2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
28 #include "xfs_mount.h"
29 #include "xfs_da_format.h"
30 #include "xfs_da_btree.h"
32 #include "xfs_inode.h"
33 #include "xfs_btree.h"
34 #include "xfs_trans.h"
35 #include "xfs_inode_item.h"
36 #include "xfs_extfree_item.h"
37 #include "xfs_alloc.h"
39 #include "xfs_bmap_util.h"
40 #include "xfs_bmap_btree.h"
41 #include "xfs_rtalloc.h"
42 #include "xfs_error.h"
43 #include "xfs_quota.h"
44 #include "xfs_trans_space.h"
45 #include "xfs_buf_item.h"
46 #include "xfs_trace.h"
47 #include "xfs_symlink.h"
48 #include "xfs_attr_leaf.h"
49 #include "xfs_dinode.h"
50 #include "xfs_filestream.h"
53 kmem_zone_t *xfs_bmap_free_item_zone;
56 * Miscellaneous helper functions
60 * Compute and fill in the value of the maximum depth of a bmap btree
61 * in this filesystem. Done once, during mount.
64 xfs_bmap_compute_maxlevels(
65 xfs_mount_t *mp, /* file system mount structure */
66 int whichfork) /* data or attr fork */
68 int level; /* btree level */
69 uint maxblocks; /* max blocks at this level */
70 uint maxleafents; /* max leaf entries possible */
71 int maxrootrecs; /* max records in root block */
72 int minleafrecs; /* min records in leaf block */
73 int minnoderecs; /* min records in node block */
74 int sz; /* root block size */
77 * The maximum number of extents in a file, hence the maximum
78 * number of leaf entries, is controlled by the type of di_nextents
79 * (a signed 32-bit number, xfs_extnum_t), or by di_anextents
80 * (a signed 16-bit number, xfs_aextnum_t).
82 * Note that we can no longer assume that if we are in ATTR1 that
83 * the fork offset of all the inodes will be
84 * (xfs_default_attroffset(ip) >> 3) because we could have mounted
85 * with ATTR2 and then mounted back with ATTR1, keeping the
86 * di_forkoff's fixed but probably at various positions. Therefore,
87 * for both ATTR1 and ATTR2 we have to assume the worst case scenario
88 * of a minimum size available.
90 if (whichfork == XFS_DATA_FORK) {
91 maxleafents = MAXEXTNUM;
92 sz = XFS_BMDR_SPACE_CALC(MINDBTPTRS);
94 maxleafents = MAXAEXTNUM;
95 sz = XFS_BMDR_SPACE_CALC(MINABTPTRS);
97 maxrootrecs = xfs_bmdr_maxrecs(mp, sz, 0);
98 minleafrecs = mp->m_bmap_dmnr[0];
99 minnoderecs = mp->m_bmap_dmnr[1];
100 maxblocks = (maxleafents + minleafrecs - 1) / minleafrecs;
101 for (level = 1; maxblocks > 1; level++) {
102 if (maxblocks <= maxrootrecs)
105 maxblocks = (maxblocks + minnoderecs - 1) / minnoderecs;
107 mp->m_bm_maxlevels[whichfork] = level;
110 STATIC int /* error */
112 struct xfs_btree_cur *cur,
116 int *stat) /* success/failure */
118 cur->bc_rec.b.br_startoff = off;
119 cur->bc_rec.b.br_startblock = bno;
120 cur->bc_rec.b.br_blockcount = len;
121 return xfs_btree_lookup(cur, XFS_LOOKUP_EQ, stat);
124 STATIC int /* error */
126 struct xfs_btree_cur *cur,
130 int *stat) /* success/failure */
132 cur->bc_rec.b.br_startoff = off;
133 cur->bc_rec.b.br_startblock = bno;
134 cur->bc_rec.b.br_blockcount = len;
135 return xfs_btree_lookup(cur, XFS_LOOKUP_GE, stat);
139 * Check if the inode needs to be converted to btree format.
141 static inline bool xfs_bmap_needs_btree(struct xfs_inode *ip, int whichfork)
143 return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
144 XFS_IFORK_NEXTENTS(ip, whichfork) >
145 XFS_IFORK_MAXEXT(ip, whichfork);
149 * Check if the inode should be converted to extent format.
151 static inline bool xfs_bmap_wants_extents(struct xfs_inode *ip, int whichfork)
153 return XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE &&
154 XFS_IFORK_NEXTENTS(ip, whichfork) <=
155 XFS_IFORK_MAXEXT(ip, whichfork);
159 * Update the record referred to by cur to the value given
160 * by [off, bno, len, state].
161 * This either works (return 0) or gets an EFSCORRUPTED error.
165 struct xfs_btree_cur *cur,
171 union xfs_btree_rec rec;
173 xfs_bmbt_disk_set_allf(&rec.bmbt, off, bno, len, state);
174 return xfs_btree_update(cur, &rec);
178 * Compute the worst-case number of indirect blocks that will be used
179 * for ip's delayed extent of length "len".
182 xfs_bmap_worst_indlen(
183 xfs_inode_t *ip, /* incore inode pointer */
184 xfs_filblks_t len) /* delayed extent length */
186 int level; /* btree level number */
187 int maxrecs; /* maximum record count at this level */
188 xfs_mount_t *mp; /* mount structure */
189 xfs_filblks_t rval; /* return value */
192 maxrecs = mp->m_bmap_dmxr[0];
193 for (level = 0, rval = 0;
194 level < XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK);
197 do_div(len, maxrecs);
200 return rval + XFS_BM_MAXLEVELS(mp, XFS_DATA_FORK) -
203 maxrecs = mp->m_bmap_dmxr[1];
209 * Calculate the default attribute fork offset for newly created inodes.
212 xfs_default_attroffset(
213 struct xfs_inode *ip)
215 struct xfs_mount *mp = ip->i_mount;
218 if (mp->m_sb.sb_inodesize == 256) {
219 offset = XFS_LITINO(mp, ip->i_d.di_version) -
220 XFS_BMDR_SPACE_CALC(MINABTPTRS);
222 offset = XFS_BMDR_SPACE_CALC(6 * MINABTPTRS);
225 ASSERT(offset < XFS_LITINO(mp, ip->i_d.di_version));
230 * Helper routine to reset inode di_forkoff field when switching
231 * attribute fork from local to extent format - we reset it where
232 * possible to make space available for inline data fork extents.
235 xfs_bmap_forkoff_reset(
240 if (whichfork == XFS_ATTR_FORK &&
241 ip->i_d.di_format != XFS_DINODE_FMT_DEV &&
242 ip->i_d.di_format != XFS_DINODE_FMT_UUID &&
243 ip->i_d.di_format != XFS_DINODE_FMT_BTREE) {
244 uint dfl_forkoff = xfs_default_attroffset(ip) >> 3;
246 if (dfl_forkoff > ip->i_d.di_forkoff)
247 ip->i_d.di_forkoff = dfl_forkoff;
252 * Debug/sanity checking code
256 xfs_bmap_sanity_check(
257 struct xfs_mount *mp,
261 struct xfs_btree_block *block = XFS_BUF_TO_BLOCK(bp);
263 if (block->bb_magic != cpu_to_be32(XFS_BMAP_CRC_MAGIC) &&
264 block->bb_magic != cpu_to_be32(XFS_BMAP_MAGIC))
267 if (be16_to_cpu(block->bb_level) != level ||
268 be16_to_cpu(block->bb_numrecs) == 0 ||
269 be16_to_cpu(block->bb_numrecs) > mp->m_bmap_dmxr[level != 0])
276 STATIC struct xfs_buf *
278 struct xfs_btree_cur *cur,
281 struct xfs_log_item_desc *lidp;
287 for (i = 0; i < XFS_BTREE_MAXLEVELS; i++) {
288 if (!cur->bc_bufs[i])
290 if (XFS_BUF_ADDR(cur->bc_bufs[i]) == bno)
291 return cur->bc_bufs[i];
294 /* Chase down all the log items to see if the bp is there */
295 list_for_each_entry(lidp, &cur->bc_tp->t_items, lid_trans) {
296 struct xfs_buf_log_item *bip;
297 bip = (struct xfs_buf_log_item *)lidp->lid_item;
298 if (bip->bli_item.li_type == XFS_LI_BUF &&
299 XFS_BUF_ADDR(bip->bli_buf) == bno)
308 struct xfs_btree_block *block,
314 __be64 *pp, *thispa; /* pointer to block address */
315 xfs_bmbt_key_t *prevp, *keyp;
317 ASSERT(be16_to_cpu(block->bb_level) > 0);
320 for( i = 1; i <= xfs_btree_get_numrecs(block); i++) {
321 dmxr = mp->m_bmap_dmxr[0];
322 keyp = XFS_BMBT_KEY_ADDR(mp, block, i);
325 ASSERT(be64_to_cpu(prevp->br_startoff) <
326 be64_to_cpu(keyp->br_startoff));
331 * Compare the block numbers to see if there are dups.
334 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, i, sz);
336 pp = XFS_BMBT_PTR_ADDR(mp, block, i, dmxr);
338 for (j = i+1; j <= be16_to_cpu(block->bb_numrecs); j++) {
340 thispa = XFS_BMAP_BROOT_PTR_ADDR(mp, block, j, sz);
342 thispa = XFS_BMBT_PTR_ADDR(mp, block, j, dmxr);
343 if (*thispa == *pp) {
344 xfs_warn(mp, "%s: thispa(%d) == pp(%d) %Ld",
346 (unsigned long long)be64_to_cpu(*thispa));
347 panic("%s: ptrs are equal in node\n",
355 * Check that the extents for the inode ip are in the right order in all
360 xfs_bmap_check_leaf_extents(
361 xfs_btree_cur_t *cur, /* btree cursor or null */
362 xfs_inode_t *ip, /* incore inode pointer */
363 int whichfork) /* data or attr fork */
365 struct xfs_btree_block *block; /* current btree block */
366 xfs_fsblock_t bno; /* block # of "block" */
367 xfs_buf_t *bp; /* buffer for "block" */
368 int error; /* error return value */
369 xfs_extnum_t i=0, j; /* index into the extents list */
370 xfs_ifork_t *ifp; /* fork structure */
371 int level; /* btree level, for checking */
372 xfs_mount_t *mp; /* file system mount structure */
373 __be64 *pp; /* pointer to block address */
374 xfs_bmbt_rec_t *ep; /* pointer to current extent */
375 xfs_bmbt_rec_t last = {0, 0}; /* last extent in prev block */
376 xfs_bmbt_rec_t *nextp; /* pointer to next extent */
379 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE) {
385 ifp = XFS_IFORK_PTR(ip, whichfork);
386 block = ifp->if_broot;
388 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
390 level = be16_to_cpu(block->bb_level);
392 xfs_check_block(block, mp, 1, ifp->if_broot_bytes);
393 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
394 bno = be64_to_cpu(*pp);
396 ASSERT(bno != NULLDFSBNO);
397 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
398 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
401 * Go down the tree until leaf level is reached, following the first
402 * pointer (leftmost) at each level.
404 while (level-- > 0) {
405 /* See if buf is in cur first */
407 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
410 error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
416 block = XFS_BUF_TO_BLOCK(bp);
417 XFS_WANT_CORRUPTED_GOTO(
418 xfs_bmap_sanity_check(mp, bp, level),
424 * Check this block for basic sanity (increasing keys and
425 * no duplicate blocks).
428 xfs_check_block(block, mp, 0, 0);
429 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
430 bno = be64_to_cpu(*pp);
431 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
434 xfs_trans_brelse(NULL, bp);
439 * Here with bp and block set to the leftmost leaf node in the tree.
444 * Loop over all leaf nodes checking that all extents are in the right order.
447 xfs_fsblock_t nextbno;
448 xfs_extnum_t num_recs;
451 num_recs = xfs_btree_get_numrecs(block);
454 * Read-ahead the next leaf block, if any.
457 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
460 * Check all the extents to make sure they are OK.
461 * If we had a previous block, the last entry should
462 * conform with the first entry in this one.
465 ep = XFS_BMBT_REC_ADDR(mp, block, 1);
467 ASSERT(xfs_bmbt_disk_get_startoff(&last) +
468 xfs_bmbt_disk_get_blockcount(&last) <=
469 xfs_bmbt_disk_get_startoff(ep));
471 for (j = 1; j < num_recs; j++) {
472 nextp = XFS_BMBT_REC_ADDR(mp, block, j + 1);
473 ASSERT(xfs_bmbt_disk_get_startoff(ep) +
474 xfs_bmbt_disk_get_blockcount(ep) <=
475 xfs_bmbt_disk_get_startoff(nextp));
483 xfs_trans_brelse(NULL, bp);
487 * If we've reached the end, stop.
489 if (bno == NULLFSBLOCK)
493 bp = xfs_bmap_get_bp(cur, XFS_FSB_TO_DADDR(mp, bno));
496 error = xfs_btree_read_bufl(mp, NULL, bno, 0, &bp,
502 block = XFS_BUF_TO_BLOCK(bp);
506 xfs_trans_brelse(NULL, bp);
511 xfs_warn(mp, "%s: at error0", __func__);
513 xfs_trans_brelse(NULL, bp);
515 xfs_warn(mp, "%s: BAD after btree leaves for %d extents",
517 panic("%s: CORRUPTED BTREE OR SOMETHING", __func__);
522 * Add bmap trace insert entries for all the contents of the extent records.
525 xfs_bmap_trace_exlist(
526 xfs_inode_t *ip, /* incore inode pointer */
527 xfs_extnum_t cnt, /* count of entries in the list */
528 int whichfork, /* data or attr fork */
529 unsigned long caller_ip)
531 xfs_extnum_t idx; /* extent record index */
532 xfs_ifork_t *ifp; /* inode fork pointer */
535 if (whichfork == XFS_ATTR_FORK)
536 state |= BMAP_ATTRFORK;
538 ifp = XFS_IFORK_PTR(ip, whichfork);
539 ASSERT(cnt == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
540 for (idx = 0; idx < cnt; idx++)
541 trace_xfs_extlist(ip, idx, whichfork, caller_ip);
545 * Validate that the bmbt_irecs being returned from bmapi are valid
546 * given the caller's original parameters. Specifically check the
547 * ranges of the returned irecs to ensure that they only extend beyond
548 * the given parameters if the XFS_BMAPI_ENTIRE flag was set.
551 xfs_bmap_validate_ret(
555 xfs_bmbt_irec_t *mval,
559 int i; /* index to map values */
561 ASSERT(ret_nmap <= nmap);
563 for (i = 0; i < ret_nmap; i++) {
564 ASSERT(mval[i].br_blockcount > 0);
565 if (!(flags & XFS_BMAPI_ENTIRE)) {
566 ASSERT(mval[i].br_startoff >= bno);
567 ASSERT(mval[i].br_blockcount <= len);
568 ASSERT(mval[i].br_startoff + mval[i].br_blockcount <=
571 ASSERT(mval[i].br_startoff < bno + len);
572 ASSERT(mval[i].br_startoff + mval[i].br_blockcount >
576 mval[i - 1].br_startoff + mval[i - 1].br_blockcount ==
577 mval[i].br_startoff);
578 ASSERT(mval[i].br_startblock != DELAYSTARTBLOCK &&
579 mval[i].br_startblock != HOLESTARTBLOCK);
580 ASSERT(mval[i].br_state == XFS_EXT_NORM ||
581 mval[i].br_state == XFS_EXT_UNWRITTEN);
586 #define xfs_bmap_check_leaf_extents(cur, ip, whichfork) do { } while (0)
587 #define xfs_bmap_validate_ret(bno,len,flags,mval,onmap,nmap)
591 * bmap free list manipulation functions
595 * Add the extent to the list of extents to be free at transaction end.
596 * The list is maintained sorted (by block number).
600 xfs_fsblock_t bno, /* fs block number of extent */
601 xfs_filblks_t len, /* length of extent */
602 xfs_bmap_free_t *flist, /* list of extents */
603 xfs_mount_t *mp) /* mount point structure */
605 xfs_bmap_free_item_t *cur; /* current (next) element */
606 xfs_bmap_free_item_t *new; /* new element */
607 xfs_bmap_free_item_t *prev; /* previous element */
612 ASSERT(bno != NULLFSBLOCK);
614 ASSERT(len <= MAXEXTLEN);
615 ASSERT(!isnullstartblock(bno));
616 agno = XFS_FSB_TO_AGNO(mp, bno);
617 agbno = XFS_FSB_TO_AGBNO(mp, bno);
618 ASSERT(agno < mp->m_sb.sb_agcount);
619 ASSERT(agbno < mp->m_sb.sb_agblocks);
620 ASSERT(len < mp->m_sb.sb_agblocks);
621 ASSERT(agbno + len <= mp->m_sb.sb_agblocks);
623 ASSERT(xfs_bmap_free_item_zone != NULL);
624 new = kmem_zone_alloc(xfs_bmap_free_item_zone, KM_SLEEP);
625 new->xbfi_startblock = bno;
626 new->xbfi_blockcount = (xfs_extlen_t)len;
627 for (prev = NULL, cur = flist->xbf_first;
629 prev = cur, cur = cur->xbfi_next) {
630 if (cur->xbfi_startblock >= bno)
634 prev->xbfi_next = new;
636 flist->xbf_first = new;
637 new->xbfi_next = cur;
642 * Remove the entry "free" from the free item list. Prev points to the
643 * previous entry, unless "free" is the head of the list.
647 xfs_bmap_free_t *flist, /* free item list header */
648 xfs_bmap_free_item_t *prev, /* previous item on list, if any */
649 xfs_bmap_free_item_t *free) /* list item to be freed */
652 prev->xbfi_next = free->xbfi_next;
654 flist->xbf_first = free->xbfi_next;
656 kmem_zone_free(xfs_bmap_free_item_zone, free);
660 * Free up any items left in the list.
664 xfs_bmap_free_t *flist) /* list of bmap_free_items */
666 xfs_bmap_free_item_t *free; /* free list item */
667 xfs_bmap_free_item_t *next;
669 if (flist->xbf_count == 0)
671 ASSERT(flist->xbf_first != NULL);
672 for (free = flist->xbf_first; free; free = next) {
673 next = free->xbfi_next;
674 xfs_bmap_del_free(flist, NULL, free);
676 ASSERT(flist->xbf_count == 0);
680 * Inode fork format manipulation functions
684 * Transform a btree format file with only one leaf node, where the
685 * extents list will fit in the inode, into an extents format file.
686 * Since the file extents are already in-core, all we have to do is
687 * give up the space for the btree root and pitch the leaf block.
689 STATIC int /* error */
690 xfs_bmap_btree_to_extents(
691 xfs_trans_t *tp, /* transaction pointer */
692 xfs_inode_t *ip, /* incore inode pointer */
693 xfs_btree_cur_t *cur, /* btree cursor */
694 int *logflagsp, /* inode logging flags */
695 int whichfork) /* data or attr fork */
698 struct xfs_btree_block *cblock;/* child btree block */
699 xfs_fsblock_t cbno; /* child block number */
700 xfs_buf_t *cbp; /* child block's buffer */
701 int error; /* error return value */
702 xfs_ifork_t *ifp; /* inode fork data */
703 xfs_mount_t *mp; /* mount point structure */
704 __be64 *pp; /* ptr to block address */
705 struct xfs_btree_block *rblock;/* root btree block */
708 ifp = XFS_IFORK_PTR(ip, whichfork);
709 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
710 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
711 rblock = ifp->if_broot;
712 ASSERT(be16_to_cpu(rblock->bb_level) == 1);
713 ASSERT(be16_to_cpu(rblock->bb_numrecs) == 1);
714 ASSERT(xfs_bmbt_maxrecs(mp, ifp->if_broot_bytes, 0) == 1);
715 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, rblock, 1, ifp->if_broot_bytes);
716 cbno = be64_to_cpu(*pp);
719 if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
722 error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
726 cblock = XFS_BUF_TO_BLOCK(cbp);
727 if ((error = xfs_btree_check_block(cur, cblock, 0, cbp)))
729 xfs_bmap_add_free(cbno, 1, cur->bc_private.b.flist, mp);
730 ip->i_d.di_nblocks--;
731 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, -1L);
732 xfs_trans_binval(tp, cbp);
733 if (cur->bc_bufs[0] == cbp)
734 cur->bc_bufs[0] = NULL;
735 xfs_iroot_realloc(ip, -1, whichfork);
736 ASSERT(ifp->if_broot == NULL);
737 ASSERT((ifp->if_flags & XFS_IFBROOT) == 0);
738 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
739 *logflagsp = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
744 * Convert an extents-format file into a btree-format file.
