1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * Copyright (c) 2016-2018 Christoph Hellwig.
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_inode.h"
15 #include "xfs_btree.h"
16 #include "xfs_bmap_btree.h"
18 #include "xfs_bmap_util.h"
19 #include "xfs_errortag.h"
20 #include "xfs_error.h"
21 #include "xfs_trans.h"
22 #include "xfs_trans_space.h"
23 #include "xfs_inode_item.h"
24 #include "xfs_iomap.h"
25 #include "xfs_trace.h"
26 #include "xfs_quota.h"
27 #include "xfs_dquot_item.h"
28 #include "xfs_dquot.h"
29 #include "xfs_reflink.h"
31 #define XFS_ALLOC_ALIGN(mp, off) \
32 (((off) >> mp->m_allocsize_log) << mp->m_allocsize_log)
35 xfs_alert_fsblock_zero(
37 xfs_bmbt_irec_t *imap)
39 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
40 "Access to block zero in inode %llu "
41 "start_block: %llx start_off: %llx "
42 "blkcnt: %llx extent-state: %x",
43 (unsigned long long)ip->i_ino,
44 (unsigned long long)imap->br_startblock,
45 (unsigned long long)imap->br_startoff,
46 (unsigned long long)imap->br_blockcount,
55 struct xfs_bmbt_irec *imap,
56 unsigned int mapping_flags,
59 struct xfs_mount *mp = ip->i_mount;
60 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
62 if (unlikely(!xfs_valid_startblock(ip, imap->br_startblock)))
63 return xfs_alert_fsblock_zero(ip, imap);
65 if (imap->br_startblock == HOLESTARTBLOCK) {
66 iomap->addr = IOMAP_NULL_ADDR;
67 iomap->type = IOMAP_HOLE;
68 } else if (imap->br_startblock == DELAYSTARTBLOCK ||
69 isnullstartblock(imap->br_startblock)) {
70 iomap->addr = IOMAP_NULL_ADDR;
71 iomap->type = IOMAP_DELALLOC;
73 iomap->addr = BBTOB(xfs_fsb_to_db(ip, imap->br_startblock));
74 if (mapping_flags & IOMAP_DAX)
75 iomap->addr += target->bt_dax_part_off;
77 if (imap->br_state == XFS_EXT_UNWRITTEN)
78 iomap->type = IOMAP_UNWRITTEN;
80 iomap->type = IOMAP_MAPPED;
83 iomap->offset = XFS_FSB_TO_B(mp, imap->br_startoff);
84 iomap->length = XFS_FSB_TO_B(mp, imap->br_blockcount);
85 if (mapping_flags & IOMAP_DAX)
86 iomap->dax_dev = target->bt_daxdev;
88 iomap->bdev = target->bt_bdev;
89 iomap->flags = iomap_flags;
91 if (xfs_ipincount(ip) &&
92 (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
93 iomap->flags |= IOMAP_F_DIRTY;
101 xfs_fileoff_t offset_fsb,
102 xfs_fileoff_t end_fsb)
104 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
106 iomap->addr = IOMAP_NULL_ADDR;
107 iomap->type = IOMAP_HOLE;
108 iomap->offset = XFS_FSB_TO_B(ip->i_mount, offset_fsb);
109 iomap->length = XFS_FSB_TO_B(ip->i_mount, end_fsb - offset_fsb);
110 iomap->bdev = target->bt_bdev;
111 iomap->dax_dev = target->bt_daxdev;
114 static inline xfs_fileoff_t
116 struct xfs_mount *mp,
120 ASSERT(offset <= mp->m_super->s_maxbytes);
121 return min(XFS_B_TO_FSB(mp, offset + count),
122 XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes));
127 struct xfs_inode *ip)
129 struct xfs_mount *mp = ip->i_mount;
130 xfs_extlen_t align = 0;
132 if (!XFS_IS_REALTIME_INODE(ip)) {
134 * Round up the allocation request to a stripe unit
135 * (m_dalign) boundary if the file size is >= stripe unit
136 * size, and we are allocating past the allocation eof.
138 * If mounted with the "-o swalloc" option the alignment is
139 * increased from the strip unit size to the stripe width.
141 if (mp->m_swidth && xfs_has_swalloc(mp))
142 align = mp->m_swidth;
143 else if (mp->m_dalign)
144 align = mp->m_dalign;
146 if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
154 * Check if last_fsb is outside the last extent, and if so grow it to the next
155 * stripe unit boundary.
158 xfs_iomap_eof_align_last_fsb(
159 struct xfs_inode *ip,
160 xfs_fileoff_t end_fsb)
162 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
163 xfs_extlen_t extsz = xfs_get_extsz_hint(ip);
164 xfs_extlen_t align = xfs_eof_alignment(ip);
165 struct xfs_bmbt_irec irec;
166 struct xfs_iext_cursor icur;
168 ASSERT(!xfs_need_iread_extents(ifp));
171 * Always round up the allocation request to the extent hint boundary.
