fs/xfs/xfs_fsops.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_types.h"
  21 #include "xfs_log.h"
  22 #include "xfs_trans.h"
  23 #include "xfs_sb.h"
  24 #include "xfs_ag.h"
  25 #include "xfs_mount.h"
  26 #include "xfs_bmap_btree.h"
  27 #include "xfs_alloc_btree.h"
  28 #include "xfs_ialloc_btree.h"
  29 #include "xfs_dinode.h"
  30 #include "xfs_inode.h"
  31 #include "xfs_inode_item.h"
  32 #include "xfs_btree.h"
  33 #include "xfs_error.h"
  34 #include "xfs_alloc.h"
  35 #include "xfs_ialloc.h"
  36 #include "xfs_fsops.h"
  37 #include "xfs_itable.h"
  38 #include "xfs_trans_space.h"
  39 #include "xfs_rtalloc.h"
  40 #include "xfs_filestream.h"
  41 #include "xfs_trace.h"
  42
  43 /*
  44  * File system operations
  45  */
  46
  47 int
  48 xfs_fs_geometry(
  49         xfs_mount_t             *mp,
  50         xfs_fsop_geom_t         *geo,
  51         int                     new_version)
  52 {
  53
  54         memset(geo, 0, sizeof(*geo));
  55
  56         geo->blocksize = mp->m_sb.sb_blocksize;
  57         geo->rtextsize = mp->m_sb.sb_rextsize;
  58         geo->agblocks = mp->m_sb.sb_agblocks;
  59         geo->agcount = mp->m_sb.sb_agcount;
  60         geo->logblocks = mp->m_sb.sb_logblocks;
  61         geo->sectsize = mp->m_sb.sb_sectsize;
  62         geo->inodesize = mp->m_sb.sb_inodesize;
  63         geo->imaxpct = mp->m_sb.sb_imax_pct;
  64         geo->datablocks = mp->m_sb.sb_dblocks;
  65         geo->rtblocks = mp->m_sb.sb_rblocks;
  66         geo->rtextents = mp->m_sb.sb_rextents;
  67         geo->logstart = mp->m_sb.sb_logstart;
  68         ASSERT(sizeof(geo->uuid)==sizeof(mp->m_sb.sb_uuid));
  69         memcpy(geo->uuid, &mp->m_sb.sb_uuid, sizeof(mp->m_sb.sb_uuid));
  70         if (new_version >= 2) {
  71                 geo->sunit = mp->m_sb.sb_unit;
  72                 geo->swidth = mp->m_sb.sb_width;
  73         }
  74         if (new_version >= 3) {
  75                 geo->version = XFS_FSOP_GEOM_VERSION;
  76                 geo->flags =
  77                         (xfs_sb_version_hasattr(&mp->m_sb) ?
  78                                 XFS_FSOP_GEOM_FLAGS_ATTR : 0) |
  79                         (xfs_sb_version_hasnlink(&mp->m_sb) ?
  80                                 XFS_FSOP_GEOM_FLAGS_NLINK : 0) |
  81                         (xfs_sb_version_hasquota(&mp->m_sb) ?
  82                                 XFS_FSOP_GEOM_FLAGS_QUOTA : 0) |
  83                         (xfs_sb_version_hasalign(&mp->m_sb) ?
  84                                 XFS_FSOP_GEOM_FLAGS_IALIGN : 0) |
  85                         (xfs_sb_version_hasdalign(&mp->m_sb) ?
  86                                 XFS_FSOP_GEOM_FLAGS_DALIGN : 0) |
  87                         (xfs_sb_version_hasshared(&mp->m_sb) ?
  88                                 XFS_FSOP_GEOM_FLAGS_SHARED : 0) |
  89                         (xfs_sb_version_hasextflgbit(&mp->m_sb) ?
  90                                 XFS_FSOP_GEOM_FLAGS_EXTFLG : 0) |
  91                         (xfs_sb_version_hasdirv2(&mp->m_sb) ?
  92                                 XFS_FSOP_GEOM_FLAGS_DIRV2 : 0) |
  93                         (xfs_sb_version_hassector(&mp->m_sb) ?
  94                                 XFS_FSOP_GEOM_FLAGS_SECTOR : 0) |
  95                         (xfs_sb_version_hasasciici(&mp->m_sb) ?
  96                                 XFS_FSOP_GEOM_FLAGS_DIRV2CI : 0) |
  97                         (xfs_sb_version_haslazysbcount(&mp->m_sb) ?