745 * The new file will have a root block (in the inode) and a single child block.
747 STATIC int /* error */
748 xfs_bmap_extents_to_btree(
749 xfs_trans_t *tp, /* transaction pointer */
750 xfs_inode_t *ip, /* incore inode pointer */
751 xfs_fsblock_t *firstblock, /* first-block-allocated */
752 xfs_bmap_free_t *flist, /* blocks freed in xaction */
753 xfs_btree_cur_t **curp, /* cursor returned to caller */
754 int wasdel, /* converting a delayed alloc */
755 int *logflagsp, /* inode logging flags */
756 int whichfork) /* data or attr fork */
758 struct xfs_btree_block *ablock; /* allocated (child) bt block */
759 xfs_buf_t *abp; /* buffer for ablock */
760 xfs_alloc_arg_t args; /* allocation arguments */
761 xfs_bmbt_rec_t *arp; /* child record pointer */
762 struct xfs_btree_block *block; /* btree root block */
763 xfs_btree_cur_t *cur; /* bmap btree cursor */
764 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
765 int error; /* error return value */
766 xfs_extnum_t i, cnt; /* extent record index */
767 xfs_ifork_t *ifp; /* inode fork pointer */
768 xfs_bmbt_key_t *kp; /* root block key pointer */
769 xfs_mount_t *mp; /* mount structure */
770 xfs_extnum_t nextents; /* number of file extents */
771 xfs_bmbt_ptr_t *pp; /* root block address pointer */
774 ifp = XFS_IFORK_PTR(ip, whichfork);
775 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS);
778 * Make space in the inode incore.
780 xfs_iroot_realloc(ip, 1, whichfork);
781 ifp->if_flags |= XFS_IFBROOT;
786 block = ifp->if_broot;
787 if (xfs_sb_version_hascrc(&mp->m_sb))
788 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
789 XFS_BMAP_CRC_MAGIC, 1, 1, ip->i_ino,
790 XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
792 xfs_btree_init_block_int(mp, block, XFS_BUF_DADDR_NULL,
793 XFS_BMAP_MAGIC, 1, 1, ip->i_ino,
794 XFS_BTREE_LONG_PTRS);
797 * Need a cursor. Can't allocate until bb_level is filled in.
799 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
800 cur->bc_private.b.firstblock = *firstblock;
801 cur->bc_private.b.flist = flist;
802 cur->bc_private.b.flags = wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
804 * Convert to a btree with two levels, one record in root.
806 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_BTREE);
807 memset(&args, 0, sizeof(args));
810 args.firstblock = *firstblock;
811 if (*firstblock == NULLFSBLOCK) {
812 args.type = XFS_ALLOCTYPE_START_BNO;
813 args.fsbno = XFS_INO_TO_FSB(mp, ip->i_ino);
814 } else if (flist->xbf_low) {
815 args.type = XFS_ALLOCTYPE_START_BNO;
816 args.fsbno = *firstblock;
818 args.type = XFS_ALLOCTYPE_NEAR_BNO;
819 args.fsbno = *firstblock;
821 args.minlen = args.maxlen = args.prod = 1;
822 args.wasdel = wasdel;
824 if ((error = xfs_alloc_vextent(&args))) {
825 xfs_iroot_realloc(ip, -1, whichfork);
826 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
830 * Allocation can't fail, the space was reserved.
832 ASSERT(args.fsbno != NULLFSBLOCK);
833 ASSERT(*firstblock == NULLFSBLOCK ||
834 args.agno == XFS_FSB_TO_AGNO(mp, *firstblock) ||
836 args.agno > XFS_FSB_TO_AGNO(mp, *firstblock)));
837 *firstblock = cur->bc_private.b.firstblock = args.fsbno;
838 cur->bc_private.b.allocated++;
839 ip->i_d.di_nblocks++;
840 xfs_trans_mod_dquot_byino(tp, ip, XFS_TRANS_DQ_BCOUNT, 1L);
841 abp = xfs_btree_get_bufl(mp, tp, args.fsbno, 0);
843 * Fill in the child block.
845 abp->b_ops = &xfs_bmbt_buf_ops;
846 ablock = XFS_BUF_TO_BLOCK(abp);
847 if (xfs_sb_version_hascrc(&mp->m_sb))
848 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
849 XFS_BMAP_CRC_MAGIC, 0, 0, ip->i_ino,
850 XFS_BTREE_LONG_PTRS | XFS_BTREE_CRC_BLOCKS);
852 xfs_btree_init_block_int(mp, ablock, abp->b_bn,
853 XFS_BMAP_MAGIC, 0, 0, ip->i_ino,
854 XFS_BTREE_LONG_PTRS);
856 arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
857 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
858 for (cnt = i = 0; i < nextents; i++) {
859 ep = xfs_iext_get_ext(ifp, i);
860 if (!isnullstartblock(xfs_bmbt_get_startblock(ep))) {
861 arp->l0 = cpu_to_be64(ep->l0);
862 arp->l1 = cpu_to_be64(ep->l1);
866 ASSERT(cnt == XFS_IFORK_NEXTENTS(ip, whichfork));
867 xfs_btree_set_numrecs(ablock, cnt);
870 * Fill in the root key and pointer.
872 kp = XFS_BMBT_KEY_ADDR(mp, block, 1);
873 arp = XFS_BMBT_REC_ADDR(mp, ablock, 1);
874 kp->br_startoff = cpu_to_be64(xfs_bmbt_disk_get_startoff(arp));
875 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, xfs_bmbt_get_maxrecs(cur,
876 be16_to_cpu(block->bb_level)));
877 *pp = cpu_to_be64(args.fsbno);
880 * Do all this logging at the end so that
881 * the root is at the right level.
883 xfs_btree_log_block(cur, abp, XFS_BB_ALL_BITS);
884 xfs_btree_log_recs(cur, abp, 1, be16_to_cpu(ablock->bb_numrecs));
885 ASSERT(*curp == NULL);
887 *logflagsp = XFS_ILOG_CORE | xfs_ilog_fbroot(whichfork);
892 * Convert a local file to an extents file.
893 * This code is out of bounds for data forks of regular files,
894 * since the file data needs to get logged so things will stay consistent.
895 * (The bmap-level manipulations are ok, though).
898 xfs_bmap_local_to_extents_empty(
899 struct xfs_inode *ip,
902 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
904 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
905 ASSERT(ifp->if_bytes == 0);
906 ASSERT(XFS_IFORK_NEXTENTS(ip, whichfork) == 0);
908 xfs_bmap_forkoff_reset(ip->i_mount, ip, whichfork);
909 ifp->if_flags &= ~XFS_IFINLINE;
910 ifp->if_flags |= XFS_IFEXTENTS;
911 XFS_IFORK_FMT_SET(ip, whichfork, XFS_DINODE_FMT_EXTENTS);
915 STATIC int /* error */
916 xfs_bmap_local_to_extents(
917 xfs_trans_t *tp, /* transaction pointer */
918 xfs_inode_t *ip, /* incore inode pointer */
919 xfs_fsblock_t *firstblock, /* first block allocated in xaction */
920 xfs_extlen_t total, /* total blocks needed by transaction */
921 int *logflagsp, /* inode logging flags */
923 void (*init_fn)(struct xfs_trans *tp,
925 struct xfs_inode *ip,
926 struct xfs_ifork *ifp))
929 int flags; /* logging flags returned */
930 xfs_ifork_t *ifp; /* inode fork pointer */
931 xfs_alloc_arg_t args; /* allocation arguments */
932 xfs_buf_t *bp; /* buffer for extent block */
933 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
936 * We don't want to deal with the case of keeping inode data inline yet.
937 * So sending the data fork of a regular inode is invalid.
939 ASSERT(!(S_ISREG(ip->i_d.di_mode) && whichfork == XFS_DATA_FORK));
940 ifp = XFS_IFORK_PTR(ip, whichfork);
941 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
943 if (!ifp->if_bytes) {
944 xfs_bmap_local_to_extents_empty(ip, whichfork);
945 flags = XFS_ILOG_CORE;
951 ASSERT((ifp->if_flags & (XFS_IFINLINE|XFS_IFEXTENTS|XFS_IFEXTIREC)) ==
953 memset(&args, 0, sizeof(args));
955 args.mp = ip->i_mount;
956 args.firstblock = *firstblock;
958 * Allocate a block. We know we need only one, since the
959 * file currently fits in an inode.
961 if (*firstblock == NULLFSBLOCK) {
962 args.fsbno = XFS_INO_TO_FSB(args.mp, ip->i_ino);
963 args.type = XFS_ALLOCTYPE_START_BNO;
965 args.fsbno = *firstblock;
966 args.type = XFS_ALLOCTYPE_NEAR_BNO;
969 args.minlen = args.maxlen = args.prod = 1;
970 error = xfs_alloc_vextent(&args);
974 /* Can't fail, the space was reserved. */
975 ASSERT(args.fsbno != NULLFSBLOCK);
976 ASSERT(args.len == 1);
977 *firstblock = args.fsbno;
978 bp = xfs_btree_get_bufl(args.mp, tp, args.fsbno, 0);
980 /* initialise the block and copy the data */
981 init_fn(tp, bp, ip, ifp);
983 /* account for the change in fork size and log everything */
984 xfs_trans_log_buf(tp, bp, 0, ifp->if_bytes - 1);
985 xfs_idata_realloc(ip, -ifp->if_bytes, whichfork);
986 xfs_bmap_local_to_extents_empty(ip, whichfork);
987 flags |= XFS_ILOG_CORE;
989 xfs_iext_add(ifp, 0, 1);
990 ep = xfs_iext_get_ext(ifp, 0);
991 xfs_bmbt_set_allf(ep, 0, args.fsbno, 1, XFS_EXT_NORM);
992 trace_xfs_bmap_post_update(ip, 0,
993 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0,
995 XFS_IFORK_NEXT_SET(ip, whichfork, 1);
996 ip->i_d.di_nblocks = 1;
997 xfs_trans_mod_dquot_byino(tp, ip,
998 XFS_TRANS_DQ_BCOUNT, 1L);
999 flags |= xfs_ilog_fext(whichfork);
1007 * Called from xfs_bmap_add_attrfork to handle btree format files.
1009 STATIC int /* error */
1010 xfs_bmap_add_attrfork_btree(
1011 xfs_trans_t *tp, /* transaction pointer */
1012 xfs_inode_t *ip, /* incore inode pointer */
1013 xfs_fsblock_t *firstblock, /* first block allocated */
1014 xfs_bmap_free_t *flist, /* blocks to free at commit */
1015 int *flags) /* inode logging flags */
1017 xfs_btree_cur_t *cur; /* btree cursor */
1018 int error; /* error return value */
1019 xfs_mount_t *mp; /* file system mount struct */
1020 int stat; /* newroot status */
1023 if (ip->i_df.if_broot_bytes <= XFS_IFORK_DSIZE(ip))
1024 *flags |= XFS_ILOG_DBROOT;
1026 cur = xfs_bmbt_init_cursor(mp, tp, ip, XFS_DATA_FORK);
1027 cur->bc_private.b.flist = flist;
1028 cur->bc_private.b.firstblock = *firstblock;
1029 if ((error = xfs_bmbt_lookup_ge(cur, 0, 0, 0, &stat)))
1031 /* must be at least one entry */
1032 XFS_WANT_CORRUPTED_GOTO(stat == 1, error0);
1033 if ((error = xfs_btree_new_iroot(cur, flags, &stat)))
1036 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1037 return XFS_ERROR(ENOSPC);
1039 *firstblock = cur->bc_private.b.firstblock;
1040 cur->bc_private.b.allocated = 0;
1041 xfs_btree_del_cursor(cur, XFS_BTREE_NOERROR);
1045 xfs_btree_del_cursor(cur, XFS_BTREE_ERROR);
1050 * Called from xfs_bmap_add_attrfork to handle extents format files.
1052 STATIC int /* error */
1053 xfs_bmap_add_attrfork_extents(
1054 xfs_trans_t *tp, /* transaction pointer */
1055 xfs_inode_t *ip, /* incore inode pointer */
1056 xfs_fsblock_t *firstblock, /* first block allocated */
1057 xfs_bmap_free_t *flist, /* blocks to free at commit */
1058 int *flags) /* inode logging flags */
1060 xfs_btree_cur_t *cur; /* bmap btree cursor */
1061 int error; /* error return value */
1063 if (ip->i_d.di_nextents * sizeof(xfs_bmbt_rec_t) <= XFS_IFORK_DSIZE(ip))
1066 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist, &cur, 0,
1067 flags, XFS_DATA_FORK);
1069 cur->bc_private.b.allocated = 0;
1070 xfs_btree_del_cursor(cur,
1071 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
1077 * Called from xfs_bmap_add_attrfork to handle local format files. Each
1078 * different data fork content type needs a different callout to do the
1079 * conversion. Some are basic and only require special block initialisation
1080 * callouts for the data formating, others (directories) are so specialised they
1081 * handle everything themselves.
1083 * XXX (dgc): investigate whether directory conversion can use the generic
1084 * formatting callout. It should be possible - it's just a very complex
1087 STATIC int /* error */
1088 xfs_bmap_add_attrfork_local(
1089 xfs_trans_t *tp, /* transaction pointer */
1090 xfs_inode_t *ip, /* incore inode pointer */
1091 xfs_fsblock_t *firstblock, /* first block allocated */
1092 xfs_bmap_free_t *flist, /* blocks to free at commit */
1093 int *flags) /* inode logging flags */
1095 xfs_da_args_t dargs; /* args for dir/attr code */
1097 if (ip->i_df.if_bytes <= XFS_IFORK_DSIZE(ip))
1100 if (S_ISDIR(ip->i_d.di_mode)) {
1101 memset(&dargs, 0, sizeof(dargs));
1103 dargs.firstblock = firstblock;
1104 dargs.flist = flist;
1105 dargs.total = ip->i_mount->m_dirblkfsbs;
1106 dargs.whichfork = XFS_DATA_FORK;
1108 return xfs_dir2_sf_to_block(&dargs);
1111 if (S_ISLNK(ip->i_d.di_mode))
1112 return xfs_bmap_local_to_extents(tp, ip, firstblock, 1,
1113 flags, XFS_DATA_FORK,
1114 xfs_symlink_local_to_remote);
1116 /* should only be called for types that support local format data */
1118 return EFSCORRUPTED;
1122 * Convert inode from non-attributed to attributed.
1123 * Must not be in a transaction, ip must not be locked.
1125 int /* error code */
1126 xfs_bmap_add_attrfork(
1127 xfs_inode_t *ip, /* incore inode pointer */
1128 int size, /* space new attribute needs */
1129 int rsvd) /* xact may use reserved blks */
1131 xfs_fsblock_t firstblock; /* 1st block/ag allocated */
1132 xfs_bmap_free_t flist; /* freed extent records */
1133 xfs_mount_t *mp; /* mount structure */
1134 xfs_trans_t *tp; /* transaction pointer */
1135 int blks; /* space reservation */
1136 int version = 1; /* superblock attr version */
1137 int committed; /* xaction was committed */
1138 int logflags; /* logging flags */
1139 int error; /* error return value */
1141 ASSERT(XFS_IFORK_Q(ip) == 0);
1144 ASSERT(!XFS_NOT_DQATTACHED(mp, ip));
1145 tp = xfs_trans_alloc(mp, XFS_TRANS_ADDAFORK);
1146 blks = XFS_ADDAFORK_SPACE_RES(mp);
1148 tp->t_flags |= XFS_TRANS_RESERVE;
1149 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_addafork, blks, 0);
1152 xfs_ilock(ip, XFS_ILOCK_EXCL);
1153 error = xfs_trans_reserve_quota_nblks(tp, ip, blks, 0, rsvd ?
1154 XFS_QMOPT_RES_REGBLKS | XFS_QMOPT_FORCE_RES :
1155 XFS_QMOPT_RES_REGBLKS);
1157 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1158 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES);
1161 if (XFS_IFORK_Q(ip))
1163 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS) {
1165 * For inodes coming from pre-6.2 filesystems.
1167 ASSERT(ip->i_d.di_aformat == 0);
1168 ip->i_d.di_aformat = XFS_DINODE_FMT_EXTENTS;
1170 ASSERT(ip->i_d.di_anextents == 0);
1172 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
1173 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1175 switch (ip->i_d.di_format) {
1176 case XFS_DINODE_FMT_DEV:
1177 ip->i_d.di_forkoff = roundup(sizeof(xfs_dev_t), 8) >> 3;
1179 case XFS_DINODE_FMT_UUID:
1180 ip->i_d.di_forkoff = roundup(sizeof(uuid_t), 8) >> 3;
1182 case XFS_DINODE_FMT_LOCAL:
1183 case XFS_DINODE_FMT_EXTENTS:
1184 case XFS_DINODE_FMT_BTREE:
1185 ip->i_d.di_forkoff = xfs_attr_shortform_bytesfit(ip, size);
1186 if (!ip->i_d.di_forkoff)
1187 ip->i_d.di_forkoff = xfs_default_attroffset(ip) >> 3;
1188 else if (mp->m_flags & XFS_MOUNT_ATTR2)
1193 error = XFS_ERROR(EINVAL);
1197 ASSERT(ip->i_afp == NULL);
1198 ip->i_afp = kmem_zone_zalloc(xfs_ifork_zone, KM_SLEEP);
1199 ip->i_afp->if_flags = XFS_IFEXTENTS;
1201 xfs_bmap_init(&flist, &firstblock);
1202 switch (ip->i_d.di_format) {
1203 case XFS_DINODE_FMT_LOCAL:
1204 error = xfs_bmap_add_attrfork_local(tp, ip, &firstblock, &flist,
1207 case XFS_DINODE_FMT_EXTENTS:
1208 error = xfs_bmap_add_attrfork_extents(tp, ip, &firstblock,
1211 case XFS_DINODE_FMT_BTREE:
1212 error = xfs_bmap_add_attrfork_btree(tp, ip, &firstblock, &flist,
1220 xfs_trans_log_inode(tp, ip, logflags);
1223 if (!xfs_sb_version_hasattr(&mp->m_sb) ||
1224 (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2)) {
1225 __int64_t sbfields = 0;
1227 spin_lock(&mp->m_sb_lock);
1228 if (!xfs_sb_version_hasattr(&mp->m_sb)) {
1229 xfs_sb_version_addattr(&mp->m_sb);
1230 sbfields |= XFS_SB_VERSIONNUM;
1232 if (!xfs_sb_version_hasattr2(&mp->m_sb) && version == 2) {
1233 xfs_sb_version_addattr2(&mp->m_sb);
1234 sbfields |= (XFS_SB_VERSIONNUM | XFS_SB_FEATURES2);
1237 spin_unlock(&mp->m_sb_lock);
1238 xfs_mod_sb(tp, sbfields);
1240 spin_unlock(&mp->m_sb_lock);
1243 error = xfs_bmap_finish(&tp, &flist, &committed);
1246 return xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1248 xfs_bmap_cancel(&flist);
1250 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1252 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
1257 * Internal and external extent tree search functions.
1261 * Read in the extents to if_extents.
1262 * All inode fields are set up by caller, we just traverse the btree
1263 * and copy the records in. If the file system cannot contain unwritten
1264 * extents, the records are checked for no "state" flags.
