1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * Copyright (c) 2012 Red Hat, Inc.
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
14 #include "xfs_mount.h"
15 #include "xfs_defer.h"
16 #include "xfs_inode.h"
17 #include "xfs_btree.h"
18 #include "xfs_trans.h"
19 #include "xfs_alloc.h"
21 #include "xfs_bmap_util.h"
22 #include "xfs_bmap_btree.h"
23 #include "xfs_rtalloc.h"
24 #include "xfs_error.h"
25 #include "xfs_quota.h"
26 #include "xfs_trans_space.h"
27 #include "xfs_trace.h"
28 #include "xfs_icache.h"
29 #include "xfs_iomap.h"
30 #include "xfs_reflink.h"
31 #include "xfs_rtbitmap.h"
33 /* Kernel only BMAP related definitions and functions */
36 * Convert the given file system block to a disk block. We have to treat it
37 * differently based on whether the file is a real time file or not, because the
41 xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb)
43 if (XFS_IS_REALTIME_INODE(ip))
44 return XFS_FSB_TO_BB(ip->i_mount, fsb);
45 return XFS_FSB_TO_DADDR(ip->i_mount, fsb);
49 * Routine to zero an extent on disk allocated to the specific inode.
51 * The VFS functions take a linearised filesystem block offset, so we have to
52 * convert the sparse xfs fsb to the right format first.
53 * VFS types are real funky, too.
58 xfs_fsblock_t start_fsb,
61 struct xfs_mount *mp = ip->i_mount;
62 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
63 xfs_daddr_t sector = xfs_fsb_to_db(ip, start_fsb);
64 sector_t block = XFS_BB_TO_FSBT(mp, sector);
66 return blkdev_issue_zeroout(target->bt_bdev,
67 block << (mp->m_super->s_blocksize_bits - 9),
68 count_fsb << (mp->m_super->s_blocksize_bits - 9),
73 * Extent tree block counting routines.
77 * Count leaf blocks given a range of extent records. Delayed allocation
78 * extents are not counted towards the totals.
81 xfs_bmap_count_leaves(
82 struct xfs_ifork *ifp,
85 struct xfs_iext_cursor icur;
86 struct xfs_bmbt_irec got;
87 xfs_extnum_t numrecs = 0;
89 for_each_xfs_iext(ifp, &icur, &got) {
90 if (!isnullstartblock(got.br_startblock)) {
91 *count += got.br_blockcount;
100 * Count fsblocks of the given fork. Delayed allocation extents are
101 * not counted towards the totals.
104 xfs_bmap_count_blocks(
105 struct xfs_trans *tp,
106 struct xfs_inode *ip,
108 xfs_extnum_t *nextents,
109 xfs_filblks_t *count)
111 struct xfs_mount *mp = ip->i_mount;
112 struct xfs_ifork *ifp = xfs_ifork_ptr(ip, whichfork);
113 struct xfs_btree_cur *cur;
114 xfs_extlen_t btblocks = 0;
123 switch (ifp->if_format) {
124 case XFS_DINODE_FMT_BTREE:
125 error = xfs_iread_extents(tp, ip, whichfork);
129 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
130 error = xfs_btree_count_blocks(cur, &btblocks);
131 xfs_btree_del_cursor(cur, error);
136 * xfs_btree_count_blocks includes the root block contained in
137 * the inode fork in @btblocks, so subtract one because we're
138 * only interested in allocated disk blocks.
140 *count += btblocks - 1;
143 case XFS_DINODE_FMT_EXTENTS:
144 *nextents = xfs_bmap_count_leaves(ifp, count);
152 xfs_getbmap_report_one(
153 struct xfs_inode *ip,
154 struct getbmapx *bmv,
155 struct kgetbmap *out,
157 struct xfs_bmbt_irec *got)
159 struct kgetbmap *p = out + bmv->bmv_entries;
163 error = xfs_reflink_trim_around_shared(ip, got, &shared);
167 if (isnullstartblock(got->br_startblock) ||
168 got->br_startblock == DELAYSTARTBLOCK) {
170 * Take the flush completion as being a point-in-time snapshot
171 * where there are no delalloc extents, and if any new ones
172 * have been created racily, just skip them as being 'after'
173 * the flush and so don't get reported.
175 if (!(bmv->bmv_iflags & BMV_IF_DELALLOC))
178 p->bmv_oflags |= BMV_OF_DELALLOC;
181 p->bmv_block = xfs_fsb_to_db(ip, got->br_startblock);
184 if (got->br_state == XFS_EXT_UNWRITTEN &&
185 (bmv->bmv_iflags & BMV_IF_PREALLOC))
186 p->bmv_oflags |= BMV_OF_PREALLOC;
189 p->bmv_oflags |= BMV_OF_SHARED;
191 p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, got->br_startoff);
192 p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, got->br_blockcount);
194 bmv->bmv_offset = p->bmv_offset + p->bmv_length;
195 bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset);
201 xfs_getbmap_report_hole(
202 struct xfs_inode *ip,
203 struct getbmapx *bmv,
204 struct kgetbmap *out,
209 struct kgetbmap *p = out + bmv->bmv_entries;
211 if (bmv->bmv_iflags & BMV_IF_NO_HOLES)
215 p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, bno);
216 p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, end - bno);
218 bmv->bmv_offset = p->bmv_offset + p->bmv_length;
219 bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset);
225 struct getbmapx *bmv)
227 return bmv->bmv_length == 0 || bmv->bmv_entries >= bmv->bmv_count - 1;
231 xfs_getbmap_next_rec(
232 struct xfs_bmbt_irec *rec,
233 xfs_fileoff_t total_end)
235 xfs_fileoff_t end = rec->br_startoff + rec->br_blockcount;
237 if (end == total_end)
240 rec->br_startoff += rec->br_blockcount;
241 if (!isnullstartblock(rec->br_startblock) &&
242 rec->br_startblock != DELAYSTARTBLOCK)
243 rec->br_startblock += rec->br_blockcount;
244 rec->br_blockcount = total_end - end;
249 * Get inode's extents as described in bmv, and format for output.
