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_types.h"
  21 #include "xfs_log.h"
  22 #include "xfs_trans.h"
  23 #include "xfs_sb.h"
  24 #include "xfs_ag.h"
  25 #include "xfs_mount.h"
  26 #include "xfs_bmap_btree.h"
  27 #include "xfs_alloc_btree.h"
  28 #include "xfs_ialloc_btree.h"
  29 #include "xfs_dinode.h"
  30 #include "xfs_inode.h"
  31 #include "xfs_inode_item.h"
  32 #include "xfs_btree.h"
  33 #include "xfs_error.h"
  34 #include "xfs_alloc.h"
  35 #include "xfs_ialloc.h"
  36 #include "xfs_fsops.h"
  37 #include "xfs_itable.h"
  38 #include "xfs_trans_space.h"
  39 #include "xfs_rtalloc.h"
  40 #include "xfs_filestream.h"
  41 #include "xfs_trace.h"
  42
  43 /*
  44  * File system operations
  45  */
  46
  47 int
  48 xfs_fs_geometry(
  49         xfs_mount_t             *mp,
  50         xfs_fsop_geom_t         *geo,
  51         int                     new_version)
  52 {
  53
  54         memset(geo, 0, sizeof(*geo));
  55
  56         geo->blocksize = mp->m_sb.sb_blocksize;
  57         geo->rtextsize = mp->m_sb.sb_rextsize;
  58         geo->agblocks = mp->m_sb.sb_agblocks;
  59         geo->agcount = mp->m_sb.sb_agcount;
  60         geo->logblocks = mp->m_sb.sb_logblocks;
  61         geo->sectsize = mp->m_sb.sb_sectsize;
  62         geo->inodesize = mp->m_sb.sb_inodesize;
  63         geo->imaxpct = mp->m_sb.sb_imax_pct;
  64         geo->datablocks = mp->m_sb.sb_dblocks;
  65         geo->rtblocks = mp->m_sb.sb_rblocks;
  66         geo->rtextents = mp->m_sb.sb_rextents;
  67         geo->logstart = mp->m_sb.sb_logstart;
  68         ASSERT(sizeof(geo->uuid)==sizeof(mp->m_sb.sb_uuid));
  69         memcpy(geo->uuid, &mp->m_sb.sb_uuid, sizeof(mp->m_sb.sb_uuid));
  70         if (new_version >= 2) {
  71                 geo->sunit = mp->m_sb.sb_unit;
  72                 geo->swidth = mp->m_sb.sb_width;
  73         }
  74         if (new_version >= 3) {
  75                 geo->version = XFS_FSOP_GEOM_VERSION;
  76                 geo->flags =
  77                         (xfs_sb_version_hasattr(&mp->m_sb) ?
  78                                 XFS_FSOP_GEOM_FLAGS_ATTR : 0) |
  79                         (xfs_sb_version_hasnlink(&mp->m_sb) ?
  80                                 XFS_FSOP_GEOM_FLAGS_NLINK : 0) |
  81                         (xfs_sb_version_hasquota(&mp->m_sb) ?
  82                                 XFS_FSOP_GEOM_FLAGS_QUOTA : 0) |
  83                         (xfs_sb_version_hasalign(&mp->m_sb) ?
  84                                 XFS_FSOP_GEOM_FLAGS_IALIGN : 0) |
  85                         (xfs_sb_version_hasdalign(&mp->m_sb) ?
  86                                 XFS_FSOP_GEOM_FLAGS_DALIGN : 0) |
  87                         (xfs_sb_version_hasshared(&mp->m_sb) ?
  88                                 XFS_FSOP_GEOM_FLAGS_SHARED : 0) |
  89                         (xfs_sb_version_hasextflgbit(&mp->m_sb) ?
  90                                 XFS_FSOP_GEOM_FLAGS_EXTFLG : 0) |
  91                         (xfs_sb_version_hasdirv2(&mp->m_sb) ?
  92                                 XFS_FSOP_GEOM_FLAGS_DIRV2 : 0) |
  93                         (xfs_sb_version_hassector(&mp->m_sb) ?