  98                                 XFS_FSOP_GEOM_FLAGS_LAZYSB : 0) |
  99                         (xfs_sb_version_hasattr2(&mp->m_sb) ?
 100                                 XFS_FSOP_GEOM_FLAGS_ATTR2 : 0) |
 101                         (xfs_sb_version_hasprojid32bit(&mp->m_sb) ?
 102                                 XFS_FSOP_GEOM_FLAGS_PROJID32 : 0) |
 103                         (xfs_sb_version_hascrc(&mp->m_sb) ?
 104                                 XFS_FSOP_GEOM_FLAGS_V5SB : 0) |
 105                         (xfs_sb_version_hasftype(&mp->m_sb) ?
 106                                 XFS_FSOP_GEOM_FLAGS_FTYPE : 0);
 107                 geo->logsectsize = xfs_sb_version_hassector(&mp->m_sb) ?
 108                                 mp->m_sb.sb_logsectsize : BBSIZE;
 109                 geo->rtsectsize = mp->m_sb.sb_blocksize;
 110                 geo->dirblocksize = mp->m_dirblksize;
 111         }
 112         if (new_version >= 4) {
 113                 geo->flags |=
 114                         (xfs_sb_version_haslogv2(&mp->m_sb) ?
 115                                 XFS_FSOP_GEOM_FLAGS_LOGV2 : 0);
 116                 geo->logsunit = mp->m_sb.sb_logsunit;
 117         }
 118         return 0;
 119 }
 120
 121 static struct xfs_buf *
 122 xfs_growfs_get_hdr_buf(
 123         struct xfs_mount        *mp,
 124         xfs_daddr_t             blkno,
 125         size_t                  numblks,
 126         int                     flags,
 127         const struct xfs_buf_ops *ops)
 128 {
 129         struct xfs_buf          *bp;
 130
 131         bp = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, flags);
 132         if (!bp)
 133                 return NULL;
 134
 135         xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
 136         bp->b_bn = blkno;
 137         bp->b_maps[0].bm_bn = blkno;
 138         bp->b_ops = ops;
 139
 140         return bp;
 141 }
 142
 143 static int
 144 xfs_growfs_data_private(
 145         xfs_mount_t             *mp,            /* mount point for filesystem */
 146         xfs_growfs_data_t       *in)            /* growfs data input struct */
 147 {
 148         xfs_agf_t               *agf;
 149         struct xfs_agfl         *agfl;
 150         xfs_agi_t               *agi;
 151         xfs_agnumber_t          agno;
 152         xfs_extlen_t            agsize;
 153         xfs_extlen_t            tmpsize;
 154         xfs_alloc_rec_t         *arec;
 155         xfs_buf_t               *bp;
 156         int                     bucket;
 157         int                     dpct;
 158         int                     error;
 159         xfs_agnumber_t          nagcount;
 160         xfs_agnumber_t          nagimax = 0;
 161         xfs_rfsblock_t          nb, nb_mod;
 162         xfs_rfsblock_t          new;
 163         xfs_rfsblock_t          nfree;
 164         xfs_agnumber_t          oagcount;
 165         int                     pct;
 166         xfs_trans_t             *tp;
 167
 168         nb = in->newblocks;
 169         pct = in->imaxpct;
 170         if (nb < mp->m_sb.sb_dblocks || pct < 0 || pct > 100)
 171                 return XFS_ERROR(EINVAL);
 172         if ((error = xfs_sb_validate_fsb_count(&mp->m_sb, nb)))
 173                 return error;
 174         dpct = pct - mp->m_sb.sb_imax_pct;
 175         bp = xfs_buf_read_uncached(mp->m_ddev_targp,
 176                                 XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1),
 177                                 XFS_FSS_TO_BB(mp, 1), 0, NULL);
 178         if (!bp)
 179                 return EIO;
 180         if (bp->b_error) {
 181                 error = bp->b_error;
 182                 xfs_buf_relse(bp);
 183                 return error;
 184         }
 185         xfs_buf_relse(bp);
 186
 187         new = nb;       /* use new as a temporary here */
 188         nb_mod = do_div(new, mp->m_sb.sb_agblocks);
 189         nagcount = new + (nb_mod != 0);
 190         if (nb_mod && nb_mod < XFS_MIN_AG_BLOCKS) {
 191                 nagcount--;
 192                 nb = (xfs_rfsblock_t)nagcount * mp->m_sb.sb_agblocks;
 193                 if (nb < mp->m_sb.sb_dblocks)
 194                         return XFS_ERROR(EINVAL);
 195         }
 196         new = nb - mp->m_sb.sb_dblocks;
 197         oagcount = mp->m_sb.sb_agcount;
 198
 199         /* allocate the new per-ag structures */
 200         if (nagcount > oagcount) {
 201                 error = xfs_initialize_perag(mp, nagcount, &nagimax);
 202                 if (error)
 203                         return error;
 204         }
 205
 206         tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFS);
 207         tp->t_flags |= XFS_TRANS_RESERVE;
 208         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_growdata,
 209                                   XFS_GROWFS_SPACE_RES(mp), 0);
 210         if (error) {
 211                 xfs_trans_cancel(tp, 0);
 212                 return error;
 213         }
 214
 215         /*
 216          * Write new AG headers to disk. Non-transactional, but written
 217          * synchronously so they are completed prior to the growfs transaction
 218          * being logged.
