2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
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.
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.
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
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
26 #include "xfs_mount.h"
27 #include "xfs_da_format.h"
28 #include "xfs_da_btree.h"
30 #include "xfs_inode.h"
31 #include "xfs_btree.h"
32 #include "xfs_trans.h"
33 #include "xfs_inode_item.h"
34 #include "xfs_extfree_item.h"
35 #include "xfs_alloc.h"
37 #include "xfs_bmap_util.h"
38 #include "xfs_bmap_btree.h"
39 #include "xfs_rtalloc.h"
40 #include "xfs_error.h"
41 #include "xfs_quota.h"
42 #include "xfs_trans_space.h"
43 #include "xfs_buf_item.h"
44 #include "xfs_trace.h"
45 #include "xfs_symlink.h"
46 #include "xfs_attr_leaf.h"
47 #include "xfs_filestream.h"
50 kmem_zone_t *xfs_bmap_free_item_zone;
53 * Miscellaneous helper functions
57 * Compute and fill in the value of the maximum depth of a bmap btree
58 * in this filesystem. Done once, during mount.
61 xfs_bmap_compute_maxlevels(
62 xfs_mount_t *mp, /* file system mount structure */
63 int whichfork) /* data or attr fork */
65 int level; /* btree level */
66 uint maxblocks; /* max blocks at this level */
67 uint maxleafents; /* max leaf entries possible */
68 int maxrootrecs; /* max records in root block */
69 int minleafrecs; /* min records in leaf block */
70 int minnoderecs; /* min records in node block */
71 int sz; /* root block size */
74 * The maximum number of extents in a file, hence the maximum
75 * number of leaf entries, is controlled by the type of di_nextents
76 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
77 * (a signed 16-bit number, xfs_aextnum_t).
79 * Note that we can no longer assume that if we are in ATTR1 that
80 * the fork offset of all the inodes will be
81 * (xfs_default_attroffset(ip) >> 3) because we could have mounted
82 * with ATTR2 and then mounted back with ATTR1, keeping the
83 * di_forkoff's fixed but probably at various positions. Therefore,
84 * for both ATTR1 and ATTR2 we have to assume the worst case scenario
85 * of a minimum size available.
87 if (whichfork == XFS_DATA_FORK) {
88 maxleafents = MAXEXTNUM;
89 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
91 maxleafents = MAXAEXTNUM;
92 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
94 maxrootrecs = xfs_bmdr_maxrecs(sz, 0);
95 minleafrecs = mp->m_bmap_dmnr[0];
96 minnoderecs = mp->m_bmap_dmnr[1];
97 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
98 for (level = 1; maxblocks > 1; level++) {
99 if (maxblocks <= maxrootrecs)
102 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
104 mp->m_bm_maxlevels[whichfork] = level;
107 STATIC int /* error */
109 struct xfs_btree_cur *cur,
113 int *stat) /* success/failure */
115 cur->bc_rec.b.br_startoff = off;
116 cur->bc_rec.b.br_startblock = bno;
117 cur->bc_rec.b.br_blockcount = len;
118 return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
121 STATIC int /* error */
123 struct xfs_btree_cur *cur,
127 int *stat) /* success/failure */
129 cur->bc_rec.b.br_startoff = off;
130 cur->bc_rec.b.br_startblock = bno;
131 cur->bc_rec.b.br_blockcount = len;
132 return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
136 * Check if the inode needs to be converted to btree format.
138 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
140 return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
141 XFS_IFORK_NEXTENTS(ip, whichfork) >
142 XFS_IFORK_MAXEXT(ip, whichfork);
146 * Check if the inode should be converted to extent format.
148 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
150 return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
151 XFS_IFORK_NEXTENTS(ip, whichfork) <=
152 XFS_IFORK_MAXEXT(ip, whichfork);
156 * Update the record referred to by cur to the value given
157 * by [off, bno, len, state].
158 * This either works (return 0) or gets an EFSCORRUPTED error.
162 struct xfs_btree_cur *cur,
168 union xfs_btree_rec rec;
170 xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
171 return xfs_btree_update(cur, &rec);
175 * Compute the worst-case number of indirect blocks that will be used
176 * for ip's delayed extent of length "len".
179 xfs_bmap_worst_indlen(
180 xfs_inode_t *ip, /* incore inode pointer */
181 xfs_filblks_t len) /* delayed extent length */
183 int level; /* btree level number */
184 int maxrecs; /* maximum record count at this level */
185 xfs_mount_t *mp; /* mount structure */
186 xfs_filblks_t rval; /* return value */
189 maxrecs = mp->m_bmap_dmxr[0];
190 for (level = 0, rval = 0;
191 level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
194 do_div(len, maxrecs);
197 return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
200 maxrecs = mp->m_bmap_dmxr[1];
206 * Calculate the default attribute fork offset for newly created inodes.
209 xfs_default_attroffset(
210 struct xfs_inode *ip)
212 struct xfs_mount *mp = ip->i_mount;
215 if (mp->m_sb.sb_inodesize == 256) {
216 offset = XFS_LITINO(mp, ip->i_d.di_version) -
217 XFS_BMDR_SPACE_CALC(MINABTPTRS);
219 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
222 ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version));
227 * Helper routine to reset inode di_forkoff field when switching
228 * attribute fork from local to extent format - we reset it where
229 * possible to make space available for inline data fork extents.
232 xfs_bmap_forkoff_reset(
236 if (whichfork == XFS_ATTR_FORK &&
237 ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
238 ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
239 ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
240 uint dfl_forkoff = xfs_default_attroffset(ip) >> 3;
242 if (dfl_forkoff > ip->i_d.di_forkoff)
243 ip->i_d.di_forkoff = dfl_forkoff;
248 * Debug/sanity checking code
252 xfs_bmap_sanity_check(
253 struct xfs_mount *mp,
257 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
259 if (block->bb_magic != cpu_to_be32(XFS_BMAP_CRC_MAGIC) &&
260 block->bb_magic != cpu_to_be32(XFS_BMAP_MAGIC))
263 if (be16_to_cpu(block->bb_level) != level ||
264 be16_to_cpu(block->bb_numrecs) == 0 ||
265 be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
272 STATIC struct xfs_buf *
274 struct xfs_btree_cur *cur,
277 struct xfs_log_item_desc *lidp;
283 for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
284 if (!cur->bc_bufs[i])
286 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
287 return cur->bc_bufs[i];
290 /* Chase down all the log items to see if the bp is there */
291 list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
292 struct xfs_buf_log_item *bip;
293 bip = (struct xfs_buf_log_item *)lidp->lid_item;
294 if (bip->bli_item.li_type == XFS_LI_BUF &&
295 XFS_BUF_ADDR(bip->bli_buf) == bno)
304 struct xfs_btree_block *block,
310 __be64 *pp, *thispa; /* pointer to block address */
311 xfs_bmbt_key_t *prevp, *keyp;
313 ASSERT(be16_to_cpu(block->bb_level) > 0);
316 for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
317 dmxr = mp->m_bmap_dmxr[0];
318 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
321 ASSERT(be64_to_cpu(prevp->br_startoff) <
322 be64_to_cpu(keyp->br_startoff));
327 * Compare the block numbers to see if there are dups.
330 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
332 pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
334 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
336 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
338 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
339 if (*thispa == *pp) {
340 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
342 (unsigned long long)be64_to_cpu(*thispa));
343 panic("%s: ptrs are equal in node\n",
351 * Check that the extents for the inode ip are in the right order in all
356 xfs_bmap_check_leaf_extents(
357 xfs_btree_cur_t *cur, /* btree cursor or null */
358 xfs_inode_t *ip, /* incore inode pointer */
359 int whichfork) /* data or attr fork */
361 struct xfs_btree_block *block; /* current btree block */
362 xfs_fsblock_t bno; /* block # of "block" */
363 xfs_buf_t *bp; /* buffer for "block" */
364 int error; /* error return value */
365 xfs_extnum_t i=0, j; /* index into the extents list */
366 xfs_ifork_t *ifp; /* fork structure */
367 int level; /* btree level, for checking */
368 xfs_mount_t *mp; /* file system mount structure */
369 __be64 *pp; /* pointer to block address */
370 xfs_bmbt_rec_t *ep; /* pointer to current extent */
371 xfs_bmbt_rec_t last = {0, 0}; /* last extent in prev block */
372 xfs_bmbt_rec_t *nextp; /* pointer to next extent */
375 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
381 ifp = XFS_IFORK_PTR(ip, whichfork);
382 block = ifp->if_broot;
384 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
386 level = be16_to_cpu(block->bb_level);
388 xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
389 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
390 bno = be64_to_cpu(*pp);
392 ASSERT(bno != NULLFSBLOCK);
393 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
394 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
397 * Go down the tree until leaf level is reached, following the first
398 * pointer (leftmost) at each level.
400 while (level-- > 0) {
401 /* See if buf is in cur first */
403 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
406 error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
412 block = XFS_BUF_TO_BLOCK(bp);
413 XFS_WANT_CORRUPTED_GOTO(mp,
414 xfs_bmap_sanity_check(mp, bp, level),
420 * Check this block for basic sanity (increasing keys and
421 * no duplicate blocks).
424 xfs_check_block(block, mp, 0, 0);
425 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
426 bno = be64_to_cpu(*pp);
427 XFS_WANT_CORRUPTED_GOTO(mp,
428 XFS_FSB_SANITY_CHECK(mp, bno), error0);
431 xfs_trans_brelse(NULL, bp);
436 * Here with bp and block set to the leftmost leaf node in the tree.
441 * Loop over all leaf nodes checking that all extents are in the right order.
444 xfs_fsblock_t nextbno;
445 xfs_extnum_t num_recs;
448 num_recs = xfs_btree_get_numrecs(block);
451 * Read-ahead the next leaf block, if any.
454 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
457 * Check all the extents to make sure they are OK.
458 * If we had a previous block, the last entry should
459 * conform with the first entry in this one.
462 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
464 ASSERT(xfs_bmbt_disk_get_startoff(&last) +
465 xfs_bmbt_disk_get_blockcount(&last) <=
466 xfs_bmbt_disk_get_startoff(ep));
468 for (j = 1; j < num_recs; j++) {
469 nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
470 ASSERT(xfs_bmbt_disk_get_startoff(ep) +
471 xfs_bmbt_disk_get_blockcount(ep) <=
472 xfs_bmbt_disk_get_startoff(nextp));
480 xfs_trans_brelse(NULL, bp);
484 * If we've reached the end, stop.
486 if (bno == NULLFSBLOCK)
490 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
493 error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
499 block = XFS_BUF_TO_BLOCK(bp);
503 xfs_trans_brelse(NULL, bp);
508 xfs_warn(mp, "%s: at error0", __func__);
510 xfs_trans_brelse(NULL, bp);
512 xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
514 panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
519 * Add bmap trace insert entries for all the contents of the extent records.
522 xfs_bmap_trace_exlist(
523 xfs_inode_t *ip, /* incore inode pointer */
524 xfs_extnum_t cnt, /* count of entries in the list */
525 int whichfork, /* data or attr fork */
526 unsigned long caller_ip)
528 xfs_extnum_t idx; /* extent record index */
529 xfs_ifork_t *ifp; /* inode fork pointer */
532 if (whichfork == XFS_ATTR_FORK)
533 state |= BMAP_ATTRFORK;
535 ifp = XFS_IFORK_PTR(ip, whichfork);
536 ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
537 for (idx = 0; idx < cnt; idx++)
538 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
542 * Validate that the bmbt_irecs being returned from bmapi are valid
543 * given the caller's original parameters. Specifically check the
544 * ranges of the returned irecs to ensure that they only extend beyond
545 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
548 xfs_bmap_validate_ret(
552 xfs_bmbt_irec_t *mval,
556 int i; /* index to map values */
558 ASSERT(ret_nmap <= nmap);
560 for (i = 0; i < ret_nmap; i++) {
561 ASSERT(mval[i].br_blockcount > 0);
562 if (!(flags & XFS_BMAPI_ENTIRE)) {
563 ASSERT(mval[i].br_startoff >= bno);
564 ASSERT(mval[i].br_blockcount <= len);
565 ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
568 ASSERT(mval[i].br_startoff < bno + len);
569 ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
573 mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
574 mval[i].br_startoff);
575 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
576 mval[i].br_startblock != HOLESTARTBLOCK);
577 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
578 mval[i].br_state == XFS_EXT_UNWRITTEN);
583 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork) do { } while (0)
584 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
588 * bmap free list manipulation functions
592 * Add the extent to the list of extents to be free at transaction end.
593 * The list is maintained sorted (by block number).
597 xfs_fsblock_t bno, /* fs block number of extent */
598 xfs_filblks_t len, /* length of extent */
599 xfs_bmap_free_t *flist, /* list of extents */
600 xfs_mount_t *mp) /* mount point structure */
602 xfs_bmap_free_item_t *cur; /* current (next) element */
603 xfs_bmap_free_item_t *new; /* new element */
604 xfs_bmap_free_item_t *prev; /* previous element */
609 ASSERT(bno != NULLFSBLOCK);
611 ASSERT(len <= MAXEXTLEN);
612 ASSERT(!isnullstartblock(bno));
613 agno = XFS_FSB_TO_AGNO(mp, bno);
614 agbno = XFS_FSB_TO_AGBNO(mp, bno);
615 ASSERT(agno < mp->m_sb.sb_agcount);
616 ASSERT(agbno < mp->m_sb.sb_agblocks);
617 ASSERT(len < mp->m_sb.sb_agblocks);
618 ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
620 ASSERT(xfs_bmap_free_item_zone != NULL);
621 new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
622 new->xbfi_startblock = bno;
623 new->xbfi_blockcount = (xfs_extlen_t)len;
624 for (prev = NULL, cur = flist->xbf_first;
626 prev = cur, cur = cur->xbfi_next) {
627 if (cur->xbfi_startblock >= bno)
631 prev->xbfi_next = new;
633 flist->xbf_first = new;
634 new->xbfi_next = cur;
639 * Remove the entry "free" from the free item list. Prev points to the
640 * previous entry, unless "free" is the head of the list.
644 xfs_bmap_free_t *flist, /* free item list header */
645 xfs_bmap_free_item_t *prev, /* previous item on list, if any */
646 xfs_bmap_free_item_t *free) /* list item to be freed */
649 prev->xbfi_next = free->xbfi_next;
651 flist->xbf_first = free->xbfi_next;
653 kmem_zone_free(xfs_bmap_free_item_zone, free);
657 * Free up any items left in the list.
661 xfs_bmap_free_t *flist) /* list of bmap_free_items */
663 xfs_bmap_free_item_t *free; /* free list item */
664 xfs_bmap_free_item_t *next;
666 if (flist->xbf_count == 0)
668 ASSERT(flist->xbf_first != NULL);
669 for (free = flist->xbf_first; free; free = next) {
670 next = free->xbfi_next;
671 xfs_bmap_del_free(flist, NULL, free);
673 ASSERT(flist->xbf_count == 0);
677 * Inode fork format manipulation functions
681 * Transform a btree format file with only one leaf node, where the
682 * extents list will fit in the inode, into an extents format file.
683 * Since the file extents are already in-core, all we have to do is
684 * give up the space for the btree root and pitch the leaf block.
686 STATIC int /* error */
687 xfs_bmap_btree_to_extents(
688 xfs_trans_t *tp, /* transaction pointer */
689 xfs_inode_t *ip, /* incore inode pointer */
690 xfs_btree_cur_t *cur, /* btree cursor */
691 int *logflagsp, /* inode logging flags */
692 int whichfork) /* data or attr fork */
695 struct xfs_btree_block *cblock;/* child btree block */
696 xfs_fsblock_t cbno; /* child block number */
697 xfs_buf_t *cbp; /* child block's buffer */
698 int error; /* error return value */
699 xfs_ifork_t *ifp; /* inode fork data */
700 xfs_mount_t *mp; /* mount point structure */
701 __be64 *pp; /* ptr to block address */
702 struct xfs_btree_block *rblock;/* root btree block */
705 ifp = XFS_IFORK_PTR(ip, whichfork);
706 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
707 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
708 rblock = ifp->if_broot;
709 ASSERT(be16_to_cpu(rblock->bb_level) == 1);
710 ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
711 ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
712 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
713 cbno = be64_to_cpu(*pp);
716 if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
719 error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
723 cblock = XFS_BUF_TO_BLOCK(cbp);
724 if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
726 xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
727 ip->i_d.di_nblocks--;
728 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
729 xfs_trans_binval(tp, cbp);
730 if (cur->bc_bufs[0] == cbp)
731 cur->bc_bufs[0] = NULL;
732 xfs_iroot_realloc(ip, -1, whichfork);
733 ASSERT(ifp->if_broot == NULL);
734 ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
735 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
736 *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
741 * Convert an extents-format file into a btree-format file.
742 * The new file will have a root block (in the inode) and a single child block.
744 STATIC int /* error */
745 xfs_bmap_extents_to_btree(
746 xfs_trans_t *tp, /* transaction pointer */
747 xfs_inode_t *ip, /* incore inode pointer */
748 xfs_fsblock_t *firstblock, /* first-block-allocated */
749 xfs_bmap_free_t *flist, /* blocks freed in xaction */
750 xfs_btree_cur_t **curp, /* cursor returned to caller */
751 int wasdel, /* converting a delayed alloc */
752 int *logflagsp, /* inode logging flags */
753 int whichfork) /* data or attr fork */
755 struct xfs_btree_block *ablock; /* allocated (child) bt block */
756 xfs_buf_t *abp; /* buffer for ablock */
757 xfs_alloc_arg_t args; /* allocation arguments */
758 xfs_bmbt_rec_t *arp; /* child record pointer */
759 struct xfs_btree_block *block; /* btree root block */
760 xfs_btree_cur_t *cur; /* bmap btree cursor */
761 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
762 int error; /* error return value */
763 xfs_extnum_t i, cnt; /* extent record index */
764 xfs_ifork_t *ifp; /* inode fork pointer */
765 xfs_bmbt_key_t *kp; /* root block key pointer */
766 xfs_mount_t *mp; /* mount structure */
767 xfs_extnum_t nextents; /* number of file extents */
768 xfs_bmbt_ptr_t *pp; /* root block address pointer */
771 ifp = XFS_IFORK_PTR(ip, whichfork);
772 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
775 * Make space in the inode incore.