  94                                 XFS_FSOP_GEOM_FLAGS_SECTOR : 0) |
  95                         (xfs_sb_version_hasasciici(&mp->m_sb) ?
  96                                 XFS_FSOP_GEOM_FLAGS_DIRV2CI : 0) |
  97                         (xfs_sb_version_haslazysbcount(&mp->m_sb) ?
  98                                 XFS_FSOP_GEOM_FLAGS_LAZYSB : 0) |
  99                         (xfs_sb_version_hasattr2(&mp->m_sb) ?
 100                                 XFS_FSOP_GEOM_FLAGS_ATTR2 : 0) |
 101                         (xfs_sb_version_hasprojid32bit(&mp->m_sb) ?
 102                                 XFS_FSOP_GEOM_FLAGS_PROJID32 : 0) |
 103                         (xfs_sb_version_hascrc(&mp->m_sb) ?
 104                                 XFS_FSOP_GEOM_FLAGS_V5SB : 0);
 105                 geo->logsectsize = xfs_sb_version_hassector(&mp->m_sb) ?
 106                                 mp->m_sb.sb_logsectsize : BBSIZE;
 107                 geo->rtsectsize = mp->m_sb.sb_blocksize;
 108                 geo->dirblocksize = mp->m_dirblksize;
 109         }
 110         if (new_version >= 4) {
 111                 geo->flags |=
 112                         (xfs_sb_version_haslogv2(&mp->m_sb) ?
 113                                 XFS_FSOP_GEOM_FLAGS_LOGV2 : 0);
 114                 geo->logsunit = mp->m_sb.sb_logsunit;
 115         }
 116         return 0;
 117 }
 118
 119 static struct xfs_buf *
 120 xfs_growfs_get_hdr_buf(
 121         struct xfs_mount        *mp,
 122         xfs_daddr_t             blkno,
 123         size_t                  numblks,
 124         int                     flags,
 125         const struct xfs_buf_ops *ops)
 126 {
 127         struct xfs_buf          *bp;
 128
 129         bp = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, flags);
 130         if (!bp)
 131                 return NULL;
 132
 133         xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
 134         bp->b_bn = blkno;
 135         bp->b_maps[0].bm_bn = blkno;
 136         bp->b_ops = ops;
 137
 138         return bp;
 139 }
 140
 141 static int
 142 xfs_growfs_data_private(
 143         xfs_mount_t             *mp,            /* mount point for filesystem */
 144         xfs_growfs_data_t       *in)            /* growfs data input struct */
 145 {
 146         xfs_agf_t               *agf;
 147         struct xfs_agfl         *agfl;
 148         xfs_agi_t               *agi;
 149         xfs_agnumber_t          agno;
 150         xfs_extlen_t            agsize;
 151         xfs_extlen_t            tmpsize;
 152         xfs_alloc_rec_t         *arec;
 153         xfs_buf_t               *bp;
 154         int                     bucket;
 155         int                     dpct;
 156         int                     error;
 157         xfs_agnumber_t          nagcount;
 158         xfs_agnumber_t          nagimax = 0;
 159         xfs_rfsblock_t          nb, nb_mod;
 160         xfs_rfsblock_t          new;
 161         xfs_rfsblock_t          nfree;
 162         xfs_agnumber_t          oagcount;
 163         int                     pct;
 164         xfs_trans_t             *tp;
 165
 166         nb = in->newblocks;
 167         pct = in->imaxpct;
 168         if (nb < mp->m_sb.sb_dblocks || pct < 0 || pct > 100)
 169                 return XFS_ERROR(EINVAL);
 170         if ((error = xfs_sb_validate_fsb_count(&mp->m_sb, nb)))
 171                 return error;
 172         dpct = pct - mp->m_sb.sb_imax_pct;
 173         bp = xfs_buf_read_uncached(mp->m_ddev_targp,
 174                                 XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1),
 175                                 XFS_FSS_TO_BB(mp, 1), 0, NULL);
 176         if (!bp)
 177                 return EIO;
 178         if (bp->b_error) {
 179                 int     error = bp->b_error;
 180                 xfs_buf_relse(bp);
 181                 return error;
 182         }
 183         xfs_buf_relse(bp);
 184
 185         new = nb;       /* use new as a temporary here */
 186         nb_mod = do_div(new, mp->m_sb.sb_agblocks);
 187         nagcount = new + (nb_mod != 0);
 188         if (nb_mod && nb_mod < XFS_MIN_AG_BLOCKS) {
 189                 nagcount--;
 190                 nb = (xfs_rfsblock_t)nagcount * mp->m_sb.sb_agblocks;
 191                 if (nb < mp->m_sb.sb_dblocks)
 192                         return XFS_ERROR(EINVAL);
 193         }
 194         new = nb - mp->m_sb.sb_dblocks;
 195         oagcount = mp->m_sb.sb_agcount;
 196
 197         /* allocate the new per-ag structures */
 198         if (nagcount > oagcount) {
 199                 error = xfs_initialize_perag(mp, nagcount, &nagimax);
 200                 if (error)
 201                         return error;
 202         }
 203
 204         tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFS);
