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 * Check if the endoff is outside the last extent. If so the caller will grow
253 * the allocation to a stripe unit boundary. All offsets are considered outside
254 * the end of file for an empty fork, so 1 is returned in *eof in that case.
258 struct xfs_inode *ip,
259 xfs_fileoff_t endoff,
263 struct xfs_bmbt_irec rec;
266 error = xfs_bmap_last_extent(NULL, ip, whichfork, &rec, eof);
270 *eof = endoff >= rec.br_startoff + rec.br_blockcount;
275 * Extent tree block counting routines.
279 * Count leaf blocks given a range of extent records.
282 xfs_bmap_count_leaves(
290 for (b = 0; b < numrecs; b++) {
291 xfs_bmbt_rec_host_t *frp = xfs_iext_get_ext(ifp, idx + b);
292 *count += xfs_bmbt_get_blockcount(frp);
297 * Count leaf blocks given a range of extent records originally
301 xfs_bmap_disk_count_leaves(
302 struct xfs_mount *mp,
303 struct xfs_btree_block *block,
310 for (b = 1; b <= numrecs; b++) {
311 frp = XFS_BMBT_REC_ADDR(mp, block, b);
312 *count += xfs_bmbt_disk_get_blockcount(frp);
317 * Recursively walks each level of a btree
318 * to count total fsblocks in use.
320 STATIC int /* error */
322 xfs_mount_t *mp, /* file system mount point */
323 xfs_trans_t *tp, /* transaction pointer */
324 xfs_ifork_t *ifp, /* inode fork pointer */
325 xfs_fsblock_t blockno, /* file system block number */
326 int levelin, /* level in btree */
327 int *count) /* Count of blocks */
333 xfs_fsblock_t bno = blockno;
334 xfs_fsblock_t nextbno;
335 struct xfs_btree_block *block, *nextblock;
338 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp, XFS_BMAP_BTREE_REF,
343 block = XFS_BUF_TO_BLOCK(bp);
346 /* Not at node above leaves, count this level of nodes */
347 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
348 while (nextbno != NULLFSBLOCK) {
349 error = xfs_btree_read_bufl(mp, tp, nextbno, 0, &nbp,
355 nextblock = XFS_BUF_TO_BLOCK(nbp);
356 nextbno = be64_to_cpu(nextblock->bb_u.l.bb_rightsib);
357 xfs_trans_brelse(tp, nbp);
360 /* Dive to the next level */
361 pp = XFS_BMBT_PTR_ADDR(mp, block, 1, mp->m_bmap_dmxr[1]);
362 bno = be64_to_cpu(*pp);
363 if (unlikely((error =
364 xfs_bmap_count_tree(mp, tp, ifp, bno, level, count)) < 0)) {
365 xfs_trans_brelse(tp, bp);
366 XFS_ERROR_REPORT("xfs_bmap_count_tree(1)",
367 XFS_ERRLEVEL_LOW, mp);
368 return -EFSCORRUPTED;
370 xfs_trans_brelse(tp, bp);
372 /* count all level 1 nodes and their leaves */
374 nextbno = be64_to_cpu(block->bb_u.l.bb_rightsib);
375 numrecs = be16_to_cpu(block->bb_numrecs);
376 xfs_bmap_disk_count_leaves(mp, block, numrecs, count);
377 xfs_trans_brelse(tp, bp);
378 if (nextbno == NULLFSBLOCK)
381 error = xfs_btree_read_bufl(mp, tp, bno, 0, &bp,
387 block = XFS_BUF_TO_BLOCK(bp);
394 * Count fsblocks of the given fork.
397 xfs_bmap_count_blocks(
398 xfs_trans_t *tp, /* transaction pointer */
399 xfs_inode_t *ip, /* incore inode */
400 int whichfork, /* data or attr fork */
401 int *count) /* out: count of blocks */
403 struct xfs_btree_block *block; /* current btree block */
404 xfs_fsblock_t bno; /* block # of "block" */
405 xfs_ifork_t *ifp; /* fork structure */
406 int level; /* btree level, for checking */
407 xfs_mount_t *mp; /* file system mount structure */
408 __be64 *pp; /* pointer to block address */
412 ifp = XFS_IFORK_PTR(ip, whichfork);
413 if ( XFS_IFORK_FORMAT(ip, whichfork) == XFS_DINODE_FMT_EXTENTS ) {
414 xfs_bmap_count_leaves(ifp, 0,
415 ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t),
421 * Root level must use BMAP_BROOT_PTR_ADDR macro to get ptr out.
423 block = ifp->if_broot;
424 level = be16_to_cpu(block->bb_level);
426 pp = XFS_BMAP_BROOT_PTR_ADDR(mp, block, 1, ifp->if_broot_bytes);
427 bno = be64_to_cpu(*pp);
428 ASSERT(bno != NULLFSBLOCK);
429 ASSERT(XFS_FSB_TO_AGNO(mp, bno) < mp->m_sb.sb_agcount);
430 ASSERT(XFS_FSB_TO_AGBNO(mp, bno) < mp->m_sb.sb_agblocks);
432 if (unlikely(xfs_bmap_count_tree(mp, tp, ifp, bno, level, count) < 0)) {
433 XFS_ERROR_REPORT("xfs_bmap_count_blocks(2)", XFS_ERRLEVEL_LOW,
435 return -EFSCORRUPTED;
442 * returns 1 for success, 0 if we failed to map the extent.