175 align = roundup_64(align, extsz);
181 xfs_fileoff_t aligned_end_fsb = roundup_64(end_fsb, align);
183 xfs_iext_last(ifp, &icur);
184 if (!xfs_iext_get_extent(ifp, &icur, &irec) ||
185 aligned_end_fsb >= irec.br_startoff + irec.br_blockcount)
186 return aligned_end_fsb;
193 xfs_iomap_write_direct(
194 struct xfs_inode *ip,
195 xfs_fileoff_t offset_fsb,
196 xfs_fileoff_t count_fsb,
198 struct xfs_bmbt_irec *imap)
200 struct xfs_mount *mp = ip->i_mount;
201 struct xfs_trans *tp;
202 xfs_filblks_t resaligned;
204 unsigned int dblocks, rblocks;
207 int bmapi_flags = XFS_BMAPI_PREALLOC;
208 int nr_exts = XFS_IEXT_ADD_NOSPLIT_CNT;
210 ASSERT(count_fsb > 0);
212 resaligned = xfs_aligned_fsb_count(offset_fsb, count_fsb,
213 xfs_get_extsz_hint(ip));
214 if (unlikely(XFS_IS_REALTIME_INODE(ip))) {
215 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
216 rblocks = resaligned;
218 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
222 error = xfs_qm_dqattach(ip);
227 * For DAX, we do not allocate unwritten extents, but instead we zero
228 * the block before we commit the transaction. Ideally we'd like to do
229 * this outside the transaction context, but if we commit and then crash
230 * we may not have zeroed the blocks and this will be exposed on
231 * recovery of the allocation. Hence we must zero before commit.
233 * Further, if we are mapping unwritten extents here, we need to zero
234 * and convert them to written so that we don't need an unwritten extent
235 * callback for DAX. This also means that we need to be able to dip into
236 * the reserve block pool for bmbt block allocation if there is no space
237 * left but we need to do unwritten extent conversion.
239 if (flags & IOMAP_DAX) {
240 bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO;
241 if (imap->br_state == XFS_EXT_UNWRITTEN) {
243 nr_exts = XFS_IEXT_WRITE_UNWRITTEN_CNT;
244 dblocks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
248 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, dblocks,
249 rblocks, force, &tp);
253 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK, nr_exts);
255 goto out_trans_cancel;
258 * From this point onwards we overwrite the imap pointer that the
262 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, bmapi_flags, 0,
265 goto out_trans_cancel;
268 * Complete the transaction
270 error = xfs_trans_commit(tp);
275 * Copy any maps to caller's array and return any error.
282 if (unlikely(!xfs_valid_startblock(ip, imap->br_startblock)))
283 error = xfs_alert_fsblock_zero(ip, imap);
286 xfs_iunlock(ip, XFS_ILOCK_EXCL);
290 xfs_trans_cancel(tp);
295 xfs_quota_need_throttle(
296 struct xfs_inode *ip,
298 xfs_fsblock_t alloc_blocks)
300 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
302 if (!dq || !xfs_this_quota_on(ip->i_mount, type))
305 /* no hi watermark, no throttle */
306 if (!dq->q_prealloc_hi_wmark)
309 /* under the lo watermark, no throttle */
310 if (dq->q_blk.reserved + alloc_blocks < dq->q_prealloc_lo_wmark)
317 xfs_quota_calc_throttle(
318 struct xfs_inode *ip,
320 xfs_fsblock_t *qblocks,
324 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
328 /* no dq, or over hi wmark, squash the prealloc completely */
329 if (!dq || dq->q_blk.reserved >= dq->q_prealloc_hi_wmark) {
335 freesp = dq->q_prealloc_hi_wmark - dq->q_blk.reserved;
336 if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
338 if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
340 if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
344 if (freesp < *qfreesp)
347 /* only overwrite the throttle values if we are more aggressive */
348 if ((freesp >> shift) < (*qblocks >> *qshift)) {
355 * If we don't have a user specified preallocation size, dynamically increase
356 * the preallocation size as the size of the file grows. Cap the maximum size
357 * at a single extent or less if the filesystem is near full. The closer the
358 * filesystem is to being full, the smaller the maximum preallocation.
361 xfs_iomap_prealloc_size(
362 struct xfs_inode *ip,
366 struct xfs_iext_cursor *icur)
368 struct xfs_iext_cursor ncur = *icur;
369 struct xfs_bmbt_irec prev, got;
370 struct xfs_mount *mp = ip->i_mount;
371 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
372 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
374 xfs_fsblock_t qblocks;
375 xfs_fsblock_t alloc_blocks = 0;
381 * As an exception we don't do any preallocation at all if the file is
382 * smaller than the minimum preallocation and we are using the default
383 * dynamic preallocation scheme, as it is likely this is the only write
384 * to the file that is going to be done.
