Back to home page

OSCL-LXR
 
 

     

0001 /* SPDX-License-Identifier: GPL-2.0 */
0002 /*
0003  * Copyright (c) 2000-2005 Silicon Graphics, Inc.
0004  * Copyright (c) 2018 Red Hat, Inc.
0005  * All rights reserved.
0006  */
0007 
0008 #include "xfs.h"
0009 #include "xfs_fs.h"
0010 #include "xfs_shared.h"
0011 #include "xfs_format.h"
0012 #include "xfs_trans_resv.h"
0013 #include "xfs_bit.h"
0014 #include "xfs_sb.h"
0015 #include "xfs_mount.h"
0016 #include "xfs_btree.h"
0017 #include "xfs_alloc_btree.h"
0018 #include "xfs_rmap_btree.h"
0019 #include "xfs_alloc.h"
0020 #include "xfs_ialloc.h"
0021 #include "xfs_rmap.h"
0022 #include "xfs_ag.h"
0023 #include "xfs_ag_resv.h"
0024 #include "xfs_health.h"
0025 #include "xfs_error.h"
0026 #include "xfs_bmap.h"
0027 #include "xfs_defer.h"
0028 #include "xfs_log_format.h"
0029 #include "xfs_trans.h"
0030 #include "xfs_trace.h"
0031 #include "xfs_inode.h"
0032 #include "xfs_icache.h"
0033 
0034 
0035 /*
0036  * Passive reference counting access wrappers to the perag structures.  If the
0037  * per-ag structure is to be freed, the freeing code is responsible for cleaning
0038  * up objects with passive references before freeing the structure. This is
0039  * things like cached buffers.
0040  */
0041 struct xfs_perag *
0042 xfs_perag_get(
0043     struct xfs_mount    *mp,
0044     xfs_agnumber_t      agno)
0045 {
0046     struct xfs_perag    *pag;
0047     int         ref = 0;
0048 
0049     rcu_read_lock();
0050     pag = radix_tree_lookup(&mp->m_perag_tree, agno);
0051     if (pag) {
0052         ASSERT(atomic_read(&pag->pag_ref) >= 0);
0053         ref = atomic_inc_return(&pag->pag_ref);
0054     }
0055     rcu_read_unlock();
0056     trace_xfs_perag_get(mp, agno, ref, _RET_IP_);
0057     return pag;
0058 }
0059 
0060 /*
0061  * search from @first to find the next perag with the given tag set.
0062  */
0063 struct xfs_perag *
0064 xfs_perag_get_tag(
0065     struct xfs_mount    *mp,
0066     xfs_agnumber_t      first,
0067     unsigned int        tag)
0068 {
0069     struct xfs_perag    *pag;
0070     int         found;
0071     int         ref;
0072 
0073     rcu_read_lock();
0074     found = radix_tree_gang_lookup_tag(&mp->m_perag_tree,
0075                     (void **)&pag, first, 1, tag);
0076     if (found <= 0) {
0077         rcu_read_unlock();
0078         return NULL;
0079     }
0080     ref = atomic_inc_return(&pag->pag_ref);
0081     rcu_read_unlock();
0082     trace_xfs_perag_get_tag(mp, pag->pag_agno, ref, _RET_IP_);
0083     return pag;
0084 }
0085 
0086 void
0087 xfs_perag_put(
0088     struct xfs_perag    *pag)
0089 {
0090     int ref;
0091 
0092     ASSERT(atomic_read(&pag->pag_ref) > 0);
0093     ref = atomic_dec_return(&pag->pag_ref);
0094     trace_xfs_perag_put(pag->pag_mount, pag->pag_agno, ref, _RET_IP_);
0095 }
0096 
0097 /*
0098  * xfs_initialize_perag_data
0099  *
0100  * Read in each per-ag structure so we can count up the number of
0101  * allocated inodes, free inodes and used filesystem blocks as this
0102  * information is no longer persistent in the superblock. Once we have
0103  * this information, write it into the in-core superblock structure.
0104  */
0105 int
0106 xfs_initialize_perag_data(
0107     struct xfs_mount    *mp,
0108     xfs_agnumber_t      agcount)
0109 {
0110     xfs_agnumber_t      index;
0111     struct xfs_perag    *pag;
0112     struct xfs_sb       *sbp = &mp->m_sb;
0113     uint64_t        ifree = 0;
0114     uint64_t        ialloc = 0;
0115     uint64_t        bfree = 0;
0116     uint64_t        bfreelst = 0;
0117     uint64_t        btree = 0;
0118     uint64_t        fdblocks;
0119     int         error = 0;
0120 
0121     for (index = 0; index < agcount; index++) {
0122         /*
0123          * Read the AGF and AGI buffers to populate the per-ag
0124          * structures for us.
0125          */
0126         pag = xfs_perag_get(mp, index);
0127         error = xfs_alloc_read_agf(pag, NULL, 0, NULL);
0128         if (!error)
0129             error = xfs_ialloc_read_agi(pag, NULL, NULL);
0130         if (error) {
0131             xfs_perag_put(pag);
0132             return error;
0133         }
0134 
0135         ifree += pag->pagi_freecount;
0136         ialloc += pag->pagi_count;
0137         bfree += pag->pagf_freeblks;
0138         bfreelst += pag->pagf_flcount;
0139         btree += pag->pagf_btreeblks;
0140         xfs_perag_put(pag);
0141     }
0142     fdblocks = bfree + bfreelst + btree;
0143 
0144     /*
0145      * If the new summary counts are obviously incorrect, fail the
0146      * mount operation because that implies the AGFs are also corrupt.
