original block number value */
4629         int             orig_flags;     /* original flags arg value */
4630         xfs_filblks_t   orig_len;       /* original value of len arg */
4631         xfs_bmbt_irec_t *orig_mval;     /* original value of mval */
4632         int             orig_nmap;      /* original value of *nmap */
4633
4634         orig_bno = bno;
4635         orig_len = len;
4636         orig_flags = flags;
4637         orig_mval = mval;
4638         orig_nmap = *nmap;
4639 #endif
4640         ASSERT(*nmap >= 1);
4641         ASSERT(*nmap <= XFS_BMAP_MAX_NMAP || !(flags & XFS_BMAPI_WRITE));
4642         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4643                 XFS_ATTR_FORK : XFS_DATA_FORK;
4644         mp = ip->i_mount;
4645         if (unlikely(XFS_TEST_ERROR(
4646             (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4647              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
4648              XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL),
4649              mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4650                 XFS_ERROR_REPORT("xfs_bmapi", XFS_ERRLEVEL_LOW, mp);
4651                 return XFS_ERROR(EFSCORRUPTED);
4652         }
4653         if (XFS_FORCED_SHUTDOWN(mp))
4654                 return XFS_ERROR(EIO);
4655         rt = XFS_IS_REALTIME_INODE(ip);
4656         ifp = XFS_IFORK_PTR(ip, whichfork);
4657         ASSERT(ifp->if_ext_max ==
4658                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
4659         if ((wr = (flags & XFS_BMAPI_WRITE)) != 0)
4660                 XFS_STATS_INC(xs_blk_mapw);
4661         else
4662                 XFS_STATS_INC(xs_blk_mapr);
4663         delay = (flags & XFS_BMAPI_DELAY) != 0;
4664         trim = (flags & XFS_BMAPI_ENTIRE) == 0;
4665         userdata = (flags & XFS_BMAPI_METADATA) == 0;
4666         exact = (flags & XFS_BMAPI_EXACT) != 0;
4667         rsvd = (flags & XFS_BMAPI_RSVBLOCKS) != 0;
4668         contig = (flags & XFS_BMAPI_CONTIG) != 0;
4669         /*
4670          * stateless is used to combine extents which
4671          * differ only due to the state of the extents.
4672          * This technique is used from xfs_getbmap()
4673          * when the caller does not wish to see the
4674          * separation (which is the default).
4675          *
4676          * This technique is also used when writing a
4677          * buffer which has been partially written,
4678          * (usually by being flushed during a chunkread),
4679          * to ensure one write takes place. This also
4680          * prevents a change in the xfs inode extents at
4681          * this time, intentionally. This change occurs
4682          * on completion of the write operation, in
4683          * xfs_strat_comp(), where the xfs_bmapi() call
4684          * is transactioned, and the extents combined.
4685          */
4686         stateless = (flags & XFS_BMAPI_IGSTATE) != 0;
4687         if (stateless && wr)    /* if writing unwritten space, no */
4688                 wr = 0;         /* allocations are allowed */
4689         ASSERT(wr || !delay);
4690         logflags = 0;
4691         nallocs = 0;
4692         cur = NULL;
4693         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
4694                 ASSERT(wr && tp);
4695                 if ((error = xfs_bmap_local_to_extents(tp, ip,
4696                                 firstblock, total, &logflags, whichfork)))
4697                         goto error0;
4698         }
4699         if (wr && *firstblock == NULLFSBLOCK) {
4700                 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4701                         minleft = INT_GET(ifp->if_broot->bb_level, ARCH_CONVERT) + 1;
4702                 else
4703                         minleft = 1;
4704         } else
4705                 minleft = 0;
4706         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
4707             (error = xfs_iread_extents(tp, ip, whichfork)))
4708                 goto error0;
4709         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
4710                 &prev);
4711         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4712         n = 0;
4713         end = bno + len;
4714         obno = bno;
4715         bma.ip = NULL;
4716         while (bno < end && n < *nmap) {
4717                 /*
4718                  * Reading past eof, act as though there's a hole
4719                  * up to end.
4720                  */
4721                 if (eof && !wr)
4722                         got.br_startoff = end;
4723                 inhole = eof || got.br_startoff > bno;
4724                 wasdelay = wr && !inhole && !delay &&
4725                         ISNULLSTARTBLOCK(got.br_startblock);
4726                 /*
4727                  * First, deal with the hole before the allocated space
4728                  * that we found, if any.
4729                  */
4730                 if (wr && (inhole || wasdelay)) {
4731                         /*
4732                          * For the wasdelay case, we could also just
4733                          * allocate the stuff asked for in this bmap call
4734                          * but that wouldn't be as good.
4735                          */
4736                         if (wasdelay && !exact) {
4737                                 alen = (xfs_extlen_t)got.br_blockcount;
4738                                 aoff = got.br_startoff;
4739                                 if (lastx != NULLEXTNUM && lastx) {
4740                                         ep = &ifp->if_u1.if_extents[lastx - 1];
4741                                         xfs_bmbt_get_all(ep, &prev);
4742                                 }
4743                         } else if (wasdelay) {
4744                                 alen = (xfs_extlen_t)
4745                                         XFS_FILBLKS_MIN(len,
4746                                                 (got.br_startoff +
4747                                                  got.br_blockcount) - bno);
4748                                 aoff = bno;
4749                         } else {
4750                                 alen = (xfs_extlen_t)
4751                                         XFS_FILBLKS_MIN(len, MAXEXTLEN);
4752                                 if (!eof)
4753                                         alen = (xfs_extlen_t)
4754                                                 XFS_FILBLKS_MIN(alen,
4755                                                         got.br_startoff - bno);
4756                                 aoff = bno;
4757                         }
4758                         minlen = contig ? alen : 1;
4759                         if (delay) {
4760                                 xfs_extlen_t    extsz = 0;
4761
4762                                 /* Figure out the extent size, adjust alen */
4763                                 if (rt) {
4764                                         if (!(extsz = ip->i_d.di_extsize))
4765                                                 extsz = mp->m_sb.sb_rextsize;
4766                                         alen = roundup(alen, extsz);
4767                                         extsz = alen / mp->m_sb.sb_rextsize;
4768                                 }
4769
4770                                 /*
4771                                  * Make a transaction-less quota reservation for
4772                                  * delayed allocation blocks. This number gets
4773                                  * adjusted later.
4774                                  * We return EDQUOT if we haven't allocated
4775                                  * blks already inside this loop;
4776                                  */
4777                                 if (XFS_TRANS_RESERVE_QUOTA_NBLKS(
4778                                                 mp, NULL, ip, (long)alen, 0,
4779                                                 rt ? XFS_QMOPT_RES_RTBLKS :
4780                                                      XFS_QMOPT_RES_REGBLKS)) {
4781                                         if (n == 0) {
4782                                                 *nmap = 0;
4783                                                 ASSERT(cur == NULL);
4784                                                 return XFS_ERROR(EDQUOT);
4785                                         }
4786                                         break;
4787                                 }
4788
4789                                 /*
4790                                  * Split changing sb for alen and indlen since
4791                                  * they could be coming from different places.
4792                                  */
4793                                 indlen = (xfs_extlen_t)
4794                                         xfs_bmap_worst_indlen(ip, alen);
4795                                 ASSERT(indlen > 0);
4796
4797                                 if (rt)
4798                                         error = xfs_mod_incore_sb(mp,
4799                                                         XFS_SBS_FREXTENTS,
4800                                                         -(extsz), rsvd);
4801                                 else
4802                                         error = xfs_mod_incore_sb(mp,
4803                                                         XFS_SBS_FDBLOCKS,
4804                                                         -(alen), rsvd);
4805                                 if (!error) {
4806                                         error = xfs_mod_incore_sb(mp,
4807                                                         XFS_SBS_FDBLOCKS,
4808                                                         -(indlen), rsvd);
4809                                         if (error && rt) {
4810                                                 xfs_mod_incore_sb(ip->i_mount,
4811                                                         XFS_SBS_FREXTENTS,
4812                                                         extsz, rsvd);
4813                                         } else if (error) {
4814                                                 xfs_mod_incore_sb(ip->i_mount,
4815                                                         XFS_SBS_FDBLOCKS,
4816                                                         alen, rsvd);
4817                                         }
4818                                 }
4819
4820                                 if (error) {
4821                                         if (XFS_IS_QUOTA_ON(ip->i_mount))
4822                                                 /* unreserve the blocks now */
4823                                                 XFS_TRANS_UNRESERVE_QUOTA_NBLKS(
4824                                                         mp, NULL, ip,
4825                                                         (long)alen, 0, rt ?
4826                                                         XFS_QMOPT_RES_RTBLKS :
4827                                                         XFS_QMOPT_RES_REGBLKS);
4828                                         break;
4829                                 }
4830
4831                                 ip->i_delayed_blks += alen;
4832                                 abno = NULLSTARTBLOCK(indlen);
4833                         } else {
4834                                 /*
4835                                  * If first time, allocate and fill in
4836                                  * once-only bma fields.
