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_mount.h"
  26 #include "xfs_defer.h"
  27 #include "xfs_da_format.h"
  28 #include "xfs_da_btree.h"
  29 #include "xfs_inode.h"
  30 #include "xfs_trans.h"
  31 #include "xfs_inode_item.h"
  32 #include "xfs_error.h"
  33 #include "xfs_btree.h"
  34 #include "xfs_alloc_btree.h"
  35 #include "xfs_alloc.h"
  36 #include "xfs_rmap_btree.h"
  37 #include "xfs_ialloc.h"
  38 #include "xfs_fsops.h"
  39 #include "xfs_itable.h"
  40 #include "xfs_trans_space.h"
  41 #include "xfs_rtalloc.h"
  42 #include "xfs_trace.h"
  43 #include "xfs_log.h"
  44 #include "xfs_filestream.h"
  45 #include "xfs_rmap.h"
  46
  47 /*
  48  * File system operations
  49  */
  50
  51 int
  52 xfs_fs_geometry(
  53         xfs_mount_t             *mp,
  54         xfs_fsop_geom_t         *geo,
  55         int                     new_version)
  56 {
  57
  58         memset(geo, 0, sizeof(*geo));
  59
  60         geo->blocksize = mp->m_sb.sb_blocksize;
  61         geo->rtextsize = mp->m_sb.sb_rextsize;
  62         geo->agblocks = mp->m_sb.sb_agblocks;
  63         geo->agcount = mp->m_sb.sb_agcount;
  64         geo->logblocks = mp->m_sb.sb_logblocks;
  65         geo->sectsize = mp->m_sb.sb_sectsize;
  66         geo->inodesize = mp->m_sb.sb_inodesize;
  67         geo->imaxpct = mp->m_sb.sb_imax_pct;
  68         geo->datablocks = mp->m_sb.sb_dblocks;
  69         geo->rtblocks = mp->m_sb.sb_rblocks;
  70         geo->rtextents = mp->m_sb.sb_rextents;
  71         geo->logstart = mp->m_sb.sb_logstart;
  72         ASSERT(sizeof(geo->uuid)==sizeof(mp->m_sb.sb_uuid));
  73         memcpy(geo->uuid, &mp->m_sb.sb_uuid, sizeof(mp->m_sb.sb_uuid));
  74         if (new_version >= 2) {
  75                 geo->sunit = mp->m_sb.sb_unit;
  76                 geo->swidth = mp->m_sb.sb_width;
  77         }
  78         if (new_version >= 3) {
  79                 geo->version = XFS_FSOP_GEOM_VERSION;
  80                 geo->flags = XFS_FSOP_GEOM_FLAGS_NLINK |
  81                              XFS_FSOP_GEOM_FLAGS_DIRV2 |
  82                         (xfs_sb_version_hasattr(&mp->m_sb) ?
  83                                 XFS_FSOP_GEOM_FLAGS_ATTR : 0) |
  84                         (xfs_sb_version_hasquota(&mp->m_sb) ?
  85                                 XFS_FSOP_GEOM_FLAGS_QUOTA : 0) |
  86                         (xfs_sb_version_hasalign(&mp->m_sb) ?
  87                                 XFS_FSOP_GEOM_FLAGS_IALIGN : 0) |
  88                         (xfs_sb_version_hasdalign(&mp->m_sb) ?
  89                                 XFS_FSOP_GEOM_FLAGS_DALIGN : 0) |
  90                         (xfs_sb_version_hasextflgbit(&mp->m_sb) ?
  91                                 XFS_FSOP_GEOM_FLAGS_EXTFLG : 0) |
  92                         (xfs_sb_version_hassector(&mp->m_sb) ?
  93                                 XFS_FSOP_GEOM_FLAGS_SECTOR : 0) |
  94                         (xfs_sb_version_hasasciici(&mp->m_sb) ?
  95                                 XFS_FSOP_GEOM_FLAGS_DIRV2CI : 0) |
  96                         (xfs_sb_version_haslazysbcount(&mp->m_sb) ?
  97                                 XFS_FSOP_GEOM_FLAGS_LAZYSB : 0) |
  98                         (xfs_sb_version_hasattr2(&mp->m_sb) ?
  99                                 XFS_FSOP_GEOM_FLAGS_ATTR2 : 0) |
 100                         (xfs_sb_version_hasprojid32bit(&mp->m_sb) ?
 101                                 XFS_FSOP_GEOM_FLAGS_PROJID32 : 0) |
 102                         (xfs_sb_version_hascrc(&mp->m_sb) ?
 103                                 XFS_FSOP_GEOM_FLAGS_V5SB : 0) |
 104                         (xfs_sb_version_hasftype(&mp->m_sb) ?
 105                                 XFS_FSOP_GEOM_FLAGS_FTYPE : 0) |
 106                         (xfs_sb_version_hasfinobt(&mp->m_sb) ?