386 if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_allocsize_blocks))
390 * Use the minimum preallocation size for small files or if we are
391 * writing right after a hole.
393 if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign) ||
394 !xfs_iext_prev_extent(ifp, &ncur, &prev) ||
395 prev.br_startoff + prev.br_blockcount < offset_fsb)
396 return mp->m_allocsize_blocks;
399 * Take the size of the preceding data extents as the basis for the
400 * preallocation size. Note that we don't care if the previous extents
401 * are written or not.
403 plen = prev.br_blockcount;
404 while (xfs_iext_prev_extent(ifp, &ncur, &got)) {
405 if (plen > MAXEXTLEN / 2 ||
406 isnullstartblock(got.br_startblock) ||
407 got.br_startoff + got.br_blockcount != prev.br_startoff ||
408 got.br_startblock + got.br_blockcount != prev.br_startblock)
410 plen += got.br_blockcount;
415 * If the size of the extents is greater than half the maximum extent
416 * length, then use the current offset as the basis. This ensures that
417 * for large files the preallocation size always extends to MAXEXTLEN
418 * rather than falling short due to things like stripe unit/width
419 * alignment of real extents.
421 alloc_blocks = plen * 2;
422 if (alloc_blocks > MAXEXTLEN)
423 alloc_blocks = XFS_B_TO_FSB(mp, offset);
424 qblocks = alloc_blocks;
427 * MAXEXTLEN is not a power of two value but we round the prealloc down
428 * to the nearest power of two value after throttling. To prevent the
429 * round down from unconditionally reducing the maximum supported
430 * prealloc size, we round up first, apply appropriate throttling,
431 * round down and cap the value to MAXEXTLEN.
433 alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN),
436 freesp = percpu_counter_read_positive(&mp->m_fdblocks);
437 if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
439 if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
441 if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
443 if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
445 if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
450 * Check each quota to cap the prealloc size, provide a shift value to
451 * throttle with and adjust amount of available space.
453 if (xfs_quota_need_throttle(ip, XFS_DQTYPE_USER, alloc_blocks))
454 xfs_quota_calc_throttle(ip, XFS_DQTYPE_USER, &qblocks, &qshift,
456 if (xfs_quota_need_throttle(ip, XFS_DQTYPE_GROUP, alloc_blocks))
457 xfs_quota_calc_throttle(ip, XFS_DQTYPE_GROUP, &qblocks, &qshift,
459 if (xfs_quota_need_throttle(ip, XFS_DQTYPE_PROJ, alloc_blocks))
460 xfs_quota_calc_throttle(ip, XFS_DQTYPE_PROJ, &qblocks, &qshift,
464 * The final prealloc size is set to the minimum of free space available
465 * in each of the quotas and the overall filesystem.
467 * The shift throttle value is set to the maximum value as determined by
468 * the global low free space values and per-quota low free space values.
470 alloc_blocks = min(alloc_blocks, qblocks);
471 shift = max(shift, qshift);
474 alloc_blocks >>= shift;
476 * rounddown_pow_of_two() returns an undefined result if we pass in
480 alloc_blocks = rounddown_pow_of_two(alloc_blocks);
481 if (alloc_blocks > MAXEXTLEN)
482 alloc_blocks = MAXEXTLEN;
485 * If we are still trying to allocate more space than is
486 * available, squash the prealloc hard. This can happen if we
487 * have a large file on a small filesystem and the above
488 * lowspace thresholds are smaller than MAXEXTLEN.
490 while (alloc_blocks && alloc_blocks >= freesp)
492 if (alloc_blocks < mp->m_allocsize_blocks)
493 alloc_blocks = mp->m_allocsize_blocks;
494 trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
495 mp->m_allocsize_blocks);
500 xfs_iomap_write_unwritten(
506 xfs_mount_t *mp = ip->i_mount;
507 xfs_fileoff_t offset_fsb;
508 xfs_filblks_t count_fsb;
509 xfs_filblks_t numblks_fsb;
512 xfs_bmbt_irec_t imap;
513 struct inode *inode = VFS_I(ip);
518 trace_xfs_unwritten_convert(ip, offset, count);
520 offset_fsb = XFS_B_TO_FSBT(mp, offset);
521 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
522 count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
525 * Reserve enough blocks in this transaction for two complete extent
526 * btree splits. We may be converting the middle part of an unwritten
527 * extent and in this case we will insert two new extents in the btree
528 * each of which could cause a full split.
530 * This reservation amount will be used in the first call to
531 * xfs_bmbt_split() to select an AG with enough space to satisfy the
532 * rest of the operation.