 219          */
 220         nfree = 0;
 221         for (agno = nagcount - 1; agno >= oagcount; agno--, new -= agsize) {
 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                 for (bucket = 0; bucket < XFS_AGFL_SIZE(mp); bucket++)
 282                         agfl->agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
 283
 284                 error = xfs_bwrite(bp);
 285                 xfs_buf_relse(bp);
 286                 if (error)
 287                         goto error0;
 288
 289                 /*
 290                  * AG inode header block
 291                  */
 292                 bp = xfs_growfs_get_hdr_buf(mp,
 293                                 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
 294                                 XFS_FSS_TO_BB(mp, 1), 0,
 295                                 &xfs_agi_buf_ops);
 296                 if (!bp) {
 297                         error = ENOMEM;
 298                         goto error0;
 299                 }
 300
 301                 agi = XFS_BUF_TO_AGI(bp);
 302                 agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
 303                 agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
 304                 agi->agi_seqno = cpu_to_be32(agno);
 305                 agi->agi_length = cpu_to_be32(agsize);
 306                 agi->agi_count = 0;
 307                 agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
 308                 agi->agi_level = cpu_to_be32(1);
 309                 agi->agi_freecount = 0;
 310                 agi->agi_newino = cpu_to_be32(NULLAGINO);
 311                 agi->agi_dirino = cpu_to_be32(NULLAGINO);
 312                 if (xfs_sb_version_hascrc(&mp->m_sb))
 313                         uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_uuid);
 314                 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
 315                         agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
 316
 317                 error = xfs_bwrite(bp);
 318                 xfs_buf_relse(bp);
 319                 if (error)
 320                         goto error0;
 321
 322                 /*
 323                  * BNO btree root block
 324                  */
 325                 bp = xfs_growfs_get_hdr_buf(mp,
 326                                 XFS_AGB_TO_DADDR(mp, agno, XFS_BNO_BLOCK(mp)),
 327                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 328                                 &xfs_allocbt_buf_ops);
 329
 330                 if (!bp) {
 331                         error = ENOMEM;
 332                         goto error0;
 333                 }
 334
 335                 if (xfs_sb_version_hascrc(&mp->m_sb))
 336                         xfs_btree_init_block(mp, bp, XFS_ABTB_CRC_MAGIC, 0, 1,
 337                                                 agno, XFS_BTREE_CRC_BLOCKS);
 338                 else
 339                         xfs_btree_init_block(mp, bp, XFS_ABTB_MAGIC, 0, 1,
 340                                                 agno, 0);
 341
 342                 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
 343                 arec->ar_startblock = cpu_to_be32(XFS_PREALLOC_BLOCKS(mp));
 344                 arec->ar_blockcount = cpu_to_be32(
 345                         agsize - be32_to_cpu(arec->ar_startblock));
 346
 347                 error = xfs_bwrite(bp);
 348                 xfs_buf_relse(bp);
 349                 if (error)
 350                         goto error0;
 351
 352                 /*
 353                  * CNT btree root block
 354                  */
 355                 bp = xfs_growfs_get_hdr_buf(mp,
 356                                 XFS_AGB_TO_DADDR(mp, agno, XFS_CNT_BLOCK(mp)),
 357                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 358                                 &xfs_allocbt_buf_ops);
 359                 if (!bp) {
 360                         error = ENOMEM;
 361                         goto error0;
 362                 }
 363
 364                 if (xfs_sb_version_hascrc(&mp->m_sb))
 365                         xfs_btree_init_block(mp, bp, XFS_ABTC_CRC_MAGIC, 0, 1,
 366                                                 agno, XFS_BTREE_CRC_BLOCKS);
 367                 else
 368                         xfs_btree_init_block(mp, bp, XFS_ABTC_MAGIC, 0, 1,
 369                                                 agno, 0);
 370
 371                 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
 372                 arec->ar_startblock = cpu_to_be32(XFS_PREALLOC_BLOCKS(mp));
 373                 arec->ar_blockcount = cpu_to_be32(
 374                         agsize - be32_to_cpu(arec->ar_startblock));
 375                 nfree += be32_to_cpu(arec->ar_blockcount);