0147      * Clear FS_COUNTERS so that we don't unmount with a dirty log, which
0148      * will prevent xfs_repair from fixing anything.
0149      */
0150     if (fdblocks > sbp->sb_dblocks || ifree > ialloc) {
0151         xfs_alert(mp, "AGF corruption. Please run xfs_repair.");
0152         error = -EFSCORRUPTED;
0153         goto out;
0154     }
0155 
0156     /* Overwrite incore superblock counters with just-read data */
0157     spin_lock(&mp->m_sb_lock);
0158     sbp->sb_ifree = ifree;
0159     sbp->sb_icount = ialloc;
0160     sbp->sb_fdblocks = fdblocks;
0161     spin_unlock(&mp->m_sb_lock);
0162 
0163     xfs_reinit_percpu_counters(mp);
0164 out:
0165     xfs_fs_mark_healthy(mp, XFS_SICK_FS_COUNTERS);
0166     return error;
0167 }
0168 
0169 STATIC void
0170 __xfs_free_perag(
0171     struct rcu_head *head)
0172 {
0173     struct xfs_perag *pag = container_of(head, struct xfs_perag, rcu_head);
0174 
0175     ASSERT(!delayed_work_pending(&pag->pag_blockgc_work));
0176     kmem_free(pag);
0177 }
0178 
0179 /*
0180  * Free up the per-ag resources associated with the mount structure.
0181  */
0182 void
0183 xfs_free_perag(
0184     struct xfs_mount    *mp)
0185 {
0186     struct xfs_perag    *pag;
0187     xfs_agnumber_t      agno;
0188 
0189     for (agno = 0; agno < mp->m_sb.sb_agcount; agno++) {
0190         spin_lock(&mp->m_perag_lock);
0191         pag = radix_tree_delete(&mp->m_perag_tree, agno);
0192         spin_unlock(&mp->m_perag_lock);
0193         ASSERT(pag);
0194         XFS_IS_CORRUPT(pag->pag_mount, atomic_read(&pag->pag_ref) != 0);
0195 
0196         cancel_delayed_work_sync(&pag->pag_blockgc_work);
0197         xfs_buf_hash_destroy(pag);
0198 
0199         call_rcu(&pag->rcu_head, __xfs_free_perag);
0200     }
0201 }
0202 
0203 /* Find the size of the AG, in blocks. */
0204 static xfs_agblock_t
0205 __xfs_ag_block_count(
0206     struct xfs_mount    *mp,
0207     xfs_agnumber_t      agno,
0208     xfs_agnumber_t      agcount,
0209     xfs_rfsblock_t      dblocks)
0210 {
0211     ASSERT(agno < agcount);
0212 
0213     if (agno < agcount - 1)
0214         return mp->m_sb.sb_agblocks;
0215     return dblocks - (agno * mp->m_sb.sb_agblocks);
0216 }
0217 
0218 xfs_agblock_t
0219 xfs_ag_block_count(
0220     struct xfs_mount    *mp,
0221     xfs_agnumber_t      agno)
0222 {
0223     return __xfs_ag_block_count(mp, agno, mp->m_sb.sb_agcount,
0224             mp->m_sb.sb_dblocks);
0225 }
0226 
0227 /* Calculate the first and last possible inode number in an AG. */
0228 static void
0229 __xfs_agino_range(
0230     struct xfs_mount    *mp,
0231     xfs_agblock_t       eoag,
0232     xfs_agino_t     *first,
0233     xfs_agino_t     *last)
0234 {
0235     xfs_agblock_t       bno;
0236 
0237     /*
0238      * Calculate the first inode, which will be in the first
0239      * cluster-aligned block after the AGFL.
0240      */
0241     bno = round_up(XFS_AGFL_BLOCK(mp) + 1, M_IGEO(mp)->cluster_align);
0242     *first = XFS_AGB_TO_AGINO(mp, bno);
0243 
0244     /*
0245      * Calculate the last inode, which will be at the end of the
0246      * last (aligned) cluster that can be allocated in the AG.
0247      */
0248     bno = round_down(eoag, M_IGEO(mp)->cluster_align);
0249     *last = XFS_AGB_TO_AGINO(mp, bno) - 1;
0250 }
0251 
0252 void
0253 xfs_agino_range(
0254     struct xfs_mount    *mp,
0255     xfs_agnumber_t      agno,
0256     xfs_agino_t     *first,
0257     xfs_agino_t     *last)
0258 {
0259     return __xfs_agino_range(mp, xfs_ag_block_count(mp, agno), first, last);
0260 }
0261 
0262 int
0263 xfs_initialize_perag(
0264     struct xfs_mount    *mp,
0265     xfs_agnumber_t      agcount,
0266     xfs_rfsblock_t      dblocks,
0267     xfs_agnumber_t      *maxagi)
0268 {
0269     struct xfs_perag    *pag;
0270     xfs_agnumber_t      index;
0271     xfs_agnumber_t      first_initialised = NULLAGNUMBER;
0272     int         error;
0273 
0274     /*
0275      * Walk the current per-ag tree so we don't try to initialise AGs
0276      * that already exist (growfs case). Allocate and insert all the
0277      * AGs we don't find ready for initialisation.