1267 xfs_bmap_read_extents(
1268 xfs_trans_t *tp, /* transaction pointer */
1269 xfs_inode_t *ip, /* incore inode */
1270 int whichfork) /* data or attr fork */
1272 struct xfs_btree_block *block; /* current btree block */
1273 xfs_fsblock_t bno; /* block # of "block" */
1274 xfs_buf_t *bp; /* buffer for "block" */
1275 int error; /* error return value */
1276 xfs_exntfmt_t exntf; /* XFS_EXTFMT_NOSTATE, if checking */
1277 xfs_extnum_t i, j; /* index into the extents list */
1278 xfs_ifork_t *ifp; /* fork structure */
1279 int level; /* btree level, for checking */
1280 xfs_mount_t *mp; /* file system mount structure */
1281 __be64 *pp; /* pointer to block address */
1283 xfs_extnum_t room; /* number of entries there's room for */
1287 ifp = XFS_IFORK_PTR(ip, whichfork);
1288 exntf = (whichfork != XFS_DATA_FORK) ? XFS_EXTFMT_NOSTATE :
1289 XFS_EXTFMT_INODE(ip);
1290 block = ifp->if_broot;
1292 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
1294 level = be16_to_cpu(block->bb_level);
1296 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
1297 bno = be64_to_cpu(*pp);
1298 ASSERT(bno != NULLDFSBNO);
1299 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
1300 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
1302 * Go down the tree until leaf level is reached, following the first
1303 * pointer (leftmost) at each level.
1305 while (level-- > 0) {
1306 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1307 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1310 block = XFS_BUF_TO_BLOCK(bp);
1311 XFS_WANT_CORRUPTED_GOTO(
1312 xfs_bmap_sanity_check(mp, bp, level),
1316 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
1317 bno = be64_to_cpu(*pp);
1318 XFS_WANT_CORRUPTED_GOTO(XFS_FSB_SANITY_CHECK(mp, bno), error0);
1319 xfs_trans_brelse(tp, bp);
1322 * Here with bp and block set to the leftmost leaf node in the tree.
1324 room = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1327 * Loop over all leaf nodes. Copy information to the extent records.
1330 xfs_bmbt_rec_t *frp;
1331 xfs_fsblock_t nextbno;
1332 xfs_extnum_t num_recs;
1335 num_recs = xfs_btree_get_numrecs(block);
1336 if (unlikely(i + num_recs > room)) {
1337 ASSERT(i + num_recs <= room);
1338 xfs_warn(ip->i_mount,
1339 "corrupt dinode %Lu, (btree extents).",
1340 (unsigned long long) ip->i_ino);
1341 XFS_CORRUPTION_ERROR("xfs_bmap_read_extents(1)",
1342 XFS_ERRLEVEL_LOW, ip->i_mount, block);
1345 XFS_WANT_CORRUPTED_GOTO(
1346 xfs_bmap_sanity_check(mp, bp, 0),
1349 * Read-ahead the next leaf block, if any.
1351 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
1352 if (nextbno != NULLFSBLOCK)
1353 xfs_btree_reada_bufl(mp, nextbno, 1,
1356 * Copy records into the extent records.
1358 frp = XFS_BMBT_REC_ADDR(mp, block, 1);
1360 for (j = 0; j < num_recs; j++, i++, frp++) {
1361 xfs_bmbt_rec_host_t *trp = xfs_iext_get_ext(ifp, i);
1362 trp->l0 = be64_to_cpu(frp->l0);
1363 trp->l1 = be64_to_cpu(frp->l1);
1365 if (exntf == XFS_EXTFMT_NOSTATE) {
1367 * Check all attribute bmap btree records and
1368 * any "older" data bmap btree records for a
1369 * set bit in the "extent flag" position.
1371 if (unlikely(xfs_check_nostate_extents(ifp,
1372 start, num_recs))) {
1373 XFS_ERROR_REPORT("xfs_bmap_read_extents(2)",
1379 xfs_trans_brelse(tp, bp);
1382 * If we've reached the end, stop.
1384 if (bno == NULLFSBLOCK)
1386 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
1387 XFS_BMAP_BTREE_REF, &xfs_bmbt_buf_ops);
1390 block = XFS_BUF_TO_BLOCK(bp);
1392 ASSERT(i == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)));
1393 ASSERT(i == XFS_IFORK_NEXTENTS(ip, whichfork));
1394 XFS_BMAP_TRACE_EXLIST(ip, i, whichfork);
1397 xfs_trans_brelse(tp, bp);
1398 return XFS_ERROR(EFSCORRUPTED);
1403 * Search the extent records for the entry containing block bno.
1404 * If bno lies in a hole, point to the next entry. If bno lies
1405 * past eof, *eofp will be set, and *prevp will contain the last
1406 * entry (null if none). Else, *lastxp will be set to the index
1407 * of the found entry; *gotp will contain the entry.
1409 STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
1410 xfs_bmap_search_multi_extents(
1411 xfs_ifork_t *ifp, /* inode fork pointer */
1412 xfs_fileoff_t bno, /* block number searched for */
1413 int *eofp, /* out: end of file found */
1414 xfs_extnum_t *lastxp, /* out: last extent index */
1415 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
1416 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
1418 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
1419 xfs_extnum_t lastx; /* last extent index */
1422 * Initialize the extent entry structure to catch access to
1423 * uninitialized br_startblock field.
1425 gotp->br_startoff = 0xffa5a5a5a5a5a5a5LL;
1426 gotp->br_blockcount = 0xa55a5a5a5a5a5a5aLL;
1427 gotp->br_state = XFS_EXT_INVALID;
1429 gotp->br_startblock = 0xffffa5a5a5a5a5a5LL;
1431 gotp->br_startblock = 0xffffa5a5;
1433 prevp->br_startoff = NULLFILEOFF;
1435 ep = xfs_iext_bno_to_ext(ifp, bno, &lastx);
1437 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx - 1), prevp);
1439 if (lastx < (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))) {
1440 xfs_bmbt_get_all(ep, gotp);
1454 * Search the extents list for the inode, for the extent containing bno.
1455 * If bno lies in a hole, point to the next entry. If bno lies past eof,
1456 * *eofp will be set, and *prevp will contain the last entry (null if none).
1457 * Else, *lastxp will be set to the index of the found
1458 * entry; *gotp will contain the entry.
1460 STATIC xfs_bmbt_rec_host_t * /* pointer to found extent entry */
1461 xfs_bmap_search_extents(
1462 xfs_inode_t *ip, /* incore inode pointer */
1463 xfs_fileoff_t bno, /* block number searched for */
1464 int fork, /* data or attr fork */
1465 int *eofp, /* out: end of file found */
1466 xfs_extnum_t *lastxp, /* out: last extent index */
1467 xfs_bmbt_irec_t *gotp, /* out: extent entry found */
1468 xfs_bmbt_irec_t *prevp) /* out: previous extent entry found */
1470 xfs_ifork_t *ifp; /* inode fork pointer */
1471 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
1473 XFS_STATS_INC(xs_look_exlist);
1474 ifp = XFS_IFORK_PTR(ip, fork);
1476 ep = xfs_bmap_search_multi_extents(ifp, bno, eofp, lastxp, gotp, prevp);
1478 if (unlikely(!(gotp->br_startblock) && (*lastxp != NULLEXTNUM) &&
1479 !(XFS_IS_REALTIME_INODE(ip) && fork == XFS_DATA_FORK))) {
1480 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
1481 "Access to block zero in inode %llu "
1482 "start_block: %llx start_off: %llx "
1483 "blkcnt: %llx extent-state: %x lastx: %x",
1484 (unsigned long long)ip->i_ino,
1485 (unsigned long long)gotp->br_startblock,
1486 (unsigned long long)gotp->br_startoff,
1487 (unsigned long long)gotp->br_blockcount,
1488 gotp->br_state, *lastxp);
1489 *lastxp = NULLEXTNUM;
1497 * Returns the file-relative block number of the first unused block(s)
1498 * in the file with at least "len" logically contiguous blocks free.
1499 * This is the lowest-address hole if the file has holes, else the first block
1500 * past the end of file.
1501 * Return 0 if the file is currently local (in-inode).
1504 xfs_bmap_first_unused(
1505 xfs_trans_t *tp, /* transaction pointer */
1506 xfs_inode_t *ip, /* incore inode */
1507 xfs_extlen_t len, /* size of hole to find */
1508 xfs_fileoff_t *first_unused, /* unused block */
1509 int whichfork) /* data or attr fork */
1511 int error; /* error return value */
1512 int idx; /* extent record index */
1513 xfs_ifork_t *ifp; /* inode fork pointer */
1514 xfs_fileoff_t lastaddr; /* last block number seen */
1515 xfs_fileoff_t lowest; /* lowest useful block */
1516 xfs_fileoff_t max; /* starting useful block */
1517 xfs_fileoff_t off; /* offset for this block */
1518 xfs_extnum_t nextents; /* number of extent entries */
1520 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE ||
1521 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ||
1522 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL);
1523 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1527 ifp = XFS_IFORK_PTR(ip, whichfork);
1528 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1529 (error = xfs_iread_extents(tp, ip, whichfork)))
1531 lowest = *first_unused;
1532 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
1533 for (idx = 0, lastaddr = 0, max = lowest; idx < nextents; idx++) {
1534 xfs_bmbt_rec_host_t *ep = xfs_iext_get_ext(ifp, idx);
1535 off = xfs_bmbt_get_startoff(ep);
1537 * See if the hole before this extent will work.
1539 if (off >= lowest + len && off - max >= len) {
1540 *first_unused = max;
1543 lastaddr = off + xfs_bmbt_get_blockcount(ep);
1544 max = XFS_FILEOFF_MAX(lastaddr, lowest);
1546 *first_unused = max;
1551 * Returns the file-relative block number of the last block - 1 before
1552 * last_block (input value) in the file.
1553 * This is not based on i_size, it is based on the extent records.
1554 * Returns 0 for local files, as they do not have extent records.
1557 xfs_bmap_last_before(
1558 xfs_trans_t *tp, /* transaction pointer */
1559 xfs_inode_t *ip, /* incore inode */
1560 xfs_fileoff_t *last_block, /* last block */
1561 int whichfork) /* data or attr fork */
1563 xfs_fileoff_t bno; /* input file offset */
1564 int eof; /* hit end of file */
1565 xfs_bmbt_rec_host_t *ep; /* pointer to last extent */
1566 int error; /* error return value */
1567 xfs_bmbt_irec_t got; /* current extent value */
1568 xfs_ifork_t *ifp; /* inode fork pointer */
1569 xfs_extnum_t lastx; /* last extent used */
1570 xfs_bmbt_irec_t prev; /* previous extent value */
1572 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1573 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
1574 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL)
1575 return XFS_ERROR(EIO);
1576 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL) {
1580 ifp = XFS_IFORK_PTR(ip, whichfork);
1581 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
1582 (error = xfs_iread_extents(tp, ip, whichfork)))
1584 bno = *last_block - 1;
1585 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
1587 if (eof || xfs_bmbt_get_startoff(ep) > bno) {
1588 if (prev.br_startoff == NULLFILEOFF)
1591 *last_block = prev.br_startoff + prev.br_blockcount;
1594 * Otherwise *last_block is already the right answer.
1600 xfs_bmap_last_extent(
1601 struct xfs_trans *tp,
1602 struct xfs_inode *ip,
1604 struct xfs_bmbt_irec *rec,
1607 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
1611 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
1612 error = xfs_iread_extents(tp, ip, whichfork);
1617 nextents = ifp->if_bytes / sizeof(xfs_bmbt_rec_t);
1618 if (nextents == 0) {
1623 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, nextents - 1), rec);
1629 * Check the last inode extent to determine whether this allocation will result
1630 * in blocks being allocated at the end of the file. When we allocate new data
1631 * blocks at the end of the file which do not start at the previous data block,
1632 * we will try to align the new blocks at stripe unit boundaries.
1634 * Returns 0 in bma->aeof if the file (fork) is empty as any new write will be
1635 * at, or past the EOF.
1639 struct xfs_bmalloca *bma,
1642 struct xfs_bmbt_irec rec;
1647 error = xfs_bmap_last_extent(NULL, bma->ip, whichfork, &rec,
1649 if (error || is_empty)
1653 * Check if we are allocation or past the last extent, or at least into
1654 * the last delayed allocated extent.
1656 bma->aeof = bma->offset >= rec.br_startoff + rec.br_blockcount ||
1657 (bma->offset >= rec.br_startoff &&
1658 isnullstartblock(rec.br_startblock));
1663 * Returns the file-relative block number of the first block past eof in
1664 * the file. This is not based on i_size, it is based on the extent records.
1665 * Returns 0 for local files, as they do not have extent records.
1668 xfs_bmap_last_offset(
1669 struct xfs_trans *tp,
1670 struct xfs_inode *ip,
1671 xfs_fileoff_t *last_block,
1674 struct xfs_bmbt_irec rec;
1680 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_LOCAL)
1683 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE &&
1684 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1685 return XFS_ERROR(EIO);
1687 error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, &is_empty);
1688 if (error || is_empty)
1691 *last_block = rec.br_startoff + rec.br_blockcount;
1696 * Returns whether the selected fork of the inode has exactly one
1697 * block or not. For the data fork we check this matches di_size,
1698 * implying the file's range is 0..bsize-1.
1700 int /* 1=>1 block, 0=>otherwise */
1702 xfs_inode_t *ip, /* incore inode */
1703 int whichfork) /* data or attr fork */
1705 xfs_bmbt_rec_host_t *ep; /* ptr to fork's extent */
1706 xfs_ifork_t *ifp; /* inode fork pointer */
1707 int rval; /* return value */
1708 xfs_bmbt_irec_t s; /* internal version of extent */
1711 if (whichfork == XFS_DATA_FORK)
1712 return XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize;
1714 if (XFS_IFORK_NEXTENTS(ip, whichfork) != 1)
1716 if (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
1718 ifp = XFS_IFORK_PTR(ip, whichfork);
1719 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
1720 ep = xfs_iext_get_ext(ifp, 0);
1721 xfs_bmbt_get_all(ep, &s);
1722 rval = s.br_startoff == 0 && s.br_blockcount == 1;
1723 if (rval && whichfork == XFS_DATA_FORK)
1724 ASSERT(XFS_ISIZE(ip) == ip->i_mount->m_sb.sb_blocksize);
1729 * Extent tree manipulation functions used during allocation.
1733 * Convert a delayed allocation to a real allocation.
1735 STATIC int /* error */
1736 xfs_bmap_add_extent_delay_real(
1737 struct xfs_bmalloca *bma)
1739 struct xfs_bmbt_irec *new = &bma->got;
1740 int diff; /* temp value */
1741 xfs_bmbt_rec_host_t *ep; /* extent entry for idx */
1742 int error; /* error return value */
1743 int i; /* temp state */
1744 xfs_ifork_t *ifp; /* inode fork pointer */
1745 xfs_fileoff_t new_endoff; /* end offset of new entry */
1746 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
1747 /* left is 0, right is 1, prev is 2 */
1748 int rval=0; /* return value (logging flags) */
1749 int state = 0;/* state bits, accessed thru macros */
1750 xfs_filblks_t da_new; /* new count del alloc blocks used */
1751 xfs_filblks_t da_old; /* old count del alloc blocks used */
1752 xfs_filblks_t temp=0; /* value for da_new calculations */
1753 xfs_filblks_t temp2=0;/* value for da_new calculations */
1754 int tmp_rval; /* partial logging flags */
1756 ifp = XFS_IFORK_PTR(bma->ip, XFS_DATA_FORK);
1758 ASSERT(bma->idx >= 0);
1759 ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
1760 ASSERT(!isnullstartblock(new->br_startblock));
1762 (bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
1764 XFS_STATS_INC(xs_add_exlist);
1771 * Set up a bunch of variables to make the tests simpler.
1773 ep = xfs_iext_get_ext(ifp, bma->idx);
1774 xfs_bmbt_get_all(ep, &PREV);
1775 new_endoff = new->br_startoff + new->br_blockcount;
1776 ASSERT(PREV.br_startoff <= new->br_startoff);
1777 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
1779 da_old = startblockval(PREV.br_startblock);
1783 * Set flags determining what part of the previous delayed allocation
1784 * extent is being replaced by a real allocation.
1786 if (PREV.br_startoff == new->br_startoff)
1787 state |= BMAP_LEFT_FILLING;
1788 if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
1789 state |= BMAP_RIGHT_FILLING;
1792 * Check and set flags if this segment has a left neighbor.
1793 * Don't set contiguous if the combined extent would be too large.
1796 state |= BMAP_LEFT_VALID;
1797 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &LEFT);
1799 if (isnullstartblock(LEFT.br_startblock))
1800 state |= BMAP_LEFT_DELAY;
1803 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
1804 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
1805 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
1806 LEFT.br_state == new->br_state &&
1807 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
1808 state |= BMAP_LEFT_CONTIG;
1811 * Check and set flags if this segment has a right neighbor.
1812 * Don't set contiguous if the combined extent would be too large.
1813 * Also check for all-three-contiguous being too large.
1815 if (bma->idx < bma->ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
1816 state |= BMAP_RIGHT_VALID;
1817 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx + 1), &RIGHT);
1819 if (isnullstartblock(RIGHT.br_startblock))
1820 state |= BMAP_RIGHT_DELAY;
1823 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
1824 new_endoff == RIGHT.br_startoff &&
1825 new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
1826 new->br_state == RIGHT.br_state &&
1827 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
1828 ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1829 BMAP_RIGHT_FILLING)) !=
1830 (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
1831 BMAP_RIGHT_FILLING) ||
1832 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
1834 state |= BMAP_RIGHT_CONTIG;
1838 * Switch out based on the FILLING and CONTIG state bits.
1840 switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1841 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
1842 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
1843 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1845 * Filling in all of a previously delayed allocation extent.
1846 * The left and right neighbors are both contiguous with new.
1849 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1850 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1851 LEFT.br_blockcount + PREV.br_blockcount +
1852 RIGHT.br_blockcount);
1853 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1855 xfs_iext_remove(bma->ip, bma->idx + 1, 2, state);
1856 bma->ip->i_d.di_nextents--;
1857 if (bma->cur == NULL)
1858 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1860 rval = XFS_ILOG_CORE;
1861 error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1862 RIGHT.br_startblock,
1863 RIGHT.br_blockcount, &i);
1866 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1867 error = xfs_btree_delete(bma->cur, &i);
1870 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1871 error = xfs_btree_decrement(bma->cur, 0, &i);
1874 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1875 error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1877 LEFT.br_blockcount +
1878 PREV.br_blockcount +
1879 RIGHT.br_blockcount, LEFT.br_state);
1885 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
1887 * Filling in all of a previously delayed allocation extent.
1888 * The left neighbor is contiguous, the right is not.
1892 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1893 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
1894 LEFT.br_blockcount + PREV.br_blockcount);
1895 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1897 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1898 if (bma->cur == NULL)
1899 rval = XFS_ILOG_DEXT;
1902 error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1903 LEFT.br_startblock, LEFT.br_blockcount,
1907 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1908 error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
1910 LEFT.br_blockcount +
1911 PREV.br_blockcount, LEFT.br_state);
1917 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
1919 * Filling in all of a previously delayed allocation extent.
1920 * The right neighbor is contiguous, the left is not.
1922 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1923 xfs_bmbt_set_startblock(ep, new->br_startblock);
1924 xfs_bmbt_set_blockcount(ep,
1925 PREV.br_blockcount + RIGHT.br_blockcount);
1926 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1928 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
1929 if (bma->cur == NULL)
1930 rval = XFS_ILOG_DEXT;
1933 error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
1934 RIGHT.br_startblock,
1935 RIGHT.br_blockcount, &i);
1938 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1939 error = xfs_bmbt_update(bma->cur, PREV.br_startoff,
1941 PREV.br_blockcount +
1942 RIGHT.br_blockcount, PREV.br_state);
1948 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
1950 * Filling in all of a previously delayed allocation extent.
1951 * Neither the left nor right neighbors are contiguous with
1954 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1955 xfs_bmbt_set_startblock(ep, new->br_startblock);
1956 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
1958 bma->ip->i_d.di_nextents++;
1959 if (bma->cur == NULL)
1960 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
1962 rval = XFS_ILOG_CORE;
1963 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
1964 new->br_startblock, new->br_blockcount,
1968 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
1969 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
1970 error = xfs_btree_insert(bma->cur, &i);
1973 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
1977 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
1979 * Filling in the first part of a previous delayed allocation.