 205         tp->t_flags |= XFS_TRANS_RESERVE;
 206         if ((error = xfs_trans_reserve(tp, XFS_GROWFS_SPACE_RES(mp),
 207                         XFS_GROWDATA_LOG_RES(mp), 0, 0, 0))) {
 208                 xfs_trans_cancel(tp, 0);
 209                 return error;
 210         }
 211
 212         /*
 213          * Write new AG headers to disk. Non-transactional, but written
 214          * synchronously so they are completed prior to the growfs transaction
 215          * being logged.
 216          */
 217         nfree = 0;
 218         for (agno = nagcount - 1; agno >= oagcount; agno--, new -= agsize) {
 219                 __be32  *agfl_bno;
 220
 221                 /*
 222                  * AG freespace header block
 223                  */
 224                 bp = xfs_growfs_get_hdr_buf(mp,
 225                                 XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
 226                                 XFS_FSS_TO_BB(mp, 1), 0,
 227                                 &xfs_agf_buf_ops);
 228                 if (!bp) {
 229                         error = ENOMEM;
 230                         goto error0;
 231                 }
 232
 233                 agf = XFS_BUF_TO_AGF(bp);
 234                 agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
 235                 agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
 236                 agf->agf_seqno = cpu_to_be32(agno);
 237                 if (agno == nagcount - 1)
 238                         agsize =
 239                                 nb -
 240                                 (agno * (xfs_rfsblock_t)mp->m_sb.sb_agblocks);
 241                 else
 242                         agsize = mp->m_sb.sb_agblocks;
 243                 agf->agf_length = cpu_to_be32(agsize);
 244                 agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
 245                 agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
 246                 agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
 247                 agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
 248                 agf->agf_flfirst = 0;
 249                 agf->agf_fllast = cpu_to_be32(XFS_AGFL_SIZE(mp) - 1);
 250                 agf->agf_flcount = 0;
 251                 tmpsize = agsize - XFS_PREALLOC_BLOCKS(mp);
 252                 agf->agf_freeblks = cpu_to_be32(tmpsize);
 253                 agf->agf_longest = cpu_to_be32(tmpsize);
 254                 if (xfs_sb_version_hascrc(&mp->m_sb))
 255                         uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_uuid);
 256
 257                 error = xfs_bwrite(bp);
 258                 xfs_buf_relse(bp);
 259                 if (error)
 260                         goto error0;
 261
 262                 /*
 263                  * AG freelist header block
 264                  */
 265                 bp = xfs_growfs_get_hdr_buf(mp,
 266                                 XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)),
 267                                 XFS_FSS_TO_BB(mp, 1), 0,
 268                                 &xfs_agfl_buf_ops);
 269                 if (!bp) {
 270                         error = ENOMEM;
 271                         goto error0;
 272                 }
 273
 274                 agfl = XFS_BUF_TO_AGFL(bp);
 275                 if (xfs_sb_version_hascrc(&mp->m_sb)) {
 276                         agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
 277                         agfl->agfl_seqno = cpu_to_be32(agno);
 278                         uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_uuid);
 279                 }
 280
 281                 agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, bp);
 282                 for (bucket = 0; bucket < XFS_AGFL_SIZE(mp); bucket++)
 283                         agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
 284
 285                 error = xfs_bwrite(bp);
 286                 xfs_buf_relse(bp);
 287                 if (error)
 288                         goto error0;
 289
 290                 /*
 291                  * AG inode header block
 292                  */
 293                 bp = xfs_growfs_get_hdr_buf(mp,
 294                                 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
 295                                 XFS_FSS_TO_BB(mp, 1), 0,
 296                                 &xfs_agi_buf_ops);
 297                 if (!bp) {
 298                         error = ENOMEM;
 299                         goto error0;
 300                 }
 301
 302                 agi = XFS_BUF_TO_AGI(bp);
 303                 agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
 304                 agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
 305                 agi->agi_seqno = cpu_to_be32(agno);
 306                 agi->agi_length = cpu_to_be32(agsize);
