fs/xfs/xfs_itable.c

   1 /*
   2  * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_bit.h"
  22 #include "xfs_log.h"
  23 #include "xfs_inum.h"
  24 #include "xfs_trans.h"
  25 #include "xfs_sb.h"
  26 #include "xfs_ag.h"
  27 #include "xfs_dir2.h"
  28 #include "xfs_dmapi.h"
  29 #include "xfs_mount.h"
  30 #include "xfs_bmap_btree.h"
  31 #include "xfs_alloc_btree.h"
  32 #include "xfs_ialloc_btree.h"
  33 #include "xfs_dir2_sf.h"
  34 #include "xfs_attr_sf.h"
  35 #include "xfs_dinode.h"
  36 #include "xfs_inode.h"
  37 #include "xfs_ialloc.h"
  38 #include "xfs_itable.h"
  39 #include "xfs_error.h"
  40 #include "xfs_btree.h"
  41
  42 STATIC int
  43 xfs_bulkstat_one_iget(
  44         xfs_mount_t     *mp,            /* mount point for filesystem */
  45         xfs_ino_t       ino,            /* inode number to get data for */
  46         xfs_daddr_t     bno,            /* starting bno of inode cluster */
  47         xfs_bstat_t     *buf,           /* return buffer */
  48         int             *stat)          /* BULKSTAT_RV_... */
  49 {
  50         xfs_dinode_core_t *dic;         /* dinode core info pointer */
  51         xfs_inode_t     *ip;            /* incore inode pointer */
  52         bhv_vnode_t     *vp;
  53         int             error;
  54
  55         error = xfs_iget(mp, NULL, ino, 0, XFS_ILOCK_SHARED, &ip, bno);
  56         if (error) {
  57                 *stat = BULKSTAT_RV_NOTHING;
  58                 return error;
  59         }
  60
  61         ASSERT(ip != NULL);
  62         ASSERT(ip->i_blkno != (xfs_daddr_t)0);
  63         if (ip->i_d.di_mode == 0) {
  64                 *stat = BULKSTAT_RV_NOTHING;
  65                 error = XFS_ERROR(ENOENT);
  66                 goto out_iput;
  67         }
  68
  69         vp = XFS_ITOV(ip);
  70         dic = &ip->i_d;
  71
  72         /* xfs_iget returns the following without needing
  73          * further change.
  74          */
  75         buf->bs_nlink = dic->di_nlink;
  76         buf->bs_projid = dic->di_projid;
  77         buf->bs_ino = ino;
  78         buf->bs_mode = dic->di_mode;
  79         buf->bs_uid = dic->di_uid;
  80         buf->bs_gid = dic->di_gid;
  81         buf->bs_size = dic->di_size;
  82         vn_atime_to_bstime(vp, &buf->bs_atime);
  83         buf->bs_mtime.tv_sec = dic->di_mtime.t_sec;
  84         buf->bs_mtime.tv_nsec = dic->di_mtime.t_nsec;
  85         buf->bs_ctime.tv_sec = dic->di_ctime.t_sec;
  86         buf->bs_ctime.tv_nsec = dic->di_ctime.t_nsec;
  87         buf->bs_xflags = xfs_ip2xflags(ip);
  88         buf->bs_extsize = dic->di_extsize << mp->m_sb.sb_blocklog;
  89         buf->bs_extents = dic->di_nextents;
  90         buf->bs_gen = dic->di_gen;
  91         memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
  92         buf->bs_dmevmask = dic->di_dmevmask;
  93         buf->bs_dmstate = dic->di_dmstate;
  94         buf->bs_aextents = dic->di_anextents;
  95
  96         switch (dic->di_format) {
  97         case XFS_DINODE_FMT_DEV:
  98                 buf->bs_rdev = ip->i_df.if_u2.if_rdev;
  99                 buf->bs_blksize = BLKDEV_IOSIZE;
 100                 buf->bs_blocks = 0;
 101                 break;
 102         case XFS_DINODE_FMT_LOCAL:
 103         case XFS_DINODE_FMT_UUID:
 104                 buf->bs_rdev = 0;
 105                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 106                 buf->bs_blocks = 0;
 107                 break;
 108         case XFS_DINODE_FMT_EXTENTS:
 109         case XFS_DINODE_FMT_BTREE:
 110                 buf->bs_rdev = 0;
 111                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 112                 buf->bs_blocks = dic->di_nblocks + ip->i_delayed_blks;
 113                 break;
 114         }
 115
 116  out_iput:
 117         xfs_iput(ip, XFS_ILOCK_SHARED);
 118         return error;
 119 }
 120
 121 STATIC int
 122 xfs_bulkstat_one_dinode(
 123         xfs_mount_t     *mp,            /* mount point for filesystem */
 124         xfs_ino_t       ino,            /* inode number to get data for */
 125         xfs_dinode_t    *dip,           /* dinode inode pointer */
 126         xfs_bstat_t     *buf)           /* return buffer */
 127 {
 128         xfs_dinode_core_t *dic;         /* dinode core info pointer */
 129
 130         dic = &dip->di_core;
 131
 132         /*
 133          * The inode format changed when we moved the link count and
 134          * made it 32 bits long.  If this is an old format inode,
 135          * convert it in memory to look like a new one.  If it gets
 136          * flushed to disk we will convert back before flushing or
 137          * logging it.  We zero out the new projid field and the old link
 138          * count field.  We'll handle clearing the pad field (the remains
 139          * of the old uuid field) when we actually convert the inode to
 140          * the new format. We don't change the version number so that we
 141          * can distinguish this from a real new format inode.