4837                                  */
4838                                 if (bma.ip == NULL) {
4839                                         bma.tp = tp;
4840                                         bma.ip = ip;
4841                                         bma.prevp = &prev;
4842                                         bma.gotp = &got;
4843                                         bma.total = total;
4844                                         bma.userdata = 0;
4845                                 }
4846                                 /* Indicate if this is the first user data
4847                                  * in the file, or just any user data.
4848                                  */
4849                                 if (userdata) {
4850                                         bma.userdata = (aoff == 0) ?
4851                                                 XFS_ALLOC_INITIAL_USER_DATA :
4852                                                 XFS_ALLOC_USERDATA;
4853                                 }
4854                                 /*
4855                                  * Fill in changeable bma fields.
4856                                  */
4857                                 bma.eof = eof;
4858                                 bma.firstblock = *firstblock;
4859                                 bma.alen = alen;
4860                                 bma.off = aoff;
4861                                 bma.wasdel = wasdelay;
4862                                 bma.minlen = minlen;
4863                                 bma.low = flist->xbf_low;
4864                                 bma.minleft = minleft;
4865                                 /*
4866                                  * Only want to do the alignment at the
4867                                  * eof if it is userdata and allocation length
4868                                  * is larger than a stripe unit.
4869                                  */
4870                                 if (mp->m_dalign && alen >= mp->m_dalign &&
4871                                     userdata && whichfork == XFS_DATA_FORK) {
4872                                         if ((error = xfs_bmap_isaeof(ip, aoff,
4873                                                         whichfork, &bma.aeof)))
4874                                                 goto error0;
4875                                 } else
4876                                         bma.aeof = 0;
4877                                 /*
4878                                  * Call allocator.
4879                                  */
4880                                 if ((error = xfs_bmap_alloc(&bma)))
4881                                         goto error0;
4882                                 /*
4883                                  * Copy out result fields.
4884                                  */
4885                                 abno = bma.rval;
4886                                 if ((flist->xbf_low = bma.low))
4887                                         minleft = 0;
4888                                 alen = bma.alen;
4889                                 aoff = bma.off;
4890                                 ASSERT(*firstblock == NULLFSBLOCK ||
4891                                        XFS_FSB_TO_AGNO(mp, *firstblock) ==
4892                                        XFS_FSB_TO_AGNO(mp, bma.firstblock) ||
4893                                        (flist->xbf_low &&
4894                                         XFS_FSB_TO_AGNO(mp, *firstblock) <
4895                                         XFS_FSB_TO_AGNO(mp, bma.firstblock)));
4896                                 *firstblock = bma.firstblock;
4897                                 if (cur)
4898                                         cur->bc_private.b.firstblock =
4899                                                 *firstblock;
4900                                 if (abno == NULLFSBLOCK)
4901                                         break;
4902                                 if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
4903                                         cur = xfs_btree_init_cursor(mp,
4904                                                 tp, NULL, 0, XFS_BTNUM_BMAP,
4905                                                 ip, whichfork);
4906                                         cur->bc_private.b.firstblock =
4907                                                 *firstblock;
4908                                         cur->bc_private.b.flist = flist;
4909                                 }
4910                                 /*
4911                                  * Bump the number of extents we've allocated
4912                                  * in this call.
4913                                  */
4914                                 nallocs++;
4915                         }
4916                         if (cur)
4917                                 cur->bc_private.b.flags =
4918                                         wasdelay ? XFS_BTCUR_BPRV_WASDEL : 0;
4919                         got.br_startoff = aoff;
4920                         got.br_startblock = abno;
4921                         got.br_blockcount = alen;
4922                         got.br_state = XFS_EXT_NORM;    /* assume normal */
4923                         /*
4924                          * Determine state of extent, and the filesystem.
4925                          * A wasdelay extent has been initialized, so
4926                          * shouldn't be flagged as unwritten.
4927                          */
4928                         if (wr && XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
4929                                 if (!wasdelay && (flags & XFS_BMAPI_PREALLOC))
4930                                         got.br_state = XFS_EXT_UNWRITTEN;
4931                         }
4932                         error = xfs_bmap_add_extent(ip, lastx, &cur, &got,
4933                                 firstblock, flist, &tmp_logflags, whichfork,
4934                                 rsvd);
4935                         logflags |= tmp_logflags;
4936                         if (error)
4937                                 goto error0;
4938                         lastx = ifp->if_lastex;
4939                         ep = &ifp->if_u1.if_extents[lastx];
4940                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
4941                         xfs_bmbt_get_all(ep, &got);
4942                         ASSERT(got.br_startoff <= aoff);
4943                         ASSERT(got.br_startoff + got.br_blockcount >=
4944                                 aoff + alen);
4945 #ifdef DEBUG
4946                         if (delay) {
4947                                 ASSERT(ISNULLSTARTBLOCK(got.br_startblock));
4948                                 ASSERT(STARTBLOCKVAL(got.br_startblock) > 0);
4949                         }
4950                         ASSERT(got.br_state == XFS_EXT_NORM ||
4951                                got.br_state == XFS_EXT_UNWRITTEN);
4952 #endif
4953                         /*
4954                          * Fall down into the found allocated space case.
4955                          */
4956                 } else if (inhole) {
4957                         /*
4958                          * Reading in a hole.
4959                          */
4960                         mval->br_startoff = bno;
4961                         mval->br_startblock = HOLESTARTBLOCK;
4962                         mval->br_blockcount =
4963                                 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4964                         mval->br_state = XFS_EXT_NORM;
4965                         bno += mval->br_blockcount;
4966                         len -= mval->br_blockcount;
4967                         mval++;
4968                         n++;
4969                         continue;
4970                 }
4971                 /*
4972                  * Then deal with the allocated space we found.
4973                  */
4974                 ASSERT(ep != NULL);
4975                 if (trim && (got.br_startoff + got.br_blockcount > obno)) {
4976                         if (obno > bno)
4977                                 bno = obno;
4978                         ASSERT((bno >= obno) || (n == 0));
4979                         ASSERT(bno < end);
4980                         mval->br_startoff = bno;
4981                         if (ISNULLSTARTBLOCK(got.br_startblock)) {
4982                                 ASSERT(!wr || delay);
4983                                 mval->br_startblock = DELAYSTARTBLOCK;
4984                         } else
4985                                 mval->br_startblock =
4986                                         got.br_startblock +
4987                                         (bno - got.br_startoff);
4988                         /*
4989                          * Return the minimum of what we got and what we
4990                          * asked for for the length.  We can use the len
4991                          * variable here because it is modified below
4992                          * and we could have been there before coming
4993                          * here if the first part of the allocation
4994                          * didn't overlap what was asked for.
4995                          */
4996                         mval->br_blockcount =
4997                                 XFS_FILBLKS_MIN(end - bno, got.br_blockcount -
4998                                         (bno - got.br_startoff));
4999                         mval->br_state = got.br_state;
5000                         ASSERT(mval->br_blockcount <= len);
5001                 } else {
5002                         *mval = got;
5003                         if (ISNULLSTARTBLOCK(mval->br_startblock)) {
5004                                 ASSERT(!wr || delay);
5005                                 mval->br_startblock = DELAYSTARTBLOCK;
5006                         }
5007                 }
5008
5009                 /*
5010                  * Check if writing previously allocated but
5011                  * unwritten extents.
5012                  */
5013                 if (wr && mval->br_state == XFS_EXT_UNWRITTEN &&
5014                     ((flags & (XFS_BMAPI_PREALLOC|XFS_BMAPI_DELAY)) == 0)) {
5015                         /*
5016                          * Modify (by adding) the state flag, if writing.
5017                          */
5018                         ASSERT(mval->br_blockcount <= len);
5019                         if ((ifp->if_flags & XFS_IFBROOT) && !cur) {
5020                                 cur = xfs_btree_init_cursor(mp,
5021                                         tp, NULL, 0, XFS_BTNUM_BMAP,
5022                                         ip, whichfork);
5023                                 cur->bc_private.b.firstblock =
5024                                         *firstblock;
5025                                 cur->bc_private.b.flist = flist;
5026                         }
5027                         mval->br_state = XFS_EXT_NORM;
5028                         error = xfs_bmap_add_extent(ip, lastx, &cur, mval,
5029                                 firstblock, flist, &tmp_logflags, whichfork,
5030                                 rsvd);
5031                         logflags |= tmp_logflags;
5032                         if (error)
5033                                 goto error0;
5034                         lastx = ifp->if_lastex;
5035                         ep = &ifp->if_u1.if_extents[lastx];
5036                         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5037                         xfs_bmbt_get_all(ep, &got);
5038                         /*
5039                          * We may have combined previously unwritten
5040                          * space with written space, so generate
5041                          * another request.
