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_log_format.h"
  22 #include "xfs_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_bmap_btree.h"
  28 #include "xfs_alloc_btree.h"
  29 #include "xfs_ialloc_btree.h"
  30 #include "xfs_dinode.h"
  31 #include "xfs_inode.h"
  32 #include "xfs_trans.h"
  33 #include "xfs_inode_item.h"
  34 #include "xfs_btree.h"
  35 #include "xfs_error.h"
  36 #include "xfs_alloc.h"
  37 #include "xfs_ialloc.h"
  38 #include "xfs_fsops.h"
  39 #include "xfs_itable.h"
  40 #include "xfs_trans_space.h"
  41 #include "xfs_rtalloc.h"
  42 #include "xfs_filestream.h"
  43 #include "xfs_trace.h"
  44 #include "xfs_log.h"
  45
  46 /*
  47  * File system operations
  48  */
  49
  50 int
  51 xfs_fs_geometry(
  52         xfs_mount_t             *mp,
  53         xfs_fsop_geom_t         *geo,
  54         int                     new_version)
  55 {
  56
  57         memset(geo, 0, sizeof(*geo));
  58
  59         geo->blocksize = mp->m_sb.sb_blocksize;
  60         geo->rtextsize = mp->m_sb.sb_rextsize;
  61         geo->agblocks = mp->m_sb.sb_agblocks;
  62         geo->agcount = mp->m_sb.sb_agcount;
  63         geo->logblocks = mp->m_sb.sb_logblocks;
  64         geo->sectsize = mp->m_sb.sb_sectsize;
  65         geo->inodesize = mp->m_sb.sb_inodesize;
  66         geo->imaxpct = mp->m_sb.sb_imax_pct;
  67         geo->datablocks = mp->m_sb.sb_dblocks;
  68         geo->rtblocks = mp->m_sb.sb_rblocks;
  69         geo->rtextents = mp->m_sb.sb_rextents;
  70         geo->logstart = mp->m_sb.sb_logstart;
  71         ASSERT(sizeof(geo->uuid)==sizeof(mp->m_sb.sb_uuid));
  72         memcpy(geo->uuid, &mp->m_sb.sb_uuid, sizeof(mp->m_sb.sb_uuid));
  73         if (new_version >= 2) {
  74                 geo->sunit = mp->m_sb.sb_unit;
  75                 geo->swidth = mp->m_sb.sb_width;
  76         }
  77         if (new_version >= 3) {
  78                 geo->version = XFS_FSOP_GEOM_VERSION;
  79                 geo->flags =
  80                         (xfs_sb_version_hasattr(&mp->m_sb) ?
  81                                 XFS_FSOP_GEOM_FLAGS_ATTR : 0) |
  82                         (xfs_sb_version_hasnlink(&mp->m_sb) ?
  83                                 XFS_FSOP_GEOM_FLAGS_NLINK : 0) |
  84                         (xfs_sb_version_hasquota(&mp->m_sb) ?
  85                                 XFS_FSOP_GEOM_FLAGS_QUOTA : 0) |
  86                         (xfs_sb_version_hasalign(&mp->m_sb) ?
  87                                 XFS_FSOP_GEOM_FLAGS_IALIGN : 0) |
  88                         (xfs_sb_version_hasdalign(&mp->m_sb) ?
  89                                 XFS_FSOP_GEOM_FLAGS_DALIGN : 0) |
  90                         (xfs_sb_version_hasshared(&mp->m_sb) ?
  91                                 XFS_FSOP_GEOM_FLAGS_SHARED : 0) |
  92                         (xfs_sb_version_hasextflgbit(&mp->m_sb) ?
  93                                 XFS_FSOP_GEOM_FLAGS_EXTFLG : 0) |
  94                         (xfs_sb_version_hasdirv2(&mp->m_sb) ?
  95                                 XFS_FSOP_GEOM_FLAGS_DIRV2 : 0) |
  96                         (xfs_sb_version_hassector(&mp->m_sb) ?
  97                                 XFS_FSOP_GEOM_FLAGS_SECTOR : 0) |
  98                         (xfs_sb_version_hasasciici(&mp->m_sb) ?
