fs/xfs/xfs_fsops.c

   1 /*
   2  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   3  * All Rights Reserved.
   4  *
   5  * This program is free software; you can redistribute it and/or
   6  * modify it under the terms of the GNU General Public License as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it would be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License
  15  * along with this program; if not, write the Free Software Foundation,
  16  * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  17  */
  18 #include "xfs.h"
  19 #include "xfs_fs.h"
  20 #include "xfs_shared.h"
  21 #include "xfs_format.h"
  22 #include "xfs_log_format.h"
  23 #include "xfs_trans_resv.h"
  24 #include "xfs_sb.h"
  25 #include "xfs_ag.h"
  26 #include "xfs_mount.h"
  27 #include "xfs_inode.h"
  28 #include "xfs_trans.h"
  29 #include "xfs_inode_item.h"
  30 #include "xfs_error.h"
  31 #include "xfs_btree.h"
  32 #include "xfs_alloc_btree.h"
  33 #include "xfs_alloc.h"
  34 #include "xfs_ialloc.h"
  35 #include "xfs_fsops.h"
  36 #include "xfs_itable.h"
  37 #include "xfs_trans_space.h"
  38 #include "xfs_rtalloc.h"
  39 #include "xfs_trace.h"
  40 #include "xfs_log.h"
  41 #include "xfs_dinode.h"
  42 #include "xfs_filestream.h"
  43
  44 /*
  45  * File system operations
  46  */
  47
  48 int
  49 xfs_fs_geometry(
  50         xfs_mount_t             *mp,
  51         xfs_fsop_geom_t         *geo,
  52         int                     new_version)
  53 {
  54
  55         memset(geo, 0, sizeof(*geo));
  56
  57         geo->blocksize = mp->m_sb.sb_blocksize;
  58         geo->rtextsize = mp->m_sb.sb_rextsize;
  59         geo->agblocks = mp->m_sb.sb_agblocks;
  60         geo->agcount = mp->m_sb.sb_agcount;
  61         geo->logblocks = mp->m_sb.sb_logblocks;
  62         geo->sectsize = mp->m_sb.sb_sectsize;
  63         geo->inodesize = mp->m_sb.sb_inodesize;
  64         geo->imaxpct = mp->m_sb.sb_imax_pct;
  65         geo->datablocks = mp->m_sb.sb_dblocks;
  66         geo->rtblocks = mp->m_sb.sb_rblocks;
  67         geo->rtextents = mp->m_sb.sb_rextents;
  68         geo->logstart = mp->m_sb.sb_logstart;
  69         ASSERT(sizeof(geo->uuid)==sizeof(mp->m_sb.sb_uuid));
  70         memcpy(geo->uuid, &mp->m_sb.sb_uuid, sizeof(mp->m_sb.sb_uuid));
  71         if (new_version >= 2) {
  72                 geo->sunit = mp->m_sb.sb_unit;
  73                 geo->swidth = mp->m_sb.sb_width;
  74         }
  75         if (new_version >= 3) {
  76                 geo->version = XFS_FSOP_GEOM_VERSION;
  77                 geo->flags = XFS_FSOP_GEOM_FLAGS_NLINK |
  78                              XFS_FSOP_GEOM_FLAGS_DIRV2 |
  79                         (xfs_sb_version_hasattr(&mp->m_sb) ?
  80                                 XFS_FSOP_GEOM_FLAGS_ATTR : 0) |
  81                         (xfs_sb_version_hasquota(&mp->m_sb) ?
  82                                 XFS_FSOP_GEOM_FLAGS_QUOTA : 0) |
  83                         (xfs_sb_version_hasalign(&mp->m_sb) ?
  84                                 XFS_FSOP_GEOM_FLAGS_IALIGN : 0) |
  85                         (xfs_sb_version_hasdalign(&mp->m_sb) ?
  86                                 XFS_FSOP_GEOM_FLAGS_DALIGN : 0) |
  87                         (xfs_sb_version_hasextflgbit(&mp->m_sb) ?
  88                                 XFS_FSOP_GEOM_FLAGS_EXTFLG : 0) |
  89                         (xfs_sb_version_hassector(&mp->m_sb) ?
  90                                 XFS_FSOP_GEOM_FLAGS_SECTOR : 0) |
  91                         (xfs_sb_version_hasasciici(&mp->m_sb) ?