534 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
536 /* Attach dquots so that bmbt splits are accounted correctly. */
537 error = xfs_qm_dqattach(ip);
543 * Set up a transaction to convert the range of extents
544 * from unwritten to real. Do allocations in a loop until
545 * we have covered the range passed in.
547 * Note that we can't risk to recursing back into the filesystem
548 * here as we might be asked to write out the same inode that we
549 * complete here and might deadlock on the iolock.
551 error = xfs_trans_alloc_inode(ip, &M_RES(mp)->tr_write, resblks,
556 error = xfs_iext_count_may_overflow(ip, XFS_DATA_FORK,
557 XFS_IEXT_WRITE_UNWRITTEN_CNT);
559 goto error_on_bmapi_transaction;
562 * Modify the unwritten extent state of the buffer.
565 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
566 XFS_BMAPI_CONVERT, resblks, &imap,
569 goto error_on_bmapi_transaction;
572 * Log the updated inode size as we go. We have to be careful
573 * to only log it up to the actual write offset if it is
574 * halfway into a block.
576 i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
577 if (i_size > offset + count)
578 i_size = offset + count;
579 if (update_isize && i_size > i_size_read(inode))
580 i_size_write(inode, i_size);
581 i_size = xfs_new_eof(ip, i_size);
583 ip->i_disk_size = i_size;
584 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
587 error = xfs_trans_commit(tp);
588 xfs_iunlock(ip, XFS_ILOCK_EXCL);
592 if (unlikely(!xfs_valid_startblock(ip, imap.br_startblock)))
593 return xfs_alert_fsblock_zero(ip, &imap);
595 if ((numblks_fsb = imap.br_blockcount) == 0) {
597 * The numblks_fsb value should always get
598 * smaller, otherwise the loop is stuck.
600 ASSERT(imap.br_blockcount);
603 offset_fsb += numblks_fsb;
604 count_fsb -= numblks_fsb;
605 } while (count_fsb > 0);
609 error_on_bmapi_transaction:
610 xfs_trans_cancel(tp);
611 xfs_iunlock(ip, XFS_ILOCK_EXCL);
619 struct xfs_bmbt_irec *imap,
622 /* don't allocate blocks when just zeroing */
623 if (flags & IOMAP_ZERO)
626 imap->br_startblock == HOLESTARTBLOCK ||
627 imap->br_startblock == DELAYSTARTBLOCK)
629 /* we convert unwritten extents before copying the data for DAX */
630 if ((flags & IOMAP_DAX) && imap->br_state == XFS_EXT_UNWRITTEN)
637 struct xfs_inode *ip,
639 struct xfs_bmbt_irec *imap,
642 if (!xfs_is_cow_inode(ip))
645 /* when zeroing we don't have to COW holes or unwritten extents */
646 if (flags & IOMAP_ZERO) {
648 imap->br_startblock == HOLESTARTBLOCK ||
649 imap->br_state == XFS_EXT_UNWRITTEN)
658 struct xfs_inode *ip,
662 unsigned mode = XFS_ILOCK_SHARED;
663 bool is_write = flags & (IOMAP_WRITE | IOMAP_ZERO);
666 * COW writes may allocate delalloc space or convert unwritten COW
667 * extents, so we need to make sure to take the lock exclusively here.
669 if (xfs_is_cow_inode(ip) && is_write)
670 mode = XFS_ILOCK_EXCL;
673 * Extents not yet cached requires exclusive access, don't block. This
674 * is an opencoded xfs_ilock_data_map_shared() call but with
675 * non-blocking behaviour.
677 if (xfs_need_iread_extents(&ip->i_df)) {
678 if (flags & IOMAP_NOWAIT)
680 mode = XFS_ILOCK_EXCL;
684 if (flags & IOMAP_NOWAIT) {
685 if (!xfs_ilock_nowait(ip, mode))
692 * The reflink iflag could have changed since the earlier unlocked
693 * check, so if we got ILOCK_SHARED for a write and but we're now a
694 * reflink inode we have to switch to ILOCK_EXCL and relock.
696 if (mode == XFS_ILOCK_SHARED && is_write && xfs_is_cow_inode(ip)) {
697 xfs_iunlock(ip, mode);
698 mode = XFS_ILOCK_EXCL;
707 * Check that the imap we are going to return to the caller spans the entire
708 * range that the caller requested for the IO.
712 struct xfs_bmbt_irec *imap,
713 xfs_fileoff_t offset_fsb,
714 xfs_fileoff_t end_fsb)
716 if (imap->br_startoff > offset_fsb)
718 if (imap->br_startoff + imap->br_blockcount < end_fsb)
724 xfs_direct_write_iomap_begin(
730 struct iomap *srcmap)
732 struct xfs_inode *ip = XFS_I(inode);
733 struct xfs_mount *mp = ip->i_mount;
734 struct xfs_bmbt_irec imap, cmap;
735 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
736 xfs_fileoff_t end_fsb = xfs_iomap_end_fsb(mp, offset, length);
737 int nimaps = 1, error = 0;
742 ASSERT(flags & (IOMAP_WRITE | IOMAP_ZERO));
744 if (xfs_is_shutdown(mp))
748 * Writes that span EOF might trigger an IO size update on completion,
749 * so consider them to be dirty for the purposes of O_DSYNC even if
750 * there is no other metadata changes pending or have been made here.
