fs/xfs/xfs_fsops.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_shared.h"
  21 #include "xfs_format.h"
  22 #include "xfs_log_format.h"
  23 #include "xfs_trans_resv.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_mount.h"
  27 #include "xfs_inode.h"
  28 #include "xfs_trans.h"
  29 #include "xfs_inode_item.h"
  30 #include "xfs_error.h"
  31 #include "xfs_btree.h"
  32 #include "xfs_alloc_btree.h"
  33 #include "xfs_alloc.h"
  34 #include "xfs_ialloc.h"
  35 #include "xfs_fsops.h"
  36 #include "xfs_itable.h"
  37 #include "xfs_trans_space.h"
  38 #include "xfs_rtalloc.h"
  39 #include "xfs_trace.h"
  40 #include "xfs_log.h"
  41 #include "xfs_dinode.h"
  42 #include "xfs_filestream.h"
  43
  44 /*
  45  * File system operations
  46  */
  47
  48 int
  49 xfs_fs_geometry(
  50         xfs_mount_t             *mp,
  51         xfs_fsop_geom_t         *geo,
  52         int                     new_version)
  53 {
  54
  55         memset(geo, 0, sizeof(*geo));
  56
  57         geo->blocksize = mp->m_sb.sb_blocksize;
  58         geo->rtextsize = mp->m_sb.sb_rextsize;
  59         geo->agblocks = mp->m_sb.sb_agblocks;
  60         geo->agcount = mp->m_sb.sb_agcount;
  61         geo->logblocks = mp->m_sb.sb_logblocks;
  62         geo->sectsize = mp->m_sb.sb_sectsize;
  63         geo->inodesize = mp->m_sb.sb_inodesize;
  64         geo->imaxpct = mp->m_sb.sb_imax_pct;
  65         geo->datablocks = mp->m_sb.sb_dblocks;
  66         geo->rtblocks = mp->m_sb.sb_rblocks;
  67         geo->rtextents = mp->m_sb.sb_rextents;
  68         geo->logstart = mp->m_sb.sb_logstart;
  69         ASSERT(sizeof(geo->uuid)==sizeof(mp->m_sb.sb_uuid));
  70         memcpy(geo->uuid, &mp->m_sb.sb_uuid, sizeof(mp->m_sb.sb_uuid));
  71         if (new_version >= 2) {
  72                 geo->sunit = mp->m_sb.sb_unit;
  73                 geo->swidth = mp->m_sb.sb_width;
  74         }
  75         if (new_version >= 3) {
  76                 geo->version = XFS_FSOP_GEOM_VERSION;
  77                 geo->flags =
  78                         (xfs_sb_version_hasattr(&mp->m_sb) ?
  79                                 XFS_FSOP_GEOM_FLAGS_ATTR : 0) |
  80                         (xfs_sb_version_hasnlink(&mp->m_sb) ?
  81                                 XFS_FSOP_GEOM_FLAGS_NLINK : 0) |
  82                         (xfs_sb_version_hasquota(&mp->m_sb) ?
  83                                 XFS_FSOP_GEOM_FLAGS_QUOTA : 0) |
  84                         (xfs_sb_version_hasalign(&mp->m_sb) ?
  85                                 XFS_FSOP_GEOM_FLAGS_IALIGN : 0) |
  86                         (xfs_sb_version_hasdalign(&mp->m_sb) ?
  87                                 XFS_FSOP_GEOM_FLAGS_DALIGN : 0) |
  88                         (xfs_sb_version_hasshared(&mp->m_sb) ?
  89                                 XFS_FSOP_GEOM_FLAGS_SHARED : 0) |
  90                         (xfs_sb_version_hasextflgbit(&mp->m_sb) ?
  91                                 XFS_FSOP_GEOM_FLAGS_EXTFLG : 0) |
  92                         (xfs_sb_version_hasdirv2(&mp->m_sb) ?
  93                                 XFS_FSOP_GEOM_FLAGS_DIRV2 : 0) |
  94                         (xfs_sb_version_hassector(&mp->m_sb) ?
  95                                 XFS_FSOP_GEOM_FLAGS_SECTOR : 0) |
  96                         (xfs_sb_version_hasasciici(&mp->m_sb) ?