 107                                 XFS_FSOP_GEOM_FLAGS_FINOBT : 0) |
 108                         (xfs_sb_version_hassparseinodes(&mp->m_sb) ?
 109                                 XFS_FSOP_GEOM_FLAGS_SPINODES : 0) |
 110                         (xfs_sb_version_hasrmapbt(&mp->m_sb) ?
 111                                 XFS_FSOP_GEOM_FLAGS_RMAPBT : 0);
 112                 geo->logsectsize = xfs_sb_version_hassector(&mp->m_sb) ?
 113                                 mp->m_sb.sb_logsectsize : BBSIZE;
 114                 geo->rtsectsize = mp->m_sb.sb_blocksize;
 115                 geo->dirblocksize = mp->m_dir_geo->blksize;
 116         }
 117         if (new_version >= 4) {
 118                 geo->flags |=
 119                         (xfs_sb_version_haslogv2(&mp->m_sb) ?
 120                                 XFS_FSOP_GEOM_FLAGS_LOGV2 : 0);
 121                 geo->logsunit = mp->m_sb.sb_logsunit;
 122         }
 123         return 0;
 124 }
 125
 126 static struct xfs_buf *
 127 xfs_growfs_get_hdr_buf(
 128         struct xfs_mount        *mp,
 129         xfs_daddr_t             blkno,
 130         size_t                  numblks,
 131         int                     flags,
 132         const struct xfs_buf_ops *ops)
 133 {
 134         struct xfs_buf          *bp;
 135
 136         bp = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, flags);
 137         if (!bp)
 138                 return NULL;
 139
 140         xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
 141         bp->b_bn = blkno;
 142         bp->b_maps[0].bm_bn = blkno;
 143         bp->b_ops = ops;
 144
 145         return bp;
 146 }
 147
 148 static int
 149 xfs_growfs_data_private(
 150         xfs_mount_t             *mp,            /* mount point for filesystem */
 151         xfs_growfs_data_t       *in)            /* growfs data input struct */
 152 {
 153         xfs_agf_t               *agf;
 154         struct xfs_agfl         *agfl;
 155         xfs_agi_t               *agi;
 156         xfs_agnumber_t          agno;
 157         xfs_extlen_t            agsize;
 158         xfs_extlen_t            tmpsize;
 159         xfs_alloc_rec_t         *arec;
 160         xfs_buf_t               *bp;
 161         int                     bucket;
 162         int                     dpct;
 163         int                     error, saved_error = 0;
 164         xfs_agnumber_t          nagcount;
 165         xfs_agnumber_t          nagimax = 0;
 166         xfs_rfsblock_t          nb, nb_mod;
 167         xfs_rfsblock_t          new;
 168         xfs_rfsblock_t          nfree;
 169         xfs_agnumber_t          oagcount;
 170         int                     pct;
 171         xfs_trans_t             *tp;
 172
 173         nb = in->newblocks;
 174         pct = in->imaxpct;
 175         if (nb < mp->m_sb.sb_dblocks || pct < 0 || pct > 100)
 176                 return -EINVAL;
 177         if ((error = xfs_sb_validate_fsb_count(&mp->m_sb, nb)))
 178                 return error;
 179         dpct = pct - mp->m_sb.sb_imax_pct;
 180         error = xfs_buf_read_uncached(mp->m_ddev_targp,
 181                                 XFS_FSB_TO_BB(mp, nb) - XFS_FSS_TO_BB(mp, 1),
 182                                 XFS_FSS_TO_BB(mp, 1), 0, &bp, NULL);
 183         if (error)
 184                 return error;
 185         xfs_buf_relse(bp);
 186
 187         new = nb;       /* use new as a temporary here */
 188         nb_mod = do_div(new, mp->m_sb.sb_agblocks);
 189         nagcount = new + (nb_mod != 0);
 190         if (nb_mod && nb_mod < XFS_MIN_AG_BLOCKS) {
 191                 nagcount--;
 192                 nb = (xfs_rfsblock_t)nagcount * mp->m_sb.sb_agblocks;
 193                 if (nb < mp->m_sb.sb_dblocks)
 194                         return -EINVAL;
 195         }
 196         new = nb - mp->m_sb.sb_dblocks;
 197         oagcount = mp->m_sb.sb_agcount;
 198
 199         /* allocate the new per-ag structures */
 200         if (nagcount > oagcount) {
 201                 error = xfs_initialize_perag(mp, nagcount, &nagimax);
 202                 if (error)
 203                         return error;
 204         }
 205
 206         error = xfs_trans_alloc(mp, &M_RES(mp)->tr_growdata,
 207                         XFS_GROWFS_SPACE_RES(mp), 0, XFS_TRANS_RESERVE, &tp);
 208         if (error)
 209                 return error;
 210
 211         /*
 212          * Write new AG headers to disk. Non-transactional, but written
 213          * synchronously so they are completed prior to the growfs transaction
 214          * being logged.