1980 * The left neighbor is contiguous.
1982 trace_xfs_bmap_pre_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1983 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx - 1),
1984 LEFT.br_blockcount + new->br_blockcount);
1985 xfs_bmbt_set_startoff(ep,
1986 PREV.br_startoff + new->br_blockcount);
1987 trace_xfs_bmap_post_update(bma->ip, bma->idx - 1, state, _THIS_IP_);
1989 temp = PREV.br_blockcount - new->br_blockcount;
1990 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
1991 xfs_bmbt_set_blockcount(ep, temp);
1992 if (bma->cur == NULL)
1993 rval = XFS_ILOG_DEXT;
1996 error = xfs_bmbt_lookup_eq(bma->cur, LEFT.br_startoff,
1997 LEFT.br_startblock, LEFT.br_blockcount,
2001 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2002 error = xfs_bmbt_update(bma->cur, LEFT.br_startoff,
2004 LEFT.br_blockcount +
2010 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2011 startblockval(PREV.br_startblock));
2012 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2013 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2018 case BMAP_LEFT_FILLING:
2020 * Filling in the first part of a previous delayed allocation.
2021 * The left neighbor is not contiguous.
2023 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2024 xfs_bmbt_set_startoff(ep, new_endoff);
2025 temp = PREV.br_blockcount - new->br_blockcount;
2026 xfs_bmbt_set_blockcount(ep, temp);
2027 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
2028 bma->ip->i_d.di_nextents++;
2029 if (bma->cur == NULL)
2030 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2032 rval = XFS_ILOG_CORE;
2033 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2034 new->br_startblock, new->br_blockcount,
2038 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2039 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2040 error = xfs_btree_insert(bma->cur, &i);
2043 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2046 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2047 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2048 bma->firstblock, bma->flist,
2049 &bma->cur, 1, &tmp_rval, XFS_DATA_FORK);
2054 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2055 startblockval(PREV.br_startblock) -
2056 (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2057 ep = xfs_iext_get_ext(ifp, bma->idx + 1);
2058 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2059 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2062 case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2064 * Filling in the last part of a previous delayed allocation.
2065 * The right neighbor is contiguous with the new allocation.
2067 temp = PREV.br_blockcount - new->br_blockcount;
2068 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2069 xfs_bmbt_set_blockcount(ep, temp);
2070 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx + 1),
2071 new->br_startoff, new->br_startblock,
2072 new->br_blockcount + RIGHT.br_blockcount,
2074 trace_xfs_bmap_post_update(bma->ip, bma->idx + 1, state, _THIS_IP_);
2075 if (bma->cur == NULL)
2076 rval = XFS_ILOG_DEXT;
2079 error = xfs_bmbt_lookup_eq(bma->cur, RIGHT.br_startoff,
2080 RIGHT.br_startblock,
2081 RIGHT.br_blockcount, &i);
2084 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2085 error = xfs_bmbt_update(bma->cur, new->br_startoff,
2087 new->br_blockcount +
2088 RIGHT.br_blockcount,
2094 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2095 startblockval(PREV.br_startblock));
2096 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2097 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2098 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2103 case BMAP_RIGHT_FILLING:
2105 * Filling in the last part of a previous delayed allocation.
2106 * The right neighbor is not contiguous.
2108 temp = PREV.br_blockcount - new->br_blockcount;
2109 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
2110 xfs_bmbt_set_blockcount(ep, temp);
2111 xfs_iext_insert(bma->ip, bma->idx + 1, 1, new, state);
2112 bma->ip->i_d.di_nextents++;
2113 if (bma->cur == NULL)
2114 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2116 rval = XFS_ILOG_CORE;
2117 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2118 new->br_startblock, new->br_blockcount,
2122 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2123 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2124 error = xfs_btree_insert(bma->cur, &i);
2127 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2130 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2131 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2132 bma->firstblock, bma->flist, &bma->cur, 1,
2133 &tmp_rval, XFS_DATA_FORK);
2138 da_new = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(bma->ip, temp),
2139 startblockval(PREV.br_startblock) -
2140 (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2141 ep = xfs_iext_get_ext(ifp, bma->idx);
2142 xfs_bmbt_set_startblock(ep, nullstartblock(da_new));
2143 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2150 * Filling in the middle part of a previous delayed allocation.
2151 * Contiguity is impossible here.
2152 * This case is avoided almost all the time.
2154 * We start with a delayed allocation:
2156 * +ddddddddddddddddddddddddddddddddddddddddddddddddddddddd+
2159 * and we are allocating:
2160 * +rrrrrrrrrrrrrrrrr+
2163 * and we set it up for insertion as:
2164 * +ddddddddddddddddddd+rrrrrrrrrrrrrrrrr+ddddddddddddddddd+
2166 * PREV @ idx LEFT RIGHT
2167 * inserted at idx + 1
2169 temp = new->br_startoff - PREV.br_startoff;
2170 temp2 = PREV.br_startoff + PREV.br_blockcount - new_endoff;
2171 trace_xfs_bmap_pre_update(bma->ip, bma->idx, 0, _THIS_IP_);
2172 xfs_bmbt_set_blockcount(ep, temp); /* truncate PREV */
2174 RIGHT.br_state = PREV.br_state;
2175 RIGHT.br_startblock = nullstartblock(
2176 (int)xfs_bmap_worst_indlen(bma->ip, temp2));
2177 RIGHT.br_startoff = new_endoff;
2178 RIGHT.br_blockcount = temp2;
2179 /* insert LEFT (r[0]) and RIGHT (r[1]) at the same time */
2180 xfs_iext_insert(bma->ip, bma->idx + 1, 2, &LEFT, state);
2181 bma->ip->i_d.di_nextents++;
2182 if (bma->cur == NULL)
2183 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2185 rval = XFS_ILOG_CORE;
2186 error = xfs_bmbt_lookup_eq(bma->cur, new->br_startoff,
2187 new->br_startblock, new->br_blockcount,
2191 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2192 bma->cur->bc_rec.b.br_state = XFS_EXT_NORM;
2193 error = xfs_btree_insert(bma->cur, &i);
2196 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2199 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2200 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2201 bma->firstblock, bma->flist, &bma->cur,
2202 1, &tmp_rval, XFS_DATA_FORK);
2207 temp = xfs_bmap_worst_indlen(bma->ip, temp);
2208 temp2 = xfs_bmap_worst_indlen(bma->ip, temp2);
2209 diff = (int)(temp + temp2 - startblockval(PREV.br_startblock) -
2210 (bma->cur ? bma->cur->bc_private.b.allocated : 0));
2212 error = xfs_icsb_modify_counters(bma->ip->i_mount,
2214 -((int64_t)diff), 0);
2220 ep = xfs_iext_get_ext(ifp, bma->idx);
2221 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
2222 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
2223 trace_xfs_bmap_pre_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2224 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, bma->idx + 2),
2225 nullstartblock((int)temp2));
2226 trace_xfs_bmap_post_update(bma->ip, bma->idx + 2, state, _THIS_IP_);
2229 da_new = temp + temp2;
2232 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2233 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2234 case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2235 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2236 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2237 case BMAP_LEFT_CONTIG:
2238 case BMAP_RIGHT_CONTIG:
2240 * These cases are all impossible.
2245 /* convert to a btree if necessary */
2246 if (xfs_bmap_needs_btree(bma->ip, XFS_DATA_FORK)) {
2247 int tmp_logflags; /* partial log flag return val */
2249 ASSERT(bma->cur == NULL);
2250 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
2251 bma->firstblock, bma->flist, &bma->cur,
2252 da_old > 0, &tmp_logflags, XFS_DATA_FORK);
2253 bma->logflags |= tmp_logflags;
2258 /* adjust for changes in reserved delayed indirect blocks */
2259 if (da_old || da_new) {
2262 temp += bma->cur->bc_private.b.allocated;
2263 ASSERT(temp <= da_old);
2265 xfs_icsb_modify_counters(bma->ip->i_mount,
2267 (int64_t)(da_old - temp), 0);
2270 /* clear out the allocated field, done with it now in any case. */
2272 bma->cur->bc_private.b.allocated = 0;
2274 xfs_bmap_check_leaf_extents(bma->cur, bma->ip, XFS_DATA_FORK);
2276 bma->logflags |= rval;
2284 * Convert an unwritten allocation to a real allocation or vice versa.
2286 STATIC int /* error */
2287 xfs_bmap_add_extent_unwritten_real(
2288 struct xfs_trans *tp,
2289 xfs_inode_t *ip, /* incore inode pointer */
2290 xfs_extnum_t *idx, /* extent number to update/insert */
2291 xfs_btree_cur_t **curp, /* if *curp is null, not a btree */
2292 xfs_bmbt_irec_t *new, /* new data to add to file extents */
2293 xfs_fsblock_t *first, /* pointer to firstblock variable */
2294 xfs_bmap_free_t *flist, /* list of extents to be freed */
2295 int *logflagsp) /* inode logging flags */
2297 xfs_btree_cur_t *cur; /* btree cursor */
2298 xfs_bmbt_rec_host_t *ep; /* extent entry for idx */
2299 int error; /* error return value */
2300 int i; /* temp state */
2301 xfs_ifork_t *ifp; /* inode fork pointer */
2302 xfs_fileoff_t new_endoff; /* end offset of new entry */
2303 xfs_exntst_t newext; /* new extent state */
2304 xfs_exntst_t oldext; /* old extent state */
2305 xfs_bmbt_irec_t r[3]; /* neighbor extent entries */
2306 /* left is 0, right is 1, prev is 2 */
2307 int rval=0; /* return value (logging flags) */
2308 int state = 0;/* state bits, accessed thru macros */
2313 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2316 ASSERT(*idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2317 ASSERT(!isnullstartblock(new->br_startblock));
2319 XFS_STATS_INC(xs_add_exlist);
2326 * Set up a bunch of variables to make the tests simpler.
2329 ep = xfs_iext_get_ext(ifp, *idx);
2330 xfs_bmbt_get_all(ep, &PREV);
2331 newext = new->br_state;
2332 oldext = (newext == XFS_EXT_UNWRITTEN) ?
2333 XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
2334 ASSERT(PREV.br_state == oldext);
2335 new_endoff = new->br_startoff + new->br_blockcount;
2336 ASSERT(PREV.br_startoff <= new->br_startoff);
2337 ASSERT(PREV.br_startoff + PREV.br_blockcount >= new_endoff);
2340 * Set flags determining what part of the previous oldext allocation
2341 * extent is being replaced by a newext allocation.
2343 if (PREV.br_startoff == new->br_startoff)
2344 state |= BMAP_LEFT_FILLING;
2345 if (PREV.br_startoff + PREV.br_blockcount == new_endoff)
2346 state |= BMAP_RIGHT_FILLING;
2349 * Check and set flags if this segment has a left neighbor.
2350 * Don't set contiguous if the combined extent would be too large.
2353 state |= BMAP_LEFT_VALID;
2354 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &LEFT);
2356 if (isnullstartblock(LEFT.br_startblock))
2357 state |= BMAP_LEFT_DELAY;
2360 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
2361 LEFT.br_startoff + LEFT.br_blockcount == new->br_startoff &&
2362 LEFT.br_startblock + LEFT.br_blockcount == new->br_startblock &&
2363 LEFT.br_state == newext &&
2364 LEFT.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2365 state |= BMAP_LEFT_CONTIG;
2368 * Check and set flags if this segment has a right neighbor.
2369 * Don't set contiguous if the combined extent would be too large.
2370 * Also check for all-three-contiguous being too large.
2372 if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t) - 1) {
2373 state |= BMAP_RIGHT_VALID;
2374 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx + 1), &RIGHT);
2375 if (isnullstartblock(RIGHT.br_startblock))
2376 state |= BMAP_RIGHT_DELAY;
2379 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
2380 new_endoff == RIGHT.br_startoff &&
2381 new->br_startblock + new->br_blockcount == RIGHT.br_startblock &&
2382 newext == RIGHT.br_state &&
2383 new->br_blockcount + RIGHT.br_blockcount <= MAXEXTLEN &&
2384 ((state & (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2385 BMAP_RIGHT_FILLING)) !=
2386 (BMAP_LEFT_CONTIG | BMAP_LEFT_FILLING |
2387 BMAP_RIGHT_FILLING) ||
2388 LEFT.br_blockcount + new->br_blockcount + RIGHT.br_blockcount
2390 state |= BMAP_RIGHT_CONTIG;
2393 * Switch out based on the FILLING and CONTIG state bits.
2395 switch (state & (BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2396 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG)) {
2397 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG |
2398 BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2400 * Setting all of a previous oldext extent to newext.
2401 * The left and right neighbors are both contiguous with new.
2405 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2406 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2407 LEFT.br_blockcount + PREV.br_blockcount +
2408 RIGHT.br_blockcount);
2409 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2411 xfs_iext_remove(ip, *idx + 1, 2, state);
2412 ip->i_d.di_nextents -= 2;
2414 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2416 rval = XFS_ILOG_CORE;
2417 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2418 RIGHT.br_startblock,
2419 RIGHT.br_blockcount, &i)))
2421 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2422 if ((error = xfs_btree_delete(cur, &i)))
2424 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2425 if ((error = xfs_btree_decrement(cur, 0, &i)))
2427 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2428 if ((error = xfs_btree_delete(cur, &i)))
2430 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2431 if ((error = xfs_btree_decrement(cur, 0, &i)))
2433 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2434 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2436 LEFT.br_blockcount + PREV.br_blockcount +
2437 RIGHT.br_blockcount, LEFT.br_state)))
2442 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2444 * Setting all of a previous oldext extent to newext.
2445 * The left neighbor is contiguous, the right is not.
2449 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2450 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx),
2451 LEFT.br_blockcount + PREV.br_blockcount);
2452 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2454 xfs_iext_remove(ip, *idx + 1, 1, state);
2455 ip->i_d.di_nextents--;
2457 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2459 rval = XFS_ILOG_CORE;
2460 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2461 PREV.br_startblock, PREV.br_blockcount,
2464 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2465 if ((error = xfs_btree_delete(cur, &i)))
2467 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2468 if ((error = xfs_btree_decrement(cur, 0, &i)))
2470 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2471 if ((error = xfs_bmbt_update(cur, LEFT.br_startoff,
2473 LEFT.br_blockcount + PREV.br_blockcount,
2479 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2481 * Setting all of a previous oldext extent to newext.
2482 * The right neighbor is contiguous, the left is not.
2484 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2485 xfs_bmbt_set_blockcount(ep,
2486 PREV.br_blockcount + RIGHT.br_blockcount);
2487 xfs_bmbt_set_state(ep, newext);
2488 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2489 xfs_iext_remove(ip, *idx + 1, 1, state);
2490 ip->i_d.di_nextents--;
2492 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2494 rval = XFS_ILOG_CORE;
2495 if ((error = xfs_bmbt_lookup_eq(cur, RIGHT.br_startoff,
2496 RIGHT.br_startblock,
2497 RIGHT.br_blockcount, &i)))
2499 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2500 if ((error = xfs_btree_delete(cur, &i)))
2502 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2503 if ((error = xfs_btree_decrement(cur, 0, &i)))
2505 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2506 if ((error = xfs_bmbt_update(cur, new->br_startoff,
2508 new->br_blockcount + RIGHT.br_blockcount,
2514 case BMAP_LEFT_FILLING | BMAP_RIGHT_FILLING:
2516 * Setting all of a previous oldext extent to newext.
2517 * Neither the left nor right neighbors are contiguous with
2520 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2521 xfs_bmbt_set_state(ep, newext);
2522 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2525 rval = XFS_ILOG_DEXT;
2528 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2529 new->br_startblock, new->br_blockcount,
2532 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2533 if ((error = xfs_bmbt_update(cur, new->br_startoff,
2534 new->br_startblock, new->br_blockcount,
2540 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG:
2542 * Setting the first part of a previous oldext extent to newext.
2543 * The left neighbor is contiguous.
2545 trace_xfs_bmap_pre_update(ip, *idx - 1, state, _THIS_IP_);
2546 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx - 1),
2547 LEFT.br_blockcount + new->br_blockcount);
2548 xfs_bmbt_set_startoff(ep,
2549 PREV.br_startoff + new->br_blockcount);
2550 trace_xfs_bmap_post_update(ip, *idx - 1, state, _THIS_IP_);
2552 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2553 xfs_bmbt_set_startblock(ep,
2554 new->br_startblock + new->br_blockcount);
2555 xfs_bmbt_set_blockcount(ep,
2556 PREV.br_blockcount - new->br_blockcount);
2557 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2562 rval = XFS_ILOG_DEXT;
2565 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2566 PREV.br_startblock, PREV.br_blockcount,
2569 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2570 if ((error = xfs_bmbt_update(cur,
2571 PREV.br_startoff + new->br_blockcount,
2572 PREV.br_startblock + new->br_blockcount,
2573 PREV.br_blockcount - new->br_blockcount,
2576 if ((error = xfs_btree_decrement(cur, 0, &i)))
2578 error = xfs_bmbt_update(cur, LEFT.br_startoff,
2580 LEFT.br_blockcount + new->br_blockcount,
2587 case BMAP_LEFT_FILLING:
2589 * Setting the first part of a previous oldext extent to newext.
2590 * The left neighbor is not contiguous.
2592 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2593 ASSERT(ep && xfs_bmbt_get_state(ep) == oldext);
2594 xfs_bmbt_set_startoff(ep, new_endoff);
2595 xfs_bmbt_set_blockcount(ep,
2596 PREV.br_blockcount - new->br_blockcount);
2597 xfs_bmbt_set_startblock(ep,
2598 new->br_startblock + new->br_blockcount);
2599 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2601 xfs_iext_insert(ip, *idx, 1, new, state);
2602 ip->i_d.di_nextents++;
2604 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2606 rval = XFS_ILOG_CORE;
2607 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2608 PREV.br_startblock, PREV.br_blockcount,
2611 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2612 if ((error = xfs_bmbt_update(cur,
2613 PREV.br_startoff + new->br_blockcount,
2614 PREV.br_startblock + new->br_blockcount,
2615 PREV.br_blockcount - new->br_blockcount,
2618 cur->bc_rec.b = *new;
2619 if ((error = xfs_btree_insert(cur, &i)))
2621 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2625 case BMAP_RIGHT_FILLING | BMAP_RIGHT_CONTIG:
2627 * Setting the last part of a previous oldext extent to newext.
2628 * The right neighbor is contiguous with the new allocation.
2630 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2631 xfs_bmbt_set_blockcount(ep,
2632 PREV.br_blockcount - new->br_blockcount);
2633 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2637 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2638 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2639 new->br_startoff, new->br_startblock,
2640 new->br_blockcount + RIGHT.br_blockcount, newext);
2641 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2644 rval = XFS_ILOG_DEXT;
2647 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2649 PREV.br_blockcount, &i)))
2651 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2652 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2654 PREV.br_blockcount - new->br_blockcount,
2657 if ((error = xfs_btree_increment(cur, 0, &i)))
2659 if ((error = xfs_bmbt_update(cur, new->br_startoff,
2661 new->br_blockcount + RIGHT.br_blockcount,
2667 case BMAP_RIGHT_FILLING:
2669 * Setting the last part of a previous oldext extent to newext.
2670 * The right neighbor is not contiguous.
2672 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2673 xfs_bmbt_set_blockcount(ep,
2674 PREV.br_blockcount - new->br_blockcount);
2675 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2678 xfs_iext_insert(ip, *idx, 1, new, state);
2680 ip->i_d.di_nextents++;
2682 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2684 rval = XFS_ILOG_CORE;
2685 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2686 PREV.br_startblock, PREV.br_blockcount,
2689 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2690 if ((error = xfs_bmbt_update(cur, PREV.br_startoff,
2692 PREV.br_blockcount - new->br_blockcount,
2695 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2696 new->br_startblock, new->br_blockcount,
2699 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2700 cur->bc_rec.b.br_state = XFS_EXT_NORM;
2701 if ((error = xfs_btree_insert(cur, &i)))
2703 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2709 * Setting the middle part of a previous oldext extent to
2710 * newext. Contiguity is impossible here.