 307                 agi->agi_count = 0;
 308                 agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
 309                 agi->agi_level = cpu_to_be32(1);
 310                 agi->agi_freecount = 0;
 311                 agi->agi_newino = cpu_to_be32(NULLAGINO);
 312                 agi->agi_dirino = cpu_to_be32(NULLAGINO);
 313                 if (xfs_sb_version_hascrc(&mp->m_sb))
 314                         uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_uuid);
 315                 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
 316                         agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
 317
 318                 error = xfs_bwrite(bp);
 319                 xfs_buf_relse(bp);
 320                 if (error)
 321                         goto error0;
 322
 323                 /*
 324                  * BNO btree root block
 325                  */
 326                 bp = xfs_growfs_get_hdr_buf(mp,
 327                                 XFS_AGB_TO_DADDR(mp, agno, XFS_BNO_BLOCK(mp)),
 328                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 329                                 &xfs_allocbt_buf_ops);
 330
 331                 if (!bp) {
 332                         error = ENOMEM;
 333                         goto error0;
 334                 }
 335
 336                 if (xfs_sb_version_hascrc(&mp->m_sb))
 337                         xfs_btree_init_block(mp, bp, XFS_ABTB_CRC_MAGIC, 0, 1,
 338                                                 agno, XFS_BTREE_CRC_BLOCKS);
 339                 else
 340                         xfs_btree_init_block(mp, bp, XFS_ABTB_MAGIC, 0, 1,
 341                                                 agno, 0);
 342
 343                 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
 344                 arec->ar_startblock = cpu_to_be32(XFS_PREALLOC_BLOCKS(mp));
 345                 arec->ar_blockcount = cpu_to_be32(
 346                         agsize - be32_to_cpu(arec->ar_startblock));
 347
 348                 error = xfs_bwrite(bp);
 349                 xfs_buf_relse(bp);
 350                 if (error)
 351                         goto error0;
 352
 353                 /*
 354                  * CNT btree root block
 355                  */
 356                 bp = xfs_growfs_get_hdr_buf(mp,
 357                                 XFS_AGB_TO_DADDR(mp, agno, XFS_CNT_BLOCK(mp)),
 358                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 359                                 &xfs_allocbt_buf_ops);
 360                 if (!bp) {
 361                         error = ENOMEM;
 362                         goto error0;
 363                 }
 364
 365                 if (xfs_sb_version_hascrc(&mp->m_sb))
 366                         xfs_btree_init_block(mp, bp, XFS_ABTC_CRC_MAGIC, 0, 1,
 367                                                 agno, XFS_BTREE_CRC_BLOCKS);
 368                 else
 369                         xfs_btree_init_block(mp, bp, XFS_ABTC_MAGIC, 0, 1,
 370                                                 agno, 0);
 371
 372                 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
 373                 arec->ar_startblock = cpu_to_be32(XFS_PREALLOC_BLOCKS(mp));
 374                 arec->ar_blockcount = cpu_to_be32(
 375                         agsize - be32_to_cpu(arec->ar_startblock));
 376                 nfree += be32_to_cpu(arec->ar_blockcount);
 377
 378                 error = xfs_bwrite(bp);
 379                 xfs_buf_relse(bp);
 380                 if (error)
 381                         goto error0;
 382
 383                 /*
 384                  * INO btree root block
 385                  */
 386                 bp = xfs_growfs_get_hdr_buf(mp,
 387                                 XFS_AGB_TO_DADDR(mp, agno, XFS_IBT_BLOCK(mp)),
 388                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 389                                 &xfs_inobt_buf_ops);
 390                 if (!bp) {
 391                         error = ENOMEM;
 392                         goto error0;
 393                 }
 394
 395                 if (xfs_sb_version_hascrc(&mp->m_sb))
 396                         xfs_btree_init_block(mp, bp, XFS_IBT_CRC_MAGIC, 0, 0,
 397                                                 agno, XFS_BTREE_CRC_BLOCKS);
 398                 else
 399                         xfs_btree_init_block(mp, bp, XFS_IBT_MAGIC, 0, 0,
 400                                                 agno, 0);
 401
 402                 error = xfs_bwrite(bp);
 403                 xfs_buf_relse(bp);
 404                 if (error)
 405                         goto error0;
 406         }
 407         xfs_trans_agblocks_delta(tp, nfree);
 408         /*
 409          * There are new blocks in the old last a.g.
