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)
76 struct xfs_mount *mp = ap->ip->i_mount;
77 xfs_fileoff_t orig_offset = ap->offset;
79 xfs_extlen_t prod = 0; /* product factor for allocators */
80 xfs_extlen_t mod = 0; /* product factor for allocators */
81 xfs_extlen_t ralen = 0; /* realtime allocation length */
82 xfs_extlen_t align; /* minimum allocation alignment */
83 xfs_extlen_t orig_length = ap->length;
84 xfs_extlen_t minlen = mp->m_sb.sb_rextsize;
85 xfs_extlen_t raminlen;
86 bool rtlocked = false;
87 bool ignore_locality = false;
90 align = xfs_get_extsz_hint(ap->ip);
92 prod = align / mp->m_sb.sb_rextsize;
93 error = xfs_bmap_extsize_align(mp, &ap->got, &ap->prev,
95 ap->conv, &ap->offset, &ap->length);
99 ASSERT(ap->length % mp->m_sb.sb_rextsize == 0);
102 * If we shifted the file offset downward to satisfy an extent size
103 * hint, increase minlen by that amount so that the allocator won't
104 * give us an allocation that's too short to cover at least one of the
105 * blocks that the caller asked for.
107 if (ap->offset != orig_offset)
108 minlen += orig_offset - ap->offset;
111 * If the offset & length are not perfectly aligned
112 * then kill prod, it will just get us in trouble.
114 div_u64_rem(ap->offset, align, &mod);
115 if (mod || ap->length % align)
118 * Set ralen to be the actual requested length in rtextents.
120 ralen = ap->length / mp->m_sb.sb_rextsize;
122 * If the old value was close enough to MAXEXTLEN that
123 * we rounded up to it, cut it back so it's valid again.
124 * Note that if it's a really large request (bigger than
125 * MAXEXTLEN), we don't hear about that number, and can't
126 * adjust the starting point to match it.
128 if (ralen * mp->m_sb.sb_rextsize >= MAXEXTLEN)
129 ralen = MAXEXTLEN / mp->m_sb.sb_rextsize;
132 * Lock out modifications to both the RT bitmap and summary inodes
135 xfs_ilock(mp->m_rbmip, XFS_ILOCK_EXCL|XFS_ILOCK_RTBITMAP);
136 xfs_trans_ijoin(ap->tp, mp->m_rbmip, XFS_ILOCK_EXCL);
137 xfs_ilock(mp->m_rsumip, XFS_ILOCK_EXCL|XFS_ILOCK_RTSUM);
138 xfs_trans_ijoin(ap->tp, mp->m_rsumip, XFS_ILOCK_EXCL);
143 * If it's an allocation to an empty file at offset 0,
144 * pick an extent that will space things out in the rt area.
146 if (ap->eof && ap->offset == 0) {
147 xfs_rtblock_t rtx; /* realtime extent no */
149 error = xfs_rtpick_extent(mp, ap->tp, ralen, &rtx);
152 ap->blkno = rtx * mp->m_sb.sb_rextsize;
157 xfs_bmap_adjacent(ap);
160 * Realtime allocation, done through xfs_rtallocate_extent.
165 do_div(ap->blkno, mp->m_sb.sb_rextsize);
168 raminlen = max_t(xfs_extlen_t, 1, minlen / mp->m_sb.sb_rextsize);
169 error = xfs_rtallocate_extent(ap->tp, ap->blkno, raminlen, ap->length,
170 &ralen, ap->wasdel, prod, &rtb);
174 if (rtb != NULLRTBLOCK) {
175 ap->blkno = rtb * mp->m_sb.sb_rextsize;
176 ap->length = ralen * mp->m_sb.sb_rextsize;
177 ap->ip->i_nblocks += ap->length;
178 xfs_trans_log_inode(ap->tp, ap->ip, XFS_ILOG_CORE);
180 ap->ip->i_delayed_blks -= ap->length;
182 * Adjust the disk quota also. This was reserved
185 xfs_trans_mod_dquot_byino(ap->tp, ap->ip,
186 ap->wasdel ? XFS_TRANS_DQ_DELRTBCOUNT :
187 XFS_TRANS_DQ_RTBCOUNT, ap->length);
191 if (align > mp->m_sb.sb_rextsize) {
193 * We previously enlarged the request length to try to satisfy
194 * an extent size hint. The allocator didn't return anything,
195 * so reset the parameters to the original values and try again
196 * without alignment criteria.
198 ap->offset = orig_offset;
199 ap->length = orig_length;
200 minlen = align = mp->m_sb.sb_rextsize;
204 if (!ignore_locality && ap->blkno != 0) {
206 * If we can't allocate near a specific rt extent, try again
207 * without locality criteria.
209 ignore_locality = true;
213 ap->blkno = NULLFSBLOCK;
217 #endif /* CONFIG_XFS_RT */
220 * Extent tree block counting routines.
224 * Count leaf blocks given a range of extent records. Delayed allocation
225 * extents are not counted towards the totals.
228 xfs_bmap_count_leaves(
229 struct xfs_ifork *ifp,
230 xfs_filblks_t *count)
232 struct xfs_iext_cursor icur;
233 struct xfs_bmbt_irec got;
234 xfs_extnum_t numrecs = 0;
236 for_each_xfs_iext(ifp, &icur, &got) {
237 if (!isnullstartblock(got.br_startblock)) {
238 *count += got.br_blockcount;
247 * Count fsblocks of the given fork. Delayed allocation extents are
248 * not counted towards the totals.
251 xfs_bmap_count_blocks(
252 struct xfs_trans *tp,
253 struct xfs_inode *ip,
255 xfs_extnum_t *nextents,
256 xfs_filblks_t *count)
258 struct xfs_mount *mp = ip->i_mount;
259 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
260 struct xfs_btree_cur *cur;
261 xfs_extlen_t btblocks = 0;
270 switch (ifp->if_format) {
271 case XFS_DINODE_FMT_BTREE:
272 error = xfs_iread_extents(tp, ip, whichfork);
276 cur = xfs_bmbt_init_cursor(mp, tp, ip, whichfork);
277 error = xfs_btree_count_blocks(cur, &btblocks);
278 xfs_btree_del_cursor(cur, error);
283 * xfs_btree_count_blocks includes the root block contained in
284 * the inode fork in @btblocks, so subtract one because we're
285 * only interested in allocated disk blocks.