250 * Calls formatter to fill the user's buffer until all extents
251 * are mapped, until the passed-in bmv->bmv_count slots have
252 * been filled, or until the formatter short-circuits the loop,
253 * if it is tracking filled-in extents on its own.
257 struct xfs_inode *ip,
258 struct getbmapx *bmv, /* user bmap structure */
259 struct kgetbmap *out)
261 struct xfs_mount *mp = ip->i_mount;
262 int iflags = bmv->bmv_iflags;
263 int whichfork, lock, error = 0;
264 int64_t bmv_end, max_len;
265 xfs_fileoff_t bno, first_bno;
266 struct xfs_ifork *ifp;
267 struct xfs_bmbt_irec got, rec;
269 struct xfs_iext_cursor icur;
271 if (bmv->bmv_iflags & ~BMV_IF_VALID)
274 /* Only allow CoW fork queries if we're debugging. */
275 if (iflags & BMV_IF_COWFORK)
278 if ((iflags & BMV_IF_ATTRFORK) && (iflags & BMV_IF_COWFORK))
281 if (bmv->bmv_length < -1)
283 bmv->bmv_entries = 0;
284 if (bmv->bmv_length == 0)
287 if (iflags & BMV_IF_ATTRFORK)
288 whichfork = XFS_ATTR_FORK;
289 else if (iflags & BMV_IF_COWFORK)
290 whichfork = XFS_COW_FORK;
292 whichfork = XFS_DATA_FORK;
294 xfs_ilock(ip, XFS_IOLOCK_SHARED);
297 lock = xfs_ilock_attr_map_shared(ip);
298 if (!xfs_inode_has_attr_fork(ip))
299 goto out_unlock_ilock;
304 lock = XFS_ILOCK_SHARED;
307 /* No CoW fork? Just return */
308 if (!xfs_ifork_ptr(ip, whichfork))
309 goto out_unlock_ilock;
311 if (xfs_get_cowextsz_hint(ip))
312 max_len = mp->m_super->s_maxbytes;
314 max_len = XFS_ISIZE(ip);
317 if (!(iflags & BMV_IF_DELALLOC) &&
318 (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_disk_size)) {
319 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
321 goto out_unlock_iolock;
324 * Even after flushing the inode, there can still be
325 * delalloc blocks on the inode beyond EOF due to
326 * speculative preallocation. These are not removed
327 * until the release function is called or the inode
328 * is inactivated. Hence we cannot assert here that
329 * ip->i_delayed_blks == 0.
333 if (xfs_get_extsz_hint(ip) ||
335 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))
336 max_len = mp->m_super->s_maxbytes;
338 max_len = XFS_ISIZE(ip);
340 lock = xfs_ilock_data_map_shared(ip);
344 ifp = xfs_ifork_ptr(ip, whichfork);
346 switch (ifp->if_format) {
347 case XFS_DINODE_FMT_EXTENTS:
348 case XFS_DINODE_FMT_BTREE:
350 case XFS_DINODE_FMT_LOCAL:
351 /* Local format inode forks report no extents. */
352 goto out_unlock_ilock;
355 goto out_unlock_ilock;
358 if (bmv->bmv_length == -1) {
359 max_len = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, max_len));
360 bmv->bmv_length = max(0LL, max_len - bmv->bmv_offset);
363 bmv_end = bmv->bmv_offset + bmv->bmv_length;
365 first_bno = bno = XFS_BB_TO_FSBT(mp, bmv->bmv_offset);
366 len = XFS_BB_TO_FSB(mp, bmv->bmv_length);
368 error = xfs_iread_extents(NULL, ip, whichfork);
370 goto out_unlock_ilock;
372 if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
374 * Report a whole-file hole if the delalloc flag is set to
375 * stay compatible with the old implementation.
377 if (iflags & BMV_IF_DELALLOC)
378 xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno,
379 XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
380 goto out_unlock_ilock;
383 while (!xfs_getbmap_full(bmv)) {
384 xfs_trim_extent(&got, first_bno, len);
387 * Report an entry for a hole if this extent doesn't directly
388 * follow the previous one.
390 if (got.br_startoff > bno) {
391 xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno,
393 if (xfs_getbmap_full(bmv))
398 * In order to report shared extents accurately, we report each
399 * distinct shared / unshared part of a single bmbt record with
400 * an individual getbmapx record.
402 bno = got.br_startoff + got.br_blockcount;
405 error = xfs_getbmap_report_one(ip, bmv, out, bmv_end,
407 if (error || xfs_getbmap_full(bmv))
408 goto out_unlock_ilock;
409 } while (xfs_getbmap_next_rec(&rec, bno));
411 if (!xfs_iext_next_extent(ifp, &icur, &got)) {
412 xfs_fileoff_t end = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
414 if (bmv->bmv_entries > 0)
415 out[bmv->bmv_entries - 1].bmv_oflags |=
418 if (whichfork != XFS_ATTR_FORK && bno < end &&
419 !xfs_getbmap_full(bmv)) {
420 xfs_getbmap_report_hole(ip, bmv, out, bmv_end,
426 if (bno >= first_bno + len)
431 xfs_iunlock(ip, lock);
433 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
438 * Dead simple method of punching delalyed allocation blocks from a range in
439 * the inode. This will always punch out both the start and end blocks, even
440 * if the ranges only partially overlap them, so it is up to the caller to
441 * ensure that partial blocks are not passed in.
