libxfs/xfs_ag.c

   1 /* SPDX-License-Identifier: GPL-2.0 */
   2 /*
   3  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
   4  * Copyright (c) 2018 Red Hat, Inc.
   5  * All rights reserved.
   6  */
   7
   8 #include "libxfs_priv.h"
   9 #include "xfs.h"
  10 #include "xfs_fs.h"
  11 #include "xfs_shared.h"
  12 #include "xfs_format.h"
  13 #include "xfs_trans_resv.h"
  14 #include "xfs_sb.h"
  15 #include "xfs_mount.h"
  16 #include "xfs_btree.h"
  17 #include "xfs_alloc_btree.h"
  18 #include "xfs_rmap_btree.h"
  19 #include "xfs_alloc.h"
  20 #include "xfs_ialloc.h"
  21 #include "xfs_rmap.h"
  22 #include "xfs_ag.h"
  23 #include "xfs_ag_resv.h"
  24 #include "xfs_health.h"
  25
  26 static struct xfs_buf *
  27 xfs_get_aghdr_buf(
  28         struct xfs_mount        *mp,
  29         xfs_daddr_t             blkno,
  30         size_t                  numblks,
  31         int                     flags,
  32         const struct xfs_buf_ops *ops)
  33 {
  34         struct xfs_buf          *bp;
  35
  36         bp = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, flags);
  37         if (!bp)
  38                 return NULL;
  39
  40         xfs_buf_zero(bp, 0, BBTOB(bp->b_length));
  41         bp->b_bn = blkno;
  42         bp->b_maps[0].bm_bn = blkno;
  43         bp->b_ops = ops;
  44
  45         return bp;
  46 }
  47
  48 /*
  49  * Generic btree root block init function
  50  */
  51 static void
  52 xfs_btroot_init(
  53         struct xfs_mount        *mp,
  54         struct xfs_buf          *bp,
  55         struct aghdr_init_data  *id)
  56 {
  57         xfs_btree_init_block(mp, bp, id->type, 0, 0, id->agno, 0);
  58 }
  59
  60 /*
  61  * Alloc btree root block init functions
  62  */
  63 static void
  64 xfs_bnoroot_init(
  65         struct xfs_mount        *mp,
  66         struct xfs_buf          *bp,
  67         struct aghdr_init_data  *id)
  68 {
  69         struct xfs_alloc_rec    *arec;
  70
  71         xfs_btree_init_block(mp, bp, XFS_BTNUM_BNO, 0, 1, id->agno, 0);
  72         arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
  73         arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
  74         arec->ar_blockcount = cpu_to_be32(id->agsize -
  75                                           be32_to_cpu(arec->ar_startblock));
  76 }
  77
  78 static void
  79 xfs_cntroot_init(
  80         struct xfs_mount        *mp,
  81         struct xfs_buf          *bp,
  82         struct aghdr_init_data  *id)
  83 {
  84         struct xfs_alloc_rec    *arec;
  85
  86         xfs_btree_init_block(mp, bp, XFS_BTNUM_CNT, 0, 1, id->agno, 0);
  87         arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
  88         arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
  89         arec->ar_blockcount = cpu_to_be32(id->agsize -
  90                                           be32_to_cpu(arec->ar_startblock));
  91 }
  92
  93 /*
  94  * Reverse map root block init
  95  */
  96 static void
  97 xfs_rmaproot_init(
  98         struct xfs_mount        *mp,
  99         struct xfs_buf          *bp,
 100         struct aghdr_init_data  *id)
 101 {
 102         struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
 103         struct xfs_rmap_rec     *rrec;
 104
 105         xfs_btree_init_block(mp, bp, XFS_BTNUM_RMAP, 0, 4, id->agno, 0);
 106
 107         /*
 108          * mark the AG header regions as static metadata The BNO
 109          * btree block is the first block after the headers, so
 110          * it's location defines the size of region the static
 111          * metadata consumes.
