fs/xfs/xfs_fsops.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_shared.h"
  21 #include "xfs_format.h"
  22 #include "xfs_log_format.h"
  23 #include "xfs_trans_resv.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_mount.h"
  27 #include "xfs_inode.h"
  28 #include "xfs_trans.h"
  29 #include "xfs_inode_item.h"
  30 #include "xfs_error.h"
  31 #include "xfs_btree.h"
  32 #include "xfs_alloc_btree.h"
  33 #include "xfs_alloc.h"
  34 #include "xfs_ialloc.h"
  35 #include "xfs_fsops.h"
  36 #include "xfs_itable.h"
  37 #include "xfs_trans_space.h"
  38 #include "xfs_rtalloc.h"
  39 #include "xfs_trace.h"
  40 #include "xfs_log.h"
  41 #include "xfs_dinode.h"
  42 #include "xfs_filestream.h"
  43
  44 /*
  45  * File system operations
  46  */
  47
  48 int
  49 xfs_fs_geometry(
  50         xfs_mount_t             *mp,
  51         xfs_fsop_geom_t         *geo,
  52         int                     new_version)
  53 {
  54
  55         memset(geo, 0, sizeof(*geo));
  56
  57         geo->blocksize = mp->m_sb.sb_blocksize;
  58         geo->rtextsize = mp->m_sb.sb_rextsize;
  59         geo->agblocks = mp->m_sb.sb_agblocks;
  60         geo->agcount = mp->m_sb.sb_agcount;
  61         geo->logblocks = mp->m_sb.sb_logblocks;
  62         geo->sectsize = mp->m_sb.sb_sectsize;
  63         geo->inodesize = mp->m_sb.sb_inodesize;
  64         geo->imaxpct = mp->m_sb.sb_imax_pct;
  65         geo->datablocks = mp->m_sb.sb_dblocks;
  66         geo->rtblocks = mp->m_sb.sb_rblocks;
  67         geo->rtextents = mp->m_sb.sb_rextents;
  68         geo->logstart = mp->m_sb.sb_logstart;
  69         ASSERT(sizeof(geo->uuid)==sizeof(mp->m_sb.sb_uuid));
  70         memcpy(geo->uuid, &mp->m_sb.sb_uuid, sizeof(mp->m_sb.sb_uuid));
  71         if (new_version >= 2) {
  72                 geo->sunit = mp->m_sb.sb_unit;
  73                 geo->swidth = mp->m_sb.sb_width;
  74         }
  75         if (new_version >= 3) {
  76                 geo->version = XFS_FSOP_GEOM_VERSION;
  77                 geo->flags =
  78                         (xfs_sb_version_hasattr(&mp->m_sb) ?
  79                                 XFS_FSOP_GEOM_FLAGS_ATTR : 0) |
  80                         (xfs_sb_version_hasnlink(&mp->m_sb) ?
  81                                 XFS_FSOP_GEOM_FLAGS_NLINK : 0) |
  82                         (xfs_sb_version_hasquota(&mp->m_sb) ?
  83                                 XFS_FSOP_GEOM_FLAGS_QUOTA : 0) |
  84                         (xfs_sb_version_hasalign(&mp->m_sb) ?
  85                                 XFS_FSOP_GEOM_FLAGS_IALIGN : 0) |
  86                         (xfs_sb_version_hasdalign(&mp->m_sb) ?
  87                                 XFS_FSOP_GEOM_FLAGS_DALIGN : 0) |
  88                         (xfs_sb_version_hasshared(&mp->m_sb) ?
  89                                 XFS_FSOP_GEOM_FLAGS_SHARED : 0) |
  90                         (xfs_sb_version_hasextflgbit(&mp->m_sb) ?
  91                                 XFS_FSOP_GEOM_FLAGS_EXTFLG : 0) |
  92                         (xfs_sb_version_hasdirv2(&mp->m_sb) ?
  93                                 XFS_FSOP_GEOM_FLAGS_DIRV2 : 0) |
  94                         (xfs_sb_version_hassector(&mp->m_sb) ?