5042                          */
5043                         if (mval->br_blockcount < len)
5044                                 continue;
5045                 }
5046
5047                 ASSERT(!trim ||
5048                        ((mval->br_startoff + mval->br_blockcount) <= end));
5049                 ASSERT(!trim || (mval->br_blockcount <= len) ||
5050                        (mval->br_startoff < obno));
5051                 bno = mval->br_startoff + mval->br_blockcount;
5052                 len = end - bno;
5053                 if (n > 0 && mval->br_startoff == mval[-1].br_startoff) {
5054                         ASSERT(mval->br_startblock == mval[-1].br_startblock);
5055                         ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
5056                         ASSERT(mval->br_state == mval[-1].br_state);
5057                         mval[-1].br_blockcount = mval->br_blockcount;
5058                         mval[-1].br_state = mval->br_state;
5059                 } else if (n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
5060                            mval[-1].br_startblock != DELAYSTARTBLOCK &&
5061                            mval[-1].br_startblock != HOLESTARTBLOCK &&
5062                            mval->br_startblock ==
5063                            mval[-1].br_startblock + mval[-1].br_blockcount &&
5064                            (stateless || mval[-1].br_state == mval->br_state)) {
5065                         ASSERT(mval->br_startoff ==
5066                                mval[-1].br_startoff + mval[-1].br_blockcount);
5067                         mval[-1].br_blockcount += mval->br_blockcount;
5068                 } else if (n > 0 &&
5069                            mval->br_startblock == DELAYSTARTBLOCK &&
5070                            mval[-1].br_startblock == DELAYSTARTBLOCK &&
5071                            mval->br_startoff ==
5072                            mval[-1].br_startoff + mval[-1].br_blockcount) {
5073                         mval[-1].br_blockcount += mval->br_blockcount;
5074                         mval[-1].br_state = mval->br_state;
5075                 } else if (!((n == 0) &&
5076                              ((mval->br_startoff + mval->br_blockcount) <=
5077                               obno))) {
5078                         mval++;
5079                         n++;
5080                 }
5081                 /*
5082                  * If we're done, stop now.  Stop when we've allocated
5083                  * XFS_BMAP_MAX_NMAP extents no matter what.  Otherwise
5084                  * the transaction may get too big.
5085                  */
5086                 if (bno >= end || n >= *nmap || nallocs >= *nmap)
5087                         break;
5088                 /*
5089                  * Else go on to the next record.
5090                  */
5091                 ep++;
5092                 lastx++;
5093                 if (lastx >= nextents) {
5094                         eof = 1;
5095                         prev = got;
5096                 } else
5097                         xfs_bmbt_get_all(ep, &got);
5098         }
5099         ifp->if_lastex = lastx;
5100         *nmap = n;
5101         /*
5102          * Transform from btree to extents, give it cur.
5103          */
5104         if (tp && XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5105             XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5106                 ASSERT(wr && cur);
5107                 error = xfs_bmap_btree_to_extents(tp, ip, cur,
5108                         &tmp_logflags, whichfork);
5109                 logflags |= tmp_logflags;
5110                 if (error)
5111                         goto error0;
5112         }
5113         ASSERT(ifp->if_ext_max ==
5114                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5115         ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
5116                XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max);
5117         error = 0;
5118
5119 error0:
5120         /*
5121          * Log everything.  Do this after conversion, there's no point in
5122          * logging the extent list if we've converted to btree format.
5123          */
5124         if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5125             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5126                 logflags &= ~XFS_ILOG_FEXT(whichfork);
5127         else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5128                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5129                 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5130         /*
5131          * Log whatever the flags say, even if error.  Otherwise we might miss
5132          * detecting a case where the data is changed, there's an error,
5133          * and it's not logged so we don't shutdown when we should.
5134          */
5135         if (logflags) {
5136                 ASSERT(tp && wr);
5137                 xfs_trans_log_inode(tp, ip, logflags);
5138         }
5139         if (cur) {
5140                 if (!error) {
5141                         ASSERT(*firstblock == NULLFSBLOCK ||
5142                                XFS_FSB_TO_AGNO(mp, *firstblock) ==
5143                                XFS_FSB_TO_AGNO(mp,
5144                                        cur->bc_private.b.firstblock) ||
5145                                (flist->xbf_low &&
5146                                 XFS_FSB_TO_AGNO(mp, *firstblock) <
5147                                 XFS_FSB_TO_AGNO(mp,
5148                                         cur->bc_private.b.firstblock)));
5149                         *firstblock = cur->bc_private.b.firstblock;
5150                 }
5151                 xfs_btree_del_cursor(cur,
5152                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5153         }
5154         if (!error)
5155                 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
5156                         orig_nmap, *nmap);
5157         return error;
5158 }
5159
5160 /*
5161  * Map file blocks to filesystem blocks, simple version.
5162  * One block (extent) only, read-only.
5163  * For flags, only the XFS_BMAPI_ATTRFORK flag is examined.
5164  * For the other flag values, the effect is as if XFS_BMAPI_METADATA
5165  * was set and all the others were clear.
5166  */
5167 int                                             /* error */
5168 xfs_bmapi_single(
5169         xfs_trans_t     *tp,            /* transaction pointer */
5170         xfs_inode_t     *ip,            /* incore inode */
5171         int             whichfork,      /* data or attr fork */
5172         xfs_fsblock_t   *fsb,           /* output: mapped block */
5173         xfs_fileoff_t   bno)            /* starting file offs. mapped */
5174 {
5175         int             eof;            /* we've hit the end of extent list */
5176         int             error;          /* error return */
5177         xfs_bmbt_irec_t got;            /* current extent list record */
5178         xfs_ifork_t     *ifp;           /* inode fork pointer */
5179         xfs_extnum_t    lastx;          /* last useful extent number */
5180         xfs_bmbt_irec_t prev;           /* previous extent list record */
5181
5182         ifp = XFS_IFORK_PTR(ip, whichfork);
5183         if (unlikely(
5184             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
5185             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)) {
5186                XFS_ERROR_REPORT("xfs_bmapi_single", XFS_ERRLEVEL_LOW,
5187                                 ip->i_mount);
5188                return XFS_ERROR(EFSCORRUPTED);
5189         }
5190         if (XFS_FORCED_SHUTDOWN(ip->i_mount))
5191                 return XFS_ERROR(EIO);
5192         XFS_STATS_INC(xs_blk_mapr);
5193         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5194             (error = xfs_iread_extents(tp, ip, whichfork)))
5195                 return error;
5196         (void)xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5197                 &prev);
5198         /*
5199          * Reading past eof, act as though there's a hole
5200          * up to end.
5201          */
5202         if (eof || got.br_startoff > bno) {
5203                 *fsb = NULLFSBLOCK;
5204                 return 0;
5205         }
5206         ASSERT(!ISNULLSTARTBLOCK(got.br_startblock));
5207         ASSERT(bno < got.br_startoff + got.br_blockcount);
5208         *fsb = got.br_startblock + (bno - got.br_startoff);
5209         ifp->if_lastex = lastx;
5210         return 0;
5211 }
5212
5213 /*
5214  * Unmap (remove) blocks from a file.
5215  * If nexts is nonzero then the number of extents to remove is limited to
5216  * that value.  If not all extents in the block range can be removed then
5217  * *done is set.
5218  */
5219 int                                             /* error */
5220 xfs_bunmapi(
5221         xfs_trans_t             *tp,            /* transaction pointer */
5222         struct xfs_inode        *ip,            /* incore inode */
5223         xfs_fileoff_t           bno,            /* starting offset to unmap */
5224         xfs_filblks_t           len,            /* length to unmap in file */
5225         int                     flags,          /* misc flags */
5226         xfs_extnum_t            nexts,          /* number of extents max */
5227         xfs_fsblock_t           *firstblock,    /* first allocated block
5228                                                    controls a.g. for allocs */
5229         xfs_bmap_free_t         *flist,         /* i/o: list extents to free */
5230         int                     *done)          /* set if not done yet */
5231 {
5232         xfs_btree_cur_t         *cur;           /* bmap btree cursor */
5233         xfs_bmbt_irec_t         del;            /* extent being deleted */
5234         int                     eof;            /* is deleting at eof */
5235         xfs_bmbt_rec_t          *ep;            /* extent list entry pointer */
5236         int                     error;          /* error return value */
5237         xfs_extnum_t            extno;          /* extent number in list */
5238         xfs_bmbt_irec_t         got;            /* current extent list entry */
5239         xfs_ifork_t             *ifp;           /* inode fork pointer */
5240         int                     isrt;           /* freeing in rt area */
5241         xfs_extnum_t            lastx;          /* last extent index used */
5242         int                     logflags;       /* transaction logging flags */
5243         xfs_extlen_t            mod;            /* rt extent offset */
5244         xfs_mount_t             *mp;            /* mount structure */
5245         xfs_extnum_t            nextents;       /* size of extent list */
5246         xfs_bmbt_irec_t         prev;           /* previous extent list entry */
5247         xfs_fileoff_t           start;          /* first file offset deleted */
5248         int                     tmp_logflags;   /* partial logging flags */
5249         int                     wasdel;         /* was a delayed alloc extent */
5250         int                     whichfork;      /* data or attribute fork */
5251         int                     rsvd;           /* OK to allocate reserved blocks */
5252         xfs_fsblock_t           sum;
5253
5254         xfs_bunmap_trace(ip, bno, len, flags, (inst_t *)__return_address);
5255         whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5256                 XFS_ATTR_FORK : XFS_DATA_FORK;
5257         ifp = XFS_IFORK_PTR(ip, whichfork);
5258         if (unlikely(
5259             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5260             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5261                 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5262                                  ip->i_mount);
5263                 return XFS_ERROR(EFSCORRUPTED);
5264         }
5265         mp = ip->i_mount;
5266         if (XFS_FORCED_SHUTDOWN(mp))
5267                 return XFS_ERROR(EIO);
5268         rsvd = (flags & XFS_BMAPI_RSVBLOCKS) != 0;
5269         ASSERT(len > 0);
5270         ASSERT(nexts >= 0);
5271         ASSERT(ifp->if_ext_max ==
5272                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5273         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5274             (error = xfs_iread_extents(tp, ip, whichfork)))
5275                 return error;
5276         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5277         if (nextents == 0) {
5278                 *done = 1;
5279                 return 0;
5280         }
5281         XFS_STATS_INC(xs_blk_unmap);
5282         isrt = (whichfork == XFS_DATA_FORK) &&
5283                (ip->i_d.di_flags & XFS_DIFLAG_REALTIME);
5284         start = bno;
5285         bno = start + len - 1;
5286         ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5287                 &prev);
5288         /*
5289          * Check to see if the given block number is past the end of the
5290          * file, back up to the last block if so...