  92                                 XFS_FSOP_GEOM_FLAGS_DIRV2CI : 0) |
  93                         (xfs_sb_version_haslazysbcount(&mp->m_sb) ?
  94                                 XFS_FSOP_GEOM_FLAGS_LAZYSB : 0) |
  95                         (xfs_sb_version_hasattr2(&mp->m_sb) ?
  96                                 XFS_FSOP_GEOM_FLAGS_ATTR2 : 0) |
  97                         (xfs_sb_version_hasprojid32bit(&mp->m_sb) ?
  98                                 XFS_FSOP_GEOM_FLAGS_PROJID32 : 0) |
  99                         (xfs_sb_version_hascrc(&mp->m_sb) ?
 100                                 XFS_FSOP_GEOM_FLAGS_V5SB : 0) |
 101                         (xfs_sb_version_hasftype(&mp->m_sb) ?
 102                                 XFS_FSOP_GEOM_FLAGS_FTYPE : 0) |
 103                         (xfs_sb_version_hasfinobt(&mp->m_sb) ?
 104                                 XFS_FSOP_GEOM_FLAGS_FINOBT : 0);
 105                 geo->logsectsize = xfs_sb_version_hassector(&mp->m_sb) ?
 106                                 mp->m_sb.sb_logsectsize : BBSIZE;
 107                 geo->rtsectsize = mp->m_sb.sb_blocksize;
 108                 geo->dirblocksize = mp->m_dirblksize;
 109         }
 110         if (new_version >= 4) {
 111                 geo->flags |=
 112                         (xfs_sb_version_haslogv2(&mp->m_sb) ?
 113                                 XFS_FSOP_GEOM_FLAGS_LOGV2 : 0);
 114                 geo->logsunit = mp->m_sb.sb_logsunit;
 115         }
 116         return 0;
 117 }
 118
 119 static struct xfs_buf *
 120 xfs_growfs_get_hdr_buf(
 121         struct xfs_mount        *mp,
 122         xfs_daddr_t             blkno,
 123         size_t                  numblks,
 124         int                     flags,
 125         const struct xfs_buf_ops *ops)
 126 {
 127         struct xfs_buf          *bp;
 128
 129         bp = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, flags);
 130         if (!bp)
 131                 return NULL;
 132
 133         xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
 134         bp->b_bn = blkno;
 135         bp->b_maps[0].bm_bn = blkno;
 136         bp->b_ops = ops;
 137
 138         return bp;
 139 }
 140
 141 static int
 142 xfs_growfs_data_private(
 143         xfs_mount_t             *mp,            /* mount point for filesystem */
 144         xfs_growfs_data_t       *in)            /* growfs data input struct */
 145 {
 146         xfs_agf_t               *agf;
 147         struct xfs_agfl         *agfl;
 148         xfs_agi_t               *agi;
 149         xfs_agnumber_t          agno;
 150         xfs_extlen_t            agsize;
 151         xfs_extlen_t            tmpsize;
 152         xfs_alloc_rec_t         *arec;
 153         xfs_buf_t               *bp;
 154         int                     bucket;
 155         int                     dpct;
 156         int                     error, saved_error = 0;
 157         xfs_agnumber_t          nagcount;
 158         xfs_agnumber_t          nagimax = 0;
 159         xfs_rfsblock_t          nb, nb_mod;
 160         xfs_rfsblock_t          new;
 161         xfs_rfsblock_t          nfree;
 162         xfs_agnumber_t          oagcount;
 163         int                     pct;
 164         xfs_trans_t             *tp;
 165
 166         nb = in->newblocks;
 167         pct = in->imaxpct;
 168         if (nb < mp->m_sb.sb_dblocks || pct < 0 || pct > 100)
 169                 return XFS_ERROR(EINVAL);
 170         if ((error = xfs_sb_validate_fsb_count(&mp->m_sb, nb)))
 171                 return error;
 172         dpct = pct - mp->m_sb.sb_imax_pct;
 173         bp = xfs_buf_read_uncached(mp->m_ddev_targp,
 174                                 XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1),
 175                                 XFS_FSS_TO_BB(mp, 1), 0, NULL);
 176         if (!bp)
 177                 return EIO;
 178         if (bp->b_error) {
 179                 error = bp->b_error;
 180                 xfs_buf_relse(bp);
 181                 return error;
 182         }
 183         xfs_buf_relse(bp);
 184
 185         new = nb;       /* use new as a temporary here */
 186         nb_mod = do_div(new, mp->m_sb.sb_agblocks);
 187         nagcount = new + (nb_mod != 0);
 188         if (nb_mod && nb_mod < XFS_MIN_AG_BLOCKS) {
 189                 nagcount--;
 190                 nb = (xfs_rfsblock_t)nagcount * mp->m_sb.sb_agblocks;
 191                 if (nb < mp->m_sb.sb_dblocks)
 192                         return XFS_ERROR(EINVAL);
 193         }
 194         new = nb - mp->m_sb.sb_dblocks;
 195         oagcount = mp->m_sb.sb_agcount;
 196
 197         /* allocate the new per-ag structures */
 198         if (nagcount > oagcount) {
 199                 error = xfs_initialize_perag(mp, nagcount, &nagimax);
 200                 if (error)
 201                         return error;
 202         }
 203