  95                                 XFS_FSOP_GEOM_FLAGS_SECTOR : 0) |
  96                         (xfs_sb_version_hasasciici(&mp->m_sb) ?
  97                                 XFS_FSOP_GEOM_FLAGS_DIRV2CI : 0) |
  98                         (xfs_sb_version_haslazysbcount(&mp->m_sb) ?
  99                                 XFS_FSOP_GEOM_FLAGS_LAZYSB : 0) |
 100                         (xfs_sb_version_hasattr2(&mp->m_sb) ?
 101                                 XFS_FSOP_GEOM_FLAGS_ATTR2 : 0) |
 102                         (xfs_sb_version_hasprojid32bit(&mp->m_sb) ?
 103                                 XFS_FSOP_GEOM_FLAGS_PROJID32 : 0) |
 104                         (xfs_sb_version_hascrc(&mp->m_sb) ?
 105                                 XFS_FSOP_GEOM_FLAGS_V5SB : 0) |
 106                         (xfs_sb_version_hasftype(&mp->m_sb) ?
 107                                 XFS_FSOP_GEOM_FLAGS_FTYPE : 0);
 108                 geo->logsectsize = xfs_sb_version_hassector(&mp->m_sb) ?
 109                                 mp->m_sb.sb_logsectsize : BBSIZE;
 110                 geo->rtsectsize = mp->m_sb.sb_blocksize;
 111                 geo->dirblocksize = mp->m_dirblksize;
 112         }
 113         if (new_version >= 4) {
 114                 geo->flags |=
 115                         (xfs_sb_version_haslogv2(&mp->m_sb) ?
 116                                 XFS_FSOP_GEOM_FLAGS_LOGV2 : 0);
 117                 geo->logsunit = mp->m_sb.sb_logsunit;
 118         }
 119         return 0;
 120 }
 121
 122 static struct xfs_buf *
 123 xfs_growfs_get_hdr_buf(
 124         struct xfs_mount        *mp,
 125         xfs_daddr_t             blkno,
 126         size_t                  numblks,
 127         int                     flags,
 128         const struct xfs_buf_ops *ops)
 129 {
 130         struct xfs_buf          *bp;
 131
 132         bp = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, flags);
 133         if (!bp)
 134                 return NULL;
 135
 136         xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
 137         bp->b_bn = blkno;
 138         bp->b_maps[0].bm_bn = blkno;
 139         bp->b_ops = ops;
 140
 141         return bp;
 142 }
 143
 144 static int
 145 xfs_growfs_data_private(
 146         xfs_mount_t             *mp,            /* mount point for filesystem */
 147         xfs_growfs_data_t       *in)            /* growfs data input struct */
 148 {
 149         xfs_agf_t               *agf;
 150         struct xfs_agfl         *agfl;
 151         xfs_agi_t               *agi;
 152         xfs_agnumber_t          agno;
 153         xfs_extlen_t            agsize;
 154         xfs_extlen_t            tmpsize;
 155         xfs_alloc_rec_t         *arec;
 156         xfs_buf_t               *bp;
 157         int                     bucket;
 158         int                     dpct;
 159         int                     error, saved_error = 0;
 160         xfs_agnumber_t          nagcount;
 161         xfs_agnumber_t          nagimax = 0;
 162         xfs_rfsblock_t          nb, nb_mod;
 163         xfs_rfsblock_t          new;
 164         xfs_rfsblock_t          nfree;
 165         xfs_agnumber_t          oagcount;
 166         int                     pct;
 167         xfs_trans_t             *tp;
 168
 169         nb = in->newblocks;
 170         pct = in->imaxpct;
 171         if (nb < mp->m_sb.sb_dblocks || pct < 0 || pct > 100)
 172                 return XFS_ERROR(EINVAL);
 173         if ((error = xfs_sb_validate_fsb_count(&mp->m_sb, nb)))
 174                 return error;
 175         dpct = pct - mp->m_sb.sb_imax_pct;
 176         bp = xfs_buf_read_uncached(mp->m_ddev_targp,
 177                                 XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1),
 178                                 XFS_FSS_TO_BB(mp, 1), 0, NULL);
 179         if (!bp)
 180                 return EIO;
 181         if (bp->b_error) {
 182                 error = bp->b_error;
 183                 xfs_buf_relse(bp);
 184                 return error;
 185         }
 186         xfs_buf_relse(bp);
 187
 188         new = nb;       /* use new as a temporary here */
 189         nb_mod = do_div(new, mp->m_sb.sb_agblocks);
 190         nagcount = new + (nb_mod != 0);
 191         if (nb_mod && nb_mod < XFS_MIN_AG_BLOCKS) {
 192                 nagcount--;
 193                 nb = (xfs_rfsblock_t)nagcount * mp->m_sb.sb_agblocks;
 194                 if (nb < mp->m_sb.sb_dblocks)