 215          */
 216         nfree = 0;
 217         for (agno = nagcount - 1; agno >= oagcount; agno--, new -= agsize) {
 218                 __be32  *agfl_bno;
 219
 220                 /*
 221                  * AG freespace header block
 222                  */
 223                 bp = xfs_growfs_get_hdr_buf(mp,
 224                                 XFS_AG_DADDR(mp, agno, XFS_AGF_DADDR(mp)),
 225                                 XFS_FSS_TO_BB(mp, 1), 0,
 226                                 &xfs_agf_buf_ops);
 227                 if (!bp) {
 228                         error = -ENOMEM;
 229                         goto error0;
 230                 }
 231
 232                 agf = XFS_BUF_TO_AGF(bp);
 233                 agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
 234                 agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
 235                 agf->agf_seqno = cpu_to_be32(agno);
 236                 if (agno == nagcount - 1)
 237                         agsize =
 238                                 nb -
 239                                 (agno * (xfs_rfsblock_t)mp->m_sb.sb_agblocks);
 240                 else
 241                         agsize = mp->m_sb.sb_agblocks;
 242                 agf->agf_length = cpu_to_be32(agsize);
 243                 agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
 244                 agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
 245                 agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
 246                 agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
 247                 if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
 248                         agf->agf_roots[XFS_BTNUM_RMAPi] =
 249                                                 cpu_to_be32(XFS_RMAP_BLOCK(mp));
 250                         agf->agf_levels[XFS_BTNUM_RMAPi] = cpu_to_be32(1);
 251                 }
 252
 253                 agf->agf_flfirst = cpu_to_be32(1);
 254                 agf->agf_fllast = 0;
 255                 agf->agf_flcount = 0;
 256                 tmpsize = agsize - mp->m_ag_prealloc_blocks;
 257                 agf->agf_freeblks = cpu_to_be32(tmpsize);
 258                 agf->agf_longest = cpu_to_be32(tmpsize);
 259                 if (xfs_sb_version_hascrc(&mp->m_sb))
 260                         uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid);
 261
 262                 error = xfs_bwrite(bp);
 263                 xfs_buf_relse(bp);
 264                 if (error)
 265                         goto error0;
 266
 267                 /*
 268                  * AG freelist header block
 269                  */
 270                 bp = xfs_growfs_get_hdr_buf(mp,
 271                                 XFS_AG_DADDR(mp, agno, XFS_AGFL_DADDR(mp)),
 272                                 XFS_FSS_TO_BB(mp, 1), 0,
 273                                 &xfs_agfl_buf_ops);
 274                 if (!bp) {
 275                         error = -ENOMEM;
 276                         goto error0;
 277                 }
 278
 279                 agfl = XFS_BUF_TO_AGFL(bp);
 280                 if (xfs_sb_version_hascrc(&mp->m_sb)) {
 281                         agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
 282                         agfl->agfl_seqno = cpu_to_be32(agno);
 283                         uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid);
 284                 }
 285
 286                 agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, bp);
 287                 for (bucket = 0; bucket < XFS_AGFL_SIZE(mp); bucket++)
 288                         agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
 289
 290                 error = xfs_bwrite(bp);
 291                 xfs_buf_relse(bp);
 292                 if (error)
 293                         goto error0;
 294
 295                 /*
 296                  * AG inode header block
 297                  */
 298                 bp = xfs_growfs_get_hdr_buf(mp,
 299                                 XFS_AG_DADDR(mp, agno, XFS_AGI_DADDR(mp)),
 300                                 XFS_FSS_TO_BB(mp, 1), 0,
 301                                 &xfs_agi_buf_ops);
 302                 if (!bp) {
 303                         error = -ENOMEM;
 304                         goto error0;
 305                 }
 306
 307                 agi = XFS_BUF_TO_AGI(bp);
 308                 agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
 309                 agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
 310                 agi->agi_seqno = cpu_to_be32(agno);
 311                 agi->agi_length = cpu_to_be32(agsize);
 312                 agi->agi_count = 0;
 313                 agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
 314                 agi->agi_level = cpu_to_be32(1);
 315                 agi->agi_freecount = 0;