5291          */
5292         if (eof) {
5293                 ep = &ifp->if_u1.if_extents[--lastx];
5294                 xfs_bmbt_get_all(ep, &got);
5295                 bno = got.br_startoff + got.br_blockcount - 1;
5296         }
5297         logflags = 0;
5298         if (ifp->if_flags & XFS_IFBROOT) {
5299                 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5300                 cur = xfs_btree_init_cursor(mp, tp, NULL, 0, XFS_BTNUM_BMAP, ip,
5301                         whichfork);
5302                 cur->bc_private.b.firstblock = *firstblock;
5303                 cur->bc_private.b.flist = flist;
5304                 cur->bc_private.b.flags = 0;
5305         } else
5306                 cur = NULL;
5307         extno = 0;
5308         while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5309                (nexts == 0 || extno < nexts)) {
5310                 /*
5311                  * Is the found extent after a hole in which bno lives?
5312                  * Just back up to the previous extent, if so.
5313                  */
5314                 if (got.br_startoff > bno) {
5315                         if (--lastx < 0)
5316                                 break;
5317                         ep--;
5318                         xfs_bmbt_get_all(ep, &got);
5319                 }
5320                 /*
5321                  * Is the last block of this extent before the range
5322                  * we're supposed to delete?  If so, we're done.
5323                  */
5324                 bno = XFS_FILEOFF_MIN(bno,
5325                         got.br_startoff + got.br_blockcount - 1);
5326                 if (bno < start)
5327                         break;
5328                 /*
5329                  * Then deal with the (possibly delayed) allocated space
5330                  * we found.
5331                  */
5332                 ASSERT(ep != NULL);
5333                 del = got;
5334                 wasdel = ISNULLSTARTBLOCK(del.br_startblock);
5335                 if (got.br_startoff < start) {
5336                         del.br_startoff = start;
5337                         del.br_blockcount -= start - got.br_startoff;
5338                         if (!wasdel)
5339                                 del.br_startblock += start - got.br_startoff;
5340                 }
5341                 if (del.br_startoff + del.br_blockcount > bno + 1)
5342                         del.br_blockcount = bno + 1 - del.br_startoff;
5343                 sum = del.br_startblock + del.br_blockcount;
5344                 if (isrt &&
5345                     (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5346                         /*
5347                          * Realtime extent not lined up at the end.
5348                          * The extent could have been split into written
5349                          * and unwritten pieces, or we could just be
5350                          * unmapping part of it.  But we can't really
5351                          * get rid of part of a realtime extent.
5352                          */
5353                         if (del.br_state == XFS_EXT_UNWRITTEN ||
5354                             !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5355                                 /*
5356                                  * This piece is unwritten, or we're not
5357                                  * using unwritten extents.  Skip over it.
5358                                  */
5359                                 ASSERT(bno >= mod);
5360                                 bno -= mod > del.br_blockcount ?
5361                                         del.br_blockcount : mod;
5362                                 if (bno < got.br_startoff) {
5363                                         if (--lastx >= 0)
5364                                                 xfs_bmbt_get_all(--ep, &got);
5365                                 }
5366                                 continue;
5367                         }
5368                         /*
5369                          * It's written, turn it unwritten.
5370                          * This is better than zeroing it.
5371                          */
5372                         ASSERT(del.br_state == XFS_EXT_NORM);
5373                         ASSERT(xfs_trans_get_block_res(tp) > 0);
5374                         /*
5375                          * If this spans a realtime extent boundary,
5376                          * chop it back to the start of the one we end at.
5377                          */
5378                         if (del.br_blockcount > mod) {
5379                                 del.br_startoff += del.br_blockcount - mod;
5380                                 del.br_startblock += del.br_blockcount - mod;
5381                                 del.br_blockcount = mod;
5382                         }
5383                         del.br_state = XFS_EXT_UNWRITTEN;
5384                         error = xfs_bmap_add_extent(ip, lastx, &cur, &del,
5385                                 firstblock, flist, &logflags, XFS_DATA_FORK, 0);
5386                         if (error)
5387                                 goto error0;
5388                         goto nodelete;
5389                 }
5390                 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5391                         /*
5392                          * Realtime extent is lined up at the end but not
5393                          * at the front.  We'll get rid of full extents if
5394                          * we can.
5395                          */
5396                         mod = mp->m_sb.sb_rextsize - mod;
5397                         if (del.br_blockcount > mod) {
5398                                 del.br_blockcount -= mod;
5399                                 del.br_startoff += mod;
5400                                 del.br_startblock += mod;
5401                         } else if ((del.br_startoff == start &&
5402                                     (del.br_state == XFS_EXT_UNWRITTEN ||
5403                                      xfs_trans_get_block_res(tp) == 0)) ||
5404                                    !XFS_SB_VERSION_HASEXTFLGBIT(&mp->m_sb)) {
5405                                 /*
5406                                  * Can't make it unwritten.  There isn't
5407                                  * a full extent here so just skip it.
5408                                  */
5409                                 ASSERT(bno >= del.br_blockcount);
5410                                 bno -= del.br_blockcount;
5411                                 if (bno < got.br_startoff) {
5412                                         if (--lastx >= 0)
5413                                                 xfs_bmbt_get_all(--ep, &got);
5414                                 }
5415                                 continue;
5416                         } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5417                                 /*
5418                                  * This one is already unwritten.
5419                                  * It must have a written left neighbor.
5420                                  * Unwrite the killed part of that one and
5421                                  * try again.
5422                                  */
5423                                 ASSERT(lastx > 0);
5424                                 xfs_bmbt_get_all(ep - 1, &prev);
5425                                 ASSERT(prev.br_state == XFS_EXT_NORM);
5426                                 ASSERT(!ISNULLSTARTBLOCK(prev.br_startblock));
5427                                 ASSERT(del.br_startblock ==
5428                                        prev.br_startblock + prev.br_blockcount);
5429                                 if (prev.br_startoff < start) {
5430                                         mod = start - prev.br_startoff;
5431                                         prev.br_blockcount -= mod;
5432                                         prev.br_startblock += mod;
5433                                         prev.br_startoff = start;
5434                                 }
5435                                 prev.br_state = XFS_EXT_UNWRITTEN;
5436                                 error = xfs_bmap_add_extent(ip, lastx - 1, &cur,
5437                                         &prev, firstblock, flist, &logflags,
5438                                         XFS_DATA_FORK, 0);
5439                                 if (error)
5440                                         goto error0;
5441                                 goto nodelete;
5442                         } else {
5443                                 ASSERT(del.br_state == XFS_EXT_NORM);
5444                                 del.br_state = XFS_EXT_UNWRITTEN;
5445                                 error = xfs_bmap_add_extent(ip, lastx, &cur,
5446                                         &del, firstblock, flist, &logflags,
5447                                         XFS_DATA_FORK, 0);
5448                                 if (error)
5449                                         goto error0;
5450                                 goto nodelete;
5451                         }
5452                 }
5453                 if (wasdel) {
5454                         ASSERT(STARTBLOCKVAL(del.br_startblock) > 0);
5455                         /* Update realtim/data freespace, unreserve quota */
5456                         if (isrt) {
5457                                 xfs_filblks_t rtexts;
5458
5459                                 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5460                                 do_div(rtexts, mp->m_sb.sb_rextsize);
5461                                 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5462                                                 (int)rtexts, rsvd);
5463                                 XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, NULL, ip,
5464                                         -((long)del.br_blockcount), 0,
5465                                         XFS_QMOPT_RES_RTBLKS);
5466                         } else {
5467                                 xfs_mod_incore_sb(mp, XFS_SBS_FDBLOCKS,
5468                                                 (int)del.br_blockcount, rsvd);
5469                                 XFS_TRANS_RESERVE_QUOTA_NBLKS(mp, NULL, ip,
5470                                         -((long)del.br_blockcount), 0,
5471                                         XFS_QMOPT_RES_REGBLKS);
5472                         }
5473                         ip->i_delayed_blks -= del.br_blockcount;
5474                         if (cur)
5475                                 cur->bc_private.b.flags |=
5476                                         XFS_BTCUR_BPRV_WASDEL;
5477                 } else if (cur)
5478                         cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5479                 /*
5480                  * If it's the case where the directory code is running
5481                  * with no block reservation, and the deleted block is in
5482                  * the middle of its extent, and the resulting insert
5483                  * of an extent would cause transformation to btree format,
5484                  * then reject it.  The calling code will then swap
5485                  * blocks around instead.