0278      */
0279     for (index = 0; index < agcount; index++) {
0280         pag = xfs_perag_get(mp, index);
0281         if (pag) {
0282             xfs_perag_put(pag);
0283             continue;
0284         }
0285 
0286         pag = kmem_zalloc(sizeof(*pag), KM_MAYFAIL);
0287         if (!pag) {
0288             error = -ENOMEM;
0289             goto out_unwind_new_pags;
0290         }
0291         pag->pag_agno = index;
0292         pag->pag_mount = mp;
0293 
0294         error = radix_tree_preload(GFP_NOFS);
0295         if (error)
0296             goto out_free_pag;
0297 
0298         spin_lock(&mp->m_perag_lock);
0299         if (radix_tree_insert(&mp->m_perag_tree, index, pag)) {
0300             WARN_ON_ONCE(1);
0301             spin_unlock(&mp->m_perag_lock);
0302             radix_tree_preload_end();
0303             error = -EEXIST;
0304             goto out_free_pag;
0305         }
0306         spin_unlock(&mp->m_perag_lock);
0307         radix_tree_preload_end();
0308 
0309 #ifdef __KERNEL__
0310         /* Place kernel structure only init below this point. */
0311         spin_lock_init(&pag->pag_ici_lock);
0312         spin_lock_init(&pag->pagb_lock);
0313         spin_lock_init(&pag->pag_state_lock);
0314         INIT_DELAYED_WORK(&pag->pag_blockgc_work, xfs_blockgc_worker);
0315         INIT_RADIX_TREE(&pag->pag_ici_root, GFP_ATOMIC);
0316         init_waitqueue_head(&pag->pagb_wait);
0317         pag->pagb_count = 0;
0318         pag->pagb_tree = RB_ROOT;
0319 #endif /* __KERNEL__ */
0320 
0321         error = xfs_buf_hash_init(pag);
0322         if (error)
0323             goto out_remove_pag;
0324 
0325         /* first new pag is fully initialized */
0326         if (first_initialised == NULLAGNUMBER)
0327             first_initialised = index;
0328 
0329         /*
0330          * Pre-calculated geometry
0331          */
0332         pag->block_count = __xfs_ag_block_count(mp, index, agcount,
0333                 dblocks);
0334         pag->min_block = XFS_AGFL_BLOCK(mp);
0335         __xfs_agino_range(mp, pag->block_count, &pag->agino_min,
0336                 &pag->agino_max);
0337     }
0338 
0339     index = xfs_set_inode_alloc(mp, agcount);
0340 
0341     if (maxagi)
0342         *maxagi = index;
0343 
0344     mp->m_ag_prealloc_blocks = xfs_prealloc_blocks(mp);
0345     return 0;
0346 
0347 out_remove_pag:
0348     radix_tree_delete(&mp->m_perag_tree, index);
0349 out_free_pag:
0350     kmem_free(pag);
0351 out_unwind_new_pags:
0352     /* unwind any prior newly initialized pags */
0353     for (index = first_initialised; index < agcount; index++) {
0354         pag = radix_tree_delete(&mp->m_perag_tree, index);
0355         if (!pag)
0356             break;
0357         xfs_buf_hash_destroy(pag);
0358         kmem_free(pag);
0359     }
0360     return error;
0361 }
0362 
0363 static int
0364 xfs_get_aghdr_buf(
0365     struct xfs_mount    *mp,
0366     xfs_daddr_t     blkno,
0367     size_t          numblks,
0368     struct xfs_buf      **bpp,
0369     const struct xfs_buf_ops *ops)
0370 {
0371     struct xfs_buf      *bp;
0372     int         error;
0373 
0374     error = xfs_buf_get_uncached(mp->m_ddev_targp, numblks, 0, &bp);
0375     if (error)
0376         return error;
0377 
0378     bp->b_maps[0].bm_bn = blkno;
0379     bp->b_ops = ops;
0380 
0381     *bpp = bp;
0382     return 0;
0383 }
0384 
0385 /*
0386  * Generic btree root block init function
0387  */
0388 static void
0389 xfs_btroot_init(
0390     struct xfs_mount    *mp,
0391     struct xfs_buf      *bp,
0392     struct aghdr_init_data  *id)
0393 {
0394     xfs_btree_init_block(mp, bp, id->type, 0, 0, id->agno);
0395 }
0396 
0397 /* Finish initializing a free space btree. */
0398 static void
0399 xfs_freesp_init_recs(
0400     struct xfs_mount    *mp,
0401     struct xfs_buf      *bp,
0402     struct aghdr_init_data  *id)
0403 {
0404     struct xfs_alloc_rec    *arec;
0405     struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
0406 
0407     arec = XFS_ALLOC_REC_ADDR(mp, XFS_BUF_TO_BLOCK(bp), 1);
0408     arec->ar_startblock = cpu_to_be32(mp->m_ag_prealloc_blocks);
0409 
0410     if (xfs_ag_contains_log(mp, id->agno)) {
0411         struct xfs_alloc_rec    *nrec;
0412         xfs_agblock_t       start = XFS_FSB_TO_AGBNO(mp,
0413                             mp->m_sb.sb_logstart);
0414 
0415         ASSERT(start >= mp->m_ag_prealloc_blocks);
0416         if (start != mp->m_ag_prealloc_blocks) {
0417             /*
0418              * Modify first record to pad stripe align of log
0419              */
0420             arec->ar_blockcount = cpu_to_be32(start -
0421                         mp->m_ag_prealloc_blocks);
0422             nrec = arec + 1;
0423 
0424             /*
0425              * Insert second record at start of internal log
0426              * which then gets trimmed.
