2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * Copyright (c) 2012 Red Hat, Inc.
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "xfs_shared.h"
22 #include "xfs_format.h"
23 #include "xfs_log_format.h"
24 #include "xfs_trans_resv.h"
28 #include "xfs_mount.h"
29 #include "xfs_da_format.h"
30 #include "xfs_inode.h"
31 #include "xfs_btree.h"
32 #include "xfs_trans.h"
33 #include "xfs_extfree_item.h"
34 #include "xfs_alloc.h"
36 #include "xfs_bmap_util.h"
37 #include "xfs_bmap_btree.h"
38 #include "xfs_rtalloc.h"
39 #include "xfs_error.h"
40 #include "xfs_quota.h"
41 #include "xfs_trans_space.h"
42 #include "xfs_trace.h"
43 #include "xfs_icache.h"
45 #include "xfs_dinode.h"
47 /* Kernel only BMAP related definitions and functions */
50 * Convert the given file system block to a disk block. We have to treat it
51 * differently based on whether the file is a real time file or not, because the
55 xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb)
57 return (XFS_IS_REALTIME_INODE(ip) ? \
58 (xfs_daddr_t)XFS_FSB_TO_BB((ip)->i_mount, (fsb)) : \
59 XFS_FSB_TO_DADDR((ip)->i_mount, (fsb)));
63 * Routine to be called at transaction's end by xfs_bmapi, xfs_bunmapi
64 * caller. Frees all the extents that need freeing, which must be done
65 * last due to locking considerations. We never free any extents in
66 * the first transaction.
68 * Return 1 if the given transaction was committed and a new one
69 * started, and 0 otherwise in the committed parameter.
73 xfs_trans_t **tp, /* transaction pointer addr */
74 xfs_bmap_free_t *flist, /* i/o: list extents to free */
75 int *committed) /* xact committed or not */
77 xfs_efd_log_item_t *efd; /* extent free data */
78 xfs_efi_log_item_t *efi; /* extent free intention */
79 int error; /* error return value */
80 xfs_bmap_free_item_t *free; /* free extent item */
81 struct xfs_trans_res tres; /* new log reservation */
82 xfs_mount_t *mp; /* filesystem mount structure */
83 xfs_bmap_free_item_t *next; /* next item on free list */
84 xfs_trans_t *ntp; /* new transaction pointer */
86 ASSERT((*tp)->t_flags & XFS_TRANS_PERM_LOG_RES);
87 if (flist->xbf_count == 0) {
92 efi = xfs_trans_get_efi(ntp, flist->xbf_count);
93 for (free = flist->xbf_first; free; free = free->xbfi_next)
94 xfs_trans_log_efi_extent(ntp, efi, free->xbfi_startblock,
95 free->xbfi_blockcount);
97 tres.tr_logres = ntp->t_log_res;
98 tres.tr_logcount = ntp->t_log_count;
99 tres.tr_logflags = XFS_TRANS_PERM_LOG_RES;
100 ntp = xfs_trans_dup(*tp);
101 error = xfs_trans_commit(*tp, 0);
105 * We have a new transaction, so we should return committed=1,
106 * even though we're returning an error.
112 * transaction commit worked ok so we can drop the extra ticket
113 * reference that we gained in xfs_trans_dup()
115 xfs_log_ticket_put(ntp->t_ticket);
117 error = xfs_trans_reserve(ntp, &tres, 0, 0);
120 efd = xfs_trans_get_efd(ntp, efi, flist->xbf_count);
121 for (free = flist->xbf_first; free != NULL; free = next) {
122 next = free->xbfi_next;
123 if ((error = xfs_free_extent(ntp, free->xbfi_startblock,
124 free->xbfi_blockcount))) {
126 * The bmap free list will be cleaned up at a
127 * higher level. The EFI will be canceled when
128 * this transaction is aborted.
129 * Need to force shutdown here to make sure it
130 * happens, since this transaction may not be
134 if (!XFS_FORCED_SHUTDOWN(mp))
135 xfs_force_shutdown(mp,
136 (error == EFSCORRUPTED) ?
137 SHUTDOWN_CORRUPT_INCORE :
138 SHUTDOWN_META_IO_ERROR);
141 xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,
142 free->xbfi_blockcount);
143 xfs_bmap_del_free(flist, NULL, free);
150 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
152 xfs_alloctype_t atype = 0; /* type for allocation routines */
153 int error; /* error return value */
154 xfs_mount_t *mp; /* mount point structure */
155 xfs_extlen_t prod = 0; /* product factor for allocators */
156 xfs_extlen_t ralen = 0; /* realtime allocation length */
157 xfs_extlen_t align; /* minimum allocation alignment */
160 mp = ap->ip->i_mount;
161 align = xfs_get_extsz_hint(ap->ip);
162 prod = align / mp->m_sb.sb_rextsize;
163 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
164 align, 1, ap->eof, 0,
165 ap->conv, &ap->offset, &ap->length);
169 ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
172 * If the offset & length are not perfectly aligned
173 * then kill prod, it will just get us in trouble.
175 if (do_mod(ap->offset, align) || ap->length % align)
178 * Set ralen to be the actual requested length in rtextents.
180 ralen = ap->length / mp->m_sb.sb_rextsize;
182 * If the old value was close enough to MAXEXTLEN that
183 * we rounded up to it, cut it back so it's valid again.
184 * Note that if it's a really large request (bigger than
185 * MAXEXTLEN), we don't hear about that number, and can't
186 * adjust the starting point to match it.
188 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
189 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
192 * Lock out other modifications to the RT bitmap inode.
194 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL);
195 xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
198 * If it's an allocation to an empty file at offset 0,
199 * pick an extent that will space things out in the rt area.