5486                  * We have to do this now, rather than waiting for the
5487                  * conversion to btree format, since the transaction
5488                  * will be dirty.
5489                  */
5490                 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5491                     XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5492                     XFS_IFORK_NEXTENTS(ip, whichfork) >= ifp->if_ext_max &&
5493                     del.br_startoff > got.br_startoff &&
5494                     del.br_startoff + del.br_blockcount <
5495                     got.br_startoff + got.br_blockcount) {
5496                         error = XFS_ERROR(ENOSPC);
5497                         goto error0;
5498                 }
5499                 error = xfs_bmap_del_extent(ip, tp, lastx, flist, cur, &del,
5500                         &tmp_logflags, whichfork, rsvd);
5501                 logflags |= tmp_logflags;
5502                 if (error)
5503                         goto error0;
5504                 bno = del.br_startoff - 1;
5505 nodelete:
5506                 lastx = ifp->if_lastex;
5507                 /*
5508                  * If not done go on to the next (previous) record.
5509                  * Reset ep in case the extents array was re-alloced.
5510                  */
5511                 ep = &ifp->if_u1.if_extents[lastx];
5512                 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5513                         if (lastx >= XFS_IFORK_NEXTENTS(ip, whichfork) ||
5514                             xfs_bmbt_get_startoff(ep) > bno) {
5515                                 lastx--;
5516                                 ep--;
5517                         }
5518                         if (lastx >= 0)
5519                                 xfs_bmbt_get_all(ep, &got);
5520                         extno++;
5521                 }
5522         }
5523         ifp->if_lastex = lastx;
5524         *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5525         ASSERT(ifp->if_ext_max ==
5526                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5527         /*
5528          * Convert to a btree if necessary.
5529          */
5530         if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5531             XFS_IFORK_NEXTENTS(ip, whichfork) > ifp->if_ext_max) {
5532                 ASSERT(cur == NULL);
5533                 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5534                         &cur, 0, &tmp_logflags, whichfork);
5535                 logflags |= tmp_logflags;
5536                 if (error)
5537                         goto error0;
5538         }
5539         /*
5540          * transform from btree to extents, give it cur
5541          */
5542         else if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
5543                  XFS_IFORK_NEXTENTS(ip, whichfork) <= ifp->if_ext_max) {
5544                 ASSERT(cur != NULL);
5545                 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5546                         whichfork);
5547                 logflags |= tmp_logflags;
5548                 if (error)
5549                         goto error0;
5550         }
5551         /*
5552          * transform from extents to local?
5553          */
5554         ASSERT(ifp->if_ext_max ==
5555                XFS_IFORK_SIZE(ip, whichfork) / (uint)sizeof(xfs_bmbt_rec_t));
5556         error = 0;
5557 error0:
5558         /*
5559          * Log everything.  Do this after conversion, there's no point in
5560          * logging the extent list if we've converted to btree format.
5561          */
5562         if ((logflags & XFS_ILOG_FEXT(whichfork)) &&
5563             XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5564                 logflags &= ~XFS_ILOG_FEXT(whichfork);
5565         else if ((logflags & XFS_ILOG_FBROOT(whichfork)) &&
5566                  XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5567                 logflags &= ~XFS_ILOG_FBROOT(whichfork);
5568         /*
5569          * Log inode even in the error case, if the transaction
5570          * is dirty we'll need to shut down the filesystem.
5571          */
5572         if (logflags)
5573                 xfs_trans_log_inode(tp, ip, logflags);
5574         if (cur) {
5575                 if (!error) {
5576                         *firstblock = cur->bc_private.b.firstblock;
5577                         cur->bc_private.b.allocated = 0;
5578                 }
5579                 xfs_btree_del_cursor(cur,
5580                         error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5581         }
5582         return error;
5583 }
5584
5585 /*
5586  * Fcntl interface to xfs_bmapi.
5587  */
5588 int                                             /* error code */
5589 xfs_getbmap(
5590         bhv_desc_t              *bdp,           /* XFS behavior descriptor*/
5591         struct getbmap          *bmv,           /* user bmap structure */
5592         void                    __user *ap,     /* pointer to user's array */
5593         int                     interface)      /* interface flags */
5594 {
5595         __int64_t               bmvend;         /* last block requested */
5596         int                     error;          /* return value */
5597         __int64_t               fixlen;         /* length for -1 case */
5598         int                     i;              /* extent number */
5599         xfs_inode_t             *ip;            /* xfs incore inode pointer */
5600         vnode_t                 *vp;            /* corresponding vnode */
5601         int                     lock;           /* lock state */
5602         xfs_bmbt_irec_t         *map;           /* buffer for user's data */
5603         xfs_mount_t             *mp;            /* file system mount point */
5604         int                     nex;            /* # of user extents can do */
5605         int                     nexleft;        /* # of user extents left */
5606         int                     subnex;         /* # of bmapi's can do */
5607         int                     nmap;           /* number of map entries */
5608         struct getbmap          out;            /* output structure */
5609         int                     whichfork;      /* data or attr fork */
5610         int                     prealloced;     /* this is a file with
5611                                                  * preallocated data space */
5612         int                     sh_unwritten;   /* true, if unwritten */
5613                                                 /* extents listed separately */
5614         int                     bmapi_flags;    /* flags for xfs_bmapi */
5615         __int32_t               oflags;         /* getbmapx bmv_oflags field */
5616
5617         vp = BHV_TO_VNODE(bdp);
5618         ip = XFS_BHVTOI(bdp);
5619         mp = ip->i_mount;
5620
5621         whichfork = interface & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
5622         sh_unwritten = (interface & BMV_IF_PREALLOC) != 0;
5623
5624         /*      If the BMV_IF_NO_DMAPI_READ interface bit specified, do not
5625          *      generate a DMAPI read event.  Otherwise, if the DM_EVENT_READ
5626          *      bit is set for the file, generate a read event in order
5627          *      that the DMAPI application may do its thing before we return
5628          *      the extents.  Usually this means restoring user file data to
5629          *      regions of the file that look like holes.
5630          *
5631          *      The "old behavior" (from XFS_IOC_GETBMAP) is to not specify
5632          *      BMV_IF_NO_DMAPI_READ so that read events are generated.
5633          *      If this were not true, callers of ioctl( XFS_IOC_GETBMAP )
5634          *      could misinterpret holes in a DMAPI file as true holes,
5635          *      when in fact they may represent offline user data.
5636          */
5637         if (   (interface & BMV_IF_NO_DMAPI_READ) == 0
5638             && DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ)
5639             && whichfork == XFS_DATA_FORK) {
5640
5641                 error = XFS_SEND_DATA(mp, DM_EVENT_READ, vp, 0, 0, 0, NULL);
5642                 if (error)
5643                         return XFS_ERROR(error);
5644         }
5645
5646         if (whichfork == XFS_ATTR_FORK) {
5647                 if (XFS_IFORK_Q(ip)) {
5648                         if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
5649                             ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
5650                             ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
5651                                 return XFS_ERROR(EINVAL);
5652                 } else if (unlikely(
5653                            ip->i_d.di_aformat != 0 &&
5654                            ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
5655                         XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
5656                                          ip->i_mount);
5657                         return XFS_ERROR(EFSCORRUPTED);
5658                 }
5659         } else if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
5660                    ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
5661                    ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
5662                 return XFS_ERROR(EINVAL);
5663         if (whichfork == XFS_DATA_FORK) {
5664                 if (ip->i_d.di_flags & XFS_DIFLAG_PREALLOC) {
5665                         prealloced = 1;
5666                         fixlen = XFS_MAXIOFFSET(mp);
5667                 } else {
5668                         prealloced = 0;
5669                         fixlen = ip->i_d.di_size;
5670                 }
5671         } else {
5672                 prealloced = 0;
5673                 fixlen = 1LL << 32;
5674         }
5675
5676         if (bmv->bmv_length == -1) {
5677                 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
5678                 bmv->bmv_length = MAX( (__int64_t)(fixlen - bmv->bmv_offset),
5679                                         (__int64_t)0);
5680         } else if (bmv->bmv_length < 0)
5681                 return XFS_ERROR(EINVAL);
5682         if (bmv->bmv_length == 0) {
5683                 bmv->bmv_entries = 0;
5684                 return 0;
5685         }
5686         nex = bmv->bmv_count - 1;
5687         if (nex <= 0)
5688                 return XFS_ERROR(EINVAL);
5689         bmvend = bmv->bmv_offset + bmv->bmv_length;
5690
5691         xfs_ilock(ip, XFS_IOLOCK_SHARED);
5692
5693         if (whichfork == XFS_DATA_FORK && ip->i_delayed_blks) {
5694                 /* xfs_fsize_t last_byte = xfs_file_last_byte(ip); */
5695                 VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1, 0, FI_REMAPF, error);
5696         }
5697
5698         ASSERT(whichfork == XFS_ATTR_FORK || ip->i_delayed_blks == 0);
5699
5700         lock = xfs_ilock_map_shared(ip);
5701
5702         /*
5703          * Don't let nex be bigger than the number of extents
5704          * we can have assuming alternating holes and real extents.