287 *count += btblocks - 1;
290 case XFS_DINODE_FMT_EXTENTS:
291 *nextents = xfs_bmap_count_leaves(ifp, count);
299 xfs_getbmap_report_one(
300 struct xfs_inode *ip,
301 struct getbmapx *bmv,
302 struct kgetbmap *out,
304 struct xfs_bmbt_irec *got)
306 struct kgetbmap *p = out + bmv->bmv_entries;
310 error = xfs_reflink_trim_around_shared(ip, got, &shared);
314 if (isnullstartblock(got->br_startblock) ||
315 got->br_startblock == DELAYSTARTBLOCK) {
317 * Delalloc extents that start beyond EOF can occur due to
318 * speculative EOF allocation when the delalloc extent is larger
319 * than the largest freespace extent at conversion time. These
320 * extents cannot be converted by data writeback, so can exist
321 * here even if we are not supposed to be finding delalloc
324 if (got->br_startoff < XFS_B_TO_FSB(ip->i_mount, XFS_ISIZE(ip)))
325 ASSERT((bmv->bmv_iflags & BMV_IF_DELALLOC) != 0);
327 p->bmv_oflags |= BMV_OF_DELALLOC;
330 p->bmv_block = xfs_fsb_to_db(ip, got->br_startblock);
333 if (got->br_state == XFS_EXT_UNWRITTEN &&
334 (bmv->bmv_iflags & BMV_IF_PREALLOC))
335 p->bmv_oflags |= BMV_OF_PREALLOC;
338 p->bmv_oflags |= BMV_OF_SHARED;
340 p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, got->br_startoff);
341 p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, got->br_blockcount);
343 bmv->bmv_offset = p->bmv_offset + p->bmv_length;
344 bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset);
350 xfs_getbmap_report_hole(
351 struct xfs_inode *ip,
352 struct getbmapx *bmv,
353 struct kgetbmap *out,
358 struct kgetbmap *p = out + bmv->bmv_entries;
360 if (bmv->bmv_iflags & BMV_IF_NO_HOLES)
364 p->bmv_offset = XFS_FSB_TO_BB(ip->i_mount, bno);
365 p->bmv_length = XFS_FSB_TO_BB(ip->i_mount, end - bno);
367 bmv->bmv_offset = p->bmv_offset + p->bmv_length;
368 bmv->bmv_length = max(0LL, bmv_end - bmv->bmv_offset);
374 struct getbmapx *bmv)
376 return bmv->bmv_length == 0 || bmv->bmv_entries >= bmv->bmv_count - 1;
380 xfs_getbmap_next_rec(
381 struct xfs_bmbt_irec *rec,
382 xfs_fileoff_t total_end)
384 xfs_fileoff_t end = rec->br_startoff + rec->br_blockcount;
386 if (end == total_end)
389 rec->br_startoff += rec->br_blockcount;
390 if (!isnullstartblock(rec->br_startblock) &&
391 rec->br_startblock != DELAYSTARTBLOCK)
392 rec->br_startblock += rec->br_blockcount;
393 rec->br_blockcount = total_end - end;
398 * Get inode's extents as described in bmv, and format for output.
399 * Calls formatter to fill the user's buffer until all extents
400 * are mapped, until the passed-in bmv->bmv_count slots have
401 * been filled, or until the formatter short-circuits the loop,
402 * if it is tracking filled-in extents on its own.
406 struct xfs_inode *ip,
407 struct getbmapx *bmv, /* user bmap structure */
408 struct kgetbmap *out)
410 struct xfs_mount *mp = ip->i_mount;
411 int iflags = bmv->bmv_iflags;
412 int whichfork, lock, error = 0;
413 int64_t bmv_end, max_len;
414 xfs_fileoff_t bno, first_bno;
415 struct xfs_ifork *ifp;
416 struct xfs_bmbt_irec got, rec;
418 struct xfs_iext_cursor icur;
420 if (bmv->bmv_iflags & ~BMV_IF_VALID)
423 /* Only allow CoW fork queries if we're debugging. */
424 if (iflags & BMV_IF_COWFORK)
427 if ((iflags & BMV_IF_ATTRFORK) && (iflags & BMV_IF_COWFORK))
430 if (bmv->bmv_length < -1)
432 bmv->bmv_entries = 0;
433 if (bmv->bmv_length == 0)
436 if (iflags & BMV_IF_ATTRFORK)
437 whichfork = XFS_ATTR_FORK;
438 else if (iflags & BMV_IF_COWFORK)
439 whichfork = XFS_COW_FORK;
441 whichfork = XFS_DATA_FORK;
442 ifp = XFS_IFORK_PTR(ip, whichfork);
444 xfs_ilock(ip, XFS_IOLOCK_SHARED);
447 if (!XFS_IFORK_Q(ip))
448 goto out_unlock_iolock;
451 lock = xfs_ilock_attr_map_shared(ip);
454 /* No CoW fork? Just return */
456 goto out_unlock_iolock;
458 if (xfs_get_cowextsz_hint(ip))
459 max_len = mp->m_super->s_maxbytes;
461 max_len = XFS_ISIZE(ip);
463 lock = XFS_ILOCK_SHARED;
467 if (!(iflags & BMV_IF_DELALLOC) &&
468 (ip->i_delayed_blks || XFS_ISIZE(ip) > ip->i_disk_size)) {
469 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
471 goto out_unlock_iolock;
474 * Even after flushing the inode, there can still be
475 * delalloc blocks on the inode beyond EOF due to
476 * speculative preallocation. These are not removed
477 * until the release function is called or the inode
478 * is inactivated. Hence we cannot assert here that
479 * ip->i_delayed_blks == 0.
483 if (xfs_get_extsz_hint(ip) ||
485 (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND)))
486 max_len = mp->m_super->s_maxbytes;
488 max_len = XFS_ISIZE(ip);
490 lock = xfs_ilock_data_map_shared(ip);
494 switch (ifp->if_format) {
495 case XFS_DINODE_FMT_EXTENTS:
496 case XFS_DINODE_FMT_BTREE:
498 case XFS_DINODE_FMT_LOCAL:
499 /* Local format inode forks report no extents. */
500 goto out_unlock_ilock;
503 goto out_unlock_ilock;
506 if (bmv->bmv_length == -1) {
507 max_len = XFS_FSB_TO_BB(mp, XFS_B_TO_FSB(mp, max_len));
508 bmv->bmv_length = max(0LL, max_len - bmv->bmv_offset);
511 bmv_end = bmv->bmv_offset + bmv->bmv_length;
513 first_bno = bno = XFS_BB_TO_FSBT(mp, bmv->bmv_offset);
514 len = XFS_BB_TO_FSB(mp, bmv->bmv_length);
516 error = xfs_iread_extents(NULL, ip, whichfork);
518 goto out_unlock_ilock;
520 if (!xfs_iext_lookup_extent(ip, ifp, bno, &icur, &got)) {
522 * Report a whole-file hole if the delalloc flag is set to
523 * stay compatible with the old implementation.