 142          */
 143         if (INT_GET(dic->di_version, ARCH_CONVERT) == XFS_DINODE_VERSION_1) {
 144                 buf->bs_nlink = INT_GET(dic->di_onlink, ARCH_CONVERT);
 145                 buf->bs_projid = 0;
 146         } else {
 147                 buf->bs_nlink = INT_GET(dic->di_nlink, ARCH_CONVERT);
 148                 buf->bs_projid = INT_GET(dic->di_projid, ARCH_CONVERT);
 149         }
 150
 151         buf->bs_ino = ino;
 152         buf->bs_mode = INT_GET(dic->di_mode, ARCH_CONVERT);
 153         buf->bs_uid = INT_GET(dic->di_uid, ARCH_CONVERT);
 154         buf->bs_gid = INT_GET(dic->di_gid, ARCH_CONVERT);
 155         buf->bs_size = INT_GET(dic->di_size, ARCH_CONVERT);
 156         buf->bs_atime.tv_sec = INT_GET(dic->di_atime.t_sec, ARCH_CONVERT);
 157         buf->bs_atime.tv_nsec = INT_GET(dic->di_atime.t_nsec, ARCH_CONVERT);
 158         buf->bs_mtime.tv_sec = INT_GET(dic->di_mtime.t_sec, ARCH_CONVERT);
 159         buf->bs_mtime.tv_nsec = INT_GET(dic->di_mtime.t_nsec, ARCH_CONVERT);
 160         buf->bs_ctime.tv_sec = INT_GET(dic->di_ctime.t_sec, ARCH_CONVERT);
 161         buf->bs_ctime.tv_nsec = INT_GET(dic->di_ctime.t_nsec, ARCH_CONVERT);
 162         buf->bs_xflags = xfs_dic2xflags(dic);
 163         buf->bs_extsize = INT_GET(dic->di_extsize, ARCH_CONVERT) << mp->m_sb.sb_blocklog;
 164         buf->bs_extents = INT_GET(dic->di_nextents, ARCH_CONVERT);
 165         buf->bs_gen = INT_GET(dic->di_gen, ARCH_CONVERT);
 166         memset(buf->bs_pad, 0, sizeof(buf->bs_pad));
 167         buf->bs_dmevmask = INT_GET(dic->di_dmevmask, ARCH_CONVERT);
 168         buf->bs_dmstate = INT_GET(dic->di_dmstate, ARCH_CONVERT);
 169         buf->bs_aextents = INT_GET(dic->di_anextents, ARCH_CONVERT);
 170
 171         switch (INT_GET(dic->di_format, ARCH_CONVERT)) {
 172         case XFS_DINODE_FMT_DEV:
 173                 buf->bs_rdev = INT_GET(dip->di_u.di_dev, ARCH_CONVERT);
 174                 buf->bs_blksize = BLKDEV_IOSIZE;
 175                 buf->bs_blocks = 0;
 176                 break;
 177         case XFS_DINODE_FMT_LOCAL:
 178         case XFS_DINODE_FMT_UUID:
 179                 buf->bs_rdev = 0;
 180                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 181                 buf->bs_blocks = 0;
 182                 break;
 183         case XFS_DINODE_FMT_EXTENTS:
 184         case XFS_DINODE_FMT_BTREE:
 185                 buf->bs_rdev = 0;
 186                 buf->bs_blksize = mp->m_sb.sb_blocksize;
 187                 buf->bs_blocks = INT_GET(dic->di_nblocks, ARCH_CONVERT);
 188                 break;
 189         }
 190
 191         return 0;
 192 }
 193
 194 /*
 195  * Return stat information for one inode.
 196  * Return 0 if ok, else errno.
