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 = XFS_FSOP_GEOM_FLAGS_NLINK |
  78                              XFS_FSOP_GEOM_FLAGS_DIRV2 |
  79                         (xfs_sb_version_hasattr(&mp->m_sb) ?
  80                                 XFS_FSOP_GEOM_FLAGS_ATTR : 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_hassector(&mp->m_sb) ?
  92                                 XFS_FSOP_GEOM_FLAGS_SECTOR : 0) |
  93                         (xfs_sb_version_hasasciici(&mp->m_sb) ?
  94                                 XFS_FSOP_GEOM_FLAGS_DIRV2CI : 0) |
  95                         (xfs_sb_version_haslazysbcount(&mp->m_sb) ?
  96                                 XFS_FSOP_GEOM_FLAGS_LAZYSB : 0) |
  97                         (xfs_sb_version_hasattr2(&mp->m_sb) ?
  98                                 XFS_FSOP_GEOM_FLAGS_ATTR2 : 0) |
  99                         (xfs_sb_version_hasprojid32bit(&mp->m_sb) ?
 100                                 XFS_FSOP_GEOM_FLAGS_PROJID32 : 0) |
 101                         (xfs_sb_version_hascrc(&mp->m_sb) ?
 102                                 XFS_FSOP_GEOM_FLAGS_V5SB : 0) |
 103                         (xfs_sb_version_hasftype(&mp->m_sb) ?
 104                                 XFS_FSOP_GEOM_FLAGS_FTYPE : 0);
 105                 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, saved_error = 0;
 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                 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         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_growdata,
 207                                   XFS_GROWFS_SPACE_RES(mp), 0);
 208         if (error) {
 209                 xfs_trans_cancel(tp, 0);
 210                 return error;
 211         }
 212
 213         /*
 214          * Write new AG headers to disk. Non-transactional, but written
 215          * synchronously so they are completed prior to the growfs transaction
 216          * being logged.
 217          */
 218         nfree = 0;
 219         for (agno = nagcount - 1; agno >= oagcount; agno--, new -= agsize) {
 220                 __be32  *agfl_bno;
 221
 222                 /*
 223                  * AG freespace header block
 224                  */
 225                 bp = xfs_growfs_get_hdr_buf(mp,
 226                                 XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
 227                                 XFS_FSS_TO_BB(mp, 1), 0,
 228                                 &xfs_agf_buf_ops);
 229                 if (!bp) {
 230                         error = ENOMEM;
 231                         goto error0;
 232                 }
 233
 234                 agf = XFS_BUF_TO_AGF(bp);
 235                 agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
 236                 agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
 237                 agf->agf_seqno = cpu_to_be32(agno);
 238                 if (agno == nagcount - 1)
 239                         agsize =
 240                                 nb -
 241                                 (agno * (xfs_rfsblock_t)mp->m_sb.sb_agblocks);
 242                 else
 243                         agsize = mp->m_sb.sb_agblocks;
 244                 agf->agf_length = cpu_to_be32(agsize);
 245                 agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
 246                 agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
 247                 agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
 248                 agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
 249                 agf->agf_flfirst = 0;
 250                 agf->agf_fllast = cpu_to_be32(XFS_AGFL_SIZE(mp) - 1);
 251                 agf->agf_flcount = 0;
 252                 tmpsize = agsize - XFS_PREALLOC_BLOCKS(mp);
 253                 agf->agf_freeblks = cpu_to_be32(tmpsize);
 254                 agf->agf_longest = cpu_to_be32(tmpsize);
 255                 if (xfs_sb_version_hascrc(&mp->m_sb))
 256                         uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_uuid);
 257
 258                 error = xfs_bwrite(bp);
 259                 xfs_buf_relse(bp);
 260                 if (error)
 261                         goto error0;
 262
 263                 /*
 264                  * AG freelist header block
 265                  */
 266                 bp = xfs_growfs_get_hdr_buf(mp,
 267                                 XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)),
 268                                 XFS_FSS_TO_BB(mp, 1), 0,
 269                                 &xfs_agfl_buf_ops);
 270                 if (!bp) {
 271                         error = ENOMEM;
 272                         goto error0;
 273                 }
 274
 275                 agfl = XFS_BUF_TO_AGFL(bp);
 276                 if (xfs_sb_version_hascrc(&mp->m_sb)) {