445 xfs_getbmapx_fix_eof_hole(
446 xfs_inode_t *ip, /* xfs incore inode pointer */
447 struct getbmapx *out, /* output structure */
448 int prealloced, /* this is a file with
449 * preallocated data space */
450 __int64_t end, /* last block requested */
451 xfs_fsblock_t startblock)
454 xfs_mount_t *mp; /* file system mount point */
455 xfs_ifork_t *ifp; /* inode fork pointer */
456 xfs_extnum_t lastx; /* last extent pointer */
457 xfs_fileoff_t fileblock;
459 if (startblock == HOLESTARTBLOCK) {
462 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
463 fixlen -= out->bmv_offset;
464 if (prealloced && out->bmv_offset + out->bmv_length == end) {
465 /* Came to hole at EOF. Trim it. */
468 out->bmv_length = fixlen;
471 if (startblock == DELAYSTARTBLOCK)
474 out->bmv_block = xfs_fsb_to_db(ip, startblock);
475 fileblock = XFS_BB_TO_FSB(ip->i_mount, out->bmv_offset);
476 ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
477 if (xfs_iext_bno_to_ext(ifp, fileblock, &lastx) &&
478 (lastx == (ifp->if_bytes / (uint)sizeof(xfs_bmbt_rec_t))-1))
479 out->bmv_oflags |= BMV_OF_LAST;
486 * Get inode's extents as described in bmv, and format for output.
487 * Calls formatter to fill the user's buffer until all extents
488 * are mapped, until the passed-in bmv->bmv_count slots have
489 * been filled, or until the formatter short-circuits the loop,
490 * if it is tracking filled-in extents on its own.
495 struct getbmapx *bmv, /* user bmap structure */
496 xfs_bmap_format_t formatter, /* format to user */
497 void *arg) /* formatter arg */
499 __int64_t bmvend; /* last block requested */
500 int error = 0; /* return value */
501 __int64_t fixlen; /* length for -1 case */
502 int i; /* extent number */
503 int lock; /* lock state */
504 xfs_bmbt_irec_t *map; /* buffer for user's data */
505 xfs_mount_t *mp; /* file system mount point */
506 int nex; /* # of user extents can do */
507 int nexleft; /* # of user extents left */
508 int subnex; /* # of bmapi's can do */
509 int nmap; /* number of map entries */
510 struct getbmapx *out; /* output structure */
511 int whichfork; /* data or attr fork */
512 int prealloced; /* this is a file with
513 * preallocated data space */
514 int iflags; /* interface flags */
515 int bmapi_flags; /* flags for xfs_bmapi */
519 iflags = bmv->bmv_iflags;
520 whichfork = iflags & BMV_IF_ATTRFORK ? XFS_ATTR_FORK : XFS_DATA_FORK;
522 if (whichfork == XFS_ATTR_FORK) {
523 if (XFS_IFORK_Q(ip)) {
524 if (ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS &&
525 ip->i_d.di_aformat != XFS_DINODE_FMT_BTREE &&
526 ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)
529 ip->i_d.di_aformat != 0 &&
530 ip->i_d.di_aformat != XFS_DINODE_FMT_EXTENTS)) {
531 XFS_ERROR_REPORT("xfs_getbmap", XFS_ERRLEVEL_LOW,
533 return -EFSCORRUPTED;
539 if (ip->i_d.di_format != XFS_DINODE_FMT_EXTENTS &&
540 ip->i_d.di_format != XFS_DINODE_FMT_BTREE &&
541 ip->i_d.di_format != XFS_DINODE_FMT_LOCAL)
544 if (xfs_get_extsz_hint(ip) ||
545 ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC|XFS_DIFLAG_APPEND)){
547 fixlen = mp->m_super->s_maxbytes;
550 fixlen = XFS_ISIZE(ip);
554 if (bmv->bmv_length == -1) {
555 fixlen = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, fixlen));
557 max_t(__int64_t, fixlen - bmv->bmv_offset, 0);
558 } else if (bmv->bmv_length == 0) {
559 bmv->bmv_entries = 0;
561 } else if (bmv->bmv_length < 0) {
565 nex = bmv->bmv_count - 1;
568 bmvend = bmv->bmv_offset + bmv->bmv_length;
571 if (bmv->bmv_count > ULONG_MAX / sizeof(struct getbmapx))
573 out = kmem_zalloc_large(bmv->bmv_count * sizeof(struct getbmapx), 0);
577 xfs_ilock(ip, XFS_IOLOCK_SHARED);
578 if (whichfork == XFS_DATA_FORK) {
579 if (!(iflags & BMV_IF_DELALLOC) &&
580 (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size)) {
581 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
583 goto out_unlock_iolock;
586 * Even after flushing the inode, there can still be
587 * delalloc blocks on the inode beyond EOF due to
588 * speculative preallocation. These are not removed
589 * until the release function is called or the inode
590 * is inactivated. Hence we cannot assert here that
591 * ip->i_delayed_blks == 0.
595 lock = xfs_ilock_data_map_shared(ip);
597 lock = xfs_ilock_attr_map_shared(ip);
601 * Don't let nex be bigger than the number of extents
602 * we can have assuming alternating holes and real extents.
604 if (nex > XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1)
605 nex = XFS_IFORK_NEXTENTS(ip, whichfork) * 2 + 1;
607 bmapi_flags = xfs_bmapi_aflag(whichfork);
608 if (!(iflags & BMV_IF_PREALLOC))
609 bmapi_flags |= XFS_BMAPI_IGSTATE;
612 * Allocate enough space to handle "subnex" maps at a time.