 195                         return XFS_ERROR(EINVAL);
 196         }
 197         new = nb - mp->m_sb.sb_dblocks;
 198         oagcount = mp->m_sb.sb_agcount;
 199
 200         /* allocate the new per-ag structures */
 201         if (nagcount > oagcount) {
 202                 error = xfs_initialize_perag(mp, nagcount, &nagimax);
 203                 if (error)
 204                         return error;
 205         }
 206
 207         tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFS);
 208         tp->t_flags |= XFS_TRANS_RESERVE;
 209         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_growdata,
 210                                   XFS_GROWFS_SPACE_RES(mp), 0);
 211         if (error) {
 212                 xfs_trans_cancel(tp, 0);
 213                 return error;
 214         }
 215
 216         /*
 217          * Write new AG headers to disk. Non-transactional, but written
 218          * synchronously so they are completed prior to the growfs transaction
 219          * being logged.
 220          */
 221         nfree = 0;
 222         for (agno = nagcount - 1; agno >= oagcount; agno--, new -= agsize) {
 223                 __be32  *agfl_bno;
 224
 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
 285                 agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, bp);
 286                 for (bucket = 0; bucket < XFS_AGFL_SIZE(mp); bucket++)
 287                         agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
 288
 289                 error = xfs_bwrite(bp);
 290                 xfs_buf_relse(bp);
 291                 if (error)
 292                         goto error0;
 293
 294                 /*
 295                  * AG inode header block
 296                  */
 297                 bp = xfs_growfs_get_hdr_buf(mp,
 298                                 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
 299                                 XFS_FSS_TO_BB(mp, 1), 0,
 300                                 &xfs_agi_buf_ops);
 301                 if (!bp) {
 302                         error = ENOMEM;
 303                         goto error0;
 304                 }
 305
 306                 agi = XFS_BUF_TO_AGI(bp);
 307                 agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
 308                 agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
 309                 agi->agi_seqno = cpu_to_be32(agno);
 310                 agi->agi_length = cpu_to_be32(agsize);
 311                 agi->agi_count = 0;
 312                 agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
 313                 agi->agi_level = cpu_to_be32(1);
 314                 agi->agi_freecount = 0;
 315                 agi->agi_newino = cpu_to_be32(NULLAGINO);
 316                 agi->agi_dirino = cpu_to_be32(NULLAGINO);
 317                 if (xfs_sb_version_hascrc(&mp->m_sb))
 318                         uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_uuid);
 319                 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
 320                         agi->agi_free_root = cpu_to_be32(XFS_FIBT_BLOCK(mp));
 321                         agi->agi_free_level = cpu_to_be32(1);
 322                 }
 323                 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
 324                         agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
 325
 326                 error = xfs_bwrite(bp);
 327                 xfs_buf_relse(bp);
 328                 if (error)
 329                         goto error0;
 330
 331                 /*
 332                  * BNO btree root block
 333                  */
 334                 bp = xfs_growfs_get_hdr_buf(mp,
 335                                 XFS_AGB_TO_DADDR(mp, agno, XFS_BNO_BLOCK(mp)),
 336                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 337                                 &xfs_allocbt_buf_ops);
 338
 339                 if (!bp) {
 340                         error = ENOMEM;
 341                         goto error0;
 342                 }
 343
 344                 if (xfs_sb_version_hascrc(&mp->m_sb))
 345                         xfs_btree_init_block(mp, bp, XFS_ABTB_CRC_MAGIC, 0, 1,
 346                                                 agno, XFS_BTREE_CRC_BLOCKS);
 347                 else
 348                         xfs_btree_init_block(mp, bp, XFS_ABTB_MAGIC, 0, 1,
 349                                                 agno, 0);