 277                         agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
 278                         agfl->agfl_seqno = cpu_to_be32(agno);
 279                         uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_uuid);
 280                 }
 281
 282                 agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, bp);
 283                 for (bucket = 0; bucket < XFS_AGFL_SIZE(mp); bucket++)
 284                         agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
 285
 286                 error = xfs_bwrite(bp);
 287                 xfs_buf_relse(bp);
 288                 if (error)
 289                         goto error0;
 290
 291                 /*
 292                  * AG inode header block
 293                  */
 294                 bp = xfs_growfs_get_hdr_buf(mp,
 295                                 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
 296                                 XFS_FSS_TO_BB(mp, 1), 0,
 297                                 &xfs_agi_buf_ops);
 298                 if (!bp) {
 299                         error = ENOMEM;
 300                         goto error0;
 301                 }
 302
 303                 agi = XFS_BUF_TO_AGI(bp);
 304                 agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
 305                 agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
 306                 agi->agi_seqno = cpu_to_be32(agno);
 307                 agi->agi_length = cpu_to_be32(agsize);
 308                 agi->agi_count = 0;
 309                 agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
 310                 agi->agi_level = cpu_to_be32(1);
 311                 agi->agi_freecount = 0;
 312                 agi->agi_newino = cpu_to_be32(NULLAGINO);
 313                 agi->agi_dirino = cpu_to_be32(NULLAGINO);
 314                 if (xfs_sb_version_hascrc(&mp->m_sb))
 315                         uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_uuid);
 316                 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
 317                         agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
 318
 319                 error = xfs_bwrite(bp);
 320                 xfs_buf_relse(bp);
 321                 if (error)
 322                         goto error0;
 323
 324                 /*
 325                  * BNO btree root block
 326                  */
 327                 bp = xfs_growfs_get_hdr_buf(mp,
 328                                 XFS_AGB_TO_DADDR(mp, agno, XFS_BNO_BLOCK(mp)),
 329                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 330                                 &xfs_allocbt_buf_ops);
 331
 332                 if (!bp) {
 333                         error = ENOMEM;
 334                         goto error0;
 335                 }
 336
 337                 if (xfs_sb_version_hascrc(&mp->m_sb))
 338                         xfs_btree_init_block(mp, bp, XFS_ABTB_CRC_MAGIC, 0, 1,
 339                                                 agno, XFS_BTREE_CRC_BLOCKS);
 340                 else
 341                         xfs_btree_init_block(mp, bp, XFS_ABTB_MAGIC, 0, 1,
 342                                                 agno, 0);
 343
 344                 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
 345                 arec->ar_startblock = cpu_to_be32(XFS_PREALLOC_BLOCKS(mp));
 346                 arec->ar_blockcount = cpu_to_be32(
 347                         agsize - be32_to_cpu(arec->ar_startblock));
 348
 349                 error = xfs_bwrite(bp);
 350                 xfs_buf_relse(bp);
 351                 if (error)
 352                         goto error0;
 353
 354                 /*
 355                  * CNT btree root block
 356                  */
 357                 bp = xfs_growfs_get_hdr_buf(mp,
 358                                 XFS_AGB_TO_DADDR(mp, agno, XFS_CNT_BLOCK(mp)),
 359                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 360                                 &xfs_allocbt_buf_ops);
 361                 if (!bp) {
 362                         error = ENOMEM;
 363                         goto error0;
 364                 }
 365
 366                 if (xfs_sb_version_hascrc(&mp->m_sb))
 367                         xfs_btree_init_block(mp, bp, XFS_ABTC_CRC_MAGIC, 0, 1,
 368                                                 agno, XFS_BTREE_CRC_BLOCKS);
 369                 else
 370                         xfs_btree_init_block(mp, bp, XFS_ABTC_MAGIC, 0, 1,
 371                                                 agno, 0);
 372
 373                 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
 374                 arec->ar_startblock = cpu_to_be32(XFS_PREALLOC_BLOCKS(mp));
 375                 arec->ar_blockcount = cpu_to_be32(
