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_format.h"
21 #include "xfs_shared.h"
25 #include "xfs_trans.h"
28 #include "xfs_mount.h"
29 #include "xfs_da_btree.h"
30 #include "xfs_dir2_format.h"
32 #include "xfs_bmap_btree.h"
33 #include "xfs_alloc_btree.h"
34 #include "xfs_ialloc_btree.h"
35 #include "xfs_dinode.h"
36 #include "xfs_inode.h"
37 #include "xfs_btree.h"
38 #include "xfs_mount.h"
39 #include "xfs_itable.h"
40 #include "xfs_inode_item.h"
41 #include "xfs_extfree_item.h"
42 #include "xfs_alloc.h"
44 #include "xfs_bmap_util.h"
45 #include "xfs_rtalloc.h"
46 #include "xfs_error.h"
47 #include "xfs_attr_leaf.h"
48 #include "xfs_quota.h"
49 #include "xfs_trans_space.h"
50 #include "xfs_buf_item.h"
51 #include "xfs_filestream.h"
52 #include "xfs_trace.h"
53 #include "xfs_symlink.h"
56 kmem_zone_t *xfs_bmap_free_item_zone;
59 * Miscellaneous helper functions
63 * Compute and fill in the value of the maximum depth of a bmap btree
64 * in this filesystem. Done once, during mount.
67 xfs_bmap_compute_maxlevels(
68 xfs_mount_t *mp, /* file system mount structure */
69 int whichfork) /* data or attr fork */
71 int level; /* btree level */
72 uint maxblocks; /* max blocks at this level */
73 uint maxleafents; /* max leaf entries possible */
74 int maxrootrecs; /* max records in root block */
75 int minleafrecs; /* min records in leaf block */
76 int minnoderecs; /* min records in node block */
77 int sz; /* root block size */
80 * The maximum number of extents in a file, hence the maximum
81 * number of leaf entries, is controlled by the type of di_nextents
82 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
83 * (a signed 16-bit number, xfs_aextnum_t).
85 * Note that we can no longer assume that if we are in ATTR1 that
86 * the fork offset of all the inodes will be
87 * (xfs_default_attroffset(ip) >> 3) because we could have mounted
88 * with ATTR2 and then mounted back with ATTR1, keeping the
89 * di_forkoff's fixed but probably at various positions. Therefore,
90 * for both ATTR1 and ATTR2 we have to assume the worst case scenario
91 * of a minimum size available.
93 if (whichfork == XFS_DATA_FORK) {
94 maxleafents = MAXEXTNUM;
95 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
97 maxleafents = MAXAEXTNUM;
98 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
100 maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
101 minleafrecs = mp->m_bmap_dmnr[0];
102 minnoderecs = mp->m_bmap_dmnr[1];
103 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
104 for (level = 1; maxblocks > 1; level++) {
105 if (maxblocks <= maxrootrecs)
108 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
110 mp->m_bm_maxlevels[whichfork] = level;
113 STATIC int /* error */
115 struct xfs_btree_cur *cur,
119 int *stat) /* success/failure */
121 cur->bc_rec.b.br_startoff = off;
122 cur->bc_rec.b.br_startblock = bno;
123 cur->bc_rec.b.br_blockcount = len;
124 return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
127 STATIC int /* error */
129 struct xfs_btree_cur *cur,
133 int *stat) /* success/failure */
135 cur->bc_rec.b.br_startoff = off;
136 cur->bc_rec.b.br_startblock = bno;
137 cur->bc_rec.b.br_blockcount = len;
138 return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
142 * Check if the inode needs to be converted to btree format.
144 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
146 return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
147 XFS_IFORK_NEXTENTS(ip, whichfork) >
148 XFS_IFORK_MAXEXT(ip, whichfork);
152 * Check if the inode should be converted to extent format.
154 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
156 return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
157 XFS_IFORK_NEXTENTS(ip, whichfork) <=
158 XFS_IFORK_MAXEXT(ip, whichfork);
162 * Update the record referred to by cur to the value given
163 * by [off, bno, len, state].
164 * This either works (return 0) or gets an EFSCORRUPTED error.
168 struct xfs_btree_cur *cur,
174 union xfs_btree_rec rec;
176 xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
177 return xfs_btree_update(cur, &rec);
181 * Compute the worst-case number of indirect blocks that will be used
182 * for ip's delayed extent of length "len".
185 xfs_bmap_worst_indlen(
186 xfs_inode_t *ip, /* incore inode pointer */
187 xfs_filblks_t len) /* delayed extent length */
189 int level; /* btree level number */
190 int maxrecs; /* maximum record count at this level */
191 xfs_mount_t *mp; /* mount structure */
192 xfs_filblks_t rval; /* return value */
195 maxrecs = mp->m_bmap_dmxr[0];
196 for (level = 0, rval = 0;
197 level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
200 do_div(len, maxrecs);
203 return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
206 maxrecs = mp->m_bmap_dmxr[1];
212 * Calculate the default attribute fork offset for newly created inodes.
215 xfs_default_attroffset(
216 struct xfs_inode *ip)
218 struct xfs_mount *mp = ip->i_mount;
221 if (mp->m_sb.sb_inodesize == 256) {
222 offset = XFS_LITINO(mp, ip->i_d.di_version) -
223 XFS_BMDR_SPACE_CALC(MINABTPTRS);
225 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
228 ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version));
233 * Helper routine to reset inode di_forkoff field when switching
234 * attribute fork from local to extent format - we reset it where
235 * possible to make space available for inline data fork extents.
238 xfs_bmap_forkoff_reset(
243 if (whichfork == XFS_ATTR_FORK &&
244 ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
245 ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
246 ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
247 uint dfl_forkoff = xfs_default_attroffset(ip) >> 3;
249 if (dfl_forkoff > ip->i_d.di_forkoff)
250 ip->i_d.di_forkoff = dfl_forkoff;
255 * Debug/sanity checking code
259 xfs_bmap_sanity_check(
260 struct xfs_mount *mp,
264 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
266 if (block->bb_magic != cpu_to_be32(XFS_BMAP_CRC_MAGIC) &&
267 block->bb_magic != cpu_to_be32(XFS_BMAP_MAGIC))
270 if (be16_to_cpu(block->bb_level) != level ||
271 be16_to_cpu(block->bb_numrecs) == 0 ||
272 be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
279 STATIC struct xfs_buf *
281 struct xfs_btree_cur *cur,
284 struct xfs_log_item_desc *lidp;
290 for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
291 if (!cur->bc_bufs[i])
293 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
294 return cur->bc_bufs[i];
297 /* Chase down all the log items to see if the bp is there */
298 list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
299 struct xfs_buf_log_item *bip;
300 bip = (struct xfs_buf_log_item *)lidp->lid_item;
301 if (bip->bli_item.li_type == XFS_LI_BUF &&
302 XFS_BUF_ADDR(bip->bli_buf) == bno)
311 struct xfs_btree_block *block,
317 __be64 *pp, *thispa; /* pointer to block address */
318 xfs_bmbt_key_t *prevp, *keyp;
320 ASSERT(be16_to_cpu(block->bb_level) > 0);
323 for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
324 dmxr = mp->m_bmap_dmxr[0];
325 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
328 ASSERT(be64_to_cpu(prevp->br_startoff) <
329 be64_to_cpu(keyp->br_startoff));
334 * Compare the block numbers to see if there are dups.
337 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
339 pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
341 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
343 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
345 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
346 if (*thispa == *pp) {
347 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
349 (unsigned long long)be64_to_cpu(*thispa));
350 panic("%s: ptrs are equal in node\n",
358 * Check that the extents for the inode ip are in the right order in all
363 xfs_bmap_check_leaf_extents(
364 xfs_btree_cur_t *cur, /* btree cursor or null */
365 xfs_inode_t *ip, /* incore inode pointer */
366 int whichfork) /* data or attr fork */
368 struct xfs_btree_block *block; /* current btree block */
369 xfs_fsblock_t bno; /* block # of "block" */
370 xfs_buf_t *bp; /* buffer for "block" */
371 int error; /* error return value */
372 xfs_extnum_t i=0, j; /* index into the extents list */
373 xfs_ifork_t *ifp; /* fork structure */
374 int level; /* btree level, for checking */
375 xfs_mount_t *mp; /* file system mount structure */
376 __be64 *pp; /* pointer to block address */
377 xfs_bmbt_rec_t *ep; /* pointer to current extent */
378 xfs_bmbt_rec_t last = {0, 0}; /* last extent in prev block */
379 xfs_bmbt_rec_t *nextp; /* pointer to next extent */
382 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
388 ifp = XFS_IFORK_PTR(ip, whichfork);
389 block = ifp->if_broot;
391 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
393 level = be16_to_cpu(block->bb_level);
395 xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
396 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
397 bno = be64_to_cpu(*pp);
399 ASSERT(bno != NULLDFSBNO);
400 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
401 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
404 * Go down the tree until leaf level is reached, following the first
405 * pointer (leftmost) at each level.
407 while (level-- > 0) {
408 /* See if buf is in cur first */
410 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
413 error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
419 block = XFS_BUF_TO_BLOCK(bp);
420 XFS_WANT_CORRUPTED_GOTO(
421 xfs_bmap_sanity_check(mp, bp, level),
427 * Check this block for basic sanity (increasing keys and
428 * no duplicate blocks).
431 xfs_check_block(block, mp, 0, 0);
432 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
433 bno = be64_to_cpu(*pp);
434 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
437 xfs_trans_brelse(NULL, bp);
442 * Here with bp and block set to the leftmost leaf node in the tree.
447 * Loop over all leaf nodes checking that all extents are in the right order.
450 xfs_fsblock_t nextbno;
451 xfs_extnum_t num_recs;
454 num_recs = xfs_btree_get_numrecs(block);
457 * Read-ahead the next leaf block, if any.
460 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
463 * Check all the extents to make sure they are OK.
464 * If we had a previous block, the last entry should
465 * conform with the first entry in this one.
468 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
470 ASSERT(xfs_bmbt_disk_get_startoff(&last) +
471 xfs_bmbt_disk_get_blockcount(&last) <=
472 xfs_bmbt_disk_get_startoff(ep));
474 for (j = 1; j < num_recs; j++) {
475 nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
476 ASSERT(xfs_bmbt_disk_get_startoff(ep) +
477 xfs_bmbt_disk_get_blockcount(ep) <=
478 xfs_bmbt_disk_get_startoff(nextp));
486 xfs_trans_brelse(NULL, bp);
490 * If we've reached the end, stop.
492 if (bno == NULLFSBLOCK)
496 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
499 error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
505 block = XFS_BUF_TO_BLOCK(bp);
509 xfs_trans_brelse(NULL, bp);
514 xfs_warn(mp, "%s: at error0", __func__);
516 xfs_trans_brelse(NULL, bp);
518 xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
520 panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
525 * Add bmap trace insert entries for all the contents of the extent records.
528 xfs_bmap_trace_exlist(
529 xfs_inode_t *ip, /* incore inode pointer */
530 xfs_extnum_t cnt, /* count of entries in the list */
531 int whichfork, /* data or attr fork */
532 unsigned long caller_ip)
534 xfs_extnum_t idx; /* extent record index */
535 xfs_ifork_t *ifp; /* inode fork pointer */
538 if (whichfork == XFS_ATTR_FORK)
539 state |= BMAP_ATTRFORK;
541 ifp = XFS_IFORK_PTR(ip, whichfork);
542 ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
543 for (idx = 0; idx < cnt; idx++)
544 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
548 * Validate that the bmbt_irecs being returned from bmapi are valid
549 * given the caller's original parameters. Specifically check the
550 * ranges of the returned irecs to ensure that they only extend beyond
551 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
554 xfs_bmap_validate_ret(
558 xfs_bmbt_irec_t *mval,
562 int i; /* index to map values */
564 ASSERT(ret_nmap <= nmap);
566 for (i = 0; i < ret_nmap; i++) {
567 ASSERT(mval[i].br_blockcount > 0);
568 if (!(flags & XFS_BMAPI_ENTIRE)) {
569 ASSERT(mval[i].br_startoff >= bno);
570 ASSERT(mval[i].br_blockcount <= len);
571 ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
574 ASSERT(mval[i].br_startoff < bno + len);
575 ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
579 mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
580 mval[i].br_startoff);
581 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
582 mval[i].br_startblock != HOLESTARTBLOCK);
583 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
584 mval[i].br_state == XFS_EXT_UNWRITTEN);
589 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork) do { } while (0)
590 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
594 * bmap free list manipulation functions
598 * Add the extent to the list of extents to be free at transaction end.
599 * The list is maintained sorted (by block number).
603 xfs_fsblock_t bno, /* fs block number of extent */
604 xfs_filblks_t len, /* length of extent */
605 xfs_bmap_free_t *flist, /* list of extents */
606 xfs_mount_t *mp) /* mount point structure */
608 xfs_bmap_free_item_t *cur; /* current (next) element */
609 xfs_bmap_free_item_t *new; /* new element */
610 xfs_bmap_free_item_t *prev; /* previous element */
615 ASSERT(bno != NULLFSBLOCK);
617 ASSERT(len <= MAXEXTLEN);
618 ASSERT(!isnullstartblock(bno));
619 agno = XFS_FSB_TO_AGNO(mp, bno);
620 agbno = XFS_FSB_TO_AGBNO(mp, bno);
621 ASSERT(agno < mp->m_sb.sb_agcount);
622 ASSERT(agbno < mp->m_sb.sb_agblocks);
623 ASSERT(len < mp->m_sb.sb_agblocks);
624 ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
626 ASSERT(xfs_bmap_free_item_zone != NULL);
627 new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
628 new->xbfi_startblock = bno;
629 new->xbfi_blockcount = (xfs_extlen_t)len;
630 for (prev = NULL, cur = flist->xbf_first;
632 prev = cur, cur = cur->xbfi_next) {
633 if (cur->xbfi_startblock >= bno)
637 prev->xbfi_next = new;
639 flist->xbf_first = new;
640 new->xbfi_next = cur;
645 * Remove the entry "free" from the free item list. Prev points to the
646 * previous entry, unless "free" is the head of the list.
650 xfs_bmap_free_t *flist, /* free item list header */
651 xfs_bmap_free_item_t *prev, /* previous item on list, if any */
652 xfs_bmap_free_item_t *free) /* list item to be freed */
655 prev->xbfi_next = free->xbfi_next;
657 flist->xbf_first = free->xbfi_next;
659 kmem_zone_free(xfs_bmap_free_item_zone, free);
663 * Free up any items left in the list.
667 xfs_bmap_free_t *flist) /* list of bmap_free_items */
669 xfs_bmap_free_item_t *free; /* free list item */
670 xfs_bmap_free_item_t *next;
672 if (flist->xbf_count == 0)
674 ASSERT(flist->xbf_first != NULL);
675 for (free = flist->xbf_first; free; free = next) {
676 next = free->xbfi_next;
677 xfs_bmap_del_free(flist, NULL, free);
679 ASSERT(flist->xbf_count == 0);
683 * Inode fork format manipulation functions
687 * Transform a btree format file with only one leaf node, where the
688 * extents list will fit in the inode, into an extents format file.
689 * Since the file extents are already in-core, all we have to do is
690 * give up the space for the btree root and pitch the leaf block.
692 STATIC int /* error */
693 xfs_bmap_btree_to_extents(
694 xfs_trans_t *tp, /* transaction pointer */
695 xfs_inode_t *ip, /* incore inode pointer */
696 xfs_btree_cur_t *cur, /* btree cursor */
697 int *logflagsp, /* inode logging flags */
698 int whichfork) /* data or attr fork */
701 struct xfs_btree_block *cblock;/* child btree block */
702 xfs_fsblock_t cbno; /* child block number */
703 xfs_buf_t *cbp; /* child block's buffer */
704 int error; /* error return value */
705 xfs_ifork_t *ifp; /* inode fork data */
706 xfs_mount_t *mp; /* mount point structure */
707 __be64 *pp; /* ptr to block address */
708 struct xfs_btree_block *rblock;/* root btree block */
711 ifp = XFS_IFORK_PTR(ip, whichfork);
712 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
713 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
714 rblock = ifp->if_broot;
715 ASSERT(be16_to_cpu(rblock->bb_level) == 1);
716 ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
717 ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
718 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
719 cbno = be64_to_cpu(*pp);
722 if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
725 error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
729 cblock = XFS_BUF_TO_BLOCK(cbp);
730 if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
732 xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
733 ip->i_d.di_nblocks--;
734 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
735 xfs_trans_binval(tp, cbp);
736 if (cur->bc_bufs[0] == cbp)
737 cur->bc_bufs[0] = NULL;
738 xfs_iroot_realloc(ip, -1, whichfork);
739 ASSERT(ifp->if_broot == NULL);
740 ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
741 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
742 *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
747 * Convert an extents-format file into a btree-format file.
