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_inode.h"
  28 #include "xfs_trans.h"
  29 #include "xfs_inode_item.h"
  30 #include "xfs_error.h"
  31 #include "xfs_btree.h"
  32 #include "xfs_alloc_btree.h"
  33 #include "xfs_alloc.h"
  34 #include "xfs_ialloc.h"
  35 #include "xfs_fsops.h"
  36 #include "xfs_itable.h"
  37 #include "xfs_trans_space.h"
  38 #include "xfs_rtalloc.h"
  39 #include "xfs_trace.h"
  40 #include "xfs_log.h"
  41 #include "xfs_dinode.h"
  42 #include "xfs_filestream.h"
  43
  44 /*
  45  * File system operations
  46  */
  47
  48 int
  49 xfs_fs_geometry(
  50         xfs_mount_t             *mp,
  51         xfs_fsop_geom_t         *geo,
  52         int                     new_version)
  53 {
  54
  55         memset(geo, 0, sizeof(*geo));
  56
  57         geo->blocksize = mp->m_sb.sb_blocksize;
  58         geo->rtextsize = mp->m_sb.sb_rextsize;
  59         geo->agblocks = mp->m_sb.sb_agblocks;
  60         geo->agcount = mp->m_sb.sb_agcount;
  61         geo->logblocks = mp->m_sb.sb_logblocks;
  62         geo->sectsize = mp->m_sb.sb_sectsize;
  63         geo->inodesize = mp->m_sb.sb_inodesize;
  64         geo->imaxpct = mp->m_sb.sb_imax_pct;
  65         geo->datablocks = mp->m_sb.sb_dblocks;
  66         geo->rtblocks = mp->m_sb.sb_rblocks;
  67         geo->rtextents = mp->m_sb.sb_rextents;
  68         geo->logstart = mp->m_sb.sb_logstart;
  69         ASSERT(sizeof(geo->uuid)==sizeof(mp->m_sb.sb_uuid));
  70         memcpy(geo->uuid, &mp->m_sb.sb_uuid, sizeof(mp->m_sb.sb_uuid));
  71         if (new_version >= 2) {
  72                 geo->sunit = mp->m_sb.sb_unit;
  73                 geo->swidth = mp->m_sb.sb_width;
  74         }
  75         if (new_version >= 3) {
  76                 geo->version = XFS_FSOP_GEOM_VERSION;
  77                 geo->flags =
  78                         (xfs_sb_version_hasattr(&mp->m_sb) ?
  79                                 XFS_FSOP_GEOM_FLAGS_ATTR : 0) |
  80                         (xfs_sb_version_hasnlink(&mp->m_sb) ?
  81                                 XFS_FSOP_GEOM_FLAGS_NLINK : 0) |
  82                         (xfs_sb_version_hasquota(&mp->m_sb) ?
  83                                 XFS_FSOP_GEOM_FLAGS_QUOTA : 0) |
  84                         (xfs_sb_version_hasalign(&mp->m_sb) ?
  85                                 XFS_FSOP_GEOM_FLAGS_IALIGN : 0) |
  86                         (xfs_sb_version_hasdalign(&mp->m_sb) ?
  87                                 XFS_FSOP_GEOM_FLAGS_DALIGN : 0) |
  88                         (xfs_sb_version_hasshared(&mp->m_sb) ?
  89                                 XFS_FSOP_GEOM_FLAGS_SHARED : 0) |
  90                         (xfs_sb_version_hasextflgbit(&mp->m_sb) ?
  91                                 XFS_FSOP_GEOM_FLAGS_EXTFLG : 0) |
  92                         (xfs_sb_version_hasdirv2(&mp->m_sb) ?
  93                                 XFS_FSOP_GEOM_FLAGS_DIRV2 : 0) |
  94                         (xfs_sb_version_hassector(&mp->m_sb) ?
  95                                 XFS_FSOP_GEOM_FLAGS_SECTOR : 0) |
  96                         (xfs_sb_version_hasasciici(&mp->m_sb) ?
  97                                 XFS_FSOP_GEOM_FLAGS_DIRV2CI : 0) |
  98                         (xfs_sb_version_haslazysbcount(&mp->m_sb) ?