  97                                 XFS_FSOP_GEOM_FLAGS_DIRV2CI : 0) |
  98                         (xfs_sb_version_haslazysbcount(&mp->m_sb) ?
  99                                 XFS_FSOP_GEOM_FLAGS_LAZYSB : 0) |
 100                         (xfs_sb_version_hasattr2(&mp->m_sb) ?
 101                                 XFS_FSOP_GEOM_FLAGS_ATTR2 : 0) |
 102                         (xfs_sb_version_hasprojid32bit(&mp->m_sb) ?
 103                                 XFS_FSOP_GEOM_FLAGS_PROJID32 : 0) |
 104                         (xfs_sb_version_hascrc(&mp->m_sb) ?
 105                                 XFS_FSOP_GEOM_FLAGS_V5SB : 0) |
 106                         (xfs_sb_version_hasftype(&mp->m_sb) ?
 107                                 XFS_FSOP_GEOM_FLAGS_FTYPE : 0) |
 108                         (xfs_sb_version_hasfinobt(&mp->m_sb) ?
 109                                 XFS_FSOP_GEOM_FLAGS_FINOBT : 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                 __be32  *agfl_bno;
 226
 227                 /*
 228                  * AG freespace header block
 229                  */
 230                 bp = xfs_growfs_get_hdr_buf(mp,
 231                                 XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
 232                                 XFS_FSS_TO_BB(mp, 1), 0,
 233                                 &xfs_agf_buf_ops);
 234                 if (!bp) {
 235                         error = ENOMEM;
 236                         goto error0;
 237                 }
 238
 239                 agf = XFS_BUF_TO_AGF(bp);
 240                 agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
 241                 agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
 242                 agf->agf_seqno = cpu_to_be32(agno);
 243                 if (agno == nagcount - 1)
 244                         agsize =
 245                                 nb -
 246                                 (agno * (xfs_rfsblock_t)mp->m_sb.sb_agblocks);
 247                 else
 248                         agsize = mp->m_sb.sb_agblocks;
 249                 agf->agf_length = cpu_to_be32(agsize);
 250                 agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
 251                 agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
 252                 agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
 253                 agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
 254                 agf->agf_flfirst = 0;
 255                 agf->agf_fllast = cpu_to_be32(XFS_AGFL_SIZE(mp) - 1);
 256                 agf->agf_flcount = 0;
 257                 tmpsize = agsize - XFS_PREALLOC_BLOCKS(mp);
 258                 agf->agf_freeblks = cpu_to_be32(tmpsize);
 259                 agf->agf_longest = cpu_to_be32(tmpsize);
 260                 if (xfs_sb_version_hascrc(&mp->m_sb))
 261                         uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_uuid);
 262
 263                 error = xfs_bwrite(bp);
 264                 xfs_buf_relse(bp);
 265                 if (error)
 266                         goto error0;
 267
 268                 /*
 269                  * AG freelist header block
 270                  */
 271                 bp = xfs_growfs_get_hdr_buf(mp,
 272                                 XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)),
 273                                 XFS_FSS_TO_BB(mp, 1), 0,
 274                                 &xfs_agfl_buf_ops);
 275                 if (!bp) {
 276                         error = ENOMEM;
 277                         goto error0;
 278                 }
 279
 280                 agfl = XFS_BUF_TO_AGFL(bp);
 281                 if (xfs_sb_version_hascrc(&mp->m_sb)) {
 282                         agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
 283                         agfl->agfl_seqno = cpu_to_be32(agno);
 284                         uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_uuid);
 285                 }
 286
 287                 agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, bp);
 288                 for (bucket = 0; bucket < XFS_AGFL_SIZE(mp); bucket++)
 289                         agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
 290
 291                 error = xfs_bwrite(bp);
 292                 xfs_buf_relse(bp);
 293                 if (error)
 294                         goto error0;
 295
 296                 /*
 297                  * AG inode header block
 298                  */
 299                 bp = xfs_growfs_get_hdr_buf(mp,
 300                                 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
 301                                 XFS_FSS_TO_BB(mp, 1), 0,
 302                                 &xfs_agi_buf_ops);
 303                 if (!bp) {
 304                         error = ENOMEM;
 305                         goto error0;
 306                 }
 307
 308                 agi = XFS_BUF_TO_AGI(bp);
 309                 agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
