fs/xfs/xfs_fsops.c

   1 /*
   2  * Copyright (c) 2000-2005 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_shared.h"
  21 #include "xfs_format.h"
  22 #include "xfs_log_format.h"
  23 #include "xfs_trans_resv.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_mount.h"
  27 #include "xfs_da_format.h"
  28 #include "xfs_da_btree.h"
  29 #include "xfs_inode.h"
  30 #include "xfs_trans.h"
  31 #include "xfs_inode_item.h"
  32 #include "xfs_error.h"
  33 #include "xfs_btree.h"
  34 #include "xfs_alloc_btree.h"
  35 #include "xfs_alloc.h"
  36 #include "xfs_ialloc.h"
  37 #include "xfs_fsops.h"
  38 #include "xfs_itable.h"
  39 #include "xfs_trans_space.h"
  40 #include "xfs_rtalloc.h"
  41 #include "xfs_trace.h"
  42 #include "xfs_log.h"
  43 #include "xfs_dinode.h"
  44 #include "xfs_filestream.h"
  45
  46 /*
  47  * File system operations
  48  */
  49
  50 int
  51 xfs_fs_geometry(
  52         xfs_mount_t             *mp,
  53         xfs_fsop_geom_t         *geo,
  54         int                     new_version)
  55 {
  56
  57         memset(geo, 0, sizeof(*geo));
  58
  59         geo->blocksize = mp->m_sb.sb_blocksize;
  60         geo->rtextsize = mp->m_sb.sb_rextsize;
  61         geo->agblocks = mp->m_sb.sb_agblocks;
  62         geo->agcount = mp->m_sb.sb_agcount;
  63         geo->logblocks = mp->m_sb.sb_logblocks;
  64         geo->sectsize = mp->m_sb.sb_sectsize;
  65         geo->inodesize = mp->m_sb.sb_inodesize;
  66         geo->imaxpct = mp->m_sb.sb_imax_pct;
  67         geo->datablocks = mp->m_sb.sb_dblocks;
  68         geo->rtblocks = mp->m_sb.sb_rblocks;
  69         geo->rtextents = mp->m_sb.sb_rextents;
  70         geo->logstart = mp->m_sb.sb_logstart;
  71         ASSERT(sizeof(geo->uuid)==sizeof(mp->m_sb.sb_uuid));
  72         memcpy(geo->uuid, &mp->m_sb.sb_uuid, sizeof(mp->m_sb.sb_uuid));
  73         if (new_version >= 2) {
  74                 geo->sunit = mp->m_sb.sb_unit;
  75                 geo->swidth = mp->m_sb.sb_width;
  76         }
  77         if (new_version >= 3) {
  78                 geo->version = XFS_FSOP_GEOM_VERSION;
  79                 geo->flags = XFS_FSOP_GEOM_FLAGS_NLINK |
  80                              XFS_FSOP_GEOM_FLAGS_DIRV2 |
  81                         (xfs_sb_version_hasattr(&mp->m_sb) ?
  82                                 XFS_FSOP_GEOM_FLAGS_ATTR : 0) |
  83                         (xfs_sb_version_hasquota(&mp->m_sb) ?
  84                                 XFS_FSOP_GEOM_FLAGS_QUOTA : 0) |
  85                         (xfs_sb_version_hasalign(&mp->m_sb) ?
  86                                 XFS_FSOP_GEOM_FLAGS_IALIGN : 0) |
  87                         (xfs_sb_version_hasdalign(&mp->m_sb) ?
  88                                 XFS_FSOP_GEOM_FLAGS_DALIGN : 0) |
  89                         (xfs_sb_version_hasextflgbit(&mp->m_sb) ?
  90                                 XFS_FSOP_GEOM_FLAGS_EXTFLG : 0) |
  91                         (xfs_sb_version_hassector(&mp->m_sb) ?
  92                                 XFS_FSOP_GEOM_FLAGS_SECTOR : 0) |
  93                         (xfs_sb_version_hasasciici(&mp->m_sb) ?
  94                                 XFS_FSOP_GEOM_FLAGS_DIRV2CI : 0) |
  95                         (xfs_sb_version_haslazysbcount(&mp->m_sb) ?
  96                                 XFS_FSOP_GEOM_FLAGS_LAZYSB : 0) |
  97                         (xfs_sb_version_hasattr2(&mp->m_sb) ?