 350
 351                 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
 352                 arec->ar_startblock = cpu_to_be32(XFS_PREALLOC_BLOCKS(mp));
 353                 arec->ar_blockcount = cpu_to_be32(
 354                         agsize - be32_to_cpu(arec->ar_startblock));
 355
 356                 error = xfs_bwrite(bp);
 357                 xfs_buf_relse(bp);
 358                 if (error)
 359                         goto error0;
 360
 361                 /*
 362                  * CNT btree root block
 363                  */
 364                 bp = xfs_growfs_get_hdr_buf(mp,
 365                                 XFS_AGB_TO_DADDR(mp, agno, XFS_CNT_BLOCK(mp)),
 366                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 367                                 &xfs_allocbt_buf_ops);
 368                 if (!bp) {
 369                         error = ENOMEM;
 370                         goto error0;
 371                 }
 372
 373                 if (xfs_sb_version_hascrc(&mp->m_sb))
 374                         xfs_btree_init_block(mp, bp, XFS_ABTC_CRC_MAGIC, 0, 1,
 375                                                 agno, XFS_BTREE_CRC_BLOCKS);
 376                 else
 377                         xfs_btree_init_block(mp, bp, XFS_ABTC_MAGIC, 0, 1,
 378                                                 agno, 0);
 379
 380                 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
 381                 arec->ar_startblock = cpu_to_be32(XFS_PREALLOC_BLOCKS(mp));
 382                 arec->ar_blockcount = cpu_to_be32(
 383                         agsize - be32_to_cpu(arec->ar_startblock));
 384                 nfree += be32_to_cpu(arec->ar_blockcount);
 385
 386                 error = xfs_bwrite(bp);
 387                 xfs_buf_relse(bp);
 388                 if (error)
 389                         goto error0;
 390
 391                 /*
 392                  * INO btree root block
 393                  */
 394                 bp = xfs_growfs_get_hdr_buf(mp,
 395                                 XFS_AGB_TO_DADDR(mp, agno, XFS_IBT_BLOCK(mp)),
 396                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 397                                 &xfs_inobt_buf_ops);
 398                 if (!bp) {
 399                         error = ENOMEM;
 400                         goto error0;
 401                 }
 402
 403                 if (xfs_sb_version_hascrc(&mp->m_sb))
 404                         xfs_btree_init_block(mp, bp, XFS_IBT_CRC_MAGIC, 0, 0,
 405                                                 agno, XFS_BTREE_CRC_BLOCKS);
 406                 else
 407                         xfs_btree_init_block(mp, bp, XFS_IBT_MAGIC, 0, 0,
 408                                                 agno, 0);
 409
 410                 error = xfs_bwrite(bp);
 411                 xfs_buf_relse(bp);
 412                 if (error)
 413                         goto error0;
 414
 415                 /*
 416                  * FINO btree root block
 417                  */
 418                 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
 419                         bp = xfs_growfs_get_hdr_buf(mp,
 420                                 XFS_AGB_TO_DADDR(mp, agno, XFS_FIBT_BLOCK(mp)),
 421                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 422                                 &xfs_inobt_buf_ops);
 423                         if (!bp) {
 424                                 error = ENOMEM;
 425                                 goto error0;
 426                         }
 427
 428                         if (xfs_sb_version_hascrc(&mp->m_sb))
 429                                 xfs_btree_init_block(mp, bp, XFS_FIBT_CRC_MAGIC,
 430                                                      0, 0, agno,
 431                                                      XFS_BTREE_CRC_BLOCKS);
 432                         else
 433                                 xfs_btree_init_block(mp, bp, XFS_FIBT_MAGIC, 0,
 434                                                      0, agno, 0);
 435
 436                         error = xfs_bwrite(bp);
 437                         xfs_buf_relse(bp);
 438                         if (error)
 439                                 goto error0;
 440                 }
 441
 442         }
 443         xfs_trans_agblocks_delta(tp, nfree);
 444         /*
 445          * There are new blocks in the old last a.g.