201 if (ap->eof && ap->offset == 0) {
202 xfs_rtblock_t uninitialized_var(rtx); /* realtime extent no */
204 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
207 ap->blkno = rtx * mp->m_sb.sb_rextsize;
212 xfs_bmap_adjacent(ap);
215 * Realtime allocation, done through xfs_rtallocate_extent.
217 atype = ap->blkno == 0 ? XFS_ALLOCTYPE_ANY_AG : XFS_ALLOCTYPE_NEAR_BNO;
218 do_div(ap->blkno, mp->m_sb.sb_rextsize);
221 if ((error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
222 &ralen, atype, ap->wasdel, prod, &rtb)))
224 if (rtb == NULLFSBLOCK && prod > 1 &&
225 (error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1,
226 ap->length, &ralen, atype,
227 ap->wasdel, 1, &rtb)))
230 if (ap->blkno != NULLFSBLOCK) {
231 ap->blkno *= mp->m_sb.sb_rextsize;
232 ralen *= mp->m_sb.sb_rextsize;
234 ap->ip->i_d.di_nblocks += ralen;
235 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
237 ap->ip->i_delayed_blks -= ralen;
239 * Adjust the disk quota also. This was reserved
242 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
243 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
244 XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
252 * Stack switching interfaces for allocation
255 xfs_bmapi_allocate_worker(
256 struct work_struct *work)
258 struct xfs_bmalloca *args = container_of(work,
259 struct xfs_bmalloca, work);
260 unsigned long pflags;
262 /* we are in a transaction context here */
263 current_set_flags_nested(&pflags, PF_FSTRANS);
265 args->result = __xfs_bmapi_allocate(args);
266 complete(args->done);
268 current_restore_flags_nested(&pflags, PF_FSTRANS);
272 * Some allocation requests often come in with little stack to work on. Push
273 * them off to a worker thread so there is lots of stack to use. Otherwise just
274 * call directly to avoid the context switch overhead here.
278 struct xfs_bmalloca *args)
280 DECLARE_COMPLETION_ONSTACK(done);
282 if (!args->stack_switch)
283 return __xfs_bmapi_allocate(args);
287 INIT_WORK_ONSTACK(&args->work, xfs_bmapi_allocate_worker);
288 queue_work(xfs_alloc_wq, &args->work);
289 wait_for_completion(&done);
290 destroy_work_on_stack(&args->work);
295 * Check if the endoff is outside the last extent. If so the caller will grow
296 * the allocation to a stripe unit boundary. All offsets are considered outside
297 * the end of file for an empty fork, so 1 is returned in *eof in that case.
301 struct xfs_inode *ip,
302 xfs_fileoff_t endoff,
306 struct xfs_bmbt_irec rec;
309 error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
313 *eof = endoff >= rec.br_startoff + rec.br_blockcount;
318 * Extent tree block counting routines.
322 * Count leaf blocks given a range of extent records.
325 xfs_bmap_count_leaves(
333 for (b = 0; b < numrecs; b++) {
334 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
335 *count += xfs_bmbt_get_blockcount(frp);
340 * Count leaf blocks given a range of extent records originally
344 xfs_bmap_disk_count_leaves(
345 struct xfs_mount *mp,
346 struct xfs_btree_block *block,
353 for (b = 1; b <= numrecs; b++) {
354 frp = XFS_BMBT_REC_ADDR(mp, block, b);
355 *count += xfs_bmbt_disk_get_blockcount(frp);
360 * Recursively walks each level of a btree
361 * to count total fsblocks in use.
363 STATIC int /* error */
365 xfs_mount_t *mp, /* file system mount point */
366 xfs_trans_t *tp, /* transaction pointer */
367 xfs_ifork_t *ifp, /* inode fork pointer */
368 xfs_fsblock_t blockno, /* file system block number */
369 int levelin, /* level in btree */
370 int *count) /* Count of blocks */
376 xfs_fsblock_t bno = blockno;
377 xfs_fsblock_t nextbno;
378 struct xfs_btree_block *block, *nextblock;
381 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF,
386 block = XFS_BUF_TO_BLOCK(bp);
389 /* Not at node above leaves, count this level of nodes */
390 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
391 while (nextbno != NULLFSBLOCK) {
392 error = xfs_btree_read_bufl(mp, tp, nextbno, 0, &nbp,
398 nextblock = XFS_BUF_TO_BLOCK(nbp);
399 nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
400 xfs_trans_brelse(tp, nbp);
403 /* Dive to the next level */
404 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
405 bno = be64_to_cpu(*pp);
406 if (unlikely((error =
407 xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
408 xfs_trans_brelse(tp, bp);
409 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
410 XFS_ERRLEVEL_LOW, mp);
411 return XFS_ERROR(EFSCORRUPTED);
413 xfs_trans_brelse(tp, bp);
415 /* count all level 1 nodes and their leaves */
417 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
418 numrecs = be16_to_cpu(block->bb_numrecs);
419 xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
420 xfs_trans_brelse(tp, bp);
421 if (nextbno == NULLFSBLOCK)
424 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
430 block = XFS_BUF_TO_BLOCK(bp);
437 * Count fsblocks of the given fork.