777 xfs_iroot_realloc(ip, 1, whichfork);
778 ifp->if_flags |= XFS_IFBROOT;
783 block = ifp->if_broot;
784 if (xfs_sb_version_hascrc(&mp->m_sb))
785 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
786 XFS_BMAP_CRC_MAGIC, 1, 1, ip->i_ino,
787 XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
789 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
790 XFS_BMAP_MAGIC, 1, 1, ip->i_ino,
791 XFS_BTREE_LONG_PTRS);
794 * Need a cursor. Can't allocate until bb_level is filled in.
796 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
797 cur->bc_private.b.firstblock = *firstblock;
798 cur->bc_private.b.flist = flist;
799 cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
801 * Convert to a btree with two levels, one record in root.
803 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
804 memset(&args, 0, sizeof(args));
807 args.firstblock = *firstblock;
808 if (*firstblock == NULLFSBLOCK) {
809 args.type = XFS_ALLOCTYPE_START_BNO;
810 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
811 } else if (flist->xbf_low) {
812 args.type = XFS_ALLOCTYPE_START_BNO;
813 args.fsbno = *firstblock;
815 args.type = XFS_ALLOCTYPE_NEAR_BNO;
816 args.fsbno = *firstblock;
818 args.minlen = args.maxlen = args.prod = 1;
819 args.wasdel = wasdel;
821 if ((error = xfs_alloc_vextent(&args))) {
822 xfs_iroot_realloc(ip, -1, whichfork);
823 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
827 * Allocation can't fail, the space was reserved.
829 ASSERT(args.fsbno != NULLFSBLOCK);
830 ASSERT(*firstblock == NULLFSBLOCK ||
831 args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
833 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
834 *firstblock = cur->bc_private.b.firstblock = args.fsbno;
835 cur->bc_private.b.allocated++;
836 ip->i_d.di_nblocks++;
837 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
838 abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
840 * Fill in the child block.
842 abp->b_ops = &xfs_bmbt_buf_ops;
843 ablock = XFS_BUF_TO_BLOCK(abp);
844 if (xfs_sb_version_hascrc(&mp->m_sb))
845 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
846 XFS_BMAP_CRC_MAGIC, 0, 0, ip->i_ino,
847 XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
849 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
850 XFS_BMAP_MAGIC, 0, 0, ip->i_ino,
851 XFS_BTREE_LONG_PTRS);
853 arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
854 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
855 for (cnt = i = 0; i < nextents; i++) {
856 ep = xfs_iext_get_ext(ifp, i);
857 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
858 arp->l0 = cpu_to_be64(ep->l0);
859 arp->l1 = cpu_to_be64(ep->l1);
863 ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
864 xfs_btree_set_numrecs(ablock, cnt);
867 * Fill in the root key and pointer.
869 kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
870 arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
871 kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
872 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
873 be16_to_cpu(block->bb_level)));
874 *pp = cpu_to_be64(args.fsbno);
877 * Do all this logging at the end so that
878 * the root is at the right level.
880 xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
881 xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
882 ASSERT(*curp == NULL);
884 *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
889 * Convert a local file to an extents file.
890 * This code is out of bounds for data forks of regular files,
891 * since the file data needs to get logged so things will stay consistent.
892 * (The bmap-level manipulations are ok, though).
895 xfs_bmap_local_to_extents_empty(
896 struct xfs_inode *ip,
899 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
901 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
902 ASSERT(ifp->if_bytes == 0);
903 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
905 xfs_bmap_forkoff_reset(ip, whichfork);
906 ifp->if_flags &= ~XFS_IFINLINE;
907 ifp->if_flags |= XFS_IFEXTENTS;
908 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
912 STATIC int /* error */
913 xfs_bmap_local_to_extents(
914 xfs_trans_t *tp, /* transaction pointer */
915 xfs_inode_t *ip, /* incore inode pointer */
916 xfs_fsblock_t *firstblock, /* first block allocated in xaction */
917 xfs_extlen_t total, /* total blocks needed by transaction */
918 int *logflagsp, /* inode logging flags */
920 void (*init_fn)(struct xfs_trans *tp,
922 struct xfs_inode *ip,
923 struct xfs_ifork *ifp))
926 int flags; /* logging flags returned */
927 xfs_ifork_t *ifp; /* inode fork pointer */
928 xfs_alloc_arg_t args; /* allocation arguments */
929 xfs_buf_t *bp; /* buffer for extent block */
930 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
933 * We don't want to deal with the case of keeping inode data inline yet.
934 * So sending the data fork of a regular inode is invalid.
936 ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
937 ifp = XFS_IFORK_PTR(ip, whichfork);
938 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
940 if (!ifp->if_bytes) {
941 xfs_bmap_local_to_extents_empty(ip, whichfork);
942 flags = XFS_ILOG_CORE;
948 ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) ==
950 memset(&args, 0, sizeof(args));
952 args.mp = ip->i_mount;
953 args.firstblock = *firstblock;
955 * Allocate a block. We know we need only one, since the
956 * file currently fits in an inode.
958 if (*firstblock == NULLFSBLOCK) {
959 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
960 args.type = XFS_ALLOCTYPE_START_BNO;
962 args.fsbno = *firstblock;
963 args.type = XFS_ALLOCTYPE_NEAR_BNO;
966 args.minlen = args.maxlen = args.prod = 1;
967 error = xfs_alloc_vextent(&args);
971 /* Can't fail, the space was reserved. */
972 ASSERT(args.fsbno != NULLFSBLOCK);
973 ASSERT(args.len == 1);
974 *firstblock = args.fsbno;
975 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
978 * Initialise the block and copy the data
980 * Note: init_fn must set the buffer log item type correctly!
982 init_fn(tp, bp, ip, ifp);
984 /* account for the change in fork size and log everything */
985 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
986 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
987 xfs_bmap_local_to_extents_empty(ip, whichfork);
988 flags |= XFS_ILOG_CORE;
990 xfs_iext_add(ifp, 0, 1);
991 ep = xfs_iext_get_ext(ifp, 0);
992 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
993 trace_xfs_bmap_post_update(ip, 0,
994 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
996 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
997 ip->i_d.di_nblocks = 1;
998 xfs_trans_mod_dquot_byino(tp, ip,
999 XFS_TRANS_DQ_BCOUNT, 1L);
1000 flags |= xfs_ilog_fext(whichfork);
1008 * Called from xfs_bmap_add_attrfork to handle btree format files.
1010 STATIC int /* error */
1011 xfs_bmap_add_attrfork_btree(
1012 xfs_trans_t *tp, /* transaction pointer */
1013 xfs_inode_t *ip, /* incore inode pointer */
1014 xfs_fsblock_t *firstblock, /* first block allocated */
1015 xfs_bmap_free_t *flist, /* blocks to free at commit */
1016 int *flags) /* inode logging flags */
1018 xfs_btree_cur_t *cur; /* btree cursor */
1019 int error; /* error return value */
1020 xfs_mount_t *mp; /* file system mount struct */
1021 int stat; /* newroot status */
1024 if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
1025 *flags |= XFS_ILOG_DBROOT;
1027 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
1028 cur->bc_private.b.flist = flist;
1029 cur->bc_private.b.firstblock = *firstblock;
1030 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
1032 /* must be at least one entry */
1033 XFS_WANT_CORRUPTED_GOTO(mp, stat == 1, error0);
1034 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
1037 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1040 *firstblock = cur->bc_private.b.firstblock;
1041 cur->bc_private.b.allocated = 0;
1042 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1046 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1051 * Called from xfs_bmap_add_attrfork to handle extents format files.
1053 STATIC int /* error */
1054 xfs_bmap_add_attrfork_extents(
1055 xfs_trans_t *tp, /* transaction pointer */
1056 xfs_inode_t *ip, /* incore inode pointer */
1057 xfs_fsblock_t *firstblock, /* first block allocated */
1058 xfs_bmap_free_t *flist, /* blocks to free at commit */
1059 int *flags) /* inode logging flags */
1061 xfs_btree_cur_t *cur; /* bmap btree cursor */
1062 int error; /* error return value */
1064 if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
1067 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
1068 flags, XFS_DATA_FORK);
1070 cur->bc_private.b.allocated = 0;
1071 xfs_btree_del_cursor(cur,
1072 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
1078 * Called from xfs_bmap_add_attrfork to handle local format files. Each
1079 * different data fork content type needs a different callout to do the
1080 * conversion. Some are basic and only require special block initialisation
1081 * callouts for the data formating, others (directories) are so specialised they
1082 * handle everything themselves.
1084 * XXX (dgc): investigate whether directory conversion can use the generic
1085 * formatting callout. It should be possible - it's just a very complex
1088 STATIC int /* error */
1089 xfs_bmap_add_attrfork_local(
1090 xfs_trans_t *tp, /* transaction pointer */
1091 xfs_inode_t *ip, /* incore inode pointer */
1092 xfs_fsblock_t *firstblock, /* first block allocated */
1093 xfs_bmap_free_t *flist, /* blocks to free at commit */
1094 int *flags) /* inode logging flags */
1096 xfs_da_args_t dargs; /* args for dir/attr code */
1098 if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1101 if (S_ISDIR(ip->i_d.di_mode)) {
1102 memset(&dargs, 0, sizeof(dargs));
1103 dargs.geo = ip->i_mount->m_dir_geo;
1105 dargs.firstblock = firstblock;
1106 dargs.flist = flist;
1107 dargs.total = dargs.geo->fsbcount;
1108 dargs.whichfork = XFS_DATA_FORK;
1110 return xfs_dir2_sf_to_block(&dargs);
1113 if (S_ISLNK(ip->i_d.di_mode))
1114 return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
1115 flags, XFS_DATA_FORK,
1116 xfs_symlink_local_to_remote);
1118 /* should only be called for types that support local format data */
1120 return -EFSCORRUPTED;
1124 * Convert inode from non-attributed to attributed.
1125 * Must not be in a transaction, ip must not be locked.
1127 int /* error code */
1128 xfs_bmap_add_attrfork(
1129 xfs_inode_t *ip, /* incore inode pointer */
1130 int size, /* space new attribute needs */
1131 int rsvd) /* xact may use reserved blks */
1133 xfs_fsblock_t firstblock; /* 1st block/ag allocated */
1134 xfs_bmap_free_t flist; /* freed extent records */
1135 xfs_mount_t *mp; /* mount structure */
1136 xfs_trans_t *tp; /* transaction pointer */
1137 int blks; /* space reservation */
1138 int version = 1; /* superblock attr version */
1139 int committed; /* xaction was committed */
1140 int logflags; /* logging flags */
1141 int error; /* error return value */
1142 int cancel_flags = 0;
1144 ASSERT(XFS_IFORK_Q(ip) == 0);
1147 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1148 tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
1149 blks = XFS_ADDAFORK_SPACE_RES(mp);
1151 tp->t_flags |= XFS_TRANS_RESERVE;
1152 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_addafork, blks, 0);
1154 xfs_trans_cancel(tp, 0);
1157 cancel_flags = XFS_TRANS_RELEASE_LOG_RES;
1158 xfs_ilock(ip, XFS_ILOCK_EXCL);
1159 error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1160 XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1161 XFS_QMOPT_RES_REGBLKS);
1164 cancel_flags |= XFS_TRANS_ABORT;
1165 if (XFS_IFORK_Q(ip))
1167 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1169 * For inodes coming from pre-6.2 filesystems.
1171 ASSERT(ip->i_d.di_aformat == 0);
1172 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1174 ASSERT(ip->i_d.di_anextents == 0);
1176 xfs_trans_ijoin(tp, ip, 0);
1177 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1179 switch (ip->i_d.di_format) {
1180 case XFS_DINODE_FMT_DEV:
1181 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1183 case XFS_DINODE_FMT_UUID:
1184 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
1186 case XFS_DINODE_FMT_LOCAL:
1187 case XFS_DINODE_FMT_EXTENTS:
1188 case XFS_DINODE_FMT_BTREE:
1189 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1190 if (!ip->i_d.di_forkoff)
1191 ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1192 else if (mp->m_flags & XFS_MOUNT_ATTR2)
1201 ASSERT(ip->i_afp == NULL);
1202 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1203 ip->i_afp->if_flags = XFS_IFEXTENTS;
1205 xfs_bmap_init(&flist, &firstblock);
1206 switch (ip->i_d.di_format) {
1207 case XFS_DINODE_FMT_LOCAL:
1208 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
1211 case XFS_DINODE_FMT_EXTENTS:
1212 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1215 case XFS_DINODE_FMT_BTREE:
1216 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
1224 xfs_trans_log_inode(tp, ip, logflags);
1227 if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1228 (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1229 bool log_sb = false;
1231 spin_lock(&mp->m_sb_lock);
1232 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1233 xfs_sb_version_addattr(&mp->m_sb);
1236 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1237 xfs_sb_version_addattr2(&mp->m_sb);
1240 spin_unlock(&mp->m_sb_lock);
1245 error = xfs_bmap_finish(&tp, &flist, &committed);
1248 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1249 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1253 xfs_bmap_cancel(&flist);
1255 xfs_trans_cancel(tp, cancel_flags);
1256 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1261 * Internal and external extent tree search functions.
1265 * Read in the extents to if_extents.
1266 * All inode fields are set up by caller, we just traverse the btree
1267 * and copy the records in. If the file system cannot contain unwritten
1268 * extents, the records are checked for no "state" flags.
1271 xfs_bmap_read_extents(
1272 xfs_trans_t *tp, /* transaction pointer */
1273 xfs_inode_t *ip, /* incore inode */
1274 int whichfork) /* data or attr fork */
1276 struct xfs_btree_block *block; /* current btree block */
1277 xfs_fsblock_t bno; /* block # of "block" */
1278 xfs_buf_t *bp; /* buffer for "block" */
1279 int error; /* error return value */
1280 xfs_exntfmt_t exntf; /* XFS_EXTFMT_NOSTATE, if checking */
1281 xfs_extnum_t i, j; /* index into the extents list */
1282 xfs_ifork_t *ifp; /* fork structure */
1283 int level; /* btree level, for checking */
1284 xfs_mount_t *mp; /* file system mount structure */
1285 __be64 *pp; /* pointer to block address */
1287 xfs_extnum_t room; /* number of entries there's room for */
1291 ifp = XFS_IFORK_PTR(ip, whichfork);
1292 exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
1293 XFS_EXTFMT_INODE(ip);
1294 block = ifp->if_broot;
1296 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1298 level = be16_to_cpu(block->bb_level);
1300 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1301 bno = be64_to_cpu(*pp);
1302 ASSERT(bno != NULLFSBLOCK);
1303 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
1304 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
1306 * Go down the tree until leaf level is reached, following the first
1307 * pointer (leftmost) at each level.
1309 while (level-- > 0) {
1310 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1311 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1314 block = XFS_BUF_TO_BLOCK(bp);
1315 XFS_WANT_CORRUPTED_GOTO(mp,
1316 xfs_bmap_sanity_check(mp, bp, level), error0);
1319 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1320 bno = be64_to_cpu(*pp);
1321 XFS_WANT_CORRUPTED_GOTO(mp,
1322 XFS_FSB_SANITY_CHECK(mp, bno), error0);
1323 xfs_trans_brelse(tp, bp);
1326 * Here with bp and block set to the leftmost leaf node in the tree.
1328 room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1331 * Loop over all leaf nodes. Copy information to the extent records.
1334 xfs_bmbt_rec_t *frp;
1335 xfs_fsblock_t nextbno;
1336 xfs_extnum_t num_recs;
1339 num_recs = xfs_btree_get_numrecs(block);
1340 if (unlikely(i + num_recs > room)) {
1341 ASSERT(i + num_recs <= room);
1342 xfs_warn(ip->i_mount,
1343 "corrupt dinode %Lu, (btree extents).",
1344 (unsigned long long) ip->i_ino);
1345 XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1346 XFS_ERRLEVEL_LOW, ip->i_mount, block);
1349 XFS_WANT_CORRUPTED_GOTO(mp,
1350 xfs_bmap_sanity_check(mp, bp, 0),
1353 * Read-ahead the next leaf block, if any.
1355 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1356 if (nextbno != NULLFSBLOCK)
1357 xfs_btree_reada_bufl(mp, nextbno, 1,
1360 * Copy records into the extent records.
1362 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1364 for (j = 0; j < num_recs; j++, i++, frp++) {
1365 xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
1366 trp->l0 = be64_to_cpu(frp->l0);
1367 trp->l1 = be64_to_cpu(frp->l1);
1369 if (exntf == XFS_EXTFMT_NOSTATE) {
1371 * Check all attribute bmap btree records and
1372 * any "older" data bmap btree records for a
1373 * set bit in the "extent flag" position.
1375 if (unlikely(xfs_check_nostate_extents(ifp,
1376 start, num_recs))) {
1377 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1383 xfs_trans_brelse(tp, bp);
1386 * If we've reached the end, stop.