0427              */
0428             nrec->ar_startblock = cpu_to_be32(
0429                     be32_to_cpu(arec->ar_startblock) +
0430                     be32_to_cpu(arec->ar_blockcount));
0431             arec = nrec;
0432             be16_add_cpu(&block->bb_numrecs, 1);
0433         }
0434         /*
0435          * Change record start to after the internal log
0436          */
0437         be32_add_cpu(&arec->ar_startblock, mp->m_sb.sb_logblocks);
0438     }
0439 
0440     /*
0441      * Calculate the record block count and check for the case where
0442      * the log might have consumed all available space in the AG. If
0443      * so, reset the record count to 0 to avoid exposure of an invalid
0444      * record start block.
0445      */
0446     arec->ar_blockcount = cpu_to_be32(id->agsize -
0447                       be32_to_cpu(arec->ar_startblock));
0448     if (!arec->ar_blockcount)
0449         block->bb_numrecs = 0;
0450 }
0451 
0452 /*
0453  * Alloc btree root block init functions
0454  */
0455 static void
0456 xfs_bnoroot_init(
0457     struct xfs_mount    *mp,
0458     struct xfs_buf      *bp,
0459     struct aghdr_init_data  *id)
0460 {
0461     xfs_btree_init_block(mp, bp, XFS_BTNUM_BNO, 0, 1, id->agno);
0462     xfs_freesp_init_recs(mp, bp, id);
0463 }
0464 
0465 static void
0466 xfs_cntroot_init(
0467     struct xfs_mount    *mp,
0468     struct xfs_buf      *bp,
0469     struct aghdr_init_data  *id)
0470 {
0471     xfs_btree_init_block(mp, bp, XFS_BTNUM_CNT, 0, 1, id->agno);
0472     xfs_freesp_init_recs(mp, bp, id);
0473 }
0474 
0475 /*
0476  * Reverse map root block init
0477  */
0478 static void
0479 xfs_rmaproot_init(
0480     struct xfs_mount    *mp,
0481     struct xfs_buf      *bp,
0482     struct aghdr_init_data  *id)
0483 {
0484     struct xfs_btree_block  *block = XFS_BUF_TO_BLOCK(bp);
0485     struct xfs_rmap_rec *rrec;
0486 
0487     xfs_btree_init_block(mp, bp, XFS_BTNUM_RMAP, 0, 4, id->agno);
0488 
0489     /*
0490      * mark the AG header regions as static metadata The BNO
0491      * btree block is the first block after the headers, so
0492      * it's location defines the size of region the static
0493      * metadata consumes.
0494      *
0495      * Note: unlike mkfs, we never have to account for log
0496      * space when growing the data regions
0497      */
0498     rrec = XFS_RMAP_REC_ADDR(block, 1);
0499     rrec->rm_startblock = 0;
0500     rrec->rm_blockcount = cpu_to_be32(XFS_BNO_BLOCK(mp));
0501     rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_FS);
0502     rrec->rm_offset = 0;
0503 
0504     /* account freespace btree root blocks */
0505     rrec = XFS_RMAP_REC_ADDR(block, 2);
0506     rrec->rm_startblock = cpu_to_be32(XFS_BNO_BLOCK(mp));
0507     rrec->rm_blockcount = cpu_to_be32(2);
0508     rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
0509     rrec->rm_offset = 0;
0510 
0511     /* account inode btree root blocks */
0512     rrec = XFS_RMAP_REC_ADDR(block, 3);
0513     rrec->rm_startblock = cpu_to_be32(XFS_IBT_BLOCK(mp));
0514     rrec->rm_blockcount = cpu_to_be32(XFS_RMAP_BLOCK(mp) -
0515                       XFS_IBT_BLOCK(mp));
0516     rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_INOBT);
0517     rrec->rm_offset = 0;
0518 
0519     /* account for rmap btree root */
0520     rrec = XFS_RMAP_REC_ADDR(block, 4);
0521     rrec->rm_startblock = cpu_to_be32(XFS_RMAP_BLOCK(mp));
0522     rrec->rm_blockcount = cpu_to_be32(1);
0523     rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_AG);
0524     rrec->rm_offset = 0;
0525 
0526     /* account for refc btree root */
0527     if (xfs_has_reflink(mp)) {
0528         rrec = XFS_RMAP_REC_ADDR(block, 5);
0529         rrec->rm_startblock = cpu_to_be32(xfs_refc_block(mp));
0530         rrec->rm_blockcount = cpu_to_be32(1);
0531         rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_REFC);
0532         rrec->rm_offset = 0;
0533         be16_add_cpu(&block->bb_numrecs, 1);
0534     }
0535 
0536     /* account for the log space */
0537     if (xfs_ag_contains_log(mp, id->agno)) {
0538         rrec = XFS_RMAP_REC_ADDR(block,
0539                 be16_to_cpu(block->bb_numrecs) + 1);
0540         rrec->rm_startblock = cpu_to_be32(
0541                 XFS_FSB_TO_AGBNO(mp, mp->m_sb.sb_logstart));
0542         rrec->rm_blockcount = cpu_to_be32(mp->m_sb.sb_logblocks);
0543         rrec->rm_owner = cpu_to_be64(XFS_RMAP_OWN_LOG);
0544         rrec->rm_offset = 0;
0545         be16_add_cpu(&block->bb_numrecs, 1);
0546     }
0547 }
0548 
0549 /*
0550  * Initialise new secondary superblocks with the pre-grow geometry, but mark
0551  * them as "in progress" so we know they haven't yet been activated. This will
0552  * get cleared when the update with the new geometry information is done after
0553  * changes to the primary are committed. This isn't strictly necessary, but we
0554  * get it for free with the delayed buffer write lists and it means we can tell
0555  * if a grow operation didn't complete properly after the fact.