5705          */
5706         if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
5707                 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
5708
5709         bmapi_flags = XFS_BMAPI_AFLAG(whichfork) |
5710                         ((sh_unwritten) ? 0 : XFS_BMAPI_IGSTATE);
5711
5712         /*
5713          * Allocate enough space to handle "subnex" maps at a time.
5714          */
5715         subnex = 16;
5716         map = kmem_alloc(subnex * sizeof(*map), KM_SLEEP);
5717
5718         bmv->bmv_entries = 0;
5719
5720         if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0) {
5721                 error = 0;
5722                 goto unlock_and_return;
5723         }
5724
5725         nexleft = nex;
5726
5727         do {
5728                 nmap = (nexleft > subnex) ? subnex : nexleft;
5729                 error = xfs_bmapi(NULL, ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
5730                                   XFS_BB_TO_FSB(mp, bmv->bmv_length),
5731                                   bmapi_flags, NULL, 0, map, &nmap, NULL);
5732                 if (error)
5733                         goto unlock_and_return;
5734                 ASSERT(nmap <= subnex);
5735
5736                 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
5737                         nexleft--;
5738                         oflags = (map[i].br_state == XFS_EXT_UNWRITTEN) ?
5739                                         BMV_OF_PREALLOC : 0;
5740                         out.bmv_offset = XFS_FSB_TO_BB(mp, map[i].br_startoff);
5741                         out.bmv_length = XFS_FSB_TO_BB(mp, map[i].br_blockcount);
5742                         ASSERT(map[i].br_startblock != DELAYSTARTBLOCK);
5743                         if (prealloced &&
5744                             map[i].br_startblock == HOLESTARTBLOCK &&
5745                             out.bmv_offset + out.bmv_length == bmvend) {
5746                                 /*
5747                                  * came to hole at end of file
5748                                  */
5749                                 goto unlock_and_return;
5750                         } else {
5751                                 out.bmv_block =
5752                                     (map[i].br_startblock == HOLESTARTBLOCK) ?
5753                                         -1 :
5754                                         XFS_FSB_TO_DB(ip, map[i].br_startblock);
5755
5756                                 /* return either getbmap/getbmapx structure. */
5757                                 if (interface & BMV_IF_EXTENDED) {
5758                                         struct  getbmapx        outx;
5759
5760                                         GETBMAP_CONVERT(out,outx);
5761                                         outx.bmv_oflags = oflags;
5762                                         outx.bmv_unused1 = outx.bmv_unused2 = 0;
5763                                         if (copy_to_user(ap, &outx,
5764                                                         sizeof(outx))) {
5765                                                 error = XFS_ERROR(EFAULT);
5766                                                 goto unlock_and_return;
5767                                         }
5768                                 } else {
5769                                         if (copy_to_user(ap, &out,
5770                                                         sizeof(out))) {
5771                                                 error = XFS_ERROR(EFAULT);
5772                                                 goto unlock_and_return;
5773                                         }
5774                                 }
5775                                 bmv->bmv_offset =
5776                                         out.bmv_offset + out.bmv_length;
5777                                 bmv->bmv_length = MAX((__int64_t)0,
5778                                         (__int64_t)(bmvend - bmv->bmv_offset));
5779                                 bmv->bmv_entries++;
5780                                 ap = (interface & BMV_IF_EXTENDED) ?
5781                                                 (void __user *)
5782                                         ((struct getbmapx __user *)ap + 1) :
5783                                                 (void __user *)
5784                                         ((struct getbmap __user *)ap + 1);
5785                         }
5786                 }
5787         } while (nmap && nexleft && bmv->bmv_length);
5788
5789 unlock_and_return:
5790         xfs_iunlock_map_shared(ip, lock);
5791         xfs_iunlock(ip, XFS_IOLOCK_SHARED);
5792
5793         kmem_free(map, subnex * sizeof(*map));
5794
5795         return error;
5796 }
5797
5798 /*
5799  * Check the last inode extent to determine whether this allocation will result
5800  * in blocks being allocated at the end of the file. When we allocate new data
5801  * blocks at the end of the file which do not start at the previous data block,
5802  * we will try to align the new blocks at stripe unit boundaries.
5803  */
5804 STATIC int                              /* error */
5805 xfs_bmap_isaeof(
5806         xfs_inode_t     *ip,            /* incore inode pointer */
5807         xfs_fileoff_t   off,            /* file offset in fsblocks */
5808         int             whichfork,      /* data or attribute fork */
5809         char            *aeof)          /* return value */
5810 {
5811         int             error;          /* error return value */
5812         xfs_ifork_t     *ifp;           /* inode fork pointer */
5813         xfs_bmbt_rec_t  *lastrec;       /* extent list entry pointer */
5814         xfs_extnum_t    nextents;       /* size of extent list */
5815         xfs_bmbt_irec_t s;              /* expanded extent list entry */
5816
5817         ASSERT(whichfork == XFS_DATA_FORK);
5818         ifp = XFS_IFORK_PTR(ip, whichfork);
5819         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5820             (error = xfs_iread_extents(NULL, ip, whichfork)))
5821                 return error;
5822         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5823         if (nextents == 0) {
5824                 *aeof = 1;
5825                 return 0;
5826         }
5827         /*
5828          * Go to the last extent
5829          */
5830         lastrec = &ifp->if_u1.if_extents[nextents - 1];
5831         xfs_bmbt_get_all(lastrec, &s);
5832         /*
5833          * Check we are allocating in the last extent (for delayed allocations)
5834          * or past the last extent for non-delayed allocations.
5835          */
5836         *aeof = (off >= s.br_startoff &&
5837                  off < s.br_startoff + s.br_blockcount &&
5838                  ISNULLSTARTBLOCK(s.br_startblock)) ||
5839                 off >= s.br_startoff + s.br_blockcount;
5840         return 0;
5841 }
5842
5843 /*
5844  * Check if the endoff is outside the last extent. If so the caller will grow
5845  * the allocation to a stripe unit boundary.
5846  */
5847 int                                     /* error */
5848 xfs_bmap_eof(
5849         xfs_inode_t     *ip,            /* incore inode pointer */
5850         xfs_fileoff_t   endoff,         /* file offset in fsblocks */
5851         int             whichfork,      /* data or attribute fork */
5852         int             *eof)           /* result value */
5853 {
5854         xfs_fsblock_t   blockcount;     /* extent block count */
5855         int             error;          /* error return value */
5856         xfs_ifork_t     *ifp;           /* inode fork pointer */
5857         xfs_bmbt_rec_t  *lastrec;       /* extent list entry pointer */
5858         xfs_extnum_t    nextents;       /* size of extent list */
5859         xfs_fileoff_t   startoff;       /* extent starting file offset */
5860
5861         ASSERT(whichfork == XFS_DATA_FORK);
5862         ifp = XFS_IFORK_PTR(ip, whichfork);
5863         if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5864             (error = xfs_iread_extents(NULL, ip, whichfork)))
5865                 return error;
5866         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5867         if (nextents == 0) {
5868                 *eof = 1;
5869                 return 0;
5870         }
5871         /*
5872          * Go to the last extent
5873          */
5874         lastrec = &ifp->if_u1.if_extents[nextents - 1];
5875         startoff = xfs_bmbt_get_startoff(lastrec);
5876         blockcount = xfs_bmbt_get_blockcount(lastrec);
5877         *eof = endoff >= startoff + blockcount;
5878         return 0;
5879 }
5880
5881 #ifdef DEBUG
5882 /*
5883  * Check that the extents list for the inode ip is in the right order.