  99                                 XFS_FSOP_GEOM_FLAGS_DIRV2CI : 0) |
 100                         (xfs_sb_version_haslazysbcount(&mp->m_sb) ?
 101                                 XFS_FSOP_GEOM_FLAGS_LAZYSB : 0) |
 102                         (xfs_sb_version_hasattr2(&mp->m_sb) ?
 103                                 XFS_FSOP_GEOM_FLAGS_ATTR2 : 0) |
 104                         (xfs_sb_version_hasprojid32bit(&mp->m_sb) ?
 105                                 XFS_FSOP_GEOM_FLAGS_PROJID32 : 0) |
 106                         (xfs_sb_version_hascrc(&mp->m_sb) ?
 107                                 XFS_FSOP_GEOM_FLAGS_V5SB : 0) |
 108                         (xfs_sb_version_hasftype(&mp->m_sb) ?
 109                                 XFS_FSOP_GEOM_FLAGS_FTYPE : 0);
 110                 geo->logsectsize = xfs_sb_version_hassector(&mp->m_sb) ?
 111                                 mp->m_sb.sb_logsectsize : BBSIZE;
 112                 geo->rtsectsize = mp->m_sb.sb_blocksize;
 113                 geo->dirblocksize = mp->m_dirblksize;
 114         }
 115         if (new_version >= 4) {
 116                 geo->flags |=
 117                         (xfs_sb_version_haslogv2(&mp->m_sb) ?
 118                                 XFS_FSOP_GEOM_FLAGS_LOGV2 : 0);
 119                 geo->logsunit = mp->m_sb.sb_logsunit;
 120         }
 121         return 0;
 122 }
 123
 124 static struct xfs_buf *
 125 xfs_growfs_get_hdr_buf(
 126         struct xfs_mount        *mp,
 127         xfs_daddr_t             blkno,
 128         size_t                  numblks,
 129         int                     flags,
 130         const struct xfs_buf_ops *ops)
 131 {
 132         struct xfs_buf          *bp;
 133
 134         bp = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, flags);
 135         if (!bp)
 136                 return NULL;
 137
 138         xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
 139         bp->b_bn = blkno;
 140         bp->b_maps[0].bm_bn = blkno;
 141         bp->b_ops = ops;
 142
 143         return bp;
 144 }
 145
 146 static int
 147 xfs_growfs_data_private(
 148         xfs_mount_t             *mp,            /* mount point for filesystem */
 149         xfs_growfs_data_t       *in)            /* growfs data input struct */
 150 {
 151         xfs_agf_t               *agf;
 152         struct xfs_agfl         *agfl;
 153         xfs_agi_t               *agi;
 154         xfs_agnumber_t          agno;
 155         xfs_extlen_t            agsize;
 156         xfs_extlen_t            tmpsize;
 157         xfs_alloc_rec_t         *arec;
 158         xfs_buf_t               *bp;
 159         int                     bucket;
 160         int                     dpct;
 161         int                     error, saved_error = 0;
 162         xfs_agnumber_t          nagcount;
 163         xfs_agnumber_t          nagimax = 0;
 164         xfs_rfsblock_t          nb, nb_mod;
 165         xfs_rfsblock_t          new;
 166         xfs_rfsblock_t          nfree;
 167         xfs_agnumber_t          oagcount;
 168         int                     pct;
 169         xfs_trans_t             *tp;
 170
 171         nb = in->newblocks;
 172         pct = in->imaxpct;
 173         if (nb < mp->m_sb.sb_dblocks || pct < 0 || pct > 100)
 174                 return XFS_ERROR(EINVAL);
 175         if ((error = xfs_sb_validate_fsb_count(&mp->m_sb, nb)))
 176                 return error;
 177         dpct = pct - mp->m_sb.sb_imax_pct;
 178         bp = xfs_buf_read_uncached(mp->m_ddev_targp,
 179                                 XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1),
 180                                 XFS_FSS_TO_BB(mp, 1), 0, NULL);
 181         if (!bp)
 182                 return EIO;
 183         if (bp->b_error) {
 184                 error = bp->b_error;
 185                 xfs_buf_relse(bp);
 186                 return error;
 187         }
 188         xfs_buf_relse(bp);
 189
 190         new = nb;       /* use new as a temporary here */
 191         nb_mod = do_div(new, mp->m_sb.sb_agblocks);
 192         nagcount = new + (nb_mod != 0);
 193         if (nb_mod && nb_mod < XFS_MIN_AG_BLOCKS) {
 194                 nagcount--;
 195                 nb = (xfs_rfsblock_t)nagcount * mp->m_sb.sb_agblocks;
 196                 if (nb < mp->m_sb.sb_dblocks)
 197                         return XFS_ERROR(EINVAL);
 198         }
 199         new = nb - mp->m_sb.sb_dblocks;
 200         oagcount = mp->m_sb.sb_agcount;
 201
 202         /* allocate the new per-ag structures */
 203         if (nagcount > oagcount) {
 204                 error = xfs_initialize_perag(mp, nagcount, &nagimax);
 205                 if (error)
 206                         return error;
 207         }
 208
 209         tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFS);