 376
 377                 error = xfs_bwrite(bp);
 378                 xfs_buf_relse(bp);
 379                 if (error)
 380                         goto error0;
 381
 382                 /*
 383                  * INO btree root block
 384                  */
 385                 bp = xfs_growfs_get_hdr_buf(mp,
 386                                 XFS_AGB_TO_DADDR(mp, agno, XFS_IBT_BLOCK(mp)),
 387                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 388                                 &xfs_inobt_buf_ops);
 389                 if (!bp) {
 390                         error = ENOMEM;
 391                         goto error0;
 392                 }
 393
 394                 if (xfs_sb_version_hascrc(&mp->m_sb))
 395                         xfs_btree_init_block(mp, bp, XFS_IBT_CRC_MAGIC, 0, 0,
 396                                                 agno, XFS_BTREE_CRC_BLOCKS);
 397                 else
 398                         xfs_btree_init_block(mp, bp, XFS_IBT_MAGIC, 0, 0,
 399                                                 agno, 0);
 400
 401                 error = xfs_bwrite(bp);
 402                 xfs_buf_relse(bp);
 403                 if (error)
 404                         goto error0;
 405         }
 406         xfs_trans_agblocks_delta(tp, nfree);
 407         /*
 408          * There are new blocks in the old last a.g.
 409          */
 410         if (new) {
 411                 /*
 412                  * Change the agi length.
 413                  */
 414                 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
 415                 if (error) {
 416                         goto error0;
 417                 }
 418                 ASSERT(bp);
 419                 agi = XFS_BUF_TO_AGI(bp);
 420                 be32_add_cpu(&agi->agi_length, new);
 421                 ASSERT(nagcount == oagcount ||
 422                        be32_to_cpu(agi->agi_length) == mp->m_sb.sb_agblocks);
 423                 xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
 424                 /*
 425                  * Change agf length.
 426                  */
 427                 error = xfs_alloc_read_agf(mp, tp, agno, 0, &bp);
 428                 if (error) {
 429                         goto error0;
 430                 }
 431                 ASSERT(bp);
 432                 agf = XFS_BUF_TO_AGF(bp);
 433                 be32_add_cpu(&agf->agf_length, new);
 434                 ASSERT(be32_to_cpu(agf->agf_length) ==
 435                        be32_to_cpu(agi->agi_length));
 436
 437                 xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
 438                 /*
 439                  * Free the new space.
 440                  */
 441                 error = xfs_free_extent(tp, XFS_AGB_TO_FSB(mp, agno,
 442                         be32_to_cpu(agf->agf_length) - new), new);
 443                 if (error) {
 444                         goto error0;
 445                 }
 446         }
 447
 448         /*
 449          * Update changed superblock fields transactionally. These are not
 450          * seen by the rest of the world until the transaction commit applies
 451          * them atomically to the superblock.
 452          */
 453         if (nagcount > oagcount)
 454                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
 455         if (nb > mp->m_sb.sb_dblocks)
 456                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS,
 457                                  nb - mp->m_sb.sb_dblocks);
 458         if (nfree)
 459                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, nfree);
 460         if (dpct)
 461                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
 462         error = xfs_trans_commit(tp, 0);
 463         if (error)
 464                 return error;
 465
 466         /* New allocation groups fully initialized, so update mount struct */
 467         if (nagimax)
 468                 mp->m_maxagi = nagimax;
 469         if (mp->m_sb.sb_imax_pct) {
 470                 __uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
 471                 do_div(icount, 100);
 472                 mp->m_maxicount = icount << mp->m_sb.sb_inopblog;
 473         } else
 474                 mp->m_maxicount = 0;
 475         xfs_set_low_space_thresholds(mp);
 476
 477         /* update secondary superblocks. */
 478         for (agno = 1; agno < nagcount; agno++) {
 479                 error = 0;
 480                 /*
 481                  * new secondary superblocks need to be zeroed, not read from
 482                  * disk as the contents of the new area we are growing into is
 483                  * completely unknown.