616 map = kmem_alloc(subnex * sizeof(*map), KM_MAYFAIL | KM_NOFS);
618 goto out_unlock_ilock;
620 bmv->bmv_entries = 0;
622 if (XFS_IFORK_NEXTENTS(ip, whichfork) == 0 &&
623 (whichfork == XFS_ATTR_FORK || !(iflags & BMV_IF_DELALLOC))) {
631 nmap = (nexleft > subnex) ? subnex : nexleft;
632 error = xfs_bmapi_read(ip, XFS_BB_TO_FSBT(mp, bmv->bmv_offset),
633 XFS_BB_TO_FSB(mp, bmv->bmv_length),
634 map, &nmap, bmapi_flags);
637 ASSERT(nmap <= subnex);
639 for (i = 0; i < nmap && nexleft && bmv->bmv_length; i++) {
640 out[cur_ext].bmv_oflags = 0;
641 if (map[i].br_state == XFS_EXT_UNWRITTEN)
642 out[cur_ext].bmv_oflags |= BMV_OF_PREALLOC;
643 else if (map[i].br_startblock == DELAYSTARTBLOCK)
644 out[cur_ext].bmv_oflags |= BMV_OF_DELALLOC;
645 out[cur_ext].bmv_offset =
646 XFS_FSB_TO_BB(mp, map[i].br_startoff);
647 out[cur_ext].bmv_length =
648 XFS_FSB_TO_BB(mp, map[i].br_blockcount);
649 out[cur_ext].bmv_unused1 = 0;
650 out[cur_ext].bmv_unused2 = 0;
653 * delayed allocation extents that start beyond EOF can
654 * occur due to speculative EOF allocation when the
655 * delalloc extent is larger than the largest freespace
656 * extent at conversion time. These extents cannot be
657 * converted by data writeback, so can exist here even
658 * if we are not supposed to be finding delalloc
661 if (map[i].br_startblock == DELAYSTARTBLOCK &&
662 map[i].br_startoff <= XFS_B_TO_FSB(mp, XFS_ISIZE(ip)))
663 ASSERT((iflags & BMV_IF_DELALLOC) != 0);
665 if (map[i].br_startblock == HOLESTARTBLOCK &&
666 whichfork == XFS_ATTR_FORK) {
667 /* came to the end of attribute fork */
668 out[cur_ext].bmv_oflags |= BMV_OF_LAST;
672 if (!xfs_getbmapx_fix_eof_hole(ip, &out[cur_ext],
674 map[i].br_startblock))
678 out[cur_ext].bmv_offset +
679 out[cur_ext].bmv_length;
681 max_t(__int64_t, 0, bmvend - bmv->bmv_offset);
684 * In case we don't want to return the hole,
685 * don't increase cur_ext so that we can reuse
686 * it in the next loop.
688 if ((iflags & BMV_IF_NO_HOLES) &&
689 map[i].br_startblock == HOLESTARTBLOCK) {
690 memset(&out[cur_ext], 0, sizeof(out[cur_ext]));
698 } while (nmap && nexleft && bmv->bmv_length);
703 xfs_iunlock(ip, lock);
705 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
707 for (i = 0; i < cur_ext; i++) {
708 int full = 0; /* user array is full */
710 /* format results & advance arg */
711 error = formatter(&arg, &out[i], &full);
721 * dead simple method of punching delalyed allocation blocks from a range in
722 * the inode. Walks a block at a time so will be slow, but is only executed in
723 * rare error cases so the overhead is not critical. This will always punch out
724 * both the start and end blocks, even if the ranges only partially overlap
725 * them, so it is up to the caller to ensure that partial blocks are not
729 xfs_bmap_punch_delalloc_range(
730 struct xfs_inode *ip,
731 xfs_fileoff_t start_fsb,
732 xfs_fileoff_t length)
734 xfs_fileoff_t remaining = length;
737 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
741 xfs_bmbt_irec_t imap;
743 xfs_fsblock_t firstblock;
744 xfs_bmap_free_t flist;
747 * Map the range first and check that it is a delalloc extent
748 * before trying to unmap the range. Otherwise we will be
749 * trying to remove a real extent (which requires a
750 * transaction) or a hole, which is probably a bad idea...
752 error = xfs_bmapi_read(ip, start_fsb, 1, &imap, &nimaps,
756 /* something screwed, just bail */
757 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
758 xfs_alert(ip->i_mount,
759 "Failed delalloc mapping lookup ino %lld fsb %lld.",
760 ip->i_ino, start_fsb);
768 if (imap.br_startblock != DELAYSTARTBLOCK) {
769 /* been converted, ignore */
772 WARN_ON(imap.br_blockcount == 0);
775 * Note: while we initialise the firstblock/flist pair, they
776 * should never be used because blocks should never be
777 * allocated or freed for a delalloc extent and hence we need
778 * don't cancel or finish them after the xfs_bunmapi() call.
780 xfs_bmap_init(&flist, &firstblock);
781 error = xfs_bunmapi(NULL, ip, start_fsb, 1, 0, 1, &firstblock,
786 ASSERT(!flist.xbf_count && !flist.xbf_first);
790 } while(remaining > 0);
796 * Test whether it is appropriate to check an inode for and free post EOF
797 * blocks. The 'force' parameter determines whether we should also consider
798 * regular files that are marked preallocated or append-only.
801 xfs_can_free_eofblocks(struct xfs_inode *ip, bool force)
803 /* prealloc/delalloc exists only on regular files */
804 if (!S_ISREG(ip->i_d.di_mode))
808 * Zero sized files with no cached pages and delalloc blocks will not
809 * have speculative prealloc/delalloc blocks to remove.
811 if (VFS_I(ip)->i_size == 0 &&
812 VFS_I(ip)->i_mapping->nrpages == 0 &&
813 ip->i_delayed_blks == 0)
816 /* If we haven't read in the extent list, then don't do it now. */
817 if (!(ip->i_df.if_flags & XFS_IFEXTENTS))
821 * Do not free real preallocated or append-only files unless the file
822 * has delalloc blocks and we are forced to remove them.