  98                                 XFS_FSOP_GEOM_FLAGS_ATTR2 : 0) |
  99                         (xfs_sb_version_hasprojid32bit(&mp->m_sb) ?
 100                                 XFS_FSOP_GEOM_FLAGS_PROJID32 : 0) |
 101                         (xfs_sb_version_hascrc(&mp->m_sb) ?
 102                                 XFS_FSOP_GEOM_FLAGS_V5SB : 0) |
 103                         (xfs_sb_version_hasftype(&mp->m_sb) ?
 104                                 XFS_FSOP_GEOM_FLAGS_FTYPE : 0) |
 105                         (xfs_sb_version_hasfinobt(&mp->m_sb) ?
 106                                 XFS_FSOP_GEOM_FLAGS_FINOBT : 0);
 107                 geo->logsectsize = xfs_sb_version_hassector(&mp->m_sb) ?
 108                                 mp->m_sb.sb_logsectsize : BBSIZE;
 109                 geo->rtsectsize = mp->m_sb.sb_blocksize;
 110                 geo->dirblocksize = mp->m_dir_geo->blksize;
 111         }
 112         if (new_version >= 4) {
 113                 geo->flags |=
 114                         (xfs_sb_version_haslogv2(&mp->m_sb) ?
 115                                 XFS_FSOP_GEOM_FLAGS_LOGV2 : 0);
 116                 geo->logsunit = mp->m_sb.sb_logsunit;
 117         }
 118         return 0;
 119 }
 120
 121 static struct xfs_buf *
 122 xfs_growfs_get_hdr_buf(
 123         struct xfs_mount        *mp,
 124         xfs_daddr_t             blkno,
 125         size_t                  numblks,
 126         int                     flags,
 127         const struct xfs_buf_ops *ops)
 128 {
 129         struct xfs_buf          *bp;
 130
 131         bp = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, flags);
 132         if (!bp)
 133                 return NULL;
 134
 135         xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
 136         bp->b_bn = blkno;
 137         bp->b_maps[0].bm_bn = blkno;
 138         bp->b_ops = ops;
 139
 140         return bp;
 141 }
 142
 143 static int
 144 xfs_growfs_data_private(
 145         xfs_mount_t             *mp,            /* mount point for filesystem */
 146         xfs_growfs_data_t       *in)            /* growfs data input struct */
 147 {
 148         xfs_agf_t               *agf;
 149         struct xfs_agfl         *agfl;
 150         xfs_agi_t               *agi;
 151         xfs_agnumber_t          agno;
 152         xfs_extlen_t            agsize;
 153         xfs_extlen_t            tmpsize;
 154         xfs_alloc_rec_t         *arec;
 155         xfs_buf_t               *bp;
 156         int                     bucket;
 157         int                     dpct;
 158         int                     error, saved_error = 0;
 159         xfs_agnumber_t          nagcount;
 160         xfs_agnumber_t          nagimax = 0;
 161         xfs_rfsblock_t          nb, nb_mod;
 162         xfs_rfsblock_t          new;
 163         xfs_rfsblock_t          nfree;
 164         xfs_agnumber_t          oagcount;
 165         int                     pct;
 166         xfs_trans_t             *tp;
 167
 168         nb = in->newblocks;
 169         pct = in->imaxpct;
 170         if (nb < mp->m_sb.sb_dblocks || pct < 0 || pct > 100)
 171                 return XFS_ERROR(EINVAL);
 172         if ((error = xfs_sb_validate_fsb_count(&mp->m_sb, nb)))
 173                 return error;
 174         dpct = pct - mp->m_sb.sb_imax_pct;
 175         bp = xfs_buf_read_uncached(mp->m_ddev_targp,
 176                                 XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1),
 177                                 XFS_FSS_TO_BB(mp, 1), 0, NULL);
 178         if (!bp)
 179                 return EIO;
 180         if (bp->b_error) {
 181                 error = bp->b_error;
 182                 xfs_buf_relse(bp);
 183                 return error;
 184         }
 185         xfs_buf_relse(bp);
 186
 187         new = nb;       /* use new as a temporary here */
 188         nb_mod = do_div(new, mp->m_sb.sb_agblocks);
 189         nagcount = new + (nb_mod != 0);
 190         if (nb_mod && nb_mod < XFS_MIN_AG_BLOCKS) {
 191                 nagcount--;
 192                 nb = (xfs_rfsblock_t)nagcount * mp->m_sb.sb_agblocks;
 193                 if (nb < mp->m_sb.sb_dblocks)
 194                         return XFS_ERROR(EINVAL);
 195         }
 196         new = nb - mp->m_sb.sb_dblocks;
 197         oagcount = mp->m_sb.sb_agcount;
 198
 199         /* allocate the new per-ag structures */
 200         if (nagcount > oagcount) {
 201                 error = xfs_initialize_perag(mp, nagcount, &nagimax);
 202                 if (error)
 203                         return error;
 204         }
 205
 206         tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFS);
 207         tp->t_flags |= XFS_TRANS_RESERVE;
 208         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_growdata,
 209                                   XFS_GROWFS_SPACE_RES(mp), 0);
 210         if (error) {
 211                 xfs_trans_cancel(tp, 0);
 212                 return error;
 213         }
 214
 215         /*
 216          * Write new AG headers to disk. Non-transactional, but written
 217          * synchronously so they are completed prior to the growfs transaction
 218          * being logged.