 484                  */
 485                 if (agno < oagcount) {
 486                         error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
 487                                   XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
 488                                   XFS_FSS_TO_BB(mp, 1), 0, &bp,
 489                                   &xfs_sb_buf_ops);
 490                 } else {
 491                         bp = xfs_trans_get_buf(NULL, mp->m_ddev_targp,
 492                                   XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
 493                                   XFS_FSS_TO_BB(mp, 1), 0);
 494                         if (bp) {
 495                                 bp->b_ops = &xfs_sb_buf_ops;
 496                                 xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
 497                         } else
 498                                 error = ENOMEM;
 499                 }
 500
 501                 if (error) {
 502                         xfs_warn(mp,
 503                 "error %d reading secondary superblock for ag %d",
 504                                 error, agno);
 505                         break;
 506                 }
 507                 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, XFS_SB_ALL_BITS);
 508
 509                 /*
 510                  * If we get an error writing out the alternate superblocks,
 511                  * just issue a warning and continue.  The real work is
 512                  * already done and committed.
 513                  */
 514                 error = xfs_bwrite(bp);
 515                 xfs_buf_relse(bp);
 516                 if (error) {
 517                         xfs_warn(mp,
 518                 "write error %d updating secondary superblock for ag %d",
 519                                 error, agno);
 520                         break; /* no point in continuing */
 521                 }
 522         }
 523         return error;
 524
 525  error0:
 526         xfs_trans_cancel(tp, XFS_TRANS_ABORT);
 527         return error;
 528 }
 529
 530 static int
 531 xfs_growfs_log_private(
 532         xfs_mount_t             *mp,    /* mount point for filesystem */
 533         xfs_growfs_log_t        *in)    /* growfs log input struct */
 534 {
 535         xfs_extlen_t            nb;
 536
 537         nb = in->newblocks;
 538         if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
 539                 return XFS_ERROR(EINVAL);
 540         if (nb == mp->m_sb.sb_logblocks &&
 541             in->isint == (mp->m_sb.sb_logstart != 0))
 542                 return XFS_ERROR(EINVAL);
 543         /*
 544          * Moving the log is hard, need new interfaces to sync
 545          * the log first, hold off all activity while moving it.
 546          * Can have shorter or longer log in the same space,
 547          * or transform internal to external log or vice versa.
 548          */
 549         return XFS_ERROR(ENOSYS);
 550 }
 551
 552 /*
 553  * protected versions of growfs function acquire and release locks on the mount
 554  * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
 555  * XFS_IOC_FSGROWFSRT
 556  */
 557
 558
 559 int
 560 xfs_growfs_data(
 561         xfs_mount_t             *mp,
 562         xfs_growfs_data_t       *in)
 563 {
 564         int error;
 565
 566         if (!capable(CAP_SYS_ADMIN))
 567                 return XFS_ERROR(EPERM);
 568         if (!mutex_trylock(&mp->m_growlock))
 569                 return XFS_ERROR(EWOULDBLOCK);
 570         error = xfs_growfs_data_private(mp, in);
 571         mutex_unlock(&mp->m_growlock);
 572         return error;
 573 }
 574
 575 int
 576 xfs_growfs_log(
 577         xfs_mount_t             *mp,
 578         xfs_growfs_log_t        *in)
 579 {
 580         int error;
 581
 582         if (!capable(CAP_SYS_ADMIN))
 583                 return XFS_ERROR(EPERM);
 584         if (!mutex_trylock(&mp->m_growlock))
 585                 return XFS_ERROR(EWOULDBLOCK);
 586         error = xfs_growfs_log_private(mp, in);
 587         mutex_unlock(&mp->m_growlock);
 588         return error;
 589 }
 590
 591 /*
 592  * exported through ioctl XFS_IOC_FSCOUNTS
 593  */
 594
 595 int
 596 xfs_fs_counts(
 597         xfs_mount_t             *mp,
 598         xfs_fsop_counts_t       *cnt)
 599 {
 600         xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
 601         spin_lock(&mp->m_sb_lock);
 602         cnt->freedata = mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
 603         cnt->freertx = mp->m_sb.sb_frextents;
 604         cnt->freeino = mp->m_sb.sb_ifree;
 605         cnt->allocino = mp->m_sb.sb_icount;
 606         spin_unlock(&mp->m_sb_lock);
 607         return 0;
 608 }
 609
 610 /*
 611  * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
 612  *
 613  * xfs_reserve_blocks is called to set m_resblks
 614  * in the in-core mount table. The number of unused reserved blocks
 615  * is kept in m_resblks_avail.
