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"
32 /* Kernel only BMAP related definitions and functions */
35 * Convert the given file system block to a disk block. We have to treat it
36 * differently based on whether the file is a real time file or not, because the
40 xfs_fsb_to_db(struct xfs_inode *ip, xfs_fsblock_t fsb)
42 if (XFS_IS_REALTIME_INODE(ip))
43 return XFS_FSB_TO_BB(ip->i_mount, fsb);
44 return XFS_FSB_TO_DADDR(ip->i_mount, fsb);
48 * Routine to zero an extent on disk allocated to the specific inode.
50 * The VFS functions take a linearised filesystem block offset, so we have to
51 * convert the sparse xfs fsb to the right format first.
52 * VFS types are real funky, too.
57 xfs_fsblock_t start_fsb,
60 struct xfs_mount *mp = ip->i_mount;
61 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
62 xfs_daddr_t sector = xfs_fsb_to_db(ip, start_fsb);
63 sector_t block = XFS_BB_TO_FSBT(mp, sector);
65 return blkdev_issue_zeroout(target->bt_bdev,
66 block << (mp->m_super->s_blocksize_bits - 9),
67 count_fsb << (mp->m_super->s_blocksize_bits - 9),
74 struct xfs_bmalloca *ap) /* bmap alloc argument struct */
76 int error; /* error return value */
77 xfs_mount_t *mp; /* mount point structure */
78 xfs_extlen_t prod = 0; /* product factor for allocators */
79 xfs_extlen_t mod = 0; /* product factor for allocators */
80 xfs_extlen_t ralen = 0; /* realtime allocation length */
81 xfs_extlen_t align; /* minimum allocation alignment */
85 align = xfs_get_extsz_hint(ap->ip);
86 prod = align / mp->m_sb.sb_rextsize;
87 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
89 ap->conv, &ap->offset, &ap->length);
93 ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
96 * If the offset & length are not perfectly aligned
97 * then kill prod, it will just get us in trouble.
99 div_u64_rem(ap->offset, align, &mod);
100 if (mod || ap->length % align)
103 * Set ralen to be the actual requested length in rtextents.
105 ralen = ap->length / mp->m_sb.sb_rextsize;
107 * If the old value was close enough to MAXEXTLEN that
108 * we rounded up to it, cut it back so it's valid again.
109 * Note that if it's a really large request (bigger than
110 * MAXEXTLEN), we don't hear about that number, and can't
111 * adjust the starting point to match it.
113 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
114 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
117 * Lock out modifications to both the RT bitmap and summary inodes
119 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
120 xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
121 xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
122 xfs_trans_ijoin(ap->tp, mp->m_rsumip, XFS_ILOCK_EXCL);
125 * If it's an allocation to an empty file at offset 0,
126 * pick an extent that will space things out in the rt area.
128 if (ap->eof && ap->offset == 0) {
129 xfs_rtblock_t rtx; /* realtime extent no */
131 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
134 ap->blkno = rtx * mp->m_sb.sb_rextsize;
139 xfs_bmap_adjacent(ap);
142 * Realtime allocation, done through xfs_rtallocate_extent.
144 do_div(ap->blkno, mp->m_sb.sb_rextsize);
147 error = xfs_rtallocate_extent(ap->tp, ap->blkno, 1, ap->length,
148 &ralen, ap->wasdel, prod, &rtb);
153 if (ap->blkno != NULLFSBLOCK) {
154 ap->blkno *= mp->m_sb.sb_rextsize;
155 ralen *= mp->m_sb.sb_rextsize;
157 ap->ip->i_d.di_nblocks += ralen;
158 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
160 ap->ip->i_delayed_blks -= ralen;
162 * Adjust the disk quota also. This was reserved
165 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
166 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
167 XFS_TRANS_DQ_RTBCOUNT, (long) ralen);
173 #endif /* CONFIG_XFS_RT */
176 * Extent tree block counting routines.
180 * Count leaf blocks given a range of extent records. Delayed allocation
181 * extents are not counted towards the totals.
184 xfs_bmap_count_leaves(
185 struct xfs_ifork *ifp,
186 xfs_filblks_t *count)
188 struct xfs_iext_cursor icur;
189 struct xfs_bmbt_irec got;
190 xfs_extnum_t numrecs = 0;
192 for_each_xfs_iext(ifp, &icur, &got) {
193 if (!isnullstartblock(got.br_startblock)) {
194 *count += got.br_blockcount;
203 * Count fsblocks of the given fork. Delayed allocation extents are
204 * not counted towards the totals.
207 xfs_bmap_count_blocks(
208 struct xfs_trans *tp,
209 struct xfs_inode *ip,
211 xfs_extnum_t *nextents,
212 xfs_filblks_t *count)
214 struct xfs_mount *mp = ip->i_mount;
215 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
216 struct xfs_btree_cur *cur;
217 xfs_extlen_t btblocks = 0;
226 switch (ifp->if_format) {
227 case XFS_DINODE_FMT_BTREE:
228 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
229 error = xfs_iread_extents(tp, ip, whichfork);
234 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
235 error = xfs_btree_count_blocks(cur, &btblocks);
236 xfs_btree_del_cursor(cur, error);
241 * xfs_btree_count_blocks includes the root block contained in
242 * the inode fork in @btblocks, so subtract one because we're
243 * only interested in allocated disk blocks.