 112          *
 113          * Note: unlike mkfs, we never have to account for log
 114          * space when growing the data regions
 115          */
 116         rrec = XFS_RMAP_REC_ADDR(block, 1);
 117         rrec->rm_startblock = 0;
 118         rrec->rm_blockcount = cpu_to_be32(XFS_BNO_BLOCK(mp));
 119         rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_FS);
 120         rrec->rm_offset = 0;
 121
 122         /* account freespace btree root blocks */
 123         rrec = XFS_RMAP_REC_ADDR(block, 2);
 124         rrec->rm_startblock = cpu_to_be32(XFS_BNO_BLOCK(mp));
 125         rrec->rm_blockcount = cpu_to_be32(2);
 126         rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
 127         rrec->rm_offset = 0;
 128
 129         /* account inode btree root blocks */
 130         rrec = XFS_RMAP_REC_ADDR(block, 3);
 131         rrec->rm_startblock = cpu_to_be32(XFS_IBT_BLOCK(mp));
 132         rrec->rm_blockcount = cpu_to_be32(XFS_RMAP_BLOCK(mp) -
 133                                           XFS_IBT_BLOCK(mp));
 134         rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_INOBT);
 135         rrec->rm_offset = 0;
 136
 137         /* account for rmap btree root */
 138         rrec = XFS_RMAP_REC_ADDR(block, 4);
 139         rrec->rm_startblock = cpu_to_be32(XFS_RMAP_BLOCK(mp));
 140         rrec->rm_blockcount = cpu_to_be32(1);
 141         rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
 142         rrec->rm_offset = 0;
 143
 144         /* account for refc btree root */
 145         if (xfs_sb_version_hasreflink(&mp->m_sb)) {
 146                 rrec = XFS_RMAP_REC_ADDR(block, 5);
 147                 rrec->rm_startblock = cpu_to_be32(xfs_refc_block(mp));
 148                 rrec->rm_blockcount = cpu_to_be32(1);
 149                 rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_REFC);
 150                 rrec->rm_offset = 0;
 151                 be16_add_cpu(&block->bb_numrecs, 1);
 152         }
 153 }
 154
 155 /*
 156  * Initialise new secondary superblocks with the pre-grow geometry, but mark
 157  * them as "in progress" so we know they haven't yet been activated. This will
 158  * get cleared when the update with the new geometry information is done after
 159  * changes to the primary are committed. This isn't strictly necessary, but we
 160  * get it for free with the delayed buffer write lists and it means we can tell
 161  * if a grow operation didn't complete properly after the fact.
 162  */
 163 static void
 164 xfs_sbblock_init(
 165         struct xfs_mount        *mp,
 166         struct xfs_buf          *bp,
 167         struct aghdr_init_data  *id)
 168 {
 169         struct xfs_dsb          *dsb = XFS_BUF_TO_SBP(bp);
 170
 171         xfs_sb_to_disk(dsb, &mp->m_sb);
 172         dsb->sb_inprogress = 1;
 173 }
 174
 175 static void
 176 xfs_agfblock_init(
 177         struct xfs_mount        *mp,
 178         struct xfs_buf          *bp,
 179         struct aghdr_init_data  *id)
 180 {
 181         struct xfs_agf          *agf = XFS_BUF_TO_AGF(bp);
 182         xfs_extlen_t            tmpsize;
 183
 184         agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
 185         agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
 186         agf->agf_seqno = cpu_to_be32(id->agno);
 187         agf->agf_length = cpu_to_be32(id->agsize);
 188         agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
 189         agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
 190         agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
 191         agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
 192         if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
 193                 agf->agf_roots[XFS_BTNUM_RMAPi] =
 194                                         cpu_to_be32(XFS_RMAP_BLOCK(mp));
 195                 agf->agf_levels[XFS_BTNUM_RMAPi] = cpu_to_be32(1);
 196                 agf->agf_rmap_blocks = cpu_to_be32(1);
 197         }
 198
 199         agf->agf_flfirst = cpu_to_be32(1);
 200         agf->agf_fllast = 0;
 201         agf->agf_flcount = 0;
 202         tmpsize = id->agsize - mp->m_ag_prealloc_blocks;
 203         agf->agf_freeblks = cpu_to_be32(tmpsize);