 410          */
 411         if (new) {
 412                 /*
 413                  * Change the agi length.
 414                  */
 415                 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
 416                 if (error) {
 417                         goto error0;
 418                 }
 419                 ASSERT(bp);
 420                 agi = XFS_BUF_TO_AGI(bp);
 421                 be32_add_cpu(&agi->agi_length, new);
 422                 ASSERT(nagcount == oagcount ||
 423                        be32_to_cpu(agi->agi_length) == mp->m_sb.sb_agblocks);
 424                 xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
 425                 /*
 426                  * Change agf length.
 427                  */
 428                 error = xfs_alloc_read_agf(mp, tp, agno, 0, &bp);
 429                 if (error) {
 430                         goto error0;
 431                 }
 432                 ASSERT(bp);
 433                 agf = XFS_BUF_TO_AGF(bp);
 434                 be32_add_cpu(&agf->agf_length, new);
 435                 ASSERT(be32_to_cpu(agf->agf_length) ==
 436                        be32_to_cpu(agi->agi_length));
 437
 438                 xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
 439                 /*
 440                  * Free the new space.
 441                  */
 442                 error = xfs_free_extent(tp, XFS_AGB_TO_FSB(mp, agno,
 443                         be32_to_cpu(agf->agf_length) - new), new);
 444                 if (error) {
 445                         goto error0;
 446                 }
 447         }
 448
 449         /*
 450          * Update changed superblock fields transactionally. These are not
 451          * seen by the rest of the world until the transaction commit applies
 452          * them atomically to the superblock.
 453          */
 454         if (nagcount > oagcount)
 455                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
 456         if (nb > mp->m_sb.sb_dblocks)
 457                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS,
 458                                  nb - mp->m_sb.sb_dblocks);
 459         if (nfree)
 460                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, nfree);
 461         if (dpct)
 462                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
 463         error = xfs_trans_commit(tp, 0);
 464         if (error)
 465                 return error;
 466
 467         /* New allocation groups fully initialized, so update mount struct */
 468         if (nagimax)
 469                 mp->m_maxagi = nagimax;
 470         if (mp->m_sb.sb_imax_pct) {
 471                 __uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
 472                 do_div(icount, 100);
 473                 mp->m_maxicount = icount << mp->m_sb.sb_inopblog;
 474         } else
 475                 mp->m_maxicount = 0;
 476         xfs_set_low_space_thresholds(mp);
 477
 478         /* update secondary superblocks. */
 479         for (agno = 1; agno < nagcount; agno++) {
 480                 error = 0;
 481                 /*
 482                  * new secondary superblocks need to be zeroed, not read from
 483                  * disk as the contents of the new area we are growing into is
 484                  * completely unknown.