2711 * One extent becomes three extents.
2713 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2714 xfs_bmbt_set_blockcount(ep,
2715 new->br_startoff - PREV.br_startoff);
2716 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2719 r[1].br_startoff = new_endoff;
2720 r[1].br_blockcount =
2721 PREV.br_startoff + PREV.br_blockcount - new_endoff;
2722 r[1].br_startblock = new->br_startblock + new->br_blockcount;
2723 r[1].br_state = oldext;
2726 xfs_iext_insert(ip, *idx, 2, &r[0], state);
2728 ip->i_d.di_nextents += 2;
2730 rval = XFS_ILOG_CORE | XFS_ILOG_DEXT;
2732 rval = XFS_ILOG_CORE;
2733 if ((error = xfs_bmbt_lookup_eq(cur, PREV.br_startoff,
2734 PREV.br_startblock, PREV.br_blockcount,
2737 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2738 /* new right extent - oldext */
2739 if ((error = xfs_bmbt_update(cur, r[1].br_startoff,
2740 r[1].br_startblock, r[1].br_blockcount,
2743 /* new left extent - oldext */
2744 cur->bc_rec.b = PREV;
2745 cur->bc_rec.b.br_blockcount =
2746 new->br_startoff - PREV.br_startoff;
2747 if ((error = xfs_btree_insert(cur, &i)))
2749 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2751 * Reset the cursor to the position of the new extent
2752 * we are about to insert as we can't trust it after
2753 * the previous insert.
2755 if ((error = xfs_bmbt_lookup_eq(cur, new->br_startoff,
2756 new->br_startblock, new->br_blockcount,
2759 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
2760 /* new middle extent - newext */
2761 cur->bc_rec.b.br_state = new->br_state;
2762 if ((error = xfs_btree_insert(cur, &i)))
2764 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
2768 case BMAP_LEFT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2769 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2770 case BMAP_LEFT_FILLING | BMAP_RIGHT_CONTIG:
2771 case BMAP_RIGHT_FILLING | BMAP_LEFT_CONTIG:
2772 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2773 case BMAP_LEFT_CONTIG:
2774 case BMAP_RIGHT_CONTIG:
2776 * These cases are all impossible.
2781 /* convert to a btree if necessary */
2782 if (xfs_bmap_needs_btree(ip, XFS_DATA_FORK)) {
2783 int tmp_logflags; /* partial log flag return val */
2785 ASSERT(cur == NULL);
2786 error = xfs_bmap_extents_to_btree(tp, ip, first, flist, &cur,
2787 0, &tmp_logflags, XFS_DATA_FORK);
2788 *logflagsp |= tmp_logflags;
2793 /* clear out the allocated field, done with it now in any case. */
2795 cur->bc_private.b.allocated = 0;
2799 xfs_bmap_check_leaf_extents(*curp, ip, XFS_DATA_FORK);
2809 * Convert a hole to a delayed allocation.
2812 xfs_bmap_add_extent_hole_delay(
2813 xfs_inode_t *ip, /* incore inode pointer */
2814 xfs_extnum_t *idx, /* extent number to update/insert */
2815 xfs_bmbt_irec_t *new) /* new data to add to file extents */
2817 xfs_ifork_t *ifp; /* inode fork pointer */
2818 xfs_bmbt_irec_t left; /* left neighbor extent entry */
2819 xfs_filblks_t newlen=0; /* new indirect size */
2820 xfs_filblks_t oldlen=0; /* old indirect size */
2821 xfs_bmbt_irec_t right; /* right neighbor extent entry */
2822 int state; /* state bits, accessed thru macros */
2823 xfs_filblks_t temp=0; /* temp for indirect calculations */
2825 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
2827 ASSERT(isnullstartblock(new->br_startblock));
2830 * Check and set flags if this segment has a left neighbor
2833 state |= BMAP_LEFT_VALID;
2834 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx - 1), &left);
2836 if (isnullstartblock(left.br_startblock))
2837 state |= BMAP_LEFT_DELAY;
2841 * Check and set flags if the current (right) segment exists.
2842 * If it doesn't exist, we're converting the hole at end-of-file.
2844 if (*idx < ip->i_df.if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2845 state |= BMAP_RIGHT_VALID;
2846 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *idx), &right);
2848 if (isnullstartblock(right.br_startblock))
2849 state |= BMAP_RIGHT_DELAY;
2853 * Set contiguity flags on the left and right neighbors.
2854 * Don't let extents get too large, even if the pieces are contiguous.
2856 if ((state & BMAP_LEFT_VALID) && (state & BMAP_LEFT_DELAY) &&
2857 left.br_startoff + left.br_blockcount == new->br_startoff &&
2858 left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
2859 state |= BMAP_LEFT_CONTIG;
2861 if ((state & BMAP_RIGHT_VALID) && (state & BMAP_RIGHT_DELAY) &&
2862 new->br_startoff + new->br_blockcount == right.br_startoff &&
2863 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
2864 (!(state & BMAP_LEFT_CONTIG) ||
2865 (left.br_blockcount + new->br_blockcount +
2866 right.br_blockcount <= MAXEXTLEN)))
2867 state |= BMAP_RIGHT_CONTIG;
2870 * Switch out based on the contiguity flags.
2872 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
2873 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
2875 * New allocation is contiguous with delayed allocations
2876 * on the left and on the right.
2877 * Merge all three into a single extent record.
2880 temp = left.br_blockcount + new->br_blockcount +
2881 right.br_blockcount;
2883 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2884 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2885 oldlen = startblockval(left.br_startblock) +
2886 startblockval(new->br_startblock) +
2887 startblockval(right.br_startblock);
2888 newlen = xfs_bmap_worst_indlen(ip, temp);
2889 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2890 nullstartblock((int)newlen));
2891 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2893 xfs_iext_remove(ip, *idx + 1, 1, state);
2896 case BMAP_LEFT_CONTIG:
2898 * New allocation is contiguous with a delayed allocation
2900 * Merge the new allocation with the left neighbor.
2903 temp = left.br_blockcount + new->br_blockcount;
2905 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2906 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, *idx), temp);
2907 oldlen = startblockval(left.br_startblock) +
2908 startblockval(new->br_startblock);
2909 newlen = xfs_bmap_worst_indlen(ip, temp);
2910 xfs_bmbt_set_startblock(xfs_iext_get_ext(ifp, *idx),
2911 nullstartblock((int)newlen));
2912 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2915 case BMAP_RIGHT_CONTIG:
2917 * New allocation is contiguous with a delayed allocation
2919 * Merge the new allocation with the right neighbor.
2921 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
2922 temp = new->br_blockcount + right.br_blockcount;
2923 oldlen = startblockval(new->br_startblock) +
2924 startblockval(right.br_startblock);
2925 newlen = xfs_bmap_worst_indlen(ip, temp);
2926 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, *idx),
2928 nullstartblock((int)newlen), temp, right.br_state);
2929 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
2934 * New allocation is not contiguous with another
2935 * delayed allocation.
2936 * Insert a new entry.
2938 oldlen = newlen = 0;
2939 xfs_iext_insert(ip, *idx, 1, new, state);
2942 if (oldlen != newlen) {
2943 ASSERT(oldlen > newlen);
2944 xfs_icsb_modify_counters(ip->i_mount, XFS_SBS_FDBLOCKS,
2945 (int64_t)(oldlen - newlen), 0);
2947 * Nothing to do for disk quota accounting here.
2953 * Convert a hole to a real allocation.
2955 STATIC int /* error */
2956 xfs_bmap_add_extent_hole_real(
2957 struct xfs_bmalloca *bma,
2960 struct xfs_bmbt_irec *new = &bma->got;
2961 int error; /* error return value */
2962 int i; /* temp state */
2963 xfs_ifork_t *ifp; /* inode fork pointer */
2964 xfs_bmbt_irec_t left; /* left neighbor extent entry */
2965 xfs_bmbt_irec_t right; /* right neighbor extent entry */
2966 int rval=0; /* return value (logging flags) */
2967 int state; /* state bits, accessed thru macros */
2969 ifp = XFS_IFORK_PTR(bma->ip, whichfork);
2971 ASSERT(bma->idx >= 0);
2972 ASSERT(bma->idx <= ifp->if_bytes / sizeof(struct xfs_bmbt_rec));
2973 ASSERT(!isnullstartblock(new->br_startblock));
2975 !(bma->cur->bc_private.b.flags & XFS_BTCUR_BPRV_WASDEL));
2977 XFS_STATS_INC(xs_add_exlist);
2980 if (whichfork == XFS_ATTR_FORK)
2981 state |= BMAP_ATTRFORK;
2984 * Check and set flags if this segment has a left neighbor.
2987 state |= BMAP_LEFT_VALID;
2988 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1), &left);
2989 if (isnullstartblock(left.br_startblock))
2990 state |= BMAP_LEFT_DELAY;
2994 * Check and set flags if this segment has a current value.
2995 * Not true if we're inserting into the "hole" at eof.
2997 if (bma->idx < ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t)) {
2998 state |= BMAP_RIGHT_VALID;
2999 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &right);
3000 if (isnullstartblock(right.br_startblock))
3001 state |= BMAP_RIGHT_DELAY;
3005 * We're inserting a real allocation between "left" and "right".
3006 * Set the contiguity flags. Don't let extents get too large.
3008 if ((state & BMAP_LEFT_VALID) && !(state & BMAP_LEFT_DELAY) &&
3009 left.br_startoff + left.br_blockcount == new->br_startoff &&
3010 left.br_startblock + left.br_blockcount == new->br_startblock &&
3011 left.br_state == new->br_state &&
3012 left.br_blockcount + new->br_blockcount <= MAXEXTLEN)
3013 state |= BMAP_LEFT_CONTIG;
3015 if ((state & BMAP_RIGHT_VALID) && !(state & BMAP_RIGHT_DELAY) &&
3016 new->br_startoff + new->br_blockcount == right.br_startoff &&
3017 new->br_startblock + new->br_blockcount == right.br_startblock &&
3018 new->br_state == right.br_state &&
3019 new->br_blockcount + right.br_blockcount <= MAXEXTLEN &&
3020 (!(state & BMAP_LEFT_CONTIG) ||
3021 left.br_blockcount + new->br_blockcount +
3022 right.br_blockcount <= MAXEXTLEN))
3023 state |= BMAP_RIGHT_CONTIG;
3027 * Select which case we're in here, and implement it.
3029 switch (state & (BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG)) {
3030 case BMAP_LEFT_CONTIG | BMAP_RIGHT_CONTIG:
3032 * New allocation is contiguous with real allocations on the
3033 * left and on the right.
3034 * Merge all three into a single extent record.
3037 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3038 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3039 left.br_blockcount + new->br_blockcount +
3040 right.br_blockcount);
3041 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3043 xfs_iext_remove(bma->ip, bma->idx + 1, 1, state);
3045 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3046 XFS_IFORK_NEXTENTS(bma->ip, whichfork) - 1);
3047 if (bma->cur == NULL) {
3048 rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3050 rval = XFS_ILOG_CORE;
3051 error = xfs_bmbt_lookup_eq(bma->cur, right.br_startoff,
3052 right.br_startblock, right.br_blockcount,
3056 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3057 error = xfs_btree_delete(bma->cur, &i);
3060 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3061 error = xfs_btree_decrement(bma->cur, 0, &i);
3064 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3065 error = xfs_bmbt_update(bma->cur, left.br_startoff,
3067 left.br_blockcount +
3068 new->br_blockcount +
3069 right.br_blockcount,
3076 case BMAP_LEFT_CONTIG:
3078 * New allocation is contiguous with a real allocation
3080 * Merge the new allocation with the left neighbor.
3083 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3084 xfs_bmbt_set_blockcount(xfs_iext_get_ext(ifp, bma->idx),
3085 left.br_blockcount + new->br_blockcount);
3086 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3088 if (bma->cur == NULL) {
3089 rval = xfs_ilog_fext(whichfork);
3092 error = xfs_bmbt_lookup_eq(bma->cur, left.br_startoff,
3093 left.br_startblock, left.br_blockcount,
3097 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3098 error = xfs_bmbt_update(bma->cur, left.br_startoff,
3100 left.br_blockcount +
3108 case BMAP_RIGHT_CONTIG:
3110 * New allocation is contiguous with a real allocation
3112 * Merge the new allocation with the right neighbor.
3114 trace_xfs_bmap_pre_update(bma->ip, bma->idx, state, _THIS_IP_);
3115 xfs_bmbt_set_allf(xfs_iext_get_ext(ifp, bma->idx),
3116 new->br_startoff, new->br_startblock,
3117 new->br_blockcount + right.br_blockcount,
3119 trace_xfs_bmap_post_update(bma->ip, bma->idx, state, _THIS_IP_);
3121 if (bma->cur == NULL) {
3122 rval = xfs_ilog_fext(whichfork);
3125 error = xfs_bmbt_lookup_eq(bma->cur,
3127 right.br_startblock,
3128 right.br_blockcount, &i);
3131 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3132 error = xfs_bmbt_update(bma->cur, new->br_startoff,
3134 new->br_blockcount +
3135 right.br_blockcount,
3144 * New allocation is not contiguous with another
3146 * Insert a new entry.
3148 xfs_iext_insert(bma->ip, bma->idx, 1, new, state);
3149 XFS_IFORK_NEXT_SET(bma->ip, whichfork,
3150 XFS_IFORK_NEXTENTS(bma->ip, whichfork) + 1);
3151 if (bma->cur == NULL) {
3152 rval = XFS_ILOG_CORE | xfs_ilog_fext(whichfork);
3154 rval = XFS_ILOG_CORE;
3155 error = xfs_bmbt_lookup_eq(bma->cur,
3158 new->br_blockcount, &i);
3161 XFS_WANT_CORRUPTED_GOTO(i == 0, done);
3162 bma->cur->bc_rec.b.br_state = new->br_state;
3163 error = xfs_btree_insert(bma->cur, &i);
3166 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
3171 /* convert to a btree if necessary */
3172 if (xfs_bmap_needs_btree(bma->ip, whichfork)) {
3173 int tmp_logflags; /* partial log flag return val */
3175 ASSERT(bma->cur == NULL);
3176 error = xfs_bmap_extents_to_btree(bma->tp, bma->ip,
3177 bma->firstblock, bma->flist, &bma->cur,
3178 0, &tmp_logflags, whichfork);
3179 bma->logflags |= tmp_logflags;
3184 /* clear out the allocated field, done with it now in any case. */
3186 bma->cur->bc_private.b.allocated = 0;
3188 xfs_bmap_check_leaf_extents(bma->cur, bma->ip, whichfork);
3190 bma->logflags |= rval;
3195 * Functions used in the extent read, allocate and remove paths
3199 * Adjust the size of the new extent based on di_extsize and rt extsize.
3202 xfs_bmap_extsize_align(
3204 xfs_bmbt_irec_t *gotp, /* next extent pointer */
3205 xfs_bmbt_irec_t *prevp, /* previous extent pointer */
3206 xfs_extlen_t extsz, /* align to this extent size */
3207 int rt, /* is this a realtime inode? */
3208 int eof, /* is extent at end-of-file? */
3209 int delay, /* creating delalloc extent? */
3210 int convert, /* overwriting unwritten extent? */
3211 xfs_fileoff_t *offp, /* in/out: aligned offset */
3212 xfs_extlen_t *lenp) /* in/out: aligned length */
3214 xfs_fileoff_t orig_off; /* original offset */
3215 xfs_extlen_t orig_alen; /* original length */
3216 xfs_fileoff_t orig_end; /* original off+len */
3217 xfs_fileoff_t nexto; /* next file offset */
3218 xfs_fileoff_t prevo; /* previous file offset */
3219 xfs_fileoff_t align_off; /* temp for offset */
3220 xfs_extlen_t align_alen; /* temp for length */
3221 xfs_extlen_t temp; /* temp for calculations */
3226 orig_off = align_off = *offp;
3227 orig_alen = align_alen = *lenp;
3228 orig_end = orig_off + orig_alen;
3231 * If this request overlaps an existing extent, then don't
3232 * attempt to perform any additional alignment.
3234 if (!delay && !eof &&
3235 (orig_off >= gotp->br_startoff) &&
3236 (orig_end <= gotp->br_startoff + gotp->br_blockcount)) {
3241 * If the file offset is unaligned vs. the extent size
3242 * we need to align it. This will be possible unless
3243 * the file was previously written with a kernel that didn't
3244 * perform this alignment, or if a truncate shot us in the
3247 temp = do_mod(orig_off, extsz);
3253 * Same adjustment for the end of the requested area.
3255 if ((temp = (align_alen % extsz))) {
3256 align_alen += extsz - temp;
3259 * If the previous block overlaps with this proposed allocation
3260 * then move the start forward without adjusting the length.
3262 if (prevp->br_startoff != NULLFILEOFF) {
3263 if (prevp->br_startblock == HOLESTARTBLOCK)
3264 prevo = prevp->br_startoff;
3266 prevo = prevp->br_startoff + prevp->br_blockcount;
3269 if (align_off != orig_off && align_off < prevo)
3272 * If the next block overlaps with this proposed allocation
3273 * then move the start back without adjusting the length,
3274 * but not before offset 0.
3275 * This may of course make the start overlap previous block,
3276 * and if we hit the offset 0 limit then the next block
3277 * can still overlap too.
3279 if (!eof && gotp->br_startoff != NULLFILEOFF) {
3280 if ((delay && gotp->br_startblock == HOLESTARTBLOCK) ||
3281 (!delay && gotp->br_startblock == DELAYSTARTBLOCK))
3282 nexto = gotp->br_startoff + gotp->br_blockcount;
3284 nexto = gotp->br_startoff;
3286 nexto = NULLFILEOFF;
3288 align_off + align_alen != orig_end &&
3289 align_off + align_alen > nexto)
3290 align_off = nexto > align_alen ? nexto - align_alen : 0;
3292 * If we're now overlapping the next or previous extent that
3293 * means we can't fit an extsz piece in this hole. Just move
3294 * the start forward to the first valid spot and set
3295 * the length so we hit the end.
3297 if (align_off != orig_off && align_off < prevo)
3299 if (align_off + align_alen != orig_end &&
3300 align_off + align_alen > nexto &&
3301 nexto != NULLFILEOFF) {
3302 ASSERT(nexto > prevo);
3303 align_alen = nexto - align_off;
3307 * If realtime, and the result isn't a multiple of the realtime
3308 * extent size we need to remove blocks until it is.
3310 if (rt && (temp = (align_alen % mp->m_sb.sb_rextsize))) {
3312 * We're not covering the original request, or
3313 * we won't be able to once we fix the length.
3315 if (orig_off < align_off ||
3316 orig_end > align_off + align_alen ||
3317 align_alen - temp < orig_alen)
3318 return XFS_ERROR(EINVAL);
3320 * Try to fix it by moving the start up.
3322 if (align_off + temp <= orig_off) {
3327 * Try to fix it by moving the end in.
3329 else if (align_off + align_alen - temp >= orig_end)
3332 * Set the start to the minimum then trim the length.
3335 align_alen -= orig_off - align_off;
3336 align_off = orig_off;
3337 align_alen -= align_alen % mp->m_sb.sb_rextsize;
3340 * Result doesn't cover the request, fail it.
3342 if (orig_off < align_off || orig_end > align_off + align_alen)
3343 return XFS_ERROR(EINVAL);
3345 ASSERT(orig_off >= align_off);
3346 ASSERT(orig_end <= align_off + align_alen);
3350 if (!eof && gotp->br_startoff != NULLFILEOFF)
3351 ASSERT(align_off + align_alen <= gotp->br_startoff);
3352 if (prevp->br_startoff != NULLFILEOFF)
3353 ASSERT(align_off >= prevp->br_startoff + prevp->br_blockcount);
3361 #define XFS_ALLOC_GAP_UNITS 4
3365 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
3367 xfs_fsblock_t adjust; /* adjustment to block numbers */
3368 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
3369 xfs_mount_t *mp; /* mount point structure */
3370 int nullfb; /* true if ap->firstblock isn't set */
3371 int rt; /* true if inode is realtime */
3373 #define ISVALID(x,y) \
3375 (x) < mp->m_sb.sb_rblocks : \
3376 XFS_FSB_TO_AGNO(mp, x) == XFS_FSB_TO_AGNO(mp, y) && \
3377 XFS_FSB_TO_AGNO(mp, x) < mp->m_sb.sb_agcount && \
3378 XFS_FSB_TO_AGBNO(mp, x) < mp->m_sb.sb_agblocks)
3380 mp = ap->ip->i_mount;
3381 nullfb = *ap->firstblock == NULLFSBLOCK;
3382 rt = XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata;
3383 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3385 * If allocating at eof, and there's a previous real block,
3386 * try to use its last block as our starting point.