 219          */
 220         nfree = 0;
 221         for (agno = nagcount - 1; agno >= oagcount; agno--, new -= agsize) {
 222                 __be32  *agfl_bno;
 223
 224                 /*
 225                  * AG freespace header block
 226                  */
 227                 bp = xfs_growfs_get_hdr_buf(mp,
 228                                 XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
 229                                 XFS_FSS_TO_BB(mp, 1), 0,
 230                                 &xfs_agf_buf_ops);
 231                 if (!bp) {
 232                         error = ENOMEM;
 233                         goto error0;
 234                 }
 235
 236                 agf = XFS_BUF_TO_AGF(bp);
 237                 agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
 238                 agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
 239                 agf->agf_seqno = cpu_to_be32(agno);
 240                 if (agno == nagcount - 1)
 241                         agsize =
 242                                 nb -
 243                                 (agno * (xfs_rfsblock_t)mp->m_sb.sb_agblocks);
 244                 else
 245                         agsize = mp->m_sb.sb_agblocks;
 246                 agf->agf_length = cpu_to_be32(agsize);
 247                 agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
 248                 agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
 249                 agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
 250                 agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
 251                 agf->agf_flfirst = 0;
 252                 agf->agf_fllast = cpu_to_be32(XFS_AGFL_SIZE(mp) - 1);
 253                 agf->agf_flcount = 0;
 254                 tmpsize = agsize - XFS_PREALLOC_BLOCKS(mp);
 255                 agf->agf_freeblks = cpu_to_be32(tmpsize);
 256                 agf->agf_longest = cpu_to_be32(tmpsize);
 257                 if (xfs_sb_version_hascrc(&mp->m_sb))
 258                         uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_uuid);
 259
 260                 error = xfs_bwrite(bp);
 261                 xfs_buf_relse(bp);
 262                 if (error)
 263                         goto error0;
 264
 265                 /*
 266                  * AG freelist header block
 267                  */
 268                 bp = xfs_growfs_get_hdr_buf(mp,
 269                                 XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)),
 270                                 XFS_FSS_TO_BB(mp, 1), 0,
 271                                 &xfs_agfl_buf_ops);
 272                 if (!bp) {
 273                         error = ENOMEM;
 274                         goto error0;
 275                 }
 276
 277                 agfl = XFS_BUF_TO_AGFL(bp);
 278                 if (xfs_sb_version_hascrc(&mp->m_sb)) {
 279                         agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
 280                         agfl->agfl_seqno = cpu_to_be32(agno);
 281                         uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_uuid);
 282                 }
 283
 284                 agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, bp);
 285                 for (bucket = 0; bucket < XFS_AGFL_SIZE(mp); bucket++)
 286                         agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
 287
 288                 error = xfs_bwrite(bp);
 289                 xfs_buf_relse(bp);
 290                 if (error)
 291                         goto error0;
 292
 293                 /*
 294                  * AG inode header block
 295                  */
 296                 bp = xfs_growfs_get_hdr_buf(mp,
 297                                 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
 298                                 XFS_FSS_TO_BB(mp, 1), 0,
 299                                 &xfs_agi_buf_ops);
 300                 if (!bp) {
 301                         error = ENOMEM;
 302                         goto error0;
 303                 }
 304
 305                 agi = XFS_BUF_TO_AGI(bp);
 306                 agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
 307                 agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
 308                 agi->agi_seqno = cpu_to_be32(agno);
 309                 agi->agi_length = cpu_to_be32(agsize);
 310                 agi->agi_count = 0;
 311                 agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
 312                 agi->agi_level = cpu_to_be32(1);
 313                 agi->agi_freecount = 0;
 314                 agi->agi_newino = cpu_to_be32(NULLAGINO);
 315                 agi->agi_dirino = cpu_to_be32(NULLAGINO);