 197  */
 198 int                             /* error status */
 199 xfs_bulkstat_one(
 200         xfs_mount_t     *mp,            /* mount point for filesystem */
 201         xfs_ino_t       ino,            /* inode number to get data for */
 202         void            __user *buffer, /* buffer to place output in */
 203         int             ubsize,         /* size of buffer */
 204         void            *private_data,  /* my private data */
 205         xfs_daddr_t     bno,            /* starting bno of inode cluster */
 206         int             *ubused,        /* bytes used by me */
 207         void            *dibuff,        /* on-disk inode buffer */
 208         int             *stat)          /* BULKSTAT_RV_... */
 209 {
 210         xfs_bstat_t     *buf;           /* return buffer */
 211         int             error = 0;      /* error value */
 212         xfs_dinode_t    *dip;           /* dinode inode pointer */
 213
 214         dip = (xfs_dinode_t *)dibuff;
 215
 216         if (!buffer || ino == mp->m_sb.sb_rbmino || ino == mp->m_sb.sb_rsumino ||
 217             (XFS_SB_VERSION_HASQUOTA(&mp->m_sb) &&
 218              (ino == mp->m_sb.sb_uquotino || ino == mp->m_sb.sb_gquotino))) {
 219                 *stat = BULKSTAT_RV_NOTHING;
 220                 return XFS_ERROR(EINVAL);
 221         }
 222         if (ubsize < sizeof(*buf)) {
 223                 *stat = BULKSTAT_RV_NOTHING;
 224                 return XFS_ERROR(ENOMEM);
 225         }
 226
 227         buf = kmem_alloc(sizeof(*buf), KM_SLEEP);
 228
 229         if (dip == NULL) {
 230                 /* We're not being passed a pointer to a dinode.  This happens
 231                  * if BULKSTAT_FG_IGET is selected.  Do the iget.
 232                  */
 233                 error = xfs_bulkstat_one_iget(mp, ino, bno, buf, stat);
 234                 if (error)
 235                         goto out_free;
 236         } else {
 237                 xfs_bulkstat_one_dinode(mp, ino, dip, buf);
 238         }
 239
 240         if (copy_to_user(buffer, buf, sizeof(*buf)))  {
 241                 *stat = BULKSTAT_RV_NOTHING;
 242                 error =  EFAULT;
 243                 goto out_free;
 244         }
 245
 246         *stat = BULKSTAT_RV_DIDONE;
 247         if (ubused)
 248                 *ubused = sizeof(*buf);
 249
 250  out_free:
 251         kmem_free(buf, sizeof(*buf));
 252         return error;
 253 }
 254
 255 /*
 256  * Test to see whether we can use the ondisk inode directly, based
 257  * on the given bulkstat flags, filling in dipp accordingly.
 258  * Returns zero if the inode is dodgey.
 259  */
 260 STATIC int
 261 xfs_bulkstat_use_dinode(
 262         xfs_mount_t     *mp,
 263         int             flags,
 264         xfs_buf_t       *bp,
 265         int             clustidx,
 266         xfs_dinode_t    **dipp)
 267 {
 268         xfs_dinode_t    *dip;
 269         unsigned int    aformat;
 270
 271         *dipp = NULL;
 272         if (!bp || (flags & BULKSTAT_FG_IGET))
 273                 return 1;
 274         dip = (xfs_dinode_t *)
 275                         xfs_buf_offset(bp, clustidx << mp->m_sb.sb_inodelog);
 276         if (INT_GET(dip->di_core.di_magic, ARCH_CONVERT) != XFS_DINODE_MAGIC ||
 277             !XFS_DINODE_GOOD_VERSION(
 278                         INT_GET(dip->di_core.di_version, ARCH_CONVERT)))
 279                 return 0;
 280         if (flags & BULKSTAT_FG_QUICK) {
 281                 *dipp = dip;
 282                 return 1;
 283         }
 284         /* BULKSTAT_FG_INLINE: if attr fork is local, or not there, use it */
 285         aformat = INT_GET(dip->di_core.di_aformat, ARCH_CONVERT);
 286         if ((XFS_CFORK_Q(&dip->di_core) == 0) ||
 287             (aformat == XFS_DINODE_FMT_LOCAL) ||
 288             (aformat == XFS_DINODE_FMT_EXTENTS && !dip->di_core.di_anextents)) {
 289                 *dipp = dip;
 290                 return 1;
 291         }
 292         return 1;
 293 }
 294
 295 /*
 296  * Return stat information in bulk (by-inode) for the filesystem.