  99                                 XFS_FSOP_GEOM_FLAGS_LAZYSB : 0) |
 100                         (xfs_sb_version_hasattr2(&mp->m_sb) ?
 101                                 XFS_FSOP_GEOM_FLAGS_ATTR2 : 0) |
 102                         (xfs_sb_version_hasprojid32bit(&mp->m_sb) ?
 103                                 XFS_FSOP_GEOM_FLAGS_PROJID32 : 0) |
 104                         (xfs_sb_version_hascrc(&mp->m_sb) ?
 105                                 XFS_FSOP_GEOM_FLAGS_V5SB : 0) |
 106                         (xfs_sb_version_hasftype(&mp->m_sb) ?
 107                                 XFS_FSOP_GEOM_FLAGS_FTYPE : 0);
 108                 geo->logsectsize = xfs_sb_version_hassector(&mp->m_sb) ?
 109                                 mp->m_sb.sb_logsectsize : BBSIZE;
 110                 geo->rtsectsize = mp->m_sb.sb_blocksize;
 111                 geo->dirblocksize = mp->m_dirblksize;
 112         }
 113         if (new_version >= 4) {
 114                 geo->flags |=
 115                         (xfs_sb_version_haslogv2(&mp->m_sb) ?
 116                                 XFS_FSOP_GEOM_FLAGS_LOGV2 : 0);
 117                 geo->logsunit = mp->m_sb.sb_logsunit;
 118         }
 119         return 0;
 120 }
 121
 122 static struct xfs_buf *
 123 xfs_growfs_get_hdr_buf(
 124         struct xfs_mount        *mp,
 125         xfs_daddr_t             blkno,
 126         size_t                  numblks,
 127         int                     flags,
 128         const struct xfs_buf_ops *ops)
 129 {
 130         struct xfs_buf          *bp;
 131
 132         bp = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, flags);
 133         if (!bp)
 134                 return NULL;
 135
 136         xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
 137         bp->b_bn = blkno;
 138         bp->b_maps[0].bm_bn = blkno;
 139         bp->b_ops = ops;
 140
 141         return bp;
 142 }
 143
 144 static int
 145 xfs_growfs_data_private(
 146         xfs_mount_t             *mp,            /* mount point for filesystem */
 147         xfs_growfs_data_t       *in)            /* growfs data input struct */
 148 {
 149         xfs_agf_t               *agf;
 150         struct xfs_agfl         *agfl;
 151         xfs_agi_t               *agi;
 152         xfs_agnumber_t          agno;
 153         xfs_extlen_t            agsize;
 154         xfs_extlen_t            tmpsize;
 155         xfs_alloc_rec_t         *arec;
 156         xfs_buf_t               *bp;
 157         int                     bucket;
 158         int                     dpct;
 159         int                     error, saved_error = 0;
 160         xfs_agnumber_t          nagcount;
 161         xfs_agnumber_t          nagimax = 0;
 162         xfs_rfsblock_t          nb, nb_mod;
 163         xfs_rfsblock_t          new;
 164         xfs_rfsblock_t          nfree;
 165         xfs_agnumber_t          oagcount;
 166         int                     pct;
 167         xfs_trans_t             *tp;
 168
 169         nb = in->newblocks;
 170         pct = in->imaxpct;
 171         if (nb < mp->m_sb.sb_dblocks || pct < 0 || pct > 100)
 172                 return XFS_ERROR(EINVAL);
 173         if ((error = xfs_sb_validate_fsb_count(&mp->m_sb, nb)))
 174                 return error;
 175         dpct = pct - mp->m_sb.sb_imax_pct;
 176         bp = xfs_buf_read_uncached(mp->m_ddev_targp,
 177                                 XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1),
 178                                 XFS_FSS_TO_BB(mp, 1), 0, NULL);
 179         if (!bp)
 180                 return EIO;
 181         if (bp->b_error) {
 182                 error = bp->b_error;
 183                 xfs_buf_relse(bp);
 184                 return error;
 185         }
 186         xfs_buf_relse(bp);
 187
 188         new = nb;       /* use new as a temporary here */
 189         nb_mod = do_div(new, mp->m_sb.sb_agblocks);
 190         nagcount = new + (nb_mod != 0);
 191         if (nb_mod && nb_mod < XFS_MIN_AG_BLOCKS) {
 192                 nagcount--;
 193                 nb = (xfs_rfsblock_t)nagcount * mp->m_sb.sb_agblocks;
 194                 if (nb < mp->m_sb.sb_dblocks)
 195                         return XFS_ERROR(EINVAL);
 196         }
 197         new = nb - mp->m_sb.sb_dblocks;
 198         oagcount = mp->m_sb.sb_agcount;
 199
 200         /* allocate the new per-ag structures */
 201         if (nagcount > oagcount) {