 210         tp->t_flags |= XFS_TRANS_RESERVE;
 211         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_growdata,
 212                                   XFS_GROWFS_SPACE_RES(mp), 0);
 213         if (error) {
 214                 xfs_trans_cancel(tp, 0);
 215                 return error;
 216         }
 217
 218         /*
 219          * Write new AG headers to disk. Non-transactional, but written
 220          * synchronously so they are completed prior to the growfs transaction
 221          * being logged.
 222          */
 223         nfree = 0;
 224         for (agno = nagcount - 1; agno >= oagcount; agno--, new -= agsize) {
 225                 /*
 226                  * AG freespace header block
 227                  */
 228                 bp = xfs_growfs_get_hdr_buf(mp,
 229                                 XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
 230                                 XFS_FSS_TO_BB(mp, 1), 0,
 231                                 &xfs_agf_buf_ops);
 232                 if (!bp) {
 233                         error = ENOMEM;
 234                         goto error0;
 235                 }
 236
 237                 agf = XFS_BUF_TO_AGF(bp);
 238                 agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
 239                 agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
 240                 agf->agf_seqno = cpu_to_be32(agno);
 241                 if (agno == nagcount - 1)
 242                         agsize =
 243                                 nb -
 244                                 (agno * (xfs_rfsblock_t)mp->m_sb.sb_agblocks);
 245                 else
 246                         agsize = mp->m_sb.sb_agblocks;
 247                 agf->agf_length = cpu_to_be32(agsize);
 248                 agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
 249                 agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
 250                 agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
 251                 agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
 252                 agf->agf_flfirst = 0;
 253                 agf->agf_fllast = cpu_to_be32(XFS_AGFL_SIZE(mp) - 1);
 254                 agf->agf_flcount = 0;
 255                 tmpsize = agsize - XFS_PREALLOC_BLOCKS(mp);
 256                 agf->agf_freeblks = cpu_to_be32(tmpsize);
 257                 agf->agf_longest = cpu_to_be32(tmpsize);
 258                 if (xfs_sb_version_hascrc(&mp->m_sb))
 259                         uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_uuid);
 260
 261                 error = xfs_bwrite(bp);
 262                 xfs_buf_relse(bp);
 263                 if (error)
 264                         goto error0;
 265
 266                 /*
 267                  * AG freelist header block
 268                  */
 269                 bp = xfs_growfs_get_hdr_buf(mp,
 270                                 XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)),
 271                                 XFS_FSS_TO_BB(mp, 1), 0,
 272                                 &xfs_agfl_buf_ops);
 273                 if (!bp) {
 274                         error = ENOMEM;
 275                         goto error0;
 276                 }
 277
 278                 agfl = XFS_BUF_TO_AGFL(bp);
 279                 if (xfs_sb_version_hascrc(&mp->m_sb)) {
 280                         agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
 281                         agfl->agfl_seqno = cpu_to_be32(agno);
 282                         uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_uuid);
 283                 }
 284                 for (bucket = 0; bucket < XFS_AGFL_SIZE(mp); bucket++)
 285                         agfl->agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
 286
 287                 error = xfs_bwrite(bp);
 288                 xfs_buf_relse(bp);
 289                 if (error)
 290                         goto error0;
 291
 292                 /*
 293                  * AG inode header block
 294                  */
 295                 bp = xfs_growfs_get_hdr_buf(mp,
 296                                 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
 297                                 XFS_FSS_TO_BB(mp, 1), 0,
 298                                 &xfs_agi_buf_ops);
 299                 if (!bp) {
 300                         error = ENOMEM;
 301                         goto error0;
 302                 }
 303
 304                 agi = XFS_BUF_TO_AGI(bp);
 305                 agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
 306                 agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
 307                 agi->agi_seqno = cpu_to_be32(agno);
 308                 agi->agi_length = cpu_to_be32(agsize);