 446          */
 447         if (new) {
 448                 /*
 449                  * Change the agi length.
 450                  */
 451                 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
 452                 if (error) {
 453                         goto error0;
 454                 }
 455                 ASSERT(bp);
 456                 agi = XFS_BUF_TO_AGI(bp);
 457                 be32_add_cpu(&agi->agi_length, new);
 458                 ASSERT(nagcount == oagcount ||
 459                        be32_to_cpu(agi->agi_length) == mp->m_sb.sb_agblocks);
 460                 xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
 461                 /*
 462                  * Change agf length.
 463                  */
 464                 error = xfs_alloc_read_agf(mp, tp, agno, 0, &bp);
 465                 if (error) {
 466                         goto error0;
 467                 }
 468                 ASSERT(bp);
 469                 agf = XFS_BUF_TO_AGF(bp);
 470                 be32_add_cpu(&agf->agf_length, new);
 471                 ASSERT(be32_to_cpu(agf->agf_length) ==
 472                        be32_to_cpu(agi->agi_length));
 473
 474                 xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
 475                 /*
 476                  * Free the new space.
 477                  */
 478                 error = xfs_free_extent(tp, XFS_AGB_TO_FSB(mp, agno,
 479                         be32_to_cpu(agf->agf_length) - new), new);
 480                 if (error) {
 481                         goto error0;
 482                 }
 483         }
 484
 485         /*
 486          * Update changed superblock fields transactionally. These are not
 487          * seen by the rest of the world until the transaction commit applies
 488          * them atomically to the superblock.
 489          */
 490         if (nagcount > oagcount)
 491                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
 492         if (nb > mp->m_sb.sb_dblocks)
 493                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS,
 494                                  nb - mp->m_sb.sb_dblocks);
 495         if (nfree)
 496                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, nfree);
 497         if (dpct)
 498                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
 499         error = xfs_trans_commit(tp, 0);
 500         if (error)
 501                 return error;
 502
 503         /* New allocation groups fully initialized, so update mount struct */
 504         if (nagimax)
 505                 mp->m_maxagi = nagimax;
 506         if (mp->m_sb.sb_imax_pct) {
 507                 __uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
 508                 do_div(icount, 100);
 509                 mp->m_maxicount = icount << mp->m_sb.sb_inopblog;
 510         } else
 511                 mp->m_maxicount = 0;
 512         xfs_set_low_space_thresholds(mp);
 513
 514         /* update secondary superblocks. */
 515         for (agno = 1; agno < nagcount; agno++) {
 516                 error = 0;
 517                 /*
 518                  * new secondary superblocks need to be zeroed, not read from
 519                  * disk as the contents of the new area we are growing into is
 520                  * completely unknown.
 521                  */
 522                 if (agno < oagcount) {
 523                         error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
 524                                   XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
 525                                   XFS_FSS_TO_BB(mp, 1), 0, &bp,
 526                                   &xfs_sb_buf_ops);
 527                 } else {
 528                         bp = xfs_trans_get_buf(NULL, mp->m_ddev_targp,
 529                                   XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
 530                                   XFS_FSS_TO_BB(mp, 1), 0);
 531                         if (bp) {
 532                                 bp->b_ops = &xfs_sb_buf_ops;
 533                                 xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
 534                         } else
 535                                 error = ENOMEM;
 536                 }
 537
 538                 /*
 539                  * If we get an error reading or writing alternate superblocks,
 540                  * continue.  xfs_repair chooses the "best" superblock based
 541                  * on most matches; if we break early, we'll leave more
 542                  * superblocks un-updated than updated, and xfs_repair may
 543                  * pick them over the properly-updated primary.