444 xfs_bmap_punch_delalloc_range(
445 struct xfs_inode *ip,
446 xfs_off_t start_byte,
449 struct xfs_mount *mp = ip->i_mount;
450 struct xfs_ifork *ifp = &ip->i_df;
451 xfs_fileoff_t start_fsb = XFS_B_TO_FSBT(mp, start_byte);
452 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, end_byte);
453 struct xfs_bmbt_irec got, del;
454 struct xfs_iext_cursor icur;
457 ASSERT(!xfs_need_iread_extents(ifp));
459 xfs_ilock(ip, XFS_ILOCK_EXCL);
460 if (!xfs_iext_lookup_extent_before(ip, ifp, &end_fsb, &icur, &got))
463 while (got.br_startoff + got.br_blockcount > start_fsb) {
465 xfs_trim_extent(&del, start_fsb, end_fsb - start_fsb);
468 * A delete can push the cursor forward. Step back to the
469 * previous extent on non-delalloc or extents outside the
472 if (!del.br_blockcount ||
473 !isnullstartblock(del.br_startblock)) {
474 if (!xfs_iext_prev_extent(ifp, &icur, &got))
479 error = xfs_bmap_del_extent_delay(ip, XFS_DATA_FORK, &icur,
481 if (error || !xfs_iext_get_extent(ifp, &icur, &got))
486 xfs_iunlock(ip, XFS_ILOCK_EXCL);
491 * Test whether it is appropriate to check an inode for and free post EOF
492 * blocks. The 'force' parameter determines whether we should also consider
493 * regular files that are marked preallocated or append-only.
496 xfs_can_free_eofblocks(
497 struct xfs_inode *ip,
500 struct xfs_bmbt_irec imap;
501 struct xfs_mount *mp = ip->i_mount;
502 xfs_fileoff_t end_fsb;
503 xfs_fileoff_t last_fsb;
508 * Caller must either hold the exclusive io lock; or be inactivating
509 * the inode, which guarantees there are no other users of the inode.
511 if (!(VFS_I(ip)->i_state & I_FREEING))
512 xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL);
514 /* prealloc/delalloc exists only on regular files */
515 if (!S_ISREG(VFS_I(ip)->i_mode))
519 * Zero sized files with no cached pages and delalloc blocks will not
520 * have speculative prealloc/delalloc blocks to remove.
522 if (VFS_I(ip)->i_size == 0 &&
523 VFS_I(ip)->i_mapping->nrpages == 0 &&
524 ip->i_delayed_blks == 0)
527 /* If we haven't read in the extent list, then don't do it now. */
528 if (xfs_need_iread_extents(&ip->i_df))
532 * Do not free real preallocated or append-only files unless the file
533 * has delalloc blocks and we are forced to remove them.
535 if (ip->i_diflags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
536 if (!force || ip->i_delayed_blks == 0)
540 * Do not try to free post-EOF blocks if EOF is beyond the end of the
541 * range supported by the page cache, because the truncation will loop
544 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
545 if (XFS_IS_REALTIME_INODE(ip) && mp->m_sb.sb_rextsize > 1)
546 end_fsb = xfs_rtb_roundup_rtx(mp, end_fsb);
547 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
548 if (last_fsb <= end_fsb)
552 * Look up the mapping for the first block past EOF. If we can't find
553 * it, there's nothing to free.
555 xfs_ilock(ip, XFS_ILOCK_SHARED);
556 error = xfs_bmapi_read(ip, end_fsb, last_fsb - end_fsb, &imap, &nimaps,
558 xfs_iunlock(ip, XFS_ILOCK_SHARED);
559 if (error || nimaps == 0)
563 * If there's a real mapping there or there are delayed allocation
564 * reservations, then we have post-EOF blocks to try to free.
566 return imap.br_startblock != HOLESTARTBLOCK || ip->i_delayed_blks;
570 * This is called to free any blocks beyond eof. The caller must hold
571 * IOLOCK_EXCL unless we are in the inode reclaim path and have the only
572 * reference to the inode.
576 struct xfs_inode *ip)
578 struct xfs_trans *tp;
579 struct xfs_mount *mp = ip->i_mount;
582 /* Attach the dquots to the inode up front. */
583 error = xfs_qm_dqattach(ip);
587 /* Wait on dio to ensure i_size has settled. */
588 inode_dio_wait(VFS_I(ip));
590 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
592 ASSERT(xfs_is_shutdown(mp));
596 xfs_ilock(ip, XFS_ILOCK_EXCL);
597 xfs_trans_ijoin(tp, ip, 0);
600 * Do not update the on-disk file size. If we update the on-disk file
601 * size and then the system crashes before the contents of the file are
602 * flushed to disk then the files may be full of holes (ie NULL files
605 error = xfs_itruncate_extents_flags(&tp, ip, XFS_DATA_FORK,
606 XFS_ISIZE(ip), XFS_BMAPI_NODISCARD);
610 error = xfs_trans_commit(tp);
614 xfs_inode_clear_eofblocks_tag(ip);
619 * If we get an error at this point we simply don't
620 * bother truncating the file.
622 xfs_trans_cancel(tp);
624 xfs_iunlock(ip, XFS_ILOCK_EXCL);
629 xfs_alloc_file_space(
630 struct xfs_inode *ip,
634 xfs_mount_t *mp = ip->i_mount;
636 xfs_filblks_t allocatesize_fsb;
637 xfs_extlen_t extsz, temp;
638 xfs_fileoff_t startoffset_fsb;
639 xfs_fileoff_t endoffset_fsb;
642 xfs_bmbt_irec_t imaps[1], *imapp;
645 trace_xfs_alloc_file_space(ip);
647 if (xfs_is_shutdown(mp))
650 error = xfs_qm_dqattach(ip);
657 rt = XFS_IS_REALTIME_INODE(ip);
658 extsz = xfs_get_extsz_hint(ip);
662 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
663 endoffset_fsb = XFS_B_TO_FSB(mp, offset + count);
664 allocatesize_fsb = endoffset_fsb - startoffset_fsb;
667 * Allocate file space until done or until there is an error
669 while (allocatesize_fsb && !error) {
671 unsigned int dblocks, rblocks, resblks;
675 * Determine space reservations for data/realtime.