 204         agf->agf_longest = cpu_to_be32(tmpsize);
 205         if (xfs_sb_version_hascrc(&mp->m_sb))
 206                 uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid);
 207         if (xfs_sb_version_hasreflink(&mp->m_sb)) {
 208                 agf->agf_refcount_root = cpu_to_be32(
 209                                 xfs_refc_block(mp));
 210                 agf->agf_refcount_level = cpu_to_be32(1);
 211                 agf->agf_refcount_blocks = cpu_to_be32(1);
 212         }
 213 }
 214
 215 static void
 216 xfs_agflblock_init(
 217         struct xfs_mount        *mp,
 218         struct xfs_buf          *bp,
 219         struct aghdr_init_data  *id)
 220 {
 221         struct xfs_agfl         *agfl = XFS_BUF_TO_AGFL(bp);
 222         __be32                  *agfl_bno;
 223         int                     bucket;
 224
 225         if (xfs_sb_version_hascrc(&mp->m_sb)) {
 226                 agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
 227                 agfl->agfl_seqno = cpu_to_be32(id->agno);
 228                 uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid);
 229         }
 230
 231         agfl_bno = XFS_BUF_TO_AGFL_BNO(mp, bp);
 232         for (bucket = 0; bucket < xfs_agfl_size(mp); bucket++)
 233                 agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
 234 }
 235
 236 static void
 237 xfs_agiblock_init(
 238         struct xfs_mount        *mp,
 239         struct xfs_buf          *bp,
 240         struct aghdr_init_data  *id)
 241 {
 242         struct xfs_agi          *agi = XFS_BUF_TO_AGI(bp);
 243         int                     bucket;
 244
 245         agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
 246         agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
 247         agi->agi_seqno = cpu_to_be32(id->agno);
 248         agi->agi_length = cpu_to_be32(id->agsize);
 249         agi->agi_count = 0;
 250         agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
 251         agi->agi_level = cpu_to_be32(1);
 252         agi->agi_freecount = 0;
 253         agi->agi_newino = cpu_to_be32(NULLAGINO);
 254         agi->agi_dirino = cpu_to_be32(NULLAGINO);
 255         if (xfs_sb_version_hascrc(&mp->m_sb))
 256                 uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid);
 257         if (xfs_sb_version_hasfinobt(&mp->m_sb)) {
 258                 agi->agi_free_root = cpu_to_be32(XFS_FIBT_BLOCK(mp));
 259                 agi->agi_free_level = cpu_to_be32(1);
 260         }
 261         for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
 262                 agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
 263 }
 264
 265 typedef void (*aghdr_init_work_f)(struct xfs_mount *mp, struct xfs_buf *bp,
 266                                   struct aghdr_init_data *id);
 267 static int
 268 xfs_ag_init_hdr(
 269         struct xfs_mount        *mp,
 270         struct aghdr_init_data  *id,
 271         aghdr_init_work_f       work,
 272         const struct xfs_buf_ops *ops)
 273
 274 {
 275         struct xfs_buf          *bp;
 276
 277         bp = xfs_get_aghdr_buf(mp, id->daddr, id->numblks, 0, ops);
 278         if (!bp)
 279                 return -ENOMEM;
 280
 281         (*work)(mp, bp, id);
 282
 283         xfs_buf_delwri_queue(bp, &id->buffer_list);
 284         xfs_buf_relse(bp);
 285         return 0;
 286 }
 287
 288 struct xfs_aghdr_grow_data {
 289         xfs_daddr_t             daddr;
 290         size_t                  numblks;
 291         const struct xfs_buf_ops *ops;
 292         aghdr_init_work_f       work;
 293         xfs_btnum_t             type;
 294         bool                    need_init;
 295 };
 296
 297 /*
 298  * Prepare new AG headers to be written to disk. We use uncached buffers here,
 299  * as it is assumed these new AG headers are currently beyond the currently
 300  * valid filesystem address space. Using cached buffers would trip over EOFS
 301  * corruption detection alogrithms in the buffer cache lookup routines.
 302  *
 303  * This is a non-transactional function, but the prepared buffers are added to a
 304  * delayed write buffer list supplied by the caller so they can submit them to
 305  * disk and wait on them as required.