525 if (iflags & BMV_IF_DELALLOC)
526 xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno,
527 XFS_B_TO_FSB(mp, XFS_ISIZE(ip)));
528 goto out_unlock_ilock;
531 while (!xfs_getbmap_full(bmv)) {
532 xfs_trim_extent(&got, first_bno, len);
535 * Report an entry for a hole if this extent doesn't directly
536 * follow the previous one.
538 if (got.br_startoff > bno) {
539 xfs_getbmap_report_hole(ip, bmv, out, bmv_end, bno,
541 if (xfs_getbmap_full(bmv))
546 * In order to report shared extents accurately, we report each
547 * distinct shared / unshared part of a single bmbt record with
548 * an individual getbmapx record.
550 bno = got.br_startoff + got.br_blockcount;
553 error = xfs_getbmap_report_one(ip, bmv, out, bmv_end,
555 if (error || xfs_getbmap_full(bmv))
556 goto out_unlock_ilock;
557 } while (xfs_getbmap_next_rec(&rec, bno));
559 if (!xfs_iext_next_extent(ifp, &icur, &got)) {
560 xfs_fileoff_t end = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
562 out[bmv->bmv_entries - 1].bmv_oflags |= BMV_OF_LAST;
564 if (whichfork != XFS_ATTR_FORK && bno < end &&
565 !xfs_getbmap_full(bmv)) {
566 xfs_getbmap_report_hole(ip, bmv, out, bmv_end,
572 if (bno >= first_bno + len)
577 xfs_iunlock(ip, lock);
579 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
584 * Dead simple method of punching delalyed allocation blocks from a range in
585 * the inode. This will always punch out both the start and end blocks, even
586 * if the ranges only partially overlap them, so it is up to the caller to
587 * ensure that partial blocks are not passed in.
590 xfs_bmap_punch_delalloc_range(
591 struct xfs_inode *ip,
592 xfs_fileoff_t start_fsb,
593 xfs_fileoff_t length)
595 struct xfs_ifork *ifp = &ip->i_df;
596 xfs_fileoff_t end_fsb = start_fsb + length;
597 struct xfs_bmbt_irec got, del;
598 struct xfs_iext_cursor icur;
601 ASSERT(!xfs_need_iread_extents(ifp));
603 xfs_ilock(ip, XFS_ILOCK_EXCL);
604 if (!xfs_iext_lookup_extent_before(ip, ifp, &end_fsb, &icur, &got))
607 while (got.br_startoff + got.br_blockcount > start_fsb) {
609 xfs_trim_extent(&del, start_fsb, length);
612 * A delete can push the cursor forward. Step back to the
613 * previous extent on non-delalloc or extents outside the
616 if (!del.br_blockcount ||
617 !isnullstartblock(del.br_startblock)) {
618 if (!xfs_iext_prev_extent(ifp, &icur, &got))
623 error = xfs_bmap_del_extent_delay(ip, XFS_DATA_FORK, &icur,
625 if (error || !xfs_iext_get_extent(ifp, &icur, &got))
630 xfs_iunlock(ip, XFS_ILOCK_EXCL);
635 * Test whether it is appropriate to check an inode for and free post EOF
636 * blocks. The 'force' parameter determines whether we should also consider
637 * regular files that are marked preallocated or append-only.
640 xfs_can_free_eofblocks(
641 struct xfs_inode *ip,
644 struct xfs_bmbt_irec imap;
645 struct xfs_mount *mp = ip->i_mount;
646 xfs_fileoff_t end_fsb;
647 xfs_fileoff_t last_fsb;
652 * Caller must either hold the exclusive io lock; or be inactivating
653 * the inode, which guarantees there are no other users of the inode.
655 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL) ||
656 (VFS_I(ip)->i_state & I_FREEING));
658 /* prealloc/delalloc exists only on regular files */
659 if (!S_ISREG(VFS_I(ip)->i_mode))
663 * Zero sized files with no cached pages and delalloc blocks will not
664 * have speculative prealloc/delalloc blocks to remove.
666 if (VFS_I(ip)->i_size == 0 &&
667 VFS_I(ip)->i_mapping->nrpages == 0 &&
668 ip->i_delayed_blks == 0)
671 /* If we haven't read in the extent list, then don't do it now. */
672 if (xfs_need_iread_extents(&ip->i_df))
676 * Do not free real preallocated or append-only files unless the file
677 * has delalloc blocks and we are forced to remove them.
679 if (ip->i_diflags & (XFS_DIFLAG_PREALLOC | XFS_DIFLAG_APPEND))
680 if (!force || ip->i_delayed_blks == 0)
684 * Do not try to free post-EOF blocks if EOF is beyond the end of the
685 * range supported by the page cache, because the truncation will loop
688 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_ISIZE(ip));
689 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
690 if (last_fsb <= end_fsb)
694 * Look up the mapping for the first block past EOF. If we can't find
695 * it, there's nothing to free.
697 xfs_ilock(ip, XFS_ILOCK_SHARED);
698 error = xfs_bmapi_read(ip, end_fsb, last_fsb - end_fsb, &imap, &nimaps,
700 xfs_iunlock(ip, XFS_ILOCK_SHARED);
701 if (error || nimaps == 0)
705 * If there's a real mapping there or there are delayed allocation
706 * reservations, then we have post-EOF blocks to try to free.
708 return imap.br_startblock != HOLESTARTBLOCK || ip->i_delayed_blks;
712 * This is called to free any blocks beyond eof. The caller must hold
713 * IOLOCK_EXCL unless we are in the inode reclaim path and have the only
714 * reference to the inode.
718 struct xfs_inode *ip)
720 struct xfs_trans *tp;
721 struct xfs_mount *mp = ip->i_mount;
724 /* Attach the dquots to the inode up front. */
725 error = xfs_qm_dqattach(ip);
729 /* Wait on dio to ensure i_size has settled. */
730 inode_dio_wait(VFS_I(ip));
732 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_itruncate, 0, 0, 0, &tp);
734 ASSERT(xfs_is_shutdown(mp));
738 xfs_ilock(ip, XFS_ILOCK_EXCL);
739 xfs_trans_ijoin(tp, ip, 0);
742 * Do not update the on-disk file size. If we update the on-disk file
743 * size and then the system crashes before the contents of the file are
744 * flushed to disk then the files may be full of holes (ie NULL files
747 error = xfs_itruncate_extents_flags(&tp, ip, XFS_DATA_FORK,
748 XFS_ISIZE(ip), XFS_BMAPI_NODISCARD);
752 error = xfs_trans_commit(tp);
756 xfs_inode_clear_eofblocks_tag(ip);
761 * If we get an error at this point we simply don't
762 * bother truncating the file.