 202                 error = xfs_initialize_perag(mp, nagcount, &nagimax);
 203                 if (error)
 204                         return error;
 205         }
 206
 207         tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFS);
 208         tp->t_flags |= XFS_TRANS_RESERVE;
 209         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_growdata,
 210                                   XFS_GROWFS_SPACE_RES(mp), 0);
 211         if (error) {
 212                 xfs_trans_cancel(tp, 0);
 213                 return error;
 214         }
 215
 216         /*
 217          * Write new AG headers to disk. Non-transactional, but written
 218          * synchronously so they are completed prior to the growfs transaction
 219          * being logged.
 220          */
 221         nfree = 0;
 222         for (agno = nagcount - 1; agno >= oagcount; agno--, new -= agsize) {
 223                 /*
 224                  * AG freespace header block
 225                  */
 226                 bp = xfs_growfs_get_hdr_buf(mp,
 227                                 XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
 228                                 XFS_FSS_TO_BB(mp, 1), 0,
 229                                 &xfs_agf_buf_ops);
 230                 if (!bp) {
 231                         error = ENOMEM;
 232                         goto error0;
 233                 }
 234
 235                 agf = XFS_BUF_TO_AGF(bp);
 236                 agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
 237                 agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
 238                 agf->agf_seqno = cpu_to_be32(agno);
 239                 if (agno == nagcount - 1)
 240                         agsize =
 241                                 nb -
 242                                 (agno * (xfs_rfsblock_t)mp->m_sb.sb_agblocks);
 243                 else
 244                         agsize = mp->m_sb.sb_agblocks;
 245                 agf->agf_length = cpu_to_be32(agsize);
 246                 agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
 247                 agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
 248                 agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
 249                 agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
 250                 agf->agf_flfirst = 0;
 251                 agf->agf_fllast = cpu_to_be32(XFS_AGFL_SIZE(mp) - 1);
 252                 agf->agf_flcount = 0;
 253                 tmpsize = agsize - XFS_PREALLOC_BLOCKS(mp);
 254                 agf->agf_freeblks = cpu_to_be32(tmpsize);
 255                 agf->agf_longest = cpu_to_be32(tmpsize);
 256                 if (xfs_sb_version_hascrc(&mp->m_sb))
 257                         uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_uuid);
 258
 259                 error = xfs_bwrite(bp);
 260                 xfs_buf_relse(bp);
 261                 if (error)
 262                         goto error0;
 263
 264                 /*
 265                  * AG freelist header block
 266                  */
 267                 bp = xfs_growfs_get_hdr_buf(mp,
 268                                 XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)),
 269                                 XFS_FSS_TO_BB(mp, 1), 0,
 270                                 &xfs_agfl_buf_ops);
 271                 if (!bp) {
 272                         error = ENOMEM;
 273                         goto error0;
 274                 }
 275
 276                 agfl = XFS_BUF_TO_AGFL(bp);
 277                 if (xfs_sb_version_hascrc(&mp->m_sb)) {
 278                         agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
 279                         agfl->agfl_seqno = cpu_to_be32(agno);
 280                         uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_uuid);
 281                 }
 282                 for (bucket = 0; bucket < XFS_AGFL_SIZE(mp); bucket++)
 283                         agfl->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                 /*
 503                  * If we get an error reading or writing alternate superblocks,
 504                  * continue.  xfs_repair chooses the "best" superblock based
 505                  * on most matches; if we break early, we'll leave more
 506                  * superblocks un-updated than updated, and xfs_repair may
 507                  * pick them over the properly-updated primary.