 616  *
 617  * Reserve the requested number of blocks if available. Otherwise return
 618  * as many as possible to satisfy the request. The actual number
 619  * reserved are returned in outval
 620  *
 621  * A null inval pointer indicates that only the current reserved blocks
 622  * available  should  be returned no settings are changed.
 623  */
 624
 625 int
 626 xfs_reserve_blocks(
 627         xfs_mount_t             *mp,
 628         __uint64_t              *inval,
 629         xfs_fsop_resblks_t      *outval)
 630 {
 631         __int64_t               lcounter, delta, fdblks_delta;
 632         __uint64_t              request;
 633
 634         /* If inval is null, report current values and return */
 635         if (inval == (__uint64_t *)NULL) {
 636                 if (!outval)
 637                         return EINVAL;
 638                 outval->resblks = mp->m_resblks;
 639                 outval->resblks_avail = mp->m_resblks_avail;
 640                 return 0;
 641         }
 642
 643         request = *inval;
 644
 645         /*
 646          * With per-cpu counters, this becomes an interesting
 647          * problem. we needto work out if we are freeing or allocation
 648          * blocks first, then we can do the modification as necessary.
 649          *
 650          * We do this under the m_sb_lock so that if we are near
 651          * ENOSPC, we will hold out any changes while we work out
 652          * what to do. This means that the amount of free space can
 653          * change while we do this, so we need to retry if we end up
 654          * trying to reserve more space than is available.
 655          *
 656          * We also use the xfs_mod_incore_sb() interface so that we
 657          * don't have to care about whether per cpu counter are
 658          * enabled, disabled or even compiled in....
 659          */
 660 retry:
 661         spin_lock(&mp->m_sb_lock);
 662         xfs_icsb_sync_counters_locked(mp, 0);
 663
 664         /*
 665          * If our previous reservation was larger than the current value,
 666          * then move any unused blocks back to the free pool.
 667          */
 668         fdblks_delta = 0;
 669         if (mp->m_resblks > request) {
 670                 lcounter = mp->m_resblks_avail - request;
 671                 if (lcounter  > 0) {            /* release unused blocks */
 672                         fdblks_delta = lcounter;
 673                         mp->m_resblks_avail -= lcounter;
 674                 }
 675                 mp->m_resblks = request;
 676         } else {
 677                 __int64_t       free;
 678
 679                 free =  mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
 680                 if (!free)
 681                         goto out; /* ENOSPC and fdblks_delta = 0 */
 682
 683                 delta = request - mp->m_resblks;
 684                 lcounter = free - delta;
 685                 if (lcounter < 0) {
 686                         /* We can't satisfy the request, just get what we can */
 687                         mp->m_resblks += free;
 688                         mp->m_resblks_avail += free;
 689                         fdblks_delta = -free;
 690                 } else {
 691                         fdblks_delta = -delta;
 692                         mp->m_resblks = request;
 693                         mp->m_resblks_avail += delta;
 694                 }
 695         }
 696 out:
 697         if (outval) {
 698                 outval->resblks = mp->m_resblks;
 699                 outval->resblks_avail = mp->m_resblks_avail;
 700         }
 701         spin_unlock(&mp->m_sb_lock);
 702
 703         if (fdblks_delta) {
 704                 /*
 705                  * If we are putting blocks back here, m_resblks_avail is
 706                  * already at its max so this will put it in the free pool.
 707                  *
 708                  * If we need space, we'll either succeed in getting it
 709                  * from the free block count or we'll get an enospc. If
 710                  * we get a ENOSPC, it means things changed while we were
 711                  * calculating fdblks_delta and so we should try again to
 712                  * see if there is anything left to reserve.
 713                  *
 714                  * Don't set the reserved flag here - we don't want to reserve
 715                  * the extra reserve blocks from the reserve.....