 316                 agi->agi_newino = cpu_to_be32(NULLAGINO);
 317                 agi->agi_dirino = cpu_to_be32(NULLAGINO);
 318                 if (xfs_sb_version_hascrc(&mp->m_sb))
 319                         uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid);
 320                 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
 321                         agi->agi_free_root = cpu_to_be32(XFS_FIBT_BLOCK(mp));
 322                         agi->agi_free_level = cpu_to_be32(1);
 323                 }
 324                 for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
 325                         agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
 326
 327                 error = xfs_bwrite(bp);
 328                 xfs_buf_relse(bp);
 329                 if (error)
 330                         goto error0;
 331
 332                 /*
 333                  * BNO btree root block
 334                  */
 335                 bp = xfs_growfs_get_hdr_buf(mp,
 336                                 XFS_AGB_TO_DADDR(mp, agno, XFS_BNO_BLOCK(mp)),
 337                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 338                                 &xfs_allocbt_buf_ops);
 339
 340                 if (!bp) {
 341                         error = -ENOMEM;
 342                         goto error0;
 343                 }
 344
 345                 if (xfs_sb_version_hascrc(&mp->m_sb))
 346                         xfs_btree_init_block(mp, bp, XFS_ABTB_CRC_MAGIC, 0, 1,
 347                                                 agno, XFS_BTREE_CRC_BLOCKS);
 348                 else
 349                         xfs_btree_init_block(mp, bp, XFS_ABTB_MAGIC, 0, 1,
 350                                                 agno, 0);
 351
 352                 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
 353                 arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
 354                 arec->ar_blockcount = cpu_to_be32(
 355                         agsize - be32_to_cpu(arec->ar_startblock));
 356
 357                 error = xfs_bwrite(bp);
 358                 xfs_buf_relse(bp);
 359                 if (error)
 360                         goto error0;
 361
 362                 /*
 363                  * CNT btree root block
 364                  */
 365                 bp = xfs_growfs_get_hdr_buf(mp,
 366                                 XFS_AGB_TO_DADDR(mp, agno, XFS_CNT_BLOCK(mp)),
 367                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 368                                 &xfs_allocbt_buf_ops);
 369                 if (!bp) {
 370                         error = -ENOMEM;
 371                         goto error0;
 372                 }
 373
 374                 if (xfs_sb_version_hascrc(&mp->m_sb))
 375                         xfs_btree_init_block(mp, bp, XFS_ABTC_CRC_MAGIC, 0, 1,
 376                                                 agno, XFS_BTREE_CRC_BLOCKS);
 377                 else
 378                         xfs_btree_init_block(mp, bp, XFS_ABTC_MAGIC, 0, 1,
 379                                                 agno, 0);
 380
 381                 arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
 382                 arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
 383                 arec->ar_blockcount = cpu_to_be32(
 384                         agsize - be32_to_cpu(arec->ar_startblock));
 385                 nfree += be32_to_cpu(arec->ar_blockcount);
 386
 387                 error = xfs_bwrite(bp);
 388                 xfs_buf_relse(bp);
 389                 if (error)
 390                         goto error0;
 391
 392                 /* RMAP btree root block */
 393                 if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
 394                         struct xfs_rmap_rec     *rrec;
 395                         struct xfs_btree_block  *block;
 396
 397                         bp = xfs_growfs_get_hdr_buf(mp,
 398                                 XFS_AGB_TO_DADDR(mp, agno, XFS_RMAP_BLOCK(mp)),
 399                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 400                                 &xfs_rmapbt_buf_ops);
 401                         if (!bp) {
 402                                 error = -ENOMEM;
 403                                 goto error0;
 404                         }
 405
 406                         xfs_btree_init_block(mp, bp, XFS_RMAP_CRC_MAGIC, 0, 0,
 407                                                 agno, XFS_BTREE_CRC_BLOCKS);
 408                         block = XFS_BUF_TO_BLOCK(bp);
 409
 410
 411                         /*
 412                          * mark the AG header regions as static metadata The BNO
 413                          * btree block is the first block after the headers, so
 414                          * it's location defines the size of region the static
 415                          * metadata consumes.