 306  */
 307 int
 308 xfs_ag_init_headers(
 309         struct xfs_mount        *mp,
 310         struct aghdr_init_data  *id)
 311
 312 {
 313         struct xfs_aghdr_grow_data aghdr_data[] = {
 314         { /* SB */
 315                 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_SB_DADDR),
 316                 .numblks = XFS_FSS_TO_BB(mp, 1),
 317                 .ops = &xfs_sb_buf_ops,
 318                 .work = &xfs_sbblock_init,
 319                 .need_init = true
 320         },
 321         { /* AGF */
 322                 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGF_DADDR(mp)),
 323                 .numblks = XFS_FSS_TO_BB(mp, 1),
 324                 .ops = &xfs_agf_buf_ops,
 325                 .work = &xfs_agfblock_init,
 326                 .need_init = true
 327         },
 328         { /* AGFL */
 329                 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGFL_DADDR(mp)),
 330                 .numblks = XFS_FSS_TO_BB(mp, 1),
 331                 .ops = &xfs_agfl_buf_ops,
 332                 .work = &xfs_agflblock_init,
 333                 .need_init = true
 334         },
 335         { /* AGI */
 336                 .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGI_DADDR(mp)),
 337                 .numblks = XFS_FSS_TO_BB(mp, 1),
 338                 .ops = &xfs_agi_buf_ops,
 339                 .work = &xfs_agiblock_init,
 340                 .need_init = true
 341         },
 342         { /* BNO root block */
 343                 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_BNO_BLOCK(mp)),
 344                 .numblks = BTOBB(mp->m_sb.sb_blocksize),
 345                 .ops = &xfs_bnobt_buf_ops,
 346                 .work = &xfs_bnoroot_init,
 347                 .need_init = true
 348         },
 349         { /* CNT root block */
 350                 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_CNT_BLOCK(mp)),
 351                 .numblks = BTOBB(mp->m_sb.sb_blocksize),
 352                 .ops = &xfs_cntbt_buf_ops,
 353                 .work = &xfs_cntroot_init,
 354                 .need_init = true
 355         },
 356         { /* INO root block */
 357                 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_IBT_BLOCK(mp)),
 358                 .numblks = BTOBB(mp->m_sb.sb_blocksize),
 359                 .ops = &xfs_inobt_buf_ops,
 360                 .work = &xfs_btroot_init,
 361                 .type = XFS_BTNUM_INO,
 362                 .need_init = true
 363         },
 364         { /* FINO root block */
 365                 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_FIBT_BLOCK(mp)),
 366                 .numblks = BTOBB(mp->m_sb.sb_blocksize),
 367                 .ops = &xfs_finobt_buf_ops,
 368                 .work = &xfs_btroot_init,
 369                 .type = XFS_BTNUM_FINO,
 370                 .need_init =  xfs_sb_version_hasfinobt(&mp->m_sb)
 371         },
 372         { /* RMAP root block */
 373                 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_RMAP_BLOCK(mp)),
 374                 .numblks = BTOBB(mp->m_sb.sb_blocksize),
 375                 .ops = &xfs_rmapbt_buf_ops,
 376                 .work = &xfs_rmaproot_init,
 377                 .need_init = xfs_sb_version_hasrmapbt(&mp->m_sb)
 378         },
 379         { /* REFC root block */
 380                 .daddr = XFS_AGB_TO_DADDR(mp, id->agno, xfs_refc_block(mp)),
 381                 .numblks = BTOBB(mp->m_sb.sb_blocksize),
 382                 .ops = &xfs_refcountbt_buf_ops,
 383                 .work = &xfs_btroot_init,
 384                 .type = XFS_BTNUM_REFC,
 385                 .need_init = xfs_sb_version_hasreflink(&mp->m_sb)
 386         },
 387         { /* NULL terminating block */
 388                 .daddr = XFS_BUF_DADDR_NULL,
 389         }
 390         };
 391         struct  xfs_aghdr_grow_data *dp;
 392         int                     error = 0;
 393
 394         /* Account for AG free space in new AG */
 395         id->nfree += id->agsize - mp->m_ag_prealloc_blocks;
 396         for (dp = &aghdr_data[0]; dp->daddr != XFS_BUF_DADDR_NULL; dp++) {
 397                 if (!dp->need_init)
 398                         continue;
 399
 400                 id->daddr = dp->daddr;
 401                 id->numblks = dp->numblks;
 402                 id->type = dp->type;
 403                 error = xfs_ag_init_hdr(mp, id, dp->work, dp->ops);
 404                 if (error)
 405                         break;
 406         }
 407         return error;
 408 }
 409
 410 /*
 411  * Extent the AG indicated by the @id by the length passed in
 412  */
 413 int
 414 xfs_ag_extend_space(
 415         struct xfs_mount        *mp,
 416         struct xfs_trans        *tp,
 417         struct aghdr_init_data  *id,
 418         xfs_extlen_t            len)
 419 {
 420         struct xfs_buf          *bp;
 421         struct xfs_agi          *agi;
 422         struct xfs_agf          *agf;
 423         int                     error;
 424
 425         /*
 426          * Change the agi length.