3388 if (ap->eof && ap->prev.br_startoff != NULLFILEOFF &&
3389 !isnullstartblock(ap->prev.br_startblock) &&
3390 ISVALID(ap->prev.br_startblock + ap->prev.br_blockcount,
3391 ap->prev.br_startblock)) {
3392 ap->blkno = ap->prev.br_startblock + ap->prev.br_blockcount;
3394 * Adjust for the gap between prevp and us.
3396 adjust = ap->offset -
3397 (ap->prev.br_startoff + ap->prev.br_blockcount);
3399 ISVALID(ap->blkno + adjust, ap->prev.br_startblock))
3400 ap->blkno += adjust;
3403 * If not at eof, then compare the two neighbor blocks.
3404 * Figure out whether either one gives us a good starting point,
3405 * and pick the better one.
3407 else if (!ap->eof) {
3408 xfs_fsblock_t gotbno; /* right side block number */
3409 xfs_fsblock_t gotdiff=0; /* right side difference */
3410 xfs_fsblock_t prevbno; /* left side block number */
3411 xfs_fsblock_t prevdiff=0; /* left side difference */
3414 * If there's a previous (left) block, select a requested
3415 * start block based on it.
3417 if (ap->prev.br_startoff != NULLFILEOFF &&
3418 !isnullstartblock(ap->prev.br_startblock) &&
3419 (prevbno = ap->prev.br_startblock +
3420 ap->prev.br_blockcount) &&
3421 ISVALID(prevbno, ap->prev.br_startblock)) {
3423 * Calculate gap to end of previous block.
3425 adjust = prevdiff = ap->offset -
3426 (ap->prev.br_startoff +
3427 ap->prev.br_blockcount);
3429 * Figure the startblock based on the previous block's
3430 * end and the gap size.
3432 * If the gap is large relative to the piece we're
3433 * allocating, or using it gives us an invalid block
3434 * number, then just use the end of the previous block.
3436 if (prevdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3437 ISVALID(prevbno + prevdiff,
3438 ap->prev.br_startblock))
3443 * If the firstblock forbids it, can't use it,
3446 if (!rt && !nullfb &&
3447 XFS_FSB_TO_AGNO(mp, prevbno) != fb_agno)
3448 prevbno = NULLFSBLOCK;
3451 * No previous block or can't follow it, just default.
3454 prevbno = NULLFSBLOCK;
3456 * If there's a following (right) block, select a requested
3457 * start block based on it.
3459 if (!isnullstartblock(ap->got.br_startblock)) {
3461 * Calculate gap to start of next block.
3463 adjust = gotdiff = ap->got.br_startoff - ap->offset;
3465 * Figure the startblock based on the next block's
3466 * start and the gap size.
3468 gotbno = ap->got.br_startblock;
3471 * If the gap is large relative to the piece we're
3472 * allocating, or using it gives us an invalid block
3473 * number, then just use the start of the next block
3474 * offset by our length.
3476 if (gotdiff <= XFS_ALLOC_GAP_UNITS * ap->length &&
3477 ISVALID(gotbno - gotdiff, gotbno))
3479 else if (ISVALID(gotbno - ap->length, gotbno)) {
3480 gotbno -= ap->length;
3481 gotdiff += adjust - ap->length;
3485 * If the firstblock forbids it, can't use it,
3488 if (!rt && !nullfb &&
3489 XFS_FSB_TO_AGNO(mp, gotbno) != fb_agno)
3490 gotbno = NULLFSBLOCK;
3493 * No next block, just default.
3496 gotbno = NULLFSBLOCK;
3498 * If both valid, pick the better one, else the only good
3499 * one, else ap->blkno is already set (to 0 or the inode block).
3501 if (prevbno != NULLFSBLOCK && gotbno != NULLFSBLOCK)
3502 ap->blkno = prevdiff <= gotdiff ? prevbno : gotbno;
3503 else if (prevbno != NULLFSBLOCK)
3504 ap->blkno = prevbno;
3505 else if (gotbno != NULLFSBLOCK)
3512 xfs_bmap_btalloc_nullfb(
3513 struct xfs_bmalloca *ap,
3514 struct xfs_alloc_arg *args,
3517 struct xfs_mount *mp = ap->ip->i_mount;
3518 struct xfs_perag *pag;
3519 xfs_agnumber_t ag, startag;
3523 if (ap->userdata && xfs_inode_is_filestream(ap->ip))
3524 args->type = XFS_ALLOCTYPE_NEAR_BNO;
3526 args->type = XFS_ALLOCTYPE_START_BNO;
3527 args->total = ap->total;
3530 * Search for an allocation group with a single extent large enough
3531 * for the request. If one isn't found, then adjust the minimum
3532 * allocation size to the largest space found.
3534 startag = ag = XFS_FSB_TO_AGNO(mp, args->fsbno);
3535 if (startag == NULLAGNUMBER)
3538 pag = xfs_perag_get(mp, ag);
3539 while (*blen < args->maxlen) {
3540 if (!pag->pagf_init) {
3541 error = xfs_alloc_pagf_init(mp, args->tp, ag,
3542 XFS_ALLOC_FLAG_TRYLOCK);
3550 * See xfs_alloc_fix_freelist...
3552 if (pag->pagf_init) {
3553 xfs_extlen_t longest;
3554 longest = xfs_alloc_longest_free_extent(mp, pag);
3555 if (*blen < longest)
3560 if (xfs_inode_is_filestream(ap->ip)) {
3561 if (*blen >= args->maxlen)
3566 * If startag is an invalid AG, we've
3567 * come here once before and
3568 * xfs_filestream_new_ag picked the
3569 * best currently available.
3571 * Don't continue looping, since we
3572 * could loop forever.
3574 if (startag == NULLAGNUMBER)
3577 error = xfs_filestream_new_ag(ap, &ag);
3582 /* loop again to set 'blen'*/
3583 startag = NULLAGNUMBER;
3584 pag = xfs_perag_get(mp, ag);
3588 if (++ag == mp->m_sb.sb_agcount)
3593 pag = xfs_perag_get(mp, ag);
3598 * Since the above loop did a BUF_TRYLOCK, it is
3599 * possible that there is space for this request.
3601 if (notinit || *blen < ap->minlen)
3602 args->minlen = ap->minlen;
3604 * If the best seen length is less than the request
3605 * length, use the best as the minimum.
3607 else if (*blen < args->maxlen)
3608 args->minlen = *blen;
3610 * Otherwise we've seen an extent as big as maxlen,
3611 * use that as the minimum.
3614 args->minlen = args->maxlen;
3617 * set the failure fallback case to look in the selected
3618 * AG as the stream may have moved.
3620 if (xfs_inode_is_filestream(ap->ip))
3621 ap->blkno = args->fsbno = XFS_AGB_TO_FSB(mp, ag, 0);
3628 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
3630 xfs_mount_t *mp; /* mount point structure */
3631 xfs_alloctype_t atype = 0; /* type for allocation routines */
3632 xfs_extlen_t align; /* minimum allocation alignment */
3633 xfs_agnumber_t fb_agno; /* ag number of ap->firstblock */
3635 xfs_alloc_arg_t args;
3637 xfs_extlen_t nextminlen = 0;
3638 int nullfb; /* true if ap->firstblock isn't set */
3645 mp = ap->ip->i_mount;
3646 align = ap->userdata ? xfs_get_extsz_hint(ap->ip) : 0;
3647 if (unlikely(align)) {
3648 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
3649 align, 0, ap->eof, 0, ap->conv,
3650 &ap->offset, &ap->length);
3654 nullfb = *ap->firstblock == NULLFSBLOCK;
3655 fb_agno = nullfb ? NULLAGNUMBER : XFS_FSB_TO_AGNO(mp, *ap->firstblock);
3657 if (ap->userdata && xfs_inode_is_filestream(ap->ip)) {
3658 ag = xfs_filestream_lookup_ag(ap->ip);
3659 ag = (ag != NULLAGNUMBER) ? ag : 0;
3660 ap->blkno = XFS_AGB_TO_FSB(mp, ag, 0);
3662 ap->blkno = XFS_INO_TO_FSB(mp, ap->ip->i_ino);
3665 ap->blkno = *ap->firstblock;
3667 xfs_bmap_adjacent(ap);
3670 * If allowed, use ap->blkno; otherwise must use firstblock since
3671 * it's in the right allocation group.
3673 if (nullfb || XFS_FSB_TO_AGNO(mp, ap->blkno) == fb_agno)
3676 ap->blkno = *ap->firstblock;
3678 * Normal allocation, done through xfs_alloc_vextent.
3680 tryagain = isaligned = 0;
3681 memset(&args, 0, sizeof(args));
3684 args.fsbno = ap->blkno;
3686 /* Trim the allocation back to the maximum an AG can fit. */
3687 args.maxlen = MIN(ap->length, XFS_ALLOC_AG_MAX_USABLE(mp));
3688 args.firstblock = *ap->firstblock;
3691 error = xfs_bmap_btalloc_nullfb(ap, &args, &blen);
3694 } else if (ap->flist->xbf_low) {
3695 if (xfs_inode_is_filestream(ap->ip))
3696 args.type = XFS_ALLOCTYPE_FIRST_AG;
3698 args.type = XFS_ALLOCTYPE_START_BNO;
3699 args.total = args.minlen = ap->minlen;
3701 args.type = XFS_ALLOCTYPE_NEAR_BNO;
3702 args.total = ap->total;
3703 args.minlen = ap->minlen;
3705 /* apply extent size hints if obtained earlier */
3706 if (unlikely(align)) {
3708 if ((args.mod = (xfs_extlen_t)do_mod(ap->offset, args.prod)))
3709 args.mod = (xfs_extlen_t)(args.prod - args.mod);
3710 } else if (mp->m_sb.sb_blocksize >= PAGE_CACHE_SIZE) {
3714 args.prod = PAGE_CACHE_SIZE >> mp->m_sb.sb_blocklog;
3715 if ((args.mod = (xfs_extlen_t)(do_mod(ap->offset, args.prod))))
3716 args.mod = (xfs_extlen_t)(args.prod - args.mod);
3719 * If we are not low on available data blocks, and the
3720 * underlying logical volume manager is a stripe, and
3721 * the file offset is zero then try to allocate data
3722 * blocks on stripe unit boundary.
3723 * NOTE: ap->aeof is only set if the allocation length
3724 * is >= the stripe unit and the allocation offset is
3725 * at the end of file.
3727 if (!ap->flist->xbf_low && ap->aeof) {
3729 args.alignment = mp->m_dalign;
3733 * Adjust for alignment
3735 if (blen > args.alignment && blen <= args.maxlen)
3736 args.minlen = blen - args.alignment;
3737 args.minalignslop = 0;
3740 * First try an exact bno allocation.
3741 * If it fails then do a near or start bno
3742 * allocation with alignment turned on.
3746 args.type = XFS_ALLOCTYPE_THIS_BNO;
3749 * Compute the minlen+alignment for the
3750 * next case. Set slop so that the value
3751 * of minlen+alignment+slop doesn't go up
3752 * between the calls.
3754 if (blen > mp->m_dalign && blen <= args.maxlen)
3755 nextminlen = blen - mp->m_dalign;
3757 nextminlen = args.minlen;
3758 if (nextminlen + mp->m_dalign > args.minlen + 1)
3760 nextminlen + mp->m_dalign -
3763 args.minalignslop = 0;
3767 args.minalignslop = 0;
3769 args.minleft = ap->minleft;
3770 args.wasdel = ap->wasdel;
3772 args.userdata = ap->userdata;
3773 if ((error = xfs_alloc_vextent(&args)))
3775 if (tryagain && args.fsbno == NULLFSBLOCK) {
3777 * Exact allocation failed. Now try with alignment
3781 args.fsbno = ap->blkno;
3782 args.alignment = mp->m_dalign;
3783 args.minlen = nextminlen;
3784 args.minalignslop = 0;
3786 if ((error = xfs_alloc_vextent(&args)))
3789 if (isaligned && args.fsbno == NULLFSBLOCK) {
3791 * allocation failed, so turn off alignment and
3795 args.fsbno = ap->blkno;
3797 if ((error = xfs_alloc_vextent(&args)))
3800 if (args.fsbno == NULLFSBLOCK && nullfb &&
3801 args.minlen > ap->minlen) {
3802 args.minlen = ap->minlen;
3803 args.type = XFS_ALLOCTYPE_START_BNO;
3804 args.fsbno = ap->blkno;
3805 if ((error = xfs_alloc_vextent(&args)))
3808 if (args.fsbno == NULLFSBLOCK && nullfb) {
3810 args.type = XFS_ALLOCTYPE_FIRST_AG;
3811 args.total = ap->minlen;
3813 if ((error = xfs_alloc_vextent(&args)))
3815 ap->flist->xbf_low = 1;
3817 if (args.fsbno != NULLFSBLOCK) {
3819 * check the allocation happened at the same or higher AG than
3820 * the first block that was allocated.
3822 ASSERT(*ap->firstblock == NULLFSBLOCK ||
3823 XFS_FSB_TO_AGNO(mp, *ap->firstblock) ==
3824 XFS_FSB_TO_AGNO(mp, args.fsbno) ||
3825 (ap->flist->xbf_low &&
3826 XFS_FSB_TO_AGNO(mp, *ap->firstblock) <
3827 XFS_FSB_TO_AGNO(mp, args.fsbno)));
3829 ap->blkno = args.fsbno;
3830 if (*ap->firstblock == NULLFSBLOCK)
3831 *ap->firstblock = args.fsbno;
3832 ASSERT(nullfb || fb_agno == args.agno ||
3833 (ap->flist->xbf_low && fb_agno < args.agno));
3834 ap->length = args.len;
3835 ap->ip->i_d.di_nblocks += args.len;
3836 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
3838 ap->ip->i_delayed_blks -= args.len;
3840 * Adjust the disk quota also. This was reserved
3843 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
3844 ap->wasdel ? XFS_TRANS_DQ_DELBCOUNT :
3845 XFS_TRANS_DQ_BCOUNT,
3848 ap->blkno = NULLFSBLOCK;
3855 * xfs_bmap_alloc is called by xfs_bmapi to allocate an extent for a file.
3856 * It figures out where to ask the underlying allocator to put the new extent.
3860 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
3862 if (XFS_IS_REALTIME_INODE(ap->ip) && ap->userdata)
3863 return xfs_bmap_rtalloc(ap);
3864 return xfs_bmap_btalloc(ap);
3868 * Trim the returned map to the required bounds
3872 struct xfs_bmbt_irec *mval,
3873 struct xfs_bmbt_irec *got,
3881 if ((flags & XFS_BMAPI_ENTIRE) ||
3882 got->br_startoff + got->br_blockcount <= obno) {
3884 if (isnullstartblock(got->br_startblock))
3885 mval->br_startblock = DELAYSTARTBLOCK;
3891 ASSERT((*bno >= obno) || (n == 0));
3893 mval->br_startoff = *bno;
3894 if (isnullstartblock(got->br_startblock))
3895 mval->br_startblock = DELAYSTARTBLOCK;
3897 mval->br_startblock = got->br_startblock +
3898 (*bno - got->br_startoff);
3900 * Return the minimum of what we got and what we asked for for
3901 * the length. We can use the len variable here because it is
3902 * modified below and we could have been there before coming
3903 * here if the first part of the allocation didn't overlap what
3906 mval->br_blockcount = XFS_FILBLKS_MIN(end - *bno,
3907 got->br_blockcount - (*bno - got->br_startoff));
3908 mval->br_state = got->br_state;
3909 ASSERT(mval->br_blockcount <= len);
3914 * Update and validate the extent map to return
3917 xfs_bmapi_update_map(
3918 struct xfs_bmbt_irec **map,
3926 xfs_bmbt_irec_t *mval = *map;
3928 ASSERT((flags & XFS_BMAPI_ENTIRE) ||
3929 ((mval->br_startoff + mval->br_blockcount) <= end));
3930 ASSERT((flags & XFS_BMAPI_ENTIRE) || (mval->br_blockcount <= *len) ||
3931 (mval->br_startoff < obno));
3933 *bno = mval->br_startoff + mval->br_blockcount;
3935 if (*n > 0 && mval->br_startoff == mval[-1].br_startoff) {
3936 /* update previous map with new information */
3937 ASSERT(mval->br_startblock == mval[-1].br_startblock);
3938 ASSERT(mval->br_blockcount > mval[-1].br_blockcount);
3939 ASSERT(mval->br_state == mval[-1].br_state);
3940 mval[-1].br_blockcount = mval->br_blockcount;
3941 mval[-1].br_state = mval->br_state;
3942 } else if (*n > 0 && mval->br_startblock != DELAYSTARTBLOCK &&
3943 mval[-1].br_startblock != DELAYSTARTBLOCK &&
3944 mval[-1].br_startblock != HOLESTARTBLOCK &&
3945 mval->br_startblock == mval[-1].br_startblock +
3946 mval[-1].br_blockcount &&
3947 ((flags & XFS_BMAPI_IGSTATE) ||
3948 mval[-1].br_state == mval->br_state)) {
3949 ASSERT(mval->br_startoff ==
3950 mval[-1].br_startoff + mval[-1].br_blockcount);
3951 mval[-1].br_blockcount += mval->br_blockcount;
3952 } else if (*n > 0 &&
3953 mval->br_startblock == DELAYSTARTBLOCK &&
3954 mval[-1].br_startblock == DELAYSTARTBLOCK &&
3955 mval->br_startoff ==
3956 mval[-1].br_startoff + mval[-1].br_blockcount) {
3957 mval[-1].br_blockcount += mval->br_blockcount;
3958 mval[-1].br_state = mval->br_state;
3959 } else if (!((*n == 0) &&
3960 ((mval->br_startoff + mval->br_blockcount) <=
3969 * Map file blocks to filesystem blocks without allocation.
3973 struct xfs_inode *ip,
3976 struct xfs_bmbt_irec *mval,
3980 struct xfs_mount *mp = ip->i_mount;
3981 struct xfs_ifork *ifp;
3982 struct xfs_bmbt_irec got;
3983 struct xfs_bmbt_irec prev;
3990 int whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
3991 XFS_ATTR_FORK : XFS_DATA_FORK;
3994 ASSERT(!(flags & ~(XFS_BMAPI_ATTRFORK|XFS_BMAPI_ENTIRE|
3995 XFS_BMAPI_IGSTATE)));
3997 if (unlikely(XFS_TEST_ERROR(
3998 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
3999 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4000 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4001 XFS_ERROR_REPORT("xfs_bmapi_read", XFS_ERRLEVEL_LOW, mp);
4002 return XFS_ERROR(EFSCORRUPTED);
4005 if (XFS_FORCED_SHUTDOWN(mp))
4006 return XFS_ERROR(EIO);
4008 XFS_STATS_INC(xs_blk_mapr);
4010 ifp = XFS_IFORK_PTR(ip, whichfork);
4012 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4013 error = xfs_iread_extents(NULL, ip, whichfork);
4018 xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got, &prev);
4022 while (bno < end && n < *nmap) {
4023 /* Reading past eof, act as though there's a hole up to end. */
4025 got.br_startoff = end;
4026 if (got.br_startoff > bno) {
4027 /* Reading in a hole. */
4028 mval->br_startoff = bno;
4029 mval->br_startblock = HOLESTARTBLOCK;
4030 mval->br_blockcount =
4031 XFS_FILBLKS_MIN(len, got.br_startoff - bno);
4032 mval->br_state = XFS_EXT_NORM;
4033 bno += mval->br_blockcount;
4034 len -= mval->br_blockcount;
4040 /* set up the extent map to return. */
4041 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4042 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4044 /* If we're done, stop now. */
4045 if (bno >= end || n >= *nmap)
4048 /* Else go on to the next record. */
4049 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4050 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4059 xfs_bmapi_reserve_delalloc(
4060 struct xfs_inode *ip,
4063 struct xfs_bmbt_irec *got,
4064 struct xfs_bmbt_irec *prev,
4065 xfs_extnum_t *lastx,
4068 struct xfs_mount *mp = ip->i_mount;
4069 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4071 xfs_extlen_t indlen;
4072 char rt = XFS_IS_REALTIME_INODE(ip);
4076 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN);
4078 alen = XFS_FILBLKS_MIN(alen, got->br_startoff - aoff);
4080 /* Figure out the extent size, adjust alen */
4081 extsz = xfs_get_extsz_hint(ip);
4084 * Make sure we don't exceed a single extent length when we
4085 * align the extent by reducing length we are going to
4086 * allocate by the maximum amount extent size aligment may
4089 alen = XFS_FILBLKS_MIN(len, MAXEXTLEN - (2 * extsz - 1));
4090 error = xfs_bmap_extsize_align(mp, got, prev, extsz, rt, eof,
4091 1, 0, &aoff, &alen);
4096 extsz = alen / mp->m_sb.sb_rextsize;
4099 * Make a transaction-less quota reservation for delayed allocation
4100 * blocks. This number gets adjusted later. We return if we haven't
4101 * allocated blocks already inside this loop.