1388 if (bno == NULLFSBLOCK)
1390 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1391 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1394 block = XFS_BUF_TO_BLOCK(bp);
1396 ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
1397 ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
1398 XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
1401 xfs_trans_brelse(tp, bp);
1402 return -EFSCORRUPTED;
1407 * Search the extent records for the entry containing block bno.
1408 * If bno lies in a hole, point to the next entry. If bno lies
1409 * past eof, *eofp will be set, and *prevp will contain the last
1410 * entry (null if none). Else, *lastxp will be set to the index
1411 * of the found entry; *gotp will contain the entry.
1413 STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
1414 xfs_bmap_search_multi_extents(
1415 xfs_ifork_t *ifp, /* inode fork pointer */
1416 xfs_fileoff_t bno, /* block number searched for */
1417 int *eofp, /* out: end of file found */
1418 xfs_extnum_t *lastxp, /* out: last extent index */
1419 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
1420 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
1422 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
1423 xfs_extnum_t lastx; /* last extent index */
1426 * Initialize the extent entry structure to catch access to
1427 * uninitialized br_startblock field.
1429 gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
1430 gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
1431 gotp->br_state = XFS_EXT_INVALID;
1432 gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
1433 prevp->br_startoff = NULLFILEOFF;
1435 ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
1437 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
1439 if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1440 xfs_bmbt_get_all(ep, gotp);
1454 * Search the extents list for the inode, for the extent containing bno.
1455 * If bno lies in a hole, point to the next entry. If bno lies past eof,
1456 * *eofp will be set, and *prevp will contain the last entry (null if none).
1457 * Else, *lastxp will be set to the index of the found
1458 * entry; *gotp will contain the entry.
1460 STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
1461 xfs_bmap_search_extents(
1462 xfs_inode_t *ip, /* incore inode pointer */
1463 xfs_fileoff_t bno, /* block number searched for */
1464 int fork, /* data or attr fork */
1465 int *eofp, /* out: end of file found */
1466 xfs_extnum_t *lastxp, /* out: last extent index */
1467 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
1468 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
1470 xfs_ifork_t *ifp; /* inode fork pointer */
1471 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
1473 XFS_STATS_INC(xs_look_exlist);
1474 ifp = XFS_IFORK_PTR(ip, fork);
1476 ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
1478 if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
1479 !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
1480 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
1481 "Access to block zero in inode %llu "
1482 "start_block: %llx start_off: %llx "
1483 "blkcnt: %llx extent-state: %x lastx: %x",
1484 (unsigned long long)ip->i_ino,
1485 (unsigned long long)gotp->br_startblock,
1486 (unsigned long long)gotp->br_startoff,
1487 (unsigned long long)gotp->br_blockcount,
1488 gotp->br_state, *lastxp);
1489 *lastxp = NULLEXTNUM;
1497 * Returns the file-relative block number of the first unused block(s)
1498 * in the file with at least "len" logically contiguous blocks free.
1499 * This is the lowest-address hole if the file has holes, else the first block
1500 * past the end of file.
1501 * Return 0 if the file is currently local (in-inode).
1504 xfs_bmap_first_unused(
1505 xfs_trans_t *tp, /* transaction pointer */
1506 xfs_inode_t *ip, /* incore inode */
1507 xfs_extlen_t len, /* size of hole to find */
1508 xfs_fileoff_t *first_unused, /* unused block */
1509 int whichfork) /* data or attr fork */
1511 int error; /* error return value */
1512 int idx; /* extent record index */
1513 xfs_ifork_t *ifp; /* inode fork pointer */
1514 xfs_fileoff_t lastaddr; /* last block number seen */
1515 xfs_fileoff_t lowest; /* lowest useful block */
1516 xfs_fileoff_t max; /* starting useful block */
1517 xfs_fileoff_t off; /* offset for this block */
1518 xfs_extnum_t nextents; /* number of extent entries */
1520 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1521 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1522 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1523 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1527 ifp = XFS_IFORK_PTR(ip, whichfork);
1528 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1529 (error = xfs_iread_extents(tp, ip, whichfork)))
1531 lowest = *first_unused;
1532 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1533 for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
1534 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
1535 off = xfs_bmbt_get_startoff(ep);
1537 * See if the hole before this extent will work.
1539 if (off >= lowest + len && off - max >= len) {
1540 *first_unused = max;
1543 lastaddr = off + xfs_bmbt_get_blockcount(ep);
1544 max = XFS_FILEOFF_MAX(lastaddr, lowest);
1546 *first_unused = max;
1551 * Returns the file-relative block number of the last block - 1 before
1552 * last_block (input value) in the file.
1553 * This is not based on i_size, it is based on the extent records.
1554 * Returns 0 for local files, as they do not have extent records.
1557 xfs_bmap_last_before(
1558 xfs_trans_t *tp, /* transaction pointer */
1559 xfs_inode_t *ip, /* incore inode */
1560 xfs_fileoff_t *last_block, /* last block */
1561 int whichfork) /* data or attr fork */
1563 xfs_fileoff_t bno; /* input file offset */
1564 int eof; /* hit end of file */
1565 xfs_bmbt_rec_host_t *ep; /* pointer to last extent */
1566 int error; /* error return value */
1567 xfs_bmbt_irec_t got; /* current extent value */
1568 xfs_ifork_t *ifp; /* inode fork pointer */
1569 xfs_extnum_t lastx; /* last extent used */
1570 xfs_bmbt_irec_t prev; /* previous extent value */
1572 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1573 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
1574 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
1576 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1580 ifp = XFS_IFORK_PTR(ip, whichfork);
1581 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1582 (error = xfs_iread_extents(tp, ip, whichfork)))
1584 bno = *last_block - 1;
1585 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
1587 if (eof || xfs_bmbt_get_startoff(ep) > bno) {
1588 if (prev.br_startoff == NULLFILEOFF)
1591 *last_block = prev.br_startoff + prev.br_blockcount;
1594 * Otherwise *last_block is already the right answer.
1600 xfs_bmap_last_extent(
1601 struct xfs_trans *tp,
1602 struct xfs_inode *ip,
1604 struct xfs_bmbt_irec *rec,
1607 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
1611 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1612 error = xfs_iread_extents(tp, ip, whichfork);
1617 nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
1618 if (nextents == 0) {
1623 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
1629 * Check the last inode extent to determine whether this allocation will result
1630 * in blocks being allocated at the end of the file. When we allocate new data
1631 * blocks at the end of the file which do not start at the previous data block,
1632 * we will try to align the new blocks at stripe unit boundaries.
1634 * Returns 1 in bma->aeof if the file (fork) is empty as any new write will be
1635 * at, or past the EOF.
1639 struct xfs_bmalloca *bma,
1642 struct xfs_bmbt_irec rec;
1647 error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1658 * Check if we are allocation or past the last extent, or at least into
1659 * the last delayed allocated extent.
1661 bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1662 (bma->offset >= rec.br_startoff &&
1663 isnullstartblock(rec.br_startblock));
1668 * Returns the file-relative block number of the first block past eof in
1669 * the file. This is not based on i_size, it is based on the extent records.
1670 * Returns 0 for local files, as they do not have extent records.
1673 xfs_bmap_last_offset(
1674 struct xfs_inode *ip,
1675 xfs_fileoff_t *last_block,
1678 struct xfs_bmbt_irec rec;
1684 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1687 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1688 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1691 error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1692 if (error || is_empty)
1695 *last_block = rec.br_startoff + rec.br_blockcount;
1700 * Returns whether the selected fork of the inode has exactly one
1701 * block or not. For the data fork we check this matches di_size,
1702 * implying the file's range is 0..bsize-1.
1704 int /* 1=>1 block, 0=>otherwise */
1706 xfs_inode_t *ip, /* incore inode */
1707 int whichfork) /* data or attr fork */
1709 xfs_bmbt_rec_host_t *ep; /* ptr to fork's extent */
1710 xfs_ifork_t *ifp; /* inode fork pointer */
1711 int rval; /* return value */
1712 xfs_bmbt_irec_t s; /* internal version of extent */
1715 if (whichfork == XFS_DATA_FORK)
1716 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1718 if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1720 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1722 ifp = XFS_IFORK_PTR(ip, whichfork);
1723 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1724 ep = xfs_iext_get_ext(ifp, 0);
1725 xfs_bmbt_get_all(ep, &s);
1726 rval = s.br_startoff == 0 && s.br_blockcount == 1;
1727 if (rval && whichfork == XFS_DATA_FORK)
1728 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1733 * Extent tree manipulation functions used during allocation.
1737 * Convert a delayed allocation to a real allocation.
1739 STATIC int /* error */
1740 xfs_bmap_add_extent_delay_real(
1741 struct xfs_bmalloca *bma)
1743 struct xfs_bmbt_irec *new = &bma->got;
1744 int diff; /* temp value */
1745 xfs_bmbt_rec_host_t *ep; /* extent entry for idx */
1746 int error; /* error return value */
1747 int i; /* temp state */
1748 xfs_ifork_t *ifp; /* inode fork pointer */
1749 xfs_fileoff_t new_endoff; /* end offset of new entry */
1750 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
1751 /* left is 0, right is 1, prev is 2 */
1752 int rval=0; /* return value (logging flags) */
1753 int state = 0;/* state bits, accessed thru macros */
1754 xfs_filblks_t da_new; /* new count del alloc blocks used */
1755 xfs_filblks_t da_old; /* old count del alloc blocks used */
1756 xfs_filblks_t temp=0; /* value for da_new calculations */
1757 xfs_filblks_t temp2=0;/* value for da_new calculations */
1758 int tmp_rval; /* partial logging flags */
1759 struct xfs_mount *mp;
1761 mp = bma->tp ? bma->tp->t_mountp : NULL;
1762 ifp = XFS_IFORK_PTR(bma->ip, XFS_DATA_FORK);
1764 ASSERT(bma->idx >= 0);
1765 ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1766 ASSERT(!isnullstartblock(new->br_startblock));
1768 (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1770 XFS_STATS_INC(xs_add_exlist);
1777 * Set up a bunch of variables to make the tests simpler.
1779 ep = xfs_iext_get_ext(ifp, bma->idx);
1780 xfs_bmbt_get_all(ep, &PREV);
1781 new_endoff = new->br_startoff + new->br_blockcount;
1782 ASSERT(PREV.br_startoff <= new->br_startoff);
1783 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1785 da_old = startblockval(PREV.br_startblock);
1789 * Set flags determining what part of the previous delayed allocation
1790 * extent is being replaced by a real allocation.
1792 if (PREV.br_startoff == new->br_startoff)
1793 state |= BMAP_LEFT_FILLING;
1794 if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1795 state |= BMAP_RIGHT_FILLING;
1798 * Check and set flags if this segment has a left neighbor.
1799 * Don't set contiguous if the combined extent would be too large.
1802 state |= BMAP_LEFT_VALID;
1803 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
1805 if (isnullstartblock(LEFT.br_startblock))
1806 state |= BMAP_LEFT_DELAY;
1809 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1810 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1811 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1812 LEFT.br_state == new->br_state &&
1813 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1814 state |= BMAP_LEFT_CONTIG;
1817 * Check and set flags if this segment has a right neighbor.
1818 * Don't set contiguous if the combined extent would be too large.
1819 * Also check for all-three-contiguous being too large.
1821 if (bma->idx < bma->ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1822 state |= BMAP_RIGHT_VALID;
1823 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
1825 if (isnullstartblock(RIGHT.br_startblock))
1826 state |= BMAP_RIGHT_DELAY;
1829 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1830 new_endoff == RIGHT.br_startoff &&
1831 new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1832 new->br_state == RIGHT.br_state &&
1833 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1834 ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1835 BMAP_RIGHT_FILLING)) !=
1836 (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1837 BMAP_RIGHT_FILLING) ||
1838 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1840 state |= BMAP_RIGHT_CONTIG;
1844 * Switch out based on the FILLING and CONTIG state bits.
1846 switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1847 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1848 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1849 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1851 * Filling in all of a previously delayed allocation extent.
1852 * The left and right neighbors are both contiguous with new.
1855 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1856 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1857 LEFT.br_blockcount + PREV.br_blockcount +
1858 RIGHT.br_blockcount);
1859 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1861 xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
1862 bma->ip->i_d.di_nextents--;
1863 if (bma->cur == NULL)
1864 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1866 rval = XFS_ILOG_CORE;
1867 error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1868 RIGHT.br_startblock,
1869 RIGHT.br_blockcount, &i);
1872 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1873 error = xfs_btree_delete(bma->cur, &i);
1876 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1877 error = xfs_btree_decrement(bma->cur, 0, &i);
1880 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1881 error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1883 LEFT.br_blockcount +
1884 PREV.br_blockcount +
1885 RIGHT.br_blockcount, LEFT.br_state);
1891 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1893 * Filling in all of a previously delayed allocation extent.
1894 * The left neighbor is contiguous, the right is not.
1898 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1899 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1900 LEFT.br_blockcount + PREV.br_blockcount);
1901 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1903 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1904 if (bma->cur == NULL)
1905 rval = XFS_ILOG_DEXT;
1908 error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1909 LEFT.br_startblock, LEFT.br_blockcount,
1913 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1914 error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1916 LEFT.br_blockcount +
1917 PREV.br_blockcount, LEFT.br_state);
1923 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1925 * Filling in all of a previously delayed allocation extent.
1926 * The right neighbor is contiguous, the left is not.
1928 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1929 xfs_bmbt_set_startblock(ep, new->br_startblock);
1930 xfs_bmbt_set_blockcount(ep,
1931 PREV.br_blockcount + RIGHT.br_blockcount);
1932 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1934 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1935 if (bma->cur == NULL)
1936 rval = XFS_ILOG_DEXT;
1939 error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1940 RIGHT.br_startblock,
1941 RIGHT.br_blockcount, &i);
1944 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1945 error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
1947 PREV.br_blockcount +
1948 RIGHT.br_blockcount, PREV.br_state);
1954 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1956 * Filling in all of a previously delayed allocation extent.
1957 * Neither the left nor right neighbors are contiguous with
1960 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1961 xfs_bmbt_set_startblock(ep, new->br_startblock);
1962 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1964 bma->ip->i_d.di_nextents++;
1965 if (bma->cur == NULL)
1966 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1968 rval = XFS_ILOG_CORE;
1969 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1970 new->br_startblock, new->br_blockcount,
1974 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
1975 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1976 error = xfs_btree_insert(bma->cur, &i);
1979 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
1983 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1985 * Filling in the first part of a previous delayed allocation.
1986 * The left neighbor is contiguous.
1988 trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1989 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
1990 LEFT.br_blockcount + new->br_blockcount);
1991 xfs_bmbt_set_startoff(ep,
1992 PREV.br_startoff + new->br_blockcount);
1993 trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1995 temp = PREV.br_blockcount - new->br_blockcount;
1996 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1997 xfs_bmbt_set_blockcount(ep, temp);
1998 if (bma->cur == NULL)
1999 rval = XFS_ILOG_DEXT;
2002 error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
2003 LEFT.br_startblock, LEFT.br_blockcount,
2007 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2008 error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
2010 LEFT.br_blockcount +
2016 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2017 startblockval(PREV.br_startblock));
2018 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2019 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2024 case BMAP_LEFT_FILLING:
2026 * Filling in the first part of a previous delayed allocation.
2027 * The left neighbor is not contiguous.
2029 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2030 xfs_bmbt_set_startoff(ep, new_endoff);
2031 temp = PREV.br_blockcount - new->br_blockcount;
2032 xfs_bmbt_set_blockcount(ep, temp);
2033 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
2034 bma->ip->i_d.di_nextents++;
2035 if (bma->cur == NULL)
2036 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2038 rval = XFS_ILOG_CORE;
2039 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2040 new->br_startblock, new->br_blockcount,
2044 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2045 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2046 error = xfs_btree_insert(bma->cur, &i);
2049 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2052 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2053 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2054 bma->firstblock, bma->flist,
2055 &bma->cur, 1, &tmp_rval, XFS_DATA_FORK);
2060 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2061 startblockval(PREV.br_startblock) -
2062 (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2063 ep = xfs_iext_get_ext(ifp, bma->idx + 1);
2064 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2065 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2068 case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2070 * Filling in the last part of a previous delayed allocation.
2071 * The right neighbor is contiguous with the new allocation.
2073 temp = PREV.br_blockcount - new->br_blockcount;
2074 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2075 xfs_bmbt_set_blockcount(ep, temp);
2076 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
2077 new->br_startoff, new->br_startblock,
2078 new->br_blockcount + RIGHT.br_blockcount,
2080 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2081 if (bma->cur == NULL)
2082 rval = XFS_ILOG_DEXT;
2085 error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
2086 RIGHT.br_startblock,
2087 RIGHT.br_blockcount, &i);
2090 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2091 error = xfs_bmbt_update(bma->cur, new->br_startoff,
2093 new->br_blockcount +
2094 RIGHT.br_blockcount,
2100 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2101 startblockval(PREV.br_startblock));
2102 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2103 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2104 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2109 case BMAP_RIGHT_FILLING:
2111 * Filling in the last part of a previous delayed allocation.
2112 * The right neighbor is not contiguous.
2114 temp = PREV.br_blockcount - new->br_blockcount;
2115 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2116 xfs_bmbt_set_blockcount(ep, temp);
2117 xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
2118 bma->ip->i_d.di_nextents++;
2119 if (bma->cur == NULL)
2120 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2122 rval = XFS_ILOG_CORE;
2123 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2124 new->br_startblock, new->br_blockcount,
2128 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2129 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2130 error = xfs_btree_insert(bma->cur, &i);
2133 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2136 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2137 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2138 bma->firstblock, bma->flist, &bma->cur, 1,
2139 &tmp_rval, XFS_DATA_FORK);
2144 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2145 startblockval(PREV.br_startblock) -
2146 (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2147 ep = xfs_iext_get_ext(ifp, bma->idx);
2148 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2149 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2156 * Filling in the middle part of a previous delayed allocation.