748 * The new file will have a root block (in the inode) and a single child block.
750 STATIC int /* error */
751 xfs_bmap_extents_to_btree(
752 xfs_trans_t *tp, /* transaction pointer */
753 xfs_inode_t *ip, /* incore inode pointer */
754 xfs_fsblock_t *firstblock, /* first-block-allocated */
755 xfs_bmap_free_t *flist, /* blocks freed in xaction */
756 xfs_btree_cur_t **curp, /* cursor returned to caller */
757 int wasdel, /* converting a delayed alloc */
758 int *logflagsp, /* inode logging flags */
759 int whichfork) /* data or attr fork */
761 struct xfs_btree_block *ablock; /* allocated (child) bt block */
762 xfs_buf_t *abp; /* buffer for ablock */
763 xfs_alloc_arg_t args; /* allocation arguments */
764 xfs_bmbt_rec_t *arp; /* child record pointer */
765 struct xfs_btree_block *block; /* btree root block */
766 xfs_btree_cur_t *cur; /* bmap btree cursor */
767 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
768 int error; /* error return value */
769 xfs_extnum_t i, cnt; /* extent record index */
770 xfs_ifork_t *ifp; /* inode fork pointer */
771 xfs_bmbt_key_t *kp; /* root block key pointer */
772 xfs_mount_t *mp; /* mount structure */
773 xfs_extnum_t nextents; /* number of file extents */
774 xfs_bmbt_ptr_t *pp; /* root block address pointer */
777 ifp = XFS_IFORK_PTR(ip, whichfork);
778 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
781 * Make space in the inode incore.
783 xfs_iroot_realloc(ip, 1, whichfork);
784 ifp->if_flags |= XFS_IFBROOT;
789 block = ifp->if_broot;
790 if (xfs_sb_version_hascrc(&mp->m_sb))
791 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
792 XFS_BMAP_CRC_MAGIC, 1, 1, ip->i_ino,
793 XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
795 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
796 XFS_BMAP_MAGIC, 1, 1, ip->i_ino,
797 XFS_BTREE_LONG_PTRS);
800 * Need a cursor. Can't allocate until bb_level is filled in.
802 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
803 cur->bc_private.b.firstblock = *firstblock;
804 cur->bc_private.b.flist = flist;
805 cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
807 * Convert to a btree with two levels, one record in root.
809 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
810 memset(&args, 0, sizeof(args));
813 args.firstblock = *firstblock;
814 if (*firstblock == NULLFSBLOCK) {
815 args.type = XFS_ALLOCTYPE_START_BNO;
816 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
817 } else if (flist->xbf_low) {
818 args.type = XFS_ALLOCTYPE_START_BNO;
819 args.fsbno = *firstblock;
821 args.type = XFS_ALLOCTYPE_NEAR_BNO;
822 args.fsbno = *firstblock;
824 args.minlen = args.maxlen = args.prod = 1;
825 args.wasdel = wasdel;
827 if ((error = xfs_alloc_vextent(&args))) {
828 xfs_iroot_realloc(ip, -1, whichfork);
829 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
833 * Allocation can't fail, the space was reserved.
835 ASSERT(args.fsbno != NULLFSBLOCK);
836 ASSERT(*firstblock == NULLFSBLOCK ||
837 args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
839 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
840 *firstblock = cur->bc_private.b.firstblock = args.fsbno;
841 cur->bc_private.b.allocated++;
842 ip->i_d.di_nblocks++;
843 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
844 abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
846 * Fill in the child block.
848 abp->b_ops = &xfs_bmbt_buf_ops;
849 ablock = XFS_BUF_TO_BLOCK(abp);
850 if (xfs_sb_version_hascrc(&mp->m_sb))
851 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
852 XFS_BMAP_CRC_MAGIC, 0, 0, ip->i_ino,
853 XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
855 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
856 XFS_BMAP_MAGIC, 0, 0, ip->i_ino,
857 XFS_BTREE_LONG_PTRS);
859 arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
860 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
861 for (cnt = i = 0; i < nextents; i++) {
862 ep = xfs_iext_get_ext(ifp, i);
863 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
864 arp->l0 = cpu_to_be64(ep->l0);
865 arp->l1 = cpu_to_be64(ep->l1);
869 ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
870 xfs_btree_set_numrecs(ablock, cnt);
873 * Fill in the root key and pointer.
875 kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
876 arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
877 kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
878 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
879 be16_to_cpu(block->bb_level)));
880 *pp = cpu_to_be64(args.fsbno);
883 * Do all this logging at the end so that
884 * the root is at the right level.
886 xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
887 xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
888 ASSERT(*curp == NULL);
890 *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
895 * Convert a local file to an extents file.
896 * This code is out of bounds for data forks of regular files,
897 * since the file data needs to get logged so things will stay consistent.
898 * (The bmap-level manipulations are ok, though).
901 xfs_bmap_local_to_extents_empty(
902 struct xfs_inode *ip,
905 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
907 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
908 ASSERT(ifp->if_bytes == 0);
909 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
911 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
912 ifp->if_flags &= ~XFS_IFINLINE;
913 ifp->if_flags |= XFS_IFEXTENTS;
914 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
918 STATIC int /* error */
919 xfs_bmap_local_to_extents(
920 xfs_trans_t *tp, /* transaction pointer */
921 xfs_inode_t *ip, /* incore inode pointer */
922 xfs_fsblock_t *firstblock, /* first block allocated in xaction */
923 xfs_extlen_t total, /* total blocks needed by transaction */
924 int *logflagsp, /* inode logging flags */
926 void (*init_fn)(struct xfs_trans *tp,
928 struct xfs_inode *ip,
929 struct xfs_ifork *ifp))
932 int flags; /* logging flags returned */
933 xfs_ifork_t *ifp; /* inode fork pointer */
934 xfs_alloc_arg_t args; /* allocation arguments */
935 xfs_buf_t *bp; /* buffer for extent block */
936 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
939 * We don't want to deal with the case of keeping inode data inline yet.
940 * So sending the data fork of a regular inode is invalid.
942 ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
943 ifp = XFS_IFORK_PTR(ip, whichfork);
944 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
946 if (!ifp->if_bytes) {
947 xfs_bmap_local_to_extents_empty(ip, whichfork);
948 flags = XFS_ILOG_CORE;
954 ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) ==
956 memset(&args, 0, sizeof(args));
958 args.mp = ip->i_mount;
959 args.firstblock = *firstblock;
961 * Allocate a block. We know we need only one, since the
962 * file currently fits in an inode.
964 if (*firstblock == NULLFSBLOCK) {
965 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
966 args.type = XFS_ALLOCTYPE_START_BNO;
968 args.fsbno = *firstblock;
969 args.type = XFS_ALLOCTYPE_NEAR_BNO;
972 args.minlen = args.maxlen = args.prod = 1;
973 error = xfs_alloc_vextent(&args);
977 /* Can't fail, the space was reserved. */
978 ASSERT(args.fsbno != NULLFSBLOCK);
979 ASSERT(args.len == 1);
980 *firstblock = args.fsbno;
981 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
983 /* initialise the block and copy the data */
984 init_fn(tp, bp, ip, ifp);
986 /* account for the change in fork size and log everything */
987 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
988 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
989 xfs_bmap_local_to_extents_empty(ip, whichfork);
990 flags |= XFS_ILOG_CORE;
992 xfs_iext_add(ifp, 0, 1);
993 ep = xfs_iext_get_ext(ifp, 0);
994 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
995 trace_xfs_bmap_post_update(ip, 0,
996 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
998 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
999 ip->i_d.di_nblocks = 1;
1000 xfs_trans_mod_dquot_byino(tp, ip,
1001 XFS_TRANS_DQ_BCOUNT, 1L);
1002 flags |= xfs_ilog_fext(whichfork);
1010 * Called from xfs_bmap_add_attrfork to handle btree format files.
1012 STATIC int /* error */
1013 xfs_bmap_add_attrfork_btree(
1014 xfs_trans_t *tp, /* transaction pointer */
1015 xfs_inode_t *ip, /* incore inode pointer */
1016 xfs_fsblock_t *firstblock, /* first block allocated */
1017 xfs_bmap_free_t *flist, /* blocks to free at commit */
1018 int *flags) /* inode logging flags */
1020 xfs_btree_cur_t *cur; /* btree cursor */
1021 int error; /* error return value */
1022 xfs_mount_t *mp; /* file system mount struct */
1023 int stat; /* newroot status */
1026 if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
1027 *flags |= XFS_ILOG_DBROOT;
1029 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
1030 cur->bc_private.b.flist = flist;
1031 cur->bc_private.b.firstblock = *firstblock;
1032 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
1034 /* must be at least one entry */
1035 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
1036 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
1039 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1040 return XFS_ERROR(ENOSPC);
1042 *firstblock = cur->bc_private.b.firstblock;
1043 cur->bc_private.b.allocated = 0;
1044 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1048 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1053 * Called from xfs_bmap_add_attrfork to handle extents format files.
1055 STATIC int /* error */
1056 xfs_bmap_add_attrfork_extents(
1057 xfs_trans_t *tp, /* transaction pointer */
1058 xfs_inode_t *ip, /* incore inode pointer */
1059 xfs_fsblock_t *firstblock, /* first block allocated */
1060 xfs_bmap_free_t *flist, /* blocks to free at commit */
1061 int *flags) /* inode logging flags */
1063 xfs_btree_cur_t *cur; /* bmap btree cursor */
1064 int error; /* error return value */
1066 if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
1069 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
1070 flags, XFS_DATA_FORK);
1072 cur->bc_private.b.allocated = 0;
1073 xfs_btree_del_cursor(cur,
1074 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
1080 * Called from xfs_bmap_add_attrfork to handle local format files. Each
1081 * different data fork content type needs a different callout to do the
1082 * conversion. Some are basic and only require special block initialisation
1083 * callouts for the data formating, others (directories) are so specialised they
1084 * handle everything themselves.
1086 * XXX (dgc): investigate whether directory conversion can use the generic
1087 * formatting callout. It should be possible - it's just a very complex
1090 STATIC int /* error */
1091 xfs_bmap_add_attrfork_local(
1092 xfs_trans_t *tp, /* transaction pointer */
1093 xfs_inode_t *ip, /* incore inode pointer */
1094 xfs_fsblock_t *firstblock, /* first block allocated */
1095 xfs_bmap_free_t *flist, /* blocks to free at commit */
1096 int *flags) /* inode logging flags */
1098 xfs_da_args_t dargs; /* args for dir/attr code */
1100 if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1103 if (S_ISDIR(ip->i_d.di_mode)) {
1104 memset(&dargs, 0, sizeof(dargs));
1106 dargs.firstblock = firstblock;
1107 dargs.flist = flist;
1108 dargs.total = ip->i_mount->m_dirblkfsbs;
1109 dargs.whichfork = XFS_DATA_FORK;
1111 return xfs_dir2_sf_to_block(&dargs);
1114 if (S_ISLNK(ip->i_d.di_mode))
1115 return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
1116 flags, XFS_DATA_FORK,
1117 xfs_symlink_local_to_remote);
1119 /* should only be called for types that support local format data */
1121 return EFSCORRUPTED;
1125 * Convert inode from non-attributed to attributed.
1126 * Must not be in a transaction, ip must not be locked.
1128 int /* error code */
1129 xfs_bmap_add_attrfork(
1130 xfs_inode_t *ip, /* incore inode pointer */
1131 int size, /* space new attribute needs */
1132 int rsvd) /* xact may use reserved blks */
1134 xfs_fsblock_t firstblock; /* 1st block/ag allocated */
1135 xfs_bmap_free_t flist; /* freed extent records */
1136 xfs_mount_t *mp; /* mount structure */
1137 xfs_trans_t *tp; /* transaction pointer */
1138 int blks; /* space reservation */
1139 int version = 1; /* superblock attr version */
1140 int committed; /* xaction was committed */
1141 int logflags; /* logging flags */
1142 int error; /* error return value */
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);
1155 xfs_ilock(ip, XFS_ILOCK_EXCL);
1156 error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1157 XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1158 XFS_QMOPT_RES_REGBLKS);
1160 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1161 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
1164 if (XFS_IFORK_Q(ip))
1166 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1168 * For inodes coming from pre-6.2 filesystems.
1170 ASSERT(ip->i_d.di_aformat == 0);
1171 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1173 ASSERT(ip->i_d.di_anextents == 0);
1175 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1176 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1178 switch (ip->i_d.di_format) {
1179 case XFS_DINODE_FMT_DEV:
1180 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1182 case XFS_DINODE_FMT_UUID:
1183 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
1185 case XFS_DINODE_FMT_LOCAL:
1186 case XFS_DINODE_FMT_EXTENTS:
1187 case XFS_DINODE_FMT_BTREE:
1188 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1189 if (!ip->i_d.di_forkoff)
1190 ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1191 else if (mp->m_flags & XFS_MOUNT_ATTR2)
1196 error = XFS_ERROR(EINVAL);
1200 ASSERT(ip->i_afp == NULL);
1201 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1202 ip->i_afp->if_flags = XFS_IFEXTENTS;
1204 xfs_bmap_init(&flist, &firstblock);
1205 switch (ip->i_d.di_format) {
1206 case XFS_DINODE_FMT_LOCAL:
1207 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
1210 case XFS_DINODE_FMT_EXTENTS:
1211 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1214 case XFS_DINODE_FMT_BTREE:
1215 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
1223 xfs_trans_log_inode(tp, ip, logflags);
1226 if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1227 (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1228 __int64_t sbfields = 0;
1230 spin_lock(&mp->m_sb_lock);
1231 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1232 xfs_sb_version_addattr(&mp->m_sb);
1233 sbfields |= XFS_SB_VERSIONNUM;
1235 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1236 xfs_sb_version_addattr2(&mp->m_sb);
1237 sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
1240 spin_unlock(&mp->m_sb_lock);
1241 xfs_mod_sb(tp, sbfields);
1243 spin_unlock(&mp->m_sb_lock);
1246 error = xfs_bmap_finish(&tp, &flist, &committed);
1249 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1251 xfs_bmap_cancel(&flist);
1253 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1255 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1260 * Internal and external extent tree search functions.
1264 * Read in the extents to if_extents.
1265 * All inode fields are set up by caller, we just traverse the btree
1266 * and copy the records in. If the file system cannot contain unwritten
1267 * extents, the records are checked for no "state" flags.
1270 xfs_bmap_read_extents(
1271 xfs_trans_t *tp, /* transaction pointer */
1272 xfs_inode_t *ip, /* incore inode */
1273 int whichfork) /* data or attr fork */
1275 struct xfs_btree_block *block; /* current btree block */
1276 xfs_fsblock_t bno; /* block # of "block" */
1277 xfs_buf_t *bp; /* buffer for "block" */
1278 int error; /* error return value */
1279 xfs_exntfmt_t exntf; /* XFS_EXTFMT_NOSTATE, if checking */
1280 xfs_extnum_t i, j; /* index into the extents list */
1281 xfs_ifork_t *ifp; /* fork structure */
1282 int level; /* btree level, for checking */
1283 xfs_mount_t *mp; /* file system mount structure */
1284 __be64 *pp; /* pointer to block address */
1286 xfs_extnum_t room; /* number of entries there's room for */
1290 ifp = XFS_IFORK_PTR(ip, whichfork);
1291 exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
1292 XFS_EXTFMT_INODE(ip);
1293 block = ifp->if_broot;
1295 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1297 level = be16_to_cpu(block->bb_level);
1299 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1300 bno = be64_to_cpu(*pp);
1301 ASSERT(bno != NULLDFSBNO);
1302 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
1303 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
1305 * Go down the tree until leaf level is reached, following the first
1306 * pointer (leftmost) at each level.
1308 while (level-- > 0) {
1309 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1310 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1313 block = XFS_BUF_TO_BLOCK(bp);
1314 XFS_WANT_CORRUPTED_GOTO(
1315 xfs_bmap_sanity_check(mp, bp, level),
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(XFS_FSB_SANITY_CHECK(mp, bno), error0);
1322 xfs_trans_brelse(tp, bp);
1325 * Here with bp and block set to the leftmost leaf node in the tree.