5884  */
5885 STATIC void
5886 xfs_bmap_check_extents(
5887         xfs_inode_t             *ip,            /* incore inode pointer */
5888         int                     whichfork)      /* data or attr fork */
5889 {
5890         xfs_bmbt_rec_t          *base;          /* base of extents list */
5891         xfs_bmbt_rec_t          *ep;            /* current extent entry */
5892         xfs_ifork_t             *ifp;           /* inode fork pointer */
5893         xfs_extnum_t            nextents;       /* number of extents in list */
5894
5895         ifp = XFS_IFORK_PTR(ip, whichfork);
5896         ASSERT(ifp->if_flags & XFS_IFEXTENTS);
5897         base = ifp->if_u1.if_extents;
5898         nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5899         for (ep = base; ep < &base[nextents - 1]; ep++) {
5900                 xfs_btree_check_rec(XFS_BTNUM_BMAP, (void *)ep,
5901                         (void *)(ep + 1));
5902         }
5903 }
5904
5905 STATIC
5906 xfs_buf_t *
5907 xfs_bmap_get_bp(
5908         xfs_btree_cur_t         *cur,
5909         xfs_fsblock_t           bno)
5910 {
5911         int i;
5912         xfs_buf_t *bp;
5913
5914         if (!cur)
5915                 return(NULL);
5916
5917         bp = NULL;
5918         for(i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
5919                 bp = cur->bc_bufs[i];
5920                 if (!bp) break;
5921                 if (XFS_BUF_ADDR(bp) == bno)
5922                         break;  /* Found it */
5923         }
5924         if (i == XFS_BTREE_MAXLEVELS)
5925                 bp = NULL;
5926
5927         if (!bp) { /* Chase down all the log items to see if the bp is there */
5928                 xfs_log_item_chunk_t    *licp;
5929                 xfs_trans_t             *tp;
5930
5931                 tp = cur->bc_tp;
5932                 licp = &tp->t_items;
5933                 while (!bp && licp != NULL) {
5934                         if (XFS_LIC_ARE_ALL_FREE(licp)) {
5935                                 licp = licp->lic_next;
5936                                 continue;
5937                         }
5938                         for (i = 0; i < licp->lic_unused; i++) {
5939                                 xfs_log_item_desc_t     *lidp;
5940                                 xfs_log_item_t          *lip;
5941                                 xfs_buf_log_item_t      *bip;
5942                                 xfs_buf_t               *lbp;
5943
5944                                 if (XFS_LIC_ISFREE(licp, i)) {
5945                                         continue;
5946                                 }
5947
5948                                 lidp = XFS_LIC_SLOT(licp, i);
5949                                 lip = lidp->lid_item;
5950                                 if (lip->li_type != XFS_LI_BUF)
5951                                         continue;
5952
5953                                 bip = (xfs_buf_log_item_t *)lip;
5954                                 lbp = bip->bli_buf;
5955
5956                                 if (XFS_BUF_ADDR(lbp) == bno) {
5957                                         bp = lbp;
5958                                         break; /* Found it */
5959                                 }
5960                         }
5961                         licp = licp->lic_next;
5962                 }
5963         }
5964         return(bp);
5965 }
5966
5967 void
5968 xfs_check_block(
5969         xfs_bmbt_block_t        *block,
5970         xfs_mount_t             *mp,
5971         int                     root,
5972         short                   sz)
5973 {
5974         int                     i, j, dmxr;
5975         xfs_bmbt_ptr_t          *pp, *thispa;   /* pointer to block address */
5976         xfs_bmbt_key_t          *prevp, *keyp;
5977
5978         ASSERT(INT_GET(block->bb_level, ARCH_CONVERT) > 0);
5979
5980         prevp = NULL;
5981         for( i = 1; i <= INT_GET(block->bb_numrecs, ARCH_CONVERT);i++) {
5982                 dmxr = mp->m_bmap_dmxr[0];
5983
5984                 if (root) {
5985                         keyp = XFS_BMAP_BROOT_KEY_ADDR(block, i, sz);
5986                 } else {
5987                         keyp = XFS_BTREE_KEY_ADDR(mp->m_sb.sb_blocksize,
5988                                 xfs_bmbt, block, i, dmxr);
5989                 }
5990
5991                 if (prevp) {
5992                         xfs_btree_check_key(XFS_BTNUM_BMAP, prevp, keyp);
5993                 }
5994                 prevp = keyp;
5995
5996                 /*
5997                  * Compare the block numbers to see if there are dups.
5998                  */
5999
6000                 if (root) {
6001                         pp = XFS_BMAP_BROOT_PTR_ADDR(block, i, sz);
6002                 } else {
6003                         pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize,
6004                                 xfs_bmbt, block, i, dmxr);
6005                 }
6006                 for (j = i+1; j <= INT_GET(block->bb_numrecs, ARCH_CONVERT); j++) {
6007                         if (root) {
6008                                 thispa = XFS_BMAP_BROOT_PTR_ADDR(block, j, sz);
6009                         } else {
6010                                 thispa = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize,
6011                                         xfs_bmbt, block, j, dmxr);
6012                         }
6013                         if (INT_GET(*thispa, ARCH_CONVERT) ==
6014                             INT_GET(*pp, ARCH_CONVERT)) {
6015                                 cmn_err(CE_WARN, "%s: thispa(%d) == pp(%d) %Ld",
6016                                         __FUNCTION__, j, i,
6017                                         INT_GET(*thispa, ARCH_CONVERT));
6018                                 panic("%s: ptrs are equal in node\n",
6019                                         __FUNCTION__);
6020                         }
6021                 }
6022         }
6023 }
6024
6025 /*
6026  * Check that the extents for the inode ip are in the right order in all
6027  * btree leaves.
6028  */
6029
6030 STATIC void
6031 xfs_bmap_check_leaf_extents(
6032         xfs_btree_cur_t         *cur,   /* btree cursor or null */
6033         xfs_inode_t             *ip,            /* incore inode pointer */
6034         int                     whichfork)      /* data or attr fork */
6035 {
6036         xfs_bmbt_block_t        *block; /* current btree block */
6037         xfs_fsblock_t           bno;    /* block # of "block" */
6038         xfs_buf_t               *bp;    /* buffer for "block" */
6039         int                     error;  /* error return value */
6040         xfs_extnum_t            i=0;    /* index into the extents list */
6041         xfs_ifork_t             *ifp;   /* fork structure */
6042         int                     level;  /* btree level, for checking */
6043         xfs_mount_t             *mp;    /* file system mount structure */
6044         xfs_bmbt_ptr_t          *pp;    /* pointer to block address */
6045         xfs_bmbt_rec_t          *ep, *lastp;    /* extent pointers in block entry */
6046         int                     bp_release = 0;
6047
6048         if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
6049                 return;
6050         }
6051
6052         bno = NULLFSBLOCK;
6053         mp = ip->i_mount;
6054         ifp = XFS_IFORK_PTR(ip, whichfork);
6055         block = ifp->if_broot;
6056         /*
6057          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6058          */
6059         ASSERT(INT_GET(block->bb_level, ARCH_CONVERT) > 0);
6060         level = INT_GET(block->bb_level, ARCH_CONVERT);
6061         xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
6062         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6063         ASSERT(INT_GET(*pp, ARCH_CONVERT) != NULLDFSBNO);
6064         ASSERT(XFS_FSB_TO_AGNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agcount);
6065         ASSERT(XFS_FSB_TO_AGBNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agblocks);
6066         bno = INT_GET(*pp, ARCH_CONVERT);
6067         /*
6068          * Go down the tree until leaf level is reached, following the first
6069          * pointer (leftmost) at each level.
6070          */
6071         while (level-- > 0) {
6072                 /* See if buf is in cur first */
6073                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6074                 if (bp) {
6075                         bp_release = 0;
6076                 } else {
6077                         bp_release = 1;
6078                 }
6079                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6080                                 XFS_BMAP_BTREE_REF)))
6081                         goto error_norelse;
6082                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6083                 XFS_WANT_CORRUPTED_GOTO(
6084                         XFS_BMAP_SANITY_CHECK(mp, block, level),
6085                         error0);
6086                 if (level == 0)
6087                         break;
6088
6089                 /*
6090                  * Check this block for basic sanity (increasing keys and
6091                  * no duplicate blocks).
6092                  */
6093
6094                 xfs_check_block(block, mp, 0, 0);
6095                 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt, block,
6096                         1, mp->m_bmap_dmxr[1]);
6097                 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, INT_GET(*pp, ARCH_CONVERT)), error0);
6098                 bno = INT_GET(*pp, ARCH_CONVERT);
6099                 if (bp_release) {
6100                         bp_release = 0;
6101                         xfs_trans_brelse(NULL, bp);
6102                 }
6103         }
6104
6105         /*
6106          * Here with bp and block set to the leftmost leaf node in the tree.
6107          */
6108         i = 0;
6109
6110         /*
6111          * Loop over all leaf nodes checking that all extents are in the right order.
6112          */
6113         lastp = NULL;
6114         for (;;) {
6115                 xfs_bmbt_rec_t  *frp;
6116                 xfs_fsblock_t   nextbno;
6117                 xfs_extnum_t    num_recs;
6118
6119
6120                 num_recs = INT_GET(block->bb_numrecs, ARCH_CONVERT);
6121
6122                 /*
6123                  * Read-ahead the next leaf block, if any.