824 if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
825 if (!force || ip->i_delayed_blks == 0)
832 * This is called by xfs_inactive to free any blocks beyond eof
833 * when the link count isn't zero and by xfs_dm_punch_hole() when
834 * punching a hole to EOF.
844 xfs_fileoff_t end_fsb;
845 xfs_fileoff_t last_fsb;
846 xfs_filblks_t map_len;
848 xfs_bmbt_irec_t imap;
851 * Figure out if there are any blocks beyond the end
852 * of the file. If not, then there is nothing to do.
854 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
855 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
856 if (last_fsb <= end_fsb)
858 map_len = last_fsb - end_fsb;
861 xfs_ilock(ip, XFS_ILOCK_SHARED);
862 error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
863 xfs_iunlock(ip, XFS_ILOCK_SHARED);
865 if (!error && (nimaps != 0) &&
866 (imap.br_startblock != HOLESTARTBLOCK ||
867 ip->i_delayed_blks)) {
869 * Attach the dquots to the inode up front.
871 error = xfs_qm_dqattach(ip, 0);
876 * There are blocks after the end of file.
877 * Free them up now by truncating the file to
880 tp = xfs_trans_alloc(mp, XFS_TRANS_INACTIVE);
883 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_EXCL)) {
884 xfs_trans_cancel(tp, 0);
889 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_itruncate, 0, 0);
891 ASSERT(XFS_FORCED_SHUTDOWN(mp));
892 xfs_trans_cancel(tp, 0);
894 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
898 xfs_ilock(ip, XFS_ILOCK_EXCL);
899 xfs_trans_ijoin(tp, ip, 0);
902 * Do not update the on-disk file size. If we update the
903 * on-disk file size and then the system crashes before the
904 * contents of the file are flushed to disk then the files
905 * may be full of holes (ie NULL files bug).
907 error = xfs_itruncate_extents(&tp, ip, XFS_DATA_FORK,
911 * If we get an error at this point we simply don't
912 * bother truncating the file.
915 (XFS_TRANS_RELEASE_LOG_RES |
918 error = xfs_trans_commit(tp,
919 XFS_TRANS_RELEASE_LOG_RES);
921 xfs_inode_clear_eofblocks_tag(ip);
924 xfs_iunlock(ip, XFS_ILOCK_EXCL);
926 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
932 xfs_alloc_file_space(
933 struct xfs_inode *ip,
938 xfs_mount_t *mp = ip->i_mount;
940 xfs_filblks_t allocated_fsb;
941 xfs_filblks_t allocatesize_fsb;
942 xfs_extlen_t extsz, temp;
943 xfs_fileoff_t startoffset_fsb;
944 xfs_fsblock_t firstfsb;
949 xfs_bmbt_irec_t imaps[1], *imapp;
950 xfs_bmap_free_t free_list;
951 uint qblocks, resblks, resrtextents;
955 trace_xfs_alloc_file_space(ip);
957 if (XFS_FORCED_SHUTDOWN(mp))
960 error = xfs_qm_dqattach(ip, 0);
967 rt = XFS_IS_REALTIME_INODE(ip);
968 extsz = xfs_get_extsz_hint(ip);
973 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
974 allocatesize_fsb = XFS_B_TO_FSB(mp, count);
977 * Allocate file space until done or until there is an error
979 while (allocatesize_fsb && !error) {
983 * Determine space reservations for data/realtime.
985 if (unlikely(extsz)) {
989 e = startoffset_fsb + allocatesize_fsb;
990 if ((temp = do_mod(startoffset_fsb, extsz)))
992 if ((temp = do_mod(e, extsz)))
996 e = allocatesize_fsb;
1000 * The transaction reservation is limited to a 32-bit block
1001 * count, hence we need to limit the number of blocks we are
1002 * trying to reserve to avoid an overflow. We can't allocate
1003 * more than @nimaps extents, and an extent is limited on disk
1004 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
1006 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
1008 resrtextents = qblocks = resblks;
1009 resrtextents /= mp->m_sb.sb_rextsize;
1010 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1011 quota_flag = XFS_QMOPT_RES_RTBLKS;
1014 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
1015 quota_flag = XFS_QMOPT_RES_REGBLKS;
1019 * Allocate and setup the transaction.
1021 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1022 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
1023 resblks, resrtextents);
1025 * Check for running out of space
1029 * Free the transaction structure.
1031 ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1032 xfs_trans_cancel(tp, 0);
1035 xfs_ilock(ip, XFS_ILOCK_EXCL);
1036 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks,
1041 xfs_trans_ijoin(tp, ip, 0);
1043 xfs_bmap_init(&free_list, &firstfsb);
1044 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
1045 allocatesize_fsb, alloc_type, &firstfsb,
1046 0, imapp, &nimaps, &free_list);
1052 * Complete the transaction
1054 error = xfs_bmap_finish(&tp, &free_list, &committed);
1059 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1060 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1065 allocated_fsb = imapp->br_blockcount;
1072 startoffset_fsb += allocated_fsb;
1073 allocatesize_fsb -= allocated_fsb;
1078 error0: /* Cancel bmap, unlock inode, unreserve quota blocks, cancel trans */
1079 xfs_bmap_cancel(&free_list);
1080 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
1082 error1: /* Just cancel transaction */
1083 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1084 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1089 * Zero file bytes between startoff and endoff inclusive.
1090 * The iolock is held exclusive and no blocks are buffered.
1092 * This function is used by xfs_free_file_space() to zero
1093 * partial blocks when the range to free is not block aligned.