2157 * Contiguity is impossible here.
2158 * This case is avoided almost all the time.
2160 * We start with a delayed allocation:
2162 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2165 * and we are allocating:
2166 * +rrrrrrrrrrrrrrrrr+
2169 * and we set it up for insertion as:
2170 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2172 * PREV @ idx LEFT RIGHT
2173 * inserted at idx + 1
2175 temp = new->br_startoff - PREV.br_startoff;
2176 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
2177 trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
2178 xfs_bmbt_set_blockcount(ep, temp); /* truncate PREV */
2180 RIGHT.br_state = PREV.br_state;
2181 RIGHT.br_startblock = nullstartblock(
2182 (int)xfs_bmap_worst_indlen(bma->ip, temp2));
2183 RIGHT.br_startoff = new_endoff;
2184 RIGHT.br_blockcount = temp2;
2185 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2186 xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
2187 bma->ip->i_d.di_nextents++;
2188 if (bma->cur == NULL)
2189 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2191 rval = XFS_ILOG_CORE;
2192 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2193 new->br_startblock, new->br_blockcount,
2197 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2198 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2199 error = xfs_btree_insert(bma->cur, &i);
2202 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2205 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2206 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2207 bma->firstblock, bma->flist, &bma->cur,
2208 1, &tmp_rval, XFS_DATA_FORK);
2213 temp = xfs_bmap_worst_indlen(bma->ip, temp);
2214 temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
2215 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
2216 (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2218 error = xfs_icsb_modify_counters(bma->ip->i_mount,
2220 -((int64_t)diff), 0);
2226 ep = xfs_iext_get_ext(ifp, bma->idx);
2227 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2228 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2229 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2230 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
2231 nullstartblock((int)temp2));
2232 trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2235 da_new = temp + temp2;
2238 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2239 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2240 case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2241 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2242 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2243 case BMAP_LEFT_CONTIG:
2244 case BMAP_RIGHT_CONTIG:
2246 * These cases are all impossible.
2251 /* convert to a btree if necessary */
2252 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2253 int tmp_logflags; /* partial log flag return val */
2255 ASSERT(bma->cur == NULL);
2256 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2257 bma->firstblock, bma->flist, &bma->cur,
2258 da_old > 0, &tmp_logflags, XFS_DATA_FORK);
2259 bma->logflags |= tmp_logflags;
2264 /* adjust for changes in reserved delayed indirect blocks */
2265 if (da_old || da_new) {
2268 temp += bma->cur->bc_private.b.allocated;
2269 ASSERT(temp <= da_old);
2271 xfs_icsb_modify_counters(bma->ip->i_mount,
2273 (int64_t)(da_old - temp), 0);
2276 /* clear out the allocated field, done with it now in any case. */
2278 bma->cur->bc_private.b.allocated = 0;
2280 xfs_bmap_check_leaf_extents(bma->cur, bma->ip, XFS_DATA_FORK);
2282 bma->logflags |= rval;
2290 * Convert an unwritten allocation to a real allocation or vice versa.
2292 STATIC int /* error */
2293 xfs_bmap_add_extent_unwritten_real(
2294 struct xfs_trans *tp,
2295 xfs_inode_t *ip, /* incore inode pointer */
2296 xfs_extnum_t *idx, /* extent number to update/insert */
2297 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
2298 xfs_bmbt_irec_t *new, /* new data to add to file extents */
2299 xfs_fsblock_t *first, /* pointer to firstblock variable */
2300 xfs_bmap_free_t *flist, /* list of extents to be freed */
2301 int *logflagsp) /* inode logging flags */
2303 xfs_btree_cur_t *cur; /* btree cursor */
2304 xfs_bmbt_rec_host_t *ep; /* extent entry for idx */
2305 int error; /* error return value */
2306 int i; /* temp state */
2307 xfs_ifork_t *ifp; /* inode fork pointer */
2308 xfs_fileoff_t new_endoff; /* end offset of new entry */
2309 xfs_exntst_t newext; /* new extent state */
2310 xfs_exntst_t oldext; /* old extent state */
2311 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
2312 /* left is 0, right is 1, prev is 2 */
2313 int rval=0; /* return value (logging flags) */
2314 int state = 0;/* state bits, accessed thru macros */
2315 struct xfs_mount *mp = tp->t_mountp;
2320 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2323 ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2324 ASSERT(!isnullstartblock(new->br_startblock));
2326 XFS_STATS_INC(xs_add_exlist);
2333 * Set up a bunch of variables to make the tests simpler.
2336 ep = xfs_iext_get_ext(ifp, *idx);
2337 xfs_bmbt_get_all(ep, &PREV);
2338 newext = new->br_state;
2339 oldext = (newext == XFS_EXT_UNWRITTEN) ?
2340 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
2341 ASSERT(PREV.br_state == oldext);
2342 new_endoff = new->br_startoff + new->br_blockcount;
2343 ASSERT(PREV.br_startoff <= new->br_startoff);
2344 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2347 * Set flags determining what part of the previous oldext allocation
2348 * extent is being replaced by a newext allocation.
2350 if (PREV.br_startoff == new->br_startoff)
2351 state |= BMAP_LEFT_FILLING;
2352 if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2353 state |= BMAP_RIGHT_FILLING;
2356 * Check and set flags if this segment has a left neighbor.
2357 * Don't set contiguous if the combined extent would be too large.
2360 state |= BMAP_LEFT_VALID;
2361 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
2363 if (isnullstartblock(LEFT.br_startblock))
2364 state |= BMAP_LEFT_DELAY;
2367 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2368 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2369 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2370 LEFT.br_state == newext &&
2371 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2372 state |= BMAP_LEFT_CONTIG;
2375 * Check and set flags if this segment has a right neighbor.
2376 * Don't set contiguous if the combined extent would be too large.
2377 * Also check for all-three-contiguous being too large.
2379 if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
2380 state |= BMAP_RIGHT_VALID;
2381 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
2382 if (isnullstartblock(RIGHT.br_startblock))
2383 state |= BMAP_RIGHT_DELAY;
2386 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2387 new_endoff == RIGHT.br_startoff &&
2388 new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2389 newext == RIGHT.br_state &&
2390 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2391 ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2392 BMAP_RIGHT_FILLING)) !=
2393 (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2394 BMAP_RIGHT_FILLING) ||
2395 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2397 state |= BMAP_RIGHT_CONTIG;
2400 * Switch out based on the FILLING and CONTIG state bits.
2402 switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2403 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2404 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2405 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2407 * Setting all of a previous oldext extent to newext.
2408 * The left and right neighbors are both contiguous with new.
2412 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2413 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2414 LEFT.br_blockcount + PREV.br_blockcount +
2415 RIGHT.br_blockcount);
2416 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2418 xfs_iext_remove(ip, *idx + 1, 2, state);
2419 ip->i_d.di_nextents -= 2;
2421 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2423 rval = XFS_ILOG_CORE;
2424 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2425 RIGHT.br_startblock,
2426 RIGHT.br_blockcount, &i)))
2428 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2429 if ((error = xfs_btree_delete(cur, &i)))
2431 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2432 if ((error = xfs_btree_decrement(cur, 0, &i)))
2434 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2435 if ((error = xfs_btree_delete(cur, &i)))
2437 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2438 if ((error = xfs_btree_decrement(cur, 0, &i)))
2440 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2441 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2443 LEFT.br_blockcount + PREV.br_blockcount +
2444 RIGHT.br_blockcount, LEFT.br_state)))
2449 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2451 * Setting all of a previous oldext extent to newext.
2452 * The left neighbor is contiguous, the right is not.
2456 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2457 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2458 LEFT.br_blockcount + PREV.br_blockcount);
2459 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2461 xfs_iext_remove(ip, *idx + 1, 1, state);
2462 ip->i_d.di_nextents--;
2464 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2466 rval = XFS_ILOG_CORE;
2467 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2468 PREV.br_startblock, PREV.br_blockcount,
2471 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2472 if ((error = xfs_btree_delete(cur, &i)))
2474 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2475 if ((error = xfs_btree_decrement(cur, 0, &i)))
2477 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2478 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2480 LEFT.br_blockcount + PREV.br_blockcount,
2486 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2488 * Setting all of a previous oldext extent to newext.
2489 * The right neighbor is contiguous, the left is not.
2491 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2492 xfs_bmbt_set_blockcount(ep,
2493 PREV.br_blockcount + RIGHT.br_blockcount);
2494 xfs_bmbt_set_state(ep, newext);
2495 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2496 xfs_iext_remove(ip, *idx + 1, 1, state);
2497 ip->i_d.di_nextents--;
2499 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2501 rval = XFS_ILOG_CORE;
2502 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2503 RIGHT.br_startblock,
2504 RIGHT.br_blockcount, &i)))
2506 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2507 if ((error = xfs_btree_delete(cur, &i)))
2509 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2510 if ((error = xfs_btree_decrement(cur, 0, &i)))
2512 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2513 if ((error = xfs_bmbt_update(cur, new->br_startoff,
2515 new->br_blockcount + RIGHT.br_blockcount,
2521 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2523 * Setting all of a previous oldext extent to newext.
2524 * Neither the left nor right neighbors are contiguous with
2527 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2528 xfs_bmbt_set_state(ep, newext);
2529 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2532 rval = XFS_ILOG_DEXT;
2535 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2536 new->br_startblock, new->br_blockcount,
2539 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2540 if ((error = xfs_bmbt_update(cur, new->br_startoff,
2541 new->br_startblock, new->br_blockcount,
2547 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2549 * Setting the first part of a previous oldext extent to newext.
2550 * The left neighbor is contiguous.
2552 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
2553 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
2554 LEFT.br_blockcount + new->br_blockcount);
2555 xfs_bmbt_set_startoff(ep,
2556 PREV.br_startoff + new->br_blockcount);
2557 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
2559 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2560 xfs_bmbt_set_startblock(ep,
2561 new->br_startblock + new->br_blockcount);
2562 xfs_bmbt_set_blockcount(ep,
2563 PREV.br_blockcount - new->br_blockcount);
2564 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2569 rval = XFS_ILOG_DEXT;
2572 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2573 PREV.br_startblock, PREV.br_blockcount,
2576 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2577 if ((error = xfs_bmbt_update(cur,
2578 PREV.br_startoff + new->br_blockcount,
2579 PREV.br_startblock + new->br_blockcount,
2580 PREV.br_blockcount - new->br_blockcount,
2583 if ((error = xfs_btree_decrement(cur, 0, &i)))
2585 error = xfs_bmbt_update(cur, LEFT.br_startoff,
2587 LEFT.br_blockcount + new->br_blockcount,
2594 case BMAP_LEFT_FILLING:
2596 * Setting the first part of a previous oldext extent to newext.
2597 * The left neighbor is not contiguous.
2599 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2600 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
2601 xfs_bmbt_set_startoff(ep, new_endoff);
2602 xfs_bmbt_set_blockcount(ep,
2603 PREV.br_blockcount - new->br_blockcount);
2604 xfs_bmbt_set_startblock(ep,
2605 new->br_startblock + new->br_blockcount);
2606 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2608 xfs_iext_insert(ip, *idx, 1, new, state);
2609 ip->i_d.di_nextents++;
2611 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2613 rval = XFS_ILOG_CORE;
2614 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2615 PREV.br_startblock, PREV.br_blockcount,
2618 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2619 if ((error = xfs_bmbt_update(cur,
2620 PREV.br_startoff + new->br_blockcount,
2621 PREV.br_startblock + new->br_blockcount,
2622 PREV.br_blockcount - new->br_blockcount,
2625 cur->bc_rec.b = *new;
2626 if ((error = xfs_btree_insert(cur, &i)))
2628 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2632 case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2634 * Setting the last part of a previous oldext extent to newext.
2635 * The right neighbor is contiguous with the new allocation.
2637 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2638 xfs_bmbt_set_blockcount(ep,
2639 PREV.br_blockcount - new->br_blockcount);
2640 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2644 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2645 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2646 new->br_startoff, new->br_startblock,
2647 new->br_blockcount + RIGHT.br_blockcount, newext);
2648 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2651 rval = XFS_ILOG_DEXT;
2654 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2656 PREV.br_blockcount, &i)))
2658 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2659 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2661 PREV.br_blockcount - new->br_blockcount,
2664 if ((error = xfs_btree_increment(cur, 0, &i)))
2666 if ((error = xfs_bmbt_update(cur, new->br_startoff,
2668 new->br_blockcount + RIGHT.br_blockcount,
2674 case BMAP_RIGHT_FILLING:
2676 * Setting the last part of a previous oldext extent to newext.
2677 * The right neighbor is not contiguous.
2679 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2680 xfs_bmbt_set_blockcount(ep,
2681 PREV.br_blockcount - new->br_blockcount);
2682 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2685 xfs_iext_insert(ip, *idx, 1, new, state);
2687 ip->i_d.di_nextents++;
2689 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2691 rval = XFS_ILOG_CORE;
2692 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2693 PREV.br_startblock, PREV.br_blockcount,
2696 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2697 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2699 PREV.br_blockcount - new->br_blockcount,
2702 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2703 new->br_startblock, new->br_blockcount,
2706 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2707 cur->bc_rec.b.br_state = XFS_EXT_NORM;
2708 if ((error = xfs_btree_insert(cur, &i)))
2710 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2716 * Setting the middle part of a previous oldext extent to
2717 * newext. Contiguity is impossible here.
2718 * One extent becomes three extents.
2720 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2721 xfs_bmbt_set_blockcount(ep,
2722 new->br_startoff - PREV.br_startoff);
2723 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2726 r[1].br_startoff = new_endoff;
2727 r[1].br_blockcount =
2728 PREV.br_startoff + PREV.br_blockcount - new_endoff;
2729 r[1].br_startblock = new->br_startblock + new->br_blockcount;
2730 r[1].br_state = oldext;
2733 xfs_iext_insert(ip, *idx, 2, &r[0], state);
2735 ip->i_d.di_nextents += 2;
2737 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2739 rval = XFS_ILOG_CORE;
2740 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2741 PREV.br_startblock, PREV.br_blockcount,
2744 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2745 /* new right extent - oldext */
2746 if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
2747 r[1].br_startblock, r[1].br_blockcount,
2750 /* new left extent - oldext */
2751 cur->bc_rec.b = PREV;
2752 cur->bc_rec.b.br_blockcount =
2753 new->br_startoff - PREV.br_startoff;
2754 if ((error = xfs_btree_insert(cur, &i)))
2756 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2758 * Reset the cursor to the position of the new extent
2759 * we are about to insert as we can't trust it after
2760 * the previous insert.
2762 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2763 new->br_startblock, new->br_blockcount,
2766 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
2767 /* new middle extent - newext */
2768 cur->bc_rec.b.br_state = new->br_state;
2769 if ((error = xfs_btree_insert(cur, &i)))
2771 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
2775 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2776 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2777 case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2778 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2779 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2780 case BMAP_LEFT_CONTIG:
2781 case BMAP_RIGHT_CONTIG:
2783 * These cases are all impossible.
2788 /* convert to a btree if necessary */
2789 if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
2790 int tmp_logflags; /* partial log flag return val */
2792 ASSERT(cur == NULL);
2793 error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
2794 0, &tmp_logflags, XFS_DATA_FORK);
2795 *logflagsp |= tmp_logflags;
2800 /* clear out the allocated field, done with it now in any case. */
2802 cur->bc_private.b.allocated = 0;
2806 xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
2816 * Convert a hole to a delayed allocation.
2819 xfs_bmap_add_extent_hole_delay(
2820 xfs_inode_t *ip, /* incore inode pointer */
2821 xfs_extnum_t *idx, /* extent number to update/insert */
2822 xfs_bmbt_irec_t *new) /* new data to add to file extents */
2824 xfs_ifork_t *ifp; /* inode fork pointer */
2825 xfs_bmbt_irec_t left; /* left neighbor extent entry */
2826 xfs_filblks_t newlen=0; /* new indirect size */
2827 xfs_filblks_t oldlen=0; /* old indirect size */
2828 xfs_bmbt_irec_t right; /* right neighbor extent entry */
2829 int state; /* state bits, accessed thru macros */
2830 xfs_filblks_t temp=0; /* temp for indirect calculations */
2832 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2834 ASSERT(isnullstartblock(new->br_startblock));
2837 * Check and set flags if this segment has a left neighbor
2840 state |= BMAP_LEFT_VALID;
2841 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2843 if (isnullstartblock(left.br_startblock))
2844 state |= BMAP_LEFT_DELAY;
2848 * Check and set flags if the current (right) segment exists.
2849 * If it doesn't exist, we're converting the hole at end-of-file.
2851 if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2852 state |= BMAP_RIGHT_VALID;
2853 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2855 if (isnullstartblock(right.br_startblock))
2856 state |= BMAP_RIGHT_DELAY;
2860 * Set contiguity flags on the left and right neighbors.
2861 * Don't let extents get too large, even if the pieces are contiguous.
2863 if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2864 left.br_startoff + left.br_blockcount == new->br_startoff &&
2865 left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2866 state |= BMAP_LEFT_CONTIG;
2868 if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2869 new->br_startoff + new->br_blockcount == right.br_startoff &&
2870 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2871 (!(state & BMAP_LEFT_CONTIG) ||
2872 (left.br_blockcount + new->br_blockcount +
2873 right.br_blockcount <= MAXEXTLEN)))
2874 state |= BMAP_RIGHT_CONTIG;
2877 * Switch out based on the contiguity flags.