752 if (offset + length > i_size_read(inode))
753 iomap_flags |= IOMAP_F_DIRTY;
755 error = xfs_ilock_for_iomap(ip, flags, &lockmode);
759 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
764 if (imap_needs_cow(ip, flags, &imap, nimaps)) {
766 if (flags & IOMAP_NOWAIT)
769 /* may drop and re-acquire the ilock */
770 error = xfs_reflink_allocate_cow(ip, &imap, &cmap, &shared,
771 &lockmode, flags & IOMAP_DIRECT);
776 end_fsb = imap.br_startoff + imap.br_blockcount;
777 length = XFS_FSB_TO_B(mp, end_fsb) - offset;
780 if (imap_needs_alloc(inode, flags, &imap, nimaps))
781 goto allocate_blocks;
784 * NOWAIT and OVERWRITE I/O needs to span the entire requested I/O with
785 * a single map so that we avoid partial IO failures due to the rest of
786 * the I/O range not covered by this map triggering an EAGAIN condition
787 * when it is subsequently mapped and aborting the I/O.
789 if (flags & (IOMAP_NOWAIT | IOMAP_OVERWRITE_ONLY)) {
791 if (!imap_spans_range(&imap, offset_fsb, end_fsb))
796 * For overwrite only I/O, we cannot convert unwritten extents without
797 * requiring sub-block zeroing. This can only be done under an
798 * exclusive IOLOCK, hence return -EAGAIN if this is not a written
799 * extent to tell the caller to try again.
801 if (flags & IOMAP_OVERWRITE_ONLY) {
803 if (imap.br_state != XFS_EXT_NORM &&
804 ((offset | length) & mp->m_blockmask))
808 xfs_iunlock(ip, lockmode);
809 trace_xfs_iomap_found(ip, offset, length, XFS_DATA_FORK, &imap);
810 return xfs_bmbt_to_iomap(ip, iomap, &imap, flags, iomap_flags);
814 if (flags & (IOMAP_NOWAIT | IOMAP_OVERWRITE_ONLY))
818 * We cap the maximum length we map to a sane size to keep the chunks
819 * of work done where somewhat symmetric with the work writeback does.
820 * This is a completely arbitrary number pulled out of thin air as a
821 * best guess for initial testing.
823 * Note that the values needs to be less than 32-bits wide until the
824 * lower level functions are updated.
826 length = min_t(loff_t, length, 1024 * PAGE_SIZE);
827 end_fsb = xfs_iomap_end_fsb(mp, offset, length);
829 if (offset + length > XFS_ISIZE(ip))
830 end_fsb = xfs_iomap_eof_align_last_fsb(ip, end_fsb);
831 else if (nimaps && imap.br_startblock == HOLESTARTBLOCK)
832 end_fsb = min(end_fsb, imap.br_startoff + imap.br_blockcount);
833 xfs_iunlock(ip, lockmode);
835 error = xfs_iomap_write_direct(ip, offset_fsb, end_fsb - offset_fsb,
840 trace_xfs_iomap_alloc(ip, offset, length, XFS_DATA_FORK, &imap);
841 return xfs_bmbt_to_iomap(ip, iomap, &imap, flags,
842 iomap_flags | IOMAP_F_NEW);
845 xfs_iunlock(ip, lockmode);
846 length = XFS_FSB_TO_B(mp, cmap.br_startoff + cmap.br_blockcount);
847 trace_xfs_iomap_found(ip, offset, length - offset, XFS_COW_FORK, &cmap);
848 if (imap.br_startblock != HOLESTARTBLOCK) {
849 error = xfs_bmbt_to_iomap(ip, srcmap, &imap, flags, 0);
853 return xfs_bmbt_to_iomap(ip, iomap, &cmap, flags, IOMAP_F_SHARED);
857 xfs_iunlock(ip, lockmode);
861 const struct iomap_ops xfs_direct_write_iomap_ops = {
862 .iomap_begin = xfs_direct_write_iomap_begin,
866 xfs_buffered_write_iomap_begin(
872 struct iomap *srcmap)
874 struct xfs_inode *ip = XFS_I(inode);
875 struct xfs_mount *mp = ip->i_mount;
876 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
877 xfs_fileoff_t end_fsb = xfs_iomap_end_fsb(mp, offset, count);
878 struct xfs_bmbt_irec imap, cmap;
879 struct xfs_iext_cursor icur, ccur;
880 xfs_fsblock_t prealloc_blocks = 0;
881 bool eof = false, cow_eof = false, shared = false;
882 int allocfork = XFS_DATA_FORK;
885 if (xfs_is_shutdown(mp))
888 /* we can't use delayed allocations when using extent size hints */
889 if (xfs_get_extsz_hint(ip))
890 return xfs_direct_write_iomap_begin(inode, offset, count,
891 flags, iomap, srcmap);
893 ASSERT(!XFS_IS_REALTIME_INODE(ip));
895 xfs_ilock(ip, XFS_ILOCK_EXCL);
897 if (XFS_IS_CORRUPT(mp, !xfs_ifork_has_extents(&ip->i_df)) ||
898 XFS_TEST_ERROR(false, mp, XFS_ERRTAG_BMAPIFORMAT)) {
899 error = -EFSCORRUPTED;
903 XFS_STATS_INC(mp, xs_blk_mapw);
905 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
910 * Search the data fork first to look up our source mapping. We
911 * always need the data fork map, as we have to return it to the
912 * iomap code so that the higher level write code can read data in to
913 * perform read-modify-write cycles for unaligned writes.