 204         tp = xfs_trans_alloc(mp, XFS_TRANS_GROWFS);
 205         tp->t_flags |= XFS_TRANS_RESERVE;
 206         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_growdata,
 207                                   XFS_GROWFS_SPACE_RES(mp), 0);
 208         if (error) {
 209                 xfs_trans_cancel(tp, 0);
 210                 return error;
 211         }
 212
 213         /*
 214          * Write new AG headers to disk. Non-transactional, but written
 215          * synchronously so they are completed prior to the growfs transaction
 216          * being logged.
 217          */
 218         nfree = 0;
 219         for (agno = nagcount - 1; agno >= oagcount; agno--, new -= agsize) {
 220                 __be32  *agfl_bno;
 221
 222                 /*
 223                  * AG freespace header block
 224                  */
 225                 bp = xfs_growfs_get_hdr_buf(mp,
 226                                 XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
 227                                 XFS_FSS_TO_BB(mp, 1), 0,
 228                                 &xfs_agf_buf_ops);
 229                 if (!bp) {
 230                         error = ENOMEM;
 231                         goto error0;
 232                 }
 233
 234                 agf = XFS_BUF_TO_AGF(bp);
 235                 agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
 236                 agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
 237                 agf->agf_seqno = cpu_to_be32(agno);
 238                 if (agno == nagcount - 1)
 239                         agsize =
 240                                 nb -
 241                                 (agno * (xfs_rfsblock_t)mp->m_sb.sb_agblocks);
 242                 else
 243                         agsize = mp->m_sb.sb_agblocks;
 244                 agf->agf_length = cpu_to_be32(agsize);
 245                 agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
 246                 agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
 247                 agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
 248                 agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
 249                 agf->agf_flfirst = 0;
 250                 agf->agf_fllast = cpu_to_be32(XFS_AGFL_SIZE(mp) - 1);
 251                 agf->agf_flcount = 0;
 252                 tmpsize = agsize - XFS_PREALLOC_BLOCKS(mp);
 253                 agf->agf_freeblks = cpu_to_be32(tmpsize);
 254                 agf->agf_longest = cpu_to_be32(tmpsize);
 255                 if (xfs_sb_version_hascrc(&mp->m_sb))
 256                         uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_uuid);
 257
 258                 error = xfs_bwrite(bp);
 259                 xfs_buf_relse(bp);
 260                 if (error)
 261                         goto error0;
 262
 263                 /*
 264                  * AG freelist header block
 265                  */
 266                 bp = xfs_growfs_get_hdr_buf(mp,
 267                                 XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)),
 268                                 XFS_FSS_TO_BB(mp, 1), 0,
 269                                 &xfs_agfl_buf_ops);
 270                 if (!bp) {
 271                         error = ENOMEM;
 272                         goto error0;
 273                 }
 274
 275                 agfl = XFS_BUF_TO_AGFL(bp);
 276                 if (xfs_sb_version_hascrc(&mp->m_sb)) {
 277                         agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
 278                         agfl->agfl_seqno = cpu_to_be32(agno);
 279                         uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_uuid);
 280                 }
 281
 282                 agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, bp);
 283                 for (bucket = 0; bucket < XFS_AGFL_SIZE(mp); bucket++)
 284                         agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
 285
 286                 error = xfs_bwrite(bp);
 287                 xfs_buf_relse(bp);
 288                 if (error)
 289                         goto error0;
 290
 291                 /*
 292                  * AG inode header block
 293                  */
 294                 bp = xfs_growfs_get_hdr_buf(mp,
 295                                 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
 296                                 XFS_FSS_TO_BB(mp, 1), 0,
 297                                 &xfs_agi_buf_ops);
 298                 if (!bp) {
 299                         error = ENOMEM;
 300                         goto error0;
 301                 }
 302
 303                 agi = XFS_BUF_TO_AGI(bp);
 304                 agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
 305                 agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
 306                 agi->agi_seqno = cpu_to_be32(agno);