2879 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2880 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2882 * New allocation is contiguous with delayed allocations
2883 * on the left and on the right.
2884 * Merge all three into a single extent record.
2887 temp = left.br_blockcount + new->br_blockcount +
2888 right.br_blockcount;
2890 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2891 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2892 oldlen = startblockval(left.br_startblock) +
2893 startblockval(new->br_startblock) +
2894 startblockval(right.br_startblock);
2895 newlen = xfs_bmap_worst_indlen(ip, temp);
2896 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2897 nullstartblock((int)newlen));
2898 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2900 xfs_iext_remove(ip, *idx + 1, 1, state);
2903 case BMAP_LEFT_CONTIG:
2905 * New allocation is contiguous with a delayed allocation
2907 * Merge the new allocation with the left neighbor.
2910 temp = left.br_blockcount + new->br_blockcount;
2912 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2913 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2914 oldlen = startblockval(left.br_startblock) +
2915 startblockval(new->br_startblock);
2916 newlen = xfs_bmap_worst_indlen(ip, temp);
2917 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2918 nullstartblock((int)newlen));
2919 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2922 case BMAP_RIGHT_CONTIG:
2924 * New allocation is contiguous with a delayed allocation
2926 * Merge the new allocation with the right neighbor.
2928 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2929 temp = new->br_blockcount + right.br_blockcount;
2930 oldlen = startblockval(new->br_startblock) +
2931 startblockval(right.br_startblock);
2932 newlen = xfs_bmap_worst_indlen(ip, temp);
2933 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2935 nullstartblock((int)newlen), temp, right.br_state);
2936 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2941 * New allocation is not contiguous with another
2942 * delayed allocation.
2943 * Insert a new entry.
2945 oldlen = newlen = 0;
2946 xfs_iext_insert(ip, *idx, 1, new, state);
2949 if (oldlen != newlen) {
2950 ASSERT(oldlen > newlen);
2951 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
2952 (int64_t)(oldlen - newlen), 0);
2954 * Nothing to do for disk quota accounting here.
2960 * Convert a hole to a real allocation.
2962 STATIC int /* error */
2963 xfs_bmap_add_extent_hole_real(
2964 struct xfs_bmalloca *bma,
2967 struct xfs_bmbt_irec *new = &bma->got;
2968 int error; /* error return value */
2969 int i; /* temp state */
2970 xfs_ifork_t *ifp; /* inode fork pointer */
2971 xfs_bmbt_irec_t left; /* left neighbor extent entry */
2972 xfs_bmbt_irec_t right; /* right neighbor extent entry */
2973 int rval=0; /* return value (logging flags) */
2974 int state; /* state bits, accessed thru macros */
2975 struct xfs_mount *mp;
2977 mp = bma->tp ? bma->tp->t_mountp : NULL;
2978 ifp = XFS_IFORK_PTR(bma->ip, whichfork);
2980 ASSERT(bma->idx >= 0);
2981 ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2982 ASSERT(!isnullstartblock(new->br_startblock));
2984 !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2986 XFS_STATS_INC(xs_add_exlist);
2989 if (whichfork == XFS_ATTR_FORK)
2990 state |= BMAP_ATTRFORK;
2993 * Check and set flags if this segment has a left neighbor.
2996 state |= BMAP_LEFT_VALID;
2997 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
2998 if (isnullstartblock(left.br_startblock))
2999 state |= BMAP_LEFT_DELAY;
3003 * Check and set flags if this segment has a current value.
3004 * Not true if we're inserting into the "hole" at eof.
3006 if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
3007 state |= BMAP_RIGHT_VALID;
3008 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
3009 if (isnullstartblock(right.br_startblock))
3010 state |= BMAP_RIGHT_DELAY;
3014 * We're inserting a real allocation between "left" and "right".
3015 * Set the contiguity flags. Don't let extents get too large.
3017 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
3018 left.br_startoff + left.br_blockcount == new->br_startoff &&
3019 left.br_startblock + left.br_blockcount == new->br_startblock &&
3020 left.br_state == new->br_state &&
3021 left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
3022 state |= BMAP_LEFT_CONTIG;
3024 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
3025 new->br_startoff + new->br_blockcount == right.br_startoff &&
3026 new->br_startblock + new->br_blockcount == right.br_startblock &&
3027 new->br_state == right.br_state &&
3028 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
3029 (!(state & BMAP_LEFT_CONTIG) ||
3030 left.br_blockcount + new->br_blockcount +
3031 right.br_blockcount <= MAXEXTLEN))
3032 state |= BMAP_RIGHT_CONTIG;
3036 * Select which case we're in here, and implement it.
3038 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
3039 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
3041 * New allocation is contiguous with real allocations on the
3042 * left and on the right.
3043 * Merge all three into a single extent record.
3046 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3047 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3048 left.br_blockcount + new->br_blockcount +
3049 right.br_blockcount);
3050 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3052 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
3054 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3055 XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
3056 if (bma->cur == NULL) {
3057 rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3059 rval = XFS_ILOG_CORE;
3060 error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
3061 right.br_startblock, right.br_blockcount,
3065 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3066 error = xfs_btree_delete(bma->cur, &i);
3069 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3070 error = xfs_btree_decrement(bma->cur, 0, &i);
3073 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3074 error = xfs_bmbt_update(bma->cur, left.br_startoff,
3076 left.br_blockcount +
3077 new->br_blockcount +
3078 right.br_blockcount,
3085 case BMAP_LEFT_CONTIG:
3087 * New allocation is contiguous with a real allocation
3089 * Merge the new allocation with the left neighbor.
3092 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3093 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3094 left.br_blockcount + new->br_blockcount);
3095 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3097 if (bma->cur == NULL) {
3098 rval = xfs_ilog_fext(whichfork);
3101 error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
3102 left.br_startblock, left.br_blockcount,
3106 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3107 error = xfs_bmbt_update(bma->cur, left.br_startoff,
3109 left.br_blockcount +
3117 case BMAP_RIGHT_CONTIG:
3119 * New allocation is contiguous with a real allocation
3121 * Merge the new allocation with the right neighbor.
3123 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3124 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
3125 new->br_startoff, new->br_startblock,
3126 new->br_blockcount + right.br_blockcount,
3128 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3130 if (bma->cur == NULL) {
3131 rval = xfs_ilog_fext(whichfork);
3134 error = xfs_bmbt_lookup_eq(bma->cur,
3136 right.br_startblock,
3137 right.br_blockcount, &i);
3140 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3141 error = xfs_bmbt_update(bma->cur, new->br_startoff,
3143 new->br_blockcount +
3144 right.br_blockcount,
3153 * New allocation is not contiguous with another
3155 * Insert a new entry.
3157 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
3158 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3159 XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
3160 if (bma->cur == NULL) {
3161 rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3163 rval = XFS_ILOG_CORE;
3164 error = xfs_bmbt_lookup_eq(bma->cur,
3167 new->br_blockcount, &i);
3170 XFS_WANT_CORRUPTED_GOTO(mp, i == 0, done);
3171 bma->cur->bc_rec.b.br_state = new->br_state;
3172 error = xfs_btree_insert(bma->cur, &i);
3175 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
3180 /* convert to a btree if necessary */
3181 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
3182 int tmp_logflags; /* partial log flag return val */
3184 ASSERT(bma->cur == NULL);
3185 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
3186 bma->firstblock, bma->flist, &bma->cur,
3187 0, &tmp_logflags, whichfork);
3188 bma->logflags |= tmp_logflags;
3193 /* clear out the allocated field, done with it now in any case. */
3195 bma->cur->bc_private.b.allocated = 0;
3197 xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
3199 bma->logflags |= rval;
3204 * Functions used in the extent read, allocate and remove paths
3208 * Adjust the size of the new extent based on di_extsize and rt extsize.
3211 xfs_bmap_extsize_align(
3213 xfs_bmbt_irec_t *gotp, /* next extent pointer */
3214 xfs_bmbt_irec_t *prevp, /* previous extent pointer */
3215 xfs_extlen_t extsz, /* align to this extent size */
3216 int rt, /* is this a realtime inode? */
3217 int eof, /* is extent at end-of-file? */
3218 int delay, /* creating delalloc extent? */
3219 int convert, /* overwriting unwritten extent? */
3220 xfs_fileoff_t *offp, /* in/out: aligned offset */
3221 xfs_extlen_t *lenp) /* in/out: aligned length */
3223 xfs_fileoff_t orig_off; /* original offset */
3224 xfs_extlen_t orig_alen; /* original length */
3225 xfs_fileoff_t orig_end; /* original off+len */
3226 xfs_fileoff_t nexto; /* next file offset */
3227 xfs_fileoff_t prevo; /* previous file offset */
3228 xfs_fileoff_t align_off; /* temp for offset */
3229 xfs_extlen_t align_alen; /* temp for length */
3230 xfs_extlen_t temp; /* temp for calculations */
3235 orig_off = align_off = *offp;
3236 orig_alen = align_alen = *lenp;
3237 orig_end = orig_off + orig_alen;
3240 * If this request overlaps an existing extent, then don't
3241 * attempt to perform any additional alignment.
3243 if (!delay && !eof &&
3244 (orig_off >= gotp->br_startoff) &&
3245 (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3250 * If the file offset is unaligned vs. the extent size
3251 * we need to align it. This will be possible unless
3252 * the file was previously written with a kernel that didn't
3253 * perform this alignment, or if a truncate shot us in the
3256 temp = do_mod(orig_off, extsz);
3262 * Same adjustment for the end of the requested area.
3264 if ((temp = (align_alen % extsz))) {
3265 align_alen += extsz - temp;
3268 * If the previous block overlaps with this proposed allocation
3269 * then move the start forward without adjusting the length.
3271 if (prevp->br_startoff != NULLFILEOFF) {
3272 if (prevp->br_startblock == HOLESTARTBLOCK)
3273 prevo = prevp->br_startoff;
3275 prevo = prevp->br_startoff + prevp->br_blockcount;
3278 if (align_off != orig_off && align_off < prevo)
3281 * If the next block overlaps with this proposed allocation
3282 * then move the start back without adjusting the length,
3283 * but not before offset 0.
3284 * This may of course make the start overlap previous block,
3285 * and if we hit the offset 0 limit then the next block
3286 * can still overlap too.
3288 if (!eof && gotp->br_startoff != NULLFILEOFF) {
3289 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3290 (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3291 nexto = gotp->br_startoff + gotp->br_blockcount;
3293 nexto = gotp->br_startoff;
3295 nexto = NULLFILEOFF;
3297 align_off + align_alen != orig_end &&
3298 align_off + align_alen > nexto)
3299 align_off = nexto > align_alen ? nexto - align_alen : 0;
3301 * If we're now overlapping the next or previous extent that
3302 * means we can't fit an extsz piece in this hole. Just move
3303 * the start forward to the first valid spot and set
3304 * the length so we hit the end.
3306 if (align_off != orig_off && align_off < prevo)
3308 if (align_off + align_alen != orig_end &&
3309 align_off + align_alen > nexto &&
3310 nexto != NULLFILEOFF) {
3311 ASSERT(nexto > prevo);
3312 align_alen = nexto - align_off;
3316 * If realtime, and the result isn't a multiple of the realtime
3317 * extent size we need to remove blocks until it is.
3319 if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3321 * We're not covering the original request, or
3322 * we won't be able to once we fix the length.
3324 if (orig_off < align_off ||
3325 orig_end > align_off + align_alen ||
3326 align_alen - temp < orig_alen)
3329 * Try to fix it by moving the start up.
3331 if (align_off + temp <= orig_off) {
3336 * Try to fix it by moving the end in.
3338 else if (align_off + align_alen - temp >= orig_end)
3341 * Set the start to the minimum then trim the length.
3344 align_alen -= orig_off - align_off;
3345 align_off = orig_off;
3346 align_alen -= align_alen % mp->m_sb.sb_rextsize;
3349 * Result doesn't cover the request, fail it.
3351 if (orig_off < align_off || orig_end > align_off + align_alen)
3354 ASSERT(orig_off >= align_off);
3355 ASSERT(orig_end <= align_off + align_alen);
3359 if (!eof && gotp->br_startoff != NULLFILEOFF)
3360 ASSERT(align_off + align_alen <= gotp->br_startoff);
3361 if (prevp->br_startoff != NULLFILEOFF)
3362 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3370 #define XFS_ALLOC_GAP_UNITS 4
3374 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
3376 xfs_fsblock_t adjust; /* adjustment to block numbers */
3377 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
3378 xfs_mount_t *mp; /* mount point structure */
3379 int nullfb; /* true if ap->firstblock isn't set */
3380 int rt; /* true if inode is realtime */
3382 #define ISVALID(x,y) \
3384 (x) < mp->m_sb.sb_rblocks : \
3385 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3386 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3387 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3389 mp = ap->ip->i_mount;
3390 nullfb = *ap->firstblock == NULLFSBLOCK;
3391 rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
3392 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3394 * If allocating at eof, and there's a previous real block,
3395 * try to use its last block as our starting point.
3397 if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3398 !isnullstartblock(ap->prev.br_startblock) &&
3399 ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3400 ap->prev.br_startblock)) {
3401 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3403 * Adjust for the gap between prevp and us.
3405 adjust = ap->offset -
3406 (ap->prev.br_startoff + ap->prev.br_blockcount);
3408 ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3409 ap->blkno += adjust;
3412 * If not at eof, then compare the two neighbor blocks.
3413 * Figure out whether either one gives us a good starting point,
3414 * and pick the better one.
3416 else if (!ap->eof) {
3417 xfs_fsblock_t gotbno; /* right side block number */
3418 xfs_fsblock_t gotdiff=0; /* right side difference */
3419 xfs_fsblock_t prevbno; /* left side block number */
3420 xfs_fsblock_t prevdiff=0; /* left side difference */
3423 * If there's a previous (left) block, select a requested
3424 * start block based on it.
3426 if (ap->prev.br_startoff != NULLFILEOFF &&
3427 !isnullstartblock(ap->prev.br_startblock) &&
3428 (prevbno = ap->prev.br_startblock +
3429 ap->prev.br_blockcount) &&
3430 ISVALID(prevbno, ap->prev.br_startblock)) {
3432 * Calculate gap to end of previous block.
3434 adjust = prevdiff = ap->offset -
3435 (ap->prev.br_startoff +
3436 ap->prev.br_blockcount);
3438 * Figure the startblock based on the previous block's
3439 * end and the gap size.
3441 * If the gap is large relative to the piece we're
3442 * allocating, or using it gives us an invalid block
3443 * number, then just use the end of the previous block.
3445 if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3446 ISVALID(prevbno + prevdiff,
3447 ap->prev.br_startblock))
3452 * If the firstblock forbids it, can't use it,
3455 if (!rt && !nullfb &&
3456 XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3457 prevbno = NULLFSBLOCK;
3460 * No previous block or can't follow it, just default.
3463 prevbno = NULLFSBLOCK;
3465 * If there's a following (right) block, select a requested
3466 * start block based on it.
3468 if (!isnullstartblock(ap->got.br_startblock)) {
3470 * Calculate gap to start of next block.
3472 adjust = gotdiff = ap->got.br_startoff - ap->offset;
3474 * Figure the startblock based on the next block's
3475 * start and the gap size.
3477 gotbno = ap->got.br_startblock;
3480 * If the gap is large relative to the piece we're
3481 * allocating, or using it gives us an invalid block
3482 * number, then just use the start of the next block
3483 * offset by our length.
3485 if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3486 ISVALID(gotbno - gotdiff, gotbno))
3488 else if (ISVALID(gotbno - ap->length, gotbno)) {
3489 gotbno -= ap->length;
3490 gotdiff += adjust - ap->length;
3494 * If the firstblock forbids it, can't use it,
3497 if (!rt && !nullfb &&
3498 XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3499 gotbno = NULLFSBLOCK;
3502 * No next block, just default.
3505 gotbno = NULLFSBLOCK;
3507 * If both valid, pick the better one, else the only good
3508 * one, else ap->blkno is already set (to 0 or the inode block).
3510 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3511 ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3512 else if (prevbno != NULLFSBLOCK)
3513 ap->blkno = prevbno;
3514 else if (gotbno != NULLFSBLOCK)
3521 xfs_bmap_longest_free_extent(
3522 struct xfs_trans *tp,
3527 struct xfs_mount *mp = tp->t_mountp;
3528 struct xfs_perag *pag;
3529 xfs_extlen_t longest;
3532 pag = xfs_perag_get(mp, ag);
3533 if (!pag->pagf_init) {
3534 error = xfs_alloc_pagf_init(mp, tp, ag, XFS_ALLOC_FLAG_TRYLOCK);
3538 if (!pag->pagf_init) {
3544 longest = xfs_alloc_longest_free_extent(mp, pag);
3545 if (*blen < longest)
3554 xfs_bmap_select_minlen(
3555 struct xfs_bmalloca *ap,
3556 struct xfs_alloc_arg *args,
3560 if (notinit || *blen < ap->minlen) {
3562 * Since we did a BUF_TRYLOCK above, it is possible that
3563 * there is space for this request.
3565 args->minlen = ap->minlen;
3566 } else if (*blen < args->maxlen) {
3568 * If the best seen length is less than the request length,
3569 * use the best as the minimum.