440 xfs_bmap_count_blocks(
441 xfs_trans_t *tp, /* transaction pointer */
442 xfs_inode_t *ip, /* incore inode */
443 int whichfork, /* data or attr fork */
444 int *count) /* out: count of blocks */
446 struct xfs_btree_block *block; /* current btree block */
447 xfs_fsblock_t bno; /* block # of "block" */
448 xfs_ifork_t *ifp; /* fork structure */
449 int level; /* btree level, for checking */
450 xfs_mount_t *mp; /* file system mount structure */
451 __be64 *pp; /* pointer to block address */
455 ifp = XFS_IFORK_PTR(ip, whichfork);
456 if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
457 xfs_bmap_count_leaves(ifp, 0,
458 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
464 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
466 block = ifp->if_broot;
467 level = be16_to_cpu(block->bb_level);
469 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
470 bno = be64_to_cpu(*pp);
471 ASSERT(bno != NULLDFSBNO);
472 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
473 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
475 if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
476 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
478 return XFS_ERROR(EFSCORRUPTED);
485 * returns 1 for success, 0 if we failed to map the extent.
488 xfs_getbmapx_fix_eof_hole(
489 xfs_inode_t *ip, /* xfs incore inode pointer */
490 struct getbmapx *out, /* output structure */
491 int prealloced, /* this is a file with
492 * preallocated data space */
493 __int64_t end, /* last block requested */
494 xfs_fsblock_t startblock)
497 xfs_mount_t *mp; /* file system mount point */
498 xfs_ifork_t *ifp; /* inode fork pointer */
499 xfs_extnum_t lastx; /* last extent pointer */
500 xfs_fileoff_t fileblock;
502 if (startblock == HOLESTARTBLOCK) {
505 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
506 fixlen -= out->bmv_offset;
507 if (prealloced && out->bmv_offset + out->bmv_length == end) {
508 /* Came to hole at EOF. Trim it. */
511 out->bmv_length = fixlen;
514 if (startblock == DELAYSTARTBLOCK)
517 out->bmv_block = xfs_fsb_to_db(ip, startblock);
518 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
519 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
520 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
521 (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
522 out->bmv_oflags |= BMV_OF_LAST;
529 * Get inode's extents as described in bmv, and format for output.
530 * Calls formatter to fill the user's buffer until all extents
531 * are mapped, until the passed-in bmv->bmv_count slots have
532 * been filled, or until the formatter short-circuits the loop,
533 * if it is tracking filled-in extents on its own.
538 struct getbmapx *bmv, /* user bmap structure */
539 xfs_bmap_format_t formatter, /* format to user */
540 void *arg) /* formatter arg */
542 __int64_t bmvend; /* last block requested */
543 int error = 0; /* return value */
544 __int64_t fixlen; /* length for -1 case */
545 int i; /* extent number */
546 int lock; /* lock state */
547 xfs_bmbt_irec_t *map; /* buffer for user's data */
548 xfs_mount_t *mp; /* file system mount point */
549 int nex; /* # of user extents can do */
550 int nexleft; /* # of user extents left */
551 int subnex; /* # of bmapi's can do */
552 int nmap; /* number of map entries */
553 struct getbmapx *out; /* output structure */
554 int whichfork; /* data or attr fork */
555 int prealloced; /* this is a file with
556 * preallocated data space */
557 int iflags; /* interface flags */
558 int bmapi_flags; /* flags for xfs_bmapi */
562 iflags = bmv->bmv_iflags;
563 whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
565 if (whichfork == XFS_ATTR_FORK) {
566 if (XFS_IFORK_Q(ip)) {
567 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
568 ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
569 ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
570 return XFS_ERROR(EINVAL);
572 ip->i_d.di_aformat != 0 &&
573 ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
574 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
576 return XFS_ERROR(EFSCORRUPTED);
582 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
583 ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
584 ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
585 return XFS_ERROR(EINVAL);
587 if (xfs_get_extsz_hint(ip) ||
588 ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
590 fixlen = mp->m_super->s_maxbytes;
593 fixlen = XFS_ISIZE(ip);
597 if (bmv->bmv_length == -1) {
598 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
600 max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
601 } else if (bmv->bmv_length == 0) {
602 bmv->bmv_entries = 0;
604 } else if (bmv->bmv_length < 0) {
605 return XFS_ERROR(EINVAL);
608 nex = bmv->bmv_count - 1;
610 return XFS_ERROR(EINVAL);
611 bmvend = bmv->bmv_offset + bmv->bmv_length;
614 if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
615 return XFS_ERROR(ENOMEM);
616 out = kmem_zalloc_large(bmv->bmv_count * sizeof(struct getbmapx), 0);
618 return XFS_ERROR(ENOMEM);
620 xfs_ilock(ip, XFS_IOLOCK_SHARED);
621 if (whichfork == XFS_DATA_FORK && !(iflags & BMV_IF_DELALLOC)) {
622 if (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size) {
623 error = -filemap_write_and_wait(VFS_I(ip)->i_mapping);
625 goto out_unlock_iolock;
628 * even after flushing the inode, there can still be delalloc
629 * blocks on the inode beyond EOF due to speculative
630 * preallocation. These are not removed until the release
631 * function is called or the inode is inactivated. Hence we
632 * cannot assert here that ip->i_delayed_blks == 0.
636 lock = xfs_ilock_map_shared(ip);
639 * Don't let nex be bigger than the number of extents
640 * we can have assuming alternating holes and real extents.
642 if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
643 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
645 bmapi_flags = xfs_bmapi_aflag(whichfork);
646 if (!(iflags & BMV_IF_PREALLOC))
647 bmapi_flags |= XFS_BMAPI_IGSTATE;
650 * Allocate enough space to handle "subnex" maps at a time.