245 *count += btblocks - 1;
248 case XFS_DINODE_FMT_EXTENTS:
249 *nextents = xfs_bmap_count_leaves(ifp, count);
257 xfs_getbmap_report_one(
258 struct xfs_inode *ip,
259 struct getbmapx *bmv,
260 struct kgetbmap *out,
262 struct xfs_bmbt_irec *got)
264 struct kgetbmap *p = out + bmv->bmv_entries;
268 error = xfs_reflink_trim_around_shared(ip, got, &shared);
272 if (isnullstartblock(got->br_startblock) ||
273 got->br_startblock == DELAYSTARTBLOCK) {
275 * Delalloc extents that start beyond EOF can occur due to
276 * speculative EOF allocation when the delalloc extent is larger
277 * than the largest freespace extent at conversion time. These
278 * extents cannot be converted by data writeback, so can exist
279 * here even if we are not supposed to be finding delalloc
282 if (got->br_startoff < XFS_B_TO_FSB(ip->i_mount, XFS_ISIZE(ip)))
283 ASSERT((bmv->bmv_iflags & BMV_IF_DELALLOC) != 0);
285 p->bmv_oflags |= BMV_OF_DELALLOC;
288 p->bmv_block = xfs_fsb_to_db(ip, got->br_startblock);
291 if (got->br_state == XFS_EXT_UNWRITTEN &&
292 (bmv->bmv_iflags & BMV_IF_PREALLOC))
293 p->bmv_oflags |= BMV_OF_PREALLOC;
296 p->bmv_oflags |= BMV_OF_SHARED;
298 p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, got->br_startoff);
299 p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, got->br_blockcount);
301 bmv->bmv_offset = p->bmv_offset + p->bmv_length;
302 bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset);
308 xfs_getbmap_report_hole(
309 struct xfs_inode *ip,
310 struct getbmapx *bmv,
311 struct kgetbmap *out,
316 struct kgetbmap *p = out + bmv->bmv_entries;
318 if (bmv->bmv_iflags & BMV_IF_NO_HOLES)
322 p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, bno);
323 p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, end - bno);
325 bmv->bmv_offset = p->bmv_offset + p->bmv_length;
326 bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset);
332 struct getbmapx *bmv)
334 return bmv->bmv_length == 0 || bmv->bmv_entries >= bmv->bmv_count - 1;
338 xfs_getbmap_next_rec(
339 struct xfs_bmbt_irec *rec,
340 xfs_fileoff_t total_end)
342 xfs_fileoff_t end = rec->br_startoff + rec->br_blockcount;
344 if (end == total_end)
347 rec->br_startoff += rec->br_blockcount;
348 if (!isnullstartblock(rec->br_startblock) &&
349 rec->br_startblock != DELAYSTARTBLOCK)
350 rec->br_startblock += rec->br_blockcount;
351 rec->br_blockcount = total_end - end;
356 * Get inode's extents as described in bmv, and format for output.
357 * Calls formatter to fill the user's buffer until all extents
358 * are mapped, until the passed-in bmv->bmv_count slots have
359 * been filled, or until the formatter short-circuits the loop,
360 * if it is tracking filled-in extents on its own.
364 struct xfs_inode *ip,
365 struct getbmapx *bmv, /* user bmap structure */
366 struct kgetbmap *out)
368 struct xfs_mount *mp = ip->i_mount;
369 int iflags = bmv->bmv_iflags;
370 int whichfork, lock, error = 0;
371 int64_t bmv_end, max_len;
372 xfs_fileoff_t bno, first_bno;
373 struct xfs_ifork *ifp;
374 struct xfs_bmbt_irec got, rec;
376 struct xfs_iext_cursor icur;
378 if (bmv->bmv_iflags & ~BMV_IF_VALID)
381 /* Only allow CoW fork queries if we're debugging. */
382 if (iflags & BMV_IF_COWFORK)
385 if ((iflags & BMV_IF_ATTRFORK) && (iflags & BMV_IF_COWFORK))
388 if (bmv->bmv_length < -1)
390 bmv->bmv_entries = 0;
391 if (bmv->bmv_length == 0)
394 if (iflags & BMV_IF_ATTRFORK)
395 whichfork = XFS_ATTR_FORK;
396 else if (iflags & BMV_IF_COWFORK)
397 whichfork = XFS_COW_FORK;
399 whichfork = XFS_DATA_FORK;
400 ifp = XFS_IFORK_PTR(ip, whichfork);
402 xfs_ilock(ip, XFS_IOLOCK_SHARED);
405 if (!XFS_IFORK_Q(ip))
406 goto out_unlock_iolock;
409 lock = xfs_ilock_attr_map_shared(ip);
412 /* No CoW fork? Just return */
414 goto out_unlock_iolock;
416 if (xfs_get_cowextsz_hint(ip))
417 max_len = mp->m_super->s_maxbytes;
419 max_len = XFS_ISIZE(ip);
421 lock = XFS_ILOCK_SHARED;
425 if (!(iflags & BMV_IF_DELALLOC) &&
426 (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_d.di_size)) {
427 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
429 goto out_unlock_iolock;
432 * Even after flushing the inode, there can still be
433 * delalloc blocks on the inode beyond EOF due to
434 * speculative preallocation. These are not removed
435 * until the release function is called or the inode
436 * is inactivated. Hence we cannot assert here that
437 * ip->i_delayed_blks == 0.
441 if (xfs_get_extsz_hint(ip) ||
443 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))
444 max_len = mp->m_super->s_maxbytes;
446 max_len = XFS_ISIZE(ip);
448 lock = xfs_ilock_data_map_shared(ip);
452 switch (ifp->if_format) {
453 case XFS_DINODE_FMT_EXTENTS:
454 case XFS_DINODE_FMT_BTREE:
456 case XFS_DINODE_FMT_LOCAL:
457 /* Local format inode forks report no extents. */
458 goto out_unlock_ilock;
461 goto out_unlock_ilock;
464 if (bmv->bmv_length == -1) {
465 max_len = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, max_len));
466 bmv->bmv_length = max(0LL, max_len - bmv->bmv_offset);
469 bmv_end = bmv->bmv_offset + bmv->bmv_length;
471 first_bno = bno = XFS_BB_TO_FSBT(mp, bmv->bmv_offset);
472 len = XFS_BB_TO_FSB(mp, bmv->bmv_length);
474 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
475 error = xfs_iread_extents(NULL, ip, whichfork);
477 goto out_unlock_ilock;
480 if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
482 * Report a whole-file hole if the delalloc flag is set to
483 * stay compatible with the old implementation.