 508                  */
 509                 if (error) {
 510                         xfs_warn(mp,
 511                 "error %d reading secondary superblock for ag %d",
 512                                 error, agno);
 513                         saved_error = error;
 514                         continue;
 515                 }
 516                 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, XFS_SB_ALL_BITS);
 517
 518                 error = xfs_bwrite(bp);
 519                 xfs_buf_relse(bp);
 520                 if (error) {
 521                         xfs_warn(mp,
 522                 "write error %d updating secondary superblock for ag %d",
 523                                 error, agno);
 524                         saved_error = error;
 525                         continue;
 526                 }
 527         }
 528         return saved_error ? saved_error : error;
 529
 530  error0:
 531         xfs_trans_cancel(tp, XFS_TRANS_ABORT);
 532         return error;
 533 }
 534
 535 static int
 536 xfs_growfs_log_private(
 537         xfs_mount_t             *mp,    /* mount point for filesystem */
 538         xfs_growfs_log_t        *in)    /* growfs log input struct */
 539 {
 540         xfs_extlen_t            nb;
 541
 542         nb = in->newblocks;
 543         if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
 544                 return XFS_ERROR(EINVAL);
 545         if (nb == mp->m_sb.sb_logblocks &&
 546             in->isint == (mp->m_sb.sb_logstart != 0))
 547                 return XFS_ERROR(EINVAL);
 548         /*
 549          * Moving the log is hard, need new interfaces to sync
 550          * the log first, hold off all activity while moving it.
 551          * Can have shorter or longer log in the same space,
 552          * or transform internal to external log or vice versa.
 553          */
 554         return XFS_ERROR(ENOSYS);
 555 }
 556
 557 /*
 558  * protected versions of growfs function acquire and release locks on the mount
 559  * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
 560  * XFS_IOC_FSGROWFSRT
 561  */
 562
 563
 564 int
 565 xfs_growfs_data(
 566         xfs_mount_t             *mp,
 567         xfs_growfs_data_t       *in)
 568 {
 569         int error;
 570
 571         if (!capable(CAP_SYS_ADMIN))
 572                 return XFS_ERROR(EPERM);
 573         if (!mutex_trylock(&mp->m_growlock))
 574                 return XFS_ERROR(EWOULDBLOCK);
 575         error = xfs_growfs_data_private(mp, in);
 576         mutex_unlock(&mp->m_growlock);
 577         return error;
 578 }
 579
 580 int
 581 xfs_growfs_log(
 582         xfs_mount_t             *mp,
 583         xfs_growfs_log_t        *in)
 584 {
 585         int error;
 586
 587         if (!capable(CAP_SYS_ADMIN))
 588                 return XFS_ERROR(EPERM);
 589         if (!mutex_trylock(&mp->m_growlock))
 590                 return XFS_ERROR(EWOULDBLOCK);
 591         error = xfs_growfs_log_private(mp, in);
 592         mutex_unlock(&mp->m_growlock);
 593         return error;
 594 }
 595
 596 /*
 597  * exported through ioctl XFS_IOC_FSCOUNTS
 598  */
 599
 600 int
 601 xfs_fs_counts(
 602         xfs_mount_t             *mp,
 603         xfs_fsop_counts_t       *cnt)
 604 {
 605         xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
 606         spin_lock(&mp->m_sb_lock);
 607         cnt->freedata = mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
 608         cnt->freertx = mp->m_sb.sb_frextents;
 609         cnt->freeino = mp->m_sb.sb_ifree;
 610         cnt->allocino = mp->m_sb.sb_icount;
 611         spin_unlock(&mp->m_sb_lock);
 612         return 0;
 613 }
 614
 615 /*
 616  * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
 617  *
 618  * xfs_reserve_blocks is called to set m_resblks
 619  * in the in-core mount table. The number of unused reserved blocks
 620  * is kept in m_resblks_avail.