654 map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
656 goto out_unlock_ilock;
658 bmv->bmv_entries = 0;
660 if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
661 (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
669 nmap = (nexleft > subnex) ? subnex : nexleft;
670 error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
671 XFS_BB_TO_FSB(mp, bmv->bmv_length),
672 map, &nmap, bmapi_flags);
675 ASSERT(nmap <= subnex);
677 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
678 out[cur_ext].bmv_oflags = 0;
679 if (map[i].br_state == XFS_EXT_UNWRITTEN)
680 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
681 else if (map[i].br_startblock == DELAYSTARTBLOCK)
682 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
683 out[cur_ext].bmv_offset =
684 XFS_FSB_TO_BB(mp, map[i].br_startoff);
685 out[cur_ext].bmv_length =
686 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
687 out[cur_ext].bmv_unused1 = 0;
688 out[cur_ext].bmv_unused2 = 0;
691 * delayed allocation extents that start beyond EOF can
692 * occur due to speculative EOF allocation when the
693 * delalloc extent is larger than the largest freespace
694 * extent at conversion time. These extents cannot be
695 * converted by data writeback, so can exist here even
696 * if we are not supposed to be finding delalloc
699 if (map[i].br_startblock == DELAYSTARTBLOCK &&
700 map[i].br_startoff <= XFS_B_TO_FSB(mp, XFS_ISIZE(ip)))
701 ASSERT((iflags & BMV_IF_DELALLOC) != 0);
703 if (map[i].br_startblock == HOLESTARTBLOCK &&
704 whichfork == XFS_ATTR_FORK) {
705 /* came to the end of attribute fork */
706 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
710 if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
712 map[i].br_startblock))
716 out[cur_ext].bmv_offset +
717 out[cur_ext].bmv_length;
719 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
722 * In case we don't want to return the hole,
723 * don't increase cur_ext so that we can reuse
724 * it in the next loop.
726 if ((iflags & BMV_IF_NO_HOLES) &&
727 map[i].br_startblock == HOLESTARTBLOCK) {
728 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
736 } while (nmap && nexleft && bmv->bmv_length);
741 xfs_iunlock_map_shared(ip, lock);
743 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
745 for (i = 0; i < cur_ext; i++) {
746 int full = 0; /* user array is full */
748 /* format results & advance arg */
749 error = formatter(&arg, &out[i], &full);
759 * dead simple method of punching delalyed allocation blocks from a range in
760 * the inode. Walks a block at a time so will be slow, but is only executed in
761 * rare error cases so the overhead is not critical. This will always punch out
762 * both the start and end blocks, even if the ranges only partially overlap
763 * them, so it is up to the caller to ensure that partial blocks are not
767 xfs_bmap_punch_delalloc_range(
768 struct xfs_inode *ip,
769 xfs_fileoff_t start_fsb,
770 xfs_fileoff_t length)
772 xfs_fileoff_t remaining = length;
775 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
779 xfs_bmbt_irec_t imap;
781 xfs_fsblock_t firstblock;
782 xfs_bmap_free_t flist;
785 * Map the range first and check that it is a delalloc extent
786 * before trying to unmap the range. Otherwise we will be
787 * trying to remove a real extent (which requires a
788 * transaction) or a hole, which is probably a bad idea...
790 error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
794 /* something screwed, just bail */
795 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
796 xfs_alert(ip->i_mount,
797 "Failed delalloc mapping lookup ino %lld fsb %lld.",
798 ip->i_ino, start_fsb);
806 if (imap.br_startblock != DELAYSTARTBLOCK) {
807 /* been converted, ignore */
810 WARN_ON(imap.br_blockcount == 0);
813 * Note: while we initialise the firstblock/flist pair, they
814 * should never be used because blocks should never be
815 * allocated or freed for a delalloc extent and hence we need
816 * don't cancel or finish them after the xfs_bunmapi() call.
818 xfs_bmap_init(&flist, &firstblock);
819 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
824 ASSERT(!flist.xbf_count && !flist.xbf_first);
828 } while(remaining > 0);
834 * Test whether it is appropriate to check an inode for and free post EOF
835 * blocks. The 'force' parameter determines whether we should also consider
836 * regular files that are marked preallocated or append-only.
839 xfs_can_free_eofblocks(struct xfs_inode *ip, bool force)
841 /* prealloc/delalloc exists only on regular files */
842 if (!S_ISREG(ip->i_d.di_mode))
846 * Zero sized files with no cached pages and delalloc blocks will not
847 * have speculative prealloc/delalloc blocks to remove.
849 if (VFS_I(ip)->i_size == 0 &&
850 VN_CACHED(VFS_I(ip)) == 0 &&
851 ip->i_delayed_blks == 0)
854 /* If we haven't read in the extent list, then don't do it now. */
855 if (!(ip->i_df.if_flags & XFS_IFEXTENTS))
859 * Do not free real preallocated or append-only files unless the file
860 * has delalloc blocks and we are forced to remove them.
862 if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
863 if (!force || ip->i_delayed_blks == 0)
870 * This is called by xfs_inactive to free any blocks beyond eof
871 * when the link count isn't zero and by xfs_dm_punch_hole() when
872 * punching a hole to EOF.
882 xfs_fileoff_t end_fsb;
883 xfs_fileoff_t last_fsb;
884 xfs_filblks_t map_len;
886 xfs_bmbt_irec_t imap;
889 * Figure out if there are any blocks beyond the end
890 * of the file. If not, then there is nothing to do.
892 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
893 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
894 if (last_fsb <= end_fsb)
896 map_len = last_fsb - end_fsb;
899 xfs_ilock(ip, XFS_ILOCK_SHARED);
900 error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
901 xfs_iunlock(ip, XFS_ILOCK_SHARED);
903 if (!error && (nimaps != 0) &&
904 (imap.br_startblock != HOLESTARTBLOCK ||
905 ip->i_delayed_blks)) {
907 * Attach the dquots to the inode up front.