 307                 agi->agi_length = cpu_to_be32(agsize);
 308                 agi->agi_count = 0;
 309                 agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
 310                 agi->agi_level = cpu_to_be32(1);
 311                 agi->agi_freecount = 0;
 312                 agi->agi_newino = cpu_to_be32(NULLAGINO);
 313                 agi->agi_dirino = cpu_to_be32(NULLAGINO);
 314                 if (xfs_sb_version_hascrc(&mp->m_sb))
 315                         uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_uuid);
 316                 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
 317                         agi->agi_free_root = cpu_to_be32(XFS_FIBT_BLOCK(mp));
 318                         agi->agi_free_level = cpu_to_be32(1);
 319                 }
 320                 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
 321                         agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
 322
 323                 error = xfs_bwrite(bp);
 324                 xfs_buf_relse(bp);
 325                 if (error)
 326                         goto error0;
 327
 328                 /*
 329                  * BNO btree root block
 330                  */
 331                 bp = xfs_growfs_get_hdr_buf(mp,
 332                                 XFS_AGB_TO_DADDR(mp, agno, XFS_BNO_BLOCK(mp)),
 333                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 334                                 &xfs_allocbt_buf_ops);
 335
 336                 if (!bp) {
 337                         error = ENOMEM;
 338                         goto error0;
 339                 }
 340
 341                 if (xfs_sb_version_hascrc(&mp->m_sb))
 342                         xfs_btree_init_block(mp, bp, XFS_ABTB_CRC_MAGIC, 0, 1,
 343                                                 agno, XFS_BTREE_CRC_BLOCKS);
 344                 else
 345                         xfs_btree_init_block(mp, bp, XFS_ABTB_MAGIC, 0, 1,
 346                                                 agno, 0);
 347
 348                 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
 349                 arec->ar_startblock = cpu_to_be32(XFS_PREALLOC_BLOCKS(mp));
 350                 arec->ar_blockcount = cpu_to_be32(
 351                         agsize - be32_to_cpu(arec->ar_startblock));
 352
 353                 error = xfs_bwrite(bp);
 354                 xfs_buf_relse(bp);
 355                 if (error)
 356                         goto error0;
 357
 358                 /*
 359                  * CNT btree root block
 360                  */
 361                 bp = xfs_growfs_get_hdr_buf(mp,
 362                                 XFS_AGB_TO_DADDR(mp, agno, XFS_CNT_BLOCK(mp)),
 363                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 364                                 &xfs_allocbt_buf_ops);
 365                 if (!bp) {
 366                         error = ENOMEM;
 367                         goto error0;
 368                 }
 369
 370                 if (xfs_sb_version_hascrc(&mp->m_sb))
 371                         xfs_btree_init_block(mp, bp, XFS_ABTC_CRC_MAGIC, 0, 1,
 372                                                 agno, XFS_BTREE_CRC_BLOCKS);
 373                 else
 374                         xfs_btree_init_block(mp, bp, XFS_ABTC_MAGIC, 0, 1,
 375                                                 agno, 0);
 376
 377                 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
 378                 arec->ar_startblock = cpu_to_be32(XFS_PREALLOC_BLOCKS(mp));
 379                 arec->ar_blockcount = cpu_to_be32(
 380                         agsize - be32_to_cpu(arec->ar_startblock));
 381                 nfree += be32_to_cpu(arec->ar_blockcount);
 382
 383                 error = xfs_bwrite(bp);
 384                 xfs_buf_relse(bp);
 385                 if (error)
 386                         goto error0;
 387
 388                 /*
 389                  * INO btree root block
 390                  */
 391                 bp = xfs_growfs_get_hdr_buf(mp,
 392                                 XFS_AGB_TO_DADDR(mp, agno, XFS_IBT_BLOCK(mp)),
 393                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 394                                 &xfs_inobt_buf_ops);
 395                 if (!bp) {
 396                         error = ENOMEM;
 397                         goto error0;
 398                 }
 399
 400                 if (xfs_sb_version_hascrc(&mp->m_sb))
 401                         xfs_btree_init_block(mp, bp, XFS_IBT_CRC_MAGIC, 0, 0,
 402                                                 agno, XFS_BTREE_CRC_BLOCKS);
 403                 else
 404                         xfs_btree_init_block(mp, bp, XFS_IBT_MAGIC, 0, 0,
 405                                                 agno, 0);
 406
 407                 error = xfs_bwrite(bp);
 408                 xfs_buf_relse(bp);