764 xfs_trans_cancel(tp);
766 xfs_iunlock(ip, XFS_ILOCK_EXCL);
771 xfs_alloc_file_space(
772 struct xfs_inode *ip,
776 xfs_mount_t *mp = ip->i_mount;
778 xfs_filblks_t allocated_fsb;
779 xfs_filblks_t allocatesize_fsb;
780 xfs_extlen_t extsz, temp;
781 xfs_fileoff_t startoffset_fsb;
782 xfs_fileoff_t endoffset_fsb;
786 xfs_bmbt_irec_t imaps[1], *imapp;
789 trace_xfs_alloc_file_space(ip);
791 if (xfs_is_shutdown(mp))
794 error = xfs_qm_dqattach(ip);
801 rt = XFS_IS_REALTIME_INODE(ip);
802 extsz = xfs_get_extsz_hint(ip);
807 startoffset_fsb = XFS_B_TO_FSBT(mp, offset);
808 endoffset_fsb = XFS_B_TO_FSB(mp, offset + count);
809 allocatesize_fsb = endoffset_fsb - startoffset_fsb;
812 * Allocate file space until done or until there is an error
814 while (allocatesize_fsb && !error) {
816 unsigned int dblocks, rblocks, resblks;
819 * Determine space reservations for data/realtime.
821 if (unlikely(extsz)) {
825 e = startoffset_fsb + allocatesize_fsb;
826 div_u64_rem(startoffset_fsb, extsz, &temp);
829 div_u64_rem(e, extsz, &temp);
834 e = allocatesize_fsb;
838 * The transaction reservation is limited to a 32-bit block
839 * count, hence we need to limit the number of blocks we are
840 * trying to reserve to avoid an overflow. We can't allocate
841 * more than @nimaps extents, and an extent is limited on disk
842 * to MAXEXTLEN (21 bits), so use that to enforce the limit.
844 resblks = min_t(xfs_fileoff_t, (e - s), (MAXEXTLEN * nimaps));
846 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
849 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, resblks);
853 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write,
854 dblocks, rblocks, false, &tp);
858 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
859 XFS_IEXT_ADD_NOSPLIT_CNT);
863 error = xfs_bmapi_write(tp, ip, startoffset_fsb,
864 allocatesize_fsb, XFS_BMAPI_PREALLOC, 0, imapp,
869 ip->i_diflags |= XFS_DIFLAG_PREALLOC;
870 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
872 error = xfs_trans_commit(tp);
873 xfs_iunlock(ip, XFS_ILOCK_EXCL);
877 allocated_fsb = imapp->br_blockcount;
884 startoffset_fsb += allocated_fsb;
885 allocatesize_fsb -= allocated_fsb;
891 xfs_trans_cancel(tp);
892 xfs_iunlock(ip, XFS_ILOCK_EXCL);
898 struct xfs_inode *ip,
899 xfs_fileoff_t startoffset_fsb,
900 xfs_filblks_t len_fsb,
903 struct xfs_mount *mp = ip->i_mount;
904 struct xfs_trans *tp;
905 uint resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
908 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, resblks, 0,
913 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
914 XFS_IEXT_PUNCH_HOLE_CNT);
916 goto out_trans_cancel;
918 error = xfs_bunmapi(tp, ip, startoffset_fsb, len_fsb, 0, 2, done);
920 goto out_trans_cancel;
922 error = xfs_trans_commit(tp);
924 xfs_iunlock(ip, XFS_ILOCK_EXCL);
928 xfs_trans_cancel(tp);
932 /* Caller must first wait for the completion of any pending DIOs if required. */
934 xfs_flush_unmap_range(
935 struct xfs_inode *ip,
939 struct xfs_mount *mp = ip->i_mount;
940 struct inode *inode = VFS_I(ip);
941 xfs_off_t rounding, start, end;
944 rounding = max_t(xfs_off_t, mp->m_sb.sb_blocksize, PAGE_SIZE);
945 start = round_down(offset, rounding);
946 end = round_up(offset + len, rounding) - 1;
948 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
951 truncate_pagecache_range(inode, start, end);
957 struct xfs_inode *ip,
961 struct xfs_mount *mp = ip->i_mount;
962 xfs_fileoff_t startoffset_fsb;
963 xfs_fileoff_t endoffset_fsb;
966 trace_xfs_free_file_space(ip);
968 error = xfs_qm_dqattach(ip);
972 if (len <= 0) /* if nothing being freed */
975 startoffset_fsb = XFS_B_TO_FSB(mp, offset);
976 endoffset_fsb = XFS_B_TO_FSBT(mp, offset + len);
978 /* We can only free complete realtime extents. */
979 if (XFS_IS_REALTIME_INODE(ip) && mp->m_sb.sb_rextsize > 1) {
980 startoffset_fsb = roundup_64(startoffset_fsb,
981 mp->m_sb.sb_rextsize);
982 endoffset_fsb = rounddown_64(endoffset_fsb,
983 mp->m_sb.sb_rextsize);
987 * Need to zero the stuff we're not freeing, on disk.
989 if (endoffset_fsb > startoffset_fsb) {
991 error = xfs_unmap_extent(ip, startoffset_fsb,
992 endoffset_fsb - startoffset_fsb, &done);
999 * Now that we've unmap all full blocks we'll have to zero out any
1000 * partial block at the beginning and/or end. xfs_zero_range is smart
1001 * enough to skip any holes, including those we just created, but we
1002 * must take care not to zero beyond EOF and enlarge i_size.
1004 if (offset >= XFS_ISIZE(ip))
1006 if (offset + len > XFS_ISIZE(ip))
1007 len = XFS_ISIZE(ip) - offset;
1008 error = xfs_zero_range(ip, offset, len, NULL);
1013 * If we zeroed right up to EOF and EOF straddles a page boundary we
1014 * must make sure that the post-EOF area is also zeroed because the
1015 * page could be mmap'd and xfs_zero_range doesn't do that for us.
1016 * Writeback of the eof page will do this, albeit clumsily.