 621  *
 622  * Reserve the requested number of blocks if available. Otherwise return
 623  * as many as possible to satisfy the request. The actual number
 624  * reserved are returned in outval
 625  *
 626  * A null inval pointer indicates that only the current reserved blocks
 627  * available  should  be returned no settings are changed.
 628  */
 629
 630 int
 631 xfs_reserve_blocks(
 632         xfs_mount_t             *mp,
 633         __uint64_t              *inval,
 634         xfs_fsop_resblks_t      *outval)
 635 {
 636         __int64_t               lcounter, delta, fdblks_delta;
 637         __uint64_t              request;
 638
 639         /* If inval is null, report current values and return */
 640         if (inval == (__uint64_t *)NULL) {
 641                 if (!outval)
 642                         return EINVAL;
 643                 outval->resblks = mp->m_resblks;
 644                 outval->resblks_avail = mp->m_resblks_avail;
 645                 return 0;
 646         }
 647
 648         request = *inval;
 649
 650         /*
 651          * With per-cpu counters, this becomes an interesting
 652          * problem. we needto work out if we are freeing or allocation
 653          * blocks first, then we can do the modification as necessary.
 654          *
 655          * We do this under the m_sb_lock so that if we are near
 656          * ENOSPC, we will hold out any changes while we work out
 657          * what to do. This means that the amount of free space can
 658          * change while we do this, so we need to retry if we end up
 659          * trying to reserve more space than is available.
 660          *
 661          * We also use the xfs_mod_incore_sb() interface so that we
 662          * don't have to care about whether per cpu counter are
 663          * enabled, disabled or even compiled in....
 664          */
 665 retry:
 666         spin_lock(&mp->m_sb_lock);
 667         xfs_icsb_sync_counters_locked(mp, 0);
 668
 669         /*
 670          * If our previous reservation was larger than the current value,
 671          * then move any unused blocks back to the free pool.
 672          */
 673         fdblks_delta = 0;
 674         if (mp->m_resblks > request) {
 675                 lcounter = mp->m_resblks_avail - request;
 676                 if (lcounter  > 0) {            /* release unused blocks */
 677                         fdblks_delta = lcounter;
 678                         mp->m_resblks_avail -= lcounter;
 679                 }
 680                 mp->m_resblks = request;
 681         } else {
 682                 __int64_t       free;
 683
 684                 free =  mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
 685                 if (!free)
 686                         goto out; /* ENOSPC and fdblks_delta = 0 */
 687
 688                 delta = request - mp->m_resblks;
 689                 lcounter = free - delta;
 690                 if (lcounter < 0) {
 691                         /* We can't satisfy the request, just get what we can */
 692                         mp->m_resblks += free;
 693                         mp->m_resblks_avail += free;
 694                         fdblks_delta = -free;
 695                 } else {
 696                         fdblks_delta = -delta;
 697                         mp->m_resblks = request;
 698                         mp->m_resblks_avail += delta;
 699                 }
 700         }
 701 out:
 702         if (outval) {
 703                 outval->resblks = mp->m_resblks;
 704                 outval->resblks_avail = mp->m_resblks_avail;
 705         }
 706         spin_unlock(&mp->m_sb_lock);
 707
 708         if (fdblks_delta) {
 709                 /*
 710                  * If we are putting blocks back here, m_resblks_avail is
 711                  * already at its max so this will put it in the free pool.
 712                  *
 713                  * If we need space, we'll either succeed in getting it
 714                  * from the free block count or we'll get an enospc. If
 715                  * we get a ENOSPC, it means things changed while we were
 716                  * calculating fdblks_delta and so we should try again to
 717                  * see if there is anything left to reserve.
 718                  *
 719                  * Don't set the reserved flag here - we don't want to reserve
 720                  * the extra reserve blocks from the reserve.....
 721                  */
 722                 int error;
 723                 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
 724                                                  fdblks_delta, 0);
 725                 if (error == ENOSPC)
 726                         goto retry;
 727         }
 728         return 0;
 729 }
 730
 731 /*
 732  * Dump a transaction into the log that contains no real change. This is needed
 733  * to be able to make the log dirty or stamp the current tail LSN into the log
 734  * during the covering operation.