 409                 if (error)
 410                         goto error0;
 411
 412                 /*
 413                  * FINO btree root block
 414                  */
 415                 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
 416                         bp = xfs_growfs_get_hdr_buf(mp,
 417                                 XFS_AGB_TO_DADDR(mp, agno, XFS_FIBT_BLOCK(mp)),
 418                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 419                                 &xfs_inobt_buf_ops);
 420                         if (!bp) {
 421                                 error = ENOMEM;
 422                                 goto error0;
 423                         }
 424
 425                         if (xfs_sb_version_hascrc(&mp->m_sb))
 426                                 xfs_btree_init_block(mp, bp, XFS_FIBT_CRC_MAGIC,
 427                                                      0, 0, agno,
 428                                                      XFS_BTREE_CRC_BLOCKS);
 429                         else
 430                                 xfs_btree_init_block(mp, bp, XFS_FIBT_MAGIC, 0,
 431                                                      0, agno, 0);
 432
 433                         error = xfs_bwrite(bp);
 434                         xfs_buf_relse(bp);
 435                         if (error)
 436                                 goto error0;
 437                 }
 438
 439         }
 440         xfs_trans_agblocks_delta(tp, nfree);
 441         /*
 442          * There are new blocks in the old last a.g.
 443          */
 444         if (new) {
 445                 /*
 446                  * Change the agi length.
 447                  */
 448                 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
 449                 if (error) {
 450                         goto error0;
 451                 }
 452                 ASSERT(bp);
 453                 agi = XFS_BUF_TO_AGI(bp);
 454                 be32_add_cpu(&agi->agi_length, new);
 455                 ASSERT(nagcount == oagcount ||
 456                        be32_to_cpu(agi->agi_length) == mp->m_sb.sb_agblocks);
 457                 xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
 458                 /*
 459                  * Change agf length.
 460                  */
 461                 error = xfs_alloc_read_agf(mp, tp, agno, 0, &bp);
 462                 if (error) {
 463                         goto error0;
 464                 }
 465                 ASSERT(bp);
 466                 agf = XFS_BUF_TO_AGF(bp);
 467                 be32_add_cpu(&agf->agf_length, new);
 468                 ASSERT(be32_to_cpu(agf->agf_length) ==
 469                        be32_to_cpu(agi->agi_length));
 470
 471                 xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
 472                 /*
 473                  * Free the new space.
 474                  */
 475                 error = xfs_free_extent(tp, XFS_AGB_TO_FSB(mp, agno,
 476                         be32_to_cpu(agf->agf_length) - new), new);
 477                 if (error) {
 478                         goto error0;
 479                 }
 480         }
 481
 482         /*
 483          * Update changed superblock fields transactionally. These are not
 484          * seen by the rest of the world until the transaction commit applies
 485          * them atomically to the superblock.
 486          */
 487         if (nagcount > oagcount)
 488                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
 489         if (nb > mp->m_sb.sb_dblocks)
 490                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS,
 491                                  nb - mp->m_sb.sb_dblocks);
 492         if (nfree)
 493                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, nfree);
 494         if (dpct)
 495                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
 496         error = xfs_trans_commit(tp, 0);
 497         if (error)
 498                 return error;
 499
 500         /* New allocation groups fully initialized, so update mount struct */
 501         if (nagimax)
 502                 mp->m_maxagi = nagimax;
 503         if (mp->m_sb.sb_imax_pct) {
 504                 __uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
 505                 do_div(icount, 100);
 506                 mp->m_maxicount = icount << mp->m_sb.sb_inopblog;
 507         } else
 508                 mp->m_maxicount = 0;
 509         xfs_set_low_space_thresholds(mp);
 510
 511         /* update secondary superblocks. */
 512         for (agno = 1; agno < nagcount; agno++) {
 513                 error = 0;
 514                 /*
 515                  * new secondary superblocks need to be zeroed, not read from
 516                  * disk as the contents of the new area we are growing into is
 517                  * completely unknown.