1327 room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1330 * Loop over all leaf nodes. Copy information to the extent records.
1333 xfs_bmbt_rec_t *frp;
1334 xfs_fsblock_t nextbno;
1335 xfs_extnum_t num_recs;
1338 num_recs = xfs_btree_get_numrecs(block);
1339 if (unlikely(i + num_recs > room)) {
1340 ASSERT(i + num_recs <= room);
1341 xfs_warn(ip->i_mount,
1342 "corrupt dinode %Lu, (btree extents).",
1343 (unsigned long long) ip->i_ino);
1344 XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1345 XFS_ERRLEVEL_LOW, ip->i_mount, block);
1348 XFS_WANT_CORRUPTED_GOTO(
1349 xfs_bmap_sanity_check(mp, bp, 0),
1352 * Read-ahead the next leaf block, if any.
1354 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1355 if (nextbno != NULLFSBLOCK)
1356 xfs_btree_reada_bufl(mp, nextbno, 1,
1359 * Copy records into the extent records.
1361 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1363 for (j = 0; j < num_recs; j++, i++, frp++) {
1364 xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
1365 trp->l0 = be64_to_cpu(frp->l0);
1366 trp->l1 = be64_to_cpu(frp->l1);
1368 if (exntf == XFS_EXTFMT_NOSTATE) {
1370 * Check all attribute bmap btree records and
1371 * any "older" data bmap btree records for a
1372 * set bit in the "extent flag" position.
1374 if (unlikely(xfs_check_nostate_extents(ifp,
1375 start, num_recs))) {
1376 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1382 xfs_trans_brelse(tp, bp);
1385 * If we've reached the end, stop.
1387 if (bno == NULLFSBLOCK)
1389 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1390 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1393 block = XFS_BUF_TO_BLOCK(bp);
1395 ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
1396 ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
1397 XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
1400 xfs_trans_brelse(tp, bp);
1401 return XFS_ERROR(EFSCORRUPTED);
1406 * Search the extent records for the entry containing block bno.
1407 * If bno lies in a hole, point to the next entry. If bno lies
1408 * past eof, *eofp will be set, and *prevp will contain the last
1409 * entry (null if none). Else, *lastxp will be set to the index
1410 * of the found entry; *gotp will contain the entry.
1412 STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
1413 xfs_bmap_search_multi_extents(
1414 xfs_ifork_t *ifp, /* inode fork pointer */
1415 xfs_fileoff_t bno, /* block number searched for */
1416 int *eofp, /* out: end of file found */
1417 xfs_extnum_t *lastxp, /* out: last extent index */
1418 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
1419 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
1421 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
1422 xfs_extnum_t lastx; /* last extent index */
1425 * Initialize the extent entry structure to catch access to
1426 * uninitialized br_startblock field.
1428 gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
1429 gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
1430 gotp->br_state = XFS_EXT_INVALID;
1432 gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
1434 gotp->br_startblock = 0xffffa5a5;
1436 prevp->br_startoff = NULLFILEOFF;
1438 ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
1440 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
1442 if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1443 xfs_bmbt_get_all(ep, gotp);
1457 * Search the extents list for the inode, for the extent containing bno.
1458 * If bno lies in a hole, point to the next entry. If bno lies past eof,
1459 * *eofp will be set, and *prevp will contain the last entry (null if none).
1460 * Else, *lastxp will be set to the index of the found
1461 * entry; *gotp will contain the entry.
1463 STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
1464 xfs_bmap_search_extents(
1465 xfs_inode_t *ip, /* incore inode pointer */
1466 xfs_fileoff_t bno, /* block number searched for */
1467 int fork, /* data or attr fork */
1468 int *eofp, /* out: end of file found */
1469 xfs_extnum_t *lastxp, /* out: last extent index */
1470 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
1471 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
1473 xfs_ifork_t *ifp; /* inode fork pointer */
1474 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
1476 XFS_STATS_INC(xs_look_exlist);
1477 ifp = XFS_IFORK_PTR(ip, fork);
1479 ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
1481 if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
1482 !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
1483 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
1484 "Access to block zero in inode %llu "
1485 "start_block: %llx start_off: %llx "
1486 "blkcnt: %llx extent-state: %x lastx: %x",
1487 (unsigned long long)ip->i_ino,
1488 (unsigned long long)gotp->br_startblock,
1489 (unsigned long long)gotp->br_startoff,
1490 (unsigned long long)gotp->br_blockcount,
1491 gotp->br_state, *lastxp);
1492 *lastxp = NULLEXTNUM;
1500 * Returns the file-relative block number of the first unused block(s)
1501 * in the file with at least "len" logically contiguous blocks free.
1502 * This is the lowest-address hole if the file has holes, else the first block
1503 * past the end of file.
1504 * Return 0 if the file is currently local (in-inode).
1507 xfs_bmap_first_unused(
1508 xfs_trans_t *tp, /* transaction pointer */
1509 xfs_inode_t *ip, /* incore inode */
1510 xfs_extlen_t len, /* size of hole to find */
1511 xfs_fileoff_t *first_unused, /* unused block */
1512 int whichfork) /* data or attr fork */
1514 int error; /* error return value */
1515 int idx; /* extent record index */
1516 xfs_ifork_t *ifp; /* inode fork pointer */
1517 xfs_fileoff_t lastaddr; /* last block number seen */
1518 xfs_fileoff_t lowest; /* lowest useful block */
1519 xfs_fileoff_t max; /* starting useful block */
1520 xfs_fileoff_t off; /* offset for this block */
1521 xfs_extnum_t nextents; /* number of extent entries */
1523 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1524 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1525 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1526 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1530 ifp = XFS_IFORK_PTR(ip, whichfork);
1531 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1532 (error = xfs_iread_extents(tp, ip, whichfork)))
1534 lowest = *first_unused;
1535 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1536 for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
1537 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
1538 off = xfs_bmbt_get_startoff(ep);
1540 * See if the hole before this extent will work.
1542 if (off >= lowest + len && off - max >= len) {
1543 *first_unused = max;
1546 lastaddr = off + xfs_bmbt_get_blockcount(ep);
1547 max = XFS_FILEOFF_MAX(lastaddr, lowest);
1549 *first_unused = max;
1554 * Returns the file-relative block number of the last block - 1 before
1555 * last_block (input value) in the file.
1556 * This is not based on i_size, it is based on the extent records.
1557 * Returns 0 for local files, as they do not have extent records.
1560 xfs_bmap_last_before(
1561 xfs_trans_t *tp, /* transaction pointer */
1562 xfs_inode_t *ip, /* incore inode */
1563 xfs_fileoff_t *last_block, /* last block */
1564 int whichfork) /* data or attr fork */
1566 xfs_fileoff_t bno; /* input file offset */
1567 int eof; /* hit end of file */
1568 xfs_bmbt_rec_host_t *ep; /* pointer to last extent */
1569 int error; /* error return value */
1570 xfs_bmbt_irec_t got; /* current extent value */
1571 xfs_ifork_t *ifp; /* inode fork pointer */
1572 xfs_extnum_t lastx; /* last extent used */
1573 xfs_bmbt_irec_t prev; /* previous extent value */
1575 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1576 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
1577 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
1578 return XFS_ERROR(EIO);
1579 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1583 ifp = XFS_IFORK_PTR(ip, whichfork);
1584 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1585 (error = xfs_iread_extents(tp, ip, whichfork)))
1587 bno = *last_block - 1;
1588 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
1590 if (eof || xfs_bmbt_get_startoff(ep) > bno) {
1591 if (prev.br_startoff == NULLFILEOFF)
1594 *last_block = prev.br_startoff + prev.br_blockcount;
1597 * Otherwise *last_block is already the right answer.
1603 xfs_bmap_last_extent(
1604 struct xfs_trans *tp,
1605 struct xfs_inode *ip,
1607 struct xfs_bmbt_irec *rec,
1610 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
1614 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1615 error = xfs_iread_extents(tp, ip, whichfork);
1620 nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
1621 if (nextents == 0) {
1626 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
1632 * Check the last inode extent to determine whether this allocation will result
1633 * in blocks being allocated at the end of the file. When we allocate new data
1634 * blocks at the end of the file which do not start at the previous data block,
1635 * we will try to align the new blocks at stripe unit boundaries.
1637 * Returns 0 in bma->aeof if the file (fork) is empty as any new write will be
1638 * at, or past the EOF.
1642 struct xfs_bmalloca *bma,
1645 struct xfs_bmbt_irec rec;
1650 error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1652 if (error || is_empty)
1656 * Check if we are allocation or past the last extent, or at least into
1657 * the last delayed allocated extent.
1659 bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1660 (bma->offset >= rec.br_startoff &&
1661 isnullstartblock(rec.br_startblock));
1666 * Returns the file-relative block number of the first block past eof in
1667 * the file. This is not based on i_size, it is based on the extent records.
1668 * Returns 0 for local files, as they do not have extent records.
1671 xfs_bmap_last_offset(
1672 struct xfs_trans *tp,
1673 struct xfs_inode *ip,
1674 xfs_fileoff_t *last_block,
1677 struct xfs_bmbt_irec rec;
1683 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1686 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1687 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1688 return XFS_ERROR(EIO);
1690 error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1691 if (error || is_empty)
1694 *last_block = rec.br_startoff + rec.br_blockcount;
1699 * Returns whether the selected fork of the inode has exactly one
1700 * block or not. For the data fork we check this matches di_size,
1701 * implying the file's range is 0..bsize-1.
1703 int /* 1=>1 block, 0=>otherwise */
1705 xfs_inode_t *ip, /* incore inode */
1706 int whichfork) /* data or attr fork */
1708 xfs_bmbt_rec_host_t *ep; /* ptr to fork's extent */
1709 xfs_ifork_t *ifp; /* inode fork pointer */
1710 int rval; /* return value */
1711 xfs_bmbt_irec_t s; /* internal version of extent */
1714 if (whichfork == XFS_DATA_FORK)
1715 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1717 if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1719 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1721 ifp = XFS_IFORK_PTR(ip, whichfork);
1722 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1723 ep = xfs_iext_get_ext(ifp, 0);
1724 xfs_bmbt_get_all(ep, &s);
1725 rval = s.br_startoff == 0 && s.br_blockcount == 1;
1726 if (rval && whichfork == XFS_DATA_FORK)
1727 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1732 * Extent tree manipulation functions used during allocation.
1736 * Convert a delayed allocation to a real allocation.
1738 STATIC int /* error */
1739 xfs_bmap_add_extent_delay_real(
1740 struct xfs_bmalloca *bma)
1742 struct xfs_bmbt_irec *new = &bma->got;
1743 int diff; /* temp value */
1744 xfs_bmbt_rec_host_t *ep; /* extent entry for idx */
1745 int error; /* error return value */
1746 int i; /* temp state */
1747 xfs_ifork_t *ifp; /* inode fork pointer */
1748 xfs_fileoff_t new_endoff; /* end offset of new entry */
1749 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
1750 /* left is 0, right is 1, prev is 2 */
1751 int rval=0; /* return value (logging flags) */
1752 int state = 0;/* state bits, accessed thru macros */
1753 xfs_filblks_t da_new; /* new count del alloc blocks used */
1754 xfs_filblks_t da_old; /* old count del alloc blocks used */
1755 xfs_filblks_t temp=0; /* value for da_new calculations */
1756 xfs_filblks_t temp2=0;/* value for da_new calculations */
1757 int tmp_rval; /* partial logging flags */
1759 ifp = XFS_IFORK_PTR(bma->ip, XFS_DATA_FORK);
1761 ASSERT(bma->idx >= 0);
1762 ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1763 ASSERT(!isnullstartblock(new->br_startblock));
1765 (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1767 XFS_STATS_INC(xs_add_exlist);
1774 * Set up a bunch of variables to make the tests simpler.
1776 ep = xfs_iext_get_ext(ifp, bma->idx);
1777 xfs_bmbt_get_all(ep, &PREV);
1778 new_endoff = new->br_startoff + new->br_blockcount;
1779 ASSERT(PREV.br_startoff <= new->br_startoff);
1780 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1782 da_old = startblockval(PREV.br_startblock);
1786 * Set flags determining what part of the previous delayed allocation
1787 * extent is being replaced by a real allocation.
1789 if (PREV.br_startoff == new->br_startoff)
1790 state |= BMAP_LEFT_FILLING;
1791 if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1792 state |= BMAP_RIGHT_FILLING;
1795 * Check and set flags if this segment has a left neighbor.
1796 * Don't set contiguous if the combined extent would be too large.
1799 state |= BMAP_LEFT_VALID;
1800 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
1802 if (isnullstartblock(LEFT.br_startblock))
1803 state |= BMAP_LEFT_DELAY;
1806 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1807 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1808 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1809 LEFT.br_state == new->br_state &&
1810 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1811 state |= BMAP_LEFT_CONTIG;
1814 * Check and set flags if this segment has a right neighbor.
1815 * Don't set contiguous if the combined extent would be too large.
1816 * Also check for all-three-contiguous being too large.
1818 if (bma->idx < bma->ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1819 state |= BMAP_RIGHT_VALID;
1820 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
1822 if (isnullstartblock(RIGHT.br_startblock))
1823 state |= BMAP_RIGHT_DELAY;
1826 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1827 new_endoff == RIGHT.br_startoff &&
1828 new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1829 new->br_state == RIGHT.br_state &&
1830 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1831 ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1832 BMAP_RIGHT_FILLING)) !=
1833 (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1834 BMAP_RIGHT_FILLING) ||
1835 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1837 state |= BMAP_RIGHT_CONTIG;
1841 * Switch out based on the FILLING and CONTIG state bits.
1843 switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1844 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1845 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1846 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1848 * Filling in all of a previously delayed allocation extent.
1849 * The left and right neighbors are both contiguous with new.
1852 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1853 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1854 LEFT.br_blockcount + PREV.br_blockcount +
1855 RIGHT.br_blockcount);
1856 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1858 xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
1859 bma->ip->i_d.di_nextents--;
1860 if (bma->cur == NULL)
1861 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1863 rval = XFS_ILOG_CORE;
1864 error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1865 RIGHT.br_startblock,
1866 RIGHT.br_blockcount, &i);
1869 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1870 error = xfs_btree_delete(bma->cur, &i);
1873 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1874 error = xfs_btree_decrement(bma->cur, 0, &i);
1877 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1878 error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1880 LEFT.br_blockcount +
1881 PREV.br_blockcount +
1882 RIGHT.br_blockcount, LEFT.br_state);
1888 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1890 * Filling in all of a previously delayed allocation extent.
1891 * The left neighbor is contiguous, the right is not.
1895 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1896 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1897 LEFT.br_blockcount + PREV.br_blockcount);
1898 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1900 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1901 if (bma->cur == NULL)
1902 rval = XFS_ILOG_DEXT;
1905 error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1906 LEFT.br_startblock, LEFT.br_blockcount,
1910 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1911 error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1913 LEFT.br_blockcount +
1914 PREV.br_blockcount, LEFT.br_state);
1920 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1922 * Filling in all of a previously delayed allocation extent.
1923 * The right neighbor is contiguous, the left is not.
1925 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1926 xfs_bmbt_set_startblock(ep, new->br_startblock);
1927 xfs_bmbt_set_blockcount(ep,
1928 PREV.br_blockcount + RIGHT.br_blockcount);
1929 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1931 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1932 if (bma->cur == NULL)
1933 rval = XFS_ILOG_DEXT;
1936 error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1937 RIGHT.br_startblock,
1938 RIGHT.br_blockcount, &i);
1941 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1942 error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
1944 PREV.br_blockcount +
1945 RIGHT.br_blockcount, PREV.br_state);
1951 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1953 * Filling in all of a previously delayed allocation extent.
1954 * Neither the left nor right neighbors are contiguous with
1957 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1958 xfs_bmbt_set_startblock(ep, new->br_startblock);
1959 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1961 bma->ip->i_d.di_nextents++;
1962 if (bma->cur == NULL)
1963 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1965 rval = XFS_ILOG_CORE;
1966 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1967 new->br_startblock, new->br_blockcount,
1971 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1972 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1973 error = xfs_btree_insert(bma->cur, &i);
1976 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1980 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1982 * Filling in the first part of a previous delayed allocation.
1983 * The left neighbor is contiguous.