 735  *
 736  * We cannot use an inode here for this - that will push dirty state back up
 737  * into the VFS and then periodic inode flushing will prevent log covering from
 738  * making progress. Hence we log a field in the superblock instead and use a
 739  * synchronous transaction to ensure the superblock is immediately unpinned
 740  * and can be written back.
 741  */
 742 int
 743 xfs_fs_log_dummy(
 744         xfs_mount_t     *mp)
 745 {
 746         xfs_trans_t     *tp;
 747         int             error;
 748
 749         tp = _xfs_trans_alloc(mp, XFS_TRANS_DUMMY1, KM_SLEEP);
 750         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
 751         if (error) {
 752                 xfs_trans_cancel(tp, 0);
 753                 return error;
 754         }
 755
 756         /* log the UUID because it is an unchanging field */
 757         xfs_mod_sb(tp, XFS_SB_UUID);
 758         xfs_trans_set_sync(tp);
 759         return xfs_trans_commit(tp, 0);
 760 }
 761
 762 int
 763 xfs_fs_goingdown(
 764         xfs_mount_t     *mp,
 765         __uint32_t      inflags)
 766 {
 767         switch (inflags) {
 768         case XFS_FSOP_GOING_FLAGS_DEFAULT: {
 769                 struct super_block *sb = freeze_bdev(mp->m_super->s_bdev);
 770
 771                 if (sb && !IS_ERR(sb)) {
 772                         xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
 773                         thaw_bdev(sb->s_bdev, sb);
 774                 }
 775
 776                 break;
 777         }
 778         case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
 779                 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
 780                 break;
 781         case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
 782                 xfs_force_shutdown(mp,
 783                                 SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
 784                 break;
 785         default:
 786                 return XFS_ERROR(EINVAL);
 787         }
 788
 789         return 0;
 790 }
 791
 792 /*
 793  * Force a shutdown of the filesystem instantly while keeping the filesystem
 794  * consistent. We don't do an unmount here; just shutdown the shop, make sure
 795  * that absolutely nothing persistent happens to this filesystem after this
 796  * point.
 797  */
 798 void
 799 xfs_do_force_shutdown(
 800         xfs_mount_t     *mp,
 801         int             flags,
 802         char            *fname,
 803         int             lnnum)
 804 {
 805         int             logerror;
 806
 807         logerror = flags & SHUTDOWN_LOG_IO_ERROR;
 808
 809         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 810                 xfs_notice(mp,
 811         "%s(0x%x) called from line %d of file %s.  Return address = 0x%p",
 812                         __func__, flags, lnnum, fname, __return_address);
 813         }
 814         /*
 815          * No need to duplicate efforts.
 816          */
 817         if (XFS_FORCED_SHUTDOWN(mp) && !logerror)
 818                 return;
 819
 820         /*
 821          * This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't
 822          * queue up anybody new on the log reservations, and wakes up
 823          * everybody who's sleeping on log reservations to tell them
 824          * the bad news.
 825          */
 826         if (xfs_log_force_umount(mp, logerror))
 827                 return;
 828
 829         if (flags & SHUTDOWN_CORRUPT_INCORE) {
 830                 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_CORRUPT,
 831     "Corruption of in-memory data detected.  Shutting down filesystem");
 832                 if (XFS_ERRLEVEL_HIGH <= xfs_error_level)
 833                         xfs_stack_trace();
 834         } else if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 835                 if (logerror) {
 836                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_LOGERROR,
 837                 "Log I/O Error Detected.  Shutting down filesystem");
 838                 } else if (flags & SHUTDOWN_DEVICE_REQ) {
 839                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
 840                 "All device paths lost.  Shutting down filesystem");
 841                 } else if (!(flags & SHUTDOWN_REMOTE_REQ)) {
 842                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
 843                 "I/O Error Detected. Shutting down filesystem");
 844                 }
 845         }
 846         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 847                 xfs_alert(mp,
 848         "Please umount the filesystem and rectify the problem(s)");
 849         }
 850 }