1094 * When unreserving space with boundaries that are not block
1095 * aligned we round up the start and round down the end
1096 * boundaries and then use this function to zero the parts of
1097 * the blocks that got dropped during the rounding.
1100 xfs_zero_remaining_bytes(
1105 xfs_bmbt_irec_t imap;
1106 xfs_fileoff_t offset_fsb;
1107 xfs_off_t lastoffset;
1110 xfs_mount_t *mp = ip->i_mount;
1115 * Avoid doing I/O beyond eof - it's not necessary
1116 * since nothing can read beyond eof. The space will
1117 * be zeroed when the file is extended anyway.
1119 if (startoff >= XFS_ISIZE(ip))
1122 if (endoff > XFS_ISIZE(ip))
1123 endoff = XFS_ISIZE(ip);
1125 bp = xfs_buf_get_uncached(XFS_IS_REALTIME_INODE(ip) ?
1126 mp->m_rtdev_targp : mp->m_ddev_targp,
1127 BTOBB(mp->m_sb.sb_blocksize), 0);
1133 for (offset = startoff; offset <= endoff; offset = lastoffset + 1) {
1136 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1139 lock_mode = xfs_ilock_data_map_shared(ip);
1140 error = xfs_bmapi_read(ip, offset_fsb, 1, &imap, &nimap, 0);
1141 xfs_iunlock(ip, lock_mode);
1143 if (error || nimap < 1)
1145 ASSERT(imap.br_blockcount >= 1);
1146 ASSERT(imap.br_startoff == offset_fsb);
1147 lastoffset = XFS_FSB_TO_B(mp, imap.br_startoff + 1) - 1;
1148 if (lastoffset > endoff)
1149 lastoffset = endoff;
1150 if (imap.br_startblock == HOLESTARTBLOCK)
1152 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1153 if (imap.br_state == XFS_EXT_UNWRITTEN)
1156 XFS_BUF_UNWRITE(bp);
1158 XFS_BUF_SET_ADDR(bp, xfs_fsb_to_db(ip, imap.br_startblock));
1160 if (XFS_FORCED_SHUTDOWN(mp)) {
1164 xfs_buf_iorequest(bp);
1165 error = xfs_buf_iowait(bp);
1167 xfs_buf_ioerror_alert(bp,
1168 "xfs_zero_remaining_bytes(read)");
1172 (offset - XFS_FSB_TO_B(mp, imap.br_startoff)),
1173 0, lastoffset - offset + 1);
1178 if (XFS_FORCED_SHUTDOWN(mp)) {
1182 xfs_buf_iorequest(bp);
1183 error = xfs_buf_iowait(bp);
1185 xfs_buf_ioerror_alert(bp,
1186 "xfs_zero_remaining_bytes(write)");
1195 xfs_free_file_space(
1196 struct xfs_inode *ip,
1202 xfs_fileoff_t endoffset_fsb;
1204 xfs_fsblock_t firstfsb;
1205 xfs_bmap_free_t free_list;
1206 xfs_bmbt_irec_t imap;
1214 xfs_fileoff_t startoffset_fsb;
1219 trace_xfs_free_file_space(ip);
1221 error = xfs_qm_dqattach(ip, 0);
1226 if (len <= 0) /* if nothing being freed */
1228 rt = XFS_IS_REALTIME_INODE(ip);
1229 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
1230 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
1232 /* wait for the completion of any pending DIOs */
1233 inode_dio_wait(VFS_I(ip));
1235 rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
1236 ioffset = offset & ~(rounding - 1);
1237 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
1241 truncate_pagecache_range(VFS_I(ip), ioffset, -1);
1244 * Need to zero the stuff we're not freeing, on disk.
1245 * If it's a realtime file & can't use unwritten extents then we
1246 * actually need to zero the extent edges. Otherwise xfs_bunmapi
1247 * will take care of it for us.
1249 if (rt && !xfs_sb_version_hasextflgbit(&mp->m_sb)) {
1251 error = xfs_bmapi_read(ip, startoffset_fsb, 1,
1255 ASSERT(nimap == 0 || nimap == 1);
1256 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1259 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1260 block = imap.br_startblock;
1261 mod = do_div(block, mp->m_sb.sb_rextsize);
1263 startoffset_fsb += mp->m_sb.sb_rextsize - mod;
1266 error = xfs_bmapi_read(ip, endoffset_fsb - 1, 1,
1270 ASSERT(nimap == 0 || nimap == 1);
1271 if (nimap && imap.br_startblock != HOLESTARTBLOCK) {
1272 ASSERT(imap.br_startblock != DELAYSTARTBLOCK);
1274 if (mod && (mod != mp->m_sb.sb_rextsize))
1275 endoffset_fsb -= mod;
1278 if ((done = (endoffset_fsb <= startoffset_fsb)))
1280 * One contiguous piece to clear
1282 error = xfs_zero_remaining_bytes(ip, offset, offset + len - 1);
1285 * Some full blocks, possibly two pieces to clear
1287 if (offset < XFS_FSB_TO_B(mp, startoffset_fsb))
1288 error = xfs_zero_remaining_bytes(ip, offset,
1289 XFS_FSB_TO_B(mp, startoffset_fsb) - 1);
1291 XFS_FSB_TO_B(mp, endoffset_fsb) < offset + len)
1292 error = xfs_zero_remaining_bytes(ip,
1293 XFS_FSB_TO_B(mp, endoffset_fsb),
1298 * free file space until done or until there is an error
1300 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
1301 while (!error && !done) {
1304 * allocate and setup the transaction. Allow this
1305 * transaction to dip into the reserve blocks to ensure
1306 * the freeing of the space succeeds at ENOSPC.