 316                 if (xfs_sb_version_hascrc(&mp->m_sb))
 317                         uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_uuid);
 318                 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
 319                         agi->agi_free_root = cpu_to_be32(XFS_FIBT_BLOCK(mp));
 320                         agi->agi_free_level = cpu_to_be32(1);
 321                 }
 322                 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
 323                         agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
 324
 325                 error = xfs_bwrite(bp);
 326                 xfs_buf_relse(bp);
 327                 if (error)
 328                         goto error0;
 329
 330                 /*
 331                  * BNO btree root block
 332                  */
 333                 bp = xfs_growfs_get_hdr_buf(mp,
 334                                 XFS_AGB_TO_DADDR(mp, agno, XFS_BNO_BLOCK(mp)),
 335                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 336                                 &xfs_allocbt_buf_ops);
 337
 338                 if (!bp) {
 339                         error = ENOMEM;
 340                         goto error0;
 341                 }
 342
 343                 if (xfs_sb_version_hascrc(&mp->m_sb))
 344                         xfs_btree_init_block(mp, bp, XFS_ABTB_CRC_MAGIC, 0, 1,
 345                                                 agno, XFS_BTREE_CRC_BLOCKS);
 346                 else
 347                         xfs_btree_init_block(mp, bp, XFS_ABTB_MAGIC, 0, 1,
 348                                                 agno, 0);
 349
 350                 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
 351                 arec->ar_startblock = cpu_to_be32(XFS_PREALLOC_BLOCKS(mp));
 352                 arec->ar_blockcount = cpu_to_be32(
 353                         agsize - be32_to_cpu(arec->ar_startblock));
 354
 355                 error = xfs_bwrite(bp);
 356                 xfs_buf_relse(bp);
 357                 if (error)
 358                         goto error0;
 359
 360                 /*
 361                  * CNT btree root block
 362                  */
 363                 bp = xfs_growfs_get_hdr_buf(mp,
 364                                 XFS_AGB_TO_DADDR(mp, agno, XFS_CNT_BLOCK(mp)),
 365                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 366                                 &xfs_allocbt_buf_ops);
 367                 if (!bp) {
 368                         error = ENOMEM;
 369                         goto error0;
 370                 }
 371
 372                 if (xfs_sb_version_hascrc(&mp->m_sb))
 373                         xfs_btree_init_block(mp, bp, XFS_ABTC_CRC_MAGIC, 0, 1,
 374                                                 agno, XFS_BTREE_CRC_BLOCKS);
 375                 else
 376                         xfs_btree_init_block(mp, bp, XFS_ABTC_MAGIC, 0, 1,
 377                                                 agno, 0);
 378
 379                 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
 380                 arec->ar_startblock = cpu_to_be32(XFS_PREALLOC_BLOCKS(mp));
 381                 arec->ar_blockcount = cpu_to_be32(
 382                         agsize - be32_to_cpu(arec->ar_startblock));
 383                 nfree += be32_to_cpu(arec->ar_blockcount);
 384
 385                 error = xfs_bwrite(bp);
 386                 xfs_buf_relse(bp);
 387                 if (error)
 388                         goto error0;
 389
 390                 /*
 391                  * INO btree root block
 392                  */
 393                 bp = xfs_growfs_get_hdr_buf(mp,
 394                                 XFS_AGB_TO_DADDR(mp, agno, XFS_IBT_BLOCK(mp)),
 395                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 396                                 &xfs_inobt_buf_ops);
 397                 if (!bp) {
 398                         error = ENOMEM;
 399                         goto error0;
 400                 }
 401
 402                 if (xfs_sb_version_hascrc(&mp->m_sb))
 403                         xfs_btree_init_block(mp, bp, XFS_IBT_CRC_MAGIC, 0, 0,
 404                                                 agno, XFS_BTREE_CRC_BLOCKS);
 405                 else
 406                         xfs_btree_init_block(mp, bp, XFS_IBT_MAGIC, 0, 0,
 407                                                 agno, 0);
 408
 409                 error = xfs_bwrite(bp);
 410                 xfs_buf_relse(bp);
 411                 if (error)
 412                         goto error0;
 413
 414                 /*
 415                  * FINO btree root block
 416                  */