 297  */
 298 int                                     /* error status */
 299 xfs_bulkstat(
 300         xfs_mount_t             *mp,    /* mount point for filesystem */
 301         xfs_ino_t               *lastinop, /* last inode returned */
 302         int                     *ubcountp, /* size of buffer/count returned */
 303         bulkstat_one_pf         formatter, /* func that'd fill a single buf */
 304         void                    *private_data,/* private data for formatter */
 305         size_t                  statstruct_size, /* sizeof struct filling */
 306         char                    __user *ubuffer, /* buffer with inode stats */
 307         int                     flags,  /* defined in xfs_itable.h */
 308         int                     *done)  /* 1 if there are more stats to get */
 309 {
 310         xfs_agblock_t           agbno=0;/* allocation group block number */
 311         xfs_buf_t               *agbp;  /* agi header buffer */
 312         xfs_agi_t               *agi;   /* agi header data */
 313         xfs_agino_t             agino;  /* inode # in allocation group */
 314         xfs_agnumber_t          agno;   /* allocation group number */
 315         xfs_daddr_t             bno;    /* inode cluster start daddr */
 316         int                     chunkidx; /* current index into inode chunk */
 317         int                     clustidx; /* current index into inode cluster */
 318         xfs_btree_cur_t         *cur;   /* btree cursor for ialloc btree */
 319         int                     end_of_ag; /* set if we've seen the ag end */
 320         int                     error;  /* error code */
 321         int                     fmterror;/* bulkstat formatter result */
 322         __int32_t               gcnt;   /* current btree rec's count */
 323         xfs_inofree_t           gfree;  /* current btree rec's free mask */
 324         xfs_agino_t             gino;   /* current btree rec's start inode */
 325         int                     i;      /* loop index */
 326         int                     icount; /* count of inodes good in irbuf */
 327         int                     irbsize; /* size of irec buffer in bytes */
 328         unsigned int            kmflags; /* flags for allocating irec buffer */
 329         xfs_ino_t               ino;    /* inode number (filesystem) */
 330         xfs_inobt_rec_incore_t  *irbp;  /* current irec buffer pointer */
 331         xfs_inobt_rec_incore_t  *irbuf; /* start of irec buffer */
 332         xfs_inobt_rec_incore_t  *irbufend; /* end of good irec buffer entries */
 333         xfs_ino_t               lastino=0; /* last inode number returned */
 334         int                     nbcluster; /* # of blocks in a cluster */
 335         int                     nicluster; /* # of inodes in a cluster */
 336         int                     nimask; /* mask for inode clusters */
 337         int                     nirbuf; /* size of irbuf */
 338         int                     rval;   /* return value error code */
 339         int                     tmp;    /* result value from btree calls */
 340         int                     ubcount; /* size of user's buffer */
 341         int                     ubleft; /* bytes left in user's buffer */
 342         char                    __user *ubufp;  /* pointer into user's buffer */
 343         int                     ubelem; /* spaces used in user's buffer */
 344         int                     ubused; /* bytes used by formatter */
 345         xfs_buf_t               *bp;    /* ptr to on-disk inode cluster buf */
 346         xfs_dinode_t            *dip;   /* ptr into bp for specific inode */
 347         xfs_inode_t             *ip;    /* ptr to in-core inode struct */
 348
 349         /*
 350          * Get the last inode value, see if there's nothing to do.
 351          */
 352         ino = (xfs_ino_t)*lastinop;
 353         dip = NULL;
 354         agno = XFS_INO_TO_AGNO(mp, ino);
 355         agino = XFS_INO_TO_AGINO(mp, ino);
 356         if (agno >= mp->m_sb.sb_agcount ||
 357             ino != XFS_AGINO_TO_INO(mp, agno, agino)) {
 358                 *done = 1;
 359                 *ubcountp = 0;
 360                 return 0;
 361         }
 362         ubcount = *ubcountp; /* statstruct's */
 363         ubleft = ubcount * statstruct_size; /* bytes */
 364         *ubcountp = ubelem = 0;
 365         *done = 0;
 366         fmterror = 0;
 367         ubufp = ubuffer;
 368         nicluster = mp->m_sb.sb_blocksize >= XFS_INODE_CLUSTER_SIZE(mp) ?
 369                 mp->m_sb.sb_inopblock :
 370                 (XFS_INODE_CLUSTER_SIZE(mp) >> mp->m_sb.sb_inodelog);
 371         nimask = ~(nicluster - 1);
 372         nbcluster = nicluster >> mp->m_sb.sb_inopblog;
 373         /*
 374          * Allocate a local buffer for inode cluster btree records.
 375          * This caps our maximum readahead window (so don't be stingy)
 376          * but we must handle the case where we can't get a contiguous
 377          * multi-page buffer, so we drop back toward pagesize; the end
 378          * case we ensure succeeds, via appropriate allocation flags.
 379          */
 380         irbsize = NBPP * 4;
 381         kmflags = KM_SLEEP | KM_MAYFAIL;
 382         while (!(irbuf = kmem_alloc(irbsize, kmflags))) {
 383                 if ((irbsize >>= 1) <= NBPP)
 384                         kmflags = KM_SLEEP;
 385         }
 386         nirbuf = irbsize / sizeof(*irbuf);
 387
 388         /*
 389          * Loop over the allocation groups, starting from the last
 390          * inode returned; 0 means start of the allocation group.
 391          */
 392         rval = 0;
 393         while (ubleft >= statstruct_size && agno < mp->m_sb.sb_agcount) {
 394                 bp = NULL;
 395                 down_read(&mp->m_peraglock);
 396                 error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
 397                 up_read(&mp->m_peraglock);
 398                 if (error) {
 399                         /*
 400                          * Skip this allocation group and go to the next one.
 401                          */
 402                         agno++;
 403                         agino = 0;
 404                         continue;
 405                 }
 406                 agi = XFS_BUF_TO_AGI(agbp);
 407                 /*
 408                  * Allocate and initialize a btree cursor for ialloc btree.