 427          */
 428         error = xfs_ialloc_read_agi(mp, tp, id->agno, &bp);
 429         if (error)
 430                 return error;
 431
 432         agi = XFS_BUF_TO_AGI(bp);
 433         be32_add_cpu(&agi->agi_length, len);
 434         ASSERT(id->agno == mp->m_sb.sb_agcount - 1 ||
 435                be32_to_cpu(agi->agi_length) == mp->m_sb.sb_agblocks);
 436         xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
 437
 438         /*
 439          * Change agf length.
 440          */
 441         error = xfs_alloc_read_agf(mp, tp, id->agno, 0, &bp);
 442         if (error)
 443                 return error;
 444
 445         agf = XFS_BUF_TO_AGF(bp);
 446         be32_add_cpu(&agf->agf_length, len);
 447         ASSERT(agf->agf_length == agi->agi_length);
 448         xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
 449
 450         /*
 451          * Free the new space.
 452          *
 453          * XFS_RMAP_OINFO_SKIP_UPDATE is used here to tell the rmap btree that
 454          * this doesn't actually exist in the rmap btree.
 455          */
 456         error = xfs_rmap_free(tp, bp, id->agno,
 457                                 be32_to_cpu(agf->agf_length) - len,
 458                                 len, &XFS_RMAP_OINFO_SKIP_UPDATE);
 459         if (error)
 460                 return error;
 461
 462         return  xfs_free_extent(tp, XFS_AGB_TO_FSB(mp, id->agno,
 463                                         be32_to_cpu(agf->agf_length) - len),
 464                                 len, &XFS_RMAP_OINFO_SKIP_UPDATE,
 465                                 XFS_AG_RESV_NONE);
 466 }
 467
 468 /* Retrieve AG geometry. */
 469 int
 470 xfs_ag_get_geometry(
 471         struct xfs_mount        *mp,
 472         xfs_agnumber_t          agno,
 473         struct xfs_ag_geometry  *ageo)
 474 {
 475         struct xfs_buf          *agi_bp;
 476         struct xfs_buf          *agf_bp;
 477         struct xfs_agi          *agi;
 478         struct xfs_agf          *agf;
 479         struct xfs_perag        *pag;
 480         unsigned int            freeblks;
 481         int                     error;
 482
 483         if (agno >= mp->m_sb.sb_agcount)
 484                 return -EINVAL;
 485
 486         /* Lock the AG headers. */
 487         error = xfs_ialloc_read_agi(mp, NULL, agno, &agi_bp);
 488         if (error)
 489                 return error;
 490         error = xfs_alloc_read_agf(mp, NULL, agno, 0, &agf_bp);
 491         if (error)
 492                 goto out_agi;
 493         pag = xfs_perag_get(mp, agno);
 494
 495         /* Fill out form. */
 496         memset(ageo, 0, sizeof(*ageo));
 497         ageo->ag_number = agno;
 498
 499         agi = XFS_BUF_TO_AGI(agi_bp);
 500         ageo->ag_icount = be32_to_cpu(agi->agi_count);
 501         ageo->ag_ifree = be32_to_cpu(agi->agi_freecount);
 502
 503         agf = XFS_BUF_TO_AGF(agf_bp);
 504         ageo->ag_length = be32_to_cpu(agf->agf_length);
 505         freeblks = pag->pagf_freeblks +
 506                    pag->pagf_flcount +
 507                    pag->pagf_btreeblks -
 508                    xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE);
 509         ageo->ag_freeblks = freeblks;
 510         xfs_ag_geom_health(pag, ageo);
 511
 512         /* Release resources. */
 513         xfs_perag_put(pag);
 514         xfs_buf_relse(agf_bp);
 515 out_agi:
 516         xfs_buf_relse(agi_bp);
 517         return error;
 518 }