1985 trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1986 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
1987 LEFT.br_blockcount + new->br_blockcount);
1988 xfs_bmbt_set_startoff(ep,
1989 PREV.br_startoff + new->br_blockcount);
1990 trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1992 temp = PREV.br_blockcount - new->br_blockcount;
1993 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1994 xfs_bmbt_set_blockcount(ep, temp);
1995 if (bma->cur == NULL)
1996 rval = XFS_ILOG_DEXT;
1999 error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
2000 LEFT.br_startblock, LEFT.br_blockcount,
2004 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2005 error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
2007 LEFT.br_blockcount +
2013 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2014 startblockval(PREV.br_startblock));
2015 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2016 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2021 case BMAP_LEFT_FILLING:
2023 * Filling in the first part of a previous delayed allocation.
2024 * The left neighbor is not contiguous.
2026 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2027 xfs_bmbt_set_startoff(ep, new_endoff);
2028 temp = PREV.br_blockcount - new->br_blockcount;
2029 xfs_bmbt_set_blockcount(ep, temp);
2030 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
2031 bma->ip->i_d.di_nextents++;
2032 if (bma->cur == NULL)
2033 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2035 rval = XFS_ILOG_CORE;
2036 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2037 new->br_startblock, new->br_blockcount,
2041 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2042 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2043 error = xfs_btree_insert(bma->cur, &i);
2046 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2049 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2050 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2051 bma->firstblock, bma->flist,
2052 &bma->cur, 1, &tmp_rval, XFS_DATA_FORK);
2057 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2058 startblockval(PREV.br_startblock) -
2059 (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2060 ep = xfs_iext_get_ext(ifp, bma->idx + 1);
2061 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2062 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2065 case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2067 * Filling in the last part of a previous delayed allocation.
2068 * The right neighbor is contiguous with the new allocation.
2070 temp = PREV.br_blockcount - new->br_blockcount;
2071 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2072 xfs_bmbt_set_blockcount(ep, temp);
2073 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
2074 new->br_startoff, new->br_startblock,
2075 new->br_blockcount + RIGHT.br_blockcount,
2077 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2078 if (bma->cur == NULL)
2079 rval = XFS_ILOG_DEXT;
2082 error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
2083 RIGHT.br_startblock,
2084 RIGHT.br_blockcount, &i);
2087 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2088 error = xfs_bmbt_update(bma->cur, new->br_startoff,
2090 new->br_blockcount +
2091 RIGHT.br_blockcount,
2097 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2098 startblockval(PREV.br_startblock));
2099 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2100 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2101 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2106 case BMAP_RIGHT_FILLING:
2108 * Filling in the last part of a previous delayed allocation.
2109 * The right neighbor is not contiguous.
2111 temp = PREV.br_blockcount - new->br_blockcount;
2112 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2113 xfs_bmbt_set_blockcount(ep, temp);
2114 xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
2115 bma->ip->i_d.di_nextents++;
2116 if (bma->cur == NULL)
2117 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2119 rval = XFS_ILOG_CORE;
2120 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2121 new->br_startblock, new->br_blockcount,
2125 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2126 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2127 error = xfs_btree_insert(bma->cur, &i);
2130 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2133 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2134 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2135 bma->firstblock, bma->flist, &bma->cur, 1,
2136 &tmp_rval, XFS_DATA_FORK);
2141 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2142 startblockval(PREV.br_startblock) -
2143 (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2144 ep = xfs_iext_get_ext(ifp, bma->idx);
2145 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2146 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2153 * Filling in the middle part of a previous delayed allocation.
2154 * Contiguity is impossible here.
2155 * This case is avoided almost all the time.
2157 * We start with a delayed allocation:
2159 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2162 * and we are allocating:
2163 * +rrrrrrrrrrrrrrrrr+
2166 * and we set it up for insertion as:
2167 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2169 * PREV @ idx LEFT RIGHT
2170 * inserted at idx + 1
2172 temp = new->br_startoff - PREV.br_startoff;
2173 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
2174 trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
2175 xfs_bmbt_set_blockcount(ep, temp); /* truncate PREV */
2177 RIGHT.br_state = PREV.br_state;
2178 RIGHT.br_startblock = nullstartblock(
2179 (int)xfs_bmap_worst_indlen(bma->ip, temp2));
2180 RIGHT.br_startoff = new_endoff;
2181 RIGHT.br_blockcount = temp2;
2182 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2183 xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
2184 bma->ip->i_d.di_nextents++;
2185 if (bma->cur == NULL)
2186 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2188 rval = XFS_ILOG_CORE;
2189 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2190 new->br_startblock, new->br_blockcount,
2194 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2195 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2196 error = xfs_btree_insert(bma->cur, &i);
2199 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2202 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2203 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2204 bma->firstblock, bma->flist, &bma->cur,
2205 1, &tmp_rval, XFS_DATA_FORK);
2210 temp = xfs_bmap_worst_indlen(bma->ip, temp);
2211 temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
2212 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
2213 (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2215 error = xfs_icsb_modify_counters(bma->ip->i_mount,
2217 -((int64_t)diff), 0);
2223 ep = xfs_iext_get_ext(ifp, bma->idx);
2224 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2225 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2226 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2227 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
2228 nullstartblock((int)temp2));
2229 trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2232 da_new = temp + temp2;
2235 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2236 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2237 case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2238 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2239 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2240 case BMAP_LEFT_CONTIG:
2241 case BMAP_RIGHT_CONTIG:
2243 * These cases are all impossible.
2248 /* convert to a btree if necessary */
2249 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2250 int tmp_logflags; /* partial log flag return val */
2252 ASSERT(bma->cur == NULL);
2253 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2254 bma->firstblock, bma->flist, &bma->cur,
2255 da_old > 0, &tmp_logflags, XFS_DATA_FORK);
2256 bma->logflags |= tmp_logflags;
2261 /* adjust for changes in reserved delayed indirect blocks */
2262 if (da_old || da_new) {
2265 temp += bma->cur->bc_private.b.allocated;
2266 ASSERT(temp <= da_old);
2268 xfs_icsb_modify_counters(bma->ip->i_mount,
2270 (int64_t)(da_old - temp), 0);
2273 /* clear out the allocated field, done with it now in any case. */
2275 bma->cur->bc_private.b.allocated = 0;
2277 xfs_bmap_check_leaf_extents(bma->cur, bma->ip, XFS_DATA_FORK);
2279 bma->logflags |= rval;
2287 * Convert an unwritten allocation to a real allocation or vice versa.
2289 STATIC int /* error */
2290 xfs_bmap_add_extent_unwritten_real(
2291 struct xfs_trans *tp,
2292 xfs_inode_t *ip, /* incore inode pointer */
2293 xfs_extnum_t *idx, /* extent number to update/insert */
2294 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
2295 xfs_bmbt_irec_t *new, /* new data to add to file extents */
2296 xfs_fsblock_t *first, /* pointer to firstblock variable */
2297 xfs_bmap_free_t *flist, /* list of extents to be freed */
2298 int *logflagsp) /* inode logging flags */
2300 xfs_btree_cur_t *cur; /* btree cursor */
2301 xfs_bmbt_rec_host_t *ep; /* extent entry for idx */
2302 int error; /* error return value */
2303 int i; /* temp state */
2304 xfs_ifork_t *ifp; /* inode fork pointer */
2305 xfs_fileoff_t new_endoff; /* end offset of new entry */
2306 xfs_exntst_t newext; /* new extent state */
2307 xfs_exntst_t oldext; /* old extent state */
2308 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
2309 /* left is 0, right is 1, prev is 2 */
2310 int rval=0; /* return value (logging flags) */
2311 int state = 0;/* state bits, accessed thru macros */
2316 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2319 ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2320 ASSERT(!isnullstartblock(new->br_startblock));
2322 XFS_STATS_INC(xs_add_exlist);
2329 * Set up a bunch of variables to make the tests simpler.
2332 ep = xfs_iext_get_ext(ifp, *idx);
2333 xfs_bmbt_get_all(ep, &PREV);
2334 newext = new->br_state;
2335 oldext = (newext == XFS_EXT_UNWRITTEN) ?
2336 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
2337 ASSERT(PREV.br_state == oldext);
2338 new_endoff = new->br_startoff + new->br_blockcount;
2339 ASSERT(PREV.br_startoff <= new->br_startoff);
2340 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2343 * Set flags determining what part of the previous oldext allocation
2344 * extent is being replaced by a newext allocation.
2346 if (PREV.br_startoff == new->br_startoff)
2347 state |= BMAP_LEFT_FILLING;
2348 if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2349 state |= BMAP_RIGHT_FILLING;
2352 * Check and set flags if this segment has a left neighbor.
2353 * Don't set contiguous if the combined extent would be too large.
2356 state |= BMAP_LEFT_VALID;
2357 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
2359 if (isnullstartblock(LEFT.br_startblock))
2360 state |= BMAP_LEFT_DELAY;
2363 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2364 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2365 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2366 LEFT.br_state == newext &&
2367 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2368 state |= BMAP_LEFT_CONTIG;
2371 * Check and set flags if this segment has a right neighbor.
2372 * Don't set contiguous if the combined extent would be too large.
2373 * Also check for all-three-contiguous being too large.
2375 if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
2376 state |= BMAP_RIGHT_VALID;
2377 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
2378 if (isnullstartblock(RIGHT.br_startblock))
2379 state |= BMAP_RIGHT_DELAY;
2382 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2383 new_endoff == RIGHT.br_startoff &&
2384 new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2385 newext == RIGHT.br_state &&
2386 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2387 ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2388 BMAP_RIGHT_FILLING)) !=
2389 (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2390 BMAP_RIGHT_FILLING) ||
2391 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2393 state |= BMAP_RIGHT_CONTIG;
2396 * Switch out based on the FILLING and CONTIG state bits.
2398 switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2399 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2400 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2401 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2403 * Setting all of a previous oldext extent to newext.
2404 * The left and right neighbors are both contiguous with new.
2408 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2409 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2410 LEFT.br_blockcount + PREV.br_blockcount +
2411 RIGHT.br_blockcount);
2412 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2414 xfs_iext_remove(ip, *idx + 1, 2, state);
2415 ip->i_d.di_nextents -= 2;
2417 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2419 rval = XFS_ILOG_CORE;
2420 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2421 RIGHT.br_startblock,
2422 RIGHT.br_blockcount, &i)))
2424 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2425 if ((error = xfs_btree_delete(cur, &i)))
2427 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2428 if ((error = xfs_btree_decrement(cur, 0, &i)))
2430 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2431 if ((error = xfs_btree_delete(cur, &i)))
2433 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2434 if ((error = xfs_btree_decrement(cur, 0, &i)))
2436 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2437 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2439 LEFT.br_blockcount + PREV.br_blockcount +
2440 RIGHT.br_blockcount, LEFT.br_state)))
2445 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2447 * Setting all of a previous oldext extent to newext.
2448 * The left neighbor is contiguous, the right is not.
2452 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2453 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2454 LEFT.br_blockcount + PREV.br_blockcount);
2455 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2457 xfs_iext_remove(ip, *idx + 1, 1, state);
2458 ip->i_d.di_nextents--;
2460 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2462 rval = XFS_ILOG_CORE;
2463 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2464 PREV.br_startblock, PREV.br_blockcount,
2467 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2468 if ((error = xfs_btree_delete(cur, &i)))
2470 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2471 if ((error = xfs_btree_decrement(cur, 0, &i)))
2473 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2474 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2476 LEFT.br_blockcount + PREV.br_blockcount,
2482 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2484 * Setting all of a previous oldext extent to newext.
2485 * The right neighbor is contiguous, the left is not.
2487 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2488 xfs_bmbt_set_blockcount(ep,
2489 PREV.br_blockcount + RIGHT.br_blockcount);
2490 xfs_bmbt_set_state(ep, newext);
2491 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2492 xfs_iext_remove(ip, *idx + 1, 1, state);
2493 ip->i_d.di_nextents--;
2495 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2497 rval = XFS_ILOG_CORE;
2498 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2499 RIGHT.br_startblock,
2500 RIGHT.br_blockcount, &i)))
2502 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2503 if ((error = xfs_btree_delete(cur, &i)))
2505 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2506 if ((error = xfs_btree_decrement(cur, 0, &i)))
2508 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2509 if ((error = xfs_bmbt_update(cur, new->br_startoff,
2511 new->br_blockcount + RIGHT.br_blockcount,
2517 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2519 * Setting all of a previous oldext extent to newext.
2520 * Neither the left nor right neighbors are contiguous with
2523 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2524 xfs_bmbt_set_state(ep, newext);
2525 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2528 rval = XFS_ILOG_DEXT;
2531 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2532 new->br_startblock, new->br_blockcount,
2535 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2536 if ((error = xfs_bmbt_update(cur, new->br_startoff,
2537 new->br_startblock, new->br_blockcount,
2543 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2545 * Setting the first part of a previous oldext extent to newext.
2546 * The left neighbor is contiguous.
2548 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
2549 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
2550 LEFT.br_blockcount + new->br_blockcount);
2551 xfs_bmbt_set_startoff(ep,
2552 PREV.br_startoff + new->br_blockcount);
2553 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
2555 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2556 xfs_bmbt_set_startblock(ep,
2557 new->br_startblock + new->br_blockcount);
2558 xfs_bmbt_set_blockcount(ep,
2559 PREV.br_blockcount - new->br_blockcount);
2560 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2565 rval = XFS_ILOG_DEXT;
2568 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2569 PREV.br_startblock, PREV.br_blockcount,
2572 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2573 if ((error = xfs_bmbt_update(cur,
2574 PREV.br_startoff + new->br_blockcount,
2575 PREV.br_startblock + new->br_blockcount,
2576 PREV.br_blockcount - new->br_blockcount,
2579 if ((error = xfs_btree_decrement(cur, 0, &i)))
2581 error = xfs_bmbt_update(cur, LEFT.br_startoff,
2583 LEFT.br_blockcount + new->br_blockcount,
2590 case BMAP_LEFT_FILLING:
2592 * Setting the first part of a previous oldext extent to newext.
2593 * The left neighbor is not contiguous.
2595 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2596 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
2597 xfs_bmbt_set_startoff(ep, new_endoff);
2598 xfs_bmbt_set_blockcount(ep,
2599 PREV.br_blockcount - new->br_blockcount);
2600 xfs_bmbt_set_startblock(ep,
2601 new->br_startblock + new->br_blockcount);
2602 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2604 xfs_iext_insert(ip, *idx, 1, new, state);
2605 ip->i_d.di_nextents++;
2607 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2609 rval = XFS_ILOG_CORE;
2610 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2611 PREV.br_startblock, PREV.br_blockcount,
2614 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2615 if ((error = xfs_bmbt_update(cur,
2616 PREV.br_startoff + new->br_blockcount,
2617 PREV.br_startblock + new->br_blockcount,
2618 PREV.br_blockcount - new->br_blockcount,
2621 cur->bc_rec.b = *new;
2622 if ((error = xfs_btree_insert(cur, &i)))
2624 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2628 case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2630 * Setting the last part of a previous oldext extent to newext.
2631 * The right neighbor is contiguous with the new allocation.
2633 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2634 xfs_bmbt_set_blockcount(ep,
2635 PREV.br_blockcount - new->br_blockcount);
2636 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2640 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2641 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2642 new->br_startoff, new->br_startblock,
2643 new->br_blockcount + RIGHT.br_blockcount, newext);
2644 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2647 rval = XFS_ILOG_DEXT;
2650 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2652 PREV.br_blockcount, &i)))
2654 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2655 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2657 PREV.br_blockcount - new->br_blockcount,
2660 if ((error = xfs_btree_increment(cur, 0, &i)))
2662 if ((error = xfs_bmbt_update(cur, new->br_startoff,
2664 new->br_blockcount + RIGHT.br_blockcount,
2670 case BMAP_RIGHT_FILLING:
2672 * Setting the last part of a previous oldext extent to newext.
2673 * The right neighbor is not contiguous.
2675 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2676 xfs_bmbt_set_blockcount(ep,
2677 PREV.br_blockcount - new->br_blockcount);
2678 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2681 xfs_iext_insert(ip, *idx, 1, new, state);
2683 ip->i_d.di_nextents++;
2685 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2687 rval = XFS_ILOG_CORE;
2688 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2689 PREV.br_startblock, PREV.br_blockcount,
2692 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2693 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2695 PREV.br_blockcount - new->br_blockcount,
2698 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2699 new->br_startblock, new->br_blockcount,
2702 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2703 cur->bc_rec.b.br_state = XFS_EXT_NORM;
2704 if ((error = xfs_btree_insert(cur, &i)))
2706 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2712 * Setting the middle part of a previous oldext extent to
2713 * newext. Contiguity is impossible here.