3571 args->minlen = *blen;
3574 * Otherwise we've seen an extent as big as maxlen, use that
3577 args->minlen = args->maxlen;
3582 xfs_bmap_btalloc_nullfb(
3583 struct xfs_bmalloca *ap,
3584 struct xfs_alloc_arg *args,
3587 struct xfs_mount *mp = ap->ip->i_mount;
3588 xfs_agnumber_t ag, startag;
3592 args->type = XFS_ALLOCTYPE_START_BNO;
3593 args->total = ap->total;
3595 startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3596 if (startag == NULLAGNUMBER)
3599 while (*blen < args->maxlen) {
3600 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3605 if (++ag == mp->m_sb.sb_agcount)
3611 xfs_bmap_select_minlen(ap, args, blen, notinit);
3616 xfs_bmap_btalloc_filestreams(
3617 struct xfs_bmalloca *ap,
3618 struct xfs_alloc_arg *args,
3621 struct xfs_mount *mp = ap->ip->i_mount;
3626 args->type = XFS_ALLOCTYPE_NEAR_BNO;
3627 args->total = ap->total;
3629 ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3630 if (ag == NULLAGNUMBER)
3633 error = xfs_bmap_longest_free_extent(args->tp, ag, blen, ¬init);
3637 if (*blen < args->maxlen) {
3638 error = xfs_filestream_new_ag(ap, &ag);
3642 error = xfs_bmap_longest_free_extent(args->tp, ag, blen,
3649 xfs_bmap_select_minlen(ap, args, blen, notinit);
3652 * Set the failure fallback case to look in the selected AG as stream
3655 ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3661 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
3663 xfs_mount_t *mp; /* mount point structure */
3664 xfs_alloctype_t atype = 0; /* type for allocation routines */
3665 xfs_extlen_t align; /* minimum allocation alignment */
3666 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
3668 xfs_alloc_arg_t args;
3670 xfs_extlen_t nextminlen = 0;
3671 int nullfb; /* true if ap->firstblock isn't set */
3679 mp = ap->ip->i_mount;
3681 /* stripe alignment for allocation is determined by mount parameters */
3683 if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
3684 stripe_align = mp->m_swidth;
3685 else if (mp->m_dalign)
3686 stripe_align = mp->m_dalign;
3688 align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
3689 if (unlikely(align)) {
3690 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3691 align, 0, ap->eof, 0, ap->conv,
3692 &ap->offset, &ap->length);
3698 nullfb = *ap->firstblock == NULLFSBLOCK;
3699 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3701 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
3702 ag = xfs_filestream_lookup_ag(ap->ip);
3703 ag = (ag != NULLAGNUMBER) ? ag : 0;
3704 ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3706 ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3709 ap->blkno = *ap->firstblock;
3711 xfs_bmap_adjacent(ap);
3714 * If allowed, use ap->blkno; otherwise must use firstblock since
3715 * it's in the right allocation group.
3717 if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3720 ap->blkno = *ap->firstblock;
3722 * Normal allocation, done through xfs_alloc_vextent.
3724 tryagain = isaligned = 0;
3725 memset(&args, 0, sizeof(args));
3728 args.fsbno = ap->blkno;
3730 /* Trim the allocation back to the maximum an AG can fit. */
3731 args.maxlen = MIN(ap->length, XFS_ALLOC_AG_MAX_USABLE(mp));
3732 args.firstblock = *ap->firstblock;
3736 * Search for an allocation group with a single extent large
3737 * enough for the request. If one isn't found, then adjust
3738 * the minimum allocation size to the largest space found.
3740 if (ap->userdata && xfs_inode_is_filestream(ap->ip))
3741 error = xfs_bmap_btalloc_filestreams(ap, &args, &blen);
3743 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3746 } else if (ap->flist->xbf_low) {
3747 if (xfs_inode_is_filestream(ap->ip))
3748 args.type = XFS_ALLOCTYPE_FIRST_AG;
3750 args.type = XFS_ALLOCTYPE_START_BNO;
3751 args.total = args.minlen = ap->minlen;
3753 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3754 args.total = ap->total;
3755 args.minlen = ap->minlen;
3757 /* apply extent size hints if obtained earlier */
3758 if (unlikely(align)) {
3760 if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
3761 args.mod = (xfs_extlen_t)(args.prod - args.mod);
3762 } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
3766 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
3767 if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
3768 args.mod = (xfs_extlen_t)(args.prod - args.mod);
3771 * If we are not low on available data blocks, and the
3772 * underlying logical volume manager is a stripe, and
3773 * the file offset is zero then try to allocate data
3774 * blocks on stripe unit boundary.
3775 * NOTE: ap->aeof is only set if the allocation length
3776 * is >= the stripe unit and the allocation offset is
3777 * at the end of file.
3779 if (!ap->flist->xbf_low && ap->aeof) {
3781 args.alignment = stripe_align;
3785 * Adjust for alignment
3787 if (blen > args.alignment && blen <= args.maxlen)
3788 args.minlen = blen - args.alignment;
3789 args.minalignslop = 0;
3792 * First try an exact bno allocation.
3793 * If it fails then do a near or start bno
3794 * allocation with alignment turned on.
3798 args.type = XFS_ALLOCTYPE_THIS_BNO;
3801 * Compute the minlen+alignment for the
3802 * next case. Set slop so that the value
3803 * of minlen+alignment+slop doesn't go up
3804 * between the calls.
3806 if (blen > stripe_align && blen <= args.maxlen)
3807 nextminlen = blen - stripe_align;
3809 nextminlen = args.minlen;
3810 if (nextminlen + stripe_align > args.minlen + 1)
3812 nextminlen + stripe_align -
3815 args.minalignslop = 0;
3819 args.minalignslop = 0;
3821 args.minleft = ap->minleft;
3822 args.wasdel = ap->wasdel;
3824 args.userdata = ap->userdata;
3825 if ((error = xfs_alloc_vextent(&args)))
3827 if (tryagain && args.fsbno == NULLFSBLOCK) {
3829 * Exact allocation failed. Now try with alignment
3833 args.fsbno = ap->blkno;
3834 args.alignment = stripe_align;
3835 args.minlen = nextminlen;
3836 args.minalignslop = 0;
3838 if ((error = xfs_alloc_vextent(&args)))
3841 if (isaligned && args.fsbno == NULLFSBLOCK) {
3843 * allocation failed, so turn off alignment and
3847 args.fsbno = ap->blkno;
3849 if ((error = xfs_alloc_vextent(&args)))
3852 if (args.fsbno == NULLFSBLOCK && nullfb &&
3853 args.minlen > ap->minlen) {
3854 args.minlen = ap->minlen;
3855 args.type = XFS_ALLOCTYPE_START_BNO;
3856 args.fsbno = ap->blkno;
3857 if ((error = xfs_alloc_vextent(&args)))
3860 if (args.fsbno == NULLFSBLOCK && nullfb) {
3862 args.type = XFS_ALLOCTYPE_FIRST_AG;
3863 args.total = ap->minlen;
3865 if ((error = xfs_alloc_vextent(&args)))
3867 ap->flist->xbf_low = 1;
3869 if (args.fsbno != NULLFSBLOCK) {
3871 * check the allocation happened at the same or higher AG than
3872 * the first block that was allocated.
3874 ASSERT(*ap->firstblock == NULLFSBLOCK ||
3875 XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
3876 XFS_FSB_TO_AGNO(mp, args.fsbno) ||
3877 (ap->flist->xbf_low &&
3878 XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
3879 XFS_FSB_TO_AGNO(mp, args.fsbno)));
3881 ap->blkno = args.fsbno;
3882 if (*ap->firstblock == NULLFSBLOCK)
3883 *ap->firstblock = args.fsbno;
3884 ASSERT(nullfb || fb_agno == args.agno ||
3885 (ap->flist->xbf_low && fb_agno < args.agno));
3886 ap->length = args.len;
3887 ap->ip->i_d.di_nblocks += args.len;
3888 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3890 ap->ip->i_delayed_blks -= args.len;
3892 * Adjust the disk quota also. This was reserved
3895 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3896 ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
3897 XFS_TRANS_DQ_BCOUNT,
3900 ap->blkno = NULLFSBLOCK;
3907 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3908 * It figures out where to ask the underlying allocator to put the new extent.
3912 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
3914 if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
3915 return xfs_bmap_rtalloc(ap);
3916 return xfs_bmap_btalloc(ap);
3920 * Trim the returned map to the required bounds
3924 struct xfs_bmbt_irec *mval,
3925 struct xfs_bmbt_irec *got,
3933 if ((flags & XFS_BMAPI_ENTIRE) ||
3934 got->br_startoff + got->br_blockcount <= obno) {
3936 if (isnullstartblock(got->br_startblock))
3937 mval->br_startblock = DELAYSTARTBLOCK;
3943 ASSERT((*bno >= obno) || (n == 0));
3945 mval->br_startoff = *bno;
3946 if (isnullstartblock(got->br_startblock))
3947 mval->br_startblock = DELAYSTARTBLOCK;
3949 mval->br_startblock = got->br_startblock +
3950 (*bno - got->br_startoff);
3952 * Return the minimum of what we got and what we asked for for
3953 * the length. We can use the len variable here because it is
3954 * modified below and we could have been there before coming
3955 * here if the first part of the allocation didn't overlap what
3958 mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3959 got->br_blockcount - (*bno - got->br_startoff));
3960 mval->br_state = got->br_state;
3961 ASSERT(mval->br_blockcount <= len);
3966 * Update and validate the extent map to return
3969 xfs_bmapi_update_map(
3970 struct xfs_bmbt_irec **map,
3978 xfs_bmbt_irec_t *mval = *map;
3980 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3981 ((mval->br_startoff + mval->br_blockcount) <= end));
3982 ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3983 (mval->br_startoff < obno));
3985 *bno = mval->br_startoff + mval->br_blockcount;
3987 if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3988 /* update previous map with new information */
3989 ASSERT(mval->br_startblock == mval[-1].br_startblock);
3990 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3991 ASSERT(mval->br_state == mval[-1].br_state);
3992 mval[-1].br_blockcount = mval->br_blockcount;
3993 mval[-1].br_state = mval->br_state;
3994 } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3995 mval[-1].br_startblock != DELAYSTARTBLOCK &&
3996 mval[-1].br_startblock != HOLESTARTBLOCK &&
3997 mval->br_startblock == mval[-1].br_startblock +
3998 mval[-1].br_blockcount &&
3999 ((flags & XFS_BMAPI_IGSTATE) ||
4000 mval[-1].br_state == mval->br_state)) {
4001 ASSERT(mval->br_startoff ==
4002 mval[-1].br_startoff + mval[-1].br_blockcount);
4003 mval[-1].br_blockcount += mval->br_blockcount;
4004 } else if (*n > 0 &&
4005 mval->br_startblock == DELAYSTARTBLOCK &&
4006 mval[-1].br_startblock == DELAYSTARTBLOCK &&
4007 mval->br_startoff ==
4008 mval[-1].br_startoff + mval[-1].br_blockcount) {
4009 mval[-1].br_blockcount += mval->br_blockcount;
4010 mval[-1].br_state = mval->br_state;
4011 } else if (!((*n == 0) &&
4012 ((mval->br_startoff + mval->br_blockcount) <=
4021 * Map file blocks to filesystem blocks without allocation.
4025 struct xfs_inode *ip,
4028 struct xfs_bmbt_irec *mval,
4032 struct xfs_mount *mp = ip->i_mount;
4033 struct xfs_ifork *ifp;
4034 struct xfs_bmbt_irec got;
4035 struct xfs_bmbt_irec prev;
4042 int whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4043 XFS_ATTR_FORK : XFS_DATA_FORK;
4046 ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
4047 XFS_BMAPI_IGSTATE)));
4048 ASSERT(xfs_isilocked(ip, XFS_ILOCK_SHARED|XFS_ILOCK_EXCL));
4050 if (unlikely(XFS_TEST_ERROR(
4051 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4052 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4053 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4054 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4055 return -EFSCORRUPTED;
4058 if (XFS_FORCED_SHUTDOWN(mp))
4061 XFS_STATS_INC(xs_blk_mapr);
4063 ifp = XFS_IFORK_PTR(ip, whichfork);
4065 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4066 error = xfs_iread_extents(NULL, ip, whichfork);
4071 xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4075 while (bno < end && n < *nmap) {
4076 /* Reading past eof, act as though there's a hole up to end. */
4078 got.br_startoff = end;
4079 if (got.br_startoff > bno) {
4080 /* Reading in a hole. */
4081 mval->br_startoff = bno;
4082 mval->br_startblock = HOLESTARTBLOCK;
4083 mval->br_blockcount =
4084 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4085 mval->br_state = XFS_EXT_NORM;
4086 bno += mval->br_blockcount;
4087 len -= mval->br_blockcount;
4093 /* set up the extent map to return. */
4094 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4095 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4097 /* If we're done, stop now. */
4098 if (bno >= end || n >= *nmap)
4101 /* Else go on to the next record. */
4102 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4103 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4112 xfs_bmapi_reserve_delalloc(
4113 struct xfs_inode *ip,
4116 struct xfs_bmbt_irec *got,
4117 struct xfs_bmbt_irec *prev,
4118 xfs_extnum_t *lastx,
4121 struct xfs_mount *mp = ip->i_mount;
4122 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4124 xfs_extlen_t indlen;
4125 char rt = XFS_IS_REALTIME_INODE(ip);
4129 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4131 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4133 /* Figure out the extent size, adjust alen */
4134 extsz = xfs_get_extsz_hint(ip);
4137 * Make sure we don't exceed a single extent length when we
4138 * align the extent by reducing length we are going to
4139 * allocate by the maximum amount extent size aligment may
4142 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN - (2 * extsz - 1));
4143 error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4144 1, 0, &aoff, &alen);
4149 extsz = alen / mp->m_sb.sb_rextsize;
4152 * Make a transaction-less quota reservation for delayed allocation
4153 * blocks. This number gets adjusted later. We return if we haven't
4154 * allocated blocks already inside this loop.
4156 error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4157 rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4162 * Split changing sb for alen and indlen since they could be coming
4163 * from different places.
4165 indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4169 error = xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
4170 -((int64_t)extsz), 0);
4172 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4173 -((int64_t)alen), 0);
4177 goto out_unreserve_quota;
4179 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4180 -((int64_t)indlen), 0);
4182 goto out_unreserve_blocks;
4185 ip->i_delayed_blks += alen;
4187 got->br_startoff = aoff;
4188 got->br_startblock = nullstartblock(indlen);
4189 got->br_blockcount = alen;
4190 got->br_state = XFS_EXT_NORM;
4191 xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4194 * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4195 * might have merged it into one of the neighbouring ones.
4197 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4199 ASSERT(got->br_startoff <= aoff);
4200 ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4201 ASSERT(isnullstartblock(got->br_startblock));
4202 ASSERT(got->br_state == XFS_EXT_NORM);
4205 out_unreserve_blocks:
4207 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS, extsz, 0);
4209 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, alen, 0);
4210 out_unreserve_quota:
4211 if (XFS_IS_QUOTA_ON(mp))
4212 xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4213 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4218 * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4222 struct xfs_inode *ip, /* incore inode */
4223 xfs_fileoff_t bno, /* starting file offs. mapped */
4224 xfs_filblks_t len, /* length to map in file */
4225 struct xfs_bmbt_irec *mval, /* output: map values */
4226 int *nmap, /* i/o: mval size/count */
4227 int flags) /* XFS_BMAPI_... */
4229 struct xfs_mount *mp = ip->i_mount;
4230 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4231 struct xfs_bmbt_irec got; /* current file extent record */
4232 struct xfs_bmbt_irec prev; /* previous file extent record */
4233 xfs_fileoff_t obno; /* old block number (offset) */
4234 xfs_fileoff_t end; /* end of mapped file region */
4235 xfs_extnum_t lastx; /* last useful extent number */
4236 int eof; /* we've hit the end of extents */
4237 int n = 0; /* current extent index */
4241 ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4242 ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4243 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4245 if (unlikely(XFS_TEST_ERROR(
4246 (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4247 XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4248 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4249 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4250 return -EFSCORRUPTED;
4253 if (XFS_FORCED_SHUTDOWN(mp))
4256 XFS_STATS_INC(xs_blk_mapw);
4258 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4259 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4264 xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4268 while (bno < end && n < *nmap) {
4269 if (eof || got.br_startoff > bno) {
4270 error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4271 &prev, &lastx, eof);
4281 /* set up the extent map to return. */
4282 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4283 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4285 /* If we're done, stop now. */
4286 if (bno >= end || n >= *nmap)
4289 /* Else go on to the next record. */
4291 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4292 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4304 struct xfs_bmalloca *bma)
4306 struct xfs_mount *mp = bma->ip->i_mount;
4307 int whichfork = (bma->flags & XFS_BMAPI_ATTRFORK) ?
4308 XFS_ATTR_FORK : XFS_DATA_FORK;
4309 struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4310 int tmp_logflags = 0;
4313 ASSERT(bma->length > 0);
4316 * For the wasdelay case, we could also just allocate the stuff asked
4317 * for in this bmap call but that wouldn't be as good.
4320 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4321 bma->offset = bma->got.br_startoff;
4322 if (bma->idx != NULLEXTNUM && bma->idx) {
4323 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4327 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4329 bma->length = XFS_FILBLKS_MIN(bma->length,
4330 bma->got.br_startoff - bma->offset);
4334 * Indicate if this is the first user data in the file, or just any
4337 if (!(bma->flags & XFS_BMAPI_METADATA)) {
4338 bma->userdata = (bma->offset == 0) ?
4339 XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4342 bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4345 * Only want to do the alignment at the eof if it is userdata and
4346 * allocation length is larger than a stripe unit.