677 if (unlikely(extsz)) {
681 e = startoffset_fsb + allocatesize_fsb;
682 div_u64_rem(startoffset_fsb, extsz, &temp);
685 div_u64_rem(e, extsz, &temp);
690 e = allocatesize_fsb;
694 * The transaction reservation is limited to a 32-bit block
695 * count, hence we need to limit the number of blocks we are
696 * trying to reserve to avoid an overflow. We can't allocate
697 * more than @nimaps extents, and an extent is limited on disk
698 * to XFS_BMBT_MAX_EXTLEN (21 bits), so use that to enforce the
701 resblks = min_t(xfs_fileoff_t, (e - s),
702 (XFS_MAX_BMBT_EXTLEN * nimaps));
704 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
707 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
711 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write,
712 dblocks, rblocks, false, &tp);
716 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
717 XFS_IEXT_ADD_NOSPLIT_CNT);
719 error = xfs_iext_count_upgrade(tp, ip,
720 XFS_IEXT_ADD_NOSPLIT_CNT);
724 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
725 allocatesize_fsb, XFS_BMAPI_PREALLOC, 0, imapp,
730 ip->i_diflags |= XFS_DIFLAG_PREALLOC;
731 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
733 error = xfs_trans_commit(tp);
734 xfs_iunlock(ip, XFS_ILOCK_EXCL);
739 * If the allocator cannot find a single free extent large
740 * enough to cover the start block of the requested range,
741 * xfs_bmapi_write will return 0 but leave *nimaps set to 0.
743 * In that case we simply need to keep looping with the same
744 * startoffset_fsb so that one of the following allocations
745 * will eventually reach the requested range.
748 startoffset_fsb += imapp->br_blockcount;
749 allocatesize_fsb -= imapp->br_blockcount;
756 xfs_trans_cancel(tp);
757 xfs_iunlock(ip, XFS_ILOCK_EXCL);
763 struct xfs_inode *ip,
764 xfs_fileoff_t startoffset_fsb,
765 xfs_filblks_t len_fsb,
768 struct xfs_mount *mp = ip->i_mount;
769 struct xfs_trans *tp;
770 uint resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
773 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, resblks, 0,
778 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
779 XFS_IEXT_PUNCH_HOLE_CNT);
781 error = xfs_iext_count_upgrade(tp, ip, XFS_IEXT_PUNCH_HOLE_CNT);
783 goto out_trans_cancel;
785 error = xfs_bunmapi(tp, ip, startoffset_fsb, len_fsb, 0, 2, done);
787 goto out_trans_cancel;
789 error = xfs_trans_commit(tp);
791 xfs_iunlock(ip, XFS_ILOCK_EXCL);
795 xfs_trans_cancel(tp);
799 /* Caller must first wait for the completion of any pending DIOs if required. */
801 xfs_flush_unmap_range(
802 struct xfs_inode *ip,
806 struct xfs_mount *mp = ip->i_mount;
807 struct inode *inode = VFS_I(ip);
808 xfs_off_t rounding, start, end;
811 rounding = max_t(xfs_off_t, mp->m_sb.sb_blocksize, PAGE_SIZE);
812 start = round_down(offset, rounding);
813 end = round_up(offset + len, rounding) - 1;
815 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
818 truncate_pagecache_range(inode, start, end);
824 struct xfs_inode *ip,
828 struct xfs_mount *mp = ip->i_mount;
829 xfs_fileoff_t startoffset_fsb;
830 xfs_fileoff_t endoffset_fsb;
833 trace_xfs_free_file_space(ip);
835 error = xfs_qm_dqattach(ip);
839 if (len <= 0) /* if nothing being freed */
842 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
843 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
845 /* We can only free complete realtime extents. */
846 if (XFS_IS_REALTIME_INODE(ip) && mp->m_sb.sb_rextsize > 1) {
847 startoffset_fsb = xfs_rtb_roundup_rtx(mp, startoffset_fsb);
848 endoffset_fsb = xfs_rtb_rounddown_rtx(mp, endoffset_fsb);
852 * Need to zero the stuff we're not freeing, on disk.
854 if (endoffset_fsb > startoffset_fsb) {
856 error = xfs_unmap_extent(ip, startoffset_fsb,
857 endoffset_fsb - startoffset_fsb, &done);
864 * Now that we've unmap all full blocks we'll have to zero out any
865 * partial block at the beginning and/or end. xfs_zero_range is smart
866 * enough to skip any holes, including those we just created, but we
867 * must take care not to zero beyond EOF and enlarge i_size.
869 if (offset >= XFS_ISIZE(ip))
871 if (offset + len > XFS_ISIZE(ip))
872 len = XFS_ISIZE(ip) - offset;
873 error = xfs_zero_range(ip, offset, len, NULL);
878 * If we zeroed right up to EOF and EOF straddles a page boundary we
879 * must make sure that the post-EOF area is also zeroed because the
880 * page could be mmap'd and xfs_zero_range doesn't do that for us.
881 * Writeback of the eof page will do this, albeit clumsily.
883 if (offset + len >= XFS_ISIZE(ip) && offset_in_page(offset + len) > 0) {
884 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
885 round_down(offset + len, PAGE_SIZE), LLONG_MAX);
893 struct xfs_inode *ip,
896 struct xfs_mount *mp = ip->i_mount;
900 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
901 * into the accessible region of the file.
903 if (xfs_can_free_eofblocks(ip, true)) {
904 error = xfs_free_eofblocks(ip);
910 * Shift operations must stabilize the start block offset boundary along
911 * with the full range of the operation. If we don't, a COW writeback
912 * completion could race with an insert, front merge with the start
913 * extent (after split) during the shift and corrupt the file. Start
914 * with the block just prior to the start to stabilize the boundary.
916 offset = round_down(offset, mp->m_sb.sb_blocksize);
918 offset -= mp->m_sb.sb_blocksize;
921 * Writeback and invalidate cache for the remainder of the file as we're
922 * about to shift down every extent from offset to EOF.
924 error = xfs_flush_unmap_range(ip, offset, XFS_ISIZE(ip));
929 * Clean out anything hanging around in the cow fork now that
930 * we've flushed all the dirty data out to disk to avoid having
931 * CoW extents at the wrong offsets.