 518                  */
 519                 if (agno < oagcount) {
 520                         error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
 521                                   XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
 522                                   XFS_FSS_TO_BB(mp, 1), 0, &bp,
 523                                   &xfs_sb_buf_ops);
 524                 } else {
 525                         bp = xfs_trans_get_buf(NULL, mp->m_ddev_targp,
 526                                   XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
 527                                   XFS_FSS_TO_BB(mp, 1), 0);
 528                         if (bp) {
 529                                 bp->b_ops = &xfs_sb_buf_ops;
 530                                 xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
 531                         } else
 532                                 error = ENOMEM;
 533                 }
 534
 535                 /*
 536                  * If we get an error reading or writing alternate superblocks,
 537                  * continue.  xfs_repair chooses the "best" superblock based
 538                  * on most matches; if we break early, we'll leave more
 539                  * superblocks un-updated than updated, and xfs_repair may
 540                  * pick them over the properly-updated primary.
 541                  */
 542                 if (error) {
 543                         xfs_warn(mp,
 544                 "error %d reading secondary superblock for ag %d",
 545                                 error, agno);
 546                         saved_error = error;
 547                         continue;
 548                 }
 549                 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb, XFS_SB_ALL_BITS);
 550
 551                 error = xfs_bwrite(bp);
 552                 xfs_buf_relse(bp);
 553                 if (error) {
 554                         xfs_warn(mp,
 555                 "write error %d updating secondary superblock for ag %d",
 556                                 error, agno);
 557                         saved_error = error;
 558                         continue;
 559                 }
 560         }
 561         return saved_error ? saved_error : error;
 562
 563  error0:
 564         xfs_trans_cancel(tp, XFS_TRANS_ABORT);
 565         return error;
 566 }
 567
 568 static int
 569 xfs_growfs_log_private(
 570         xfs_mount_t             *mp,    /* mount point for filesystem */
 571         xfs_growfs_log_t        *in)    /* growfs log input struct */
 572 {
 573         xfs_extlen_t            nb;
 574
 575         nb = in->newblocks;
 576         if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
 577                 return XFS_ERROR(EINVAL);
 578         if (nb == mp->m_sb.sb_logblocks &&
 579             in->isint == (mp->m_sb.sb_logstart != 0))
 580                 return XFS_ERROR(EINVAL);
 581         /*
 582          * Moving the log is hard, need new interfaces to sync
 583          * the log first, hold off all activity while moving it.
 584          * Can have shorter or longer log in the same space,
 585          * or transform internal to external log or vice versa.
 586          */
 587         return XFS_ERROR(ENOSYS);
 588 }
 589
 590 /*
 591  * protected versions of growfs function acquire and release locks on the mount
 592  * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
 593  * XFS_IOC_FSGROWFSRT
 594  */
 595
 596
 597 int
 598 xfs_growfs_data(
 599         xfs_mount_t             *mp,
 600         xfs_growfs_data_t       *in)
 601 {
 602         int error;
 603
 604         if (!capable(CAP_SYS_ADMIN))
 605                 return XFS_ERROR(EPERM);
 606         if (!mutex_trylock(&mp->m_growlock))
 607                 return XFS_ERROR(EWOULDBLOCK);
 608         error = xfs_growfs_data_private(mp, in);
 609         mutex_unlock(&mp->m_growlock);
 610         return error;
 611 }
 612
 613 int
 614 xfs_growfs_log(
 615         xfs_mount_t             *mp,
 616         xfs_growfs_log_t        *in)
 617 {
 618         int error;
 619
 620         if (!capable(CAP_SYS_ADMIN))
 621                 return XFS_ERROR(EPERM);
 622         if (!mutex_trylock(&mp->m_growlock))
 623                 return XFS_ERROR(EWOULDBLOCK);
 624         error = xfs_growfs_log_private(mp, in);
 625         mutex_unlock(&mp->m_growlock);
 626         return error;
 627 }
 628
 629 /*
 630  * exported through ioctl XFS_IOC_FSCOUNTS
 631  */
 632
 633 int
 634 xfs_fs_counts(
 635         xfs_mount_t             *mp,
 636         xfs_fsop_counts_t       *cnt)
 637 {
 638         xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
 639         spin_lock(&mp->m_sb_lock);
 640         cnt->freedata = mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
 641         cnt->freertx = mp->m_sb.sb_frextents;
 642         cnt->freeino = mp->m_sb.sb_ifree;
 643         cnt->allocino = mp->m_sb.sb_icount;
 644         spin_unlock(&mp->m_sb_lock);
 645         return 0;
 646 }
 647
 648 /*
 649  * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
 650  *
 651  * xfs_reserve_blocks is called to set m_resblks
 652  * in the in-core mount table. The number of unused reserved blocks
 653  * is kept in m_resblks_avail.