 310                 agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
 311                 agi->agi_seqno = cpu_to_be32(agno);
 312                 agi->agi_length = cpu_to_be32(agsize);
 313                 agi->agi_count = 0;
 314                 agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
 315                 agi->agi_level = cpu_to_be32(1);
 316                 agi->agi_freecount = 0;
 317                 agi->agi_newino = cpu_to_be32(NULLAGINO);
 318                 agi->agi_dirino = cpu_to_be32(NULLAGINO);
 319                 if (xfs_sb_version_hascrc(&mp->m_sb))
 320                         uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_uuid);
 321                 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
 322                         agi->agi_free_root = cpu_to_be32(XFS_FIBT_BLOCK(mp));
 323                         agi->agi_free_level = cpu_to_be32(1);
 324                 }
 325                 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
 326                         agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
 327
 328                 error = xfs_bwrite(bp);
 329                 xfs_buf_relse(bp);
 330                 if (error)
 331                         goto error0;
 332
 333                 /*
 334                  * BNO btree root block
 335                  */
 336                 bp = xfs_growfs_get_hdr_buf(mp,
 337                                 XFS_AGB_TO_DADDR(mp, agno, XFS_BNO_BLOCK(mp)),
 338                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 339                                 &xfs_allocbt_buf_ops);
 340
 341                 if (!bp) {
 342                         error = ENOMEM;
 343                         goto error0;
 344                 }
 345
 346                 if (xfs_sb_version_hascrc(&mp->m_sb))
 347                         xfs_btree_init_block(mp, bp, XFS_ABTB_CRC_MAGIC, 0, 1,
 348                                                 agno, XFS_BTREE_CRC_BLOCKS);
 349                 else
 350                         xfs_btree_init_block(mp, bp, XFS_ABTB_MAGIC, 0, 1,
 351                                                 agno, 0);
 352
 353                 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
 354                 arec->ar_startblock = cpu_to_be32(XFS_PREALLOC_BLOCKS(mp));
 355                 arec->ar_blockcount = cpu_to_be32(
 356                         agsize - be32_to_cpu(arec->ar_startblock));
 357
 358                 error = xfs_bwrite(bp);
 359                 xfs_buf_relse(bp);
 360                 if (error)
 361                         goto error0;
 362
 363                 /*
 364                  * CNT btree root block
 365                  */
 366                 bp = xfs_growfs_get_hdr_buf(mp,
 367                                 XFS_AGB_TO_DADDR(mp, agno, XFS_CNT_BLOCK(mp)),
 368                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 369                                 &xfs_allocbt_buf_ops);
 370                 if (!bp) {
 371                         error = ENOMEM;
 372                         goto error0;
 373                 }
 374
 375                 if (xfs_sb_version_hascrc(&mp->m_sb))
 376                         xfs_btree_init_block(mp, bp, XFS_ABTC_CRC_MAGIC, 0, 1,
 377                                                 agno, XFS_BTREE_CRC_BLOCKS);
 378                 else
 379                         xfs_btree_init_block(mp, bp, XFS_ABTC_MAGIC, 0, 1,
 380                                                 agno, 0);
 381
 382                 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
 383                 arec->ar_startblock = cpu_to_be32(XFS_PREALLOC_BLOCKS(mp));
 384                 arec->ar_blockcount = cpu_to_be32(
 385                         agsize - be32_to_cpu(arec->ar_startblock));
 386                 nfree += be32_to_cpu(arec->ar_blockcount);
 387
 388                 error = xfs_bwrite(bp);
 389                 xfs_buf_relse(bp);
 390                 if (error)
 391                         goto error0;
 392
 393                 /*
 394                  * INO btree root block
 395                  */
 396                 bp = xfs_growfs_get_hdr_buf(mp,
 397                                 XFS_AGB_TO_DADDR(mp, agno, XFS_IBT_BLOCK(mp)),
 398                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 399                                 &xfs_inobt_buf_ops);
 400                 if (!bp) {
 401                         error = ENOMEM;
 402                         goto error0;
 403                 }
 404
 405                 if (xfs_sb_version_hascrc(&mp->m_sb))
 406                         xfs_btree_init_block(mp, bp, XFS_IBT_CRC_MAGIC, 0, 0,
 407                                                 agno, XFS_BTREE_CRC_BLOCKS);
 408                 else
 409                         xfs_btree_init_block(mp, bp, XFS_IBT_MAGIC, 0, 0,
 410                                                 agno, 0);