1018 if (offset + len >= XFS_ISIZE(ip) && offset_in_page(offset + len) > 0) {
1019 error = filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
1020 round_down(offset + len, PAGE_SIZE), LLONG_MAX);
1028 struct xfs_inode *ip,
1031 struct xfs_mount *mp = ip->i_mount;
1035 * Trim eofblocks to avoid shifting uninitialized post-eof preallocation
1036 * into the accessible region of the file.
1038 if (xfs_can_free_eofblocks(ip, true)) {
1039 error = xfs_free_eofblocks(ip);
1045 * Shift operations must stabilize the start block offset boundary along
1046 * with the full range of the operation. If we don't, a COW writeback
1047 * completion could race with an insert, front merge with the start
1048 * extent (after split) during the shift and corrupt the file. Start
1049 * with the block just prior to the start to stabilize the boundary.
1051 offset = round_down(offset, mp->m_sb.sb_blocksize);
1053 offset -= mp->m_sb.sb_blocksize;
1056 * Writeback and invalidate cache for the remainder of the file as we're
1057 * about to shift down every extent from offset to EOF.
1059 error = xfs_flush_unmap_range(ip, offset, XFS_ISIZE(ip));
1064 * Clean out anything hanging around in the cow fork now that
1065 * we've flushed all the dirty data out to disk to avoid having
1066 * CoW extents at the wrong offsets.
1068 if (xfs_inode_has_cow_data(ip)) {
1069 error = xfs_reflink_cancel_cow_range(ip, offset, NULLFILEOFF,
1079 * xfs_collapse_file_space()
1080 * This routine frees disk space and shift extent for the given file.
1081 * The first thing we do is to free data blocks in the specified range
1082 * by calling xfs_free_file_space(). It would also sync dirty data
1083 * and invalidate page cache over the region on which collapse range
1084 * is working. And Shift extent records to the left to cover a hole.
1091 xfs_collapse_file_space(
1092 struct xfs_inode *ip,
1096 struct xfs_mount *mp = ip->i_mount;
1097 struct xfs_trans *tp;
1099 xfs_fileoff_t next_fsb = XFS_B_TO_FSB(mp, offset + len);
1100 xfs_fileoff_t shift_fsb = XFS_B_TO_FSB(mp, len);
1103 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1104 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
1106 trace_xfs_collapse_file_space(ip);
1108 error = xfs_free_file_space(ip, offset, len);
1112 error = xfs_prepare_shift(ip, offset);
1116 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0, 0, 0, &tp);
1120 xfs_ilock(ip, XFS_ILOCK_EXCL);
1121 xfs_trans_ijoin(tp, ip, 0);
1124 error = xfs_bmap_collapse_extents(tp, ip, &next_fsb, shift_fsb,
1127 goto out_trans_cancel;
1131 /* finish any deferred frees and roll the transaction */
1132 error = xfs_defer_finish(&tp);
1134 goto out_trans_cancel;
1137 error = xfs_trans_commit(tp);
1138 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1142 xfs_trans_cancel(tp);
1143 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1148 * xfs_insert_file_space()
1149 * This routine create hole space by shifting extents for the given file.
1150 * The first thing we do is to sync dirty data and invalidate page cache
1151 * over the region on which insert range is working. And split an extent
1152 * to two extents at given offset by calling xfs_bmap_split_extent.
1153 * And shift all extent records which are laying between [offset,
1154 * last allocated extent] to the right to reserve hole range.
1160 xfs_insert_file_space(
1161 struct xfs_inode *ip,
1165 struct xfs_mount *mp = ip->i_mount;
1166 struct xfs_trans *tp;
1168 xfs_fileoff_t stop_fsb = XFS_B_TO_FSB(mp, offset);
1169 xfs_fileoff_t next_fsb = NULLFSBLOCK;
1170 xfs_fileoff_t shift_fsb = XFS_B_TO_FSB(mp, len);
1173 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
1174 ASSERT(xfs_isilocked(ip, XFS_MMAPLOCK_EXCL));
1176 trace_xfs_insert_file_space(ip);
1178 error = xfs_bmap_can_insert_extents(ip, stop_fsb, shift_fsb);
1182 error = xfs_prepare_shift(ip, offset);
1186 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write,
1187 XFS_DIOSTRAT_SPACE_RES(mp, 0), 0, 0, &tp);
1191 xfs_ilock(ip, XFS_ILOCK_EXCL);
1192 xfs_trans_ijoin(tp, ip, 0);
1194 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
1195 XFS_IEXT_PUNCH_HOLE_CNT);
1197 goto out_trans_cancel;
1200 * The extent shifting code works on extent granularity. So, if stop_fsb
1201 * is not the starting block of extent, we need to split the extent at
1204 error = xfs_bmap_split_extent(tp, ip, stop_fsb);
1206 goto out_trans_cancel;
1209 error = xfs_defer_finish(&tp);
1211 goto out_trans_cancel;
1213 error = xfs_bmap_insert_extents(tp, ip, &next_fsb, shift_fsb,
1216 goto out_trans_cancel;
1219 error = xfs_trans_commit(tp);
1220 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1224 xfs_trans_cancel(tp);
1225 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1230 * We need to check that the format of the data fork in the temporary inode is
1231 * valid for the target inode before doing the swap. This is not a problem with
1232 * attr1 because of the fixed fork offset, but attr2 has a dynamically sized
1233 * data fork depending on the space the attribute fork is taking so we can get
1234 * invalid formats on the target inode.
1236 * E.g. target has space for 7 extents in extent format, temp inode only has
1237 * space for 6. If we defragment down to 7 extents, then the tmp format is a
1238 * btree, but when swapped it needs to be in extent format. Hence we can't just
1239 * blindly swap data forks on attr2 filesystems.
1241 * Note that we check the swap in both directions so that we don't end up with
1242 * a corrupt temporary inode, either.
1244 * Note that fixing the way xfs_fsr sets up the attribute fork in the source
1245 * inode will prevent this situation from occurring, so all we do here is
1246 * reject and log the attempt. basically we are putting the responsibility on
1247 * userspace to get this right.
1250 xfs_swap_extents_check_format(
1251 struct xfs_inode *ip, /* target inode */
1252 struct xfs_inode *tip) /* tmp inode */
1254 struct xfs_ifork *ifp = &ip->i_df;
1255 struct xfs_ifork *tifp = &tip->i_df;
1257 /* User/group/project quota ids must match if quotas are enforced. */
1258 if (XFS_IS_QUOTA_ON(ip->i_mount) &&
1259 (!uid_eq(VFS_I(ip)->i_uid, VFS_I(tip)->i_uid) ||
1260 !gid_eq(VFS_I(ip)->i_gid, VFS_I(tip)->i_gid) ||
1261 ip->i_projid != tip->i_projid))
1264 /* Should never get a local format */
1265 if (ifp->if_format == XFS_DINODE_FMT_LOCAL ||
1266 tifp->if_format == XFS_DINODE_FMT_LOCAL)
1270 * if the target inode has less extents that then temporary inode then
1271 * why did userspace call us?