 376                         agsize - be32_to_cpu(arec->ar_startblock));
 377                 nfree += be32_to_cpu(arec->ar_blockcount);
 378
 379                 error = xfs_bwrite(bp);
 380                 xfs_buf_relse(bp);
 381                 if (error)
 382                         goto error0;
 383
 384                 /*
 385                  * INO btree root block
 386                  */
 387                 bp = xfs_growfs_get_hdr_buf(mp,
 388                                 XFS_AGB_TO_DADDR(mp, agno, XFS_IBT_BLOCK(mp)),
 389                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 390                                 &xfs_inobt_buf_ops);
 391                 if (!bp) {
 392                         error = ENOMEM;
 393                         goto error0;
 394                 }
 395
 396                 if (xfs_sb_version_hascrc(&mp->m_sb))
 397                         xfs_btree_init_block(mp, bp, XFS_IBT_CRC_MAGIC, 0, 0,
 398                                                 agno, XFS_BTREE_CRC_BLOCKS);
 399                 else
 400                         xfs_btree_init_block(mp, bp, XFS_IBT_MAGIC, 0, 0,
 401                                                 agno, 0);
 402
 403                 error = xfs_bwrite(bp);
 404                 xfs_buf_relse(bp);
 405                 if (error)
 406                         goto error0;
 407         }
 408         xfs_trans_agblocks_delta(tp, nfree);
 409         /*
 410          * There are new blocks in the old last a.g.
 411          */
 412         if (new) {
 413                 /*
 414                  * Change the agi length.
 415                  */
 416                 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
 417                 if (error) {
 418                         goto error0;
 419                 }
 420                 ASSERT(bp);
 421                 agi = XFS_BUF_TO_AGI(bp);
 422                 be32_add_cpu(&agi->agi_length, new);
 423                 ASSERT(nagcount == oagcount ||
 424                        be32_to_cpu(agi->agi_length) == mp->m_sb.sb_agblocks);
 425                 xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
 426                 /*
 427                  * Change agf length.
 428                  */
 429                 error = xfs_alloc_read_agf(mp, tp, agno, 0, &bp);
 430                 if (error) {
 431                         goto error0;
 432                 }
 433                 ASSERT(bp);
 434                 agf = XFS_BUF_TO_AGF(bp);
 435                 be32_add_cpu(&agf->agf_length, new);
 436                 ASSERT(be32_to_cpu(agf->agf_length) ==
 437                        be32_to_cpu(agi->agi_length));
 438
 439                 xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
 440                 /*
 441                  * Free the new space.
 442                  */
 443                 error = xfs_free_extent(tp, XFS_AGB_TO_FSB(mp, agno,
 444                         be32_to_cpu(agf->agf_length) - new), new);
 445                 if (error) {
 446                         goto error0;
 447                 }
 448         }
 449
 450         /*
 451          * Update changed superblock fields transactionally. These are not
 452          * seen by the rest of the world until the transaction commit applies
 453          * them atomically to the superblock.
 454          */
 455         if (nagcount > oagcount)
 456                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
 457         if (nb > mp->m_sb.sb_dblocks)
 458                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS,
 459                                  nb - mp->m_sb.sb_dblocks);
 460         if (nfree)
 461                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, nfree);
 462         if (dpct)
 463                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
 464         error = xfs_trans_commit(tp, 0);
 465         if (error)
 466                 return error;
 467
 468         /* New allocation groups fully initialized, so update mount struct */
 469         if (nagimax)
 470                 mp->m_maxagi = nagimax;
 471         if (mp->m_sb.sb_imax_pct) {
 472                 __uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
 473                 do_div(icount, 100);
 474                 mp->m_maxicount = icount << mp->m_sb.sb_inopblog;
 475         } else
 476                 mp->m_maxicount = 0;
 477         xfs_set_low_space_thresholds(mp);
 478
 479         /* update secondary superblocks. */
 480         for (agno = 1; agno < nagcount; agno++) {
 481                 error = 0;
 482                 /*
 483                  * new secondary superblocks need to be zeroed, not read from
 484                  * disk as the contents of the new area we are growing into is
 485                  * completely unknown.