 485                  */
 486                 if (agno < oagcount) {
 487                         error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
 488                                   XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
 489                                   XFS_FSS_TO_BB(mp, 1), 0, &bp,
 490                                   &xfs_sb_buf_ops);
 491                 } else {
 492                         bp = xfs_trans_get_buf(NULL, mp->m_ddev_targp,
 493                                   XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
 494                                   XFS_FSS_TO_BB(mp, 1), 0);
 495                         if (bp) {
 496                                 bp->b_ops = &xfs_sb_buf_ops;
 497                                 xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
 498                         } else
 499                                 error = ENOMEM;
 500                 }
 501
 502                 if (error) {
 503                         xfs_warn(mp,
 504                 "error %d reading secondary superblock for ag %d",
 505                                 error, agno);
 506                         break;
 507                 }
 508                 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, XFS_SB_ALL_BITS);
 509
 510                 /*
 511                  * If we get an error writing out the alternate superblocks,
 512                  * just issue a warning and continue.  The real work is
 513                  * already done and committed.
 514                  */
 515                 error = xfs_bwrite(bp);
 516                 xfs_buf_relse(bp);
 517                 if (error) {
 518                         xfs_warn(mp,
 519                 "write error %d updating secondary superblock for ag %d",
 520                                 error, agno);
 521                         break; /* no point in continuing */
 522                 }
 523         }
 524         return error;
 525
 526  error0:
 527         xfs_trans_cancel(tp, XFS_TRANS_ABORT);
 528         return error;
 529 }
 530
 531 static int
 532 xfs_growfs_log_private(
 533         xfs_mount_t             *mp,    /* mount point for filesystem */
 534         xfs_growfs_log_t        *in)    /* growfs log input struct */
 535 {
 536         xfs_extlen_t            nb;
 537
 538         nb = in->newblocks;
 539         if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
 540                 return XFS_ERROR(EINVAL);
 541         if (nb == mp->m_sb.sb_logblocks &&
 542             in->isint == (mp->m_sb.sb_logstart != 0))
 543                 return XFS_ERROR(EINVAL);
 544         /*
 545          * Moving the log is hard, need new interfaces to sync
 546          * the log first, hold off all activity while moving it.
 547          * Can have shorter or longer log in the same space,
 548          * or transform internal to external log or vice versa.
 549          */
 550         return XFS_ERROR(ENOSYS);
 551 }
 552
 553 /*
 554  * protected versions of growfs function acquire and release locks on the mount
 555  * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
 556  * XFS_IOC_FSGROWFSRT
 557  */
 558
 559
 560 int
 561 xfs_growfs_data(
 562         xfs_mount_t             *mp,
 563         xfs_growfs_data_t       *in)
 564 {
 565         int error;
 566
 567         if (!capable(CAP_SYS_ADMIN))
 568                 return XFS_ERROR(EPERM);
 569         if (!mutex_trylock(&mp->m_growlock))
 570                 return XFS_ERROR(EWOULDBLOCK);
 571         error = xfs_growfs_data_private(mp, in);
 572         mutex_unlock(&mp->m_growlock);
 573         return error;
 574 }
 575
 576 int
 577 xfs_growfs_log(
 578         xfs_mount_t             *mp,
 579         xfs_growfs_log_t        *in)
 580 {
 581         int error;
 582
 583         if (!capable(CAP_SYS_ADMIN))
 584                 return XFS_ERROR(EPERM);
 585         if (!mutex_trylock(&mp->m_growlock))
 586                 return XFS_ERROR(EWOULDBLOCK);
 587         error = xfs_growfs_log_private(mp, in);
 588         mutex_unlock(&mp->m_growlock);
 589         return error;
 590 }
 591
 592 /*
 593  * exported through ioctl XFS_IOC_FSCOUNTS
 594  */
 595
 596 int
 597 xfs_fs_counts(
 598         xfs_mount_t             *mp,
 599         xfs_fsop_counts_t       *cnt)
 600 {
 601         xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
 602         spin_lock(&mp->m_sb_lock);
 603         cnt->freedata = mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
 604         cnt->freertx = mp->m_sb.sb_frextents;
 605         cnt->freeino = mp->m_sb.sb_ifree;
 606         cnt->allocino = mp->m_sb.sb_icount;
 607         spin_unlock(&mp->m_sb_lock);
 608         return 0;
 609 }
 610
 611 /*
 612  * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
 613  *
 614  * xfs_reserve_blocks is called to set m_resblks
 615  * in the in-core mount table. The number of unused reserved blocks
 616  * is kept in m_resblks_avail.