485 if (iflags & BMV_IF_DELALLOC)
486 xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno,
487 XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
488 goto out_unlock_ilock;
491 while (!xfs_getbmap_full(bmv)) {
492 xfs_trim_extent(&got, first_bno, len);
495 * Report an entry for a hole if this extent doesn't directly
496 * follow the previous one.
498 if (got.br_startoff > bno) {
499 xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno,
501 if (xfs_getbmap_full(bmv))
506 * In order to report shared extents accurately, we report each
507 * distinct shared / unshared part of a single bmbt record with
508 * an individual getbmapx record.
510 bno = got.br_startoff + got.br_blockcount;
513 error = xfs_getbmap_report_one(ip, bmv, out, bmv_end,
515 if (error || xfs_getbmap_full(bmv))
516 goto out_unlock_ilock;
517 } while (xfs_getbmap_next_rec(&rec, bno));
519 if (!xfs_iext_next_extent(ifp, &icur, &got)) {
520 xfs_fileoff_t end = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
522 out[bmv->bmv_entries - 1].bmv_oflags |= BMV_OF_LAST;
524 if (whichfork != XFS_ATTR_FORK && bno < end &&
525 !xfs_getbmap_full(bmv)) {
526 xfs_getbmap_report_hole(ip, bmv, out, bmv_end,
532 if (bno >= first_bno + len)
537 xfs_iunlock(ip, lock);
539 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
544 * Dead simple method of punching delalyed allocation blocks from a range in
545 * the inode. This will always punch out both the start and end blocks, even
546 * if the ranges only partially overlap them, so it is up to the caller to
547 * ensure that partial blocks are not passed in.
550 xfs_bmap_punch_delalloc_range(
551 struct xfs_inode *ip,
552 xfs_fileoff_t start_fsb,
553 xfs_fileoff_t length)
555 struct xfs_ifork *ifp = &ip->i_df;
556 xfs_fileoff_t end_fsb = start_fsb + length;
557 struct xfs_bmbt_irec got, del;
558 struct xfs_iext_cursor icur;
561 ASSERT(ifp->if_flags & XFS_IFEXTENTS);
563 xfs_ilock(ip, XFS_ILOCK_EXCL);
564 if (!xfs_iext_lookup_extent_before(ip, ifp, &end_fsb, &icur, &got))
567 while (got.br_startoff + got.br_blockcount > start_fsb) {
569 xfs_trim_extent(&del, start_fsb, length);
572 * A delete can push the cursor forward. Step back to the
573 * previous extent on non-delalloc or extents outside the
576 if (!del.br_blockcount ||
577 !isnullstartblock(del.br_startblock)) {
578 if (!xfs_iext_prev_extent(ifp, &icur, &got))
583 error = xfs_bmap_del_extent_delay(ip, XFS_DATA_FORK, &icur,
585 if (error || !xfs_iext_get_extent(ifp, &icur, &got))
590 xfs_iunlock(ip, XFS_ILOCK_EXCL);
595 * Test whether it is appropriate to check an inode for and free post EOF
596 * blocks. The 'force' parameter determines whether we should also consider
597 * regular files that are marked preallocated or append-only.
600 xfs_can_free_eofblocks(struct xfs_inode *ip, bool force)
602 /* prealloc/delalloc exists only on regular files */
603 if (!S_ISREG(VFS_I(ip)->i_mode))
607 * Zero sized files with no cached pages and delalloc blocks will not
608 * have speculative prealloc/delalloc blocks to remove.
610 if (VFS_I(ip)->i_size == 0 &&
611 VFS_I(ip)->i_mapping->nrpages == 0 &&
612 ip->i_delayed_blks == 0)
615 /* If we haven't read in the extent list, then don't do it now. */
616 if (!(ip->i_df.if_flags & XFS_IFEXTENTS))
620 * Do not free real preallocated or append-only files unless the file
621 * has delalloc blocks and we are forced to remove them.
623 if (ip->i_d.di_flags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
624 if (!force || ip->i_delayed_blks == 0)
631 * This is called to free any blocks beyond eof. The caller must hold
632 * IOLOCK_EXCL unless we are in the inode reclaim path and have the only
633 * reference to the inode.
637 struct xfs_inode *ip)
639 struct xfs_trans *tp;
641 xfs_fileoff_t end_fsb;
642 xfs_fileoff_t last_fsb;
643 xfs_filblks_t map_len;
645 struct xfs_bmbt_irec imap;
646 struct xfs_mount *mp = ip->i_mount;
649 * Figure out if there are any blocks beyond the end
650 * of the file. If not, then there is nothing to do.
652 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
653 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
654 if (last_fsb <= end_fsb)
656 map_len = last_fsb - end_fsb;
659 xfs_ilock(ip, XFS_ILOCK_SHARED);
660 error = xfs_bmapi_read(ip, end_fsb, map_len, &imap, &nimaps, 0);
661 xfs_iunlock(ip, XFS_ILOCK_SHARED);
664 * If there are blocks after the end of file, truncate the file to its
665 * current size to free them up.
667 if (!error && (nimaps != 0) &&
668 (imap.br_startblock != HOLESTARTBLOCK ||
669 ip->i_delayed_blks)) {
671 * Attach the dquots to the inode up front.
673 error = xfs_qm_dqattach(ip);
677 /* wait on dio to ensure i_size has settled */
678 inode_dio_wait(VFS_I(ip));
680 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0,
683 ASSERT(XFS_FORCED_SHUTDOWN(mp));
687 xfs_ilock(ip, XFS_ILOCK_EXCL);
688 xfs_trans_ijoin(tp, ip, 0);
691 * Do not update the on-disk file size. If we update the
692 * on-disk file size and then the system crashes before the
693 * contents of the file are flushed to disk then the files
694 * may be full of holes (ie NULL files bug).
696 error = xfs_itruncate_extents_flags(&tp, ip, XFS_DATA_FORK,
697 XFS_ISIZE(ip), XFS_BMAPI_NODISCARD);
700 * If we get an error at this point we simply don't
701 * bother truncating the file.