 544                  */
 545                 if (error) {
 546                         xfs_warn(mp,
 547                 "error %d reading secondary superblock for ag %d",
 548                                 error, agno);
 549                         saved_error = error;
 550                         continue;
 551                 }
 552                 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, XFS_SB_ALL_BITS);
 553
 554                 error = xfs_bwrite(bp);
 555                 xfs_buf_relse(bp);
 556                 if (error) {
 557                         xfs_warn(mp,
 558                 "write error %d updating secondary superblock for ag %d",
 559                                 error, agno);
 560                         saved_error = error;
 561                         continue;
 562                 }
 563         }
 564         return saved_error ? saved_error : error;
 565
 566  error0:
 567         xfs_trans_cancel(tp, XFS_TRANS_ABORT);
 568         return error;
 569 }
 570
 571 static int
 572 xfs_growfs_log_private(
 573         xfs_mount_t             *mp,    /* mount point for filesystem */
 574         xfs_growfs_log_t        *in)    /* growfs log input struct */
 575 {
 576         xfs_extlen_t            nb;
 577
 578         nb = in->newblocks;
 579         if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
 580                 return XFS_ERROR(EINVAL);
 581         if (nb == mp->m_sb.sb_logblocks &&
 582             in->isint == (mp->m_sb.sb_logstart != 0))
 583                 return XFS_ERROR(EINVAL);
 584         /*
 585          * Moving the log is hard, need new interfaces to sync
 586          * the log first, hold off all activity while moving it.
 587          * Can have shorter or longer log in the same space,
 588          * or transform internal to external log or vice versa.
 589          */
 590         return XFS_ERROR(ENOSYS);
 591 }
 592
 593 /*
 594  * protected versions of growfs function acquire and release locks on the mount
 595  * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
 596  * XFS_IOC_FSGROWFSRT
 597  */
 598
 599
 600 int
 601 xfs_growfs_data(
 602         xfs_mount_t             *mp,
 603         xfs_growfs_data_t       *in)
 604 {
 605         int error;
 606
 607         if (!capable(CAP_SYS_ADMIN))
 608                 return XFS_ERROR(EPERM);
 609         if (!mutex_trylock(&mp->m_growlock))
 610                 return XFS_ERROR(EWOULDBLOCK);
 611         error = xfs_growfs_data_private(mp, in);
 612         mutex_unlock(&mp->m_growlock);
 613         return error;
 614 }
 615
 616 int
 617 xfs_growfs_log(
 618         xfs_mount_t             *mp,
 619         xfs_growfs_log_t        *in)
 620 {
 621         int error;
 622
 623         if (!capable(CAP_SYS_ADMIN))
 624                 return XFS_ERROR(EPERM);
 625         if (!mutex_trylock(&mp->m_growlock))
 626                 return XFS_ERROR(EWOULDBLOCK);
 627         error = xfs_growfs_log_private(mp, in);
 628         mutex_unlock(&mp->m_growlock);
 629         return error;
 630 }
 631
 632 /*
 633  * exported through ioctl XFS_IOC_FSCOUNTS
 634  */
 635
 636 int
 637 xfs_fs_counts(
 638         xfs_mount_t             *mp,
 639         xfs_fsop_counts_t       *cnt)
 640 {
 641         xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
 642         spin_lock(&mp->m_sb_lock);
 643         cnt->freedata = mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
 644         cnt->freertx = mp->m_sb.sb_frextents;
 645         cnt->freeino = mp->m_sb.sb_ifree;
 646         cnt->allocino = mp->m_sb.sb_icount;
 647         spin_unlock(&mp->m_sb_lock);
 648         return 0;
 649 }
 650
 651 /*
 652  * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
 653  *
 654  * xfs_reserve_blocks is called to set m_resblks
 655  * in the in-core mount table. The number of unused reserved blocks
 656  * is kept in m_resblks_avail.