933 if (xfs_inode_has_cow_data(ip)) {
934 error = xfs_reflink_cancel_cow_range(ip, offset, NULLFILEOFF,
944 * xfs_collapse_file_space()
945 * This routine frees disk space and shift extent for the given file.
946 * The first thing we do is to free data blocks in the specified range
947 * by calling xfs_free_file_space(). It would also sync dirty data
948 * and invalidate page cache over the region on which collapse range
949 * is working. And Shift extent records to the left to cover a hole.
956 xfs_collapse_file_space(
957 struct xfs_inode *ip,
961 struct xfs_mount *mp = ip->i_mount;
962 struct xfs_trans *tp;
964 xfs_fileoff_t next_fsb = XFS_B_TO_FSB(mp, offset + len);
965 xfs_fileoff_t shift_fsb = XFS_B_TO_FSB(mp, len);
968 xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL);
970 trace_xfs_collapse_file_space(ip);
972 error = xfs_free_file_space(ip, offset, len);
976 error = xfs_prepare_shift(ip, offset);
980 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
984 xfs_ilock(ip, XFS_ILOCK_EXCL);
985 xfs_trans_ijoin(tp, ip, 0);
988 error = xfs_bmap_collapse_extents(tp, ip, &next_fsb, shift_fsb,
991 goto out_trans_cancel;
995 /* finish any deferred frees and roll the transaction */
996 error = xfs_defer_finish(&tp);
998 goto out_trans_cancel;
1001 error = xfs_trans_commit(tp);
1002 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1006 xfs_trans_cancel(tp);
1007 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1012 * xfs_insert_file_space()
1013 * This routine create hole space by shifting extents for the given file.
1014 * The first thing we do is to sync dirty data and invalidate page cache
1015 * over the region on which insert range is working. And split an extent
1016 * to two extents at given offset by calling xfs_bmap_split_extent.
1017 * And shift all extent records which are laying between [offset,
1018 * last allocated extent] to the right to reserve hole range.
1024 xfs_insert_file_space(
1025 struct xfs_inode *ip,
1029 struct xfs_mount *mp = ip->i_mount;
1030 struct xfs_trans *tp;
1032 xfs_fileoff_t stop_fsb = XFS_B_TO_FSB(mp, offset);
1033 xfs_fileoff_t next_fsb = NULLFSBLOCK;
1034 xfs_fileoff_t shift_fsb = XFS_B_TO_FSB(mp, len);
1037 xfs_assert_ilocked(ip, XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL);
1039 trace_xfs_insert_file_space(ip);
1041 error = xfs_bmap_can_insert_extents(ip, stop_fsb, shift_fsb);
1045 error = xfs_prepare_shift(ip, offset);
1049 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write,
1050 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp);
1054 xfs_ilock(ip, XFS_ILOCK_EXCL);
1055 xfs_trans_ijoin(tp, ip, 0);
1057 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
1058 XFS_IEXT_PUNCH_HOLE_CNT);
1059 if (error == -EFBIG)
1060 error = xfs_iext_count_upgrade(tp, ip, XFS_IEXT_PUNCH_HOLE_CNT);
1062 goto out_trans_cancel;
1065 * The extent shifting code works on extent granularity. So, if stop_fsb
1066 * is not the starting block of extent, we need to split the extent at
1069 error = xfs_bmap_split_extent(tp, ip, stop_fsb);
1071 goto out_trans_cancel;
1074 error = xfs_defer_finish(&tp);
1076 goto out_trans_cancel;
1078 error = xfs_bmap_insert_extents(tp, ip, &next_fsb, shift_fsb,
1081 goto out_trans_cancel;
1084 error = xfs_trans_commit(tp);
1085 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1089 xfs_trans_cancel(tp);
1090 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1095 * We need to check that the format of the data fork in the temporary inode is
1096 * valid for the target inode before doing the swap. This is not a problem with
1097 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1098 * data fork depending on the space the attribute fork is taking so we can get
1099 * invalid formats on the target inode.
1101 * E.g. target has space for 7 extents in extent format, temp inode only has
1102 * space for 6. If we defragment down to 7 extents, then the tmp format is a
1103 * btree, but when swapped it needs to be in extent format. Hence we can't just
1104 * blindly swap data forks on attr2 filesystems.
1106 * Note that we check the swap in both directions so that we don't end up with
1107 * a corrupt temporary inode, either.
1109 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1110 * inode will prevent this situation from occurring, so all we do here is
1111 * reject and log the attempt. basically we are putting the responsibility on
1112 * userspace to get this right.
1115 xfs_swap_extents_check_format(
1116 struct xfs_inode *ip, /* target inode */
1117 struct xfs_inode *tip) /* tmp inode */
1119 struct xfs_ifork *ifp = &ip->i_df;
1120 struct xfs_ifork *tifp = &tip->i_df;
1122 /* User/group/project quota ids must match if quotas are enforced. */
1123 if (XFS_IS_QUOTA_ON(ip->i_mount) &&
1124 (!uid_eq(VFS_I(ip)->i_uid, VFS_I(tip)->i_uid) ||
1125 !gid_eq(VFS_I(ip)->i_gid, VFS_I(tip)->i_gid) ||
1126 ip->i_projid != tip->i_projid))
1129 /* Should never get a local format */
1130 if (ifp->if_format == XFS_DINODE_FMT_LOCAL ||
1131 tifp->if_format == XFS_DINODE_FMT_LOCAL)
1135 * if the target inode has less extents that then temporary inode then
1136 * why did userspace call us?
1138 if (ifp->if_nextents < tifp->if_nextents)
1142 * If we have to use the (expensive) rmap swap method, we can
1143 * handle any number of extents and any format.
1145 if (xfs_has_rmapbt(ip->i_mount))
1149 * if the target inode is in extent form and the temp inode is in btree
1150 * form then we will end up with the target inode in the wrong format
1151 * as we already know there are less extents in the temp inode.