0556  */
0557 static void
0558 xfs_sbblock_init(
0559     struct xfs_mount    *mp,
0560     struct xfs_buf      *bp,
0561     struct aghdr_init_data  *id)
0562 {
0563     struct xfs_dsb      *dsb = bp->b_addr;
0564 
0565     xfs_sb_to_disk(dsb, &mp->m_sb);
0566     dsb->sb_inprogress = 1;
0567 }
0568 
0569 static void
0570 xfs_agfblock_init(
0571     struct xfs_mount    *mp,
0572     struct xfs_buf      *bp,
0573     struct aghdr_init_data  *id)
0574 {
0575     struct xfs_agf      *agf = bp->b_addr;
0576     xfs_extlen_t        tmpsize;
0577 
0578     agf->agf_magicnum = cpu_to_be32(XFS_AGF_MAGIC);
0579     agf->agf_versionnum = cpu_to_be32(XFS_AGF_VERSION);
0580     agf->agf_seqno = cpu_to_be32(id->agno);
0581     agf->agf_length = cpu_to_be32(id->agsize);
0582     agf->agf_roots[XFS_BTNUM_BNOi] = cpu_to_be32(XFS_BNO_BLOCK(mp));
0583     agf->agf_roots[XFS_BTNUM_CNTi] = cpu_to_be32(XFS_CNT_BLOCK(mp));
0584     agf->agf_levels[XFS_BTNUM_BNOi] = cpu_to_be32(1);
0585     agf->agf_levels[XFS_BTNUM_CNTi] = cpu_to_be32(1);
0586     if (xfs_has_rmapbt(mp)) {
0587         agf->agf_roots[XFS_BTNUM_RMAPi] =
0588                     cpu_to_be32(XFS_RMAP_BLOCK(mp));
0589         agf->agf_levels[XFS_BTNUM_RMAPi] = cpu_to_be32(1);
0590         agf->agf_rmap_blocks = cpu_to_be32(1);
0591     }
0592 
0593     agf->agf_flfirst = cpu_to_be32(1);
0594     agf->agf_fllast = 0;
0595     agf->agf_flcount = 0;
0596     tmpsize = id->agsize - mp->m_ag_prealloc_blocks;
0597     agf->agf_freeblks = cpu_to_be32(tmpsize);
0598     agf->agf_longest = cpu_to_be32(tmpsize);
0599     if (xfs_has_crc(mp))
0600         uuid_copy(&agf->agf_uuid, &mp->m_sb.sb_meta_uuid);
0601     if (xfs_has_reflink(mp)) {
0602         agf->agf_refcount_root = cpu_to_be32(
0603                 xfs_refc_block(mp));
0604         agf->agf_refcount_level = cpu_to_be32(1);
0605         agf->agf_refcount_blocks = cpu_to_be32(1);
0606     }
0607 
0608     if (xfs_ag_contains_log(mp, id->agno)) {
0609         int64_t logblocks = mp->m_sb.sb_logblocks;
0610 
0611         be32_add_cpu(&agf->agf_freeblks, -logblocks);
0612         agf->agf_longest = cpu_to_be32(id->agsize -
0613             XFS_FSB_TO_AGBNO(mp, mp->m_sb.sb_logstart) - logblocks);
0614     }
0615 }
0616 
0617 static void
0618 xfs_agflblock_init(
0619     struct xfs_mount    *mp,
0620     struct xfs_buf      *bp,
0621     struct aghdr_init_data  *id)
0622 {
0623     struct xfs_agfl     *agfl = XFS_BUF_TO_AGFL(bp);
0624     __be32          *agfl_bno;
0625     int         bucket;
0626 
0627     if (xfs_has_crc(mp)) {
0628         agfl->agfl_magicnum = cpu_to_be32(XFS_AGFL_MAGIC);
0629         agfl->agfl_seqno = cpu_to_be32(id->agno);
0630         uuid_copy(&agfl->agfl_uuid, &mp->m_sb.sb_meta_uuid);
0631     }
0632 
0633     agfl_bno = xfs_buf_to_agfl_bno(bp);
0634     for (bucket = 0; bucket < xfs_agfl_size(mp); bucket++)
0635         agfl_bno[bucket] = cpu_to_be32(NULLAGBLOCK);
0636 }
0637 
0638 static void
0639 xfs_agiblock_init(
0640     struct xfs_mount    *mp,
0641     struct xfs_buf      *bp,
0642     struct aghdr_init_data  *id)
0643 {
0644     struct xfs_agi      *agi = bp->b_addr;
0645     int         bucket;
0646 
0647     agi->agi_magicnum = cpu_to_be32(XFS_AGI_MAGIC);
0648     agi->agi_versionnum = cpu_to_be32(XFS_AGI_VERSION);
0649     agi->agi_seqno = cpu_to_be32(id->agno);
0650     agi->agi_length = cpu_to_be32(id->agsize);
0651     agi->agi_count = 0;
0652     agi->agi_root = cpu_to_be32(XFS_IBT_BLOCK(mp));
0653     agi->agi_level = cpu_to_be32(1);
0654     agi->agi_freecount = 0;
0655     agi->agi_newino = cpu_to_be32(NULLAGINO);
0656     agi->agi_dirino = cpu_to_be32(NULLAGINO);
0657     if (xfs_has_crc(mp))
0658         uuid_copy(&agi->agi_uuid, &mp->m_sb.sb_meta_uuid);
0659     if (xfs_has_finobt(mp)) {
0660         agi->agi_free_root = cpu_to_be32(XFS_FIBT_BLOCK(mp));
0661         agi->agi_free_level = cpu_to_be32(1);
0662     }
0663     for (bucket = 0; bucket < XFS_AGI_UNLINKED_BUCKETS; bucket++)
0664         agi->agi_unlinked[bucket] = cpu_to_be32(NULLAGINO);
0665     if (xfs_has_inobtcounts(mp)) {
0666         agi->agi_iblocks = cpu_to_be32(1);
0667         if (xfs_has_finobt(mp))
0668             agi->agi_fblocks = cpu_to_be32(1);
0669     }
0670 }
0671 
0672 typedef void (*aghdr_init_work_f)(struct xfs_mount *mp, struct xfs_buf *bp,