 657  *
 658  * Reserve the requested number of blocks if available. Otherwise return
 659  * as many as possible to satisfy the request. The actual number
 660  * reserved are returned in outval
 661  *
 662  * A null inval pointer indicates that only the current reserved blocks
 663  * available  should  be returned no settings are changed.
 664  */
 665
 666 int
 667 xfs_reserve_blocks(
 668         xfs_mount_t             *mp,
 669         __uint64_t              *inval,
 670         xfs_fsop_resblks_t      *outval)
 671 {
 672         __int64_t               lcounter, delta, fdblks_delta;
 673         __uint64_t              request;
 674
 675         /* If inval is null, report current values and return */
 676         if (inval == (__uint64_t *)NULL) {
 677                 if (!outval)
 678                         return EINVAL;
 679                 outval->resblks = mp->m_resblks;
 680                 outval->resblks_avail = mp->m_resblks_avail;
 681                 return 0;
 682         }
 683
 684         request = *inval;
 685
 686         /*
 687          * With per-cpu counters, this becomes an interesting
 688          * problem. we needto work out if we are freeing or allocation
 689          * blocks first, then we can do the modification as necessary.
 690          *
 691          * We do this under the m_sb_lock so that if we are near
 692          * ENOSPC, we will hold out any changes while we work out
 693          * what to do. This means that the amount of free space can
 694          * change while we do this, so we need to retry if we end up
 695          * trying to reserve more space than is available.
 696          *
 697          * We also use the xfs_mod_incore_sb() interface so that we
 698          * don't have to care about whether per cpu counter are
 699          * enabled, disabled or even compiled in....
 700          */
 701 retry:
 702         spin_lock(&mp->m_sb_lock);
 703         xfs_icsb_sync_counters_locked(mp, 0);
 704
 705         /*
 706          * If our previous reservation was larger than the current value,
 707          * then move any unused blocks back to the free pool.
 708          */
 709         fdblks_delta = 0;
 710         if (mp->m_resblks > request) {
 711                 lcounter = mp->m_resblks_avail - request;
 712                 if (lcounter  > 0) {            /* release unused blocks */
 713                         fdblks_delta = lcounter;
 714                         mp->m_resblks_avail -= lcounter;
 715                 }
 716                 mp->m_resblks = request;
 717         } else {
 718                 __int64_t       free;
 719
 720                 free =  mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
 721                 if (!free)
 722                         goto out; /* ENOSPC and fdblks_delta = 0 */
 723
 724                 delta = request - mp->m_resblks;
 725                 lcounter = free - delta;
 726                 if (lcounter < 0) {
 727                         /* We can't satisfy the request, just get what we can */
 728                         mp->m_resblks += free;
 729                         mp->m_resblks_avail += free;
 730                         fdblks_delta = -free;
 731                 } else {
 732                         fdblks_delta = -delta;
 733                         mp->m_resblks = request;
 734                         mp->m_resblks_avail += delta;
 735                 }
 736         }
 737 out:
 738         if (outval) {
 739                 outval->resblks = mp->m_resblks;
 740                 outval->resblks_avail = mp->m_resblks_avail;
 741         }
 742         spin_unlock(&mp->m_sb_lock);
 743
 744         if (fdblks_delta) {
 745                 /*
 746                  * If we are putting blocks back here, m_resblks_avail is
 747                  * already at its max so this will put it in the free pool.
 748                  *
 749                  * If we need space, we'll either succeed in getting it
 750                  * from the free block count or we'll get an enospc. If
 751                  * we get a ENOSPC, it means things changed while we were
 752                  * calculating fdblks_delta and so we should try again to
 753                  * see if there is anything left to reserve.
 754                  *
 755                  * Don't set the reserved flag here - we don't want to reserve
 756                  * the extra reserve blocks from the reserve.....
 757                  */
 758                 int error;
 759                 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
 760                                                  fdblks_delta, 0);
 761                 if (error == ENOSPC)
 762                         goto retry;
 763         }
 764         return 0;
 765 }
 766
 767 /*
 768  * Dump a transaction into the log that contains no real change. This is needed
 769  * to be able to make the log dirty or stamp the current tail LSN into the log
 770  * during the covering operation.