 654  *
 655  * Reserve the requested number of blocks if available. Otherwise return
 656  * as many as possible to satisfy the request. The actual number
 657  * reserved are returned in outval
 658  *
 659  * A null inval pointer indicates that only the current reserved blocks
 660  * available  should  be returned no settings are changed.
 661  */
 662
 663 int
 664 xfs_reserve_blocks(
 665         xfs_mount_t             *mp,
 666         __uint64_t              *inval,
 667         xfs_fsop_resblks_t      *outval)
 668 {
 669         __int64_t               lcounter, delta, fdblks_delta;
 670         __uint64_t              request;
 671
 672         /* If inval is null, report current values and return */
 673         if (inval == (__uint64_t *)NULL) {
 674                 if (!outval)
 675                         return EINVAL;
 676                 outval->resblks = mp->m_resblks;
 677                 outval->resblks_avail = mp->m_resblks_avail;
 678                 return 0;
 679         }
 680
 681         request = *inval;
 682
 683         /*
 684          * With per-cpu counters, this becomes an interesting
 685          * problem. we needto work out if we are freeing or allocation
 686          * blocks first, then we can do the modification as necessary.
 687          *
 688          * We do this under the m_sb_lock so that if we are near
 689          * ENOSPC, we will hold out any changes while we work out
 690          * what to do. This means that the amount of free space can
 691          * change while we do this, so we need to retry if we end up
 692          * trying to reserve more space than is available.
 693          *
 694          * We also use the xfs_mod_incore_sb() interface so that we
 695          * don't have to care about whether per cpu counter are
 696          * enabled, disabled or even compiled in....
 697          */
 698 retry:
 699         spin_lock(&mp->m_sb_lock);
 700         xfs_icsb_sync_counters_locked(mp, 0);
 701
 702         /*
 703          * If our previous reservation was larger than the current value,
 704          * then move any unused blocks back to the free pool.
 705          */
 706         fdblks_delta = 0;
 707         if (mp->m_resblks > request) {
 708                 lcounter = mp->m_resblks_avail - request;
 709                 if (lcounter  > 0) {            /* release unused blocks */
 710                         fdblks_delta = lcounter;
 711                         mp->m_resblks_avail -= lcounter;
 712                 }
 713                 mp->m_resblks = request;
 714         } else {
 715                 __int64_t       free;
 716
 717                 free =  mp->m_sb.sb_fdblocks - XFS_ALLOC_SET_ASIDE(mp);
 718                 if (!free)
 719                         goto out; /* ENOSPC and fdblks_delta = 0 */
 720
 721                 delta = request - mp->m_resblks;
 722                 lcounter = free - delta;
 723                 if (lcounter < 0) {
 724                         /* We can't satisfy the request, just get what we can */
 725                         mp->m_resblks += free;
 726                         mp->m_resblks_avail += free;
 727                         fdblks_delta = -free;
 728                 } else {
 729                         fdblks_delta = -delta;
 730                         mp->m_resblks = request;
 731                         mp->m_resblks_avail += delta;
 732                 }
 733         }
 734 out:
 735         if (outval) {
 736                 outval->resblks = mp->m_resblks;
 737                 outval->resblks_avail = mp->m_resblks_avail;
 738         }
 739         spin_unlock(&mp->m_sb_lock);
 740
 741         if (fdblks_delta) {
 742                 /*
 743                  * If we are putting blocks back here, m_resblks_avail is
 744                  * already at its max so this will put it in the free pool.
 745                  *
 746                  * If we need space, we'll either succeed in getting it
 747                  * from the free block count or we'll get an enospc. If
 748                  * we get a ENOSPC, it means things changed while we were
 749                  * calculating fdblks_delta and so we should try again to
 750                  * see if there is anything left to reserve.
 751                  *
 752                  * Don't set the reserved flag here - we don't want to reserve
 753                  * the extra reserve blocks from the reserve.....