909 error = xfs_qm_dqattach(ip, 0);
914 * There are blocks after the end of file.
915 * Free them up now by truncating the file to
918 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
921 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
922 xfs_trans_cancel(tp, 0);
927 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
929 ASSERT(XFS_FORCED_SHUTDOWN(mp));
930 xfs_trans_cancel(tp, 0);
932 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
936 xfs_ilock(ip, XFS_ILOCK_EXCL);
937 xfs_trans_ijoin(tp, ip, 0);
940 * Do not update the on-disk file size. If we update the
941 * on-disk file size and then the system crashes before the
942 * contents of the file are flushed to disk then the files
943 * may be full of holes (ie NULL files bug).
945 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK,
949 * If we get an error at this point we simply don't
950 * bother truncating the file.
953 (XFS_TRANS_RELEASE_LOG_RES |
956 error = xfs_trans_commit(tp,
957 XFS_TRANS_RELEASE_LOG_RES);
959 xfs_inode_clear_eofblocks_tag(ip);
962 xfs_iunlock(ip, XFS_ILOCK_EXCL);
964 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
970 xfs_alloc_file_space(
971 struct xfs_inode *ip,
976 xfs_mount_t *mp = ip->i_mount;
978 xfs_filblks_t allocated_fsb;
979 xfs_filblks_t allocatesize_fsb;
980 xfs_extlen_t extsz, temp;
981 xfs_fileoff_t startoffset_fsb;
982 xfs_fsblock_t firstfsb;
987 xfs_bmbt_irec_t imaps[1], *imapp;
988 xfs_bmap_free_t free_list;
989 uint qblocks, resblks, resrtextents;
993 trace_xfs_alloc_file_space(ip);
995 if (XFS_FORCED_SHUTDOWN(mp))
996 return XFS_ERROR(EIO);
998 error = xfs_qm_dqattach(ip, 0);
1003 return XFS_ERROR(EINVAL);
1005 rt = XFS_IS_REALTIME_INODE(ip);
1006 extsz = xfs_get_extsz_hint(ip);
1011 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
1012 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
1015 * Allocate file space until done or until there is an error
1017 while (allocatesize_fsb && !error) {
1021 * Determine space reservations for data/realtime.
1023 if (unlikely(extsz)) {
1024 s = startoffset_fsb;
1027 e = startoffset_fsb + allocatesize_fsb;
1028 if ((temp = do_mod(startoffset_fsb, extsz)))
1030 if ((temp = do_mod(e, extsz)))
1034 e = allocatesize_fsb;
1038 * The transaction reservation is limited to a 32-bit block
1039 * count, hence we need to limit the number of blocks we are
1040 * trying to reserve to avoid an overflow. We can't allocate
1041 * more than @nimaps extents, and an extent is limited on disk
1042 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
1044 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
1046 resrtextents = qblocks = resblks;
1047 resrtextents /= mp->m_sb.sb_rextsize;
1048 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1049 quota_flag = XFS_QMOPT_RES_RTBLKS;
1052 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
1053 quota_flag = XFS_QMOPT_RES_REGBLKS;
1057 * Allocate and setup the transaction.
1059 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1060 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
1061 resblks, resrtextents);
1063 * Check for running out of space
1067 * Free the transaction structure.
1069 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1070 xfs_trans_cancel(tp, 0);
1073 xfs_ilock(ip, XFS_ILOCK_EXCL);
1074 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
1079 xfs_trans_ijoin(tp, ip, 0);
1081 xfs_bmap_init(&free_list, &firstfsb);
1082 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
1083 allocatesize_fsb, alloc_type, &firstfsb,
1084 0, imapp, &nimaps, &free_list);
1090 * Complete the transaction
1092 error = xfs_bmap_finish(&tp, &free_list, &committed);
1097 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1098 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1103 allocated_fsb = imapp->br_blockcount;
1106 error = XFS_ERROR(ENOSPC);
1110 startoffset_fsb += allocated_fsb;
1111 allocatesize_fsb -= allocated_fsb;
1116 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1117 xfs_bmap_cancel(&free_list);
1118 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
1120 error1: /* Just cancel transaction */
1121 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1122 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1127 * Zero file bytes between startoff and endoff inclusive.
1128 * The iolock is held exclusive and no blocks are buffered.
1130 * This function is used by xfs_free_file_space() to zero
1131 * partial blocks when the range to free is not block aligned.
1132 * When unreserving space with boundaries that are not block
1133 * aligned we round up the start and round down the end
1134 * boundaries and then use this function to zero the parts of
1135 * the blocks that got dropped during the rounding.
1138 xfs_zero_remaining_bytes(
1143 xfs_bmbt_irec_t imap;
1144 xfs_fileoff_t offset_fsb;
1145 xfs_off_t lastoffset;
1148 xfs_mount_t *mp = ip->i_mount;
1153 * Avoid doing I/O beyond eof - it's not necessary
1154 * since nothing can read beyond eof. The space will
1155 * be zeroed when the file is extended anyway.
1157 if (startoff >= XFS_ISIZE(ip))
1160 if (endoff > XFS_ISIZE(ip))
1161 endoff = XFS_ISIZE(ip);
1163 bp = xfs_buf_get_uncached(XFS_IS_REALTIME_INODE(ip) ?