 411
 412                 error = xfs_bwrite(bp);
 413                 xfs_buf_relse(bp);
 414                 if (error)
 415                         goto error0;
 416
 417                 /*
 418                  * FINO btree root block
 419                  */
 420                 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
 421                         bp = xfs_growfs_get_hdr_buf(mp,
 422                                 XFS_AGB_TO_DADDR(mp, agno, XFS_FIBT_BLOCK(mp)),
 423                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 424                                 &xfs_inobt_buf_ops);
 425                         if (!bp) {
 426                                 error = ENOMEM;
 427                                 goto error0;
 428                         }
 429
 430                         if (xfs_sb_version_hascrc(&mp->m_sb))
 431                                 xfs_btree_init_block(mp, bp, XFS_FIBT_CRC_MAGIC,
 432                                                      0, 0, agno,
 433                                                      XFS_BTREE_CRC_BLOCKS);
 434                         else
 435                                 xfs_btree_init_block(mp, bp, XFS_FIBT_MAGIC, 0,
 436                                                      0, agno, 0);
 437
 438                         error = xfs_bwrite(bp);
 439                         xfs_buf_relse(bp);
 440                         if (error)
 441                                 goto error0;
 442                 }
 443
 444         }
 445         xfs_trans_agblocks_delta(tp, nfree);
 446         /*
 447          * There are new blocks in the old last a.g.
 448          */
 449         if (new) {
 450                 /*
 451                  * Change the agi length.
 452                  */
 453                 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
 454                 if (error) {
 455                         goto error0;
 456                 }
 457                 ASSERT(bp);
 458                 agi = XFS_BUF_TO_AGI(bp);
 459                 be32_add_cpu(&agi->agi_length, new);
 460                 ASSERT(nagcount == oagcount ||
 461                        be32_to_cpu(agi->agi_length) == mp->m_sb.sb_agblocks);
 462                 xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
 463                 /*
 464                  * Change agf length.
 465                  */
 466                 error = xfs_alloc_read_agf(mp, tp, agno, 0, &bp);
 467                 if (error) {
 468                         goto error0;
 469                 }
 470                 ASSERT(bp);
 471                 agf = XFS_BUF_TO_AGF(bp);
 472                 be32_add_cpu(&agf->agf_length, new);
 473                 ASSERT(be32_to_cpu(agf->agf_length) ==
 474                        be32_to_cpu(agi->agi_length));
 475
 476                 xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
 477                 /*
 478                  * Free the new space.
 479                  */
 480                 error = xfs_free_extent(tp, XFS_AGB_TO_FSB(mp, agno,
 481                         be32_to_cpu(agf->agf_length) - new), new);
 482                 if (error) {
 483                         goto error0;
 484                 }
 485         }
 486
 487         /*
 488          * Update changed superblock fields transactionally. These are not
 489          * seen by the rest of the world until the transaction commit applies
 490          * them atomically to the superblock.
 491          */
 492         if (nagcount > oagcount)
 493                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
 494         if (nb > mp->m_sb.sb_dblocks)
 495                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS,
 496                                  nb - mp->m_sb.sb_dblocks);
 497         if (nfree)
 498                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, nfree);
 499         if (dpct)
 500                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
 501         error = xfs_trans_commit(tp, 0);
 502         if (error)
 503                 return error;
 504
 505         /* New allocation groups fully initialized, so update mount struct */
 506         if (nagimax)
 507                 mp->m_maxagi = nagimax;
 508         if (mp->m_sb.sb_imax_pct) {
 509                 __uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
 510                 do_div(icount, 100);
 511                 mp->m_maxicount = icount << mp->m_sb.sb_inopblog;
 512         } else
 513                 mp->m_maxicount = 0;
 514         xfs_set_low_space_thresholds(mp);
 515
 516         /* update secondary superblocks. */
 517         for (agno = 1; agno < nagcount; agno++) {
 518                 error = 0;
 519                 /*
 520                  * new secondary superblocks need to be zeroed, not read from
 521                  * disk as the contents of the new area we are growing into is
 522                  * completely unknown.