0673                   struct aghdr_init_data *id);
0674 static int
0675 xfs_ag_init_hdr(
0676     struct xfs_mount    *mp,
0677     struct aghdr_init_data  *id,
0678     aghdr_init_work_f   work,
0679     const struct xfs_buf_ops *ops)
0680 {
0681     struct xfs_buf      *bp;
0682     int         error;
0683 
0684     error = xfs_get_aghdr_buf(mp, id->daddr, id->numblks, &bp, ops);
0685     if (error)
0686         return error;
0687 
0688     (*work)(mp, bp, id);
0689 
0690     xfs_buf_delwri_queue(bp, &id->buffer_list);
0691     xfs_buf_relse(bp);
0692     return 0;
0693 }
0694 
0695 struct xfs_aghdr_grow_data {
0696     xfs_daddr_t     daddr;
0697     size_t          numblks;
0698     const struct xfs_buf_ops *ops;
0699     aghdr_init_work_f   work;
0700     xfs_btnum_t     type;
0701     bool            need_init;
0702 };
0703 
0704 /*
0705  * Prepare new AG headers to be written to disk. We use uncached buffers here,
0706  * as it is assumed these new AG headers are currently beyond the currently
0707  * valid filesystem address space. Using cached buffers would trip over EOFS
0708  * corruption detection alogrithms in the buffer cache lookup routines.
0709  *
0710  * This is a non-transactional function, but the prepared buffers are added to a
0711  * delayed write buffer list supplied by the caller so they can submit them to
0712  * disk and wait on them as required.
0713  */
0714 int
0715 xfs_ag_init_headers(
0716     struct xfs_mount    *mp,
0717     struct aghdr_init_data  *id)
0718 
0719 {
0720     struct xfs_aghdr_grow_data aghdr_data[] = {
0721     { /* SB */
0722         .daddr = XFS_AG_DADDR(mp, id->agno, XFS_SB_DADDR),
0723         .numblks = XFS_FSS_TO_BB(mp, 1),
0724         .ops = &xfs_sb_buf_ops,
0725         .work = &xfs_sbblock_init,
0726         .need_init = true
0727     },
0728     { /* AGF */
0729         .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGF_DADDR(mp)),
0730         .numblks = XFS_FSS_TO_BB(mp, 1),
0731         .ops = &xfs_agf_buf_ops,
0732         .work = &xfs_agfblock_init,
0733         .need_init = true
0734     },
0735     { /* AGFL */
0736         .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGFL_DADDR(mp)),
0737         .numblks = XFS_FSS_TO_BB(mp, 1),
0738         .ops = &xfs_agfl_buf_ops,
0739         .work = &xfs_agflblock_init,
0740         .need_init = true
0741     },
0742     { /* AGI */
0743         .daddr = XFS_AG_DADDR(mp, id->agno, XFS_AGI_DADDR(mp)),
0744         .numblks = XFS_FSS_TO_BB(mp, 1),
0745         .ops = &xfs_agi_buf_ops,
0746         .work = &xfs_agiblock_init,
0747         .need_init = true
0748     },
0749     { /* BNO root block */
0750         .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_BNO_BLOCK(mp)),
0751         .numblks = BTOBB(mp->m_sb.sb_blocksize),
0752         .ops = &xfs_bnobt_buf_ops,
0753         .work = &xfs_bnoroot_init,
0754         .need_init = true
0755     },
0756     { /* CNT root block */
0757         .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_CNT_BLOCK(mp)),
0758         .numblks = BTOBB(mp->m_sb.sb_blocksize),
0759         .ops = &xfs_cntbt_buf_ops,
0760         .work = &xfs_cntroot_init,
0761         .need_init = true
0762     },
0763     { /* INO root block */
0764         .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_IBT_BLOCK(mp)),
0765         .numblks = BTOBB(mp->m_sb.sb_blocksize),
0766         .ops = &xfs_inobt_buf_ops,
0767         .work = &xfs_btroot_init,
0768         .type = XFS_BTNUM_INO,
0769         .need_init = true
0770     },
0771     { /* FINO root block */
0772         .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_FIBT_BLOCK(mp)),
0773         .numblks = BTOBB(mp->m_sb.sb_blocksize),
0774         .ops = &xfs_finobt_buf_ops,
0775         .work = &xfs_btroot_init,
0776         .type = XFS_BTNUM_FINO,
0777         .need_init =  xfs_has_finobt(mp)
0778     },
0779     { /* RMAP root block */
0780         .daddr = XFS_AGB_TO_DADDR(mp, id->agno, XFS_RMAP_BLOCK(mp)),
0781         .numblks = BTOBB(mp->m_sb.sb_blocksize),
0782         .ops = &xfs_rmapbt_buf_ops,
0783         .work = &xfs_rmaproot_init,
0784         .need_init = xfs_has_rmapbt(mp)
0785     },
0786     { /* REFC root block */
0787         .daddr = XFS_AGB_TO_DADDR(mp, id->agno, xfs_refc_block(mp)),