1308 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1309 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write, resblks, 0);
1312 * check for running out of space
1316 * Free the transaction structure.
1318 ASSERT(error == -ENOSPC || XFS_FORCED_SHUTDOWN(mp));
1319 xfs_trans_cancel(tp, 0);
1322 xfs_ilock(ip, XFS_ILOCK_EXCL);
1323 error = xfs_trans_reserve_quota(tp, mp,
1324 ip->i_udquot, ip->i_gdquot, ip->i_pdquot,
1325 resblks, 0, XFS_QMOPT_RES_REGBLKS);
1329 xfs_trans_ijoin(tp, ip, 0);
1332 * issue the bunmapi() call to free the blocks
1334 xfs_bmap_init(&free_list, &firstfsb);
1335 error = xfs_bunmapi(tp, ip, startoffset_fsb,
1336 endoffset_fsb - startoffset_fsb,
1337 0, 2, &firstfsb, &free_list, &done);
1343 * complete the transaction
1345 error = xfs_bmap_finish(&tp, &free_list, &committed);
1350 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1351 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1358 xfs_bmap_cancel(&free_list);
1360 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1361 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1367 xfs_zero_file_space(
1368 struct xfs_inode *ip,
1372 struct xfs_mount *mp = ip->i_mount;
1374 xfs_off_t start_boundary;
1375 xfs_off_t end_boundary;
1378 trace_xfs_zero_file_space(ip);
1380 granularity = max_t(uint, 1 << mp->m_sb.sb_blocklog, PAGE_CACHE_SIZE);
1383 * Round the range of extents we are going to convert inwards. If the
1384 * offset is aligned, then it doesn't get changed so we zero from the
1385 * start of the block offset points to.
1387 start_boundary = round_up(offset, granularity);
1388 end_boundary = round_down(offset + len, granularity);
1390 ASSERT(start_boundary >= offset);
1391 ASSERT(end_boundary <= offset + len);
1393 if (start_boundary < end_boundary - 1) {
1395 * punch out delayed allocation blocks and the page cache over
1396 * the conversion range
1398 xfs_ilock(ip, XFS_ILOCK_EXCL);
1399 error = xfs_bmap_punch_delalloc_range(ip,
1400 XFS_B_TO_FSBT(mp, start_boundary),
1401 XFS_B_TO_FSB(mp, end_boundary - start_boundary));
1402 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1403 truncate_pagecache_range(VFS_I(ip), start_boundary,
1406 /* convert the blocks */
1407 error = xfs_alloc_file_space(ip, start_boundary,
1408 end_boundary - start_boundary - 1,
1409 XFS_BMAPI_PREALLOC | XFS_BMAPI_CONVERT);
1413 /* We've handled the interior of the range, now for the edges */
1414 if (start_boundary != offset) {
1415 error = xfs_iozero(ip, offset, start_boundary - offset);
1420 if (end_boundary != offset + len)
1421 error = xfs_iozero(ip, end_boundary,
1422 offset + len - end_boundary);
1426 * It's either a sub-granularity range or the range spanned lies
1427 * partially across two adjacent blocks.
1429 error = xfs_iozero(ip, offset, len);
1438 * xfs_collapse_file_space()
1439 * This routine frees disk space and shift extent for the given file.
1440 * The first thing we do is to free data blocks in the specified range
1441 * by calling xfs_free_file_space(). It would also sync dirty data
1442 * and invalidate page cache over the region on which collapse range
1443 * is working. And Shift extent records to the left to cover a hole.
1450 xfs_collapse_file_space(
1451 struct xfs_inode *ip,
1456 struct xfs_mount *mp = ip->i_mount;
1457 struct xfs_trans *tp;
1459 struct xfs_bmap_free free_list;
1460 xfs_fsblock_t first_block;
1462 xfs_fileoff_t start_fsb;
1463 xfs_fileoff_t next_fsb;
1464 xfs_fileoff_t shift_fsb;
1466 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1468 trace_xfs_collapse_file_space(ip);
1470 next_fsb = XFS_B_TO_FSB(mp, offset + len);
1471 shift_fsb = XFS_B_TO_FSB(mp, len);
1474 * Writeback the entire file and force remove any post-eof blocks. The
1475 * writeback prevents changes to the extent list via concurrent
1476 * writeback and the eofblocks trim prevents the extent shift algorithm
1477 * from running into a post-eof delalloc extent.
1479 * XXX: This is a temporary fix until the extent shift loop below is
1480 * converted to use offsets and lookups within the ILOCK rather than
1481 * carrying around the index into the extent list for the next
1484 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1487 if (xfs_can_free_eofblocks(ip, true)) {
1488 error = xfs_free_eofblocks(mp, ip, false);
1493 error = xfs_free_file_space(ip, offset, len);
1497 while (!error && !done) {
1498 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
1500 * We would need to reserve permanent block for transaction.
1501 * This will come into picture when after shifting extent into
1502 * hole we found that adjacent extents can be merged which
1503 * may lead to freeing of a block during record update.
1505 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_write,
1506 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0);
1508 xfs_trans_cancel(tp, 0);
1512 xfs_ilock(ip, XFS_ILOCK_EXCL);
1513 error = xfs_trans_reserve_quota(tp, mp, ip->i_udquot,
1514 ip->i_gdquot, ip->i_pdquot,
1515 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0,
1516 XFS_QMOPT_RES_REGBLKS);
1520 xfs_trans_ijoin(tp, ip, 0);
1522 xfs_bmap_init(&free_list, &first_block);
1525 * We are using the write transaction in which max 2 bmbt
1526 * updates are allowed
1528 start_fsb = next_fsb;
1529 error = xfs_bmap_shift_extents(tp, ip, start_fsb, shift_fsb,
1530 &done, &next_fsb, &first_block, &free_list,
1531 XFS_BMAP_MAX_SHIFT_EXTENTS);
1535 error = xfs_bmap_finish(&tp, &free_list, &committed);
1539 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
1540 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1546 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
1547 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1552 * We need to check that the format of the data fork in the temporary inode is
1553 * valid for the target inode before doing the swap. This is not a problem with
1554 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1555 * data fork depending on the space the attribute fork is taking so we can get
1556 * invalid formats on the target inode.