 523                  */
 524                 if (agno < oagcount) {
 525                         error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
 526                                   XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
 527                                   XFS_FSS_TO_BB(mp, 1), 0, &bp,
 528                                   &xfs_sb_buf_ops);
 529                 } else {
 530                         bp = xfs_trans_get_buf(NULL, mp->m_ddev_targp,
 531                                   XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
 532                                   XFS_FSS_TO_BB(mp, 1), 0);
 533                         if (bp) {
 534                                 bp->b_ops = &xfs_sb_buf_ops;
 535                                 xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
 536                         } else
 537                                 error = ENOMEM;
 538                 }
 539
 540                 /*
 541                  * If we get an error reading or writing alternate superblocks,
 542                  * continue.  xfs_repair chooses the "best" superblock based
 543                  * on most matches; if we break early, we'll leave more
 544                  * superblocks un-updated than updated, and xfs_repair may
 545                  * pick them over the properly-updated primary.
 546                  */
 547                 if (error) {
 548                         xfs_warn(mp,
 549                 "error %d reading secondary superblock for ag %d",
 550                                 error, agno);
 551                         saved_error = error;
 552                         continue;
 553                 }
 554                 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, XFS_SB_ALL_BITS);
 555
 556                 error = xfs_bwrite(bp);
 557                 xfs_buf_relse(bp);
 558                 if (error) {
 559                         xfs_warn(mp,
 560                 "write error %d updating secondary superblock for ag %d",
 561                                 error, agno);
 562                         saved_error = error;
 563                         continue;
 564                 }
 565         }
 566         return saved_error ? saved_error : error;
 567
 568  error0:
 569         xfs_trans_cancel(tp, XFS_TRANS_ABORT);
 570         return error;
 571 }
 572
 573 static int
 574 xfs_growfs_log_private(
 575         xfs_mount_t             *mp,    /* mount point for filesystem */
 576         xfs_growfs_log_t        *in)    /* growfs log input struct */
 577 {
 578         xfs_extlen_t            nb;
 579
 580         nb = in->newblocks;
 581         if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
 582                 return XFS_ERROR(EINVAL);
 583         if (nb == mp->m_sb.sb_logblocks &&
 584             in->isint == (mp->m_sb.sb_logstart != 0))
 585                 return XFS_ERROR(EINVAL);
 586         /*
 587          * Moving the log is hard, need new interfaces to sync
 588          * the log first, hold off all activity while moving it.
 589          * Can have shorter or longer log in the same space,
 590          * or transform internal to external log or vice versa.
 591          */
 592         return XFS_ERROR(ENOSYS);
 593 }
 594
 595 /*
 596  * protected versions of growfs function acquire and release locks on the mount
 597  * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
 598  * XFS_IOC_FSGROWFSRT
 599  */
 600
 601
 602 int
 603 xfs_growfs_data(
 604         xfs_mount_t             *mp,
 605         xfs_growfs_data_t       *in)
 606 {
 607         int error;
 608
 609         if (!capable(CAP_SYS_ADMIN))
 610                 return XFS_ERROR(EPERM);
 611         if (!mutex_trylock(&mp->m_growlock))
 612                 return XFS_ERROR(EWOULDBLOCK);
 613         error = xfs_growfs_data_private(mp, in);
 614         mutex_unlock(&mp->m_growlock);
 615         return error;
 616 }
 617
 618 int
 619 xfs_growfs_log(
 620         xfs_mount_t             *mp,
 621         xfs_growfs_log_t        *in)
 622 {
 623         int error;
 624
 625         if (!capable(CAP_SYS_ADMIN))
 626                 return XFS_ERROR(EPERM);
 627         if (!mutex_trylock(&mp->m_growlock))
 628                 return XFS_ERROR(EWOULDBLOCK);
 629         error = xfs_growfs_log_private(mp, in);
 630         mutex_unlock(&mp->m_growlock);
 631         return error;
 632 }
 633
 634 /*
 635  * exported through ioctl XFS_IOC_FSCOUNTS
 636  */
 637
 638 int
 639 xfs_fs_counts(
 640         xfs_mount_t             *mp,
 641         xfs_fsop_counts_t       *cnt)
 642 {
 643         xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
 644         spin_lock(&mp->m_sb_lock);
 645         cnt->freedata = mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
 646         cnt->freertx = mp->m_sb.sb_frextents;
 647         cnt->freeino = mp->m_sb.sb_ifree;
 648         cnt->allocino = mp->m_sb.sb_icount;
 649         spin_unlock(&mp->m_sb_lock);
 650         return 0;
 651 }
 652
 653 /*
 654  * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
 655  *
 656  * xfs_reserve_blocks is called to set m_resblks
 657  * in the in-core mount table. The number of unused reserved blocks
 658  * is kept in m_resblks_avail.