915 eof = !xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap);
917 imap.br_startoff = end_fsb; /* fake hole until the end */
919 /* We never need to allocate blocks for zeroing a hole. */
920 if ((flags & IOMAP_ZERO) && imap.br_startoff > offset_fsb) {
921 xfs_hole_to_iomap(ip, iomap, offset_fsb, imap.br_startoff);
926 * Search the COW fork extent list even if we did not find a data fork
927 * extent. This serves two purposes: first this implements the
928 * speculative preallocation using cowextsize, so that we also unshare
929 * block adjacent to shared blocks instead of just the shared blocks
930 * themselves. Second the lookup in the extent list is generally faster
931 * than going out to the shared extent tree.
933 if (xfs_is_cow_inode(ip)) {
935 ASSERT(!xfs_is_reflink_inode(ip));
936 xfs_ifork_init_cow(ip);
938 cow_eof = !xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb,
940 if (!cow_eof && cmap.br_startoff <= offset_fsb) {
941 trace_xfs_reflink_cow_found(ip, &cmap);
946 if (imap.br_startoff <= offset_fsb) {
948 * For reflink files we may need a delalloc reservation when
949 * overwriting shared extents. This includes zeroing of
950 * existing extents that contain data.
952 if (!xfs_is_cow_inode(ip) ||
953 ((flags & IOMAP_ZERO) && imap.br_state != XFS_EXT_NORM)) {
954 trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
959 xfs_trim_extent(&imap, offset_fsb, end_fsb - offset_fsb);
961 /* Trim the mapping to the nearest shared extent boundary. */
962 error = xfs_bmap_trim_cow(ip, &imap, &shared);
966 /* Not shared? Just report the (potentially capped) extent. */
968 trace_xfs_iomap_found(ip, offset, count, XFS_DATA_FORK,
974 * Fork all the shared blocks from our write offset until the
977 allocfork = XFS_COW_FORK;
978 end_fsb = imap.br_startoff + imap.br_blockcount;
981 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES
982 * pages to keep the chunks of work done where somewhat
983 * symmetric with the work writeback does. This is a completely
984 * arbitrary number pulled out of thin air.
986 * Note that the values needs to be less than 32-bits wide until
987 * the lower level functions are updated.
989 count = min_t(loff_t, count, 1024 * PAGE_SIZE);
990 end_fsb = xfs_iomap_end_fsb(mp, offset, count);
992 if (xfs_is_always_cow_inode(ip))
993 allocfork = XFS_COW_FORK;
996 error = xfs_qm_dqattach_locked(ip, false);
1000 if (eof && offset + count > XFS_ISIZE(ip)) {
1002 * Determine the initial size of the preallocation.
1003 * We clean up any extra preallocation when the file is closed.