4103 error = xfs_trans_reserve_quota_nblks(NULL, ip, (long)alen, 0,
4104 rt ? XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4109 * Split changing sb for alen and indlen since they could be coming
4110 * from different places.
4112 indlen = (xfs_extlen_t)xfs_bmap_worst_indlen(ip, alen);
4116 error = xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
4117 -((int64_t)extsz), 0);
4119 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4120 -((int64_t)alen), 0);
4124 goto out_unreserve_quota;
4126 error = xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4127 -((int64_t)indlen), 0);
4129 goto out_unreserve_blocks;
4132 ip->i_delayed_blks += alen;
4134 got->br_startoff = aoff;
4135 got->br_startblock = nullstartblock(indlen);
4136 got->br_blockcount = alen;
4137 got->br_state = XFS_EXT_NORM;
4138 xfs_bmap_add_extent_hole_delay(ip, lastx, got);
4141 * Update our extent pointer, given that xfs_bmap_add_extent_hole_delay
4142 * might have merged it into one of the neighbouring ones.
4144 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, *lastx), got);
4146 ASSERT(got->br_startoff <= aoff);
4147 ASSERT(got->br_startoff + got->br_blockcount >= aoff + alen);
4148 ASSERT(isnullstartblock(got->br_startblock));
4149 ASSERT(got->br_state == XFS_EXT_NORM);
4152 out_unreserve_blocks:
4154 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS, extsz, 0);
4156 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS, alen, 0);
4157 out_unreserve_quota:
4158 if (XFS_IS_QUOTA_ON(mp))
4159 xfs_trans_unreserve_quota_nblks(NULL, ip, (long)alen, 0, rt ?
4160 XFS_QMOPT_RES_RTBLKS : XFS_QMOPT_RES_REGBLKS);
4165 * Map file blocks to filesystem blocks, adding delayed allocations as needed.
4169 struct xfs_inode *ip, /* incore inode */
4170 xfs_fileoff_t bno, /* starting file offs. mapped */
4171 xfs_filblks_t len, /* length to map in file */
4172 struct xfs_bmbt_irec *mval, /* output: map values */
4173 int *nmap, /* i/o: mval size/count */
4174 int flags) /* XFS_BMAPI_... */
4176 struct xfs_mount *mp = ip->i_mount;
4177 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
4178 struct xfs_bmbt_irec got; /* current file extent record */
4179 struct xfs_bmbt_irec prev; /* previous file extent record */
4180 xfs_fileoff_t obno; /* old block number (offset) */
4181 xfs_fileoff_t end; /* end of mapped file region */
4182 xfs_extnum_t lastx; /* last useful extent number */
4183 int eof; /* we've hit the end of extents */
4184 int n = 0; /* current extent index */
4188 ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4189 ASSERT(!(flags & ~XFS_BMAPI_ENTIRE));
4191 if (unlikely(XFS_TEST_ERROR(
4192 (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
4193 XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
4194 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4195 XFS_ERROR_REPORT("xfs_bmapi_delay", XFS_ERRLEVEL_LOW, mp);
4196 return XFS_ERROR(EFSCORRUPTED);
4199 if (XFS_FORCED_SHUTDOWN(mp))
4200 return XFS_ERROR(EIO);
4202 XFS_STATS_INC(xs_blk_mapw);
4204 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4205 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
4210 xfs_bmap_search_extents(ip, bno, XFS_DATA_FORK, &eof, &lastx, &got, &prev);
4214 while (bno < end && n < *nmap) {
4215 if (eof || got.br_startoff > bno) {
4216 error = xfs_bmapi_reserve_delalloc(ip, bno, len, &got,
4217 &prev, &lastx, eof);
4227 /* set up the extent map to return. */
4228 xfs_bmapi_trim_map(mval, &got, &bno, len, obno, end, n, flags);
4229 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4231 /* If we're done, stop now. */
4232 if (bno >= end || n >= *nmap)
4235 /* Else go on to the next record. */
4237 if (++lastx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t))
4238 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, lastx), &got);
4249 __xfs_bmapi_allocate(
4250 struct xfs_bmalloca *bma)
4252 struct xfs_mount *mp = bma->ip->i_mount;
4253 int whichfork = (bma->flags & XFS_BMAPI_ATTRFORK) ?
4254 XFS_ATTR_FORK : XFS_DATA_FORK;
4255 struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4256 int tmp_logflags = 0;
4259 ASSERT(bma->length > 0);
4262 * For the wasdelay case, we could also just allocate the stuff asked
4263 * for in this bmap call but that wouldn't be as good.
4266 bma->length = (xfs_extlen_t)bma->got.br_blockcount;
4267 bma->offset = bma->got.br_startoff;
4268 if (bma->idx != NULLEXTNUM && bma->idx) {
4269 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx - 1),
4273 bma->length = XFS_FILBLKS_MIN(bma->length, MAXEXTLEN);
4275 bma->length = XFS_FILBLKS_MIN(bma->length,
4276 bma->got.br_startoff - bma->offset);
4280 * Indicate if this is the first user data in the file, or just any
4283 if (!(bma->flags & XFS_BMAPI_METADATA)) {
4284 bma->userdata = (bma->offset == 0) ?
4285 XFS_ALLOC_INITIAL_USER_DATA : XFS_ALLOC_USERDATA;
4288 bma->minlen = (bma->flags & XFS_BMAPI_CONTIG) ? bma->length : 1;
4291 * Only want to do the alignment at the eof if it is userdata and
4292 * allocation length is larger than a stripe unit.
4294 if (mp->m_dalign && bma->length >= mp->m_dalign &&
4295 !(bma->flags & XFS_BMAPI_METADATA) && whichfork == XFS_DATA_FORK) {
4296 error = xfs_bmap_isaeof(bma, whichfork);
4301 error = xfs_bmap_alloc(bma);
4305 if (bma->flist->xbf_low)
4308 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4309 if (bma->blkno == NULLFSBLOCK)
4311 if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4312 bma->cur = xfs_bmbt_init_cursor(mp, bma->tp, bma->ip, whichfork);
4313 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4314 bma->cur->bc_private.b.flist = bma->flist;
4317 * Bump the number of extents we've allocated
4323 bma->cur->bc_private.b.flags =
4324 bma->wasdel ? XFS_BTCUR_BPRV_WASDEL : 0;
4326 bma->got.br_startoff = bma->offset;
4327 bma->got.br_startblock = bma->blkno;
4328 bma->got.br_blockcount = bma->length;
4329 bma->got.br_state = XFS_EXT_NORM;
4332 * A wasdelay extent has been initialized, so shouldn't be flagged
4335 if (!bma->wasdel && (bma->flags & XFS_BMAPI_PREALLOC) &&
4336 xfs_sb_version_hasextflgbit(&mp->m_sb))
4337 bma->got.br_state = XFS_EXT_UNWRITTEN;
4340 error = xfs_bmap_add_extent_delay_real(bma);
4342 error = xfs_bmap_add_extent_hole_real(bma, whichfork);
4344 bma->logflags |= tmp_logflags;
4349 * Update our extent pointer, given that xfs_bmap_add_extent_delay_real
4350 * or xfs_bmap_add_extent_hole_real might have merged it into one of
4351 * the neighbouring ones.
4353 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4355 ASSERT(bma->got.br_startoff <= bma->offset);
4356 ASSERT(bma->got.br_startoff + bma->got.br_blockcount >=
4357 bma->offset + bma->length);
4358 ASSERT(bma->got.br_state == XFS_EXT_NORM ||
4359 bma->got.br_state == XFS_EXT_UNWRITTEN);
4364 xfs_bmapi_convert_unwritten(
4365 struct xfs_bmalloca *bma,
4366 struct xfs_bmbt_irec *mval,
4370 int whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4371 XFS_ATTR_FORK : XFS_DATA_FORK;
4372 struct xfs_ifork *ifp = XFS_IFORK_PTR(bma->ip, whichfork);
4373 int tmp_logflags = 0;
4376 /* check if we need to do unwritten->real conversion */
4377 if (mval->br_state == XFS_EXT_UNWRITTEN &&
4378 (flags & XFS_BMAPI_PREALLOC))
4381 /* check if we need to do real->unwritten conversion */
4382 if (mval->br_state == XFS_EXT_NORM &&
4383 (flags & (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT)) !=
4384 (XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT))
4388 * Modify (by adding) the state flag, if writing.
4390 ASSERT(mval->br_blockcount <= len);
4391 if ((ifp->if_flags & XFS_IFBROOT) && !bma->cur) {
4392 bma->cur = xfs_bmbt_init_cursor(bma->ip->i_mount, bma->tp,
4393 bma->ip, whichfork);
4394 bma->cur->bc_private.b.firstblock = *bma->firstblock;
4395 bma->cur->bc_private.b.flist = bma->flist;
4397 mval->br_state = (mval->br_state == XFS_EXT_UNWRITTEN)
4398 ? XFS_EXT_NORM : XFS_EXT_UNWRITTEN;
4400 error = xfs_bmap_add_extent_unwritten_real(bma->tp, bma->ip, &bma->idx,
4401 &bma->cur, mval, bma->firstblock, bma->flist,
4403 bma->logflags |= tmp_logflags;
4408 * Update our extent pointer, given that
4409 * xfs_bmap_add_extent_unwritten_real might have merged it into one
4410 * of the neighbouring ones.
4412 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma->idx), &bma->got);
4415 * We may have combined previously unwritten space with written space,
4416 * so generate another request.
4418 if (mval->br_blockcount < len)
4424 * Map file blocks to filesystem blocks, and allocate blocks or convert the
4425 * extent state if necessary. Details behaviour is controlled by the flags
4426 * parameter. Only allocates blocks from a single allocation group, to avoid
4429 * The returned value in "firstblock" from the first call in a transaction
4430 * must be remembered and presented to subsequent calls in "firstblock".
4431 * An upper bound for the number of blocks to be allocated is supplied to
4432 * the first call in "total"; if no allocation group has that many free
4433 * blocks then the call will fail (return NULLFSBLOCK in "firstblock").
4437 struct xfs_trans *tp, /* transaction pointer */
4438 struct xfs_inode *ip, /* incore inode */
4439 xfs_fileoff_t bno, /* starting file offs. mapped */
4440 xfs_filblks_t len, /* length to map in file */
4441 int flags, /* XFS_BMAPI_... */
4442 xfs_fsblock_t *firstblock, /* first allocated block
4443 controls a.g. for allocs */
4444 xfs_extlen_t total, /* total blocks needed */
4445 struct xfs_bmbt_irec *mval, /* output: map values */
4446 int *nmap, /* i/o: mval size/count */
4447 struct xfs_bmap_free *flist) /* i/o: list extents to free */
4449 struct xfs_mount *mp = ip->i_mount;
4450 struct xfs_ifork *ifp;
4451 struct xfs_bmalloca bma = { NULL }; /* args for xfs_bmap_alloc */
4452 xfs_fileoff_t end; /* end of mapped file region */
4453 int eof; /* after the end of extents */
4454 int error; /* error return */
4455 int n; /* current extent index */
4456 xfs_fileoff_t obno; /* old block number (offset) */
4457 int whichfork; /* data or attr fork */
4458 char inhole; /* current location is hole in file */
4459 char wasdelay; /* old extent was delayed */
4462 xfs_fileoff_t orig_bno; /* original block number value */
4463 int orig_flags; /* original flags arg value */
4464 xfs_filblks_t orig_len; /* original value of len arg */
4465 struct xfs_bmbt_irec *orig_mval; /* original value of mval */
4466 int orig_nmap; /* original value of *nmap */
4474 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
4475 XFS_ATTR_FORK : XFS_DATA_FORK;
4478 ASSERT(*nmap <= XFS_BMAP_MAX_NMAP);
4479 ASSERT(!(flags & XFS_BMAPI_IGSTATE));
4482 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_LOCAL);
4484 if (unlikely(XFS_TEST_ERROR(
4485 (XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
4486 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE),
4487 mp, XFS_ERRTAG_BMAPIFORMAT, XFS_RANDOM_BMAPIFORMAT))) {
4488 XFS_ERROR_REPORT("xfs_bmapi_write", XFS_ERRLEVEL_LOW, mp);
4489 return XFS_ERROR(EFSCORRUPTED);
4492 if (XFS_FORCED_SHUTDOWN(mp))
4493 return XFS_ERROR(EIO);
4495 ifp = XFS_IFORK_PTR(ip, whichfork);
4497 XFS_STATS_INC(xs_blk_mapw);
4499 if (*firstblock == NULLFSBLOCK) {
4500 if (XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE)
4501 bma.minleft = be16_to_cpu(ifp->if_broot->bb_level) + 1;
4508 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
4509 error = xfs_iread_extents(tp, ip, whichfork);
4514 xfs_bmap_search_extents(ip, bno, whichfork, &eof, &bma.idx, &bma.got,
4525 bma.firstblock = firstblock;
4527 if (flags & XFS_BMAPI_STACK_SWITCH)
4528 bma.stack_switch = 1;
4530 while (bno < end && n < *nmap) {
4531 inhole = eof || bma.got.br_startoff > bno;
4532 wasdelay = !inhole && isnullstartblock(bma.got.br_startblock);
4535 * First, deal with the hole before the allocated space
4536 * that we found, if any.
4538 if (inhole || wasdelay) {
4540 bma.conv = !!(flags & XFS_BMAPI_CONVERT);
4541 bma.wasdel = wasdelay;
4546 * There's a 32/64 bit type mismatch between the
4547 * allocation length request (which can be 64 bits in
4548 * length) and the bma length request, which is
4549 * xfs_extlen_t and therefore 32 bits. Hence we have to
4550 * check for 32-bit overflows and handle them here.
4552 if (len > (xfs_filblks_t)MAXEXTLEN)
4553 bma.length = MAXEXTLEN;
4558 ASSERT(bma.length > 0);
4559 error = xfs_bmapi_allocate(&bma);
4562 if (bma.blkno == NULLFSBLOCK)
4566 /* Deal with the allocated space we found. */
4567 xfs_bmapi_trim_map(mval, &bma.got, &bno, len, obno,
4570 /* Execute unwritten extent conversion if necessary */
4571 error = xfs_bmapi_convert_unwritten(&bma, mval, len, flags);
4572 if (error == EAGAIN)
4577 /* update the extent map to return */
4578 xfs_bmapi_update_map(&mval, &bno, &len, obno, end, &n, flags);
4581 * If we're done, stop now. Stop when we've allocated
4582 * XFS_BMAP_MAX_NMAP extents no matter what. Otherwise
4583 * the transaction may get too big.
4585 if (bno >= end || n >= *nmap || bma.nallocs >= *nmap)
4588 /* Else go on to the next record. */
4590 if (++bma.idx < ifp->if_bytes / sizeof(xfs_bmbt_rec_t)) {
4591 xfs_bmbt_get_all(xfs_iext_get_ext(ifp, bma.idx),
4599 * Transform from btree to extents, give it cur.
4601 if (xfs_bmap_wants_extents(ip, whichfork)) {
4602 int tmp_logflags = 0;
4605 error = xfs_bmap_btree_to_extents(tp, ip, bma.cur,
4606 &tmp_logflags, whichfork);
4607 bma.logflags |= tmp_logflags;
4612 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE ||
4613 XFS_IFORK_NEXTENTS(ip, whichfork) >
4614 XFS_IFORK_MAXEXT(ip, whichfork));
4618 * Log everything. Do this after conversion, there's no point in
4619 * logging the extent records if we've converted to btree format.
4621 if ((bma.logflags & xfs_ilog_fext(whichfork)) &&
4622 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
4623 bma.logflags &= ~xfs_ilog_fext(whichfork);
4624 else if ((bma.logflags & xfs_ilog_fbroot(whichfork)) &&
4625 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
4626 bma.logflags &= ~xfs_ilog_fbroot(whichfork);
4628 * Log whatever the flags say, even if error. Otherwise we might miss
4629 * detecting a case where the data is changed, there's an error,
4630 * and it's not logged so we don't shutdown when we should.
4633 xfs_trans_log_inode(tp, ip, bma.logflags);
4637 ASSERT(*firstblock == NULLFSBLOCK ||
4638 XFS_FSB_TO_AGNO(mp, *firstblock) ==
4640 bma.cur->bc_private.b.firstblock) ||
4642 XFS_FSB_TO_AGNO(mp, *firstblock) <
4644 bma.cur->bc_private.b.firstblock)));
4645 *firstblock = bma.cur->bc_private.b.firstblock;
4647 xfs_btree_del_cursor(bma.cur,
4648 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);
4651 xfs_bmap_validate_ret(orig_bno, orig_len, orig_flags, orig_mval,
4657 * Called by xfs_bmapi to update file extent records and the btree
4658 * after removing space (or undoing a delayed allocation).
4660 STATIC int /* error */
4661 xfs_bmap_del_extent(
4662 xfs_inode_t *ip, /* incore inode pointer */
4663 xfs_trans_t *tp, /* current transaction pointer */
4664 xfs_extnum_t *idx, /* extent number to update/delete */
4665 xfs_bmap_free_t *flist, /* list of extents to be freed */
4666 xfs_btree_cur_t *cur, /* if null, not a btree */
4667 xfs_bmbt_irec_t *del, /* data to remove from extents */
4668 int *logflagsp, /* inode logging flags */
4669 int whichfork) /* data or attr fork */
4671 xfs_filblks_t da_new; /* new delay-alloc indirect blocks */
4672 xfs_filblks_t da_old; /* old delay-alloc indirect blocks */
4673 xfs_fsblock_t del_endblock=0; /* first block past del */
4674 xfs_fileoff_t del_endoff; /* first offset past del */
4675 int delay; /* current block is delayed allocated */
4676 int do_fx; /* free extent at end of routine */
4677 xfs_bmbt_rec_host_t *ep; /* current extent entry pointer */
4678 int error; /* error return value */
4679 int flags; /* inode logging flags */
4680 xfs_bmbt_irec_t got; /* current extent entry */
4681 xfs_fileoff_t got_endoff; /* first offset past got */
4682 int i; /* temp state */
4683 xfs_ifork_t *ifp; /* inode fork pointer */
4684 xfs_mount_t *mp; /* mount structure */
4685 xfs_filblks_t nblks; /* quota/sb block count */
4686 xfs_bmbt_irec_t new; /* new record to be inserted */
4688 uint qfield; /* quota field to update */
4689 xfs_filblks_t temp; /* for indirect length calculations */
4690 xfs_filblks_t temp2; /* for indirect length calculations */
4693 XFS_STATS_INC(xs_del_exlist);
4695 if (whichfork == XFS_ATTR_FORK)
4696 state |= BMAP_ATTRFORK;
4699 ifp = XFS_IFORK_PTR(ip, whichfork);
4700 ASSERT((*idx >= 0) && (*idx < ifp->if_bytes /
4701 (uint)sizeof(xfs_bmbt_rec_t)));
4702 ASSERT(del->br_blockcount > 0);
4703 ep = xfs_iext_get_ext(ifp, *idx);
4704 xfs_bmbt_get_all(ep, &got);
4705 ASSERT(got.br_startoff <= del->br_startoff);
4706 del_endoff = del->br_startoff + del->br_blockcount;
4707 got_endoff = got.br_startoff + got.br_blockcount;
4708 ASSERT(got_endoff >= del_endoff);
4709 delay = isnullstartblock(got.br_startblock);
4710 ASSERT(isnullstartblock(del->br_startblock) == delay);
4715 * If deleting a real allocation, must free up the disk space.