 309                 agi->agi_count = 0;
 310                 agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
 311                 agi->agi_level = cpu_to_be32(1);
 312                 agi->agi_freecount = 0;
 313                 agi->agi_newino = cpu_to_be32(NULLAGINO);
 314                 agi->agi_dirino = cpu_to_be32(NULLAGINO);
 315                 if (xfs_sb_version_hascrc(&mp->m_sb))
 316                         uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_uuid);
 317                 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
 318                         agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
 319
 320                 error = xfs_bwrite(bp);
 321                 xfs_buf_relse(bp);
 322                 if (error)
 323                         goto error0;
 324
 325                 /*
 326                  * BNO btree root block
 327                  */
 328                 bp = xfs_growfs_get_hdr_buf(mp,
 329                                 XFS_AGB_TO_DADDR(mp, agno, XFS_BNO_BLOCK(mp)),
 330                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 331                                 &xfs_allocbt_buf_ops);
 332
 333                 if (!bp) {
 334                         error = ENOMEM;
 335                         goto error0;
 336                 }
 337
 338                 if (xfs_sb_version_hascrc(&mp->m_sb))
 339                         xfs_btree_init_block(mp, bp, XFS_ABTB_CRC_MAGIC, 0, 1,
 340                                                 agno, XFS_BTREE_CRC_BLOCKS);
 341                 else
 342                         xfs_btree_init_block(mp, bp, XFS_ABTB_MAGIC, 0, 1,
 343                                                 agno, 0);
 344
 345                 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
 346                 arec->ar_startblock = cpu_to_be32(XFS_PREALLOC_BLOCKS(mp));
 347                 arec->ar_blockcount = cpu_to_be32(
 348                         agsize - be32_to_cpu(arec->ar_startblock));
 349
 350                 error = xfs_bwrite(bp);
 351                 xfs_buf_relse(bp);
 352                 if (error)
 353                         goto error0;
 354
 355                 /*
 356                  * CNT btree root block
 357                  */
 358                 bp = xfs_growfs_get_hdr_buf(mp,
 359                                 XFS_AGB_TO_DADDR(mp, agno, XFS_CNT_BLOCK(mp)),
 360                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 361                                 &xfs_allocbt_buf_ops);
 362                 if (!bp) {
 363                         error = ENOMEM;
 364                         goto error0;
 365                 }
 366
 367                 if (xfs_sb_version_hascrc(&mp->m_sb))
 368                         xfs_btree_init_block(mp, bp, XFS_ABTC_CRC_MAGIC, 0, 1,
 369                                                 agno, XFS_BTREE_CRC_BLOCKS);
 370                 else
 371                         xfs_btree_init_block(mp, bp, XFS_ABTC_MAGIC, 0, 1,
 372                                                 agno, 0);
 373
 374                 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
 375                 arec->ar_startblock = cpu_to_be32(XFS_PREALLOC_BLOCKS(mp));
 376                 arec->ar_blockcount = cpu_to_be32(
 377                         agsize - be32_to_cpu(arec->ar_startblock));
 378                 nfree += be32_to_cpu(arec->ar_blockcount);
 379
 380                 error = xfs_bwrite(bp);
 381                 xfs_buf_relse(bp);
 382                 if (error)
 383                         goto error0;
 384
 385                 /*
 386                  * INO btree root block
 387                  */
 388                 bp = xfs_growfs_get_hdr_buf(mp,
 389                                 XFS_AGB_TO_DADDR(mp, agno, XFS_IBT_BLOCK(mp)),
 390                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 391                                 &xfs_inobt_buf_ops);
 392                 if (!bp) {
 393                         error = ENOMEM;
 394                         goto error0;
 395                 }
 396
 397                 if (xfs_sb_version_hascrc(&mp->m_sb))
 398                         xfs_btree_init_block(mp, bp, XFS_IBT_CRC_MAGIC, 0, 0,
 399                                                 agno, XFS_BTREE_CRC_BLOCKS);
 400                 else
 401                         xfs_btree_init_block(mp, bp, XFS_IBT_MAGIC, 0, 0,
 402                                                 agno, 0);
 403
 404                 error = xfs_bwrite(bp);
 405                 xfs_buf_relse(bp);
 406                 if (error)
 407                         goto error0;
 408         }
 409         xfs_trans_agblocks_delta(tp, nfree);
 410         /*
 411          * There are new blocks in the old last a.g.