 659  *
 660  * Reserve the requested number of blocks if available. Otherwise return
 661  * as many as possible to satisfy the request. The actual number
 662  * reserved are returned in outval
 663  *
 664  * A null inval pointer indicates that only the current reserved blocks
 665  * available  should  be returned no settings are changed.
 666  */
 667
 668 int
 669 xfs_reserve_blocks(
 670         xfs_mount_t             *mp,
 671         __uint64_t              *inval,
 672         xfs_fsop_resblks_t      *outval)
 673 {
 674         __int64_t               lcounter, delta, fdblks_delta;
 675         __uint64_t              request;
 676
 677         /* If inval is null, report current values and return */
 678         if (inval == (__uint64_t *)NULL) {
 679                 if (!outval)
 680                         return EINVAL;
 681                 outval->resblks = mp->m_resblks;
 682                 outval->resblks_avail = mp->m_resblks_avail;
 683                 return 0;
 684         }
 685
 686         request = *inval;
 687
 688         /*
 689          * With per-cpu counters, this becomes an interesting
 690          * problem. we needto work out if we are freeing or allocation
 691          * blocks first, then we can do the modification as necessary.
 692          *
 693          * We do this under the m_sb_lock so that if we are near
 694          * ENOSPC, we will hold out any changes while we work out
 695          * what to do. This means that the amount of free space can
 696          * change while we do this, so we need to retry if we end up
 697          * trying to reserve more space than is available.
 698          *
 699          * We also use the xfs_mod_incore_sb() interface so that we
 700          * don't have to care about whether per cpu counter are
 701          * enabled, disabled or even compiled in....
 702          */
 703 retry:
 704         spin_lock(&mp->m_sb_lock);
 705         xfs_icsb_sync_counters_locked(mp, 0);
 706
 707         /*
 708          * If our previous reservation was larger than the current value,
 709          * then move any unused blocks back to the free pool.
 710          */
 711         fdblks_delta = 0;
 712         if (mp->m_resblks > request) {
 713                 lcounter = mp->m_resblks_avail - request;
 714                 if (lcounter  > 0) {            /* release unused blocks */
 715                         fdblks_delta = lcounter;
 716                         mp->m_resblks_avail -= lcounter;
 717                 }
 718                 mp->m_resblks = request;
 719         } else {
 720                 __int64_t       free;
 721
 722                 free =  mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
 723                 if (!free)
 724                         goto out; /* ENOSPC and fdblks_delta = 0 */
 725
 726                 delta = request - mp->m_resblks;
 727                 lcounter = free - delta;
 728                 if (lcounter < 0) {
 729                         /* We can't satisfy the request, just get what we can */
 730                         mp->m_resblks += free;
 731                         mp->m_resblks_avail += free;
 732                         fdblks_delta = -free;
 733                 } else {
 734                         fdblks_delta = -delta;
 735                         mp->m_resblks = request;
 736                         mp->m_resblks_avail += delta;
 737                 }
 738         }
 739 out:
 740         if (outval) {
 741                 outval->resblks = mp->m_resblks;
 742                 outval->resblks_avail = mp->m_resblks_avail;
 743         }
 744         spin_unlock(&mp->m_sb_lock);
 745
 746         if (fdblks_delta) {
 747                 /*
 748                  * If we are putting blocks back here, m_resblks_avail is
 749                  * already at its max so this will put it in the free pool.
 750                  *
 751                  * If we need space, we'll either succeed in getting it
 752                  * from the free block count or we'll get an enospc. If
 753                  * we get a ENOSPC, it means things changed while we were
 754                  * calculating fdblks_delta and so we should try again to
 755                  * see if there is anything left to reserve.
 756                  *
 757                  * Don't set the reserved flag here - we don't want to reserve
 758                  * the extra reserve blocks from the reserve.....