0788         .numblks = BTOBB(mp->m_sb.sb_blocksize),
0789         .ops = &xfs_refcountbt_buf_ops,
0790         .work = &xfs_btroot_init,
0791         .type = XFS_BTNUM_REFC,
0792         .need_init = xfs_has_reflink(mp)
0793     },
0794     { /* NULL terminating block */
0795         .daddr = XFS_BUF_DADDR_NULL,
0796     }
0797     };
0798     struct  xfs_aghdr_grow_data *dp;
0799     int         error = 0;
0800 
0801     /* Account for AG free space in new AG */
0802     id->nfree += id->agsize - mp->m_ag_prealloc_blocks;
0803     for (dp = &aghdr_data[0]; dp->daddr != XFS_BUF_DADDR_NULL; dp++) {
0804         if (!dp->need_init)
0805             continue;
0806 
0807         id->daddr = dp->daddr;
0808         id->numblks = dp->numblks;
0809         id->type = dp->type;
0810         error = xfs_ag_init_hdr(mp, id, dp->work, dp->ops);
0811         if (error)
0812             break;
0813     }
0814     return error;
0815 }
0816 
0817 int
0818 xfs_ag_shrink_space(
0819     struct xfs_perag    *pag,
0820     struct xfs_trans    **tpp,
0821     xfs_extlen_t        delta)
0822 {
0823     struct xfs_mount    *mp = pag->pag_mount;
0824     struct xfs_alloc_arg    args = {
0825         .tp = *tpp,
0826         .mp = mp,
0827         .type   = XFS_ALLOCTYPE_THIS_BNO,
0828         .minlen = delta,
0829         .maxlen = delta,
0830         .oinfo  = XFS_RMAP_OINFO_SKIP_UPDATE,
0831         .resv   = XFS_AG_RESV_NONE,
0832         .prod   = 1
0833     };
0834     struct xfs_buf      *agibp, *agfbp;
0835     struct xfs_agi      *agi;
0836     struct xfs_agf      *agf;
0837     xfs_agblock_t       aglen;
0838     int         error, err2;
0839 
0840     ASSERT(pag->pag_agno == mp->m_sb.sb_agcount - 1);
0841     error = xfs_ialloc_read_agi(pag, *tpp, &agibp);
0842     if (error)
0843         return error;
0844 
0845     agi = agibp->b_addr;
0846 
0847     error = xfs_alloc_read_agf(pag, *tpp, 0, &agfbp);
0848     if (error)
0849         return error;
0850 
0851     agf = agfbp->b_addr;
0852     aglen = be32_to_cpu(agi->agi_length);
0853     /* some extra paranoid checks before we shrink the ag */
0854     if (XFS_IS_CORRUPT(mp, agf->agf_length != agi->agi_length))
0855         return -EFSCORRUPTED;
0856     if (delta >= aglen)
0857         return -EINVAL;
0858 
0859     args.fsbno = XFS_AGB_TO_FSB(mp, pag->pag_agno, aglen - delta);
0860 
0861     /*
0862      * Make sure that the last inode cluster cannot overlap with the new
0863      * end of the AG, even if it's sparse.
0864      */
0865     error = xfs_ialloc_check_shrink(*tpp, pag->pag_agno, agibp,
0866             aglen - delta);
0867     if (error)
0868         return error;
0869 
0870     /*
0871      * Disable perag reservations so it doesn't cause the allocation request
0872      * to fail. We'll reestablish reservation before we return.
0873      */
0874     error = xfs_ag_resv_free(pag);
0875     if (error)
0876         return error;
0877 
0878     /* internal log shouldn't also show up in the free space btrees */
0879     error = xfs_alloc_vextent(&args);
0880     if (!error && args.agbno == NULLAGBLOCK)
0881         error = -ENOSPC;
0882 
0883     if (error) {
0884         /*
0885          * if extent allocation fails, need to roll the transaction to
0886          * ensure that the AGFL fixup has been committed anyway.
0887          */
0888         xfs_trans_bhold(*tpp, agfbp);
0889         err2 = xfs_trans_roll(tpp);
0890         if (err2)
0891             return err2;
0892         xfs_trans_bjoin(*tpp, agfbp);
0893         goto resv_init_out;
0894     }
0895 
0896     /*
0897      * if successfully deleted from freespace btrees, need to confirm
0898      * per-AG reservation works as expected.
0899      */
0900     be32_add_cpu(&agi->agi_length, -delta);
0901     be32_add_cpu(&agf->agf_length, -delta);
0902 
0903     err2 = xfs_ag_resv_init(pag, *tpp);
0904     if (err2) {
0905         be32_add_cpu(&agi->agi_length, delta);
0906         be32_add_cpu(&agf->agf_length, delta);
0907         if (err2 != -ENOSPC)
0908             goto resv_err;
0909 
0910         __xfs_free_extent_later(*tpp, args.fsbno, delta, NULL, true);
0911 
0912         /*
0913          * Roll the transaction before trying to re-init the per-ag
0914          * reservation. The new transaction is clean so it will cancel
0915          * without any side effects.