 417                 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
 418                         bp = xfs_growfs_get_hdr_buf(mp,
 419                                 XFS_AGB_TO_DADDR(mp, agno, XFS_FIBT_BLOCK(mp)),
 420                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 421                                 &xfs_inobt_buf_ops);
 422                         if (!bp) {
 423                                 error = ENOMEM;
 424                                 goto error0;
 425                         }
 426
 427                         if (xfs_sb_version_hascrc(&mp->m_sb))
 428                                 xfs_btree_init_block(mp, bp, XFS_FIBT_CRC_MAGIC,
 429                                                      0, 0, agno,
 430                                                      XFS_BTREE_CRC_BLOCKS);
 431                         else
 432                                 xfs_btree_init_block(mp, bp, XFS_FIBT_MAGIC, 0,
 433                                                      0, agno, 0);
 434
 435                         error = xfs_bwrite(bp);
 436                         xfs_buf_relse(bp);
 437                         if (error)
 438                                 goto error0;
 439                 }
 440
 441         }
 442         xfs_trans_agblocks_delta(tp, nfree);
 443         /*
 444          * There are new blocks in the old last a.g.
 445          */
 446         if (new) {
 447                 /*
 448                  * Change the agi length.
 449                  */
 450                 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
 451                 if (error) {
 452                         goto error0;
 453                 }
 454                 ASSERT(bp);
 455                 agi = XFS_BUF_TO_AGI(bp);
 456                 be32_add_cpu(&agi->agi_length, new);
 457                 ASSERT(nagcount == oagcount ||
 458                        be32_to_cpu(agi->agi_length) == mp->m_sb.sb_agblocks);
 459                 xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
 460                 /*
 461                  * Change agf length.
 462                  */
 463                 error = xfs_alloc_read_agf(mp, tp, agno, 0, &bp);
 464                 if (error) {
 465                         goto error0;
 466                 }
 467                 ASSERT(bp);
 468                 agf = XFS_BUF_TO_AGF(bp);
 469                 be32_add_cpu(&agf->agf_length, new);
 470                 ASSERT(be32_to_cpu(agf->agf_length) ==
 471                        be32_to_cpu(agi->agi_length));
 472
 473                 xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
 474                 /*
 475                  * Free the new space.
 476                  */
 477                 error = xfs_free_extent(tp, XFS_AGB_TO_FSB(mp, agno,
 478                         be32_to_cpu(agf->agf_length) - new), new);
 479                 if (error) {
 480                         goto error0;
 481                 }
 482         }
 483
 484         /*
 485          * Update changed superblock fields transactionally. These are not
 486          * seen by the rest of the world until the transaction commit applies
 487          * them atomically to the superblock.
 488          */
 489         if (nagcount > oagcount)
 490                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
 491         if (nb > mp->m_sb.sb_dblocks)
 492                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS,
 493                                  nb - mp->m_sb.sb_dblocks);
 494         if (nfree)
 495                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, nfree);
 496         if (dpct)
 497                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
 498         error = xfs_trans_commit(tp, 0);
 499         if (error)
 500                 return error;
 501
 502         /* New allocation groups fully initialized, so update mount struct */
 503         if (nagimax)
 504                 mp->m_maxagi = nagimax;
 505         if (mp->m_sb.sb_imax_pct) {
 506                 __uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
 507                 do_div(icount, 100);
 508                 mp->m_maxicount = icount << mp->m_sb.sb_inopblog;
 509         } else
 510                 mp->m_maxicount = 0;
 511         xfs_set_low_space_thresholds(mp);
 512
 513         /* update secondary superblocks. */
 514         for (agno = 1; agno < nagcount; agno++) {
 515                 error = 0;
 516                 /*
 517                  * new secondary superblocks need to be zeroed, not read from
 518                  * disk as the contents of the new area we are growing into is
 519                  * completely unknown.