1164 mp->m_rtdev_targp : mp->m_ddev_targp,
1165 BTOBB(mp->m_sb.sb_blocksize), 0);
1167 return XFS_ERROR(ENOMEM);
1171 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
1172 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1174 error = xfs_bmapi_read(ip, offset_fsb, 1, &imap, &nimap, 0);
1175 if (error || nimap < 1)
1177 ASSERT(imap.br_blockcount >= 1);
1178 ASSERT(imap.br_startoff == offset_fsb);
1179 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
1180 if (lastoffset > endoff)
1181 lastoffset = endoff;
1182 if (imap.br_startblock == HOLESTARTBLOCK)
1184 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1185 if (imap.br_state == XFS_EXT_UNWRITTEN)
1188 XFS_BUF_UNWRITE(bp);
1190 XFS_BUF_SET_ADDR(bp, xfs_fsb_to_db(ip, imap.br_startblock));
1192 if (XFS_FORCED_SHUTDOWN(mp)) {
1193 error = XFS_ERROR(EIO);
1196 xfs_buf_iorequest(bp);
1197 error = xfs_buf_iowait(bp);
1199 xfs_buf_ioerror_alert(bp,
1200 "xfs_zero_remaining_bytes(read)");
1204 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
1205 0, lastoffset - offset + 1);
1210 if (XFS_FORCED_SHUTDOWN(mp)) {
1211 error = XFS_ERROR(EIO);
1214 xfs_buf_iorequest(bp);
1215 error = xfs_buf_iowait(bp);
1217 xfs_buf_ioerror_alert(bp,
1218 "xfs_zero_remaining_bytes(write)");
1227 xfs_free_file_space(
1228 struct xfs_inode *ip,
1234 xfs_fileoff_t endoffset_fsb;
1236 xfs_fsblock_t firstfsb;
1237 xfs_bmap_free_t free_list;
1238 xfs_bmbt_irec_t imap;
1246 xfs_fileoff_t startoffset_fsb;
1251 trace_xfs_free_file_space(ip);
1253 error = xfs_qm_dqattach(ip, 0);
1258 if (len <= 0) /* if nothing being freed */
1260 rt = XFS_IS_REALTIME_INODE(ip);
1261 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
1262 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
1264 /* wait for the completion of any pending DIOs */
1265 inode_dio_wait(VFS_I(ip));
1267 rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
1268 ioffset = offset & ~(rounding - 1);
1269 error = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
1273 truncate_pagecache_range(VFS_I(ip), ioffset, -1);
1276 * Need to zero the stuff we're not freeing, on disk.
1277 * If it's a realtime file & can't use unwritten extents then we
1278 * actually need to zero the extent edges. Otherwise xfs_bunmapi
1279 * will take care of it for us.
1281 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
1283 error = xfs_bmapi_read(ip, startoffset_fsb, 1,
1287 ASSERT(nimap == 0 || nimap == 1);
1288 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1291 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1292 block = imap.br_startblock;
1293 mod = do_div(block, mp->m_sb.sb_rextsize);
1295 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
1298 error = xfs_bmapi_read(ip, endoffset_fsb - 1, 1,
1302 ASSERT(nimap == 0 || nimap == 1);
1303 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1304 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1306 if (mod && (mod != mp->m_sb.sb_rextsize))
1307 endoffset_fsb -= mod;
1310 if ((done = (endoffset_fsb <= startoffset_fsb)))
1312 * One contiguous piece to clear
1314 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
1317 * Some full blocks, possibly two pieces to clear
1319 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
1320 error = xfs_zero_remaining_bytes(ip, offset,
1321 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
1323 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
1324 error = xfs_zero_remaining_bytes(ip,
1325 XFS_FSB_TO_B(mp, endoffset_fsb),
1330 * free file space until done or until there is an error
1332 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1333 while (!error && !done) {
1336 * allocate and setup the transaction. Allow this
1337 * transaction to dip into the reserve blocks to ensure
1338 * the freeing of the space succeeds at ENOSPC.
1340 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1341 tp->t_flags |= XFS_TRANS_RESERVE;
1342 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write, resblks, 0);
1345 * check for running out of space
1349 * Free the transaction structure.
1351 ASSERT(error == ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1352 xfs_trans_cancel(tp, 0);
1355 xfs_ilock(ip, XFS_ILOCK_EXCL);
1356 error = xfs_trans_reserve_quota(tp, mp,
1357 ip->i_udquot, ip->i_gdquot, ip->i_pdquot,
1358 resblks, 0, XFS_QMOPT_RES_REGBLKS);
1362 xfs_trans_ijoin(tp, ip, 0);
1365 * issue the bunmapi() call to free the blocks
1367 xfs_bmap_init(&free_list, &firstfsb);
1368 error = xfs_bunmapi(tp, ip, startoffset_fsb,
1369 endoffset_fsb - startoffset_fsb,
1370 0, 2, &firstfsb, &free_list, &done);
1376 * complete the transaction
1378 error = xfs_bmap_finish(&tp, &free_list, &committed);
1383 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1384 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1391 xfs_bmap_cancel(&free_list);
1393 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1394 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1400 xfs_zero_file_space(
1401 struct xfs_inode *ip,
1405 struct xfs_mount *mp = ip->i_mount;
1407 xfs_off_t start_boundary;
1408 xfs_off_t end_boundary;
1411 granularity = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
1414 * Round the range of extents we are going to convert inwards. If the
1415 * offset is aligned, then it doesn't get changed so we zero from the
1416 * start of the block offset points to.
1418 start_boundary = round_up(offset, granularity);
1419 end_boundary = round_down(offset + len, granularity);
1421 ASSERT(start_boundary >= offset);
1422 ASSERT(end_boundary <= offset + len);
1424 if (start_boundary < end_boundary - 1) {
1425 /* punch out the page cache over the conversion range */
1426 truncate_pagecache_range(VFS_I(ip), start_boundary,
1428 /* convert the blocks */
1429 error = xfs_alloc_file_space(ip, start_boundary,
1430 end_boundary - start_boundary - 1,
1431 XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT);
1435 /* We've handled the interior of the range, now for the edges */
1436 if (start_boundary != offset) {
1437 error = xfs_iozero(ip, offset, start_boundary - offset);
1442 if (end_boundary != offset + len)
1443 error = xfs_iozero(ip, end_boundary,
1444 offset + len - end_boundary);
1448 * It's either a sub-granularity range or the range spanned lies
1449 * partially across two adjacent blocks.