1273 if (ifp->if_nextents < tifp->if_nextents)
1277 * If we have to use the (expensive) rmap swap method, we can
1278 * handle any number of extents and any format.
1280 if (xfs_has_rmapbt(ip->i_mount))
1284 * if the target inode is in extent form and the temp inode is in btree
1285 * form then we will end up with the target inode in the wrong format
1286 * as we already know there are less extents in the temp inode.
1288 if (ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
1289 tifp->if_format == XFS_DINODE_FMT_BTREE)
1292 /* Check temp in extent form to max in target */
1293 if (tifp->if_format == XFS_DINODE_FMT_EXTENTS &&
1294 tifp->if_nextents > XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1297 /* Check target in extent form to max in temp */
1298 if (ifp->if_format == XFS_DINODE_FMT_EXTENTS &&
1299 ifp->if_nextents > XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1303 * If we are in a btree format, check that the temp root block will fit
1304 * in the target and that it has enough extents to be in btree format
1307 * Note that we have to be careful to allow btree->extent conversions
1308 * (a common defrag case) which will occur when the temp inode is in
1311 if (tifp->if_format == XFS_DINODE_FMT_BTREE) {
1312 if (XFS_IFORK_Q(ip) &&
1313 XFS_BMAP_BMDR_SPACE(tifp->if_broot) > XFS_IFORK_BOFF(ip))
1315 if (tifp->if_nextents <= XFS_IFORK_MAXEXT(ip, XFS_DATA_FORK))
1319 /* Reciprocal target->temp btree format checks */
1320 if (ifp->if_format == XFS_DINODE_FMT_BTREE) {
1321 if (XFS_IFORK_Q(tip) &&
1322 XFS_BMAP_BMDR_SPACE(ip->i_df.if_broot) > XFS_IFORK_BOFF(tip))
1324 if (ifp->if_nextents <= XFS_IFORK_MAXEXT(tip, XFS_DATA_FORK))
1332 xfs_swap_extent_flush(
1333 struct xfs_inode *ip)
1337 error = filemap_write_and_wait(VFS_I(ip)->i_mapping);
1340 truncate_pagecache_range(VFS_I(ip), 0, -1);
1342 /* Verify O_DIRECT for ftmp */
1343 if (VFS_I(ip)->i_mapping->nrpages)
1349 * Move extents from one file to another, when rmap is enabled.
1352 xfs_swap_extent_rmap(
1353 struct xfs_trans **tpp,
1354 struct xfs_inode *ip,
1355 struct xfs_inode *tip)
1357 struct xfs_trans *tp = *tpp;
1358 struct xfs_bmbt_irec irec;
1359 struct xfs_bmbt_irec uirec;
1360 struct xfs_bmbt_irec tirec;
1361 xfs_fileoff_t offset_fsb;
1362 xfs_fileoff_t end_fsb;
1363 xfs_filblks_t count_fsb;
1368 uint64_t tip_flags2;
1371 * If the source file has shared blocks, we must flag the donor
1372 * file as having shared blocks so that we get the shared-block
1373 * rmap functions when we go to fix up the rmaps. The flags
1374 * will be switch for reals later.
1376 tip_flags2 = tip->i_diflags2;
1377 if (ip->i_diflags2 & XFS_DIFLAG2_REFLINK)
1378 tip->i_diflags2 |= XFS_DIFLAG2_REFLINK;
1381 end_fsb = XFS_B_TO_FSB(ip->i_mount, i_size_read(VFS_I(ip)));
1382 count_fsb = (xfs_filblks_t)(end_fsb - offset_fsb);
1385 /* Read extent from the donor file */
1387 error = xfs_bmapi_read(tip, offset_fsb, count_fsb, &tirec,
1391 ASSERT(nimaps == 1);
1392 ASSERT(tirec.br_startblock != DELAYSTARTBLOCK);
1394 trace_xfs_swap_extent_rmap_remap(tip, &tirec);
1395 ilen = tirec.br_blockcount;
1397 /* Unmap the old blocks in the source file. */
1398 while (tirec.br_blockcount) {
1399 ASSERT(tp->t_firstblock == NULLFSBLOCK);
1400 trace_xfs_swap_extent_rmap_remap_piece(tip, &tirec);
1402 /* Read extent from the source file */
1404 error = xfs_bmapi_read(ip, tirec.br_startoff,
1405 tirec.br_blockcount, &irec,
1409 ASSERT(nimaps == 1);
1410 ASSERT(tirec.br_startoff == irec.br_startoff);
1411 trace_xfs_swap_extent_rmap_remap_piece(ip, &irec);
1413 /* Trim the extent. */
1415 uirec.br_blockcount = rlen = min_t(xfs_filblks_t,
1416 tirec.br_blockcount,
1417 irec.br_blockcount);
1418 trace_xfs_swap_extent_rmap_remap_piece(tip, &uirec);
1420 if (xfs_bmap_is_real_extent(&uirec)) {
1421 error = xfs_iext_count_may_overflow(ip,
1423 XFS_IEXT_SWAP_RMAP_CNT);
1428 if (xfs_bmap_is_real_extent(&irec)) {
1429 error = xfs_iext_count_may_overflow(tip,
1431 XFS_IEXT_SWAP_RMAP_CNT);
1436 /* Remove the mapping from the donor file. */
1437 xfs_bmap_unmap_extent(tp, tip, &uirec);
1439 /* Remove the mapping from the source file. */
1440 xfs_bmap_unmap_extent(tp, ip, &irec);
1442 /* Map the donor file's blocks into the source file. */
1443 xfs_bmap_map_extent(tp, ip, &uirec);
1445 /* Map the source file's blocks into the donor file. */
1446 xfs_bmap_map_extent(tp, tip, &irec);
1448 error = xfs_defer_finish(tpp);
1453 tirec.br_startoff += rlen;
1454 if (tirec.br_startblock != HOLESTARTBLOCK &&
1455 tirec.br_startblock != DELAYSTARTBLOCK)
1456 tirec.br_startblock += rlen;
1457 tirec.br_blockcount -= rlen;
1465 tip->i_diflags2 = tip_flags2;
1469 trace_xfs_swap_extent_rmap_error(ip, error, _RET_IP_);
1470 tip->i_diflags2 = tip_flags2;
1474 /* Swap the extents of two files by swapping data forks. */
1476 xfs_swap_extent_forks(
1477 struct xfs_trans *tp,
1478 struct xfs_inode *ip,
1479 struct xfs_inode *tip,
1481 int *target_log_flags)
1483 xfs_filblks_t aforkblks = 0;
1484 xfs_filblks_t taforkblks = 0;
1490 * Count the number of extended attribute blocks
1492 if (XFS_IFORK_Q(ip) && ip->i_afp->if_nextents > 0 &&
1493 ip->i_afp->if_format != XFS_DINODE_FMT_LOCAL) {
1494 error = xfs_bmap_count_blocks(tp, ip, XFS_ATTR_FORK, &junk,
1499 if (XFS_IFORK_Q(tip) && tip->i_afp->if_nextents > 0 &&
1500 tip->i_afp->if_format != XFS_DINODE_FMT_LOCAL) {
1501 error = xfs_bmap_count_blocks(tp, tip, XFS_ATTR_FORK, &junk,
1508 * Btree format (v3) inodes have the inode number stamped in the bmbt
1509 * block headers. We can't start changing the bmbt blocks until the
1510 * inode owner change is logged so recovery does the right thing in the
1511 * event of a crash. Set the owner change log flags now and leave the
1512 * bmbt scan as the last step.