 416                          *
 417                          * Note: unlike mkfs, we never have to account for log
 418                          * space when growing the data regions
 419                          */
 420                         rrec = XFS_RMAP_REC_ADDR(block, 1);
 421                         rrec->rm_startblock = 0;
 422                         rrec->rm_blockcount = cpu_to_be32(XFS_BNO_BLOCK(mp));
 423                         rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_FS);
 424                         rrec->rm_offset = 0;
 425                         be16_add_cpu(&block->bb_numrecs, 1);
 426
 427                         /* account freespace btree root blocks */
 428                         rrec = XFS_RMAP_REC_ADDR(block, 2);
 429                         rrec->rm_startblock = cpu_to_be32(XFS_BNO_BLOCK(mp));
 430                         rrec->rm_blockcount = cpu_to_be32(2);
 431                         rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
 432                         rrec->rm_offset = 0;
 433                         be16_add_cpu(&block->bb_numrecs, 1);
 434
 435                         /* account inode btree root blocks */
 436                         rrec = XFS_RMAP_REC_ADDR(block, 3);
 437                         rrec->rm_startblock = cpu_to_be32(XFS_IBT_BLOCK(mp));
 438                         rrec->rm_blockcount = cpu_to_be32(XFS_RMAP_BLOCK(mp) -
 439                                                         XFS_IBT_BLOCK(mp));
 440                         rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_INOBT);
 441                         rrec->rm_offset = 0;
 442                         be16_add_cpu(&block->bb_numrecs, 1);
 443
 444                         /* account for rmap btree root */
 445                         rrec = XFS_RMAP_REC_ADDR(block, 4);
 446                         rrec->rm_startblock = cpu_to_be32(XFS_RMAP_BLOCK(mp));
 447                         rrec->rm_blockcount = cpu_to_be32(1);
 448                         rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
 449                         rrec->rm_offset = 0;
 450                         be16_add_cpu(&block->bb_numrecs, 1);
 451
 452                         error = xfs_bwrite(bp);
 453                         xfs_buf_relse(bp);
 454                         if (error)
 455                                 goto error0;
 456                 }
 457
 458                 /*
 459                  * INO btree root block
 460                  */
 461                 bp = xfs_growfs_get_hdr_buf(mp,
 462                                 XFS_AGB_TO_DADDR(mp, agno, XFS_IBT_BLOCK(mp)),
 463                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 464                                 &xfs_inobt_buf_ops);
 465                 if (!bp) {
 466                         error = -ENOMEM;
 467                         goto error0;
 468                 }
 469
 470                 if (xfs_sb_version_hascrc(&mp->m_sb))
 471                         xfs_btree_init_block(mp, bp, XFS_IBT_CRC_MAGIC, 0, 0,
 472                                                 agno, XFS_BTREE_CRC_BLOCKS);
 473                 else
 474                         xfs_btree_init_block(mp, bp, XFS_IBT_MAGIC, 0, 0,
 475                                                 agno, 0);
 476
 477                 error = xfs_bwrite(bp);
 478                 xfs_buf_relse(bp);
 479                 if (error)
 480                         goto error0;
 481
 482                 /*
 483                  * FINO btree root block
 484                  */
 485                 if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
 486                         bp = xfs_growfs_get_hdr_buf(mp,
 487                                 XFS_AGB_TO_DADDR(mp, agno, XFS_FIBT_BLOCK(mp)),
 488                                 BTOBB(mp->m_sb.sb_blocksize), 0,
 489                                 &xfs_inobt_buf_ops);
 490                         if (!bp) {
 491                                 error = -ENOMEM;
 492                                 goto error0;
 493                         }
 494
 495                         if (xfs_sb_version_hascrc(&mp->m_sb))
 496                                 xfs_btree_init_block(mp, bp, XFS_FIBT_CRC_MAGIC,
 497                                                      0, 0, agno,
 498                                                      XFS_BTREE_CRC_BLOCKS);
 499                         else
 500                                 xfs_btree_init_block(mp, bp, XFS_FIBT_MAGIC, 0,
 501                                                      0, agno, 0);
 502
 503                         error = xfs_bwrite(bp);
 504                         xfs_buf_relse(bp);
 505                         if (error)
 506                                 goto error0;
 507                 }
 508
 509         }
 510         xfs_trans_agblocks_delta(tp, nfree);
 511         /*
 512          * There are new blocks in the old last a.g.
 513          */
 514         if (new) {
 515                 struct xfs_owner_info   oinfo;
 516
 517                 /*
 518                  * Change the agi length.
 519                  */
 520                 error = xfs_ialloc_read_agi(mp, tp, agno, &bp);
 521                 if (error) {
 522                         goto error0;
 523                 }
 524                 ASSERT(bp);
 525                 agi = XFS_BUF_TO_AGI(bp);
 526                 be32_add_cpu(&agi->agi_length, new);
 527                 ASSERT(nagcount == oagcount ||
 528                        be32_to_cpu(agi->agi_length) == mp->m_sb.sb_agblocks);
 529                 xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
 530                 /*
 531                  * Change agf length.
 532                  */
 533                 error = xfs_alloc_read_agf(mp, tp, agno, 0, &bp);
 534                 if (error) {
 535                         goto error0;
 536                 }
 537                 ASSERT(bp);
 538                 agf = XFS_BUF_TO_AGF(bp);
 539                 be32_add_cpu(&agf->agf_length, new);
 540                 ASSERT(be32_to_cpu(agf->agf_length) ==
 541                        be32_to_cpu(agi->agi_length));
 542
 543                 xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
 544
 545                 /*
 546                  * Free the new space.
 547                  *
 548                  * XFS_RMAP_OWN_NULL is used here to tell the rmap btree that
 549                  * this doesn't actually exist in the rmap btree.
 550                  */
 551                 xfs_rmap_ag_owner(&oinfo, XFS_RMAP_OWN_NULL);
 552                 error = xfs_free_extent(tp,
 553                                 XFS_AGB_TO_FSB(mp, agno,
 554                                         be32_to_cpu(agf->agf_length) - new),
 555                                 new, &oinfo);
 556                 if (error)
 557                         goto error0;
 558         }
 559
 560         /*
 561          * Update changed superblock fields transactionally. These are not
 562          * seen by the rest of the world until the transaction commit applies
 563          * them atomically to the superblock.