6124                  */
6125
6126                 nextbno = INT_GET(block->bb_rightsib, ARCH_CONVERT);
6127
6128                 /*
6129                  * Check all the extents to make sure they are OK.
6130                  * If we had a previous block, the last entry should
6131                  * conform with the first entry in this one.
6132                  */
6133
6134                 frp = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize, xfs_bmbt,
6135                         block, 1, mp->m_bmap_dmxr[0]);
6136
6137                 for (ep = frp;ep < frp + (num_recs - 1); ep++) {
6138                         if (lastp) {
6139                                 xfs_btree_check_rec(XFS_BTNUM_BMAP,
6140                                         (void *)lastp, (void *)ep);
6141                         }
6142                         xfs_btree_check_rec(XFS_BTNUM_BMAP, (void *)ep,
6143                                 (void *)(ep + 1));
6144                 }
6145                 lastp = frp + num_recs - 1; /* For the next iteration */
6146
6147                 i += num_recs;
6148                 if (bp_release) {
6149                         bp_release = 0;
6150                         xfs_trans_brelse(NULL, bp);
6151                 }
6152                 bno = nextbno;
6153                 /*
6154                  * If we've reached the end, stop.
6155                  */
6156                 if (bno == NULLFSBLOCK)
6157                         break;
6158
6159                 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
6160                 if (bp) {
6161                         bp_release = 0;
6162                 } else {
6163                         bp_release = 1;
6164                 }
6165                 if (!bp && (error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
6166                                 XFS_BMAP_BTREE_REF)))
6167                         goto error_norelse;
6168                 block = XFS_BUF_TO_BMBT_BLOCK(bp);
6169         }
6170         if (bp_release) {
6171                 bp_release = 0;
6172                 xfs_trans_brelse(NULL, bp);
6173         }
6174         return;
6175
6176 error0:
6177         cmn_err(CE_WARN, "%s: at error0", __FUNCTION__);
6178         if (bp_release)
6179                 xfs_trans_brelse(NULL, bp);
6180 error_norelse:
6181         cmn_err(CE_WARN, "%s: BAD after btree leaves for %d extents",
6182                 __FUNCTION__, i);
6183         panic("%s: CORRUPTED BTREE OR SOMETHING", __FUNCTION__);
6184         return;
6185 }
6186 #endif
6187
6188 /*
6189  * Count fsblocks of the given fork.
6190  */
6191 int                                             /* error */
6192 xfs_bmap_count_blocks(
6193         xfs_trans_t             *tp,            /* transaction pointer */
6194         xfs_inode_t             *ip,            /* incore inode */
6195         int                     whichfork,      /* data or attr fork */
6196         int                     *count)         /* out: count of blocks */
6197 {
6198         xfs_bmbt_block_t        *block; /* current btree block */
6199         xfs_fsblock_t           bno;    /* block # of "block" */
6200         xfs_ifork_t             *ifp;   /* fork structure */
6201         int                     level;  /* btree level, for checking */
6202         xfs_mount_t             *mp;    /* file system mount structure */
6203         xfs_bmbt_ptr_t          *pp;    /* pointer to block address */
6204
6205         bno = NULLFSBLOCK;
6206         mp = ip->i_mount;
6207         ifp = XFS_IFORK_PTR(ip, whichfork);
6208         if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
6209                 if (unlikely(xfs_bmap_count_leaves(ifp->if_u1.if_extents,
6210                         ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
6211                         count) < 0)) {
6212                         XFS_ERROR_REPORT("xfs_bmap_count_blocks(1)",
6213                                          XFS_ERRLEVEL_LOW, mp);
6214                         return XFS_ERROR(EFSCORRUPTED);
6215                 }
6216                 return 0;
6217         }
6218
6219         /*
6220          * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
6221          */
6222         block = ifp->if_broot;
6223         ASSERT(INT_GET(block->bb_level, ARCH_CONVERT) > 0);
6224         level = INT_GET(block->bb_level, ARCH_CONVERT);
6225         pp = XFS_BMAP_BROOT_PTR_ADDR(block, 1, ifp->if_broot_bytes);
6226         ASSERT(INT_GET(*pp, ARCH_CONVERT) != NULLDFSBNO);
6227         ASSERT(XFS_FSB_TO_AGNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agcount);
6228         ASSERT(XFS_FSB_TO_AGBNO(mp, INT_GET(*pp, ARCH_CONVERT)) < mp->m_sb.sb_agblocks);
6229         bno = INT_GET(*pp, ARCH_CONVERT);
6230
6231         if (unlikely(xfs_bmap_count_tree(mp, tp, bno, level, count) < 0)) {
6232                 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
6233                                  mp);
6234                 return XFS_ERROR(EFSCORRUPTED);
6235         }
6236
6237         return 0;
6238 }
6239
6240 /*
6241  * Recursively walks each level of a btree
6242  * to count total fsblocks is use.
6243  */
6244 int                                     /* error */
6245 xfs_bmap_count_tree(
6246         xfs_mount_t     *mp,            /* file system mount point */
6247         xfs_trans_t     *tp,            /* transaction pointer */
6248         xfs_fsblock_t   blockno,        /* file system block number */
6249         int             levelin,        /* level in btree */
6250         int             *count)         /* Count of blocks */
6251 {
6252         int                     error;
6253         xfs_buf_t               *bp, *nbp;
6254         int                     level = levelin;
6255         xfs_bmbt_ptr_t          *pp;
6256         xfs_fsblock_t           bno = blockno;
6257         xfs_fsblock_t           nextbno;
6258         xfs_bmbt_block_t        *block, *nextblock;
6259         int                     numrecs;
6260         xfs_bmbt_rec_t          *frp;
6261
6262         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF)))
6263                 return error;
6264         *count += 1;
6265         block = XFS_BUF_TO_BMBT_BLOCK(bp);
6266
6267         if (--level) {
6268                 /* Not at node above leafs, count this level of nodes */
6269                 nextbno = INT_GET(block->bb_rightsib, ARCH_CONVERT);
6270                 while (nextbno != NULLFSBLOCK) {
6271                         if ((error = xfs_btree_read_bufl(mp, tp, nextbno,
6272                                 0, &nbp, XFS_BMAP_BTREE_REF)))
6273                                 return error;
6274                         *count += 1;
6275                         nextblock = XFS_BUF_TO_BMBT_BLOCK(nbp);
6276                         nextbno = INT_GET(nextblock->bb_rightsib, ARCH_CONVERT);
6277                         xfs_trans_brelse(tp, nbp);
6278                 }
6279
6280                 /* Dive to the next level */
6281                 pp = XFS_BTREE_PTR_ADDR(mp->m_sb.sb_blocksize,
6282                         xfs_bmbt, block, 1, mp->m_bmap_dmxr[1]);
6283                 bno = INT_GET(*pp, ARCH_CONVERT);
6284                 if (unlikely((error =
6285                      xfs_bmap_count_tree(mp, tp, bno, level, count)) < 0)) {
6286                         xfs_trans_brelse(tp, bp);
6287                         XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
6288                                          XFS_ERRLEVEL_LOW, mp);
6289                         return XFS_ERROR(EFSCORRUPTED);
6290                 }
6291                 xfs_trans_brelse(tp, bp);
6292         } else {
6293                 /* count all level 1 nodes and their leaves */
6294                 for (;;) {
6295                         nextbno = INT_GET(block->bb_rightsib, ARCH_CONVERT);
6296                         numrecs = INT_GET(block->bb_numrecs, ARCH_CONVERT);
6297                         frp = XFS_BTREE_REC_ADDR(mp->m_sb.sb_blocksize,
6298                                 xfs_bmbt, block, 1, mp->m_bmap_dmxr[0]);
6299                         if (unlikely(xfs_bmap_count_leaves(frp, numrecs, count) < 0)) {
6300                                 xfs_trans_brelse(tp, bp);
6301                                 XFS_ERROR_REPORT("xfs_bmap_count_tree(2)",
6302                                                  XFS_ERRLEVEL_LOW, mp);
6303                                 return XFS_ERROR(EFSCORRUPTED);
6304                         }
6305                         xfs_trans_brelse(tp, bp);
6306                         if (nextbno == NULLFSBLOCK)
6307                                 break;
6308                         bno = nextbno;
6309                         if ((error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
6310                                 XFS_BMAP_BTREE_REF)))
6311                                 return error;
6312                         *count += 1;
6313                         block = XFS_BUF_TO_BMBT_BLOCK(bp);
6314                 }
6315         }
6316         return 0;
6317 }
6318
6319 /*
6320  * Count leaf blocks given a pointer to an extent list.
6321  */
6322 int
6323 xfs_bmap_count_leaves(
6324         xfs_bmbt_rec_t          *frp,
6325         int                     numrecs,
6326         int                     *count)
6327 {
6328         int             b;
6329
6330         for ( b = 1; b <= numrecs; b++, frp++)
6331                 *count += xfs_bmbt_disk_get_blockcount(frp);
6332         return 0;
6333 }