4348 if (mp->m_dalign && bma->length >= mp->m_dalign &&
4349 !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4350 error = xfs_bmap_isaeof(bma, whichfork);
4355 error = xfs_bmap_alloc(bma);
4359 if (bma->flist->xbf_low)
4362 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4363 if (bma->blkno == NULLFSBLOCK)
4365 if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4366 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4367 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4368 bma->cur->bc_private.b.flist = bma->flist;
4371 * Bump the number of extents we've allocated
4377 bma->cur->bc_private.b.flags =
4378 bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4380 bma->got.br_startoff = bma->offset;
4381 bma->got.br_startblock = bma->blkno;
4382 bma->got.br_blockcount = bma->length;
4383 bma->got.br_state = XFS_EXT_NORM;
4386 * A wasdelay extent has been initialized, so shouldn't be flagged
4389 if (!bma->wasdel && (bma->flags & XFS_BMAPI_PREALLOC) &&
4390 xfs_sb_version_hasextflgbit(&mp->m_sb))
4391 bma->got.br_state = XFS_EXT_UNWRITTEN;
4394 error = xfs_bmap_add_extent_delay_real(bma);
4396 error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4398 bma->logflags |= tmp_logflags;
4403 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4404 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4405 * the neighbouring ones.
4407 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4409 ASSERT(bma->got.br_startoff <= bma->offset);
4410 ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4411 bma->offset + bma->length);
4412 ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4413 bma->got.br_state == XFS_EXT_UNWRITTEN);
4418 xfs_bmapi_convert_unwritten(
4419 struct xfs_bmalloca *bma,
4420 struct xfs_bmbt_irec *mval,
4424 int whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4425 XFS_ATTR_FORK : XFS_DATA_FORK;
4426 struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4427 int tmp_logflags = 0;
4430 /* check if we need to do unwritten->real conversion */
4431 if (mval->br_state == XFS_EXT_UNWRITTEN &&
4432 (flags & XFS_BMAPI_PREALLOC))
4435 /* check if we need to do real->unwritten conversion */
4436 if (mval->br_state == XFS_EXT_NORM &&
4437 (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4438 (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4442 * Modify (by adding) the state flag, if writing.
4444 ASSERT(mval->br_blockcount <= len);
4445 if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4446 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4447 bma->ip, whichfork);
4448 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4449 bma->cur->bc_private.b.flist = bma->flist;
4451 mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4452 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4454 error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4455 &bma->cur, mval, bma->firstblock, bma->flist,
4457 bma->logflags |= tmp_logflags;
4462 * Update our extent pointer, given that
4463 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4464 * of the neighbouring ones.
4466 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4469 * We may have combined previously unwritten space with written space,
4470 * so generate another request.
4472 if (mval->br_blockcount < len)
4478 * Map file blocks to filesystem blocks, and allocate blocks or convert the
4479 * extent state if necessary. Details behaviour is controlled by the flags
4480 * parameter. Only allocates blocks from a single allocation group, to avoid
4483 * The returned value in "firstblock" from the first call in a transaction
4484 * must be remembered and presented to subsequent calls in "firstblock".
4485 * An upper bound for the number of blocks to be allocated is supplied to
4486 * the first call in "total"; if no allocation group has that many free
4487 * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4491 struct xfs_trans *tp, /* transaction pointer */
4492 struct xfs_inode *ip, /* incore inode */
4493 xfs_fileoff_t bno, /* starting file offs. mapped */
4494 xfs_filblks_t len, /* length to map in file */
4495 int flags, /* XFS_BMAPI_... */
4496 xfs_fsblock_t *firstblock, /* first allocated block
4497 controls a.g. for allocs */
4498 xfs_extlen_t total, /* total blocks needed */
4499 struct xfs_bmbt_irec *mval, /* output: map values */
4500 int *nmap, /* i/o: mval size/count */
4501 struct xfs_bmap_free *flist) /* i/o: list extents to free */
4503 struct xfs_mount *mp = ip->i_mount;
4504 struct xfs_ifork *ifp;
4505 struct xfs_bmalloca bma = { NULL }; /* args for xfs_bmap_alloc */
4506 xfs_fileoff_t end; /* end of mapped file region */
4507 int eof; /* after the end of extents */
4508 int error; /* error return */
4509 int n; /* current extent index */
4510 xfs_fileoff_t obno; /* old block number (offset) */
4511 int whichfork; /* data or attr fork */
4512 char inhole; /* current location is hole in file */
4513 char wasdelay; /* old extent was delayed */
4516 xfs_fileoff_t orig_bno; /* original block number value */
4517 int orig_flags; /* original flags arg value */
4518 xfs_filblks_t orig_len; /* original value of len arg */
4519 struct xfs_bmbt_irec *orig_mval; /* original value of mval */
4520 int orig_nmap; /* original value of *nmap */
4528 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4529 XFS_ATTR_FORK : XFS_DATA_FORK;
4532 ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4533 ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4536 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4537 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
4539 if (unlikely(XFS_TEST_ERROR(
4540 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4541 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4542 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4543 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4544 return -EFSCORRUPTED;
4547 if (XFS_FORCED_SHUTDOWN(mp))
4550 ifp = XFS_IFORK_PTR(ip, whichfork);
4552 XFS_STATS_INC(xs_blk_mapw);
4554 if (*firstblock == NULLFSBLOCK) {
4555 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4556 bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4563 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4564 error = xfs_iread_extents(tp, ip, whichfork);
4569 xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
4580 bma.firstblock = firstblock;
4582 while (bno < end && n < *nmap) {
4583 inhole = eof || bma.got.br_startoff > bno;
4584 wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
4587 * First, deal with the hole before the allocated space
4588 * that we found, if any.
4590 if (inhole || wasdelay) {
4592 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4593 bma.wasdel = wasdelay;
4598 * There's a 32/64 bit type mismatch between the
4599 * allocation length request (which can be 64 bits in
4600 * length) and the bma length request, which is
4601 * xfs_extlen_t and therefore 32 bits. Hence we have to
4602 * check for 32-bit overflows and handle them here.
4604 if (len > (xfs_filblks_t)MAXEXTLEN)
4605 bma.length = MAXEXTLEN;
4610 ASSERT(bma.length > 0);
4611 error = xfs_bmapi_allocate(&bma);
4614 if (bma.blkno == NULLFSBLOCK)
4618 /* Deal with the allocated space we found. */
4619 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4622 /* Execute unwritten extent conversion if necessary */
4623 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4624 if (error == -EAGAIN)
4629 /* update the extent map to return */
4630 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4633 * If we're done, stop now. Stop when we've allocated
4634 * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
4635 * the transaction may get too big.
4637 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4640 /* Else go on to the next record. */
4642 if (++bma.idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
4643 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
4651 * Transform from btree to extents, give it cur.
4653 if (xfs_bmap_wants_extents(ip, whichfork)) {
4654 int tmp_logflags = 0;
4657 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4658 &tmp_logflags, whichfork);
4659 bma.logflags |= tmp_logflags;
4664 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4665 XFS_IFORK_NEXTENTS(ip, whichfork) >
4666 XFS_IFORK_MAXEXT(ip, whichfork));
4670 * Log everything. Do this after conversion, there's no point in
4671 * logging the extent records if we've converted to btree format.
4673 if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
4674 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4675 bma.logflags &= ~xfs_ilog_fext(whichfork);
4676 else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
4677 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4678 bma.logflags &= ~xfs_ilog_fbroot(whichfork);
4680 * Log whatever the flags say, even if error. Otherwise we might miss
4681 * detecting a case where the data is changed, there's an error,
4682 * and it's not logged so we don't shutdown when we should.
4685 xfs_trans_log_inode(tp, ip, bma.logflags);
4689 ASSERT(*firstblock == NULLFSBLOCK ||
4690 XFS_FSB_TO_AGNO(mp, *firstblock) ==
4692 bma.cur->bc_private.b.firstblock) ||
4694 XFS_FSB_TO_AGNO(mp, *firstblock) <
4696 bma.cur->bc_private.b.firstblock)));
4697 *firstblock = bma.cur->bc_private.b.firstblock;
4699 xfs_btree_del_cursor(bma.cur,
4700 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4703 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4709 * Called by xfs_bmapi to update file extent records and the btree
4710 * after removing space (or undoing a delayed allocation).
4712 STATIC int /* error */
4713 xfs_bmap_del_extent(
4714 xfs_inode_t *ip, /* incore inode pointer */
4715 xfs_trans_t *tp, /* current transaction pointer */
4716 xfs_extnum_t *idx, /* extent number to update/delete */
4717 xfs_bmap_free_t *flist, /* list of extents to be freed */
4718 xfs_btree_cur_t *cur, /* if null, not a btree */
4719 xfs_bmbt_irec_t *del, /* data to remove from extents */
4720 int *logflagsp, /* inode logging flags */
4721 int whichfork) /* data or attr fork */
4723 xfs_filblks_t da_new; /* new delay-alloc indirect blocks */
4724 xfs_filblks_t da_old; /* old delay-alloc indirect blocks */
4725 xfs_fsblock_t del_endblock=0; /* first block past del */
4726 xfs_fileoff_t del_endoff; /* first offset past del */
4727 int delay; /* current block is delayed allocated */
4728 int do_fx; /* free extent at end of routine */
4729 xfs_bmbt_rec_host_t *ep; /* current extent entry pointer */
4730 int error; /* error return value */
4731 int flags; /* inode logging flags */
4732 xfs_bmbt_irec_t got; /* current extent entry */
4733 xfs_fileoff_t got_endoff; /* first offset past got */
4734 int i; /* temp state */
4735 xfs_ifork_t *ifp; /* inode fork pointer */
4736 xfs_mount_t *mp; /* mount structure */
4737 xfs_filblks_t nblks; /* quota/sb block count */
4738 xfs_bmbt_irec_t new; /* new record to be inserted */
4740 uint qfield; /* quota field to update */
4741 xfs_filblks_t temp; /* for indirect length calculations */
4742 xfs_filblks_t temp2; /* for indirect length calculations */
4745 XFS_STATS_INC(xs_del_exlist);
4747 if (whichfork == XFS_ATTR_FORK)
4748 state |= BMAP_ATTRFORK;
4751 ifp = XFS_IFORK_PTR(ip, whichfork);
4752 ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
4753 (uint)sizeof(xfs_bmbt_rec_t)));
4754 ASSERT(del->br_blockcount > 0);
4755 ep = xfs_iext_get_ext(ifp, *idx);
4756 xfs_bmbt_get_all(ep, &got);
4757 ASSERT(got.br_startoff <= del->br_startoff);
4758 del_endoff = del->br_startoff + del->br_blockcount;
4759 got_endoff = got.br_startoff + got.br_blockcount;
4760 ASSERT(got_endoff >= del_endoff);
4761 delay = isnullstartblock(got.br_startblock);
4762 ASSERT(isnullstartblock(del->br_startblock) == delay);
4767 * If deleting a real allocation, must free up the disk space.
4770 flags = XFS_ILOG_CORE;
4772 * Realtime allocation. Free it and record di_nblocks update.
4774 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
4778 ASSERT(do_mod(del->br_blockcount,
4779 mp->m_sb.sb_rextsize) == 0);
4780 ASSERT(do_mod(del->br_startblock,
4781 mp->m_sb.sb_rextsize) == 0);
4782 bno = del->br_startblock;
4783 len = del->br_blockcount;
4784 do_div(bno, mp->m_sb.sb_rextsize);
4785 do_div(len, mp->m_sb.sb_rextsize);
4786 error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
4790 nblks = len * mp->m_sb.sb_rextsize;
4791 qfield = XFS_TRANS_DQ_RTBCOUNT;
4794 * Ordinary allocation.
4798 nblks = del->br_blockcount;
4799 qfield = XFS_TRANS_DQ_BCOUNT;
4802 * Set up del_endblock and cur for later.
4804 del_endblock = del->br_startblock + del->br_blockcount;
4806 if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
4807 got.br_startblock, got.br_blockcount,
4810 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
4812 da_old = da_new = 0;
4814 da_old = startblockval(got.br_startblock);
4820 * Set flag value to use in switch statement.
4821 * Left-contig is 2, right-contig is 1.
4823 switch (((got.br_startoff == del->br_startoff) << 1) |
4824 (got_endoff == del_endoff)) {
4827 * Matches the whole extent. Delete the entry.
4829 xfs_iext_remove(ip, *idx, 1,
4830 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
4835 XFS_IFORK_NEXT_SET(ip, whichfork,
4836 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
4837 flags |= XFS_ILOG_CORE;
4839 flags |= xfs_ilog_fext(whichfork);
4842 if ((error = xfs_btree_delete(cur, &i)))
4844 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
4849 * Deleting the first part of the extent.
4851 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4852 xfs_bmbt_set_startoff(ep, del_endoff);
4853 temp = got.br_blockcount - del->br_blockcount;
4854 xfs_bmbt_set_blockcount(ep, temp);
4856 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4858 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4859 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4863 xfs_bmbt_set_startblock(ep, del_endblock);
4864 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4866 flags |= xfs_ilog_fext(whichfork);
4869 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
4870 got.br_blockcount - del->br_blockcount,
4877 * Deleting the last part of the extent.
4879 temp = got.br_blockcount - del->br_blockcount;
4880 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4881 xfs_bmbt_set_blockcount(ep, temp);
4883 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4885 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4886 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4890 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4892 flags |= xfs_ilog_fext(whichfork);
4895 if ((error = xfs_bmbt_update(cur, got.br_startoff,
4897 got.br_blockcount - del->br_blockcount,
4904 * Deleting the middle of the extent.
4906 temp = del->br_startoff - got.br_startoff;
4907 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4908 xfs_bmbt_set_blockcount(ep, temp);
4909 new.br_startoff = del_endoff;
4910 temp2 = got_endoff - del_endoff;
4911 new.br_blockcount = temp2;
4912 new.br_state = got.br_state;
4914 new.br_startblock = del_endblock;
4915 flags |= XFS_ILOG_CORE;
4917 if ((error = xfs_bmbt_update(cur,
4919 got.br_startblock, temp,
4922 if ((error = xfs_btree_increment(cur, 0, &i)))
4924 cur->bc_rec.b = new;
4925 error = xfs_btree_insert(cur, &i);
4926 if (error && error != -ENOSPC)
4929 * If get no-space back from btree insert,
4930 * it tried a split, and we have a zero
4931 * block reservation.
4932 * Fix up our state and return the error.
4934 if (error == -ENOSPC) {
4936 * Reset the cursor, don't trust
4937 * it after any insert operation.
4939 if ((error = xfs_bmbt_lookup_eq(cur,
4944 XFS_WANT_CORRUPTED_GOTO(mp,
4947 * Update the btree record back
4948 * to the original value.
4950 if ((error = xfs_bmbt_update(cur,
4957 * Reset the extent record back
4958 * to the original value.
4960 xfs_bmbt_set_blockcount(ep,
4966 XFS_WANT_CORRUPTED_GOTO(mp, i == 1, done);
4968 flags |= xfs_ilog_fext(whichfork);
4969 XFS_IFORK_NEXT_SET(ip, whichfork,
4970 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
4972 ASSERT(whichfork == XFS_DATA_FORK);
4973 temp = xfs_bmap_worst_indlen(ip, temp);
4974 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4975 temp2 = xfs_bmap_worst_indlen(ip, temp2);
4976 new.br_startblock = nullstartblock((int)temp2);
4977 da_new = temp + temp2;
4978 while (da_new > da_old) {
4982 xfs_bmbt_set_startblock(ep,
4983 nullstartblock((int)temp));
4985 if (da_new == da_old)
4991 nullstartblock((int)temp2);
4995 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4996 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
5001 * If we need to, add to list of extents to delete.
5004 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
5007 * Adjust inode # blocks in the file.
5010 ip->i_d.di_nblocks -= nblks;
5012 * Adjust quota data.
5015 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
5018 * Account for change in delayed indirect blocks.
5019 * Nothing to do for disk quota accounting here.
5021 ASSERT(da_old >= da_new);
5022 if (da_old > da_new) {
5023 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5024 (int64_t)(da_old - da_new), 0);
5032 * Unmap (remove) blocks from a file.
5033 * If nexts is nonzero then the number of extents to remove is limited to
5034 * that value. If not all extents in the block range can be removed then
5039 xfs_trans_t *tp, /* transaction pointer */
5040 struct xfs_inode *ip, /* incore inode */
5041 xfs_fileoff_t bno, /* starting offset to unmap */
5042 xfs_filblks_t len, /* length to unmap in file */
5043 int flags, /* misc flags */
5044 xfs_extnum_t nexts, /* number of extents max */
5045 xfs_fsblock_t *firstblock, /* first allocated block
5046 controls a.g. for allocs */
5047 xfs_bmap_free_t *flist, /* i/o: list extents to free */
5048 int *done) /* set if not done yet */
5050 xfs_btree_cur_t *cur; /* bmap btree cursor */
5051 xfs_bmbt_irec_t del; /* extent being deleted */
5052 int eof; /* is deleting at eof */
5053 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
5054 int error; /* error return value */
5055 xfs_extnum_t extno; /* extent number in list */
5056 xfs_bmbt_irec_t got; /* current extent record */
5057 xfs_ifork_t *ifp; /* inode fork pointer */
5058 int isrt; /* freeing in rt area */
5059 xfs_extnum_t lastx; /* last extent index used */
5060 int logflags; /* transaction logging flags */
5061 xfs_extlen_t mod; /* rt extent offset */
5062 xfs_mount_t *mp; /* mount structure */
5063 xfs_extnum_t nextents; /* number of file extents */
5064 xfs_bmbt_irec_t prev; /* previous extent record */
5065 xfs_fileoff_t start; /* first file offset deleted */
5066 int tmp_logflags; /* partial logging flags */
5067 int wasdel; /* was a delayed alloc extent */
5068 int whichfork; /* data or attribute fork */
5071 trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5073 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5074 XFS_ATTR_FORK : XFS_DATA_FORK;
5075 ifp = XFS_IFORK_PTR(ip, whichfork);
5077 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5078 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5079 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5081 return -EFSCORRUPTED;
5084 if (XFS_FORCED_SHUTDOWN(mp))
5087 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5091 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5092 (error = xfs_iread_extents(tp, ip, whichfork)))
5094 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5095 if (nextents == 0) {
5099 XFS_STATS_INC(xs_blk_unmap);
5100 isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5102 bno = start + len - 1;
5103 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5107 * Check to see if the given block number is past the end of the
5108 * file, back up to the last block if so...