1153 if (ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
1154 tifp->if_format == XFS_DINODE_FMT_BTREE)
1157 /* Check temp in extent form to max in target */
1158 if (tifp->if_format == XFS_DINODE_FMT_EXTENTS &&
1159 tifp->if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1162 /* Check target in extent form to max in temp */
1163 if (ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
1164 ifp->if_nextents > XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1168 * If we are in a btree format, check that the temp root block will fit
1169 * in the target and that it has enough extents to be in btree format
1172 * Note that we have to be careful to allow btree->extent conversions
1173 * (a common defrag case) which will occur when the temp inode is in
1176 if (tifp->if_format == XFS_DINODE_FMT_BTREE) {
1177 if (xfs_inode_has_attr_fork(ip) &&
1178 XFS_BMAP_BMDR_SPACE(tifp->if_broot) > xfs_inode_fork_boff(ip))
1180 if (tifp->if_nextents <= XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1184 /* Reciprocal target->temp btree format checks */
1185 if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
1186 if (xfs_inode_has_attr_fork(tip) &&
1187 XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > xfs_inode_fork_boff(tip))
1189 if (ifp->if_nextents <= XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1197 xfs_swap_extent_flush(
1198 struct xfs_inode *ip)
1202 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1205 truncate_pagecache_range(VFS_I(ip), 0, -1);
1207 /* Verify O_DIRECT for ftmp */
1208 if (VFS_I(ip)->i_mapping->nrpages)
1214 * Move extents from one file to another, when rmap is enabled.
1217 xfs_swap_extent_rmap(
1218 struct xfs_trans **tpp,
1219 struct xfs_inode *ip,
1220 struct xfs_inode *tip)
1222 struct xfs_trans *tp = *tpp;
1223 struct xfs_bmbt_irec irec;
1224 struct xfs_bmbt_irec uirec;
1225 struct xfs_bmbt_irec tirec;
1226 xfs_fileoff_t offset_fsb;
1227 xfs_fileoff_t end_fsb;
1228 xfs_filblks_t count_fsb;
1233 uint64_t tip_flags2;
1236 * If the source file has shared blocks, we must flag the donor
1237 * file as having shared blocks so that we get the shared-block
1238 * rmap functions when we go to fix up the rmaps. The flags
1239 * will be switch for reals later.
1241 tip_flags2 = tip->i_diflags2;
1242 if (ip->i_diflags2 & XFS_DIFLAG2_REFLINK)
1243 tip->i_diflags2 |= XFS_DIFLAG2_REFLINK;
1246 end_fsb = XFS_B_TO_FSB(ip->i_mount, i_size_read(VFS_I(ip)));
1247 count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb);
1250 /* Read extent from the donor file */
1252 error = xfs_bmapi_read(tip, offset_fsb, count_fsb, &tirec,
1256 ASSERT(nimaps == 1);
1257 ASSERT(tirec.br_startblock != DELAYSTARTBLOCK);
1259 trace_xfs_swap_extent_rmap_remap(tip, &tirec);
1260 ilen = tirec.br_blockcount;
1262 /* Unmap the old blocks in the source file. */
1263 while (tirec.br_blockcount) {
1264 ASSERT(tp->t_highest_agno == NULLAGNUMBER);
1265 trace_xfs_swap_extent_rmap_remap_piece(tip, &tirec);
1267 /* Read extent from the source file */
1269 error = xfs_bmapi_read(ip, tirec.br_startoff,
1270 tirec.br_blockcount, &irec,
1274 ASSERT(nimaps == 1);
1275 ASSERT(tirec.br_startoff == irec.br_startoff);
1276 trace_xfs_swap_extent_rmap_remap_piece(ip, &irec);
1278 /* Trim the extent. */
1280 uirec.br_blockcount = rlen = min_t(xfs_filblks_t,
1281 tirec.br_blockcount,
1282 irec.br_blockcount);
1283 trace_xfs_swap_extent_rmap_remap_piece(tip, &uirec);
1285 if (xfs_bmap_is_real_extent(&uirec)) {
1286 error = xfs_iext_count_may_overflow(ip,
1288 XFS_IEXT_SWAP_RMAP_CNT);
1289 if (error == -EFBIG)
1290 error = xfs_iext_count_upgrade(tp, ip,
1291 XFS_IEXT_SWAP_RMAP_CNT);
1296 if (xfs_bmap_is_real_extent(&irec)) {
1297 error = xfs_iext_count_may_overflow(tip,
1299 XFS_IEXT_SWAP_RMAP_CNT);
1300 if (error == -EFBIG)
1301 error = xfs_iext_count_upgrade(tp, ip,
1302 XFS_IEXT_SWAP_RMAP_CNT);
1307 /* Remove the mapping from the donor file. */
1308 xfs_bmap_unmap_extent(tp, tip, XFS_DATA_FORK, &uirec);
1310 /* Remove the mapping from the source file. */
1311 xfs_bmap_unmap_extent(tp, ip, XFS_DATA_FORK, &irec);
1313 /* Map the donor file's blocks into the source file. */
1314 xfs_bmap_map_extent(tp, ip, XFS_DATA_FORK, &uirec);
1316 /* Map the source file's blocks into the donor file. */
1317 xfs_bmap_map_extent(tp, tip, XFS_DATA_FORK, &irec);
1319 error = xfs_defer_finish(tpp);
1324 tirec.br_startoff += rlen;
1325 if (tirec.br_startblock != HOLESTARTBLOCK &&
1326 tirec.br_startblock != DELAYSTARTBLOCK)
1327 tirec.br_startblock += rlen;
1328 tirec.br_blockcount -= rlen;
1336 tip->i_diflags2 = tip_flags2;
1340 trace_xfs_swap_extent_rmap_error(ip, error, _RET_IP_);
1341 tip->i_diflags2 = tip_flags2;
1345 /* Swap the extents of two files by swapping data forks. */
1347 xfs_swap_extent_forks(
1348 struct xfs_trans *tp,
1349 struct xfs_inode *ip,
1350 struct xfs_inode *tip,
1352 int *target_log_flags)
1354 xfs_filblks_t aforkblks = 0;
1355 xfs_filblks_t taforkblks = 0;
1361 * Count the number of extended attribute blocks
1363 if (xfs_inode_has_attr_fork(ip) && ip->i_af.if_nextents > 0 &&
1364 ip->i_af.if_format != XFS_DINODE_FMT_LOCAL) {
1365 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &junk,
1370 if (xfs_inode_has_attr_fork(tip) && tip->i_af.if_nextents > 0 &&
1371 tip->i_af.if_format != XFS_DINODE_FMT_LOCAL) {
1372 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK, &junk,
1379 * Btree format (v3) inodes have the inode number stamped in the bmbt
1380 * block headers. We can't start changing the bmbt blocks until the
1381 * inode owner change is logged so recovery does the right thing in the
1382 * event of a crash. Set the owner change log flags now and leave the
1383 * bmbt scan as the last step.