1451 error = xfs_iozero(ip, offset, len);
1460 * We need to check that the format of the data fork in the temporary inode is
1461 * valid for the target inode before doing the swap. This is not a problem with
1462 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1463 * data fork depending on the space the attribute fork is taking so we can get
1464 * invalid formats on the target inode.
1466 * E.g. target has space for 7 extents in extent format, temp inode only has
1467 * space for 6. If we defragment down to 7 extents, then the tmp format is a
1468 * btree, but when swapped it needs to be in extent format. Hence we can't just
1469 * blindly swap data forks on attr2 filesystems.
1471 * Note that we check the swap in both directions so that we don't end up with
1472 * a corrupt temporary inode, either.
1474 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1475 * inode will prevent this situation from occurring, so all we do here is
1476 * reject and log the attempt. basically we are putting the responsibility on
1477 * userspace to get this right.
1480 xfs_swap_extents_check_format(
1481 xfs_inode_t *ip, /* target inode */
1482 xfs_inode_t *tip) /* tmp inode */
1485 /* Should never get a local format */
1486 if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
1487 tip->i_d.di_format == XFS_DINODE_FMT_LOCAL)
1491 * if the target inode has less extents that then temporary inode then
1492 * why did userspace call us?
1494 if (ip->i_d.di_nextents < tip->i_d.di_nextents)
1498 * if the target inode is in extent form and the temp inode is in btree
1499 * form then we will end up with the target inode in the wrong format
1500 * as we already know there are less extents in the temp inode.
1502 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1503 tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1506 /* Check temp in extent form to max in target */
1507 if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1508 XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) >
1509 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1512 /* Check target in extent form to max in temp */
1513 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1514 XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) >
1515 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1519 * If we are in a btree format, check that the temp root block will fit
1520 * in the target and that it has enough extents to be in btree format
1523 * Note that we have to be careful to allow btree->extent conversions
1524 * (a common defrag case) which will occur when the temp inode is in
1527 if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1528 if (XFS_IFORK_BOFF(ip) &&
1529 XFS_BMAP_BMDR_SPACE(tip->i_df.if_broot) > XFS_IFORK_BOFF(ip))
1531 if (XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <=
1532 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1536 /* Reciprocal target->temp btree format checks */
1537 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1538 if (XFS_IFORK_BOFF(tip) &&
1539 XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip))
1541 if (XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <=
1542 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1551 xfs_inode_t *ip, /* target inode */
1552 xfs_inode_t *tip, /* tmp inode */
1555 xfs_mount_t *mp = ip->i_mount;
1557 xfs_bstat_t *sbp = &sxp->sx_stat;
1558 xfs_ifork_t *tempifp, *ifp, *tifp;
1559 int src_log_flags, target_log_flags;
1565 tempifp = kmem_alloc(sizeof(xfs_ifork_t), KM_MAYFAIL);
1567 error = XFS_ERROR(ENOMEM);
1572 * we have to do two separate lock calls here to keep lockdep
1573 * happy. If we try to get all the locks in one call, lock will
1574 * report false positives when we drop the ILOCK and regain them
1577 xfs_lock_two_inodes(ip, tip, XFS_IOLOCK_EXCL);
1578 xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
1580 /* Verify that both files have the same format */
1581 if ((ip->i_d.di_mode & S_IFMT) != (tip->i_d.di_mode & S_IFMT)) {
1582 error = XFS_ERROR(EINVAL);
1586 /* Verify both files are either real-time or non-realtime */
1587 if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
1588 error = XFS_ERROR(EINVAL);
1592 error = -filemap_write_and_wait(VFS_I(tip)->i_mapping);
1595 truncate_pagecache_range(VFS_I(tip), 0, -1);
1597 /* Verify O_DIRECT for ftmp */
1598 if (VN_CACHED(VFS_I(tip)) != 0) {
1599 error = XFS_ERROR(EINVAL);
1603 /* Verify all data are being swapped */
1604 if (sxp->sx_offset != 0 ||
1605 sxp->sx_length != ip->i_d.di_size ||
1606 sxp->sx_length != tip->i_d.di_size) {
1607 error = XFS_ERROR(EFAULT);
1611 trace_xfs_swap_extent_before(ip, 0);
1612 trace_xfs_swap_extent_before(tip, 1);
1614 /* check inode formats now that data is flushed */
1615 error = xfs_swap_extents_check_format(ip, tip);
1618 "%s: inode 0x%llx format is incompatible for exchanging.",
1619 __func__, ip->i_ino);
1624 * Compare the current change & modify times with that
1625 * passed in. If they differ, we abort this swap.
1626 * This is the mechanism used to ensure the calling
1627 * process that the file was not changed out from
1630 if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
1631 (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
1632 (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
1633 (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
1634 error = XFS_ERROR(EBUSY);
1638 /* We need to fail if the file is memory mapped. Once we have tossed
1639 * all existing pages, the page fault will have no option
1640 * but to go to the filesystem for pages. By making the page fault call
1641 * vop_read (or write in the case of autogrow) they block on the iolock
1642 * until we have switched the extents.