 520                  */
 521                 if (agno < oagcount) {
 522                         error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
 523                                   XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
 524                                   XFS_FSS_TO_BB(mp, 1), 0, &bp,
 525                                   &xfs_sb_buf_ops);
 526                 } else {
 527                         bp = xfs_trans_get_buf(NULL, mp->m_ddev_targp,
 528                                   XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
 529                                   XFS_FSS_TO_BB(mp, 1), 0);
 530                         if (bp) {
 531                                 bp->b_ops = &xfs_sb_buf_ops;
 532                                 xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
 533                         } else
 534                                 error = ENOMEM;
 535                 }
 536
 537                 /*
 538                  * If we get an error reading or writing alternate superblocks,
 539                  * continue.  xfs_repair chooses the "best" superblock based
 540                  * on most matches; if we break early, we'll leave more
 541                  * superblocks un-updated than updated, and xfs_repair may
 542                  * pick them over the properly-updated primary.
 543                  */
 544                 if (error) {
 545                         xfs_warn(mp,
 546                 "error %d reading secondary superblock for ag %d",
 547                                 error, agno);
 548                         saved_error = error;
 549                         continue;
 550                 }
 551                 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, XFS_SB_ALL_BITS);
 552
 553                 error = xfs_bwrite(bp);
 554                 xfs_buf_relse(bp);
 555                 if (error) {
 556                         xfs_warn(mp,
 557                 "write error %d updating secondary superblock for ag %d",
 558                                 error, agno);
 559                         saved_error = error;
 560                         continue;
 561                 }
 562         }
 563         return saved_error ? saved_error : error;
 564
 565  error0:
 566         xfs_trans_cancel(tp, XFS_TRANS_ABORT);
 567         return error;
 568 }
 569
 570 static int
 571 xfs_growfs_log_private(
 572         xfs_mount_t             *mp,    /* mount point for filesystem */
 573         xfs_growfs_log_t        *in)    /* growfs log input struct */
 574 {
 575         xfs_extlen_t            nb;
 576
 577         nb = in->newblocks;
 578         if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
 579                 return XFS_ERROR(EINVAL);
 580         if (nb == mp->m_sb.sb_logblocks &&
 581             in->isint == (mp->m_sb.sb_logstart != 0))
 582                 return XFS_ERROR(EINVAL);
 583         /*
 584          * Moving the log is hard, need new interfaces to sync
 585          * the log first, hold off all activity while moving it.
 586          * Can have shorter or longer log in the same space,
 587          * or transform internal to external log or vice versa.
 588          */
 589         return XFS_ERROR(ENOSYS);
 590 }
 591
 592 /*
 593  * protected versions of growfs function acquire and release locks on the mount
 594  * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
 595  * XFS_IOC_FSGROWFSRT
 596  */
 597
 598
 599 int
 600 xfs_growfs_data(
 601         xfs_mount_t             *mp,
 602         xfs_growfs_data_t       *in)
 603 {
 604         int error;
 605
 606         if (!capable(CAP_SYS_ADMIN))
 607                 return XFS_ERROR(EPERM);
 608         if (!mutex_trylock(&mp->m_growlock))
 609                 return XFS_ERROR(EWOULDBLOCK);
 610         error = xfs_growfs_data_private(mp, in);
 611         mutex_unlock(&mp->m_growlock);
 612         return error;
 613 }
 614
 615 int
 616 xfs_growfs_log(
 617         xfs_mount_t             *mp,
 618         xfs_growfs_log_t        *in)
 619 {
 620         int error;
 621
 622         if (!capable(CAP_SYS_ADMIN))
 623                 return XFS_ERROR(EPERM);
 624         if (!mutex_trylock(&mp->m_growlock))
 625                 return XFS_ERROR(EWOULDBLOCK);
 626         error = xfs_growfs_log_private(mp, in);
 627         mutex_unlock(&mp->m_growlock);
 628         return error;
 629 }
 630
 631 /*
 632  * exported through ioctl XFS_IOC_FSCOUNTS
 633  */
 634
 635 int
 636 xfs_fs_counts(
 637         xfs_mount_t             *mp,
 638         xfs_fsop_counts_t       *cnt)
 639 {
 640         xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
 641         spin_lock(&mp->m_sb_lock);
 642         cnt->freedata = mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
 643         cnt->freertx = mp->m_sb.sb_frextents;
 644         cnt->freeino = mp->m_sb.sb_ifree;
 645         cnt->allocino = mp->m_sb.sb_icount;
 646         spin_unlock(&mp->m_sb_lock);
 647         return 0;
 648 }
 649
 650 /*
 651  * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
 652  *
 653  * xfs_reserve_blocks is called to set m_resblks
 654  * in the in-core mount table. The number of unused reserved blocks
 655  * is kept in m_resblks_avail.