 412          */
 413         if (new) {
 414                 /*
 415                  * Change the agi length.
 416                  */
 417                 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
 418                 if (error) {
 419                         goto error0;
 420                 }
 421                 ASSERT(bp);
 422                 agi = XFS_BUF_TO_AGI(bp);
 423                 be32_add_cpu(&agi->agi_length, new);
 424                 ASSERT(nagcount == oagcount ||
 425                        be32_to_cpu(agi->agi_length) == mp->m_sb.sb_agblocks);
 426                 xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
 427                 /*
 428                  * Change agf length.
 429                  */
 430                 error = xfs_alloc_read_agf(mp, tp, agno, 0, &bp);
 431                 if (error) {
 432                         goto error0;
 433                 }
 434                 ASSERT(bp);
 435                 agf = XFS_BUF_TO_AGF(bp);
 436                 be32_add_cpu(&agf->agf_length, new);
 437                 ASSERT(be32_to_cpu(agf->agf_length) ==
 438                        be32_to_cpu(agi->agi_length));
 439
 440                 xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
 441                 /*
 442                  * Free the new space.
 443                  */
 444                 error = xfs_free_extent(tp, XFS_AGB_TO_FSB(mp, agno,
 445                         be32_to_cpu(agf->agf_length) - new), new);
 446                 if (error) {
 447                         goto error0;
 448                 }
 449         }
 450
 451         /*
 452          * Update changed superblock fields transactionally. These are not
 453          * seen by the rest of the world until the transaction commit applies
 454          * them atomically to the superblock.
 455          */
 456         if (nagcount > oagcount)
 457                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
 458         if (nb > mp->m_sb.sb_dblocks)
 459                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS,
 460                                  nb - mp->m_sb.sb_dblocks);
 461         if (nfree)
 462                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, nfree);
 463         if (dpct)
 464                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
 465         error = xfs_trans_commit(tp, 0);
 466         if (error)
 467                 return error;
 468
 469         /* New allocation groups fully initialized, so update mount struct */
 470         if (nagimax)
 471                 mp->m_maxagi = nagimax;
 472         if (mp->m_sb.sb_imax_pct) {
 473                 __uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
 474                 do_div(icount, 100);
 475                 mp->m_maxicount = icount << mp->m_sb.sb_inopblog;
 476         } else
 477                 mp->m_maxicount = 0;
 478         xfs_set_low_space_thresholds(mp);
 479
 480         /* update secondary superblocks. */
 481         for (agno = 1; agno < nagcount; agno++) {
 482                 error = 0;
 483                 /*
 484                  * new secondary superblocks need to be zeroed, not read from
 485                  * disk as the contents of the new area we are growing into is
 486                  * completely unknown.
 487                  */
 488                 if (agno < oagcount) {
 489                         error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
 490                                   XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
 491                                   XFS_FSS_TO_BB(mp, 1), 0, &bp,
 492                                   &xfs_sb_buf_ops);
 493                 } else {
 494                         bp = xfs_trans_get_buf(NULL, mp->m_ddev_targp,
 495                                   XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
 496                                   XFS_FSS_TO_BB(mp, 1), 0);
 497                         if (bp) {
 498                                 bp->b_ops = &xfs_sb_buf_ops;
 499                                 xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
 500                         } else
 501                                 error = ENOMEM;
 502                 }
 503
 504                 /*
 505                  * If we get an error reading or writing alternate superblocks,
 506                  * continue.  xfs_repair chooses the "best" superblock based
 507                  * on most matches; if we break early, we'll leave more
 508                  * superblocks un-updated than updated, and xfs_repair may
 509                  * pick them over the properly-updated primary.