 409                  */
 410                 cur = xfs_btree_init_cursor(mp, NULL, agbp, agno, XFS_BTNUM_INO,
 411                                                 (xfs_inode_t *)0, 0);
 412                 irbp = irbuf;
 413                 irbufend = irbuf + nirbuf;
 414                 end_of_ag = 0;
 415                 /*
 416                  * If we're returning in the middle of an allocation group,
 417                  * we need to get the remainder of the chunk we're in.
 418                  */
 419                 if (agino > 0) {
 420                         /*
 421                          * Lookup the inode chunk that this inode lives in.
 422                          */
 423                         error = xfs_inobt_lookup_le(cur, agino, 0, 0, &tmp);
 424                         if (!error &&   /* no I/O error */
 425                             tmp &&      /* lookup succeeded */
 426                                         /* got the record, should always work */
 427                             !(error = xfs_inobt_get_rec(cur, &gino, &gcnt,
 428                                     &gfree, &i)) &&
 429                             i == 1 &&
 430                                         /* this is the right chunk */
 431                             agino < gino + XFS_INODES_PER_CHUNK &&
 432                                         /* lastino was not last in chunk */
 433                             (chunkidx = agino - gino + 1) <
 434                                     XFS_INODES_PER_CHUNK &&
 435                                         /* there are some left allocated */
 436                             XFS_INOBT_MASKN(chunkidx,
 437                                     XFS_INODES_PER_CHUNK - chunkidx) & ~gfree) {
 438                                 /*
 439                                  * Grab the chunk record.  Mark all the
 440                                  * uninteresting inodes (because they're
 441                                  * before our start point) free.
 442                                  */
 443                                 for (i = 0; i < chunkidx; i++) {
 444                                         if (XFS_INOBT_MASK(i) & ~gfree)
 445                                                 gcnt++;
 446                                 }
 447                                 gfree |= XFS_INOBT_MASKN(0, chunkidx);
 448                                 irbp->ir_startino = gino;
 449                                 irbp->ir_freecount = gcnt;
 450                                 irbp->ir_free = gfree;
 451                                 irbp++;
 452                                 agino = gino + XFS_INODES_PER_CHUNK;
 453                                 icount = XFS_INODES_PER_CHUNK - gcnt;
 454                         } else {
 455                                 /*
 456                                  * If any of those tests failed, bump the
 457                                  * inode number (just in case).
 458                                  */
 459                                 agino++;
 460                                 icount = 0;
 461                         }
 462                         /*
 463                          * In any case, increment to the next record.
 464                          */
 465                         if (!error)
 466                                 error = xfs_inobt_increment(cur, 0, &tmp);
 467                 } else {
 468                         /*
 469                          * Start of ag.  Lookup the first inode chunk.
 470                          */
 471                         error = xfs_inobt_lookup_ge(cur, 0, 0, 0, &tmp);
 472                         icount = 0;
 473                 }
 474                 /*
 475                  * Loop through inode btree records in this ag,
 476                  * until we run out of inodes or space in the buffer.
 477                  */
 478                 while (irbp < irbufend && icount < ubcount) {
 479                         /*
 480                          * Loop as long as we're unable to read the
 481                          * inode btree.
 482                          */
 483                         while (error) {
 484                                 agino += XFS_INODES_PER_CHUNK;
 485                                 if (XFS_AGINO_TO_AGBNO(mp, agino) >=
 486                                                 be32_to_cpu(agi->agi_length))
 487                                         break;
 488                                 error = xfs_inobt_lookup_ge(cur, agino, 0, 0,
 489                                                             &tmp);
 490                         }
 491                         /*
 492                          * If ran off the end of the ag either with an error,
 493                          * or the normal way, set end and stop collecting.
 494                          */
 495                         if (error ||
 496                             (error = xfs_inobt_get_rec(cur, &gino, &gcnt,
 497                                     &gfree, &i)) ||
 498                             i == 0) {
 499                                 end_of_ag = 1;
 500                                 break;
 501                         }
 502                         /*
 503                          * If this chunk has any allocated inodes, save it.
 504                          * Also start read-ahead now for this chunk.
 505                          */
 506                         if (gcnt < XFS_INODES_PER_CHUNK) {
 507                                 /*
 508                                  * Loop over all clusters in the next chunk.
 509                                  * Do a readahead if there are any allocated
 510                                  * inodes in that cluster.
 511                                  */
 512                                 for (agbno = XFS_AGINO_TO_AGBNO(mp, gino),
 513                                      chunkidx = 0;
 514                                      chunkidx < XFS_INODES_PER_CHUNK;
 515                                      chunkidx += nicluster,
 516                                      agbno += nbcluster) {
 517                                         if (XFS_INOBT_MASKN(chunkidx,
 518                                                             nicluster) & ~gfree)
 519                                                 xfs_btree_reada_bufs(mp, agno,
 520                                                         agbno, nbcluster);
 521                                 }
 522                                 irbp->ir_startino = gino;
 523                                 irbp->ir_freecount = gcnt;
 524                                 irbp->ir_free = gfree;
 525                                 irbp++;
 526                                 icount += XFS_INODES_PER_CHUNK - gcnt;
 527                         }
 528                         /*
 529                          * Set agino to after this chunk and bump the cursor.