 486                  */
 487                 if (agno < oagcount) {
 488                         error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
 489                                   XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
 490                                   XFS_FSS_TO_BB(mp, 1), 0, &bp,
 491                                   &xfs_sb_buf_ops);
 492                 } else {
 493                         bp = xfs_trans_get_buf(NULL, mp->m_ddev_targp,
 494                                   XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
 495                                   XFS_FSS_TO_BB(mp, 1), 0);
 496                         if (bp) {
 497                                 bp->b_ops = &xfs_sb_buf_ops;
 498                                 xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
 499                         } else
 500                                 error = ENOMEM;
 501                 }
 502
 503                 /*
 504                  * If we get an error reading or writing alternate superblocks,
 505                  * continue.  xfs_repair chooses the "best" superblock based
 506                  * on most matches; if we break early, we'll leave more
 507                  * superblocks un-updated than updated, and xfs_repair may
 508                  * pick them over the properly-updated primary.
 509                  */
 510                 if (error) {
 511                         xfs_warn(mp,
 512                 "error %d reading secondary superblock for ag %d",
 513                                 error, agno);
 514                         saved_error = error;
 515                         continue;
 516                 }
 517                 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, XFS_SB_ALL_BITS);
 518
 519                 error = xfs_bwrite(bp);
 520                 xfs_buf_relse(bp);
 521                 if (error) {
 522                         xfs_warn(mp,
 523                 "write error %d updating secondary superblock for ag %d",
 524                                 error, agno);
 525                         saved_error = error;
 526                         continue;
 527                 }
 528         }
 529         return saved_error ? saved_error : error;
 530
 531  error0:
 532         xfs_trans_cancel(tp, XFS_TRANS_ABORT);
 533         return error;
 534 }
 535
 536 static int
 537 xfs_growfs_log_private(
 538         xfs_mount_t             *mp,    /* mount point for filesystem */
 539         xfs_growfs_log_t        *in)    /* growfs log input struct */
 540 {
 541         xfs_extlen_t            nb;
 542
 543         nb = in->newblocks;
 544         if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
 545                 return XFS_ERROR(EINVAL);
 546         if (nb == mp->m_sb.sb_logblocks &&
 547             in->isint == (mp->m_sb.sb_logstart != 0))
 548                 return XFS_ERROR(EINVAL);
 549         /*
 550          * Moving the log is hard, need new interfaces to sync
 551          * the log first, hold off all activity while moving it.
 552          * Can have shorter or longer log in the same space,
 553          * or transform internal to external log or vice versa.
 554          */
 555         return XFS_ERROR(ENOSYS);
 556 }
 557
 558 /*
 559  * protected versions of growfs function acquire and release locks on the mount
 560  * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
 561  * XFS_IOC_FSGROWFSRT
 562  */
 563
 564
 565 int
 566 xfs_growfs_data(
 567         xfs_mount_t             *mp,
 568         xfs_growfs_data_t       *in)
 569 {
 570         int error;
 571
 572         if (!capable(CAP_SYS_ADMIN))
 573                 return XFS_ERROR(EPERM);
 574         if (!mutex_trylock(&mp->m_growlock))
 575                 return XFS_ERROR(EWOULDBLOCK);
 576         error = xfs_growfs_data_private(mp, in);
 577         mutex_unlock(&mp->m_growlock);
 578         return error;
 579 }
 580
 581 int
 582 xfs_growfs_log(
 583         xfs_mount_t             *mp,
 584         xfs_growfs_log_t        *in)
 585 {
 586         int error;
 587
 588         if (!capable(CAP_SYS_ADMIN))
 589                 return XFS_ERROR(EPERM);
 590         if (!mutex_trylock(&mp->m_growlock))
 591                 return XFS_ERROR(EWOULDBLOCK);
 592         error = xfs_growfs_log_private(mp, in);
 593         mutex_unlock(&mp->m_growlock);
 594         return error;
 595 }
 596
 597 /*
 598  * exported through ioctl XFS_IOC_FSCOUNTS
 599  */
 600
 601 int
 602 xfs_fs_counts(
 603         xfs_mount_t             *mp,
 604         xfs_fsop_counts_t       *cnt)
 605 {
 606         xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
 607         spin_lock(&mp->m_sb_lock);
 608         cnt->freedata = mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
 609         cnt->freertx = mp->m_sb.sb_frextents;
 610         cnt->freeino = mp->m_sb.sb_ifree;
 611         cnt->allocino = mp->m_sb.sb_icount;
 612         spin_unlock(&mp->m_sb_lock);
 613         return 0;
 614 }
 615
 616 /*
 617  * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
 618  *
 619  * xfs_reserve_blocks is called to set m_resblks
 620  * in the in-core mount table. The number of unused reserved blocks
 621  * is kept in m_resblks_avail.