5111 ep = xfs_iext_get_ext(ifp, --lastx);
5112 xfs_bmbt_get_all(ep, &got);
5113 bno = got.br_startoff + got.br_blockcount - 1;
5116 if (ifp->if_flags & XFS_IFBROOT) {
5117 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5118 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5119 cur->bc_private.b.firstblock = *firstblock;
5120 cur->bc_private.b.flist = flist;
5121 cur->bc_private.b.flags = 0;
5127 * Synchronize by locking the bitmap inode.
5129 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
5130 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5134 while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5135 (nexts == 0 || extno < nexts)) {
5137 * Is the found extent after a hole in which bno lives?
5138 * Just back up to the previous extent, if so.
5140 if (got.br_startoff > bno) {
5143 ep = xfs_iext_get_ext(ifp, lastx);
5144 xfs_bmbt_get_all(ep, &got);
5147 * Is the last block of this extent before the range
5148 * we're supposed to delete? If so, we're done.
5150 bno = XFS_FILEOFF_MIN(bno,
5151 got.br_startoff + got.br_blockcount - 1);
5155 * Then deal with the (possibly delayed) allocated space
5160 wasdel = isnullstartblock(del.br_startblock);
5161 if (got.br_startoff < start) {
5162 del.br_startoff = start;
5163 del.br_blockcount -= start - got.br_startoff;
5165 del.br_startblock += start - got.br_startoff;
5167 if (del.br_startoff + del.br_blockcount > bno + 1)
5168 del.br_blockcount = bno + 1 - del.br_startoff;
5169 sum = del.br_startblock + del.br_blockcount;
5171 (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5173 * Realtime extent not lined up at the end.
5174 * The extent could have been split into written
5175 * and unwritten pieces, or we could just be
5176 * unmapping part of it. But we can't really
5177 * get rid of part of a realtime extent.
5179 if (del.br_state == XFS_EXT_UNWRITTEN ||
5180 !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5182 * This piece is unwritten, or we're not
5183 * using unwritten extents. Skip over it.
5186 bno -= mod > del.br_blockcount ?
5187 del.br_blockcount : mod;
5188 if (bno < got.br_startoff) {
5190 xfs_bmbt_get_all(xfs_iext_get_ext(
5196 * It's written, turn it unwritten.
5197 * This is better than zeroing it.
5199 ASSERT(del.br_state == XFS_EXT_NORM);
5200 ASSERT(xfs_trans_get_block_res(tp) > 0);
5202 * If this spans a realtime extent boundary,
5203 * chop it back to the start of the one we end at.
5205 if (del.br_blockcount > mod) {
5206 del.br_startoff += del.br_blockcount - mod;
5207 del.br_startblock += del.br_blockcount - mod;
5208 del.br_blockcount = mod;
5210 del.br_state = XFS_EXT_UNWRITTEN;
5211 error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5212 &lastx, &cur, &del, firstblock, flist,
5218 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5220 * Realtime extent is lined up at the end but not
5221 * at the front. We'll get rid of full extents if
5224 mod = mp->m_sb.sb_rextsize - mod;
5225 if (del.br_blockcount > mod) {
5226 del.br_blockcount -= mod;
5227 del.br_startoff += mod;
5228 del.br_startblock += mod;
5229 } else if ((del.br_startoff == start &&
5230 (del.br_state == XFS_EXT_UNWRITTEN ||
5231 xfs_trans_get_block_res(tp) == 0)) ||
5232 !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5234 * Can't make it unwritten. There isn't
5235 * a full extent here so just skip it.
5237 ASSERT(bno >= del.br_blockcount);
5238 bno -= del.br_blockcount;
5239 if (got.br_startoff > bno) {
5241 ep = xfs_iext_get_ext(ifp,
5243 xfs_bmbt_get_all(ep, &got);
5247 } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5249 * This one is already unwritten.
5250 * It must have a written left neighbor.
5251 * Unwrite the killed part of that one and
5255 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5257 ASSERT(prev.br_state == XFS_EXT_NORM);
5258 ASSERT(!isnullstartblock(prev.br_startblock));
5259 ASSERT(del.br_startblock ==
5260 prev.br_startblock + prev.br_blockcount);
5261 if (prev.br_startoff < start) {
5262 mod = start - prev.br_startoff;
5263 prev.br_blockcount -= mod;
5264 prev.br_startblock += mod;
5265 prev.br_startoff = start;
5267 prev.br_state = XFS_EXT_UNWRITTEN;
5269 error = xfs_bmap_add_extent_unwritten_real(tp,
5270 ip, &lastx, &cur, &prev,
5271 firstblock, flist, &logflags);
5276 ASSERT(del.br_state == XFS_EXT_NORM);
5277 del.br_state = XFS_EXT_UNWRITTEN;
5278 error = xfs_bmap_add_extent_unwritten_real(tp,
5279 ip, &lastx, &cur, &del,
5280 firstblock, flist, &logflags);
5287 ASSERT(startblockval(del.br_startblock) > 0);
5288 /* Update realtime/data freespace, unreserve quota */
5290 xfs_filblks_t rtexts;
5292 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5293 do_div(rtexts, mp->m_sb.sb_rextsize);
5294 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5295 (int64_t)rtexts, 0);
5296 (void)xfs_trans_reserve_quota_nblks(NULL,
5297 ip, -((long)del.br_blockcount), 0,
5298 XFS_QMOPT_RES_RTBLKS);
5300 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5301 (int64_t)del.br_blockcount, 0);
5302 (void)xfs_trans_reserve_quota_nblks(NULL,
5303 ip, -((long)del.br_blockcount), 0,
5304 XFS_QMOPT_RES_REGBLKS);
5306 ip->i_delayed_blks -= del.br_blockcount;
5308 cur->bc_private.b.flags |=
5309 XFS_BTCUR_BPRV_WASDEL;
5311 cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5313 * If it's the case where the directory code is running
5314 * with no block reservation, and the deleted block is in
5315 * the middle of its extent, and the resulting insert
5316 * of an extent would cause transformation to btree format,
5317 * then reject it. The calling code will then swap
5318 * blocks around instead.
5319 * We have to do this now, rather than waiting for the
5320 * conversion to btree format, since the transaction
5323 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5324 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5325 XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5326 XFS_IFORK_MAXEXT(ip, whichfork) &&
5327 del.br_startoff > got.br_startoff &&
5328 del.br_startoff + del.br_blockcount <
5329 got.br_startoff + got.br_blockcount) {
5333 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5334 &tmp_logflags, whichfork);
5335 logflags |= tmp_logflags;
5338 bno = del.br_startoff - 1;
5341 * If not done go on to the next (previous) record.
5343 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5345 ep = xfs_iext_get_ext(ifp, lastx);
5346 if (xfs_bmbt_get_startoff(ep) > bno) {
5348 ep = xfs_iext_get_ext(ifp,
5351 xfs_bmbt_get_all(ep, &got);
5356 *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5359 * Convert to a btree if necessary.
5361 if (xfs_bmap_needs_btree(ip, whichfork)) {
5362 ASSERT(cur == NULL);
5363 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5364 &cur, 0, &tmp_logflags, whichfork);
5365 logflags |= tmp_logflags;
5370 * transform from btree to extents, give it cur
5372 else if (xfs_bmap_wants_extents(ip, whichfork)) {
5373 ASSERT(cur != NULL);
5374 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5376 logflags |= tmp_logflags;
5381 * transform from extents to local?
5386 * Log everything. Do this after conversion, there's no point in
5387 * logging the extent records if we've converted to btree format.
5389 if ((logflags & xfs_ilog_fext(whichfork)) &&
5390 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5391 logflags &= ~xfs_ilog_fext(whichfork);
5392 else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5393 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5394 logflags &= ~xfs_ilog_fbroot(whichfork);
5396 * Log inode even in the error case, if the transaction
5397 * is dirty we'll need to shut down the filesystem.
5400 xfs_trans_log_inode(tp, ip, logflags);
5403 *firstblock = cur->bc_private.b.firstblock;
5404 cur->bc_private.b.allocated = 0;
5406 xfs_btree_del_cursor(cur,
5407 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5413 * Determine whether an extent shift can be accomplished by a merge with the
5414 * extent that precedes the target hole of the shift.
5418 struct xfs_bmbt_irec *left, /* preceding extent */
5419 struct xfs_bmbt_irec *got, /* current extent to shift */
5420 xfs_fileoff_t shift) /* shift fsb */
5422 xfs_fileoff_t startoff;
5424 startoff = got->br_startoff - shift;
5427 * The extent, once shifted, must be adjacent in-file and on-disk with
5428 * the preceding extent.
5430 if ((left->br_startoff + left->br_blockcount != startoff) ||
5431 (left->br_startblock + left->br_blockcount != got->br_startblock) ||
5432 (left->br_state != got->br_state) ||
5433 (left->br_blockcount + got->br_blockcount > MAXEXTLEN))
5440 * A bmap extent shift adjusts the file offset of an extent to fill a preceding
5441 * hole in the file. If an extent shift would result in the extent being fully
5442 * adjacent to the extent that currently precedes the hole, we can merge with
5443 * the preceding extent rather than do the shift.
5445 * This function assumes the caller has verified a shift-by-merge is possible
5446 * with the provided extents via xfs_bmse_can_merge().
5450 struct xfs_inode *ip,
5452 xfs_fileoff_t shift, /* shift fsb */
5453 int current_ext, /* idx of gotp */
5454 struct xfs_bmbt_rec_host *gotp, /* extent to shift */
5455 struct xfs_bmbt_rec_host *leftp, /* preceding extent */
5456 struct xfs_btree_cur *cur,
5457 int *logflags) /* output */
5459 struct xfs_bmbt_irec got;
5460 struct xfs_bmbt_irec left;
5461 xfs_filblks_t blockcount;
5463 struct xfs_mount *mp = ip->i_mount;
5465 xfs_bmbt_get_all(gotp, &got);
5466 xfs_bmbt_get_all(leftp, &left);
5467 blockcount = left.br_blockcount + got.br_blockcount;
5469 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5470 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5471 ASSERT(xfs_bmse_can_merge(&left, &got, shift));
5474 * Merge the in-core extents. Note that the host record pointers and
5475 * current_ext index are invalid once the extent has been removed via
5476 * xfs_iext_remove().
5478 xfs_bmbt_set_blockcount(leftp, blockcount);
5479 xfs_iext_remove(ip, current_ext, 1, 0);
5482 * Update the on-disk extent count, the btree if necessary and log the
5485 XFS_IFORK_NEXT_SET(ip, whichfork,
5486 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
5487 *logflags |= XFS_ILOG_CORE;
5489 *logflags |= XFS_ILOG_DEXT;
5493 /* lookup and remove the extent to merge */
5494 error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
5495 got.br_blockcount, &i);
5498 XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5500 error = xfs_btree_delete(cur, &i);
5503 XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5505 /* lookup and update size of the previous extent */
5506 error = xfs_bmbt_lookup_eq(cur, left.br_startoff, left.br_startblock,
5507 left.br_blockcount, &i);
5510 XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5512 left.br_blockcount = blockcount;
5514 return xfs_bmbt_update(cur, left.br_startoff, left.br_startblock,
5515 left.br_blockcount, left.br_state);
5519 * Shift a single extent.
5523 struct xfs_inode *ip,
5525 xfs_fileoff_t offset_shift_fsb,
5527 struct xfs_bmbt_rec_host *gotp,
5528 struct xfs_btree_cur *cur,
5531 struct xfs_ifork *ifp;
5532 struct xfs_mount *mp;
5533 xfs_fileoff_t startoff;
5534 struct xfs_bmbt_rec_host *leftp;
5535 struct xfs_bmbt_irec got;
5536 struct xfs_bmbt_irec left;
5541 ifp = XFS_IFORK_PTR(ip, whichfork);
5543 xfs_bmbt_get_all(gotp, &got);
5544 startoff = got.br_startoff - offset_shift_fsb;
5546 /* delalloc extents should be prevented by caller */
5547 XFS_WANT_CORRUPTED_RETURN(mp, !isnullstartblock(got.br_startblock));
5550 * Check for merge if we've got an extent to the left, otherwise make
5551 * sure there's enough room at the start of the file for the shift.
5554 /* grab the left extent and check for a large enough hole */
5555 leftp = xfs_iext_get_ext(ifp, *current_ext - 1);
5556 xfs_bmbt_get_all(leftp, &left);
5558 if (startoff < left.br_startoff + left.br_blockcount)
5561 /* check whether to merge the extent or shift it down */
5562 if (xfs_bmse_can_merge(&left, &got, offset_shift_fsb)) {
5563 return xfs_bmse_merge(ip, whichfork, offset_shift_fsb,
5564 *current_ext, gotp, leftp, cur,
5567 } else if (got.br_startoff < offset_shift_fsb)
5571 * Increment the extent index for the next iteration, update the start
5572 * offset of the in-core extent and update the btree if applicable.
5575 xfs_bmbt_set_startoff(gotp, startoff);
5576 *logflags |= XFS_ILOG_CORE;
5578 *logflags |= XFS_ILOG_DEXT;
5582 error = xfs_bmbt_lookup_eq(cur, got.br_startoff, got.br_startblock,
5583 got.br_blockcount, &i);
5586 XFS_WANT_CORRUPTED_RETURN(mp, i == 1);
5588 got.br_startoff = startoff;
5589 return xfs_bmbt_update(cur, got.br_startoff, got.br_startblock,
5590 got.br_blockcount, got.br_state);
5594 * Shift extent records to the left to cover a hole.
5596 * The maximum number of extents to be shifted in a single operation is
5597 * @num_exts. @start_fsb specifies the file offset to start the shift and the
5598 * file offset where we've left off is returned in @next_fsb. @offset_shift_fsb
5599 * is the length by which each extent is shifted. If there is no hole to shift
5600 * the extents into, this will be considered invalid operation and we abort
5604 xfs_bmap_shift_extents(
5605 struct xfs_trans *tp,
5606 struct xfs_inode *ip,
5607 xfs_fileoff_t start_fsb,
5608 xfs_fileoff_t offset_shift_fsb,
5610 xfs_fileoff_t *next_fsb,
5611 xfs_fsblock_t *firstblock,
5612 struct xfs_bmap_free *flist,
5615 struct xfs_btree_cur *cur = NULL;
5616 struct xfs_bmbt_rec_host *gotp;
5617 struct xfs_bmbt_irec got;
5618 struct xfs_mount *mp = ip->i_mount;
5619 struct xfs_ifork *ifp;
5620 xfs_extnum_t nexts = 0;
5621 xfs_extnum_t current_ext;
5623 int whichfork = XFS_DATA_FORK;
5627 if (unlikely(XFS_TEST_ERROR(
5628 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5629 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
5630 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
5631 XFS_ERROR_REPORT("xfs_bmap_shift_extents",
5632 XFS_ERRLEVEL_LOW, mp);
5633 return -EFSCORRUPTED;
5636 if (XFS_FORCED_SHUTDOWN(mp))
5639 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
5640 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
5642 ifp = XFS_IFORK_PTR(ip, whichfork);
5643 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
5644 /* Read in all the extents */
5645 error = xfs_iread_extents(tp, ip, whichfork);
5650 if (ifp->if_flags & XFS_IFBROOT) {
5651 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5652 cur->bc_private.b.firstblock = *firstblock;
5653 cur->bc_private.b.flist = flist;
5654 cur->bc_private.b.flags = 0;
5658 * Look up the extent index for the fsb where we start shifting. We can
5659 * henceforth iterate with current_ext as extent list changes are locked
5662 * gotp can be null in 2 cases: 1) if there are no extents or 2)
5663 * start_fsb lies in a hole beyond which there are no extents. Either
5666 gotp = xfs_iext_bno_to_ext(ifp, start_fsb, ¤t_ext);
5673 * There may be delalloc extents in the data fork before the range we
5674 * are collapsing out, so we cannot use the count of real extents here.
5675 * Instead we have to calculate it from the incore fork.
5677 total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
5678 while (nexts++ < num_exts && current_ext < total_extents) {
5679 error = xfs_bmse_shift_one(ip, whichfork, offset_shift_fsb,
5680 ¤t_ext, gotp, cur, &logflags);
5684 /* update total extent count and grab the next record */
5685 total_extents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
5686 if (current_ext >= total_extents)
5688 gotp = xfs_iext_get_ext(ifp, current_ext);
5691 /* Check if we are done */
5692 if (current_ext == total_extents) {
5694 } else if (next_fsb) {
5695 xfs_bmbt_get_all(gotp, &got);
5696 *next_fsb = got.br_startoff;
5701 xfs_btree_del_cursor(cur,
5702 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
5705 xfs_trans_log_inode(tp, ip, logflags);