 510                  */
 511                 if (error) {
 512                         xfs_warn(mp,
 513                 "error %d reading secondary superblock for ag %d",
 514                                 error, agno);
 515                         saved_error = error;
 516                         continue;
 517                 }
 518                 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, XFS_SB_ALL_BITS);
 519
 520                 error = xfs_bwrite(bp);
 521                 xfs_buf_relse(bp);
 522                 if (error) {
 523                         xfs_warn(mp,
 524                 "write error %d updating secondary superblock for ag %d",
 525                                 error, agno);
 526                         saved_error = error;
 527                         continue;
 528                 }
 529         }
 530         return saved_error ? saved_error : error;
 531
 532  error0:
 533         xfs_trans_cancel(tp, XFS_TRANS_ABORT);
 534         return error;
 535 }
 536
 537 static int
 538 xfs_growfs_log_private(
 539         xfs_mount_t             *mp,    /* mount point for filesystem */
 540         xfs_growfs_log_t        *in)    /* growfs log input struct */
 541 {
 542         xfs_extlen_t            nb;
 543
 544         nb = in->newblocks;
 545         if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
 546                 return XFS_ERROR(EINVAL);
 547         if (nb == mp->m_sb.sb_logblocks &&
 548             in->isint == (mp->m_sb.sb_logstart != 0))
 549                 return XFS_ERROR(EINVAL);
 550         /*
 551          * Moving the log is hard, need new interfaces to sync
 552          * the log first, hold off all activity while moving it.
 553          * Can have shorter or longer log in the same space,
 554          * or transform internal to external log or vice versa.
 555          */
 556         return XFS_ERROR(ENOSYS);
 557 }
 558
 559 /*
 560  * protected versions of growfs function acquire and release locks on the mount
 561  * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
 562  * XFS_IOC_FSGROWFSRT
 563  */
 564
 565
 566 int
 567 xfs_growfs_data(
 568         xfs_mount_t             *mp,
 569         xfs_growfs_data_t       *in)
 570 {
 571         int error;
 572
 573         if (!capable(CAP_SYS_ADMIN))
 574                 return XFS_ERROR(EPERM);
 575         if (!mutex_trylock(&mp->m_growlock))
 576                 return XFS_ERROR(EWOULDBLOCK);
 577         error = xfs_growfs_data_private(mp, in);
 578         mutex_unlock(&mp->m_growlock);
 579         return error;
 580 }
 581
 582 int
 583 xfs_growfs_log(
 584         xfs_mount_t             *mp,
 585         xfs_growfs_log_t        *in)
 586 {
 587         int error;
 588
 589         if (!capable(CAP_SYS_ADMIN))
 590                 return XFS_ERROR(EPERM);
 591         if (!mutex_trylock(&mp->m_growlock))
 592                 return XFS_ERROR(EWOULDBLOCK);
 593         error = xfs_growfs_log_private(mp, in);
 594         mutex_unlock(&mp->m_growlock);
 595         return error;
 596 }
 597
 598 /*
 599  * exported through ioctl XFS_IOC_FSCOUNTS
 600  */
 601
 602 int
 603 xfs_fs_counts(
 604         xfs_mount_t             *mp,
 605         xfs_fsop_counts_t       *cnt)
 606 {
 607         xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
 608         spin_lock(&mp->m_sb_lock);
 609         cnt->freedata = mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
 610         cnt->freertx = mp->m_sb.sb_frextents;
 611         cnt->freeino = mp->m_sb.sb_ifree;
 612         cnt->allocino = mp->m_sb.sb_icount;
 613         spin_unlock(&mp->m_sb_lock);
 614         return 0;
 615 }
 616
 617 /*
 618  * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
 619  *
 620  * xfs_reserve_blocks is called to set m_resblks
 621  * in the in-core mount table. The number of unused reserved blocks
 622  * is kept in m_resblks_avail.