 622  *
 623  * Reserve the requested number of blocks if available. Otherwise return
 624  * as many as possible to satisfy the request. The actual number
 625  * reserved are returned in outval
 626  *
 627  * A null inval pointer indicates that only the current reserved blocks
 628  * available  should  be returned no settings are changed.
 629  */
 630
 631 int
 632 xfs_reserve_blocks(
 633         xfs_mount_t             *mp,
 634         __uint64_t              *inval,
 635         xfs_fsop_resblks_t      *outval)
 636 {
 637         __int64_t               lcounter, delta, fdblks_delta;
 638         __uint64_t              request;
 639
 640         /* If inval is null, report current values and return */
 641         if (inval == (__uint64_t *)NULL) {
 642                 if (!outval)
 643                         return EINVAL;
 644                 outval->resblks = mp->m_resblks;
 645                 outval->resblks_avail = mp->m_resblks_avail;
 646                 return 0;
 647         }
 648
 649         request = *inval;
 650
 651         /*
 652          * With per-cpu counters, this becomes an interesting
 653          * problem. we needto work out if we are freeing or allocation
 654          * blocks first, then we can do the modification as necessary.
 655          *
 656          * We do this under the m_sb_lock so that if we are near
 657          * ENOSPC, we will hold out any changes while we work out
 658          * what to do. This means that the amount of free space can
 659          * change while we do this, so we need to retry if we end up
 660          * trying to reserve more space than is available.
 661          *
 662          * We also use the xfs_mod_incore_sb() interface so that we
 663          * don't have to care about whether per cpu counter are
 664          * enabled, disabled or even compiled in....
 665          */
 666 retry:
 667         spin_lock(&mp->m_sb_lock);
 668         xfs_icsb_sync_counters_locked(mp, 0);
 669
 670         /*
 671          * If our previous reservation was larger than the current value,
 672          * then move any unused blocks back to the free pool.
 673          */
 674         fdblks_delta = 0;
 675         if (mp->m_resblks > request) {
 676                 lcounter = mp->m_resblks_avail - request;
 677                 if (lcounter  > 0) {            /* release unused blocks */
 678                         fdblks_delta = lcounter;
 679                         mp->m_resblks_avail -= lcounter;
 680                 }
 681                 mp->m_resblks = request;
 682         } else {
 683                 __int64_t       free;
 684
 685                 free =  mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
 686                 if (!free)
 687                         goto out; /* ENOSPC and fdblks_delta = 0 */
 688
 689                 delta = request - mp->m_resblks;
 690                 lcounter = free - delta;
 691                 if (lcounter < 0) {
 692                         /* We can't satisfy the request, just get what we can */
 693                         mp->m_resblks += free;
 694                         mp->m_resblks_avail += free;
 695                         fdblks_delta = -free;
 696                 } else {
 697                         fdblks_delta = -delta;
 698                         mp->m_resblks = request;
 699                         mp->m_resblks_avail += delta;
 700                 }
 701         }
 702 out:
 703         if (outval) {
 704                 outval->resblks = mp->m_resblks;
 705                 outval->resblks_avail = mp->m_resblks_avail;
 706         }
 707         spin_unlock(&mp->m_sb_lock);
 708
 709         if (fdblks_delta) {
 710                 /*
 711                  * If we are putting blocks back here, m_resblks_avail is
 712                  * already at its max so this will put it in the free pool.
 713                  *
 714                  * If we need space, we'll either succeed in getting it
 715                  * from the free block count or we'll get an enospc. If
 716                  * we get a ENOSPC, it means things changed while we were
 717                  * calculating fdblks_delta and so we should try again to
 718                  * see if there is anything left to reserve.
 719                  *
 720                  * Don't set the reserved flag here - we don't want to reserve
 721                  * the extra reserve blocks from the reserve.....