 564          */
 565         if (nagcount > oagcount)
 566                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_AGCOUNT, nagcount - oagcount);
 567         if (nb > mp->m_sb.sb_dblocks)
 568                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_DBLOCKS,
 569                                  nb - mp->m_sb.sb_dblocks);
 570         if (nfree)
 571                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_FDBLOCKS, nfree);
 572         if (dpct)
 573                 xfs_trans_mod_sb(tp, XFS_TRANS_SB_IMAXPCT, dpct);
 574         xfs_trans_set_sync(tp);
 575         error = xfs_trans_commit(tp);
 576         if (error)
 577                 return error;
 578
 579         /* New allocation groups fully initialized, so update mount struct */
 580         if (nagimax)
 581                 mp->m_maxagi = nagimax;
 582         if (mp->m_sb.sb_imax_pct) {
 583                 __uint64_t icount = mp->m_sb.sb_dblocks * mp->m_sb.sb_imax_pct;
 584                 do_div(icount, 100);
 585                 mp->m_maxicount = icount << mp->m_sb.sb_inopblog;
 586         } else
 587                 mp->m_maxicount = 0;
 588         xfs_set_low_space_thresholds(mp);
 589         mp->m_alloc_set_aside = xfs_alloc_set_aside(mp);
 590
 591         /* update secondary superblocks. */
 592         for (agno = 1; agno < nagcount; agno++) {
 593                 error = 0;
 594                 /*
 595                  * new secondary superblocks need to be zeroed, not read from
 596                  * disk as the contents of the new area we are growing into is
 597                  * completely unknown.
 598                  */
 599                 if (agno < oagcount) {
 600                         error = xfs_trans_read_buf(mp, NULL, mp->m_ddev_targp,
 601                                   XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
 602                                   XFS_FSS_TO_BB(mp, 1), 0, &bp,
 603                                   &xfs_sb_buf_ops);
 604                 } else {
 605                         bp = xfs_trans_get_buf(NULL, mp->m_ddev_targp,
 606                                   XFS_AGB_TO_DADDR(mp, agno, XFS_SB_BLOCK(mp)),
 607                                   XFS_FSS_TO_BB(mp, 1), 0);
 608                         if (bp) {
 609                                 bp->b_ops = &xfs_sb_buf_ops;
 610                                 xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
 611                         } else
 612                                 error = -ENOMEM;
 613                 }
 614
 615                 /*
 616                  * If we get an error reading or writing alternate superblocks,
 617                  * continue.  xfs_repair chooses the "best" superblock based
 618                  * on most matches; if we break early, we'll leave more
 619                  * superblocks un-updated than updated, and xfs_repair may
 620                  * pick them over the properly-updated primary.
 621                  */
 622                 if (error) {
 623                         xfs_warn(mp,
 624                 "error %d reading secondary superblock for ag %d",
 625                                 error, agno);
 626                         saved_error = error;
 627                         continue;
 628                 }
 629                 xfs_sb_to_disk(XFS_BUF_TO_SBP(bp), &mp->m_sb);
 630
 631                 error = xfs_bwrite(bp);
 632                 xfs_buf_relse(bp);
 633                 if (error) {
 634                         xfs_warn(mp,
 635                 "write error %d updating secondary superblock for ag %d",
 636                                 error, agno);
 637                         saved_error = error;
 638                         continue;
 639                 }
 640         }
 641         return saved_error ? saved_error : error;
 642
 643  error0:
 644         xfs_trans_cancel(tp);
 645         return error;
 646 }
 647
 648 static int
 649 xfs_growfs_log_private(
 650         xfs_mount_t             *mp,    /* mount point for filesystem */
 651         xfs_growfs_log_t        *in)    /* growfs log input struct */
 652 {
 653         xfs_extlen_t            nb;
 654
 655         nb = in->newblocks;
 656         if (nb < XFS_MIN_LOG_BLOCKS || nb < XFS_B_TO_FSB(mp, XFS_MIN_LOG_BYTES))
 657                 return -EINVAL;
 658         if (nb == mp->m_sb.sb_logblocks &&
 659             in->isint == (mp->m_sb.sb_logstart != 0))
 660                 return -EINVAL;
 661         /*
 662          * Moving the log is hard, need new interfaces to sync
 663          * the log first, hold off all activity while moving it.
 664          * Can have shorter or longer log in the same space,
 665          * or transform internal to external log or vice versa.
