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