 623  *
 624  * Reserve the requested number of blocks if available. Otherwise return
 625  * as many as possible to satisfy the request. The actual number
 626  * reserved are returned in outval
 627  *
 628  * A null inval pointer indicates that only the current reserved blocks
 629  * available  should  be returned no settings are changed.
 630  */
 631
 632 int
 633 xfs_reserve_blocks(
 634         xfs_mount_t             *mp,
 635         __uint64_t              *inval,
 636         xfs_fsop_resblks_t      *outval)
 637 {
 638         __int64_t               lcounter, delta, fdblks_delta;
 639         __uint64_t              request;
 640
 641         /* If inval is null, report current values and return */
 642         if (inval == (__uint64_t *)NULL) {
 643                 if (!outval)
 644                         return EINVAL;
 645                 outval->resblks = mp->m_resblks;
 646                 outval->resblks_avail = mp->m_resblks_avail;
 647                 return 0;
 648         }
 649
 650         request = *inval;
 651
 652         /*
 653          * With per-cpu counters, this becomes an interesting
 654          * problem. we needto work out if we are freeing or allocation
 655          * blocks first, then we can do the modification as necessary.
 656          *
 657          * We do this under the m_sb_lock so that if we are near
 658          * ENOSPC, we will hold out any changes while we work out
 659          * what to do. This means that the amount of free space can
 660          * change while we do this, so we need to retry if we end up
 661          * trying to reserve more space than is available.
 662          *
 663          * We also use the xfs_mod_incore_sb() interface so that we
 664          * don't have to care about whether per cpu counter are
 665          * enabled, disabled or even compiled in....
 666          */
 667 retry:
 668         spin_lock(&mp->m_sb_lock);
 669         xfs_icsb_sync_counters_locked(mp, 0);
 670
 671         /*
 672          * If our previous reservation was larger than the current value,
 673          * then move any unused blocks back to the free pool.
 674          */
 675         fdblks_delta = 0;
 676         if (mp->m_resblks > request) {
 677                 lcounter = mp->m_resblks_avail - request;
 678                 if (lcounter  > 0) {            /* release unused blocks */
 679                         fdblks_delta = lcounter;
 680                         mp->m_resblks_avail -= lcounter;
 681                 }
 682                 mp->m_resblks = request;
 683         } else {
 684                 __int64_t       free;
 685
 686                 free =  mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
 687                 if (!free)
 688                         goto out; /* ENOSPC and fdblks_delta = 0 */
 689
 690                 delta = request - mp->m_resblks;
 691                 lcounter = free - delta;
 692                 if (lcounter < 0) {
 693                         /* We can't satisfy the request, just get what we can */
 694                         mp->m_resblks += free;
 695                         mp->m_resblks_avail += free;
 696                         fdblks_delta = -free;
 697                 } else {
 698                         fdblks_delta = -delta;
 699                         mp->m_resblks = request;
 700                         mp->m_resblks_avail += delta;
 701                 }
 702         }
 703 out:
 704         if (outval) {
 705                 outval->resblks = mp->m_resblks;
 706                 outval->resblks_avail = mp->m_resblks_avail;
 707         }
 708         spin_unlock(&mp->m_sb_lock);
 709
 710         if (fdblks_delta) {
 711                 /*
 712                  * If we are putting blocks back here, m_resblks_avail is
 713                  * already at its max so this will put it in the free pool.
 714                  *
 715                  * If we need space, we'll either succeed in getting it
 716                  * from the free block count or we'll get an enospc. If
 717                  * we get a ENOSPC, it means things changed while we were
 718                  * calculating fdblks_delta and so we should try again to
 719                  * see if there is anything left to reserve.
 720                  *
 721                  * Don't set the reserved flag here - we don't want to reserve
 722                  * the extra reserve blocks from the reserve.....
 723                  */
 724                 int error;
 725                 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
 726                                                  fdblks_delta, 0);
 727                 if (error == ENOSPC)
 728                         goto retry;
 729         }
 730         return 0;
 731 }
 732
 733 /*
 734  * Dump a transaction into the log that contains no real change. This is needed
 735  * to be able to make the log dirty or stamp the current tail LSN into the log
 736  * during the covering operation.