703 xfs_trans_cancel(tp);
705 error = xfs_trans_commit(tp);
707 xfs_inode_clear_eofblocks_tag(ip);
710 xfs_iunlock(ip, XFS_ILOCK_EXCL);
716 xfs_alloc_file_space(
717 struct xfs_inode *ip,
722 xfs_mount_t *mp = ip->i_mount;
724 xfs_filblks_t allocated_fsb;
725 xfs_filblks_t allocatesize_fsb;
726 xfs_extlen_t extsz, temp;
727 xfs_fileoff_t startoffset_fsb;
728 xfs_fileoff_t endoffset_fsb;
732 xfs_bmbt_irec_t imaps[1], *imapp;
735 trace_xfs_alloc_file_space(ip);
737 if (XFS_FORCED_SHUTDOWN(mp))
740 error = xfs_qm_dqattach(ip);
747 rt = XFS_IS_REALTIME_INODE(ip);
748 extsz = xfs_get_extsz_hint(ip);
753 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
754 endoffset_fsb = XFS_B_TO_FSB(mp, offset + count);
755 allocatesize_fsb = endoffset_fsb - startoffset_fsb;
758 * Allocate file space until done or until there is an error
760 while (allocatesize_fsb && !error) {
762 unsigned int dblocks, rblocks, resblks;
765 * Determine space reservations for data/realtime.
767 if (unlikely(extsz)) {
771 e = startoffset_fsb + allocatesize_fsb;
772 div_u64_rem(startoffset_fsb, extsz, &temp);
775 div_u64_rem(e, extsz, &temp);
780 e = allocatesize_fsb;
784 * The transaction reservation is limited to a 32-bit block
785 * count, hence we need to limit the number of blocks we are
786 * trying to reserve to avoid an overflow. We can't allocate
787 * more than @nimaps extents, and an extent is limited on disk
788 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
790 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
792 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
795 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
800 * Allocate and setup the transaction.
802 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write,
803 dblocks, rblocks, false, &tp);
807 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
808 XFS_IEXT_ADD_NOSPLIT_CNT);
812 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
813 allocatesize_fsb, alloc_type, 0, imapp,
819 * Complete the transaction
821 error = xfs_trans_commit(tp);
822 xfs_iunlock(ip, XFS_ILOCK_EXCL);
826 allocated_fsb = imapp->br_blockcount;
833 startoffset_fsb += allocated_fsb;
834 allocatesize_fsb -= allocated_fsb;
840 xfs_trans_cancel(tp);
841 xfs_iunlock(ip, XFS_ILOCK_EXCL);
847 struct xfs_inode *ip,
848 xfs_fileoff_t startoffset_fsb,
849 xfs_filblks_t len_fsb,
852 struct xfs_mount *mp = ip->i_mount;
853 struct xfs_trans *tp;
854 uint resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
857 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, resblks, 0,
862 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
863 XFS_IEXT_PUNCH_HOLE_CNT);
865 goto out_trans_cancel;
867 error = xfs_bunmapi(tp, ip, startoffset_fsb, len_fsb, 0, 2, done);
869 goto out_trans_cancel;
871 error = xfs_trans_commit(tp);
873 xfs_iunlock(ip, XFS_ILOCK_EXCL);
877 xfs_trans_cancel(tp);
881 /* Caller must first wait for the completion of any pending DIOs if required. */
883 xfs_flush_unmap_range(
884 struct xfs_inode *ip,
888 struct xfs_mount *mp = ip->i_mount;
889 struct inode *inode = VFS_I(ip);
890 xfs_off_t rounding, start, end;
893 rounding = max_t(xfs_off_t, 1 << mp->m_sb.sb_blocklog, PAGE_SIZE);
894 start = round_down(offset, rounding);
895 end = round_up(offset + len, rounding) - 1;
897 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
900 truncate_pagecache_range(inode, start, end);
906 struct xfs_inode *ip,
910 struct xfs_mount *mp = ip->i_mount;
911 xfs_fileoff_t startoffset_fsb;
912 xfs_fileoff_t endoffset_fsb;
915 trace_xfs_free_file_space(ip);
917 error = xfs_qm_dqattach(ip);
921 if (len <= 0) /* if nothing being freed */
924 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
925 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
927 /* We can only free complete realtime extents. */
928 if (XFS_IS_REALTIME_INODE(ip) && mp->m_sb.sb_rextsize > 1) {
929 startoffset_fsb = roundup_64(startoffset_fsb,
930 mp->m_sb.sb_rextsize);
931 endoffset_fsb = rounddown_64(endoffset_fsb,
932 mp->m_sb.sb_rextsize);
936 * Need to zero the stuff we're not freeing, on disk.
938 if (endoffset_fsb > startoffset_fsb) {
940 error = xfs_unmap_extent(ip, startoffset_fsb,
941 endoffset_fsb - startoffset_fsb, &done);
948 * Now that we've unmap all full blocks we'll have to zero out any
949 * partial block at the beginning and/or end. iomap_zero_range is smart
950 * enough to skip any holes, including those we just created, but we
951 * must take care not to zero beyond EOF and enlarge i_size.
953 if (offset >= XFS_ISIZE(ip))
955 if (offset + len > XFS_ISIZE(ip))
956 len = XFS_ISIZE(ip) - offset;
957 error = iomap_zero_range(VFS_I(ip), offset, len, NULL,
958 &xfs_buffered_write_iomap_ops);
963 * If we zeroed right up to EOF and EOF straddles a page boundary we
964 * must make sure that the post-EOF area is also zeroed because the
965 * page could be mmap'd and iomap_zero_range doesn't do that for us.
966 * Writeback of the eof page will do this, albeit clumsily.
968 if (offset + len >= XFS_ISIZE(ip) && offset_in_page(offset + len) > 0) {
969 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
970 round_down(offset + len, PAGE_SIZE), LLONG_MAX);
978 struct xfs_inode *ip,
981 struct xfs_mount *mp = ip->i_mount;
985 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
986 * into the accessible region of the file.