 666          */
 667         return -ENOSYS;
 668 }
 669
 670 /*
 671  * protected versions of growfs function acquire and release locks on the mount
 672  * point - exported through ioctls: XFS_IOC_FSGROWFSDATA, XFS_IOC_FSGROWFSLOG,
 673  * XFS_IOC_FSGROWFSRT
 674  */
 675
 676
 677 int
 678 xfs_growfs_data(
 679         xfs_mount_t             *mp,
 680         xfs_growfs_data_t       *in)
 681 {
 682         int error;
 683
 684         if (!capable(CAP_SYS_ADMIN))
 685                 return -EPERM;
 686         if (!mutex_trylock(&mp->m_growlock))
 687                 return -EWOULDBLOCK;
 688         error = xfs_growfs_data_private(mp, in);
 689         /*
 690          * Increment the generation unconditionally, the error could be from
 691          * updating the secondary superblocks, in which case the new size
 692          * is live already.
 693          */
 694         mp->m_generation++;
 695         mutex_unlock(&mp->m_growlock);
 696         return error;
 697 }
 698
 699 int
 700 xfs_growfs_log(
 701         xfs_mount_t             *mp,
 702         xfs_growfs_log_t        *in)
 703 {
 704         int error;
 705
 706         if (!capable(CAP_SYS_ADMIN))
 707                 return -EPERM;
 708         if (!mutex_trylock(&mp->m_growlock))
 709                 return -EWOULDBLOCK;
 710         error = xfs_growfs_log_private(mp, in);
 711         mutex_unlock(&mp->m_growlock);
 712         return error;
 713 }
 714
 715 /*
 716  * exported through ioctl XFS_IOC_FSCOUNTS
 717  */
 718
 719 int
 720 xfs_fs_counts(
 721         xfs_mount_t             *mp,
 722         xfs_fsop_counts_t       *cnt)
 723 {
 724         cnt->allocino = percpu_counter_read_positive(&mp->m_icount);
 725         cnt->freeino = percpu_counter_read_positive(&mp->m_ifree);
 726         cnt->freedata = percpu_counter_read_positive(&mp->m_fdblocks) -
 727                                                 mp->m_alloc_set_aside;
 728
 729         spin_lock(&mp->m_sb_lock);
 730         cnt->freertx = mp->m_sb.sb_frextents;
 731         spin_unlock(&mp->m_sb_lock);
 732         return 0;
 733 }
 734
 735 /*
 736  * exported through ioctl XFS_IOC_SET_RESBLKS & XFS_IOC_GET_RESBLKS
 737  *
 738  * xfs_reserve_blocks is called to set m_resblks
 739  * in the in-core mount table. The number of unused reserved blocks
 740  * is kept in m_resblks_avail.
 741  *
 742  * Reserve the requested number of blocks if available. Otherwise return
 743  * as many as possible to satisfy the request. The actual number
 744  * reserved are returned in outval
 745  *
 746  * A null inval pointer indicates that only the current reserved blocks
 747  * available  should  be returned no settings are changed.
 748  */
 749
 750 int
 751 xfs_reserve_blocks(
 752         xfs_mount_t             *mp,
 753         __uint64_t              *inval,
 754         xfs_fsop_resblks_t      *outval)
 755 {
 756         __int64_t               lcounter, delta;
 757         __int64_t               fdblks_delta = 0;
 758         __uint64_t              request;
 759         __int64_t               free;
 760         int                     error = 0;
 761
 762         /* If inval is null, report current values and return */
 763         if (inval == (__uint64_t *)NULL) {
 764                 if (!outval)
 765                         return -EINVAL;
 766                 outval->resblks = mp->m_resblks;
 767                 outval->resblks_avail = mp->m_resblks_avail;
 768                 return 0;
 769         }
 770
 771         request = *inval;
 772
 773         /*
 774          * With per-cpu counters, this becomes an interesting problem. we need
 775          * to work out if we are freeing or allocation blocks first, then we can
 776          * do the modification as necessary.
 777          *
 778          * We do this under the m_sb_lock so that if we are near ENOSPC, we will
 779          * hold out any changes while we work out what to do. This means that
 780          * the amount of free space can change while we do this, so we need to
 781          * retry if we end up trying to reserve more space than is available.
 782          */
 783         spin_lock(&mp->m_sb_lock);
 784
 785         /*
 786          * If our previous reservation was larger than the current value,
 787          * then move any unused blocks back to the free pool. Modify the resblks
 788          * counters directly since we shouldn't have any problems unreserving
 789          * space.
 790          */
 791         if (mp->m_resblks > request) {
 792                 lcounter = mp->m_resblks_avail - request;
 793                 if (lcounter  > 0) {            /* release unused blocks */
 794                         fdblks_delta = lcounter;
 795                         mp->m_resblks_avail -= lcounter;
 796                 }
 797                 mp->m_resblks = request;
 798                 if (fdblks_delta) {
 799                         spin_unlock(&mp->m_sb_lock);
 800                         error = xfs_mod_fdblocks(mp, fdblks_delta, 0);
 801                         spin_lock(&mp->m_sb_lock);
 802                 }
 803
 804                 goto out;
 805         }
 806
 807         /*
 808          * If the request is larger than the current reservation, reserve the
 809          * blocks before we update the reserve counters. Sample m_fdblocks and
 810          * perform a partial reservation if the request exceeds free space.