2714 * One extent becomes three extents.
2716 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2717 xfs_bmbt_set_blockcount(ep,
2718 new->br_startoff - PREV.br_startoff);
2719 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2722 r[1].br_startoff = new_endoff;
2723 r[1].br_blockcount =
2724 PREV.br_startoff + PREV.br_blockcount - new_endoff;
2725 r[1].br_startblock = new->br_startblock + new->br_blockcount;
2726 r[1].br_state = oldext;
2729 xfs_iext_insert(ip, *idx, 2, &r[0], state);
2731 ip->i_d.di_nextents += 2;
2733 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2735 rval = XFS_ILOG_CORE;
2736 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2737 PREV.br_startblock, PREV.br_blockcount,
2740 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2741 /* new right extent - oldext */
2742 if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
2743 r[1].br_startblock, r[1].br_blockcount,
2746 /* new left extent - oldext */
2747 cur->bc_rec.b = PREV;
2748 cur->bc_rec.b.br_blockcount =
2749 new->br_startoff - PREV.br_startoff;
2750 if ((error = xfs_btree_insert(cur, &i)))
2752 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2754 * Reset the cursor to the position of the new extent
2755 * we are about to insert as we can't trust it after
2756 * the previous insert.
2758 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2759 new->br_startblock, new->br_blockcount,
2762 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2763 /* new middle extent - newext */
2764 cur->bc_rec.b.br_state = new->br_state;
2765 if ((error = xfs_btree_insert(cur, &i)))
2767 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2771 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2772 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2773 case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2774 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2775 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2776 case BMAP_LEFT_CONTIG:
2777 case BMAP_RIGHT_CONTIG:
2779 * These cases are all impossible.
2784 /* convert to a btree if necessary */
2785 if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
2786 int tmp_logflags; /* partial log flag return val */
2788 ASSERT(cur == NULL);
2789 error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
2790 0, &tmp_logflags, XFS_DATA_FORK);
2791 *logflagsp |= tmp_logflags;
2796 /* clear out the allocated field, done with it now in any case. */
2798 cur->bc_private.b.allocated = 0;
2802 xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
2812 * Convert a hole to a delayed allocation.
2815 xfs_bmap_add_extent_hole_delay(
2816 xfs_inode_t *ip, /* incore inode pointer */
2817 xfs_extnum_t *idx, /* extent number to update/insert */
2818 xfs_bmbt_irec_t *new) /* new data to add to file extents */
2820 xfs_ifork_t *ifp; /* inode fork pointer */
2821 xfs_bmbt_irec_t left; /* left neighbor extent entry */
2822 xfs_filblks_t newlen=0; /* new indirect size */
2823 xfs_filblks_t oldlen=0; /* old indirect size */
2824 xfs_bmbt_irec_t right; /* right neighbor extent entry */
2825 int state; /* state bits, accessed thru macros */
2826 xfs_filblks_t temp=0; /* temp for indirect calculations */
2828 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2830 ASSERT(isnullstartblock(new->br_startblock));
2833 * Check and set flags if this segment has a left neighbor
2836 state |= BMAP_LEFT_VALID;
2837 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2839 if (isnullstartblock(left.br_startblock))
2840 state |= BMAP_LEFT_DELAY;
2844 * Check and set flags if the current (right) segment exists.
2845 * If it doesn't exist, we're converting the hole at end-of-file.
2847 if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2848 state |= BMAP_RIGHT_VALID;
2849 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2851 if (isnullstartblock(right.br_startblock))
2852 state |= BMAP_RIGHT_DELAY;
2856 * Set contiguity flags on the left and right neighbors.
2857 * Don't let extents get too large, even if the pieces are contiguous.
2859 if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2860 left.br_startoff + left.br_blockcount == new->br_startoff &&
2861 left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2862 state |= BMAP_LEFT_CONTIG;
2864 if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2865 new->br_startoff + new->br_blockcount == right.br_startoff &&
2866 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2867 (!(state & BMAP_LEFT_CONTIG) ||
2868 (left.br_blockcount + new->br_blockcount +
2869 right.br_blockcount <= MAXEXTLEN)))
2870 state |= BMAP_RIGHT_CONTIG;
2873 * Switch out based on the contiguity flags.
2875 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2876 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2878 * New allocation is contiguous with delayed allocations
2879 * on the left and on the right.
2880 * Merge all three into a single extent record.
2883 temp = left.br_blockcount + new->br_blockcount +
2884 right.br_blockcount;
2886 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2887 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2888 oldlen = startblockval(left.br_startblock) +
2889 startblockval(new->br_startblock) +
2890 startblockval(right.br_startblock);
2891 newlen = xfs_bmap_worst_indlen(ip, temp);
2892 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2893 nullstartblock((int)newlen));
2894 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2896 xfs_iext_remove(ip, *idx + 1, 1, state);
2899 case BMAP_LEFT_CONTIG:
2901 * New allocation is contiguous with a delayed allocation
2903 * Merge the new allocation with the left neighbor.
2906 temp = left.br_blockcount + new->br_blockcount;
2908 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2909 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2910 oldlen = startblockval(left.br_startblock) +
2911 startblockval(new->br_startblock);
2912 newlen = xfs_bmap_worst_indlen(ip, temp);
2913 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2914 nullstartblock((int)newlen));
2915 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2918 case BMAP_RIGHT_CONTIG:
2920 * New allocation is contiguous with a delayed allocation
2922 * Merge the new allocation with the right neighbor.
2924 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2925 temp = new->br_blockcount + right.br_blockcount;
2926 oldlen = startblockval(new->br_startblock) +
2927 startblockval(right.br_startblock);
2928 newlen = xfs_bmap_worst_indlen(ip, temp);
2929 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2931 nullstartblock((int)newlen), temp, right.br_state);
2932 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2937 * New allocation is not contiguous with another
2938 * delayed allocation.
2939 * Insert a new entry.
2941 oldlen = newlen = 0;
2942 xfs_iext_insert(ip, *idx, 1, new, state);
2945 if (oldlen != newlen) {
2946 ASSERT(oldlen > newlen);
2947 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
2948 (int64_t)(oldlen - newlen), 0);
2950 * Nothing to do for disk quota accounting here.
2956 * Convert a hole to a real allocation.
2958 STATIC int /* error */
2959 xfs_bmap_add_extent_hole_real(
2960 struct xfs_bmalloca *bma,
2963 struct xfs_bmbt_irec *new = &bma->got;
2964 int error; /* error return value */
2965 int i; /* temp state */
2966 xfs_ifork_t *ifp; /* inode fork pointer */
2967 xfs_bmbt_irec_t left; /* left neighbor extent entry */
2968 xfs_bmbt_irec_t right; /* right neighbor extent entry */
2969 int rval=0; /* return value (logging flags) */
2970 int state; /* state bits, accessed thru macros */
2972 ifp = XFS_IFORK_PTR(bma->ip, whichfork);
2974 ASSERT(bma->idx >= 0);
2975 ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2976 ASSERT(!isnullstartblock(new->br_startblock));
2978 !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2980 XFS_STATS_INC(xs_add_exlist);
2983 if (whichfork == XFS_ATTR_FORK)
2984 state |= BMAP_ATTRFORK;
2987 * Check and set flags if this segment has a left neighbor.
2990 state |= BMAP_LEFT_VALID;
2991 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
2992 if (isnullstartblock(left.br_startblock))
2993 state |= BMAP_LEFT_DELAY;
2997 * Check and set flags if this segment has a current value.
2998 * Not true if we're inserting into the "hole" at eof.
3000 if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
3001 state |= BMAP_RIGHT_VALID;
3002 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
3003 if (isnullstartblock(right.br_startblock))
3004 state |= BMAP_RIGHT_DELAY;
3008 * We're inserting a real allocation between "left" and "right".
3009 * Set the contiguity flags. Don't let extents get too large.
3011 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
3012 left.br_startoff + left.br_blockcount == new->br_startoff &&
3013 left.br_startblock + left.br_blockcount == new->br_startblock &&
3014 left.br_state == new->br_state &&
3015 left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
3016 state |= BMAP_LEFT_CONTIG;
3018 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
3019 new->br_startoff + new->br_blockcount == right.br_startoff &&
3020 new->br_startblock + new->br_blockcount == right.br_startblock &&
3021 new->br_state == right.br_state &&
3022 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
3023 (!(state & BMAP_LEFT_CONTIG) ||
3024 left.br_blockcount + new->br_blockcount +
3025 right.br_blockcount <= MAXEXTLEN))
3026 state |= BMAP_RIGHT_CONTIG;
3030 * Select which case we're in here, and implement it.
3032 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
3033 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
3035 * New allocation is contiguous with real allocations on the
3036 * left and on the right.
3037 * Merge all three into a single extent record.
3040 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3041 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3042 left.br_blockcount + new->br_blockcount +
3043 right.br_blockcount);
3044 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3046 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
3048 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3049 XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
3050 if (bma->cur == NULL) {
3051 rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3053 rval = XFS_ILOG_CORE;
3054 error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
3055 right.br_startblock, right.br_blockcount,
3059 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3060 error = xfs_btree_delete(bma->cur, &i);
3063 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3064 error = xfs_btree_decrement(bma->cur, 0, &i);
3067 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3068 error = xfs_bmbt_update(bma->cur, left.br_startoff,
3070 left.br_blockcount +
3071 new->br_blockcount +
3072 right.br_blockcount,
3079 case BMAP_LEFT_CONTIG:
3081 * New allocation is contiguous with a real allocation
3083 * Merge the new allocation with the left neighbor.
3086 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3087 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3088 left.br_blockcount + new->br_blockcount);
3089 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3091 if (bma->cur == NULL) {
3092 rval = xfs_ilog_fext(whichfork);
3095 error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
3096 left.br_startblock, left.br_blockcount,
3100 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3101 error = xfs_bmbt_update(bma->cur, left.br_startoff,
3103 left.br_blockcount +
3111 case BMAP_RIGHT_CONTIG:
3113 * New allocation is contiguous with a real allocation
3115 * Merge the new allocation with the right neighbor.
3117 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3118 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
3119 new->br_startoff, new->br_startblock,
3120 new->br_blockcount + right.br_blockcount,
3122 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3124 if (bma->cur == NULL) {
3125 rval = xfs_ilog_fext(whichfork);
3128 error = xfs_bmbt_lookup_eq(bma->cur,
3130 right.br_startblock,
3131 right.br_blockcount, &i);
3134 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3135 error = xfs_bmbt_update(bma->cur, new->br_startoff,
3137 new->br_blockcount +
3138 right.br_blockcount,
3147 * New allocation is not contiguous with another
3149 * Insert a new entry.
3151 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
3152 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3153 XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
3154 if (bma->cur == NULL) {
3155 rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3157 rval = XFS_ILOG_CORE;
3158 error = xfs_bmbt_lookup_eq(bma->cur,
3161 new->br_blockcount, &i);
3164 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
3165 bma->cur->bc_rec.b.br_state = new->br_state;
3166 error = xfs_btree_insert(bma->cur, &i);
3169 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3174 /* convert to a btree if necessary */
3175 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
3176 int tmp_logflags; /* partial log flag return val */
3178 ASSERT(bma->cur == NULL);
3179 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
3180 bma->firstblock, bma->flist, &bma->cur,
3181 0, &tmp_logflags, whichfork);
3182 bma->logflags |= tmp_logflags;
3187 /* clear out the allocated field, done with it now in any case. */
3189 bma->cur->bc_private.b.allocated = 0;
3191 xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
3193 bma->logflags |= rval;
3198 * Functions used in the extent read, allocate and remove paths
3202 * Adjust the size of the new extent based on di_extsize and rt extsize.
3205 xfs_bmap_extsize_align(
3207 xfs_bmbt_irec_t *gotp, /* next extent pointer */
3208 xfs_bmbt_irec_t *prevp, /* previous extent pointer */
3209 xfs_extlen_t extsz, /* align to this extent size */
3210 int rt, /* is this a realtime inode? */
3211 int eof, /* is extent at end-of-file? */
3212 int delay, /* creating delalloc extent? */
3213 int convert, /* overwriting unwritten extent? */
3214 xfs_fileoff_t *offp, /* in/out: aligned offset */
3215 xfs_extlen_t *lenp) /* in/out: aligned length */
3217 xfs_fileoff_t orig_off; /* original offset */
3218 xfs_extlen_t orig_alen; /* original length */
3219 xfs_fileoff_t orig_end; /* original off+len */
3220 xfs_fileoff_t nexto; /* next file offset */
3221 xfs_fileoff_t prevo; /* previous file offset */
3222 xfs_fileoff_t align_off; /* temp for offset */
3223 xfs_extlen_t align_alen; /* temp for length */
3224 xfs_extlen_t temp; /* temp for calculations */
3229 orig_off = align_off = *offp;
3230 orig_alen = align_alen = *lenp;
3231 orig_end = orig_off + orig_alen;
3234 * If this request overlaps an existing extent, then don't
3235 * attempt to perform any additional alignment.
3237 if (!delay && !eof &&
3238 (orig_off >= gotp->br_startoff) &&
3239 (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3244 * If the file offset is unaligned vs. the extent size
3245 * we need to align it. This will be possible unless
3246 * the file was previously written with a kernel that didn't
3247 * perform this alignment, or if a truncate shot us in the
3250 temp = do_mod(orig_off, extsz);
3256 * Same adjustment for the end of the requested area.
3258 if ((temp = (align_alen % extsz))) {
3259 align_alen += extsz - temp;
3262 * If the previous block overlaps with this proposed allocation
3263 * then move the start forward without adjusting the length.
3265 if (prevp->br_startoff != NULLFILEOFF) {
3266 if (prevp->br_startblock == HOLESTARTBLOCK)
3267 prevo = prevp->br_startoff;
3269 prevo = prevp->br_startoff + prevp->br_blockcount;
3272 if (align_off != orig_off && align_off < prevo)
3275 * If the next block overlaps with this proposed allocation
3276 * then move the start back without adjusting the length,
3277 * but not before offset 0.
3278 * This may of course make the start overlap previous block,
3279 * and if we hit the offset 0 limit then the next block
3280 * can still overlap too.
3282 if (!eof && gotp->br_startoff != NULLFILEOFF) {
3283 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3284 (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3285 nexto = gotp->br_startoff + gotp->br_blockcount;
3287 nexto = gotp->br_startoff;
3289 nexto = NULLFILEOFF;
3291 align_off + align_alen != orig_end &&
3292 align_off + align_alen > nexto)
3293 align_off = nexto > align_alen ? nexto - align_alen : 0;
3295 * If we're now overlapping the next or previous extent that
3296 * means we can't fit an extsz piece in this hole. Just move
3297 * the start forward to the first valid spot and set
3298 * the length so we hit the end.
3300 if (align_off != orig_off && align_off < prevo)
3302 if (align_off + align_alen != orig_end &&
3303 align_off + align_alen > nexto &&
3304 nexto != NULLFILEOFF) {
3305 ASSERT(nexto > prevo);
3306 align_alen = nexto - align_off;
3310 * If realtime, and the result isn't a multiple of the realtime
3311 * extent size we need to remove blocks until it is.
3313 if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3315 * We're not covering the original request, or
3316 * we won't be able to once we fix the length.
3318 if (orig_off < align_off ||
3319 orig_end > align_off + align_alen ||
3320 align_alen - temp < orig_alen)
3321 return XFS_ERROR(EINVAL);
3323 * Try to fix it by moving the start up.
3325 if (align_off + temp <= orig_off) {
3330 * Try to fix it by moving the end in.
3332 else if (align_off + align_alen - temp >= orig_end)
3335 * Set the start to the minimum then trim the length.
3338 align_alen -= orig_off - align_off;
3339 align_off = orig_off;
3340 align_alen -= align_alen % mp->m_sb.sb_rextsize;
3343 * Result doesn't cover the request, fail it.