 754                  */
 755                 int error;
 756                 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
 757                                                  fdblks_delta, 0);
 758                 if (error == ENOSPC)
 759                         goto retry;
 760         }
 761         return 0;
 762 }
 763
 764 /*
 765  * Dump a transaction into the log that contains no real change. This is needed
 766  * to be able to make the log dirty or stamp the current tail LSN into the log
 767  * during the covering operation.
 768  *
 769  * We cannot use an inode here for this - that will push dirty state back up
 770  * into the VFS and then periodic inode flushing will prevent log covering from
 771  * making progress. Hence we log a field in the superblock instead and use a
 772  * synchronous transaction to ensure the superblock is immediately unpinned
 773  * and can be written back.
 774  */
 775 int
 776 xfs_fs_log_dummy(
 777         xfs_mount_t     *mp)
 778 {
 779         xfs_trans_t     *tp;
 780         int             error;
 781
 782         tp = _xfs_trans_alloc(mp, XFS_TRANS_DUMMY1, KM_SLEEP);
 783         error = xfs_trans_reserve(tp, &M_RES(mp)->tr_sb, 0, 0);
 784         if (error) {
 785                 xfs_trans_cancel(tp, 0);
 786                 return error;
 787         }
 788
 789         /* log the UUID because it is an unchanging field */
 790         xfs_mod_sb(tp, XFS_SB_UUID);
 791         xfs_trans_set_sync(tp);
 792         return xfs_trans_commit(tp, 0);
 793 }
 794
 795 int
 796 xfs_fs_goingdown(
 797         xfs_mount_t     *mp,
 798         __uint32_t      inflags)
 799 {
 800         switch (inflags) {
 801         case XFS_FSOP_GOING_FLAGS_DEFAULT: {
 802                 struct super_block *sb = freeze_bdev(mp->m_super->s_bdev);
 803
 804                 if (sb && !IS_ERR(sb)) {
 805                         xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
 806                         thaw_bdev(sb->s_bdev, sb);
 807                 }
 808
 809                 break;
 810         }
 811         case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
 812                 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
 813                 break;
 814         case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
 815                 xfs_force_shutdown(mp,
 816                                 SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
 817                 break;
 818         default:
 819                 return XFS_ERROR(EINVAL);
 820         }
 821
 822         return 0;
 823 }
 824
 825 /*
 826  * Force a shutdown of the filesystem instantly while keeping the filesystem
 827  * consistent. We don't do an unmount here; just shutdown the shop, make sure
 828  * that absolutely nothing persistent happens to this filesystem after this
 829  * point.
 830  */
 831 void
 832 xfs_do_force_shutdown(
 833         xfs_mount_t     *mp,
 834         int             flags,
 835         char            *fname,
 836         int             lnnum)
 837 {
 838         int             logerror;
 839
 840         logerror = flags & SHUTDOWN_LOG_IO_ERROR;
 841
 842         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 843                 xfs_notice(mp,
 844         "%s(0x%x) called from line %d of file %s.  Return address = 0x%p",
 845                         __func__, flags, lnnum, fname, __return_address);
 846         }
 847         /*
 848          * No need to duplicate efforts.
 849          */
 850         if (XFS_FORCED_SHUTDOWN(mp) && !logerror)
 851                 return;
 852
 853         /*
 854          * This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't
 855          * queue up anybody new on the log reservations, and wakes up
 856          * everybody who's sleeping on log reservations to tell them
 857          * the bad news.
 858          */
 859         if (xfs_log_force_umount(mp, logerror))
 860                 return;
 861
 862         if (flags & SHUTDOWN_CORRUPT_INCORE) {
 863                 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_CORRUPT,
 864     "Corruption of in-memory data detected.  Shutting down filesystem");
 865                 if (XFS_ERRLEVEL_HIGH <= xfs_error_level)
 866                         xfs_stack_trace();
 867         } else if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 868                 if (logerror) {
 869                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_LOGERROR,
 870                 "Log I/O Error Detected.  Shutting down filesystem");
 871                 } else if (flags & SHUTDOWN_DEVICE_REQ) {
 872                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
 873                 "All device paths lost.  Shutting down filesystem");
 874                 } else if (!(flags & SHUTDOWN_REMOTE_REQ)) {
 875                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
 876                 "I/O Error Detected. Shutting down filesystem");
 877                 }
 878         }
 879         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 880                 xfs_alert(mp,
 881         "Please umount the filesystem and rectify the problem(s)");
 882         }
 883 }