988 if (xfs_can_free_eofblocks(ip, true)) {
989 error = xfs_free_eofblocks(ip);
995 * Shift operations must stabilize the start block offset boundary along
996 * with the full range of the operation. If we don't, a COW writeback
997 * completion could race with an insert, front merge with the start
998 * extent (after split) during the shift and corrupt the file. Start
999 * with the block just prior to the start to stabilize the boundary.
1001 offset = round_down(offset, 1 << mp->m_sb.sb_blocklog);
1003 offset -= (1 << mp->m_sb.sb_blocklog);
1006 * Writeback and invalidate cache for the remainder of the file as we're
1007 * about to shift down every extent from offset to EOF.
1009 error = xfs_flush_unmap_range(ip, offset, XFS_ISIZE(ip));
1014 * Clean out anything hanging around in the cow fork now that
1015 * we've flushed all the dirty data out to disk to avoid having
1016 * CoW extents at the wrong offsets.
1018 if (xfs_inode_has_cow_data(ip)) {
1019 error = xfs_reflink_cancel_cow_range(ip, offset, NULLFILEOFF,
1029 * xfs_collapse_file_space()
1030 * This routine frees disk space and shift extent for the given file.
1031 * The first thing we do is to free data blocks in the specified range
1032 * by calling xfs_free_file_space(). It would also sync dirty data
1033 * and invalidate page cache over the region on which collapse range
1034 * is working. And Shift extent records to the left to cover a hole.
1041 xfs_collapse_file_space(
1042 struct xfs_inode *ip,
1046 struct xfs_mount *mp = ip->i_mount;
1047 struct xfs_trans *tp;
1049 xfs_fileoff_t next_fsb = XFS_B_TO_FSB(mp, offset + len);
1050 xfs_fileoff_t shift_fsb = XFS_B_TO_FSB(mp, len);
1053 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1054 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
1056 trace_xfs_collapse_file_space(ip);
1058 error = xfs_free_file_space(ip, offset, len);
1062 error = xfs_prepare_shift(ip, offset);
1066 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
1070 xfs_ilock(ip, XFS_ILOCK_EXCL);
1071 xfs_trans_ijoin(tp, ip, 0);
1074 error = xfs_bmap_collapse_extents(tp, ip, &next_fsb, shift_fsb,
1077 goto out_trans_cancel;
1081 /* finish any deferred frees and roll the transaction */
1082 error = xfs_defer_finish(&tp);
1084 goto out_trans_cancel;
1087 error = xfs_trans_commit(tp);
1088 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1092 xfs_trans_cancel(tp);
1093 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1098 * xfs_insert_file_space()
1099 * This routine create hole space by shifting extents for the given file.
1100 * The first thing we do is to sync dirty data and invalidate page cache
1101 * over the region on which insert range is working. And split an extent
1102 * to two extents at given offset by calling xfs_bmap_split_extent.
1103 * And shift all extent records which are laying between [offset,
1104 * last allocated extent] to the right to reserve hole range.
1110 xfs_insert_file_space(
1111 struct xfs_inode *ip,
1115 struct xfs_mount *mp = ip->i_mount;
1116 struct xfs_trans *tp;
1118 xfs_fileoff_t stop_fsb = XFS_B_TO_FSB(mp, offset);
1119 xfs_fileoff_t next_fsb = NULLFSBLOCK;
1120 xfs_fileoff_t shift_fsb = XFS_B_TO_FSB(mp, len);
1123 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1124 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
1126 trace_xfs_insert_file_space(ip);
1128 error = xfs_bmap_can_insert_extents(ip, stop_fsb, shift_fsb);
1132 error = xfs_prepare_shift(ip, offset);
1136 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write,
1137 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp);
1141 xfs_ilock(ip, XFS_ILOCK_EXCL);
1142 xfs_trans_ijoin(tp, ip, 0);
1144 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
1145 XFS_IEXT_PUNCH_HOLE_CNT);
1147 goto out_trans_cancel;
1150 * The extent shifting code works on extent granularity. So, if stop_fsb
1151 * is not the starting block of extent, we need to split the extent at
1154 error = xfs_bmap_split_extent(tp, ip, stop_fsb);
1156 goto out_trans_cancel;
1159 error = xfs_defer_finish(&tp);
1161 goto out_trans_cancel;
1163 error = xfs_bmap_insert_extents(tp, ip, &next_fsb, shift_fsb,
1166 goto out_trans_cancel;
1169 error = xfs_trans_commit(tp);
1170 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1174 xfs_trans_cancel(tp);
1175 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1180 * We need to check that the format of the data fork in the temporary inode is
1181 * valid for the target inode before doing the swap. This is not a problem with
1182 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1183 * data fork depending on the space the attribute fork is taking so we can get
1184 * invalid formats on the target inode.
1186 * E.g. target has space for 7 extents in extent format, temp inode only has
1187 * space for 6. If we defragment down to 7 extents, then the tmp format is a
1188 * btree, but when swapped it needs to be in extent format. Hence we can't just
1189 * blindly swap data forks on attr2 filesystems.
1191 * Note that we check the swap in both directions so that we don't end up with
1192 * a corrupt temporary inode, either.
1194 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1195 * inode will prevent this situation from occurring, so all we do here is
1196 * reject and log the attempt. basically we are putting the responsibility on
1197 * userspace to get this right.
1200 xfs_swap_extents_check_format(
1201 struct xfs_inode *ip, /* target inode */
1202 struct xfs_inode *tip) /* tmp inode */
1204 struct xfs_ifork *ifp = &ip->i_df;
1205 struct xfs_ifork *tifp = &tip->i_df;
1207 /* User/group/project quota ids must match if quotas are enforced. */
1208 if (XFS_IS_QUOTA_ON(ip->i_mount) &&
1209 (!uid_eq(VFS_I(ip)->i_uid, VFS_I(tip)->i_uid) ||
1210 !gid_eq(VFS_I(ip)->i_gid, VFS_I(tip)->i_gid) ||
1211 ip->i_d.di_projid != tip->i_d.di_projid))
1214 /* Should never get a local format */
1215 if (ifp->if_format == XFS_DINODE_FMT_LOCAL ||
1216 tifp->if_format == XFS_DINODE_FMT_LOCAL)
1220 * if the target inode has less extents that then temporary inode then
1221 * why did userspace call us?