1558 * E.g. target has space for 7 extents in extent format, temp inode only has
1559 * space for 6. If we defragment down to 7 extents, then the tmp format is a
1560 * btree, but when swapped it needs to be in extent format. Hence we can't just
1561 * blindly swap data forks on attr2 filesystems.
1563 * Note that we check the swap in both directions so that we don't end up with
1564 * a corrupt temporary inode, either.
1566 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1567 * inode will prevent this situation from occurring, so all we do here is
1568 * reject and log the attempt. basically we are putting the responsibility on
1569 * userspace to get this right.
1572 xfs_swap_extents_check_format(
1573 xfs_inode_t *ip, /* target inode */
1574 xfs_inode_t *tip) /* tmp inode */
1577 /* Should never get a local format */
1578 if (ip->i_d.di_format == XFS_DINODE_FMT_LOCAL ||
1579 tip->i_d.di_format == XFS_DINODE_FMT_LOCAL)
1583 * if the target inode has less extents that then temporary inode then
1584 * why did userspace call us?
1586 if (ip->i_d.di_nextents < tip->i_d.di_nextents)
1590 * if the target inode is in extent form and the temp inode is in btree
1591 * form then we will end up with the target inode in the wrong format
1592 * as we already know there are less extents in the temp inode.
1594 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1595 tip->i_d.di_format == XFS_DINODE_FMT_BTREE)
1598 /* Check temp in extent form to max in target */
1599 if (tip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1600 XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) >
1601 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1604 /* Check target in extent form to max in temp */
1605 if (ip->i_d.di_format == XFS_DINODE_FMT_EXTENTS &&
1606 XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) >
1607 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1611 * If we are in a btree format, check that the temp root block will fit
1612 * in the target and that it has enough extents to be in btree format
1615 * Note that we have to be careful to allow btree->extent conversions
1616 * (a common defrag case) which will occur when the temp inode is in
1619 if (tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1620 if (XFS_IFORK_BOFF(ip) &&
1621 XFS_BMAP_BMDR_SPACE(tip->i_df.if_broot) > XFS_IFORK_BOFF(ip))
1623 if (XFS_IFORK_NEXTENTS(tip, XFS_DATA_FORK) <=
1624 XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1628 /* Reciprocal target->temp btree format checks */
1629 if (ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1630 if (XFS_IFORK_BOFF(tip) &&
1631 XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip))
1633 if (XFS_IFORK_NEXTENTS(ip, XFS_DATA_FORK) <=
1634 XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1642 xfs_swap_extent_flush(
1643 struct xfs_inode *ip)
1647 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1650 truncate_pagecache_range(VFS_I(ip), 0, -1);
1652 /* Verify O_DIRECT for ftmp */
1653 if (VFS_I(ip)->i_mapping->nrpages)
1657 * Don't try to swap extents on mmap()d files because we can't lock
1658 * out races against page faults safely.
1660 if (mapping_mapped(VFS_I(ip)->i_mapping))
1667 xfs_inode_t *ip, /* target inode */
1668 xfs_inode_t *tip, /* tmp inode */
1671 xfs_mount_t *mp = ip->i_mount;
1673 xfs_bstat_t *sbp = &sxp->sx_stat;
1674 xfs_ifork_t *tempifp, *ifp, *tifp;
1675 int src_log_flags, target_log_flags;
1682 tempifp = kmem_alloc(sizeof(xfs_ifork_t), KM_MAYFAIL);
1689 * Lock up the inodes against other IO and truncate to begin with.
1690 * Then we can ensure the inodes are flushed and have no page cache
1691 * safely. Once we have done this we can take the ilocks and do the rest
1694 lock_flags = XFS_IOLOCK_EXCL;
1695 xfs_lock_two_inodes(ip, tip, XFS_IOLOCK_EXCL);
1697 /* Verify that both files have the same format */
1698 if ((ip->i_d.di_mode & S_IFMT) != (tip->i_d.di_mode & S_IFMT)) {
1703 /* Verify both files are either real-time or non-realtime */
1704 if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
1709 error = xfs_swap_extent_flush(ip);
1712 error = xfs_swap_extent_flush(tip);
1716 tp = xfs_trans_alloc(mp, XFS_TRANS_SWAPEXT);
1717 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_ichange, 0, 0);
1719 xfs_trans_cancel(tp, 0);
1722 xfs_lock_two_inodes(ip, tip, XFS_ILOCK_EXCL);
1723 lock_flags |= XFS_ILOCK_EXCL;
1725 /* Verify all data are being swapped */
1726 if (sxp->sx_offset != 0 ||
1727 sxp->sx_length != ip->i_d.di_size ||
1728 sxp->sx_length != tip->i_d.di_size) {
1730 goto out_trans_cancel;
1733 trace_xfs_swap_extent_before(ip, 0);
1734 trace_xfs_swap_extent_before(tip, 1);
1736 /* check inode formats now that data is flushed */
1737 error = xfs_swap_extents_check_format(ip, tip);
1740 "%s: inode 0x%llx format is incompatible for exchanging.",
1741 __func__, ip->i_ino);
1742 goto out_trans_cancel;
1746 * Compare the current change & modify times with that
1747 * passed in. If they differ, we abort this swap.