 530                          */
 531                         agino = gino + XFS_INODES_PER_CHUNK;
 532                         error = xfs_inobt_increment(cur, 0, &tmp);
 533                 }
 534                 /*
 535                  * Drop the btree buffers and the agi buffer.
 536                  * We can't hold any of the locks these represent
 537                  * when calling iget.
 538                  */
 539                 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 540                 xfs_buf_relse(agbp);
 541                 /*
 542                  * Now format all the good inodes into the user's buffer.
 543                  */
 544                 irbufend = irbp;
 545                 for (irbp = irbuf;
 546                      irbp < irbufend && ubleft >= statstruct_size; irbp++) {
 547                         /*
 548                          * Now process this chunk of inodes.
 549                          */
 550                         for (agino = irbp->ir_startino, chunkidx = clustidx = 0;
 551                              ubleft > 0 &&
 552                                 irbp->ir_freecount < XFS_INODES_PER_CHUNK;
 553                              chunkidx++, clustidx++, agino++) {
 554                                 ASSERT(chunkidx < XFS_INODES_PER_CHUNK);
 555                                 /*
 556                                  * Recompute agbno if this is the
 557                                  * first inode of the cluster.
 558                                  *
 559                                  * Careful with clustidx.   There can be
 560                                  * multple clusters per chunk, a single
 561                                  * cluster per chunk or a cluster that has
 562                                  * inodes represented from several different
 563                                  * chunks (if blocksize is large).
 564                                  *
 565                                  * Because of this, the starting clustidx is
 566                                  * initialized to zero in this loop but must
 567                                  * later be reset after reading in the cluster
 568                                  * buffer.
 569                                  */
 570                                 if ((chunkidx & (nicluster - 1)) == 0) {
 571                                         agbno = XFS_AGINO_TO_AGBNO(mp,
 572                                                         irbp->ir_startino) +
 573                                                 ((chunkidx & nimask) >>
 574                                                  mp->m_sb.sb_inopblog);
 575
 576                                         if (flags & (BULKSTAT_FG_QUICK |
 577                                                      BULKSTAT_FG_INLINE)) {
 578                                                 ino = XFS_AGINO_TO_INO(mp, agno,
 579                                                                        agino);
 580                                                 bno = XFS_AGB_TO_DADDR(mp, agno,
 581                                                                        agbno);
 582
 583                                                 /*
 584                                                  * Get the inode cluster buffer
 585                                                  */
 586                                                 ASSERT(xfs_inode_zone != NULL);
 587                                                 ip = kmem_zone_zalloc(xfs_inode_zone,
 588                                                                       KM_SLEEP);
 589                                                 ip->i_ino = ino;
 590                                                 ip->i_mount = mp;
 591                                                 if (bp)
 592                                                         xfs_buf_relse(bp);
 593                                                 error = xfs_itobp(mp, NULL, ip,
 594                                                                 &dip, &bp, bno,
 595                                                                 XFS_IMAP_BULKSTAT);
 596                                                 if (!error)
 597                                                         clustidx = ip->i_boffset / mp->m_sb.sb_inodesize;
 598                                                 kmem_zone_free(xfs_inode_zone, ip);
 599                                                 if (XFS_TEST_ERROR(error != 0,
 600                                                                    mp, XFS_ERRTAG_BULKSTAT_READ_CHUNK,
 601                                                                    XFS_RANDOM_BULKSTAT_READ_CHUNK)) {
 602                                                         bp = NULL;
 603                                                         ubleft = 0;
 604                                                         rval = error;
 605                                                         break;
 606                                                 }
 607                                         }
 608                                 }
 609                                 /*
 610                                  * Skip if this inode is free.
 611                                  */
 612                                 if (XFS_INOBT_MASK(chunkidx) & irbp->ir_free)
 613                                         continue;
 614                                 /*
 615                                  * Count used inodes as free so we can tell
 616                                  * when the chunk is used up.
 617                                  */
 618                                 irbp->ir_freecount++;
 619                                 ino = XFS_AGINO_TO_INO(mp, agno, agino);
 620                                 bno = XFS_AGB_TO_DADDR(mp, agno, agbno);
 621                                 if (!xfs_bulkstat_use_dinode(mp, flags, bp,
 622                                                              clustidx, &dip))
 623                                         continue;
 624                                 /*
 625                                  * If we need to do an iget, cannot hold bp.
 626                                  * Drop it, until starting the next cluster.
 627                                  */
 628                                 if ((flags & BULKSTAT_FG_INLINE) && !dip) {
 629                                         if (bp)
 630                                                 xfs_buf_relse(bp);
 631                                         bp = NULL;
 632                                 }
 633
 634                                 /*
 635                                  * Get the inode and fill in a single buffer.