3345 if (orig_off < align_off || orig_end > align_off + align_alen)
3346 return XFS_ERROR(EINVAL);
3348 ASSERT(orig_off >= align_off);
3349 ASSERT(orig_end <= align_off + align_alen);
3353 if (!eof && gotp->br_startoff != NULLFILEOFF)
3354 ASSERT(align_off + align_alen <= gotp->br_startoff);
3355 if (prevp->br_startoff != NULLFILEOFF)
3356 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3364 #define XFS_ALLOC_GAP_UNITS 4
3368 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
3370 xfs_fsblock_t adjust; /* adjustment to block numbers */
3371 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
3372 xfs_mount_t *mp; /* mount point structure */
3373 int nullfb; /* true if ap->firstblock isn't set */
3374 int rt; /* true if inode is realtime */
3376 #define ISVALID(x,y) \
3378 (x) < mp->m_sb.sb_rblocks : \
3379 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3380 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3381 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3383 mp = ap->ip->i_mount;
3384 nullfb = *ap->firstblock == NULLFSBLOCK;
3385 rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
3386 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3388 * If allocating at eof, and there's a previous real block,
3389 * try to use its last block as our starting point.
3391 if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3392 !isnullstartblock(ap->prev.br_startblock) &&
3393 ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3394 ap->prev.br_startblock)) {
3395 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3397 * Adjust for the gap between prevp and us.
3399 adjust = ap->offset -
3400 (ap->prev.br_startoff + ap->prev.br_blockcount);
3402 ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3403 ap->blkno += adjust;
3406 * If not at eof, then compare the two neighbor blocks.
3407 * Figure out whether either one gives us a good starting point,
3408 * and pick the better one.
3410 else if (!ap->eof) {
3411 xfs_fsblock_t gotbno; /* right side block number */
3412 xfs_fsblock_t gotdiff=0; /* right side difference */
3413 xfs_fsblock_t prevbno; /* left side block number */
3414 xfs_fsblock_t prevdiff=0; /* left side difference */
3417 * If there's a previous (left) block, select a requested
3418 * start block based on it.
3420 if (ap->prev.br_startoff != NULLFILEOFF &&
3421 !isnullstartblock(ap->prev.br_startblock) &&
3422 (prevbno = ap->prev.br_startblock +
3423 ap->prev.br_blockcount) &&
3424 ISVALID(prevbno, ap->prev.br_startblock)) {
3426 * Calculate gap to end of previous block.
3428 adjust = prevdiff = ap->offset -
3429 (ap->prev.br_startoff +
3430 ap->prev.br_blockcount);
3432 * Figure the startblock based on the previous block's
3433 * end and the gap size.
3435 * If the gap is large relative to the piece we're
3436 * allocating, or using it gives us an invalid block
3437 * number, then just use the end of the previous block.
3439 if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3440 ISVALID(prevbno + prevdiff,
3441 ap->prev.br_startblock))
3446 * If the firstblock forbids it, can't use it,
3449 if (!rt && !nullfb &&
3450 XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3451 prevbno = NULLFSBLOCK;
3454 * No previous block or can't follow it, just default.
3457 prevbno = NULLFSBLOCK;
3459 * If there's a following (right) block, select a requested
3460 * start block based on it.
3462 if (!isnullstartblock(ap->got.br_startblock)) {
3464 * Calculate gap to start of next block.
3466 adjust = gotdiff = ap->got.br_startoff - ap->offset;
3468 * Figure the startblock based on the next block's
3469 * start and the gap size.
3471 gotbno = ap->got.br_startblock;
3474 * If the gap is large relative to the piece we're
3475 * allocating, or using it gives us an invalid block
3476 * number, then just use the start of the next block
3477 * offset by our length.
3479 if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3480 ISVALID(gotbno - gotdiff, gotbno))
3482 else if (ISVALID(gotbno - ap->length, gotbno)) {
3483 gotbno -= ap->length;
3484 gotdiff += adjust - ap->length;
3488 * If the firstblock forbids it, can't use it,
3491 if (!rt && !nullfb &&
3492 XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3493 gotbno = NULLFSBLOCK;
3496 * No next block, just default.
3499 gotbno = NULLFSBLOCK;
3501 * If both valid, pick the better one, else the only good
3502 * one, else ap->blkno is already set (to 0 or the inode block).
3504 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3505 ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3506 else if (prevbno != NULLFSBLOCK)
3507 ap->blkno = prevbno;
3508 else if (gotbno != NULLFSBLOCK)
3515 xfs_bmap_btalloc_nullfb(
3516 struct xfs_bmalloca *ap,
3517 struct xfs_alloc_arg *args,
3520 struct xfs_mount *mp = ap->ip->i_mount;
3521 struct xfs_perag *pag;
3522 xfs_agnumber_t ag, startag;
3526 if (ap->userdata && xfs_inode_is_filestream(ap->ip))
3527 args->type = XFS_ALLOCTYPE_NEAR_BNO;
3529 args->type = XFS_ALLOCTYPE_START_BNO;
3530 args->total = ap->total;
3533 * Search for an allocation group with a single extent large enough
3534 * for the request. If one isn't found, then adjust the minimum
3535 * allocation size to the largest space found.
3537 startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3538 if (startag == NULLAGNUMBER)
3541 pag = xfs_perag_get(mp, ag);
3542 while (*blen < args->maxlen) {
3543 if (!pag->pagf_init) {
3544 error = xfs_alloc_pagf_init(mp, args->tp, ag,
3545 XFS_ALLOC_FLAG_TRYLOCK);
3553 * See xfs_alloc_fix_freelist...
3555 if (pag->pagf_init) {
3556 xfs_extlen_t longest;
3557 longest = xfs_alloc_longest_free_extent(mp, pag);
3558 if (*blen < longest)
3563 if (xfs_inode_is_filestream(ap->ip)) {
3564 if (*blen >= args->maxlen)
3569 * If startag is an invalid AG, we've
3570 * come here once before and
3571 * xfs_filestream_new_ag picked the
3572 * best currently available.
3574 * Don't continue looping, since we
3575 * could loop forever.
3577 if (startag == NULLAGNUMBER)
3580 error = xfs_filestream_new_ag(ap, &ag);
3585 /* loop again to set 'blen'*/
3586 startag = NULLAGNUMBER;
3587 pag = xfs_perag_get(mp, ag);
3591 if (++ag == mp->m_sb.sb_agcount)
3596 pag = xfs_perag_get(mp, ag);
3601 * Since the above loop did a BUF_TRYLOCK, it is
3602 * possible that there is space for this request.
3604 if (notinit || *blen < ap->minlen)
3605 args->minlen = ap->minlen;
3607 * If the best seen length is less than the request
3608 * length, use the best as the minimum.
3610 else if (*blen < args->maxlen)
3611 args->minlen = *blen;
3613 * Otherwise we've seen an extent as big as maxlen,
3614 * use that as the minimum.
3617 args->minlen = args->maxlen;
3620 * set the failure fallback case to look in the selected
3621 * AG as the stream may have moved.
3623 if (xfs_inode_is_filestream(ap->ip))
3624 ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3631 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
3633 xfs_mount_t *mp; /* mount point structure */
3634 xfs_alloctype_t atype = 0; /* type for allocation routines */
3635 xfs_extlen_t align; /* minimum allocation alignment */
3636 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
3638 xfs_alloc_arg_t args;
3640 xfs_extlen_t nextminlen = 0;
3641 int nullfb; /* true if ap->firstblock isn't set */
3648 mp = ap->ip->i_mount;
3649 align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
3650 if (unlikely(align)) {
3651 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3652 align, 0, ap->eof, 0, ap->conv,
3653 &ap->offset, &ap->length);
3657 nullfb = *ap->firstblock == NULLFSBLOCK;
3658 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3660 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
3661 ag = xfs_filestream_lookup_ag(ap->ip);
3662 ag = (ag != NULLAGNUMBER) ? ag : 0;
3663 ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3665 ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3668 ap->blkno = *ap->firstblock;
3670 xfs_bmap_adjacent(ap);
3673 * If allowed, use ap->blkno; otherwise must use firstblock since
3674 * it's in the right allocation group.
3676 if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3679 ap->blkno = *ap->firstblock;
3681 * Normal allocation, done through xfs_alloc_vextent.
3683 tryagain = isaligned = 0;
3684 memset(&args, 0, sizeof(args));
3687 args.fsbno = ap->blkno;
3689 /* Trim the allocation back to the maximum an AG can fit. */
3690 args.maxlen = MIN(ap->length, XFS_ALLOC_AG_MAX_USABLE(mp));
3691 args.firstblock = *ap->firstblock;
3694 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3697 } else if (ap->flist->xbf_low) {
3698 if (xfs_inode_is_filestream(ap->ip))
3699 args.type = XFS_ALLOCTYPE_FIRST_AG;
3701 args.type = XFS_ALLOCTYPE_START_BNO;
3702 args.total = args.minlen = ap->minlen;
3704 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3705 args.total = ap->total;
3706 args.minlen = ap->minlen;
3708 /* apply extent size hints if obtained earlier */
3709 if (unlikely(align)) {
3711 if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
3712 args.mod = (xfs_extlen_t)(args.prod - args.mod);
3713 } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
3717 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
3718 if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
3719 args.mod = (xfs_extlen_t)(args.prod - args.mod);
3722 * If we are not low on available data blocks, and the
3723 * underlying logical volume manager is a stripe, and
3724 * the file offset is zero then try to allocate data
3725 * blocks on stripe unit boundary.
3726 * NOTE: ap->aeof is only set if the allocation length
3727 * is >= the stripe unit and the allocation offset is
3728 * at the end of file.
3730 if (!ap->flist->xbf_low && ap->aeof) {
3732 args.alignment = mp->m_dalign;
3736 * Adjust for alignment
3738 if (blen > args.alignment && blen <= args.maxlen)
3739 args.minlen = blen - args.alignment;
3740 args.minalignslop = 0;
3743 * First try an exact bno allocation.
3744 * If it fails then do a near or start bno
3745 * allocation with alignment turned on.
3749 args.type = XFS_ALLOCTYPE_THIS_BNO;
3752 * Compute the minlen+alignment for the
3753 * next case. Set slop so that the value
3754 * of minlen+alignment+slop doesn't go up
3755 * between the calls.
3757 if (blen > mp->m_dalign && blen <= args.maxlen)
3758 nextminlen = blen - mp->m_dalign;
3760 nextminlen = args.minlen;
3761 if (nextminlen + mp->m_dalign > args.minlen + 1)
3763 nextminlen + mp->m_dalign -
3766 args.minalignslop = 0;
3770 args.minalignslop = 0;
3772 args.minleft = ap->minleft;
3773 args.wasdel = ap->wasdel;
3775 args.userdata = ap->userdata;
3776 if ((error = xfs_alloc_vextent(&args)))
3778 if (tryagain && args.fsbno == NULLFSBLOCK) {
3780 * Exact allocation failed. Now try with alignment
3784 args.fsbno = ap->blkno;
3785 args.alignment = mp->m_dalign;
3786 args.minlen = nextminlen;
3787 args.minalignslop = 0;
3789 if ((error = xfs_alloc_vextent(&args)))
3792 if (isaligned && args.fsbno == NULLFSBLOCK) {
3794 * allocation failed, so turn off alignment and
3798 args.fsbno = ap->blkno;
3800 if ((error = xfs_alloc_vextent(&args)))
3803 if (args.fsbno == NULLFSBLOCK && nullfb &&
3804 args.minlen > ap->minlen) {
3805 args.minlen = ap->minlen;
3806 args.type = XFS_ALLOCTYPE_START_BNO;
3807 args.fsbno = ap->blkno;
3808 if ((error = xfs_alloc_vextent(&args)))
3811 if (args.fsbno == NULLFSBLOCK && nullfb) {
3813 args.type = XFS_ALLOCTYPE_FIRST_AG;
3814 args.total = ap->minlen;
3816 if ((error = xfs_alloc_vextent(&args)))
3818 ap->flist->xbf_low = 1;
3820 if (args.fsbno != NULLFSBLOCK) {
3822 * check the allocation happened at the same or higher AG than
3823 * the first block that was allocated.
3825 ASSERT(*ap->firstblock == NULLFSBLOCK ||
3826 XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
3827 XFS_FSB_TO_AGNO(mp, args.fsbno) ||
3828 (ap->flist->xbf_low &&
3829 XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
3830 XFS_FSB_TO_AGNO(mp, args.fsbno)));
3832 ap->blkno = args.fsbno;
3833 if (*ap->firstblock == NULLFSBLOCK)
3834 *ap->firstblock = args.fsbno;
3835 ASSERT(nullfb || fb_agno == args.agno ||
3836 (ap->flist->xbf_low && fb_agno < args.agno));
3837 ap->length = args.len;
3838 ap->ip->i_d.di_nblocks += args.len;
3839 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3841 ap->ip->i_delayed_blks -= args.len;
3843 * Adjust the disk quota also. This was reserved
3846 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3847 ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
3848 XFS_TRANS_DQ_BCOUNT,
3851 ap->blkno = NULLFSBLOCK;
3858 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3859 * It figures out where to ask the underlying allocator to put the new extent.
3863 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
3865 if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
3866 return xfs_bmap_rtalloc(ap);
3867 return xfs_bmap_btalloc(ap);
3871 * Trim the returned map to the required bounds
3875 struct xfs_bmbt_irec *mval,
3876 struct xfs_bmbt_irec *got,
3884 if ((flags & XFS_BMAPI_ENTIRE) ||
3885 got->br_startoff + got->br_blockcount <= obno) {
3887 if (isnullstartblock(got->br_startblock))
3888 mval->br_startblock = DELAYSTARTBLOCK;
3894 ASSERT((*bno >= obno) || (n == 0));
3896 mval->br_startoff = *bno;
3897 if (isnullstartblock(got->br_startblock))
3898 mval->br_startblock = DELAYSTARTBLOCK;
3900 mval->br_startblock = got->br_startblock +
3901 (*bno - got->br_startoff);
3903 * Return the minimum of what we got and what we asked for for
3904 * the length. We can use the len variable here because it is
3905 * modified below and we could have been there before coming
3906 * here if the first part of the allocation didn't overlap what
3909 mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3910 got->br_blockcount - (*bno - got->br_startoff));
3911 mval->br_state = got->br_state;
3912 ASSERT(mval->br_blockcount <= len);
3917 * Update and validate the extent map to return
3920 xfs_bmapi_update_map(
3921 struct xfs_bmbt_irec **map,
3929 xfs_bmbt_irec_t *mval = *map;
3931 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3932 ((mval->br_startoff + mval->br_blockcount) <= end));
3933 ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3934 (mval->br_startoff < obno));
3936 *bno = mval->br_startoff + mval->br_blockcount;
3938 if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3939 /* update previous map with new information */
3940 ASSERT(mval->br_startblock == mval[-1].br_startblock);
3941 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3942 ASSERT(mval->br_state == mval[-1].br_state);
3943 mval[-1].br_blockcount = mval->br_blockcount;
3944 mval[-1].br_state = mval->br_state;
3945 } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3946 mval[-1].br_startblock != DELAYSTARTBLOCK &&
3947 mval[-1].br_startblock != HOLESTARTBLOCK &&
3948 mval->br_startblock == mval[-1].br_startblock +
3949 mval[-1].br_blockcount &&
3950 ((flags & XFS_BMAPI_IGSTATE) ||
3951 mval[-1].br_state == mval->br_state)) {
3952 ASSERT(mval->br_startoff ==
3953 mval[-1].br_startoff + mval[-1].br_blockcount);
3954 mval[-1].br_blockcount += mval->br_blockcount;
3955 } else if (*n > 0 &&
3956 mval->br_startblock == DELAYSTARTBLOCK &&
3957 mval[-1].br_startblock == DELAYSTARTBLOCK &&
3958 mval->br_startoff ==
3959 mval[-1].br_startoff + mval[-1].br_blockcount) {
3960 mval[-1].br_blockcount += mval->br_blockcount;
3961 mval[-1].br_state = mval->br_state;
3962 } else if (!((*n == 0) &&
3963 ((mval->br_startoff + mval->br_blockcount) <=
3972 * Map file blocks to filesystem blocks without allocation.