1005 if (xfs_has_allocsize(mp))
1006 prealloc_blocks = mp->m_allocsize_blocks;
1008 prealloc_blocks = xfs_iomap_prealloc_size(ip, allocfork,
1009 offset, count, &icur);
1010 if (prealloc_blocks) {
1012 xfs_off_t end_offset;
1013 xfs_fileoff_t p_end_fsb;
1015 end_offset = XFS_ALLOC_ALIGN(mp, offset + count - 1);
1016 p_end_fsb = XFS_B_TO_FSBT(mp, end_offset) +
1019 align = xfs_eof_alignment(ip);
1021 p_end_fsb = roundup_64(p_end_fsb, align);
1023 p_end_fsb = min(p_end_fsb,
1024 XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes));
1025 ASSERT(p_end_fsb > offset_fsb);
1026 prealloc_blocks = p_end_fsb - end_fsb;
1031 error = xfs_bmapi_reserve_delalloc(ip, allocfork, offset_fsb,
1032 end_fsb - offset_fsb, prealloc_blocks,
1033 allocfork == XFS_DATA_FORK ? &imap : &cmap,
1034 allocfork == XFS_DATA_FORK ? &icur : &ccur,
1035 allocfork == XFS_DATA_FORK ? eof : cow_eof);
1041 /* retry without any preallocation */
1042 trace_xfs_delalloc_enospc(ip, offset, count);
1043 if (prealloc_blocks) {
1044 prealloc_blocks = 0;
1052 if (allocfork == XFS_COW_FORK) {
1053 trace_xfs_iomap_alloc(ip, offset, count, allocfork, &cmap);
1058 * Flag newly allocated delalloc blocks with IOMAP_F_NEW so we punch
1059 * them out if the write happens to fail.
1061 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1062 trace_xfs_iomap_alloc(ip, offset, count, allocfork, &imap);
1063 return xfs_bmbt_to_iomap(ip, iomap, &imap, flags, IOMAP_F_NEW);
1066 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1067 return xfs_bmbt_to_iomap(ip, iomap, &imap, flags, 0);
1070 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1071 if (imap.br_startoff <= offset_fsb) {
1072 error = xfs_bmbt_to_iomap(ip, srcmap, &imap, flags, 0);
1075 return xfs_bmbt_to_iomap(ip, iomap, &cmap, flags,
1079 xfs_trim_extent(&cmap, offset_fsb, imap.br_startoff - offset_fsb);
1080 return xfs_bmbt_to_iomap(ip, iomap, &cmap, flags, 0);
1083 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1088 xfs_buffered_write_iomap_end(
1089 struct inode *inode,
1094 struct iomap *iomap)
1096 struct xfs_inode *ip = XFS_I(inode);
1097 struct xfs_mount *mp = ip->i_mount;
1098 xfs_fileoff_t start_fsb;
1099 xfs_fileoff_t end_fsb;
1102 if (iomap->type != IOMAP_DELALLOC)
1106 * Behave as if the write failed if drop writes is enabled. Set the NEW
1107 * flag to force delalloc cleanup.
1109 if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_DROP_WRITES)) {
1110 iomap->flags |= IOMAP_F_NEW;
1115 * start_fsb refers to the first unused block after a short write. If
1116 * nothing was written, round offset down to point at the first block in
1119 if (unlikely(!written))
1120 start_fsb = XFS_B_TO_FSBT(mp, offset);
1122 start_fsb = XFS_B_TO_FSB(mp, offset + written);
1123 end_fsb = XFS_B_TO_FSB(mp, offset + length);
1126 * Trim delalloc blocks if they were allocated by this write and we
1127 * didn't manage to write the whole range.
1129 * We don't need to care about racing delalloc as we hold i_mutex
1130 * across the reserve/allocate/unreserve calls. If there are delalloc
1131 * blocks in the range, they are ours.
1133 if ((iomap->flags & IOMAP_F_NEW) && start_fsb < end_fsb) {
1134 truncate_pagecache_range(VFS_I(ip), XFS_FSB_TO_B(mp, start_fsb),
1135 XFS_FSB_TO_B(mp, end_fsb) - 1);
1137 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1138 end_fsb - start_fsb);
1139 if (error && !xfs_is_shutdown(mp)) {
1140 xfs_alert(mp, "%s: unable to clean up ino %lld",
1141 __func__, ip->i_ino);
1149 const struct iomap_ops xfs_buffered_write_iomap_ops = {
1150 .iomap_begin = xfs_buffered_write_iomap_begin,
1151 .iomap_end = xfs_buffered_write_iomap_end,
1155 xfs_read_iomap_begin(
1156 struct inode *inode,
1160 struct iomap *iomap,
1161 struct iomap *srcmap)
1163 struct xfs_inode *ip = XFS_I(inode);
1164 struct xfs_mount *mp = ip->i_mount;
1165 struct xfs_bmbt_irec imap;
1166 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
1167 xfs_fileoff_t end_fsb = xfs_iomap_end_fsb(mp, offset, length);
1168 int nimaps = 1, error = 0;
1169 bool shared = false;
1172 ASSERT(!(flags & (IOMAP_WRITE | IOMAP_ZERO)));
1174 if (xfs_is_shutdown(mp))
1177 error = xfs_ilock_for_iomap(ip, flags, &lockmode);
1180 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1182 if (!error && (flags & IOMAP_REPORT))
1183 error = xfs_reflink_trim_around_shared(ip, &imap, &shared);
1184 xfs_iunlock(ip, lockmode);
1188 trace_xfs_iomap_found(ip, offset, length, XFS_DATA_FORK, &imap);
1189 return xfs_bmbt_to_iomap(ip, iomap, &imap, flags,
1190 shared ? IOMAP_F_SHARED : 0);
1193 const struct iomap_ops xfs_read_iomap_ops = {
1194 .iomap_begin = xfs_read_iomap_begin,
1198 xfs_seek_iomap_begin(
1199 struct inode *inode,
1203 struct iomap *iomap,
1204 struct iomap *srcmap)
1206 struct xfs_inode *ip = XFS_I(inode);
1207 struct xfs_mount *mp = ip->i_mount;
1208 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
1209 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + length);
1210 xfs_fileoff_t cow_fsb = NULLFILEOFF, data_fsb = NULLFILEOFF;
1211 struct xfs_iext_cursor icur;
1212 struct xfs_bmbt_irec imap, cmap;
1216 if (xfs_is_shutdown(mp))
1219 lockmode = xfs_ilock_data_map_shared(ip);
1220 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
1224 if (xfs_iext_lookup_extent(ip, &ip->i_df, offset_fsb, &icur, &imap)) {
1226 * If we found a data extent we are done.