 656  *
 657  * Reserve the requested number of blocks if available. Otherwise return
 658  * as many as possible to satisfy the request. The actual number
 659  * reserved are returned in outval
 660  *
 661  * A null inval pointer indicates that only the current reserved blocks
 662  * available  should  be returned no settings are changed.
 663  */
 664
 665 int
 666 xfs_reserve_blocks(
 667         xfs_mount_t             *mp,
 668         __uint64_t              *inval,
 669         xfs_fsop_resblks_t      *outval)
 670 {
 671         __int64_t               lcounter, delta, fdblks_delta;
 672         __uint64_t              request;
 673
 674         /* If inval is null, report current values and return */
 675         if (inval == (__uint64_t *)NULL) {
 676                 if (!outval)
 677                         return EINVAL;
 678                 outval->resblks = mp->m_resblks;
 679                 outval->resblks_avail = mp->m_resblks_avail;
 680                 return 0;
 681         }
 682
 683         request = *inval;
 684
 685         /*
 686          * With per-cpu counters, this becomes an interesting
 687          * problem. we needto work out if we are freeing or allocation
 688          * blocks first, then we can do the modification as necessary.
 689          *
 690          * We do this under the m_sb_lock so that if we are near
 691          * ENOSPC, we will hold out any changes while we work out
 692          * what to do. This means that the amount of free space can
 693          * change while we do this, so we need to retry if we end up
 694          * trying to reserve more space than is available.
 695          *
 696          * We also use the xfs_mod_incore_sb() interface so that we
 697          * don't have to care about whether per cpu counter are
 698          * enabled, disabled or even compiled in....
 699          */
 700 retry:
 701         spin_lock(&mp->m_sb_lock);
 702         xfs_icsb_sync_counters_locked(mp, 0);
 703
 704         /*
 705          * If our previous reservation was larger than the current value,
 706          * then move any unused blocks back to the free pool.
 707          */
 708         fdblks_delta = 0;
 709         if (mp->m_resblks > request) {
 710                 lcounter = mp->m_resblks_avail - request;
 711                 if (lcounter  > 0) {            /* release unused blocks */
 712                         fdblks_delta = lcounter;
 713                         mp->m_resblks_avail -= lcounter;
 714                 }
 715                 mp->m_resblks = request;
 716         } else {
 717                 __int64_t       free;
 718
 719                 free =  mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
 720                 if (!free)
 721                         goto out; /* ENOSPC and fdblks_delta = 0 */
 722
 723                 delta = request - mp->m_resblks;
 724                 lcounter = free - delta;
 725                 if (lcounter < 0) {
 726                         /* We can't satisfy the request, just get what we can */
 727                         mp->m_resblks += free;
 728                         mp->m_resblks_avail += free;
 729                         fdblks_delta = -free;
 730                 } else {
 731                         fdblks_delta = -delta;
 732                         mp->m_resblks = request;
 733                         mp->m_resblks_avail += delta;
 734                 }
 735         }
 736 out:
 737         if (outval) {
 738                 outval->resblks = mp->m_resblks;
 739                 outval->resblks_avail = mp->m_resblks_avail;
 740         }
 741         spin_unlock(&mp->m_sb_lock);
 742
 743         if (fdblks_delta) {
 744                 /*
 745                  * If we are putting blocks back here, m_resblks_avail is
 746                  * already at its max so this will put it in the free pool.
 747                  *
 748                  * If we need space, we'll either succeed in getting it
 749                  * from the free block count or we'll get an enospc. If
 750                  * we get a ENOSPC, it means things changed while we were
 751                  * calculating fdblks_delta and so we should try again to
 752                  * see if there is anything left to reserve.
 753                  *
 754                  * Don't set the reserved flag here - we don't want to reserve
 755                  * the extra reserve blocks from the reserve.....