1223 if (ifp->if_nextents < tifp->if_nextents)
1227 * If we have to use the (expensive) rmap swap method, we can
1228 * handle any number of extents and any format.
1230 if (xfs_sb_version_hasrmapbt(&ip->i_mount->m_sb))
1234 * if the target inode is in extent form and the temp inode is in btree
1235 * form then we will end up with the target inode in the wrong format
1236 * as we already know there are less extents in the temp inode.
1238 if (ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
1239 tifp->if_format == XFS_DINODE_FMT_BTREE)
1242 /* Check temp in extent form to max in target */
1243 if (tifp->if_format == XFS_DINODE_FMT_EXTENTS &&
1244 tifp->if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1247 /* Check target in extent form to max in temp */
1248 if (ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
1249 ifp->if_nextents > XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1253 * If we are in a btree format, check that the temp root block will fit
1254 * in the target and that it has enough extents to be in btree format
1257 * Note that we have to be careful to allow btree->extent conversions
1258 * (a common defrag case) which will occur when the temp inode is in
1261 if (tifp->if_format == XFS_DINODE_FMT_BTREE) {
1262 if (XFS_IFORK_Q(ip) &&
1263 XFS_BMAP_BMDR_SPACE(tifp->if_broot) > XFS_IFORK_BOFF(ip))
1265 if (tifp->if_nextents <= XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1269 /* Reciprocal target->temp btree format checks */
1270 if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
1271 if (XFS_IFORK_Q(tip) &&
1272 XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip))
1274 if (ifp->if_nextents <= XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1282 xfs_swap_extent_flush(
1283 struct xfs_inode *ip)
1287 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1290 truncate_pagecache_range(VFS_I(ip), 0, -1);
1292 /* Verify O_DIRECT for ftmp */
1293 if (VFS_I(ip)->i_mapping->nrpages)
1299 * Move extents from one file to another, when rmap is enabled.
1302 xfs_swap_extent_rmap(
1303 struct xfs_trans **tpp,
1304 struct xfs_inode *ip,
1305 struct xfs_inode *tip)
1307 struct xfs_trans *tp = *tpp;
1308 struct xfs_bmbt_irec irec;
1309 struct xfs_bmbt_irec uirec;
1310 struct xfs_bmbt_irec tirec;
1311 xfs_fileoff_t offset_fsb;
1312 xfs_fileoff_t end_fsb;
1313 xfs_filblks_t count_fsb;
1318 uint64_t tip_flags2;
1321 * If the source file has shared blocks, we must flag the donor
1322 * file as having shared blocks so that we get the shared-block
1323 * rmap functions when we go to fix up the rmaps. The flags
1324 * will be switch for reals later.
1326 tip_flags2 = tip->i_d.di_flags2;
1327 if (ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)
1328 tip->i_d.di_flags2 |= XFS_DIFLAG2_REFLINK;
1331 end_fsb = XFS_B_TO_FSB(ip->i_mount, i_size_read(VFS_I(ip)));
1332 count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb);
1335 /* Read extent from the donor file */
1337 error = xfs_bmapi_read(tip, offset_fsb, count_fsb, &tirec,
1341 ASSERT(nimaps == 1);
1342 ASSERT(tirec.br_startblock != DELAYSTARTBLOCK);
1344 trace_xfs_swap_extent_rmap_remap(tip, &tirec);
1345 ilen = tirec.br_blockcount;
1347 /* Unmap the old blocks in the source file. */
1348 while (tirec.br_blockcount) {
1349 ASSERT(tp->t_firstblock == NULLFSBLOCK);
1350 trace_xfs_swap_extent_rmap_remap_piece(tip, &tirec);
1352 /* Read extent from the source file */
1354 error = xfs_bmapi_read(ip, tirec.br_startoff,
1355 tirec.br_blockcount, &irec,
1359 ASSERT(nimaps == 1);
1360 ASSERT(tirec.br_startoff == irec.br_startoff);
1361 trace_xfs_swap_extent_rmap_remap_piece(ip, &irec);
1363 /* Trim the extent. */
1365 uirec.br_blockcount = rlen = min_t(xfs_filblks_t,
1366 tirec.br_blockcount,
1367 irec.br_blockcount);
1368 trace_xfs_swap_extent_rmap_remap_piece(tip, &uirec);
1370 if (xfs_bmap_is_real_extent(&uirec)) {
1371 error = xfs_iext_count_may_overflow(ip,
1373 XFS_IEXT_SWAP_RMAP_CNT);
1378 if (xfs_bmap_is_real_extent(&irec)) {
1379 error = xfs_iext_count_may_overflow(tip,
1381 XFS_IEXT_SWAP_RMAP_CNT);
1386 /* Remove the mapping from the donor file. */
1387 xfs_bmap_unmap_extent(tp, tip, &uirec);
1389 /* Remove the mapping from the source file. */
1390 xfs_bmap_unmap_extent(tp, ip, &irec);
1392 /* Map the donor file's blocks into the source file. */
1393 xfs_bmap_map_extent(tp, ip, &uirec);
1395 /* Map the source file's blocks into the donor file. */
1396 xfs_bmap_map_extent(tp, tip, &irec);
1398 error = xfs_defer_finish(tpp);
1403 tirec.br_startoff += rlen;
1404 if (tirec.br_startblock != HOLESTARTBLOCK &&
1405 tirec.br_startblock != DELAYSTARTBLOCK)
1406 tirec.br_startblock += rlen;
1407 tirec.br_blockcount -= rlen;
1415 tip->i_d.di_flags2 = tip_flags2;
1419 trace_xfs_swap_extent_rmap_error(ip, error, _RET_IP_);
1420 tip->i_d.di_flags2 = tip_flags2;
1424 /* Swap the extents of two files by swapping data forks. */
1426 xfs_swap_extent_forks(
1427 struct xfs_trans *tp,
1428 struct xfs_inode *ip,
1429 struct xfs_inode *tip,
1431 int *target_log_flags)
1433 xfs_filblks_t aforkblks = 0;
1434 xfs_filblks_t taforkblks = 0;
1440 * Count the number of extended attribute blocks
1442 if (XFS_IFORK_Q(ip) && ip->i_afp->if_nextents > 0 &&
1443 ip->i_afp->if_format != XFS_DINODE_FMT_LOCAL) {
1444 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &junk,
1449 if (XFS_IFORK_Q(tip) && tip->i_afp->if_nextents > 0 &&
1450 tip->i_afp->if_format != XFS_DINODE_FMT_LOCAL) {
1451 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK, &junk,
1458 * Btree format (v3) inodes have the inode number stamped in the bmbt
1459 * block headers. We can't start changing the bmbt blocks until the
1460 * inode owner change is logged so recovery does the right thing in the
1461 * event of a crash. Set the owner change log flags now and leave the
1462 * bmbt scan as the last step.