1385 if (xfs_has_v3inodes(ip->i_mount)) {
1386 if (ip->i_df.if_format == XFS_DINODE_FMT_BTREE)
1387 (*target_log_flags) |= XFS_ILOG_DOWNER;
1388 if (tip->i_df.if_format == XFS_DINODE_FMT_BTREE)
1389 (*src_log_flags) |= XFS_ILOG_DOWNER;
1393 * Swap the data forks of the inodes
1395 swap(ip->i_df, tip->i_df);
1398 * Fix the on-disk inode values
1400 tmp = (uint64_t)ip->i_nblocks;
1401 ip->i_nblocks = tip->i_nblocks - taforkblks + aforkblks;
1402 tip->i_nblocks = tmp + taforkblks - aforkblks;
1405 * The extents in the source inode could still contain speculative
1406 * preallocation beyond EOF (e.g. the file is open but not modified
1407 * while defrag is in progress). In that case, we need to copy over the
1408 * number of delalloc blocks the data fork in the source inode is
1409 * tracking beyond EOF so that when the fork is truncated away when the
1410 * temporary inode is unlinked we don't underrun the i_delayed_blks
1411 * counter on that inode.
1413 ASSERT(tip->i_delayed_blks == 0);
1414 tip->i_delayed_blks = ip->i_delayed_blks;
1415 ip->i_delayed_blks = 0;
1417 switch (ip->i_df.if_format) {
1418 case XFS_DINODE_FMT_EXTENTS:
1419 (*src_log_flags) |= XFS_ILOG_DEXT;
1421 case XFS_DINODE_FMT_BTREE:
1422 ASSERT(!xfs_has_v3inodes(ip->i_mount) ||
1423 (*src_log_flags & XFS_ILOG_DOWNER));
1424 (*src_log_flags) |= XFS_ILOG_DBROOT;
1428 switch (tip->i_df.if_format) {
1429 case XFS_DINODE_FMT_EXTENTS:
1430 (*target_log_flags) |= XFS_ILOG_DEXT;
1432 case XFS_DINODE_FMT_BTREE:
1433 (*target_log_flags) |= XFS_ILOG_DBROOT;
1434 ASSERT(!xfs_has_v3inodes(ip->i_mount) ||
1435 (*target_log_flags & XFS_ILOG_DOWNER));
1443 * Fix up the owners of the bmbt blocks to refer to the current inode. The
1444 * change owner scan attempts to order all modified buffers in the current
1445 * transaction. In the event of ordered buffer failure, the offending buffer is
1446 * physically logged as a fallback and the scan returns -EAGAIN. We must roll
1447 * the transaction in this case to replenish the fallback log reservation and
1448 * restart the scan. This process repeats until the scan completes.
1451 xfs_swap_change_owner(
1452 struct xfs_trans **tpp,
1453 struct xfs_inode *ip,
1454 struct xfs_inode *tmpip)
1457 struct xfs_trans *tp = *tpp;
1460 error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK, ip->i_ino,
1462 /* success or fatal error */
1463 if (error != -EAGAIN)
1466 error = xfs_trans_roll(tpp);
1472 * Redirty both inodes so they can relog and keep the log tail
1475 xfs_trans_ijoin(tp, ip, 0);
1476 xfs_trans_ijoin(tp, tmpip, 0);
1477 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1478 xfs_trans_log_inode(tp, tmpip, XFS_ILOG_CORE);
1486 struct xfs_inode *ip, /* target inode */
1487 struct xfs_inode *tip, /* tmp inode */
1488 struct xfs_swapext *sxp)
1490 struct xfs_mount *mp = ip->i_mount;
1491 struct xfs_trans *tp;
1492 struct xfs_bstat *sbp = &sxp->sx_stat;
1493 int src_log_flags, target_log_flags;
1497 unsigned int flags = 0;
1498 struct timespec64 ctime, mtime;
1501 * Lock the inodes against other IO, page faults and truncate to
1502 * begin with. Then we can ensure the inodes are flushed and have no
1503 * page cache safely. Once we have done this we can take the ilocks and
1504 * do the rest of the checks.
1506 lock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1507 filemap_invalidate_lock_two(VFS_I(ip)->i_mapping,
1508 VFS_I(tip)->i_mapping);
1510 /* Verify that both files have the same format */
1511 if ((VFS_I(ip)->i_mode & S_IFMT) != (VFS_I(tip)->i_mode & S_IFMT)) {
1516 /* Verify both files are either real-time or non-realtime */
1517 if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
1522 error = xfs_qm_dqattach(ip);
1526 error = xfs_qm_dqattach(tip);
1530 error = xfs_swap_extent_flush(ip);
1533 error = xfs_swap_extent_flush(tip);
1537 if (xfs_inode_has_cow_data(tip)) {
1538 error = xfs_reflink_cancel_cow_range(tip, 0, NULLFILEOFF, true);
1544 * Extent "swapping" with rmap requires a permanent reservation and
1545 * a block reservation because it's really just a remap operation
1546 * performed with log redo items!
1548 if (xfs_has_rmapbt(mp)) {
1549 int w = XFS_DATA_FORK;
1550 uint32_t ipnext = ip->i_df.if_nextents;
1551 uint32_t tipnext = tip->i_df.if_nextents;
1554 * Conceptually this shouldn't affect the shape of either bmbt,
1555 * but since we atomically move extents one by one, we reserve
1556 * enough space to rebuild both trees.