 737  *
 738  * We cannot use an inode here for this - that will push dirty state back up
 739  * into the VFS and then periodic inode flushing will prevent log covering from
 740  * making progress. Hence we log a field in the superblock instead and use a
 741  * synchronous transaction to ensure the superblock is immediately unpinned
 742  * and can be written back.
 743  */
 744 int
 745 xfs_fs_log_dummy(
 746         xfs_mount_t     *mp)
 747 {
 748         xfs_trans_t     *tp;
 749         int             error;
 750
 751         tp = _xfs_trans_alloc(mp, XFS_TRANS_DUMMY1, KM_SLEEP);
 752         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
 753         if (error) {
 754                 xfs_trans_cancel(tp, 0);
 755                 return error;
 756         }
 757
 758         /* log the UUID because it is an unchanging field */
 759         xfs_mod_sb(tp, XFS_SB_UUID);
 760         xfs_trans_set_sync(tp);
 761         return xfs_trans_commit(tp, 0);
 762 }
 763
 764 int
 765 xfs_fs_goingdown(
 766         xfs_mount_t     *mp,
 767         __uint32_t      inflags)
 768 {
 769         switch (inflags) {
 770         case XFS_FSOP_GOING_FLAGS_DEFAULT: {
 771                 struct super_block *sb = freeze_bdev(mp->m_super->s_bdev);
 772
 773                 if (sb && !IS_ERR(sb)) {
 774                         xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
 775                         thaw_bdev(sb->s_bdev, sb);
 776                 }
 777
 778                 break;
 779         }
 780         case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
 781                 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
 782                 break;
 783         case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
 784                 xfs_force_shutdown(mp,
 785                                 SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
 786                 break;
 787         default:
 788                 return XFS_ERROR(EINVAL);
 789         }
 790
 791         return 0;
 792 }
 793
 794 /*
 795  * Force a shutdown of the filesystem instantly while keeping the filesystem
 796  * consistent. We don't do an unmount here; just shutdown the shop, make sure
 797  * that absolutely nothing persistent happens to this filesystem after this
 798  * point.
 799  */
 800 void
 801 xfs_do_force_shutdown(
 802         xfs_mount_t     *mp,
 803         int             flags,
 804         char            *fname,
 805         int             lnnum)
 806 {
 807         int             logerror;
 808
 809         logerror = flags & SHUTDOWN_LOG_IO_ERROR;
 810
 811         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 812                 xfs_notice(mp,
 813         "%s(0x%x) called from line %d of file %s.  Return address = 0x%p",
 814                         __func__, flags, lnnum, fname, __return_address);
 815         }
 816         /*
 817          * No need to duplicate efforts.
 818          */
 819         if (XFS_FORCED_SHUTDOWN(mp) && !logerror)
 820                 return;
 821
 822         /*
 823          * This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't
 824          * queue up anybody new on the log reservations, and wakes up
 825          * everybody who's sleeping on log reservations to tell them
 826          * the bad news.
 827          */
 828         if (xfs_log_force_umount(mp, logerror))
 829                 return;
 830
 831         if (flags & SHUTDOWN_CORRUPT_INCORE) {
 832                 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_CORRUPT,
 833     "Corruption of in-memory data detected.  Shutting down filesystem");
 834                 if (XFS_ERRLEVEL_HIGH <= xfs_error_level)
 835                         xfs_stack_trace();
 836         } else if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 837                 if (logerror) {
 838                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_LOGERROR,
 839                 "Log I/O Error Detected.  Shutting down filesystem");
 840                 } else if (flags & SHUTDOWN_DEVICE_REQ) {
 841                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
 842                 "All device paths lost.  Shutting down filesystem");
 843                 } else if (!(flags & SHUTDOWN_REMOTE_REQ)) {
 844                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
 845                 "I/O Error Detected. Shutting down filesystem");
 846                 }
 847         }
 848         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 849                 xfs_alert(mp,
 850         "Please umount the filesystem and rectify the problem(s)");
 851         }
 852 }