 617  *
 618  * Reserve the requested number of blocks if available. Otherwise return
 619  * as many as possible to satisfy the request. The actual number
 620  * reserved are returned in outval
 621  *
 622  * A null inval pointer indicates that only the current reserved blocks
 623  * available  should  be returned no settings are changed.
 624  */
 625
 626 int
 627 xfs_reserve_blocks(
 628         xfs_mount_t             *mp,
 629         __uint64_t              *inval,
 630         xfs_fsop_resblks_t      *outval)
 631 {
 632         __int64_t               lcounter, delta, fdblks_delta;
 633         __uint64_t              request;
 634
 635         /* If inval is null, report current values and return */
 636         if (inval == (__uint64_t *)NULL) {
 637                 if (!outval)
 638                         return EINVAL;
 639                 outval->resblks = mp->m_resblks;
 640                 outval->resblks_avail = mp->m_resblks_avail;
 641                 return 0;
 642         }
 643
 644         request = *inval;
 645
 646         /*
 647          * With per-cpu counters, this becomes an interesting
 648          * problem. we needto work out if we are freeing or allocation
 649          * blocks first, then we can do the modification as necessary.
 650          *
 651          * We do this under the m_sb_lock so that if we are near
 652          * ENOSPC, we will hold out any changes while we work out
 653          * what to do. This means that the amount of free space can
 654          * change while we do this, so we need to retry if we end up
 655          * trying to reserve more space than is available.
 656          *
 657          * We also use the xfs_mod_incore_sb() interface so that we
 658          * don't have to care about whether per cpu counter are
 659          * enabled, disabled or even compiled in....
 660          */
 661 retry:
 662         spin_lock(&mp->m_sb_lock);
 663         xfs_icsb_sync_counters_locked(mp, 0);
 664
 665         /*
 666          * If our previous reservation was larger than the current value,
 667          * then move any unused blocks back to the free pool.
 668          */
 669         fdblks_delta = 0;
 670         if (mp->m_resblks > request) {
 671                 lcounter = mp->m_resblks_avail - request;
 672                 if (lcounter  > 0) {            /* release unused blocks */
 673                         fdblks_delta = lcounter;
 674                         mp->m_resblks_avail -= lcounter;
 675                 }
 676                 mp->m_resblks = request;
 677         } else {
 678                 __int64_t       free;
 679
 680                 free =  mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
 681                 if (!free)
 682                         goto out; /* ENOSPC and fdblks_delta = 0 */
 683
 684                 delta = request - mp->m_resblks;
 685                 lcounter = free - delta;
 686                 if (lcounter < 0) {
 687                         /* We can't satisfy the request, just get what we can */
 688                         mp->m_resblks += free;
 689                         mp->m_resblks_avail += free;
 690                         fdblks_delta = -free;
 691                 } else {
 692                         fdblks_delta = -delta;
 693                         mp->m_resblks = request;
 694                         mp->m_resblks_avail += delta;
 695                 }
 696         }
 697 out:
 698         if (outval) {
 699                 outval->resblks = mp->m_resblks;
 700                 outval->resblks_avail = mp->m_resblks_avail;
 701         }
 702         spin_unlock(&mp->m_sb_lock);
 703
 704         if (fdblks_delta) {
 705                 /*
 706                  * If we are putting blocks back here, m_resblks_avail is
 707                  * already at its max so this will put it in the free pool.
 708                  *
 709                  * If we need space, we'll either succeed in getting it
 710                  * from the free block count or we'll get an enospc. If
 711                  * we get a ENOSPC, it means things changed while we were
 712                  * calculating fdblks_delta and so we should try again to
 713                  * see if there is anything left to reserve.
 714                  *
 715                  * Don't set the reserved flag here - we don't want to reserve
 716                  * the extra reserve blocks from the reserve.....
 717                  */
 718                 int error;
 719                 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
 720                                                  fdblks_delta, 0);
 721                 if (error == ENOSPC)
 722                         goto retry;
 723         }
 724         return 0;
 725 }
 726
 727 /*
 728  * Dump a transaction into the log that contains no real change. This is needed
 729  * to be able to make the log dirty or stamp the current tail LSN into the log
 730  * during the covering operation.