 636                                  * BULKSTAT_FG_QUICK uses dip to fill it in.
 637                                  * BULKSTAT_FG_IGET uses igets.
 638                                  * BULKSTAT_FG_INLINE uses dip if we have an
 639                                  * inline attr fork, else igets.
 640                                  * See: xfs_bulkstat_one & xfs_dm_bulkstat_one.
 641                                  * This is also used to count inodes/blks, etc
 642                                  * in xfs_qm_quotacheck.
 643                                  */
 644                                 ubused = statstruct_size;
 645                                 error = formatter(mp, ino, ubufp,
 646                                                 ubleft, private_data,
 647                                                 bno, &ubused, dip, &fmterror);
 648                                 if (fmterror == BULKSTAT_RV_NOTHING) {
 649                                         if (error == ENOMEM)
 650                                                 ubleft = 0;
 651                                         continue;
 652                                 }
 653                                 if (fmterror == BULKSTAT_RV_GIVEUP) {
 654                                         ubleft = 0;
 655                                         ASSERT(error);
 656                                         rval = error;
 657                                         break;
 658                                 }
 659                                 if (ubufp)
 660                                         ubufp += ubused;
 661                                 ubleft -= ubused;
 662                                 ubelem++;
 663                                 lastino = ino;
 664                         }
 665                 }
 666
 667                 if (bp)
 668                         xfs_buf_relse(bp);
 669
 670                 /*
 671                  * Set up for the next loop iteration.
 672                  */
 673                 if (ubleft > 0) {
 674                         if (end_of_ag) {
 675                                 agno++;
 676                                 agino = 0;
 677                         } else
 678                                 agino = XFS_INO_TO_AGINO(mp, lastino);
 679                 } else
 680                         break;
 681         }
 682         /*
 683          * Done, we're either out of filesystem or space to put the data.
 684          */
 685         kmem_free(irbuf, irbsize);
 686         *ubcountp = ubelem;
 687         if (agno >= mp->m_sb.sb_agcount) {
 688                 /*
 689                  * If we ran out of filesystem, mark lastino as off
 690                  * the end of the filesystem, so the next call
 691                  * will return immediately.
 692                  */
 693                 *lastinop = (xfs_ino_t)XFS_AGINO_TO_INO(mp, agno, 0);
 694                 *done = 1;
 695         } else
 696                 *lastinop = (xfs_ino_t)lastino;
 697
 698         return rval;
 699 }
 700
 701 /*
 702  * Return stat information in bulk (by-inode) for the filesystem.
 703  * Special case for non-sequential one inode bulkstat.
 704  */
 705 int                                     /* error status */
 706 xfs_bulkstat_single(
 707         xfs_mount_t             *mp,    /* mount point for filesystem */
 708         xfs_ino_t               *lastinop, /* inode to return */
 709         char                    __user *buffer, /* buffer with inode stats */
 710         int                     *done)  /* 1 if there are more stats to get */
 711 {
 712         int                     count;  /* count value for bulkstat call */
 713         int                     error;  /* return value */
 714         xfs_ino_t               ino;    /* filesystem inode number */
 715         int                     res;    /* result from bs1 */
 716
 717         /*
 718          * note that requesting valid inode numbers which are not allocated
 719          * to inodes will most likely cause xfs_itobp to generate warning
 720          * messages about bad magic numbers. This is ok. The fact that
 721          * the inode isn't actually an inode is handled by the
 722          * error check below. Done this way to make the usual case faster
 723          * at the expense of the error case.
 724          */
 725
 726         ino = (xfs_ino_t)*lastinop;
 727         error = xfs_bulkstat_one(mp, ino, buffer, sizeof(xfs_bstat_t),
 728                                  NULL, 0, NULL, NULL, &res);
 729         if (error) {
 730                 /*
 731                  * Special case way failed, do it the "long" way
 732                  * to see if that works.
 733                  */
 734                 (*lastinop)--;
 735                 count = 1;
 736                 if (xfs_bulkstat(mp, lastinop, &count, xfs_bulkstat_one,
 737                                 NULL, sizeof(xfs_bstat_t), buffer,
 738                                 BULKSTAT_FG_IGET, done))
 739                         return error;
 740                 if (count == 0 || (xfs_ino_t)*lastinop != ino)
 741                         return error == EFSCORRUPTED ?
 742                                 XFS_ERROR(EINVAL) : error;
 743                 else
 744                         return 0;
 745         }
 746         *done = 0;
 747         return 0;
 748 }
 749
 750 /*
 751  * Return inode number table for the filesystem.