1748 * This is the mechanism used to ensure the calling
1749 * process that the file was not changed out from
1752 if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
1753 (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
1754 (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
1755 (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
1757 goto out_trans_cancel;
1760 * Count the number of extended attribute blocks
1762 if ( ((XFS_IFORK_Q(ip) != 0) && (ip->i_d.di_anextents > 0)) &&
1763 (ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1764 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &aforkblks);
1766 goto out_trans_cancel;
1768 if ( ((XFS_IFORK_Q(tip) != 0) && (tip->i_d.di_anextents > 0)) &&
1769 (tip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL)) {
1770 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK,
1773 goto out_trans_cancel;
1776 xfs_trans_ijoin(tp, ip, lock_flags);
1777 xfs_trans_ijoin(tp, tip, lock_flags);
1780 * Before we've swapped the forks, lets set the owners of the forks
1781 * appropriately. We have to do this as we are demand paging the btree
1782 * buffers, and so the validation done on read will expect the owner
1783 * field to be correctly set. Once we change the owners, we can swap the
1786 * Note the trickiness in setting the log flags - we set the owner log
1787 * flag on the opposite inode (i.e. the inode we are setting the new
1788 * owner to be) because once we swap the forks and log that, log
1789 * recovery is going to see the fork as owned by the swapped inode,
1790 * not the pre-swapped inodes.
1792 src_log_flags = XFS_ILOG_CORE;
1793 target_log_flags = XFS_ILOG_CORE;
1794 if (ip->i_d.di_version == 3 &&
1795 ip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1796 target_log_flags |= XFS_ILOG_DOWNER;
1797 error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK,
1800 goto out_trans_cancel;
1803 if (tip->i_d.di_version == 3 &&
1804 tip->i_d.di_format == XFS_DINODE_FMT_BTREE) {
1805 src_log_flags |= XFS_ILOG_DOWNER;
1806 error = xfs_bmbt_change_owner(tp, tip, XFS_DATA_FORK,
1809 goto out_trans_cancel;
1813 * Swap the data forks of the inodes
1817 *tempifp = *ifp; /* struct copy */
1818 *ifp = *tifp; /* struct copy */
1819 *tifp = *tempifp; /* struct copy */
1822 * Fix the on-disk inode values
1824 tmp = (__uint64_t)ip->i_d.di_nblocks;
1825 ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
1826 tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
1828 tmp = (__uint64_t) ip->i_d.di_nextents;
1829 ip->i_d.di_nextents = tip->i_d.di_nextents;
1830 tip->i_d.di_nextents = tmp;
1832 tmp = (__uint64_t) ip->i_d.di_format;
1833 ip->i_d.di_format = tip->i_d.di_format;
1834 tip->i_d.di_format = tmp;
1837 * The extents in the source inode could still contain speculative
1838 * preallocation beyond EOF (e.g. the file is open but not modified
1839 * while defrag is in progress). In that case, we need to copy over the
1840 * number of delalloc blocks the data fork in the source inode is
1841 * tracking beyond EOF so that when the fork is truncated away when the
1842 * temporary inode is unlinked we don't underrun the i_delayed_blks
1843 * counter on that inode.
1845 ASSERT(tip->i_delayed_blks == 0);
1846 tip->i_delayed_blks = ip->i_delayed_blks;
1847 ip->i_delayed_blks = 0;
1849 switch (ip->i_d.di_format) {
1850 case XFS_DINODE_FMT_EXTENTS:
1851 /* If the extents fit in the inode, fix the
1852 * pointer. Otherwise it's already NULL or
1853 * pointing to the extent.
1855 if (ip->i_d.di_nextents <= XFS_INLINE_EXTS) {
1856 ifp->if_u1.if_extents =
1857 ifp->if_u2.if_inline_ext;
1859 src_log_flags |= XFS_ILOG_DEXT;
1861 case XFS_DINODE_FMT_BTREE:
1862 ASSERT(ip->i_d.di_version < 3 ||
1863 (src_log_flags & XFS_ILOG_DOWNER));
1864 src_log_flags |= XFS_ILOG_DBROOT;
1868 switch (tip->i_d.di_format) {
1869 case XFS_DINODE_FMT_EXTENTS:
1870 /* If the extents fit in the inode, fix the
1871 * pointer. Otherwise it's already NULL or
1872 * pointing to the extent.
1874 if (tip->i_d.di_nextents <= XFS_INLINE_EXTS) {
1875 tifp->if_u1.if_extents =
1876 tifp->if_u2.if_inline_ext;
1878 target_log_flags |= XFS_ILOG_DEXT;
1880 case XFS_DINODE_FMT_BTREE:
1881 target_log_flags |= XFS_ILOG_DBROOT;
1882 ASSERT(tip->i_d.di_version < 3 ||
1883 (target_log_flags & XFS_ILOG_DOWNER));
1887 xfs_trans_log_inode(tp, ip, src_log_flags);
1888 xfs_trans_log_inode(tp, tip, target_log_flags);
1891 * If this is a synchronous mount, make sure that the
1892 * transaction goes to disk before returning to the user.
1894 if (mp->m_flags & XFS_MOUNT_WSYNC)
1895 xfs_trans_set_sync(tp);
1897 error = xfs_trans_commit(tp, 0);
1899 trace_xfs_swap_extent_after(ip, 0);
1900 trace_xfs_swap_extent_after(tip, 1);
1906 xfs_iunlock(ip, lock_flags);
1907 xfs_iunlock(tip, lock_flags);
1911 xfs_trans_cancel(tp, 0);