3976 struct xfs_inode *ip,
3979 struct xfs_bmbt_irec *mval,
3983 struct xfs_mount *mp = ip->i_mount;
3984 struct xfs_ifork *ifp;
3985 struct xfs_bmbt_irec got;
3986 struct xfs_bmbt_irec prev;
3993 int whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
3994 XFS_ATTR_FORK : XFS_DATA_FORK;
3997 ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
3998 XFS_BMAPI_IGSTATE)));
4000 if (unlikely(XFS_TEST_ERROR(
4001 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4002 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4003 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4004 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4005 return XFS_ERROR(EFSCORRUPTED);
4008 if (XFS_FORCED_SHUTDOWN(mp))
4009 return XFS_ERROR(EIO);
4011 XFS_STATS_INC(xs_blk_mapr);
4013 ifp = XFS_IFORK_PTR(ip, whichfork);
4015 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4016 error = xfs_iread_extents(NULL, ip, whichfork);
4021 xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4025 while (bno < end && n < *nmap) {
4026 /* Reading past eof, act as though there's a hole up to end. */
4028 got.br_startoff = end;
4029 if (got.br_startoff > bno) {
4030 /* Reading in a hole. */
4031 mval->br_startoff = bno;
4032 mval->br_startblock = HOLESTARTBLOCK;
4033 mval->br_blockcount =
4034 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4035 mval->br_state = XFS_EXT_NORM;
4036 bno += mval->br_blockcount;
4037 len -= mval->br_blockcount;
4043 /* set up the extent map to return. */
4044 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4045 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4047 /* If we're done, stop now. */
4048 if (bno >= end || n >= *nmap)
4051 /* Else go on to the next record. */
4052 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4053 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4062 xfs_bmapi_reserve_delalloc(
4063 struct xfs_inode *ip,
4066 struct xfs_bmbt_irec *got,
4067 struct xfs_bmbt_irec *prev,
4068 xfs_extnum_t *lastx,
4071 struct xfs_mount *mp = ip->i_mount;
4072 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4074 xfs_extlen_t indlen;
4075 char rt = XFS_IS_REALTIME_INODE(ip);
4079 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4081 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4083 /* Figure out the extent size, adjust alen */
4084 extsz = xfs_get_extsz_hint(ip);
4087 * Make sure we don't exceed a single extent length when we
4088 * align the extent by reducing length we are going to
4089 * allocate by the maximum amount extent size aligment may
4092 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN - (2 * extsz - 1));
4093 error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4094 1, 0, &aoff, &alen);
4099 extsz = alen / mp->m_sb.sb_rextsize;
4102 * Make a transaction-less quota reservation for delayed allocation
4103 * blocks. This number gets adjusted later. We return if we haven't
4104 * allocated blocks already inside this loop.
4106 error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4107 rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4112 * Split changing sb for alen and indlen since they could be coming
4113 * from different places.
4115 indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4119 error = xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
4120 -((int64_t)extsz), 0);
4122 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4123 -((int64_t)alen), 0);
4127 goto out_unreserve_quota;
4129 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4130 -((int64_t)indlen), 0);
4132 goto out_unreserve_blocks;
4135 ip->i_delayed_blks += alen;
4137 got->br_startoff = aoff;
4138 got->br_startblock = nullstartblock(indlen);
4139 got->br_blockcount = alen;
4140 got->br_state = XFS_EXT_NORM;
4141 xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4144 * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4145 * might have merged it into one of the neighbouring ones.
4147 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4149 ASSERT(got->br_startoff <= aoff);
4150 ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4151 ASSERT(isnullstartblock(got->br_startblock));
4152 ASSERT(got->br_state == XFS_EXT_NORM);
4155 out_unreserve_blocks:
4157 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS, extsz, 0);
4159 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, alen, 0);
4160 out_unreserve_quota:
4161 if (XFS_IS_QUOTA_ON(mp))
4162 xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4163 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4168 * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4172 struct xfs_inode *ip, /* incore inode */
4173 xfs_fileoff_t bno, /* starting file offs. mapped */
4174 xfs_filblks_t len, /* length to map in file */
4175 struct xfs_bmbt_irec *mval, /* output: map values */
4176 int *nmap, /* i/o: mval size/count */
4177 int flags) /* XFS_BMAPI_... */
4179 struct xfs_mount *mp = ip->i_mount;
4180 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4181 struct xfs_bmbt_irec got; /* current file extent record */
4182 struct xfs_bmbt_irec prev; /* previous file extent record */
4183 xfs_fileoff_t obno; /* old block number (offset) */
4184 xfs_fileoff_t end; /* end of mapped file region */
4185 xfs_extnum_t lastx; /* last useful extent number */
4186 int eof; /* we've hit the end of extents */
4187 int n = 0; /* current extent index */
4191 ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4192 ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4194 if (unlikely(XFS_TEST_ERROR(
4195 (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4196 XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4197 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4198 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4199 return XFS_ERROR(EFSCORRUPTED);
4202 if (XFS_FORCED_SHUTDOWN(mp))
4203 return XFS_ERROR(EIO);
4205 XFS_STATS_INC(xs_blk_mapw);
4207 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4208 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4213 xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4217 while (bno < end && n < *nmap) {
4218 if (eof || got.br_startoff > bno) {
4219 error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4220 &prev, &lastx, eof);
4230 /* set up the extent map to return. */
4231 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4232 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4234 /* If we're done, stop now. */
4235 if (bno >= end || n >= *nmap)
4238 /* Else go on to the next record. */
4240 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4241 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4252 __xfs_bmapi_allocate(
4253 struct xfs_bmalloca *bma)
4255 struct xfs_mount *mp = bma->ip->i_mount;
4256 int whichfork = (bma->flags & XFS_BMAPI_ATTRFORK) ?
4257 XFS_ATTR_FORK : XFS_DATA_FORK;
4258 struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4259 int tmp_logflags = 0;
4262 ASSERT(bma->length > 0);
4265 * For the wasdelay case, we could also just allocate the stuff asked
4266 * for in this bmap call but that wouldn't be as good.
4269 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4270 bma->offset = bma->got.br_startoff;
4271 if (bma->idx != NULLEXTNUM && bma->idx) {
4272 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4276 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4278 bma->length = XFS_FILBLKS_MIN(bma->length,
4279 bma->got.br_startoff - bma->offset);
4283 * Indicate if this is the first user data in the file, or just any
4286 if (!(bma->flags & XFS_BMAPI_METADATA)) {
4287 bma->userdata = (bma->offset == 0) ?
4288 XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4291 bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4294 * Only want to do the alignment at the eof if it is userdata and
4295 * allocation length is larger than a stripe unit.
4297 if (mp->m_dalign && bma->length >= mp->m_dalign &&
4298 !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4299 error = xfs_bmap_isaeof(bma, whichfork);
4304 error = xfs_bmap_alloc(bma);
4308 if (bma->flist->xbf_low)
4311 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4312 if (bma->blkno == NULLFSBLOCK)
4314 if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4315 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4316 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4317 bma->cur->bc_private.b.flist = bma->flist;
4320 * Bump the number of extents we've allocated
4326 bma->cur->bc_private.b.flags =
4327 bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4329 bma->got.br_startoff = bma->offset;
4330 bma->got.br_startblock = bma->blkno;
4331 bma->got.br_blockcount = bma->length;
4332 bma->got.br_state = XFS_EXT_NORM;
4335 * A wasdelay extent has been initialized, so shouldn't be flagged
4338 if (!bma->wasdel && (bma->flags & XFS_BMAPI_PREALLOC) &&
4339 xfs_sb_version_hasextflgbit(&mp->m_sb))
4340 bma->got.br_state = XFS_EXT_UNWRITTEN;
4343 error = xfs_bmap_add_extent_delay_real(bma);
4345 error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4347 bma->logflags |= tmp_logflags;
4352 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4353 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4354 * the neighbouring ones.
4356 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4358 ASSERT(bma->got.br_startoff <= bma->offset);
4359 ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4360 bma->offset + bma->length);
4361 ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4362 bma->got.br_state == XFS_EXT_UNWRITTEN);
4367 xfs_bmapi_convert_unwritten(
4368 struct xfs_bmalloca *bma,
4369 struct xfs_bmbt_irec *mval,
4373 int whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4374 XFS_ATTR_FORK : XFS_DATA_FORK;
4375 struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4376 int tmp_logflags = 0;
4379 /* check if we need to do unwritten->real conversion */
4380 if (mval->br_state == XFS_EXT_UNWRITTEN &&
4381 (flags & XFS_BMAPI_PREALLOC))
4384 /* check if we need to do real->unwritten conversion */
4385 if (mval->br_state == XFS_EXT_NORM &&
4386 (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4387 (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4391 * Modify (by adding) the state flag, if writing.
4393 ASSERT(mval->br_blockcount <= len);
4394 if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4395 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4396 bma->ip, whichfork);
4397 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4398 bma->cur->bc_private.b.flist = bma->flist;
4400 mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4401 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4403 error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4404 &bma->cur, mval, bma->firstblock, bma->flist,
4406 bma->logflags |= tmp_logflags;
4411 * Update our extent pointer, given that
4412 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4413 * of the neighbouring ones.
4415 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4418 * We may have combined previously unwritten space with written space,
4419 * so generate another request.
4421 if (mval->br_blockcount < len)
4427 * Map file blocks to filesystem blocks, and allocate blocks or convert the
4428 * extent state if necessary. Details behaviour is controlled by the flags
4429 * parameter. Only allocates blocks from a single allocation group, to avoid
4432 * The returned value in "firstblock" from the first call in a transaction
4433 * must be remembered and presented to subsequent calls in "firstblock".
4434 * An upper bound for the number of blocks to be allocated is supplied to
4435 * the first call in "total"; if no allocation group has that many free
4436 * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4440 struct xfs_trans *tp, /* transaction pointer */
4441 struct xfs_inode *ip, /* incore inode */
4442 xfs_fileoff_t bno, /* starting file offs. mapped */
4443 xfs_filblks_t len, /* length to map in file */
4444 int flags, /* XFS_BMAPI_... */
4445 xfs_fsblock_t *firstblock, /* first allocated block
4446 controls a.g. for allocs */
4447 xfs_extlen_t total, /* total blocks needed */
4448 struct xfs_bmbt_irec *mval, /* output: map values */
4449 int *nmap, /* i/o: mval size/count */
4450 struct xfs_bmap_free *flist) /* i/o: list extents to free */
4452 struct xfs_mount *mp = ip->i_mount;
4453 struct xfs_ifork *ifp;
4454 struct xfs_bmalloca bma = { NULL }; /* args for xfs_bmap_alloc */
4455 xfs_fileoff_t end; /* end of mapped file region */
4456 int eof; /* after the end of extents */
4457 int error; /* error return */
4458 int n; /* current extent index */
4459 xfs_fileoff_t obno; /* old block number (offset) */
4460 int whichfork; /* data or attr fork */
4461 char inhole; /* current location is hole in file */
4462 char wasdelay; /* old extent was delayed */
4465 xfs_fileoff_t orig_bno; /* original block number value */
4466 int orig_flags; /* original flags arg value */
4467 xfs_filblks_t orig_len; /* original value of len arg */
4468 struct xfs_bmbt_irec *orig_mval; /* original value of mval */
4469 int orig_nmap; /* original value of *nmap */
4477 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4478 XFS_ATTR_FORK : XFS_DATA_FORK;
4481 ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4482 ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4485 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4487 if (unlikely(XFS_TEST_ERROR(
4488 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4489 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4490 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4491 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4492 return XFS_ERROR(EFSCORRUPTED);
4495 if (XFS_FORCED_SHUTDOWN(mp))
4496 return XFS_ERROR(EIO);
4498 ifp = XFS_IFORK_PTR(ip, whichfork);
4500 XFS_STATS_INC(xs_blk_mapw);
4502 if (*firstblock == NULLFSBLOCK) {
4503 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4504 bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4511 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4512 error = xfs_iread_extents(tp, ip, whichfork);
4517 xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
4528 bma.firstblock = firstblock;
4530 if (flags & XFS_BMAPI_STACK_SWITCH)
4531 bma.stack_switch = 1;
4533 while (bno < end && n < *nmap) {
4534 inhole = eof || bma.got.br_startoff > bno;
4535 wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
4538 * First, deal with the hole before the allocated space
4539 * that we found, if any.
4541 if (inhole || wasdelay) {
4543 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4544 bma.wasdel = wasdelay;
4549 * There's a 32/64 bit type mismatch between the
4550 * allocation length request (which can be 64 bits in
4551 * length) and the bma length request, which is
4552 * xfs_extlen_t and therefore 32 bits. Hence we have to
4553 * check for 32-bit overflows and handle them here.
4555 if (len > (xfs_filblks_t)MAXEXTLEN)
4556 bma.length = MAXEXTLEN;
4561 ASSERT(bma.length > 0);
4562 error = xfs_bmapi_allocate(&bma);
4565 if (bma.blkno == NULLFSBLOCK)
4569 /* Deal with the allocated space we found. */
4570 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4573 /* Execute unwritten extent conversion if necessary */
4574 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4575 if (error == EAGAIN)
4580 /* update the extent map to return */
4581 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4584 * If we're done, stop now. Stop when we've allocated
4585 * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
4586 * the transaction may get too big.
4588 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4591 /* Else go on to the next record. */
4593 if (++bma.idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
4594 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
4602 * Transform from btree to extents, give it cur.
4604 if (xfs_bmap_wants_extents(ip, whichfork)) {
4605 int tmp_logflags = 0;
4608 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4609 &tmp_logflags, whichfork);
4610 bma.logflags |= tmp_logflags;
4615 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4616 XFS_IFORK_NEXTENTS(ip, whichfork) >
4617 XFS_IFORK_MAXEXT(ip, whichfork));
4621 * Log everything. Do this after conversion, there's no point in
4622 * logging the extent records if we've converted to btree format.
4624 if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
4625 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4626 bma.logflags &= ~xfs_ilog_fext(whichfork);
4627 else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
4628 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4629 bma.logflags &= ~xfs_ilog_fbroot(whichfork);
4631 * Log whatever the flags say, even if error. Otherwise we might miss
4632 * detecting a case where the data is changed, there's an error,
4633 * and it's not logged so we don't shutdown when we should.
4636 xfs_trans_log_inode(tp, ip, bma.logflags);
4640 ASSERT(*firstblock == NULLFSBLOCK ||
4641 XFS_FSB_TO_AGNO(mp, *firstblock) ==
4643 bma.cur->bc_private.b.firstblock) ||
4645 XFS_FSB_TO_AGNO(mp, *firstblock) <
4647 bma.cur->bc_private.b.firstblock)));
4648 *firstblock = bma.cur->bc_private.b.firstblock;
4650 xfs_btree_del_cursor(bma.cur,
4651 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4654 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4660 * Called by xfs_bmapi to update file extent records and the btree
4661 * after removing space (or undoing a delayed allocation).
4663 STATIC int /* error */
4664 xfs_bmap_del_extent(
4665 xfs_inode_t *ip, /* incore inode pointer */
4666 xfs_trans_t *tp, /* current transaction pointer */
4667 xfs_extnum_t *idx, /* extent number to update/delete */
4668 xfs_bmap_free_t *flist, /* list of extents to be freed */
4669 xfs_btree_cur_t *cur, /* if null, not a btree */
4670 xfs_bmbt_irec_t *del, /* data to remove from extents */
4671 int *logflagsp, /* inode logging flags */
4672 int whichfork) /* data or attr fork */
4674 xfs_filblks_t da_new; /* new delay-alloc indirect blocks */
4675 xfs_filblks_t da_old; /* old delay-alloc indirect blocks */
4676 xfs_fsblock_t del_endblock=0; /* first block past del */
4677 xfs_fileoff_t del_endoff; /* first offset past del */
4678 int delay; /* current block is delayed allocated */
4679 int do_fx; /* free extent at end of routine */
4680 xfs_bmbt_rec_host_t *ep; /* current extent entry pointer */
4681 int error; /* error return value */
4682 int flags; /* inode logging flags */
4683 xfs_bmbt_irec_t got; /* current extent entry */
4684 xfs_fileoff_t got_endoff; /* first offset past got */
4685 int i; /* temp state */
4686 xfs_ifork_t *ifp; /* inode fork pointer */
4687 xfs_mount_t *mp; /* mount structure */
4688 xfs_filblks_t nblks; /* quota/sb block count */
4689 xfs_bmbt_irec_t new; /* new record to be inserted */
4691 uint qfield; /* quota field to update */
4692 xfs_filblks_t temp; /* for indirect length calculations */
4693 xfs_filblks_t temp2; /* for indirect length calculations */
4696 XFS_STATS_INC(xs_del_exlist);
4698 if (whichfork == XFS_ATTR_FORK)
4699 state |= BMAP_ATTRFORK;
4702 ifp = XFS_IFORK_PTR(ip, whichfork);
4703 ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
4704 (uint)sizeof(xfs_bmbt_rec_t)));
4705 ASSERT(del->br_blockcount > 0);
4706 ep = xfs_iext_get_ext(ifp, *idx);
4707 xfs_bmbt_get_all(ep, &got);
4708 ASSERT(got.br_startoff <= del->br_startoff);
4709 del_endoff = del->br_startoff + del->br_blockcount;
4710 got_endoff = got.br_startoff + got.br_blockcount;
4711 ASSERT(got_endoff >= del_endoff);
4712 delay = isnullstartblock(got.br_startblock);
4713 ASSERT(isnullstartblock(del->br_startblock) == delay);
4718 * If deleting a real allocation, must free up the disk space.