 811          */
 812         error = -ENOSPC;
 813         do {
 814                 free = percpu_counter_sum(&mp->m_fdblocks) -
 815                                                 mp->m_alloc_set_aside;
 816                 if (!free)
 817                         break;
 818
 819                 delta = request - mp->m_resblks;
 820                 lcounter = free - delta;
 821                 if (lcounter < 0)
 822                         /* We can't satisfy the request, just get what we can */
 823                         fdblks_delta = free;
 824                 else
 825                         fdblks_delta = delta;
 826
 827                 /*
 828                  * We'll either succeed in getting space from the free block
 829                  * count or we'll get an ENOSPC. If we get a ENOSPC, it means
 830                  * things changed while we were calculating fdblks_delta and so
 831                  * we should try again to see if there is anything left to
 832                  * reserve.
 833                  *
 834                  * Don't set the reserved flag here - we don't want to reserve
 835                  * the extra reserve blocks from the reserve.....
 836                  */
 837                 spin_unlock(&mp->m_sb_lock);
 838                 error = xfs_mod_fdblocks(mp, -fdblks_delta, 0);
 839                 spin_lock(&mp->m_sb_lock);
 840         } while (error == -ENOSPC);
 841
 842         /*
 843          * Update the reserve counters if blocks have been successfully
 844          * allocated.
 845          */
 846         if (!error && fdblks_delta) {
 847                 mp->m_resblks += fdblks_delta;
 848                 mp->m_resblks_avail += fdblks_delta;
 849         }
 850
 851 out:
 852         if (outval) {
 853                 outval->resblks = mp->m_resblks;
 854                 outval->resblks_avail = mp->m_resblks_avail;
 855         }
 856
 857         spin_unlock(&mp->m_sb_lock);
 858         return error;
 859 }
 860
 861 int
 862 xfs_fs_goingdown(
 863         xfs_mount_t     *mp,
 864         __uint32_t      inflags)
 865 {
 866         switch (inflags) {
 867         case XFS_FSOP_GOING_FLAGS_DEFAULT: {
 868                 struct super_block *sb = freeze_bdev(mp->m_super->s_bdev);
 869
 870                 if (sb && !IS_ERR(sb)) {
 871                         xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
 872                         thaw_bdev(sb->s_bdev, sb);
 873                 }
 874
 875                 break;
 876         }
 877         case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
 878                 xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
 879                 break;
 880         case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
 881                 xfs_force_shutdown(mp,
 882                                 SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
 883                 break;
 884         default:
 885                 return -EINVAL;
 886         }
 887
 888         return 0;
 889 }
 890
 891 /*
 892  * Force a shutdown of the filesystem instantly while keeping the filesystem
 893  * consistent. We don't do an unmount here; just shutdown the shop, make sure
 894  * that absolutely nothing persistent happens to this filesystem after this
 895  * point.
 896  */
 897 void
 898 xfs_do_force_shutdown(
 899         xfs_mount_t     *mp,
 900         int             flags,
 901         char            *fname,
 902         int             lnnum)
 903 {
 904         int             logerror;
 905
 906         logerror = flags & SHUTDOWN_LOG_IO_ERROR;
 907
 908         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 909                 xfs_notice(mp,
 910         "%s(0x%x) called from line %d of file %s.  Return address = 0x%p",
 911                         __func__, flags, lnnum, fname, __return_address);
 912         }
 913         /*
 914          * No need to duplicate efforts.
 915          */
 916         if (XFS_FORCED_SHUTDOWN(mp) && !logerror)
 917                 return;
 918
 919         /*
 920          * This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't
 921          * queue up anybody new on the log reservations, and wakes up
 922          * everybody who's sleeping on log reservations to tell them
 923          * the bad news.
 924          */
 925         if (xfs_log_force_umount(mp, logerror))
 926                 return;
 927
 928         if (flags & SHUTDOWN_CORRUPT_INCORE) {
 929                 xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_CORRUPT,
 930     "Corruption of in-memory data detected.  Shutting down filesystem");
 931                 if (XFS_ERRLEVEL_HIGH <= xfs_error_level)
 932                         xfs_stack_trace();
 933         } else if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 934                 if (logerror) {
 935                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_LOGERROR,
 936                 "Log I/O Error Detected.  Shutting down filesystem");
 937                 } else if (flags & SHUTDOWN_DEVICE_REQ) {
 938                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
 939                 "All device paths lost.  Shutting down filesystem");
 940                 } else if (!(flags & SHUTDOWN_REMOTE_REQ)) {
 941                         xfs_alert_tag(mp, XFS_PTAG_SHUTDOWN_IOERROR,
 942                 "I/O Error Detected. Shutting down filesystem");
 943                 }
 944         }
 945         if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 946                 xfs_alert(mp,
 947         "Please umount the filesystem and rectify the problem(s)");
 948         }
 949 }