 716                  */
 717                 int error;
 718                 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
 719                                                  fdblks_delta, 0);
 720                 if (error == ENOSPC)
 721                         goto retry;
 722         }
 723         return 0;
 724 }
 725
 726 /*
 727  * Dump a transaction into the log that contains no real change. This is needed
 728  * to be able to make the log dirty or stamp the current tail LSN into the log
 729  * during the covering operation.
 730  *
 731  * We cannot use an inode here for this - that will push dirty state back up
 732  * into the VFS and then periodic inode flushing will prevent log covering from
 733  * making progress. Hence we log a field in the superblock instead and use a
 734  * synchronous transaction to ensure the superblock is immediately unpinned
 735  * and can be written back.
 736  */
 737 int
 738 xfs_fs_log_dummy(
 739         xfs_mount_t     *mp)
 740 {
 741         xfs_trans_t     *tp;
 742         int             error;
 743
 744         tp = _xfs_trans_alloc(mp, XFS_TRANS_DUMMY1, KM_SLEEP);
 745         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
 746         if (error) {
 747                 xfs_trans_cancel(tp, 0);
 748                 return error;
 749         }
 750
 751         /* log the UUID because it is an unchanging field */
 752         xfs_mod_sb(tp, XFS_SB_UUID);
 753         xfs_trans_set_sync(tp);
 754         return xfs_trans_commit(tp, 0);
 755 }
 756
 757 int
 758 xfs_fs_goingdown(
 759         xfs_mount_t     *mp,
 760         __uint32_t      inflags)
 761 {
 762         switch (inflags) {
 763         case XFS_FSOP_GOING_FLAGS_DEFAULT: {
 764                 struct super_block *sb = freeze_bdev(mp->m_super->s_bdev);
 765
 766                 if (sb && !IS_ERR(sb)) {
 767                         xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
 768                         thaw_bdev(sb->s_bdev, sb);
 769                 }
 770
 771                 break;
 772         }
 773         case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
 774                 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
 775                 break;
 776         case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
 777                 xfs_force_shutdown(mp,
 778                                 SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
 779                 break;
 780         default:
 781                 return XFS_ERROR(EINVAL);
 782         }
 783
 784         return 0;
 785 }
 786
 787 /*
 788  * Force a shutdown of the filesystem instantly while keeping the filesystem
 789  * consistent. We don't do an unmount here; just shutdown the shop, make sure
 790  * that absolutely nothing persistent happens to this filesystem after this
 791  * point.
 792  */
 793 void
 794 xfs_do_force_shutdown(
 795         xfs_mount_t     *mp,
 796         int             flags,
 797         char            *fname,
 798         int             lnnum)
 799 {
 800         int             logerror;
 801
 802         logerror = flags & SHUTDOWN_LOG_IO_ERROR;
 803
 804         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 805                 xfs_notice(mp,
 806         "%s(0x%x) called from line %d of file %s.  Return address = 0x%p",
 807                         __func__, flags, lnnum, fname, __return_address);
 808         }
 809         /*
 810          * No need to duplicate efforts.
 811          */
 812         if (XFS_FORCED_SHUTDOWN(mp) && !logerror)
 813                 return;
 814
 815         /*
 816          * This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't
 817          * queue up anybody new on the log reservations, and wakes up
 818          * everybody who's sleeping on log reservations to tell them
 819          * the bad news.
 820          */
 821         if (xfs_log_force_umount(mp, logerror))
 822                 return;
 823
 824         if (flags & SHUTDOWN_CORRUPT_INCORE) {
 825                 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_CORRUPT,
 826     "Corruption of in-memory data detected.  Shutting down filesystem");
 827                 if (XFS_ERRLEVEL_HIGH <= xfs_error_level)
 828                         xfs_stack_trace();
 829         } else if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 830                 if (logerror) {
 831                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_LOGERROR,
 832                 "Log I/O Error Detected.  Shutting down filesystem");
 833                 } else if (flags & SHUTDOWN_DEVICE_REQ) {
 834                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
 835                 "All device paths lost.  Shutting down filesystem");
 836                 } else if (!(flags & SHUTDOWN_REMOTE_REQ)) {
 837                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
 838                 "I/O Error Detected. Shutting down filesystem");
 839                 }
 840         }
 841         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 842                 xfs_alert(mp,
 843         "Please umount the filesystem and rectify the problem(s)");
 844         }
 845 }