4718 flags = XFS_ILOG_CORE;
4720 * Realtime allocation. Free it and record di_nblocks update.
4722 if (whichfork == XFS_DATA_FORK && XFS_IS_REALTIME_INODE(ip)) {
4726 ASSERT(do_mod(del->br_blockcount,
4727 mp->m_sb.sb_rextsize) == 0);
4728 ASSERT(do_mod(del->br_startblock,
4729 mp->m_sb.sb_rextsize) == 0);
4730 bno = del->br_startblock;
4731 len = del->br_blockcount;
4732 do_div(bno, mp->m_sb.sb_rextsize);
4733 do_div(len, mp->m_sb.sb_rextsize);
4734 error = xfs_rtfree_extent(tp, bno, (xfs_extlen_t)len);
4738 nblks = len * mp->m_sb.sb_rextsize;
4739 qfield = XFS_TRANS_DQ_RTBCOUNT;
4742 * Ordinary allocation.
4746 nblks = del->br_blockcount;
4747 qfield = XFS_TRANS_DQ_BCOUNT;
4750 * Set up del_endblock and cur for later.
4752 del_endblock = del->br_startblock + del->br_blockcount;
4754 if ((error = xfs_bmbt_lookup_eq(cur, got.br_startoff,
4755 got.br_startblock, got.br_blockcount,
4758 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4760 da_old = da_new = 0;
4762 da_old = startblockval(got.br_startblock);
4768 * Set flag value to use in switch statement.
4769 * Left-contig is 2, right-contig is 1.
4771 switch (((got.br_startoff == del->br_startoff) << 1) |
4772 (got_endoff == del_endoff)) {
4775 * Matches the whole extent. Delete the entry.
4777 xfs_iext_remove(ip, *idx, 1,
4778 whichfork == XFS_ATTR_FORK ? BMAP_ATTRFORK : 0);
4783 XFS_IFORK_NEXT_SET(ip, whichfork,
4784 XFS_IFORK_NEXTENTS(ip, whichfork) - 1);
4785 flags |= XFS_ILOG_CORE;
4787 flags |= xfs_ilog_fext(whichfork);
4790 if ((error = xfs_btree_delete(cur, &i)))
4792 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4797 * Deleting the first part of the extent.
4799 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4800 xfs_bmbt_set_startoff(ep, del_endoff);
4801 temp = got.br_blockcount - del->br_blockcount;
4802 xfs_bmbt_set_blockcount(ep, temp);
4804 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4806 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4807 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4811 xfs_bmbt_set_startblock(ep, del_endblock);
4812 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4814 flags |= xfs_ilog_fext(whichfork);
4817 if ((error = xfs_bmbt_update(cur, del_endoff, del_endblock,
4818 got.br_blockcount - del->br_blockcount,
4825 * Deleting the last part of the extent.
4827 temp = got.br_blockcount - del->br_blockcount;
4828 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4829 xfs_bmbt_set_blockcount(ep, temp);
4831 temp = XFS_FILBLKS_MIN(xfs_bmap_worst_indlen(ip, temp),
4833 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4834 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4838 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4840 flags |= xfs_ilog_fext(whichfork);
4843 if ((error = xfs_bmbt_update(cur, got.br_startoff,
4845 got.br_blockcount - del->br_blockcount,
4852 * Deleting the middle of the extent.
4854 temp = del->br_startoff - got.br_startoff;
4855 trace_xfs_bmap_pre_update(ip, *idx, state, _THIS_IP_);
4856 xfs_bmbt_set_blockcount(ep, temp);
4857 new.br_startoff = del_endoff;
4858 temp2 = got_endoff - del_endoff;
4859 new.br_blockcount = temp2;
4860 new.br_state = got.br_state;
4862 new.br_startblock = del_endblock;
4863 flags |= XFS_ILOG_CORE;
4865 if ((error = xfs_bmbt_update(cur,
4867 got.br_startblock, temp,
4870 if ((error = xfs_btree_increment(cur, 0, &i)))
4872 cur->bc_rec.b = new;
4873 error = xfs_btree_insert(cur, &i);
4874 if (error && error != ENOSPC)
4877 * If get no-space back from btree insert,
4878 * it tried a split, and we have a zero
4879 * block reservation.
4880 * Fix up our state and return the error.
4882 if (error == ENOSPC) {
4884 * Reset the cursor, don't trust
4885 * it after any insert operation.
4887 if ((error = xfs_bmbt_lookup_eq(cur,
4892 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4894 * Update the btree record back
4895 * to the original value.
4897 if ((error = xfs_bmbt_update(cur,
4904 * Reset the extent record back
4905 * to the original value.
4907 xfs_bmbt_set_blockcount(ep,
4910 error = XFS_ERROR(ENOSPC);
4913 XFS_WANT_CORRUPTED_GOTO(i == 1, done);
4915 flags |= xfs_ilog_fext(whichfork);
4916 XFS_IFORK_NEXT_SET(ip, whichfork,
4917 XFS_IFORK_NEXTENTS(ip, whichfork) + 1);
4919 ASSERT(whichfork == XFS_DATA_FORK);
4920 temp = xfs_bmap_worst_indlen(ip, temp);
4921 xfs_bmbt_set_startblock(ep, nullstartblock((int)temp));
4922 temp2 = xfs_bmap_worst_indlen(ip, temp2);
4923 new.br_startblock = nullstartblock((int)temp2);
4924 da_new = temp + temp2;
4925 while (da_new > da_old) {
4929 xfs_bmbt_set_startblock(ep,
4930 nullstartblock((int)temp));
4932 if (da_new == da_old)
4938 nullstartblock((int)temp2);
4942 trace_xfs_bmap_post_update(ip, *idx, state, _THIS_IP_);
4943 xfs_iext_insert(ip, *idx + 1, 1, &new, state);
4948 * If we need to, add to list of extents to delete.
4951 xfs_bmap_add_free(del->br_startblock, del->br_blockcount, flist,
4954 * Adjust inode # blocks in the file.
4957 ip->i_d.di_nblocks -= nblks;
4959 * Adjust quota data.
4962 xfs_trans_mod_dquot_byino(tp, ip, qfield, (long)-nblks);
4965 * Account for change in delayed indirect blocks.
4966 * Nothing to do for disk quota accounting here.
4968 ASSERT(da_old >= da_new);
4969 if (da_old > da_new) {
4970 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
4971 (int64_t)(da_old - da_new), 0);
4979 * Unmap (remove) blocks from a file.
4980 * If nexts is nonzero then the number of extents to remove is limited to
4981 * that value. If not all extents in the block range can be removed then
4986 xfs_trans_t *tp, /* transaction pointer */
4987 struct xfs_inode *ip, /* incore inode */
4988 xfs_fileoff_t bno, /* starting offset to unmap */
4989 xfs_filblks_t len, /* length to unmap in file */
4990 int flags, /* misc flags */
4991 xfs_extnum_t nexts, /* number of extents max */
4992 xfs_fsblock_t *firstblock, /* first allocated block
4993 controls a.g. for allocs */
4994 xfs_bmap_free_t *flist, /* i/o: list extents to free */
4995 int *done) /* set if not done yet */
4997 xfs_btree_cur_t *cur; /* bmap btree cursor */
4998 xfs_bmbt_irec_t del; /* extent being deleted */
4999 int eof; /* is deleting at eof */
5000 xfs_bmbt_rec_host_t *ep; /* extent record pointer */
5001 int error; /* error return value */
5002 xfs_extnum_t extno; /* extent number in list */
5003 xfs_bmbt_irec_t got; /* current extent record */
5004 xfs_ifork_t *ifp; /* inode fork pointer */
5005 int isrt; /* freeing in rt area */
5006 xfs_extnum_t lastx; /* last extent index used */
5007 int logflags; /* transaction logging flags */
5008 xfs_extlen_t mod; /* rt extent offset */
5009 xfs_mount_t *mp; /* mount structure */
5010 xfs_extnum_t nextents; /* number of file extents */
5011 xfs_bmbt_irec_t prev; /* previous extent record */
5012 xfs_fileoff_t start; /* first file offset deleted */
5013 int tmp_logflags; /* partial logging flags */
5014 int wasdel; /* was a delayed alloc extent */
5015 int whichfork; /* data or attribute fork */
5018 trace_xfs_bunmap(ip, bno, len, flags, _RET_IP_);
5020 whichfork = (flags & XFS_BMAPI_ATTRFORK) ?
5021 XFS_ATTR_FORK : XFS_DATA_FORK;
5022 ifp = XFS_IFORK_PTR(ip, whichfork);
5024 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS &&
5025 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)) {
5026 XFS_ERROR_REPORT("xfs_bunmapi", XFS_ERRLEVEL_LOW,
5028 return XFS_ERROR(EFSCORRUPTED);
5031 if (XFS_FORCED_SHUTDOWN(mp))
5032 return XFS_ERROR(EIO);
5037 if (!(ifp->if_flags & XFS_IFEXTENTS) &&
5038 (error = xfs_iread_extents(tp, ip, whichfork)))
5040 nextents = ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t);
5041 if (nextents == 0) {
5045 XFS_STATS_INC(xs_blk_unmap);
5046 isrt = (whichfork == XFS_DATA_FORK) && XFS_IS_REALTIME_INODE(ip);
5048 bno = start + len - 1;
5049 ep = xfs_bmap_search_extents(ip, bno, whichfork, &eof, &lastx, &got,
5053 * Check to see if the given block number is past the end of the
5054 * file, back up to the last block if so...
5057 ep = xfs_iext_get_ext(ifp, --lastx);
5058 xfs_bmbt_get_all(ep, &got);
5059 bno = got.br_startoff + got.br_blockcount - 1;
5062 if (ifp->if_flags & XFS_IFBROOT) {
5063 ASSERT(XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_BTREE);
5064 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
5065 cur->bc_private.b.firstblock = *firstblock;
5066 cur->bc_private.b.flist = flist;
5067 cur->bc_private.b.flags = 0;
5073 * Synchronize by locking the bitmap inode.
5075 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
5076 xfs_trans_ijoin(tp, mp->m_rbmip, XFS_ILOCK_EXCL);
5080 while (bno != (xfs_fileoff_t)-1 && bno >= start && lastx >= 0 &&
5081 (nexts == 0 || extno < nexts)) {
5083 * Is the found extent after a hole in which bno lives?
5084 * Just back up to the previous extent, if so.
5086 if (got.br_startoff > bno) {
5089 ep = xfs_iext_get_ext(ifp, lastx);
5090 xfs_bmbt_get_all(ep, &got);
5093 * Is the last block of this extent before the range
5094 * we're supposed to delete? If so, we're done.
5096 bno = XFS_FILEOFF_MIN(bno,
5097 got.br_startoff + got.br_blockcount - 1);
5101 * Then deal with the (possibly delayed) allocated space
5106 wasdel = isnullstartblock(del.br_startblock);
5107 if (got.br_startoff < start) {
5108 del.br_startoff = start;
5109 del.br_blockcount -= start - got.br_startoff;
5111 del.br_startblock += start - got.br_startoff;
5113 if (del.br_startoff + del.br_blockcount > bno + 1)
5114 del.br_blockcount = bno + 1 - del.br_startoff;
5115 sum = del.br_startblock + del.br_blockcount;
5117 (mod = do_mod(sum, mp->m_sb.sb_rextsize))) {
5119 * Realtime extent not lined up at the end.
5120 * The extent could have been split into written
5121 * and unwritten pieces, or we could just be
5122 * unmapping part of it. But we can't really
5123 * get rid of part of a realtime extent.
5125 if (del.br_state == XFS_EXT_UNWRITTEN ||
5126 !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5128 * This piece is unwritten, or we're not
5129 * using unwritten extents. Skip over it.
5132 bno -= mod > del.br_blockcount ?
5133 del.br_blockcount : mod;
5134 if (bno < got.br_startoff) {
5136 xfs_bmbt_get_all(xfs_iext_get_ext(
5142 * It's written, turn it unwritten.
5143 * This is better than zeroing it.
5145 ASSERT(del.br_state == XFS_EXT_NORM);
5146 ASSERT(xfs_trans_get_block_res(tp) > 0);
5148 * If this spans a realtime extent boundary,
5149 * chop it back to the start of the one we end at.
5151 if (del.br_blockcount > mod) {
5152 del.br_startoff += del.br_blockcount - mod;
5153 del.br_startblock += del.br_blockcount - mod;
5154 del.br_blockcount = mod;
5156 del.br_state = XFS_EXT_UNWRITTEN;
5157 error = xfs_bmap_add_extent_unwritten_real(tp, ip,
5158 &lastx, &cur, &del, firstblock, flist,
5164 if (isrt && (mod = do_mod(del.br_startblock, mp->m_sb.sb_rextsize))) {
5166 * Realtime extent is lined up at the end but not
5167 * at the front. We'll get rid of full extents if
5170 mod = mp->m_sb.sb_rextsize - mod;
5171 if (del.br_blockcount > mod) {
5172 del.br_blockcount -= mod;
5173 del.br_startoff += mod;
5174 del.br_startblock += mod;
5175 } else if ((del.br_startoff == start &&
5176 (del.br_state == XFS_EXT_UNWRITTEN ||
5177 xfs_trans_get_block_res(tp) == 0)) ||
5178 !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
5180 * Can't make it unwritten. There isn't
5181 * a full extent here so just skip it.
5183 ASSERT(bno >= del.br_blockcount);
5184 bno -= del.br_blockcount;
5185 if (got.br_startoff > bno) {
5187 ep = xfs_iext_get_ext(ifp,
5189 xfs_bmbt_get_all(ep, &got);
5193 } else if (del.br_state == XFS_EXT_UNWRITTEN) {
5195 * This one is already unwritten.
5196 * It must have a written left neighbor.
5197 * Unwrite the killed part of that one and
5201 xfs_bmbt_get_all(xfs_iext_get_ext(ifp,
5203 ASSERT(prev.br_state == XFS_EXT_NORM);
5204 ASSERT(!isnullstartblock(prev.br_startblock));
5205 ASSERT(del.br_startblock ==
5206 prev.br_startblock + prev.br_blockcount);
5207 if (prev.br_startoff < start) {
5208 mod = start - prev.br_startoff;
5209 prev.br_blockcount -= mod;
5210 prev.br_startblock += mod;
5211 prev.br_startoff = start;
5213 prev.br_state = XFS_EXT_UNWRITTEN;
5215 error = xfs_bmap_add_extent_unwritten_real(tp,
5216 ip, &lastx, &cur, &prev,
5217 firstblock, flist, &logflags);
5222 ASSERT(del.br_state == XFS_EXT_NORM);
5223 del.br_state = XFS_EXT_UNWRITTEN;
5224 error = xfs_bmap_add_extent_unwritten_real(tp,
5225 ip, &lastx, &cur, &del,
5226 firstblock, flist, &logflags);
5233 ASSERT(startblockval(del.br_startblock) > 0);
5234 /* Update realtime/data freespace, unreserve quota */
5236 xfs_filblks_t rtexts;
5238 rtexts = XFS_FSB_TO_B(mp, del.br_blockcount);
5239 do_div(rtexts, mp->m_sb.sb_rextsize);
5240 xfs_mod_incore_sb(mp, XFS_SBS_FREXTENTS,
5241 (int64_t)rtexts, 0);
5242 (void)xfs_trans_reserve_quota_nblks(NULL,
5243 ip, -((long)del.br_blockcount), 0,
5244 XFS_QMOPT_RES_RTBLKS);
5246 xfs_icsb_modify_counters(mp, XFS_SBS_FDBLOCKS,
5247 (int64_t)del.br_blockcount, 0);
5248 (void)xfs_trans_reserve_quota_nblks(NULL,
5249 ip, -((long)del.br_blockcount), 0,
5250 XFS_QMOPT_RES_REGBLKS);
5252 ip->i_delayed_blks -= del.br_blockcount;
5254 cur->bc_private.b.flags |=
5255 XFS_BTCUR_BPRV_WASDEL;
5257 cur->bc_private.b.flags &= ~XFS_BTCUR_BPRV_WASDEL;
5259 * If it's the case where the directory code is running
5260 * with no block reservation, and the deleted block is in
5261 * the middle of its extent, and the resulting insert
5262 * of an extent would cause transformation to btree format,
5263 * then reject it. The calling code will then swap
5264 * blocks around instead.
5265 * We have to do this now, rather than waiting for the
5266 * conversion to btree format, since the transaction
5269 if (!wasdel && xfs_trans_get_block_res(tp) == 0 &&
5270 XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS &&
5271 XFS_IFORK_NEXTENTS(ip, whichfork) >= /* Note the >= */
5272 XFS_IFORK_MAXEXT(ip, whichfork) &&
5273 del.br_startoff > got.br_startoff &&
5274 del.br_startoff + del.br_blockcount <
5275 got.br_startoff + got.br_blockcount) {
5276 error = XFS_ERROR(ENOSPC);
5279 error = xfs_bmap_del_extent(ip, tp, &lastx, flist, cur, &del,
5280 &tmp_logflags, whichfork);
5281 logflags |= tmp_logflags;
5284 bno = del.br_startoff - 1;
5287 * If not done go on to the next (previous) record.
5289 if (bno != (xfs_fileoff_t)-1 && bno >= start) {
5291 ep = xfs_iext_get_ext(ifp, lastx);
5292 if (xfs_bmbt_get_startoff(ep) > bno) {
5294 ep = xfs_iext_get_ext(ifp,
5297 xfs_bmbt_get_all(ep, &got);
5302 *done = bno == (xfs_fileoff_t)-1 || bno < start || lastx < 0;
5305 * Convert to a btree if necessary.
5307 if (xfs_bmap_needs_btree(ip, whichfork)) {
5308 ASSERT(cur == NULL);
5309 error = xfs_bmap_extents_to_btree(tp, ip, firstblock, flist,
5310 &cur, 0, &tmp_logflags, whichfork);
5311 logflags |= tmp_logflags;
5316 * transform from btree to extents, give it cur
5318 else if (xfs_bmap_wants_extents(ip, whichfork)) {
5319 ASSERT(cur != NULL);
5320 error = xfs_bmap_btree_to_extents(tp, ip, cur, &tmp_logflags,
5322 logflags |= tmp_logflags;
5327 * transform from extents to local?
5332 * Log everything. Do this after conversion, there's no point in
5333 * logging the extent records if we've converted to btree format.
5335 if ((logflags & xfs_ilog_fext(whichfork)) &&
5336 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_EXTENTS)
5337 logflags &= ~xfs_ilog_fext(whichfork);
5338 else if ((logflags & xfs_ilog_fbroot(whichfork)) &&
5339 XFS_IFORK_FORMAT(ip, whichfork) != XFS_DINODE_FMT_BTREE)
5340 logflags &= ~xfs_ilog_fbroot(whichfork);
5342 * Log inode even in the error case, if the transaction
5343 * is dirty we'll need to shut down the filesystem.
5346 xfs_trans_log_inode(tp, ip, logflags);
5349 *firstblock = cur->bc_private.b.firstblock;
5350 cur->bc_private.b.allocated = 0;
5352 xfs_btree_del_cursor(cur,
5353 error ? XFS_BTREE_ERROR : XFS_BTREE_NOERROR);