0916          */
0917         error = xfs_defer_finish(tpp);
0918         if (error)
0919             return error;
0920 
0921         error = -ENOSPC;
0922         goto resv_init_out;
0923     }
0924     xfs_ialloc_log_agi(*tpp, agibp, XFS_AGI_LENGTH);
0925     xfs_alloc_log_agf(*tpp, agfbp, XFS_AGF_LENGTH);
0926     return 0;
0927 
0928 resv_init_out:
0929     err2 = xfs_ag_resv_init(pag, *tpp);
0930     if (!err2)
0931         return error;
0932 resv_err:
0933     xfs_warn(mp, "Error %d reserving per-AG metadata reserve pool.", err2);
0934     xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
0935     return err2;
0936 }
0937 
0938 /*
0939  * Extent the AG indicated by the @id by the length passed in
0940  */
0941 int
0942 xfs_ag_extend_space(
0943     struct xfs_perag    *pag,
0944     struct xfs_trans    *tp,
0945     xfs_extlen_t        len)
0946 {
0947     struct xfs_buf      *bp;
0948     struct xfs_agi      *agi;
0949     struct xfs_agf      *agf;
0950     int         error;
0951 
0952     ASSERT(pag->pag_agno == pag->pag_mount->m_sb.sb_agcount - 1);
0953 
0954     error = xfs_ialloc_read_agi(pag, tp, &bp);
0955     if (error)
0956         return error;
0957 
0958     agi = bp->b_addr;
0959     be32_add_cpu(&agi->agi_length, len);
0960     xfs_ialloc_log_agi(tp, bp, XFS_AGI_LENGTH);
0961 
0962     /*
0963      * Change agf length.
0964      */
0965     error = xfs_alloc_read_agf(pag, tp, 0, &bp);
0966     if (error)
0967         return error;
0968 
0969     agf = bp->b_addr;
0970     be32_add_cpu(&agf->agf_length, len);
0971     ASSERT(agf->agf_length == agi->agi_length);
0972     xfs_alloc_log_agf(tp, bp, XFS_AGF_LENGTH);
0973 
0974     /*
0975      * Free the new space.
0976      *
0977      * XFS_RMAP_OINFO_SKIP_UPDATE is used here to tell the rmap btree that
0978      * this doesn't actually exist in the rmap btree.
0979      */
0980     error = xfs_rmap_free(tp, bp, pag, be32_to_cpu(agf->agf_length) - len,
0981                 len, &XFS_RMAP_OINFO_SKIP_UPDATE);
0982     if (error)
0983         return error;
0984 
0985     error = xfs_free_extent(tp, XFS_AGB_TO_FSB(pag->pag_mount, pag->pag_agno,
0986                     be32_to_cpu(agf->agf_length) - len),
0987                 len, &XFS_RMAP_OINFO_SKIP_UPDATE,
0988                 XFS_AG_RESV_NONE);
0989     if (error)
0990         return error;
0991 
0992     /* Update perag geometry */
0993     pag->block_count = be32_to_cpu(agf->agf_length);
0994     __xfs_agino_range(pag->pag_mount, pag->block_count, &pag->agino_min,
0995                 &pag->agino_max);
0996     return 0;
0997 }
0998 
0999 /* Retrieve AG geometry. */
1000 int
1001 xfs_ag_get_geometry(
1002     struct xfs_perag    *pag,
1003     struct xfs_ag_geometry  *ageo)
1004 {
1005     struct xfs_buf      *agi_bp;
1006     struct xfs_buf      *agf_bp;
1007     struct xfs_agi      *agi;
1008     struct xfs_agf      *agf;
1009     unsigned int        freeblks;
1010     int         error;
1011 
1012     /* Lock the AG headers. */
1013     error = xfs_ialloc_read_agi(pag, NULL, &agi_bp);
1014     if (error)
1015         return error;
1016     error = xfs_alloc_read_agf(pag, NULL, 0, &agf_bp);
1017     if (error)
1018         goto out_agi;
1019 
1020     /* Fill out form. */
1021     memset(ageo, 0, sizeof(*ageo));
1022     ageo->ag_number = pag->pag_agno;
1023 
1024     agi = agi_bp->b_addr;
1025     ageo->ag_icount = be32_to_cpu(agi->agi_count);
1026     ageo->ag_ifree = be32_to_cpu(agi->agi_freecount);
1027 
1028     agf = agf_bp->b_addr;
1029     ageo->ag_length = be32_to_cpu(agf->agf_length);
1030     freeblks = pag->pagf_freeblks +
1031            pag->pagf_flcount +
1032            pag->pagf_btreeblks -
1033            xfs_ag_resv_needed(pag, XFS_AG_RESV_NONE);
1034     ageo->ag_freeblks = freeblks;
1035     xfs_ag_geom_health(pag, ageo);
1036 
1037     /* Release resources. */
1038     xfs_buf_relse(agf_bp);
1039 out_agi:
1040     xfs_buf_relse(agi_bp);
1041     return error;
1042 }
[ Source navigation ] [ Diff markup ] [ Identifier search ] [ general search ]

This page was automatically generated by the 2.1.0 LXR engine.

The LXR team
Valid CSS 2.1! Valid HTML 4.01!