1514 if (xfs_has_v3inodes(ip->i_mount)) {
1515 if (ip->i_df.if_format == XFS_DINODE_FMT_BTREE)
1516 (*target_log_flags) |= XFS_ILOG_DOWNER;
1517 if (tip->i_df.if_format == XFS_DINODE_FMT_BTREE)
1518 (*src_log_flags) |= XFS_ILOG_DOWNER;
1522 * Swap the data forks of the inodes
1524 swap(ip->i_df, tip->i_df);
1527 * Fix the on-disk inode values
1529 tmp = (uint64_t)ip->i_nblocks;
1530 ip->i_nblocks = tip->i_nblocks - taforkblks + aforkblks;
1531 tip->i_nblocks = tmp + taforkblks - aforkblks;
1534 * The extents in the source inode could still contain speculative
1535 * preallocation beyond EOF (e.g. the file is open but not modified
1536 * while defrag is in progress). In that case, we need to copy over the
1537 * number of delalloc blocks the data fork in the source inode is
1538 * tracking beyond EOF so that when the fork is truncated away when the
1539 * temporary inode is unlinked we don't underrun the i_delayed_blks
1540 * counter on that inode.
1542 ASSERT(tip->i_delayed_blks == 0);
1543 tip->i_delayed_blks = ip->i_delayed_blks;
1544 ip->i_delayed_blks = 0;
1546 switch (ip->i_df.if_format) {
1547 case XFS_DINODE_FMT_EXTENTS:
1548 (*src_log_flags) |= XFS_ILOG_DEXT;
1550 case XFS_DINODE_FMT_BTREE:
1551 ASSERT(!xfs_has_v3inodes(ip->i_mount) ||
1552 (*src_log_flags & XFS_ILOG_DOWNER));
1553 (*src_log_flags) |= XFS_ILOG_DBROOT;
1557 switch (tip->i_df.if_format) {
1558 case XFS_DINODE_FMT_EXTENTS:
1559 (*target_log_flags) |= XFS_ILOG_DEXT;
1561 case XFS_DINODE_FMT_BTREE:
1562 (*target_log_flags) |= XFS_ILOG_DBROOT;
1563 ASSERT(!xfs_has_v3inodes(ip->i_mount) ||
1564 (*target_log_flags & XFS_ILOG_DOWNER));
1572 * Fix up the owners of the bmbt blocks to refer to the current inode. The
1573 * change owner scan attempts to order all modified buffers in the current
1574 * transaction. In the event of ordered buffer failure, the offending buffer is
1575 * physically logged as a fallback and the scan returns -EAGAIN. We must roll
1576 * the transaction in this case to replenish the fallback log reservation and
1577 * restart the scan. This process repeats until the scan completes.
1580 xfs_swap_change_owner(
1581 struct xfs_trans **tpp,
1582 struct xfs_inode *ip,
1583 struct xfs_inode *tmpip)
1586 struct xfs_trans *tp = *tpp;
1589 error = xfs_bmbt_change_owner(tp, ip, XFS_DATA_FORK, ip->i_ino,
1591 /* success or fatal error */
1592 if (error != -EAGAIN)
1595 error = xfs_trans_roll(tpp);
1601 * Redirty both inodes so they can relog and keep the log tail
1604 xfs_trans_ijoin(tp, ip, 0);
1605 xfs_trans_ijoin(tp, tmpip, 0);
1606 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
1607 xfs_trans_log_inode(tp, tmpip, XFS_ILOG_CORE);
1615 struct xfs_inode *ip, /* target inode */
1616 struct xfs_inode *tip, /* tmp inode */
1617 struct xfs_swapext *sxp)
1619 struct xfs_mount *mp = ip->i_mount;
1620 struct xfs_trans *tp;
1621 struct xfs_bstat *sbp = &sxp->sx_stat;
1622 int src_log_flags, target_log_flags;
1626 unsigned int flags = 0;
1629 * Lock the inodes against other IO, page faults and truncate to
1630 * begin with. Then we can ensure the inodes are flushed and have no
1631 * page cache safely. Once we have done this we can take the ilocks and
1632 * do the rest of the checks.
1634 lock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1635 filemap_invalidate_lock_two(VFS_I(ip)->i_mapping,
1636 VFS_I(tip)->i_mapping);
1638 /* Verify that both files have the same format */
1639 if ((VFS_I(ip)->i_mode & S_IFMT) != (VFS_I(tip)->i_mode & S_IFMT)) {
1644 /* Verify both files are either real-time or non-realtime */
1645 if (XFS_IS_REALTIME_INODE(ip) != XFS_IS_REALTIME_INODE(tip)) {
1650 error = xfs_qm_dqattach(ip);
1654 error = xfs_qm_dqattach(tip);
1658 error = xfs_swap_extent_flush(ip);
1661 error = xfs_swap_extent_flush(tip);
1665 if (xfs_inode_has_cow_data(tip)) {
1666 error = xfs_reflink_cancel_cow_range(tip, 0, NULLFILEOFF, true);
1672 * Extent "swapping" with rmap requires a permanent reservation and
1673 * a block reservation because it's really just a remap operation
1674 * performed with log redo items!