4721 flags = XFS_ILOG_CORE;
4723 * Realtime allocation. Free it and record di_nblocks update.
4725 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
4729 ASSERT(do_mod(del->br_blockcount,
4730 mp->m_sb.sb_rextsize) == 0);
4731 ASSERT(do_mod(del->br_startblock,
4732 mp->m_sb.sb_rextsize) == 0);
4733 bno = del->br_startblock;
4734 len = del->br_blockcount;
4735 do_div(bno, mp->m_sb.sb_rextsize);
4736 do_div(len, mp->m_sb.sb_rextsize);
4737 error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
4741 nblks = len * mp->m_sb.sb_rextsize;
4742 qfield = XFS_TRANS_DQ_RTBCOUNT;
4745 * Ordinary allocation.
4749 nblks = del->br_blockcount;
4750 qfield = XFS_TRANS_DQ_BCOUNT;
4753 * Set up del_endblock and cur for later.
4755 del_endblock = del->br_startblock + del->br_blockcount;
4757 if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
4758 got.br_startblock, got.br_blockcount,
4761 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4763 da_old = da_new = 0;
4765 da_old = startblockval(got.br_startblock);
4771 * Set flag value to use in switch statement.
4772 * Left-contig is 2, right-contig is 1.
4774 switch (((got.br_startoff == del->br_startoff) << 1) |
4775 (got_endoff == del_endoff)) {
4778 * Matches the whole extent. Delete the entry.
4780 xfs_iext_remove(ip, *idx, 1,
4781 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
4786 XFS_IFORK_NEXT_SET(ip, whichfork,
4787 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
4788 flags |= XFS_ILOG_CORE;
4790 flags |= xfs_ilog_fext(whichfork);
4793 if ((error = xfs_btree_delete(cur, &i)))
4795 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4800 * Deleting the first part of the extent.
4802 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4803 xfs_bmbt_set_startoff(ep, del_endoff);
4804 temp = got.br_blockcount - del->br_blockcount;
4805 xfs_bmbt_set_blockcount(ep, temp);
4807 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4809 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4810 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4814 xfs_bmbt_set_startblock(ep, del_endblock);
4815 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4817 flags |= xfs_ilog_fext(whichfork);
4820 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
4821 got.br_blockcount - del->br_blockcount,
4828 * Deleting the last part of the extent.
4830 temp = got.br_blockcount - del->br_blockcount;
4831 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4832 xfs_bmbt_set_blockcount(ep, temp);
4834 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4836 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4837 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4841 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4843 flags |= xfs_ilog_fext(whichfork);
4846 if ((error = xfs_bmbt_update(cur, got.br_startoff,
4848 got.br_blockcount - del->br_blockcount,
4855 * Deleting the middle of the extent.
4857 temp = del->br_startoff - got.br_startoff;
4858 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4859 xfs_bmbt_set_blockcount(ep, temp);
4860 new.br_startoff = del_endoff;
4861 temp2 = got_endoff - del_endoff;
4862 new.br_blockcount = temp2;
4863 new.br_state = got.br_state;
4865 new.br_startblock = del_endblock;
4866 flags |= XFS_ILOG_CORE;
4868 if ((error = xfs_bmbt_update(cur,
4870 got.br_startblock, temp,
4873 if ((error = xfs_btree_increment(cur, 0, &i)))
4875 cur->bc_rec.b = new;
4876 error = xfs_btree_insert(cur, &i);
4877 if (error && error != ENOSPC)
4880 * If get no-space back from btree insert,
4881 * it tried a split, and we have a zero
4882 * block reservation.
4883 * Fix up our state and return the error.
4885 if (error == ENOSPC) {
4887 * Reset the cursor, don't trust
4888 * it after any insert operation.
4890 if ((error = xfs_bmbt_lookup_eq(cur,
4895 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4897 * Update the btree record back
4898 * to the original value.
4900 if ((error = xfs_bmbt_update(cur,
4907 * Reset the extent record back
4908 * to the original value.
4910 xfs_bmbt_set_blockcount(ep,
4913 error = XFS_ERROR(ENOSPC);
4916 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4918 flags |= xfs_ilog_fext(whichfork);
4919 XFS_IFORK_NEXT_SET(ip, whichfork,
4920 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
4922 ASSERT(whichfork == XFS_DATA_FORK);
4923 temp = xfs_bmap_worst_indlen(ip, temp);
4924 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4925 temp2 = xfs_bmap_worst_indlen(ip, temp2);
4926 new.br_startblock = nullstartblock((int)temp2);
4927 da_new = temp + temp2;
4928 while (da_new > da_old) {
4932 xfs_bmbt_set_startblock(ep,
4933 nullstartblock((int)temp));
4935 if (da_new == da_old)
4941 nullstartblock((int)temp2);
4945 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4946 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
4951 * If we need to, add to list of extents to delete.
4954 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
4957 * Adjust inode # blocks in the file.
4960 ip->i_d.di_nblocks -= nblks;
4962 * Adjust quota data.
4965 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
4968 * Account for change in delayed indirect blocks.
4969 * Nothing to do for disk quota accounting here.
4971 ASSERT(da_old >= da_new);
4972 if (da_old > da_new) {
4973 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4974 (int64_t)(da_old - da_new), 0);
4982 * Unmap (remove) blocks from a file.
4983 * If nexts is nonzero then the number of extents to remove is limited to
4984 * that value. If not all extents in the block range can be removed then
4989 xfs_trans_t *tp, /* transaction pointer */
4990 struct xfs_inode *ip, /* incore inode */
4991 xfs_fileoff_t bno, /* starting offset to unmap */
4992 xfs_filblks_t len, /* length to unmap in file */
4993 int flags, /* misc flags */
4994 xfs_extnum_t nexts, /* number of extents max */
4995 xfs_fsblock_t *firstblock, /* first allocated block
4996 controls a.g. for allocs */
4997 xfs_bmap_free_t *flist, /* i/o: list extents to free */
4998 int *done) /* set if not done yet */
5000 xfs_btree_cur_t *cur; /* bmap btree cursor */
5001 xfs_bmbt_irec_t del; /* extent being deleted */
5002 int eof; /* is deleting at eof */
5003 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
5004 int error; /* error return value */
5005 xfs_extnum_t extno; /* extent number in list */
5006 xfs_bmbt_irec_t got; /* current extent record */
5007 xfs_ifork_t *ifp; /* inode fork pointer */
5008 int isrt; /* freeing in rt area */
5009 xfs_extnum_t lastx; /* last extent index used */
5010 int logflags; /* transaction logging flags */
5011 xfs_extlen_t mod; /* rt extent offset */
5012 xfs_mount_t *mp; /* mount structure */
5013 xfs_extnum_t nextents; /* number of file extents */
5014 xfs_bmbt_irec_t prev; /* previous extent record */
5015 xfs_fileoff_t start; /* first file offset deleted */
5016 int tmp_logflags; /* partial logging flags */
5017 int wasdel; /* was a delayed alloc extent */
5018 int whichfork; /* data or attribute fork */
5021 trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5023 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5024 XFS_ATTR_FORK : XFS_DATA_FORK;
5025 ifp = XFS_IFORK_PTR(ip, whichfork);
5027 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5028 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5029 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5031 return XFS_ERROR(EFSCORRUPTED);
5034 if (XFS_FORCED_SHUTDOWN(mp))
5035 return XFS_ERROR(EIO);
5040 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5041 (error = xfs_iread_extents(tp, ip, whichfork)))
5043 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5044 if (nextents == 0) {
5048 XFS_STATS_INC(xs_blk_unmap);
5049 isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5051 bno = start + len - 1;
5052 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5056 * Check to see if the given block number is past the end of the
5057 * file, back up to the last block if so...
5060 ep = xfs_iext_get_ext(ifp, --lastx);
5061 xfs_bmbt_get_all(ep, &got);
5062 bno = got.br_startoff + got.br_blockcount - 1;
5065 if (ifp->if_flags & XFS_IFBROOT) {
5066 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5067 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5068 cur->bc_private.b.firstblock = *firstblock;
5069 cur->bc_private.b.flist = flist;
5070 cur->bc_private.b.flags = 0;
5076 * Synchronize by locking the bitmap inode.
5078 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
5079 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5083 while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5084 (nexts == 0 || extno < nexts)) {
5086 * Is the found extent after a hole in which bno lives?
5087 * Just back up to the previous extent, if so.
5089 if (got.br_startoff > bno) {
5092 ep = xfs_iext_get_ext(ifp, lastx);
5093 xfs_bmbt_get_all(ep, &got);
5096 * Is the last block of this extent before the range
5097 * we're supposed to delete? If so, we're done.
5099 bno = XFS_FILEOFF_MIN(bno,
5100 got.br_startoff + got.br_blockcount - 1);
5104 * Then deal with the (possibly delayed) allocated space
5109 wasdel = isnullstartblock(del.br_startblock);
5110 if (got.br_startoff < start) {
5111 del.br_startoff = start;
5112 del.br_blockcount -= start - got.br_startoff;
5114 del.br_startblock += start - got.br_startoff;
5116 if (del.br_startoff + del.br_blockcount > bno + 1)
5117 del.br_blockcount = bno + 1 - del.br_startoff;
5118 sum = del.br_startblock + del.br_blockcount;
5120 (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5122 * Realtime extent not lined up at the end.
5123 * The extent could have been split into written
5124 * and unwritten pieces, or we could just be
5125 * unmapping part of it. But we can't really
5126 * get rid of part of a realtime extent.
5128 if (del.br_state == XFS_EXT_UNWRITTEN ||
5129 !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5131 * This piece is unwritten, or we're not
5132 * using unwritten extents. Skip over it.
5135 bno -= mod > del.br_blockcount ?
5136 del.br_blockcount : mod;
5137 if (bno < got.br_startoff) {
5139 xfs_bmbt_get_all(xfs_iext_get_ext(
5145 * It's written, turn it unwritten.
5146 * This is better than zeroing it.
5148 ASSERT(del.br_state == XFS_EXT_NORM);
5149 ASSERT(xfs_trans_get_block_res(tp) > 0);
5151 * If this spans a realtime extent boundary,
5152 * chop it back to the start of the one we end at.
5154 if (del.br_blockcount > mod) {
5155 del.br_startoff += del.br_blockcount - mod;
5156 del.br_startblock += del.br_blockcount - mod;
5157 del.br_blockcount = mod;
5159 del.br_state = XFS_EXT_UNWRITTEN;
5160 error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5161 &lastx, &cur, &del, firstblock, flist,
5167 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5169 * Realtime extent is lined up at the end but not
5170 * at the front. We'll get rid of full extents if
5173 mod = mp->m_sb.sb_rextsize - mod;
5174 if (del.br_blockcount > mod) {
5175 del.br_blockcount -= mod;
5176 del.br_startoff += mod;
5177 del.br_startblock += mod;
5178 } else if ((del.br_startoff == start &&
5179 (del.br_state == XFS_EXT_UNWRITTEN ||
5180 xfs_trans_get_block_res(tp) == 0)) ||
5181 !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5183 * Can't make it unwritten. There isn't
5184 * a full extent here so just skip it.
5186 ASSERT(bno >= del.br_blockcount);
5187 bno -= del.br_blockcount;
5188 if (got.br_startoff > bno) {
5190 ep = xfs_iext_get_ext(ifp,
5192 xfs_bmbt_get_all(ep, &got);
5196 } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5198 * This one is already unwritten.
5199 * It must have a written left neighbor.
5200 * Unwrite the killed part of that one and
5204 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5206 ASSERT(prev.br_state == XFS_EXT_NORM);
5207 ASSERT(!isnullstartblock(prev.br_startblock));
5208 ASSERT(del.br_startblock ==
5209 prev.br_startblock + prev.br_blockcount);
5210 if (prev.br_startoff < start) {
5211 mod = start - prev.br_startoff;
5212 prev.br_blockcount -= mod;
5213 prev.br_startblock += mod;
5214 prev.br_startoff = start;
5216 prev.br_state = XFS_EXT_UNWRITTEN;
5218 error = xfs_bmap_add_extent_unwritten_real(tp,
5219 ip, &lastx, &cur, &prev,
5220 firstblock, flist, &logflags);
5225 ASSERT(del.br_state == XFS_EXT_NORM);
5226 del.br_state = XFS_EXT_UNWRITTEN;
5227 error = xfs_bmap_add_extent_unwritten_real(tp,
5228 ip, &lastx, &cur, &del,
5229 firstblock, flist, &logflags);
5236 ASSERT(startblockval(del.br_startblock) > 0);
5237 /* Update realtime/data freespace, unreserve quota */
5239 xfs_filblks_t rtexts;
5241 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5242 do_div(rtexts, mp->m_sb.sb_rextsize);
5243 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5244 (int64_t)rtexts, 0);
5245 (void)xfs_trans_reserve_quota_nblks(NULL,
5246 ip, -((long)del.br_blockcount), 0,
5247 XFS_QMOPT_RES_RTBLKS);
5249 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5250 (int64_t)del.br_blockcount, 0);
5251 (void)xfs_trans_reserve_quota_nblks(NULL,
5252 ip, -((long)del.br_blockcount), 0,
5253 XFS_QMOPT_RES_REGBLKS);
5255 ip->i_delayed_blks -= del.br_blockcount;
5257 cur->bc_private.b.flags |=
5258 XFS_BTCUR_BPRV_WASDEL;
5260 cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5262 * If it's the case where the directory code is running
5263 * with no block reservation, and the deleted block is in
5264 * the middle of its extent, and the resulting insert
5265 * of an extent would cause transformation to btree format,
5266 * then reject it. The calling code will then swap
5267 * blocks around instead.
5268 * We have to do this now, rather than waiting for the
5269 * conversion to btree format, since the transaction
5272 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5273 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5274 XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5275 XFS_IFORK_MAXEXT(ip, whichfork) &&
5276 del.br_startoff > got.br_startoff &&
5277 del.br_startoff + del.br_blockcount <
5278 got.br_startoff + got.br_blockcount) {
5279 error = XFS_ERROR(ENOSPC);
5282 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5283 &tmp_logflags, whichfork);
5284 logflags |= tmp_logflags;
5287 bno = del.br_startoff - 1;
5290 * If not done go on to the next (previous) record.
5292 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5294 ep = xfs_iext_get_ext(ifp, lastx);
5295 if (xfs_bmbt_get_startoff(ep) > bno) {
5297 ep = xfs_iext_get_ext(ifp,
5300 xfs_bmbt_get_all(ep, &got);
5305 *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5308 * Convert to a btree if necessary.
5310 if (xfs_bmap_needs_btree(ip, whichfork)) {
5311 ASSERT(cur == NULL);
5312 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5313 &cur, 0, &tmp_logflags, whichfork);
5314 logflags |= tmp_logflags;
5319 * transform from btree to extents, give it cur
5321 else if (xfs_bmap_wants_extents(ip, whichfork)) {
5322 ASSERT(cur != NULL);
5323 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5325 logflags |= tmp_logflags;
5330 * transform from extents to local?
5335 * Log everything. Do this after conversion, there's no point in
5336 * logging the extent records if we've converted to btree format.
5338 if ((logflags & xfs_ilog_fext(whichfork)) &&
5339 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5340 logflags &= ~xfs_ilog_fext(whichfork);
5341 else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5342 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5343 logflags &= ~xfs_ilog_fbroot(whichfork);
5345 * Log inode even in the error case, if the transaction
5346 * is dirty we'll need to shut down the filesystem.
5349 xfs_trans_log_inode(tp, ip, logflags);
5352 *firstblock = cur->bc_private.b.firstblock;
5353 cur->bc_private.b.allocated = 0;
5355 xfs_btree_del_cursor(cur,
5356 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);