1644 if (VN_MAPPED(VFS_I(ip))) {
1645 error = XFS_ERROR(EBUSY);
1649 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1650 xfs_iunlock(tip, XFS_ILOCK_EXCL);
1653 * There is a race condition here since we gave up the
1654 * ilock. However, the data fork will not change since
1655 * we have the iolock (locked for truncation too) so we
1656 * are safe. We don't really care if non-io related
1659 truncate_pagecache_range(VFS_I(ip), 0, -1);
1661 tp = xfs_trans_alloc(mp, XFS_TRANS_SWAPEXT);
1662 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0);
1664 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
1665 xfs_iunlock(tip, XFS_IOLOCK_EXCL);
1666 xfs_trans_cancel(tp, 0);
1669 xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
1672 * Count the number of extended attribute blocks
1674 if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
1675 (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1676 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks);
1678 goto out_trans_cancel;
1680 if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
1681 (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1682 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
1685 goto out_trans_cancel;
1688 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1689 xfs_trans_ijoin(tp, tip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1692 * Before we've swapped the forks, lets set the owners of the forks
1693 * appropriately. We have to do this as we are demand paging the btree
1694 * buffers, and so the validation done on read will expect the owner
1695 * field to be correctly set. Once we change the owners, we can swap the
1698 * Note the trickiness in setting the log flags - we set the owner log
1699 * flag on the opposite inode (i.e. the inode we are setting the new
1700 * owner to be) because once we swap the forks and log that, log
1701 * recovery is going to see the fork as owned by the swapped inode,
1702 * not the pre-swapped inodes.
1704 src_log_flags = XFS_ILOG_CORE;
1705 target_log_flags = XFS_ILOG_CORE;
1706 if (ip->i_d.di_version == 3 &&
1707 ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1708 target_log_flags |= XFS_ILOG_DOWNER;
1709 error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK,
1712 goto out_trans_cancel;
1715 if (tip->i_d.di_version == 3 &&
1716 tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1717 src_log_flags |= XFS_ILOG_DOWNER;
1718 error = xfs_bmbt_change_owner(tp, tip, XFS_DATA_FORK,
1721 goto out_trans_cancel;
1725 * Swap the data forks of the inodes
1729 *tempifp = *ifp; /* struct copy */
1730 *ifp = *tifp; /* struct copy */
1731 *tifp = *tempifp; /* struct copy */
1734 * Fix the on-disk inode values
1736 tmp = (__uint64_t)ip->i_d.di_nblocks;
1737 ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
1738 tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
1740 tmp = (__uint64_t) ip->i_d.di_nextents;
1741 ip->i_d.di_nextents = tip->i_d.di_nextents;
1742 tip->i_d.di_nextents = tmp;
1744 tmp = (__uint64_t) ip->i_d.di_format;
1745 ip->i_d.di_format = tip->i_d.di_format;
1746 tip->i_d.di_format = tmp;
1749 * The extents in the source inode could still contain speculative
1750 * preallocation beyond EOF (e.g. the file is open but not modified
1751 * while defrag is in progress). In that case, we need to copy over the
1752 * number of delalloc blocks the data fork in the source inode is
1753 * tracking beyond EOF so that when the fork is truncated away when the
1754 * temporary inode is unlinked we don't underrun the i_delayed_blks
1755 * counter on that inode.
1757 ASSERT(tip->i_delayed_blks == 0);
1758 tip->i_delayed_blks = ip->i_delayed_blks;
1759 ip->i_delayed_blks = 0;
1761 switch (ip->i_d.di_format) {
1762 case XFS_DINODE_FMT_EXTENTS:
1763 /* If the extents fit in the inode, fix the
1764 * pointer. Otherwise it's already NULL or
1765 * pointing to the extent.
1767 if (ip->i_d.di_nextents <= XFS_INLINE_EXTS) {
1768 ifp->if_u1.if_extents =
1769 ifp->if_u2.if_inline_ext;
1771 src_log_flags |= XFS_ILOG_DEXT;
1773 case XFS_DINODE_FMT_BTREE:
1774 ASSERT(ip->i_d.di_version < 3 ||
1775 (src_log_flags & XFS_ILOG_DOWNER));
1776 src_log_flags |= XFS_ILOG_DBROOT;
1780 switch (tip->i_d.di_format) {
1781 case XFS_DINODE_FMT_EXTENTS:
1782 /* If the extents fit in the inode, fix the
1783 * pointer. Otherwise it's already NULL or
1784 * pointing to the extent.
1786 if (tip->i_d.di_nextents <= XFS_INLINE_EXTS) {
1787 tifp->if_u1.if_extents =
1788 tifp->if_u2.if_inline_ext;
1790 target_log_flags |= XFS_ILOG_DEXT;
1792 case XFS_DINODE_FMT_BTREE:
1793 target_log_flags |= XFS_ILOG_DBROOT;
1794 ASSERT(tip->i_d.di_version < 3 ||
1795 (target_log_flags & XFS_ILOG_DOWNER));
1799 xfs_trans_log_inode(tp, ip, src_log_flags);
1800 xfs_trans_log_inode(tp, tip, target_log_flags);
1803 * If this is a synchronous mount, make sure that the
1804 * transaction goes to disk before returning to the user.
1806 if (mp->m_flags & XFS_MOUNT_WSYNC)
1807 xfs_trans_set_sync(tp);
1809 error = xfs_trans_commit(tp, 0);
1811 trace_xfs_swap_extent_after(ip, 0);
1812 trace_xfs_swap_extent_after(tip, 1);
1818 xfs_iunlock(ip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1819 xfs_iunlock(tip, XFS_ILOCK_EXCL | XFS_IOLOCK_EXCL);
1823 xfs_trans_cancel(tp, 0);