 759                  */
 760                 int error;
 761                 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
 762                                                  fdblks_delta, 0);
 763                 if (error == ENOSPC)
 764                         goto retry;
 765         }
 766         return 0;
 767 }
 768
 769 /*
 770  * Dump a transaction into the log that contains no real change. This is needed
 771  * to be able to make the log dirty or stamp the current tail LSN into the log
 772  * during the covering operation.
 773  *
 774  * We cannot use an inode here for this - that will push dirty state back up
 775  * into the VFS and then periodic inode flushing will prevent log covering from
 776  * making progress. Hence we log a field in the superblock instead and use a
 777  * synchronous transaction to ensure the superblock is immediately unpinned
 778  * and can be written back.
 779  */
 780 int
 781 xfs_fs_log_dummy(
 782         xfs_mount_t     *mp)
 783 {
 784         xfs_trans_t     *tp;
 785         int             error;
 786
 787         tp = _xfs_trans_alloc(mp, XFS_TRANS_DUMMY1, KM_SLEEP);
 788         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
 789         if (error) {
 790                 xfs_trans_cancel(tp, 0);
 791                 return error;
 792         }
 793
 794         /* log the UUID because it is an unchanging field */
 795         xfs_mod_sb(tp, XFS_SB_UUID);
 796         xfs_trans_set_sync(tp);
 797         return xfs_trans_commit(tp, 0);
 798 }
 799
 800 int
 801 xfs_fs_goingdown(
 802         xfs_mount_t     *mp,
 803         __uint32_t      inflags)
 804 {
 805         switch (inflags) {
 806         case XFS_FSOP_GOING_FLAGS_DEFAULT: {
 807                 struct super_block *sb = freeze_bdev(mp->m_super->s_bdev);
 808
 809                 if (sb && !IS_ERR(sb)) {
 810                         xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
 811                         thaw_bdev(sb->s_bdev, sb);
 812                 }
 813
 814                 break;
 815         }
 816         case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
 817                 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
 818                 break;
 819         case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
 820                 xfs_force_shutdown(mp,
 821                                 SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
 822                 break;
 823         default:
 824                 return XFS_ERROR(EINVAL);
 825         }
 826
 827         return 0;
 828 }
 829
 830 /*
 831  * Force a shutdown of the filesystem instantly while keeping the filesystem
 832  * consistent. We don't do an unmount here; just shutdown the shop, make sure
 833  * that absolutely nothing persistent happens to this filesystem after this
 834  * point.
 835  */
 836 void
 837 xfs_do_force_shutdown(
 838         xfs_mount_t     *mp,
 839         int             flags,
 840         char            *fname,
 841         int             lnnum)
 842 {
 843         int             logerror;
 844
 845         logerror = flags & SHUTDOWN_LOG_IO_ERROR;
 846
 847         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 848                 xfs_notice(mp,
 849         "%s(0x%x) called from line %d of file %s.  Return address = 0x%p",
 850                         __func__, flags, lnnum, fname, __return_address);
 851         }
 852         /*
 853          * No need to duplicate efforts.
 854          */
 855         if (XFS_FORCED_SHUTDOWN(mp) && !logerror)
 856                 return;
 857
 858         /*
 859          * This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't
 860          * queue up anybody new on the log reservations, and wakes up
 861          * everybody who's sleeping on log reservations to tell them
 862          * the bad news.
 863          */
 864         if (xfs_log_force_umount(mp, logerror))
 865                 return;
 866
 867         if (flags & SHUTDOWN_CORRUPT_INCORE) {
 868                 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_CORRUPT,
 869     "Corruption of in-memory data detected.  Shutting down filesystem");
 870                 if (XFS_ERRLEVEL_HIGH <= xfs_error_level)
 871                         xfs_stack_trace();
 872         } else if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 873                 if (logerror) {
 874                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_LOGERROR,
 875                 "Log I/O Error Detected.  Shutting down filesystem");
 876                 } else if (flags & SHUTDOWN_DEVICE_REQ) {
 877                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
 878                 "All device paths lost.  Shutting down filesystem");
 879                 } else if (!(flags & SHUTDOWN_REMOTE_REQ)) {
 880                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
 881                 "I/O Error Detected. Shutting down filesystem");
 882                 }
 883         }
 884         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 885                 xfs_alert(mp,
 886         "Please umount the filesystem and rectify the problem(s)");
 887         }
 888 }