1676 if (xfs_has_rmapbt(mp)) {
1677 int w = XFS_DATA_FORK;
1678 uint32_t ipnext = ip->i_df.if_nextents;
1679 uint32_t tipnext = tip->i_df.if_nextents;
1682 * Conceptually this shouldn't affect the shape of either bmbt,
1683 * but since we atomically move extents one by one, we reserve
1684 * enough space to rebuild both trees.
1686 resblks = XFS_SWAP_RMAP_SPACE_RES(mp, ipnext, w);
1687 resblks += XFS_SWAP_RMAP_SPACE_RES(mp, tipnext, w);
1690 * If either inode straddles a bmapbt block allocation boundary,
1691 * the rmapbt algorithm triggers repeated allocs and frees as
1692 * extents are remapped. This can exhaust the block reservation
1693 * prematurely and cause shutdown. Return freed blocks to the
1694 * transaction reservation to counter this behavior.
1696 flags |= XFS_TRANS_RES_FDBLKS;
1698 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0, flags,
1704 * Lock and join the inodes to the tansaction so that transaction commit
1705 * or cancel will unlock the inodes from this point onwards.
1707 xfs_lock_two_inodes(ip, XFS_ILOCK_EXCL, tip, XFS_ILOCK_EXCL);
1708 xfs_trans_ijoin(tp, ip, 0);
1709 xfs_trans_ijoin(tp, tip, 0);
1712 /* Verify all data are being swapped */
1713 if (sxp->sx_offset != 0 ||
1714 sxp->sx_length != ip->i_disk_size ||
1715 sxp->sx_length != tip->i_disk_size) {
1717 goto out_trans_cancel;
1720 trace_xfs_swap_extent_before(ip, 0);
1721 trace_xfs_swap_extent_before(tip, 1);
1723 /* check inode formats now that data is flushed */
1724 error = xfs_swap_extents_check_format(ip, tip);
1727 "%s: inode 0x%llx format is incompatible for exchanging.",
1728 __func__, ip->i_ino);
1729 goto out_trans_cancel;
1733 * Compare the current change & modify times with that
1734 * passed in. If they differ, we abort this swap.
1735 * This is the mechanism used to ensure the calling
1736 * process that the file was not changed out from
1739 if ((sbp->bs_ctime.tv_sec != VFS_I(ip)->i_ctime.tv_sec) ||
1740 (sbp->bs_ctime.tv_nsec != VFS_I(ip)->i_ctime.tv_nsec) ||
1741 (sbp->bs_mtime.tv_sec != VFS_I(ip)->i_mtime.tv_sec) ||
1742 (sbp->bs_mtime.tv_nsec != VFS_I(ip)->i_mtime.tv_nsec)) {
1744 goto out_trans_cancel;
1748 * Note the trickiness in setting the log flags - we set the owner log
1749 * flag on the opposite inode (i.e. the inode we are setting the new
1750 * owner to be) because once we swap the forks and log that, log
1751 * recovery is going to see the fork as owned by the swapped inode,
1752 * not the pre-swapped inodes.
1754 src_log_flags = XFS_ILOG_CORE;
1755 target_log_flags = XFS_ILOG_CORE;
1757 if (xfs_has_rmapbt(mp))
1758 error = xfs_swap_extent_rmap(&tp, ip, tip);
1760 error = xfs_swap_extent_forks(tp, ip, tip, &src_log_flags,
1763 goto out_trans_cancel;
1765 /* Do we have to swap reflink flags? */
1766 if ((ip->i_diflags2 & XFS_DIFLAG2_REFLINK) ^
1767 (tip->i_diflags2 & XFS_DIFLAG2_REFLINK)) {
1768 f = ip->i_diflags2 & XFS_DIFLAG2_REFLINK;
1769 ip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
1770 ip->i_diflags2 |= tip->i_diflags2 & XFS_DIFLAG2_REFLINK;
1771 tip->i_diflags2 &= ~XFS_DIFLAG2_REFLINK;
1772 tip->i_diflags2 |= f & XFS_DIFLAG2_REFLINK;
1775 /* Swap the cow forks. */
1776 if (xfs_has_reflink(mp)) {
1777 ASSERT(!ip->i_cowfp ||
1778 ip->i_cowfp->if_format == XFS_DINODE_FMT_EXTENTS);
1779 ASSERT(!tip->i_cowfp ||
1780 tip->i_cowfp->if_format == XFS_DINODE_FMT_EXTENTS);
1782 swap(ip->i_cowfp, tip->i_cowfp);
1784 if (ip->i_cowfp && ip->i_cowfp->if_bytes)
1785 xfs_inode_set_cowblocks_tag(ip);
1787 xfs_inode_clear_cowblocks_tag(ip);
1788 if (tip->i_cowfp && tip->i_cowfp->if_bytes)
1789 xfs_inode_set_cowblocks_tag(tip);
1791 xfs_inode_clear_cowblocks_tag(tip);
1794 xfs_trans_log_inode(tp, ip, src_log_flags);
1795 xfs_trans_log_inode(tp, tip, target_log_flags);
1798 * The extent forks have been swapped, but crc=1,rmapbt=0 filesystems
1799 * have inode number owner values in the bmbt blocks that still refer to
1800 * the old inode. Scan each bmbt to fix up the owner values with the
1801 * inode number of the current inode.
1803 if (src_log_flags & XFS_ILOG_DOWNER) {
1804 error = xfs_swap_change_owner(&tp, ip, tip);
1806 goto out_trans_cancel;
1808 if (target_log_flags & XFS_ILOG_DOWNER) {
1809 error = xfs_swap_change_owner(&tp, tip, ip);
1811 goto out_trans_cancel;
1815 * If this is a synchronous mount, make sure that the
1816 * transaction goes to disk before returning to the user.
1818 if (xfs_has_wsync(mp))
1819 xfs_trans_set_sync(tp);
1821 error = xfs_trans_commit(tp);
1823 trace_xfs_swap_extent_after(ip, 0);
1824 trace_xfs_swap_extent_after(tip, 1);
1827 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1828 xfs_iunlock(tip, XFS_ILOCK_EXCL);
1830 filemap_invalidate_unlock_two(VFS_I(ip)->i_mapping,
1831 VFS_I(tip)->i_mapping);
1832 unlock_two_nondirectories(VFS_I(ip), VFS_I(tip));
1836 xfs_trans_cancel(tp);
1837 goto out_unlock_ilock;