1464 if (xfs_sb_version_has_v3inode(&ip->i_mount->m_sb)) {
1465 if (ip->i_df.if_format == XFS_DINODE_FMT_BTREE)
1466 (*target_log_flags) |= XFS_ILOG_DOWNER;
1467 if (tip->i_df.if_format == XFS_DINODE_FMT_BTREE)
1468 (*src_log_flags) |= XFS_ILOG_DOWNER;
1472 * Swap the data forks of the inodes
1474 swap(ip->i_df, tip->i_df);
1477 * Fix the on-disk inode values
1479 tmp = (uint64_t)ip->i_d.di_nblocks;
1480 ip->i_d.di_nblocks = tip->i_d.di_nblocks - taforkblks + aforkblks;
1481 tip->i_d.di_nblocks = tmp + taforkblks - aforkblks;
1484 * The extents in the source inode could still contain speculative
1485 * preallocation beyond EOF (e.g. the file is open but not modified
1486 * while defrag is in progress). In that case, we need to copy over the
1487 * number of delalloc blocks the data fork in the source inode is
1488 * tracking beyond EOF so that when the fork is truncated away when the
1489 * temporary inode is unlinked we don't underrun the i_delayed_blks
1490 * counter on that inode.
1492 ASSERT(tip->i_delayed_blks == 0);
1493 tip->i_delayed_blks = ip->i_delayed_blks;
1494 ip->i_delayed_blks = 0;
1496 switch (ip->i_df.if_format) {
1497 case XFS_DINODE_FMT_EXTENTS:
1498 (*src_log_flags) |= XFS_ILOG_DEXT;
1500 case XFS_DINODE_FMT_BTREE:
1501 ASSERT(!xfs_sb_version_has_v3inode(&ip->i_mount->m_sb) ||
1502 (*src_log_flags & XFS_ILOG_DOWNER));
1503 (*src_log_flags) |= XFS_ILOG_DBROOT;
1507 switch (tip->i_df.if_format) {
1508 case XFS_DINODE_FMT_EXTENTS:
1509 (*target_log_flags) |= XFS_ILOG_DEXT;
1511 case XFS_DINODE_FMT_BTREE:
1512 (*target_log_flags) |= XFS_ILOG_DBROOT;
1513 ASSERT(!xfs_sb_version_has_v3inode(&ip->i_mount->m_sb) ||
1514 (*target_log_flags & XFS_ILOG_DOWNER));
1522 * Fix up the owners of the bmbt blocks to refer to the current inode. The
1523 * change owner scan attempts to order all modified buffers in the current
1524 * transaction. In the event of ordered buffer failure, the offending buffer is
1525 * physically logged as a fallback and the scan returns -EAGAIN. We must roll
1526 * the transaction in this case to replenish the fallback log reservation and
1527 * restart the scan. This process repeats until the scan completes.
1530 xfs_swap_change_owner(
1531 struct xfs_trans **tpp,
1532 struct xfs_inode *ip,
1533 struct xfs_inode *tmpip)
1536 struct xfs_trans *tp = *tpp;
1539 error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK, ip->i_ino,
1541 /* success or fatal error */
1542 if (error != -EAGAIN)
1545 error = xfs_trans_roll(tpp);
1551 * Redirty both inodes so they can relog and keep the log tail
1554 xfs_trans_ijoin(tp, ip, 0);
1555 xfs_trans_ijoin(tp, tmpip, 0);
1556 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1557 xfs_trans_log_inode(tp, tmpip, XFS_ILOG_CORE);
1565 struct xfs_inode *ip, /* target inode */
1566 struct xfs_inode *tip, /* tmp inode */
1567 struct xfs_swapext *sxp)
1569 struct xfs_mount *mp = ip->i_mount;
1570 struct xfs_trans *tp;
1571 struct xfs_bstat *sbp = &sxp->sx_stat;
1572 int src_log_flags, target_log_flags;
1577 unsigned int flags = 0;
1580 * Lock the inodes against other IO, page faults and truncate to
1581 * begin with. Then we can ensure the inodes are flushed and have no
1582 * page cache safely. Once we have done this we can take the ilocks and
1583 * do the rest of the checks.
1585 lock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1586 lock_flags = XFS_MMAPLOCK_EXCL;
1587 xfs_lock_two_inodes(ip, XFS_MMAPLOCK_EXCL, tip, XFS_MMAPLOCK_EXCL);
1589 /* Verify that both files have the same format */
1590 if ((VFS_I(ip)->i_mode & S_IFMT) != (VFS_I(tip)->i_mode & S_IFMT)) {
1595 /* Verify both files are either real-time or non-realtime */
1596 if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
1601 error = xfs_qm_dqattach(ip);
1605 error = xfs_qm_dqattach(tip);
1609 error = xfs_swap_extent_flush(ip);
1612 error = xfs_swap_extent_flush(tip);
1616 if (xfs_inode_has_cow_data(tip)) {
1617 error = xfs_reflink_cancel_cow_range(tip, 0, NULLFILEOFF, true);
1623 * Extent "swapping" with rmap requires a permanent reservation and
1624 * a block reservation because it's really just a remap operation
1625 * performed with log redo items!