 771  *
 772  * We cannot use an inode here for this - that will push dirty state back up
 773  * into the VFS and then periodic inode flushing will prevent log covering from
 774  * making progress. Hence we log a field in the superblock instead and use a
 775  * synchronous transaction to ensure the superblock is immediately unpinned
 776  * and can be written back.
 777  */
 778 int
 779 xfs_fs_log_dummy(
 780         xfs_mount_t     *mp)
 781 {
 782         xfs_trans_t     *tp;
 783         int             error;
 784
 785         tp = _xfs_trans_alloc(mp, XFS_TRANS_DUMMY1, KM_SLEEP);
 786         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
 787         if (error) {
 788                 xfs_trans_cancel(tp, 0);
 789                 return error;
 790         }
 791
 792         /* log the UUID because it is an unchanging field */
 793         xfs_mod_sb(tp, XFS_SB_UUID);
 794         xfs_trans_set_sync(tp);
 795         return xfs_trans_commit(tp, 0);
 796 }
 797
 798 int
 799 xfs_fs_goingdown(
 800         xfs_mount_t     *mp,
 801         __uint32_t      inflags)
 802 {
 803         switch (inflags) {
 804         case XFS_FSOP_GOING_FLAGS_DEFAULT: {
 805                 struct super_block *sb = freeze_bdev(mp->m_super->s_bdev);
 806
 807                 if (sb && !IS_ERR(sb)) {
 808                         xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
 809                         thaw_bdev(sb->s_bdev, sb);
 810                 }
 811
 812                 break;
 813         }
 814         case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
 815                 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
 816                 break;
 817         case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
 818                 xfs_force_shutdown(mp,
 819                                 SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
 820                 break;
 821         default:
 822                 return XFS_ERROR(EINVAL);
 823         }
 824
 825         return 0;
 826 }
 827
 828 /*
 829  * Force a shutdown of the filesystem instantly while keeping the filesystem
 830  * consistent. We don't do an unmount here; just shutdown the shop, make sure
 831  * that absolutely nothing persistent happens to this filesystem after this
 832  * point.
 833  */
 834 void
 835 xfs_do_force_shutdown(
 836         xfs_mount_t     *mp,
 837         int             flags,
 838         char            *fname,
 839         int             lnnum)
 840 {
 841         int             logerror;
 842
 843         logerror = flags & SHUTDOWN_LOG_IO_ERROR;
 844
 845         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 846                 xfs_notice(mp,
 847         "%s(0x%x) called from line %d of file %s.  Return address = 0x%p",
 848                         __func__, flags, lnnum, fname, __return_address);
 849         }
 850         /*
 851          * No need to duplicate efforts.
 852          */
 853         if (XFS_FORCED_SHUTDOWN(mp) && !logerror)
 854                 return;
 855
 856         /*
 857          * This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't
 858          * queue up anybody new on the log reservations, and wakes up
 859          * everybody who's sleeping on log reservations to tell them
 860          * the bad news.
 861          */
 862         if (xfs_log_force_umount(mp, logerror))
 863                 return;
 864
 865         if (flags & SHUTDOWN_CORRUPT_INCORE) {
 866                 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_CORRUPT,
 867     "Corruption of in-memory data detected.  Shutting down filesystem");
 868                 if (XFS_ERRLEVEL_HIGH <= xfs_error_level)
 869                         xfs_stack_trace();
 870         } else if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 871                 if (logerror) {
 872                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_LOGERROR,
 873                 "Log I/O Error Detected.  Shutting down filesystem");
 874                 } else if (flags & SHUTDOWN_DEVICE_REQ) {
 875                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
 876                 "All device paths lost.  Shutting down filesystem");
 877                 } else if (!(flags & SHUTDOWN_REMOTE_REQ)) {
 878                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
 879                 "I/O Error Detected. Shutting down filesystem");
 880                 }
 881         }
 882         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 883                 xfs_alert(mp,
 884         "Please umount the filesystem and rectify the problem(s)");
 885         }
 886 }