 756                  */
 757                 int error;
 758                 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
 759                                                  fdblks_delta, 0);
 760                 if (error == ENOSPC)
 761                         goto retry;
 762         }
 763         return 0;
 764 }
 765
 766 /*
 767  * Dump a transaction into the log that contains no real change. This is needed
 768  * to be able to make the log dirty or stamp the current tail LSN into the log
 769  * during the covering operation.
 770  *
 771  * We cannot use an inode here for this - that will push dirty state back up
 772  * into the VFS and then periodic inode flushing will prevent log covering from
 773  * making progress. Hence we log a field in the superblock instead and use a
 774  * synchronous transaction to ensure the superblock is immediately unpinned
 775  * and can be written back.
 776  */
 777 int
 778 xfs_fs_log_dummy(
 779         xfs_mount_t     *mp)
 780 {
 781         xfs_trans_t     *tp;
 782         int             error;
 783
 784         tp = _xfs_trans_alloc(mp, XFS_TRANS_DUMMY1, KM_SLEEP);
 785         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
 786         if (error) {
 787                 xfs_trans_cancel(tp, 0);
 788                 return error;
 789         }
 790
 791         /* log the UUID because it is an unchanging field */
 792         xfs_mod_sb(tp, XFS_SB_UUID);
 793         xfs_trans_set_sync(tp);
 794         return xfs_trans_commit(tp, 0);
 795 }
 796
 797 int
 798 xfs_fs_goingdown(
 799         xfs_mount_t     *mp,
 800         __uint32_t      inflags)
 801 {
 802         switch (inflags) {
 803         case XFS_FSOP_GOING_FLAGS_DEFAULT: {
 804                 struct super_block *sb = freeze_bdev(mp->m_super->s_bdev);
 805
 806                 if (sb && !IS_ERR(sb)) {
 807                         xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
 808                         thaw_bdev(sb->s_bdev, sb);
 809                 }
 810
 811                 break;
 812         }
 813         case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
 814                 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
 815                 break;
 816         case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
 817                 xfs_force_shutdown(mp,
 818                                 SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
 819                 break;
 820         default:
 821                 return XFS_ERROR(EINVAL);
 822         }
 823
 824         return 0;
 825 }
 826
 827 /*
 828  * Force a shutdown of the filesystem instantly while keeping the filesystem
 829  * consistent. We don't do an unmount here; just shutdown the shop, make sure
 830  * that absolutely nothing persistent happens to this filesystem after this
 831  * point.
 832  */
 833 void
 834 xfs_do_force_shutdown(
 835         xfs_mount_t     *mp,
 836         int             flags,
 837         char            *fname,
 838         int             lnnum)
 839 {
 840         int             logerror;
 841
 842         logerror = flags & SHUTDOWN_LOG_IO_ERROR;
 843
 844         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 845                 xfs_notice(mp,
 846         "%s(0x%x) called from line %d of file %s.  Return address = 0x%p",
 847                         __func__, flags, lnnum, fname, __return_address);
 848         }
 849         /*
 850          * No need to duplicate efforts.
 851          */
 852         if (XFS_FORCED_SHUTDOWN(mp) && !logerror)
 853                 return;
 854
 855         /*
 856          * This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't
 857          * queue up anybody new on the log reservations, and wakes up
 858          * everybody who's sleeping on log reservations to tell them
 859          * the bad news.
 860          */
 861         if (xfs_log_force_umount(mp, logerror))
 862                 return;
 863
 864         if (flags & SHUTDOWN_CORRUPT_INCORE) {
 865                 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_CORRUPT,
 866     "Corruption of in-memory data detected.  Shutting down filesystem");
 867                 if (XFS_ERRLEVEL_HIGH <= xfs_error_level)
 868                         xfs_stack_trace();
 869         } else if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 870                 if (logerror) {
 871                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_LOGERROR,
 872                 "Log I/O Error Detected.  Shutting down filesystem");
 873                 } else if (flags & SHUTDOWN_DEVICE_REQ) {
 874                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
 875                 "All device paths lost.  Shutting down filesystem");
 876                 } else if (!(flags & SHUTDOWN_REMOTE_REQ)) {
 877                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
 878                 "I/O Error Detected. Shutting down filesystem");
 879                 }
 880         }
 881         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 882                 xfs_alert(mp,
 883         "Please umount the filesystem and rectify the problem(s)");
 884         }
 885 }