1558 resblks = XFS_SWAP_RMAP_SPACE_RES(mp, ipnext, w);
1559 resblks += XFS_SWAP_RMAP_SPACE_RES(mp, tipnext, w);
1562 * If either inode straddles a bmapbt block allocation boundary,
1563 * the rmapbt algorithm triggers repeated allocs and frees as
1564 * extents are remapped. This can exhaust the block reservation
1565 * prematurely and cause shutdown. Return freed blocks to the
1566 * transaction reservation to counter this behavior.
1568 flags |= XFS_TRANS_RES_FDBLKS;
1570 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, flags,
1576 * Lock and join the inodes to the tansaction so that transaction commit
1577 * or cancel will unlock the inodes from this point onwards.
1579 xfs_lock_two_inodes(ip, XFS_ILOCK_EXCL, tip, XFS_ILOCK_EXCL);
1580 xfs_trans_ijoin(tp, ip, 0);
1581 xfs_trans_ijoin(tp, tip, 0);
1584 /* Verify all data are being swapped */
1585 if (sxp->sx_offset != 0 ||
1586 sxp->sx_length != ip->i_disk_size ||
1587 sxp->sx_length != tip->i_disk_size) {
1589 goto out_trans_cancel;
1592 trace_xfs_swap_extent_before(ip, 0);
1593 trace_xfs_swap_extent_before(tip, 1);
1595 /* check inode formats now that data is flushed */
1596 error = xfs_swap_extents_check_format(ip, tip);
1599 "%s: inode 0x%llx format is incompatible for exchanging.",
1600 __func__, ip->i_ino);
1601 goto out_trans_cancel;
1605 * Compare the current change & modify times with that
1606 * passed in. If they differ, we abort this swap.
1607 * This is the mechanism used to ensure the calling
1608 * process that the file was not changed out from
1611 ctime = inode_get_ctime(VFS_I(ip));
1612 mtime = inode_get_mtime(VFS_I(ip));
1613 if ((sbp->bs_ctime.tv_sec != ctime.tv_sec) ||
1614 (sbp->bs_ctime.tv_nsec != ctime.tv_nsec) ||
1615 (sbp->bs_mtime.tv_sec != mtime.tv_sec) ||
1616 (sbp->bs_mtime.tv_nsec != mtime.tv_nsec)) {
1618 goto out_trans_cancel;
1622 * Note the trickiness in setting the log flags - we set the owner log
1623 * flag on the opposite inode (i.e. the inode we are setting the new
1624 * owner to be) because once we swap the forks and log that, log
1625 * recovery is going to see the fork as owned by the swapped inode,
1626 * not the pre-swapped inodes.
1628 src_log_flags = XFS_ILOG_CORE;
1629 target_log_flags = XFS_ILOG_CORE;
1631 if (xfs_has_rmapbt(mp))
1632 error = xfs_swap_extent_rmap(&tp, ip, tip);
1634 error = xfs_swap_extent_forks(tp, ip, tip, &src_log_flags,
1637 goto out_trans_cancel;
1639 /* Do we have to swap reflink flags? */
1640 if ((ip->i_diflags2 & XFS_DIFLAG2_REFLINK) ^
1641 (tip->i_diflags2 & XFS_DIFLAG2_REFLINK)) {
1642 f = ip->i_diflags2 & XFS_DIFLAG2_REFLINK;
1643 ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
1644 ip->i_diflags2 |= tip->i_diflags2 & XFS_DIFLAG2_REFLINK;
1645 tip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
1646 tip->i_diflags2 |= f & XFS_DIFLAG2_REFLINK;
1649 /* Swap the cow forks. */
1650 if (xfs_has_reflink(mp)) {
1651 ASSERT(!ip->i_cowfp ||
1652 ip->i_cowfp->if_format == XFS_DINODE_FMT_EXTENTS);
1653 ASSERT(!tip->i_cowfp ||
1654 tip->i_cowfp->if_format == XFS_DINODE_FMT_EXTENTS);
1656 swap(ip->i_cowfp, tip->i_cowfp);
1658 if (ip->i_cowfp && ip->i_cowfp->if_bytes)
1659 xfs_inode_set_cowblocks_tag(ip);
1661 xfs_inode_clear_cowblocks_tag(ip);
1662 if (tip->i_cowfp && tip->i_cowfp->if_bytes)
1663 xfs_inode_set_cowblocks_tag(tip);
1665 xfs_inode_clear_cowblocks_tag(tip);
1668 xfs_trans_log_inode(tp, ip, src_log_flags);
1669 xfs_trans_log_inode(tp, tip, target_log_flags);
1672 * The extent forks have been swapped, but crc=1,rmapbt=0 filesystems
1673 * have inode number owner values in the bmbt blocks that still refer to
1674 * the old inode. Scan each bmbt to fix up the owner values with the
1675 * inode number of the current inode.
1677 if (src_log_flags & XFS_ILOG_DOWNER) {
1678 error = xfs_swap_change_owner(&tp, ip, tip);
1680 goto out_trans_cancel;
1682 if (target_log_flags & XFS_ILOG_DOWNER) {
1683 error = xfs_swap_change_owner(&tp, tip, ip);
1685 goto out_trans_cancel;
1689 * If this is a synchronous mount, make sure that the
1690 * transaction goes to disk before returning to the user.
1692 if (xfs_has_wsync(mp))
1693 xfs_trans_set_sync(tp);
1695 error = xfs_trans_commit(tp);
1697 trace_xfs_swap_extent_after(ip, 0);
1698 trace_xfs_swap_extent_after(tip, 1);
1701 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1702 xfs_iunlock(tip, XFS_ILOCK_EXCL);
1704 filemap_invalidate_unlock_two(VFS_I(ip)->i_mapping,
1705 VFS_I(tip)->i_mapping);
1706 unlock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1710 xfs_trans_cancel(tp);
1711 goto out_unlock_ilock;