 752  */
 753 int                                     /* error status */
 754 xfs_inumbers(
 755         xfs_mount_t     *mp,            /* mount point for filesystem */
 756         xfs_ino_t       *lastino,       /* last inode returned */
 757         int             *count,         /* size of buffer/count returned */
 758         xfs_inogrp_t    __user *ubuffer)/* buffer with inode descriptions */
 759 {
 760         xfs_buf_t       *agbp;
 761         xfs_agino_t     agino;
 762         xfs_agnumber_t  agno;
 763         int             bcount;
 764         xfs_inogrp_t    *buffer;
 765         int             bufidx;
 766         xfs_btree_cur_t *cur;
 767         int             error;
 768         __int32_t       gcnt;
 769         xfs_inofree_t   gfree;
 770         xfs_agino_t     gino;
 771         int             i;
 772         xfs_ino_t       ino;
 773         int             left;
 774         int             tmp;
 775
 776         ino = (xfs_ino_t)*lastino;
 777         agno = XFS_INO_TO_AGNO(mp, ino);
 778         agino = XFS_INO_TO_AGINO(mp, ino);
 779         left = *count;
 780         *count = 0;
 781         bcount = MIN(left, (int)(NBPP / sizeof(*buffer)));
 782         buffer = kmem_alloc(bcount * sizeof(*buffer), KM_SLEEP);
 783         error = bufidx = 0;
 784         cur = NULL;
 785         agbp = NULL;
 786         while (left > 0 && agno < mp->m_sb.sb_agcount) {
 787                 if (agbp == NULL) {
 788                         down_read(&mp->m_peraglock);
 789                         error = xfs_ialloc_read_agi(mp, NULL, agno, &agbp);
 790                         up_read(&mp->m_peraglock);
 791                         if (error) {
 792                                 /*
 793                                  * If we can't read the AGI of this ag,
 794                                  * then just skip to the next one.
 795                                  */
 796                                 ASSERT(cur == NULL);
 797                                 agbp = NULL;
 798                                 agno++;
 799                                 agino = 0;
 800                                 continue;
 801                         }
 802                         cur = xfs_btree_init_cursor(mp, NULL, agbp, agno,
 803                                 XFS_BTNUM_INO, (xfs_inode_t *)0, 0);
 804                         error = xfs_inobt_lookup_ge(cur, agino, 0, 0, &tmp);
 805                         if (error) {
 806                                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 807                                 cur = NULL;
 808                                 xfs_buf_relse(agbp);
 809                                 agbp = NULL;
 810                                 /*
 811                                  * Move up the the last inode in the current
 812                                  * chunk.  The lookup_ge will always get
 813                                  * us the first inode in the next chunk.
 814                                  */
 815                                 agino += XFS_INODES_PER_CHUNK - 1;
 816                                 continue;
 817                         }
 818                 }
 819                 if ((error = xfs_inobt_get_rec(cur, &gino, &gcnt, &gfree,
 820                         &i)) ||
 821                     i == 0) {
 822                         xfs_buf_relse(agbp);
 823                         agbp = NULL;
 824                         xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
 825                         cur = NULL;
 826                         agno++;
 827                         agino = 0;
 828                         continue;
 829                 }
 830                 agino = gino + XFS_INODES_PER_CHUNK - 1;
 831                 buffer[bufidx].xi_startino = XFS_AGINO_TO_INO(mp, agno, gino);
 832                 buffer[bufidx].xi_alloccount = XFS_INODES_PER_CHUNK - gcnt;
 833                 buffer[bufidx].xi_allocmask = ~gfree;
 834                 bufidx++;
 835                 left--;
 836                 if (bufidx == bcount) {
 837                         if (copy_to_user(ubuffer, buffer,
 838                                         bufidx * sizeof(*buffer))) {
 839                                 error = XFS_ERROR(EFAULT);
 840                                 break;
 841                         }
 842                         ubuffer += bufidx;
 843                         *count += bufidx;
 844                         bufidx = 0;
 845                 }
 846                 if (left) {
 847                         error = xfs_inobt_increment(cur, 0, &tmp);
 848                         if (error) {
 849                                 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
 850                                 cur = NULL;
 851                                 xfs_buf_relse(agbp);
 852                                 agbp = NULL;
 853                                 /*
 854                                  * The agino value has already been bumped.
 855                                  * Just try to skip up to it.
 856                                  */
 857                                 agino += XFS_INODES_PER_CHUNK;
 858                                 continue;
 859                         }
 860                 }
 861         }
 862         if (!error) {
 863                 if (bufidx) {
 864                         if (copy_to_user(ubuffer, buffer,
 865                                         bufidx * sizeof(*buffer)))
 866                                 error = XFS_ERROR(EFAULT);
 867                         else
 868                                 *count += bufidx;
 869                 }
 870                 *lastino = XFS_AGINO_TO_INO(mp, agno, agino);
 871         }
 872         kmem_free(buffer, bcount * sizeof(*buffer));
 873         if (cur)
 874                 xfs_btree_del_cursor(cur, (error ? XFS_BTREE_ERROR :
 875                                            XFS_BTREE_NOERROR));
 876         if (agbp)
 877                 xfs_buf_relse(agbp);
 878         return error;
 879 }