 722                  */
 723                 int error;
 724                 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
 725                                                  fdblks_delta, 0);
 726                 if (error == ENOSPC)
 727                         goto retry;
 728         }
 729         return 0;
 730 }
 731
 732 /*
 733  * Dump a transaction into the log that contains no real change. This is needed
 734  * to be able to make the log dirty or stamp the current tail LSN into the log
 735  * during the covering operation.
 736  *
 737  * We cannot use an inode here for this - that will push dirty state back up
 738  * into the VFS and then periodic inode flushing will prevent log covering from
 739  * making progress. Hence we log a field in the superblock instead and use a
 740  * synchronous transaction to ensure the superblock is immediately unpinned
 741  * and can be written back.
 742  */
 743 int
 744 xfs_fs_log_dummy(
 745         xfs_mount_t     *mp)
 746 {
 747         xfs_trans_t     *tp;
 748         int             error;
 749
 750         tp = _xfs_trans_alloc(mp, XFS_TRANS_DUMMY1, KM_SLEEP);
 751         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
 752         if (error) {
 753                 xfs_trans_cancel(tp, 0);
 754                 return error;
 755         }
 756
 757         /* log the UUID because it is an unchanging field */
 758         xfs_mod_sb(tp, XFS_SB_UUID);
 759         xfs_trans_set_sync(tp);
 760         return xfs_trans_commit(tp, 0);
 761 }
 762
 763 int
 764 xfs_fs_goingdown(
 765         xfs_mount_t     *mp,
 766         __uint32_t      inflags)
 767 {
 768         switch (inflags) {
 769         case XFS_FSOP_GOING_FLAGS_DEFAULT: {
 770                 struct super_block *sb = freeze_bdev(mp->m_super->s_bdev);
 771
 772                 if (sb && !IS_ERR(sb)) {
 773                         xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
 774                         thaw_bdev(sb->s_bdev, sb);
 775                 }
 776
 777                 break;
 778         }
 779         case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
 780                 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
 781                 break;
 782         case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
 783                 xfs_force_shutdown(mp,
 784                                 SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
 785                 break;
 786         default:
 787                 return XFS_ERROR(EINVAL);
 788         }
 789
 790         return 0;
 791 }
 792
 793 /*
 794  * Force a shutdown of the filesystem instantly while keeping the filesystem
 795  * consistent. We don't do an unmount here; just shutdown the shop, make sure
 796  * that absolutely nothing persistent happens to this filesystem after this
 797  * point.
 798  */
 799 void
 800 xfs_do_force_shutdown(
 801         xfs_mount_t     *mp,
 802         int             flags,
 803         char            *fname,
 804         int             lnnum)
 805 {
 806         int             logerror;
 807
 808         logerror = flags & SHUTDOWN_LOG_IO_ERROR;
 809
 810         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 811                 xfs_notice(mp,
 812         "%s(0x%x) called from line %d of file %s.  Return address = 0x%p",
 813                         __func__, flags, lnnum, fname, __return_address);
 814         }
 815         /*
 816          * No need to duplicate efforts.
 817          */
 818         if (XFS_FORCED_SHUTDOWN(mp) && !logerror)
 819                 return;
 820
 821         /*
 822          * This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't
 823          * queue up anybody new on the log reservations, and wakes up
 824          * everybody who's sleeping on log reservations to tell them
 825          * the bad news.
 826          */
 827         if (xfs_log_force_umount(mp, logerror))
 828                 return;
 829
 830         if (flags & SHUTDOWN_CORRUPT_INCORE) {
 831                 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_CORRUPT,
 832     "Corruption of in-memory data detected.  Shutting down filesystem");
 833                 if (XFS_ERRLEVEL_HIGH <= xfs_error_level)
 834                         xfs_stack_trace();
 835         } else if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 836                 if (logerror) {
 837                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_LOGERROR,
 838                 "Log I/O Error Detected.  Shutting down filesystem");
 839                 } else if (flags & SHUTDOWN_DEVICE_REQ) {
 840                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
 841                 "All device paths lost.  Shutting down filesystem");
 842                 } else if (!(flags & SHUTDOWN_REMOTE_REQ)) {
 843                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
 844                 "I/O Error Detected. Shutting down filesystem");
 845                 }
 846         }
 847         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 848                 xfs_alert(mp,
 849         "Please umount the filesystem and rectify the problem(s)");
 850         }
 851 }