 731  *
 732  * We cannot use an inode here for this - that will push dirty state back up
 733  * into the VFS and then periodic inode flushing will prevent log covering from
 734  * making progress. Hence we log a field in the superblock instead and use a
 735  * synchronous transaction to ensure the superblock is immediately unpinned
 736  * and can be written back.
 737  */
 738 int
 739 xfs_fs_log_dummy(
 740         xfs_mount_t     *mp)
 741 {
 742         xfs_trans_t     *tp;
 743         int             error;
 744
 745         tp = _xfs_trans_alloc(mp, XFS_TRANS_DUMMY1, KM_SLEEP);
 746         error = xfs_trans_reserve(tp, 0, XFS_SB_LOG_RES(mp), 0, 0,
 747                                   XFS_DEFAULT_LOG_COUNT);
 748         if (error) {
 749                 xfs_trans_cancel(tp, 0);
 750                 return error;
 751         }
 752
 753         /* log the UUID because it is an unchanging field */
 754         xfs_mod_sb(tp, XFS_SB_UUID);
 755         xfs_trans_set_sync(tp);
 756         return xfs_trans_commit(tp, 0);
 757 }
 758
 759 int
 760 xfs_fs_goingdown(
 761         xfs_mount_t     *mp,
 762         __uint32_t      inflags)
 763 {
 764         switch (inflags) {
 765         case XFS_FSOP_GOING_FLAGS_DEFAULT: {
 766                 struct super_block *sb = freeze_bdev(mp->m_super->s_bdev);
 767
 768                 if (sb && !IS_ERR(sb)) {
 769                         xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
 770                         thaw_bdev(sb->s_bdev, sb);
 771                 }
 772
 773                 break;
 774         }
 775         case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
 776                 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
 777                 break;
 778         case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
 779                 xfs_force_shutdown(mp,
 780                                 SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
 781                 break;
 782         default:
 783                 return XFS_ERROR(EINVAL);
 784         }
 785
 786         return 0;
 787 }
 788
 789 /*
 790  * Force a shutdown of the filesystem instantly while keeping the filesystem
 791  * consistent. We don't do an unmount here; just shutdown the shop, make sure
 792  * that absolutely nothing persistent happens to this filesystem after this
 793  * point.
 794  */
 795 void
 796 xfs_do_force_shutdown(
 797         xfs_mount_t     *mp,
 798         int             flags,
 799         char            *fname,
 800         int             lnnum)
 801 {
 802         int             logerror;
 803
 804         logerror = flags & SHUTDOWN_LOG_IO_ERROR;
 805
 806         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 807                 xfs_notice(mp,
 808         "%s(0x%x) called from line %d of file %s.  Return address = 0x%p",
 809                         __func__, flags, lnnum, fname, __return_address);
 810         }
 811         /*
 812          * No need to duplicate efforts.
 813          */
 814         if (XFS_FORCED_SHUTDOWN(mp) && !logerror)
 815                 return;
 816
 817         /*
 818          * This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't
 819          * queue up anybody new on the log reservations, and wakes up
 820          * everybody who's sleeping on log reservations to tell them
 821          * the bad news.
 822          */
 823         if (xfs_log_force_umount(mp, logerror))
 824                 return;
 825
 826         if (flags & SHUTDOWN_CORRUPT_INCORE) {
 827                 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_CORRUPT,
 828     "Corruption of in-memory data detected.  Shutting down filesystem");
 829                 if (XFS_ERRLEVEL_HIGH <= xfs_error_level)
 830                         xfs_stack_trace();
 831         } else if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 832                 if (logerror) {
 833                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_LOGERROR,
 834                 "Log I/O Error Detected.  Shutting down filesystem");
 835                 } else if (flags & SHUTDOWN_DEVICE_REQ) {
 836                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
 837                 "All device paths lost.  Shutting down filesystem");
 838                 } else if (!(flags & SHUTDOWN_REMOTE_REQ)) {
 839                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
 840                 "I/O Error Detected. Shutting down filesystem");
 841                 }
 842         }
 843         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 844                 xfs_alert(mp,
 845         "Please umount the filesystem and rectify the problem(s)");
 846         }
 847 }