1228 if (imap.br_startoff <= offset_fsb)
1230 data_fsb = imap.br_startoff;
1233 * Fake a hole until the end of the file.
1235 data_fsb = xfs_iomap_end_fsb(mp, offset, length);
1239 * If a COW fork extent covers the hole, report it - capped to the next
1242 if (xfs_inode_has_cow_data(ip) &&
1243 xfs_iext_lookup_extent(ip, ip->i_cowfp, offset_fsb, &icur, &cmap))
1244 cow_fsb = cmap.br_startoff;
1245 if (cow_fsb != NULLFILEOFF && cow_fsb <= offset_fsb) {
1246 if (data_fsb < cow_fsb + cmap.br_blockcount)
1247 end_fsb = min(end_fsb, data_fsb);
1248 xfs_trim_extent(&cmap, offset_fsb, end_fsb);
1249 error = xfs_bmbt_to_iomap(ip, iomap, &cmap, flags,
1252 * This is a COW extent, so we must probe the page cache
1253 * because there could be dirty page cache being backed
1256 iomap->type = IOMAP_UNWRITTEN;
1261 * Else report a hole, capped to the next found data or COW extent.
1263 if (cow_fsb != NULLFILEOFF && cow_fsb < data_fsb)
1264 imap.br_blockcount = cow_fsb - offset_fsb;
1266 imap.br_blockcount = data_fsb - offset_fsb;
1267 imap.br_startoff = offset_fsb;
1268 imap.br_startblock = HOLESTARTBLOCK;
1269 imap.br_state = XFS_EXT_NORM;
1271 xfs_trim_extent(&imap, offset_fsb, end_fsb);
1272 error = xfs_bmbt_to_iomap(ip, iomap, &imap, flags, 0);
1274 xfs_iunlock(ip, lockmode);
1278 const struct iomap_ops xfs_seek_iomap_ops = {
1279 .iomap_begin = xfs_seek_iomap_begin,
1283 xfs_xattr_iomap_begin(
1284 struct inode *inode,
1288 struct iomap *iomap,
1289 struct iomap *srcmap)
1291 struct xfs_inode *ip = XFS_I(inode);
1292 struct xfs_mount *mp = ip->i_mount;
1293 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
1294 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + length);
1295 struct xfs_bmbt_irec imap;
1296 int nimaps = 1, error = 0;
1299 if (xfs_is_shutdown(mp))
1302 lockmode = xfs_ilock_attr_map_shared(ip);
1304 /* if there are no attribute fork or extents, return ENOENT */
1305 if (!XFS_IFORK_Q(ip) || !ip->i_afp->if_nextents) {
1310 ASSERT(ip->i_afp->if_format != XFS_DINODE_FMT_LOCAL);
1311 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1312 &nimaps, XFS_BMAPI_ATTRFORK);
1314 xfs_iunlock(ip, lockmode);
1319 return xfs_bmbt_to_iomap(ip, iomap, &imap, flags, 0);
1322 const struct iomap_ops xfs_xattr_iomap_ops = {
1323 .iomap_begin = xfs_xattr_iomap_begin,
1328 struct xfs_inode *ip,
1333 struct inode *inode = VFS_I(ip);
1336 return dax_zero_range(inode, pos, len, did_zero,
1337 &xfs_direct_write_iomap_ops);
1338 return iomap_zero_range(inode, pos, len, did_zero,
1339 &xfs_buffered_write_iomap_ops);
1344 struct xfs_inode *ip,
1348 struct inode *inode = VFS_I(ip);
1351 return dax_truncate_page(inode, pos, did_zero,
1352 &xfs_direct_write_iomap_ops);
1353 return iomap_truncate_page(inode, pos, did_zero,
1354 &xfs_buffered_write_iomap_ops);