1627 if (xfs_sb_version_hasrmapbt(&mp->m_sb)) {
1628 int w = XFS_DATA_FORK;
1629 uint32_t ipnext = ip->i_df.if_nextents;
1630 uint32_t tipnext = tip->i_df.if_nextents;
1633 * Conceptually this shouldn't affect the shape of either bmbt,
1634 * but since we atomically move extents one by one, we reserve
1635 * enough space to rebuild both trees.
1637 resblks = XFS_SWAP_RMAP_SPACE_RES(mp, ipnext, w);
1638 resblks += XFS_SWAP_RMAP_SPACE_RES(mp, tipnext, w);
1641 * If either inode straddles a bmapbt block allocation boundary,
1642 * the rmapbt algorithm triggers repeated allocs and frees as
1643 * extents are remapped. This can exhaust the block reservation
1644 * prematurely and cause shutdown. Return freed blocks to the
1645 * transaction reservation to counter this behavior.
1647 flags |= XFS_TRANS_RES_FDBLKS;
1649 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, flags,
1655 * Lock and join the inodes to the tansaction so that transaction commit
1656 * or cancel will unlock the inodes from this point onwards.
1658 xfs_lock_two_inodes(ip, XFS_ILOCK_EXCL, tip, XFS_ILOCK_EXCL);
1659 lock_flags |= XFS_ILOCK_EXCL;
1660 xfs_trans_ijoin(tp, ip, 0);
1661 xfs_trans_ijoin(tp, tip, 0);
1664 /* Verify all data are being swapped */
1665 if (sxp->sx_offset != 0 ||
1666 sxp->sx_length != ip->i_d.di_size ||
1667 sxp->sx_length != tip->i_d.di_size) {
1669 goto out_trans_cancel;
1672 trace_xfs_swap_extent_before(ip, 0);
1673 trace_xfs_swap_extent_before(tip, 1);
1675 /* check inode formats now that data is flushed */
1676 error = xfs_swap_extents_check_format(ip, tip);
1679 "%s: inode 0x%llx format is incompatible for exchanging.",
1680 __func__, ip->i_ino);
1681 goto out_trans_cancel;
1685 * Compare the current change & modify times with that
1686 * passed in. If they differ, we abort this swap.
1687 * This is the mechanism used to ensure the calling
1688 * process that the file was not changed out from
1691 if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
1692 (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
1693 (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
1694 (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
1696 goto out_trans_cancel;
1700 * Note the trickiness in setting the log flags - we set the owner log
1701 * flag on the opposite inode (i.e. the inode we are setting the new
1702 * owner to be) because once we swap the forks and log that, log
1703 * recovery is going to see the fork as owned by the swapped inode,
1704 * not the pre-swapped inodes.
1706 src_log_flags = XFS_ILOG_CORE;
1707 target_log_flags = XFS_ILOG_CORE;
1709 if (xfs_sb_version_hasrmapbt(&mp->m_sb))
1710 error = xfs_swap_extent_rmap(&tp, ip, tip);
1712 error = xfs_swap_extent_forks(tp, ip, tip, &src_log_flags,
1715 goto out_trans_cancel;
1717 /* Do we have to swap reflink flags? */
1718 if ((ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK) ^
1719 (tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK)) {
1720 f = ip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK;
1721 ip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1722 ip->i_d.di_flags2 |= tip->i_d.di_flags2 & XFS_DIFLAG2_REFLINK;
1723 tip->i_d.di_flags2 &= ~XFS_DIFLAG2_REFLINK;
1724 tip->i_d.di_flags2 |= f & XFS_DIFLAG2_REFLINK;
1727 /* Swap the cow forks. */
1728 if (xfs_sb_version_hasreflink(&mp->m_sb)) {
1729 ASSERT(!ip->i_cowfp ||
1730 ip->i_cowfp->if_format == XFS_DINODE_FMT_EXTENTS);
1731 ASSERT(!tip->i_cowfp ||
1732 tip->i_cowfp->if_format == XFS_DINODE_FMT_EXTENTS);
1734 swap(ip->i_cowfp, tip->i_cowfp);
1736 if (ip->i_cowfp && ip->i_cowfp->if_bytes)
1737 xfs_inode_set_cowblocks_tag(ip);
1739 xfs_inode_clear_cowblocks_tag(ip);
1740 if (tip->i_cowfp && tip->i_cowfp->if_bytes)
1741 xfs_inode_set_cowblocks_tag(tip);
1743 xfs_inode_clear_cowblocks_tag(tip);
1746 xfs_trans_log_inode(tp, ip, src_log_flags);
1747 xfs_trans_log_inode(tp, tip, target_log_flags);
1750 * The extent forks have been swapped, but crc=1,rmapbt=0 filesystems
1751 * have inode number owner values in the bmbt blocks that still refer to
1752 * the old inode. Scan each bmbt to fix up the owner values with the
1753 * inode number of the current inode.
1755 if (src_log_flags & XFS_ILOG_DOWNER) {
1756 error = xfs_swap_change_owner(&tp, ip, tip);
1758 goto out_trans_cancel;
1760 if (target_log_flags & XFS_ILOG_DOWNER) {
1761 error = xfs_swap_change_owner(&tp, tip, ip);
1763 goto out_trans_cancel;
1767 * If this is a synchronous mount, make sure that the
1768 * transaction goes to disk before returning to the user.
1770 if (mp->m_flags & XFS_MOUNT_WSYNC)
1771 xfs_trans_set_sync(tp);
1773 error = xfs_trans_commit(tp);
1775 trace_xfs_swap_extent_after(ip, 0);
1776 trace_xfs_swap_extent_after(tip, 1);
1779 xfs_iunlock(ip, lock_flags);
1780 xfs_iunlock(tip, lock_flags);
1781 unlock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1785 xfs_trans_cancel(tp);