1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * Copyright (c) 2016-2018 Christoph Hellwig.
7 #include <linux/iomap.h>
10 #include "xfs_shared.h"
11 #include "xfs_format.h"
12 #include "xfs_log_format.h"
13 #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_bmap_btree.h"
20 #include "xfs_bmap_util.h"
21 #include "xfs_errortag.h"
22 #include "xfs_error.h"
23 #include "xfs_trans.h"
24 #include "xfs_trans_space.h"
25 #include "xfs_inode_item.h"
26 #include "xfs_iomap.h"
27 #include "xfs_trace.h"
28 #include "xfs_icache.h"
29 #include "xfs_quota.h"
30 #include "xfs_dquot_item.h"
31 #include "xfs_dquot.h"
32 #include "xfs_reflink.h"
35 #define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \
42 struct xfs_bmbt_irec *imap)
44 struct xfs_mount *mp = ip->i_mount;
46 if (imap->br_startblock == HOLESTARTBLOCK) {
47 iomap->addr = IOMAP_NULL_ADDR;
48 iomap->type = IOMAP_HOLE;
49 } else if (imap->br_startblock == DELAYSTARTBLOCK) {
50 iomap->addr = IOMAP_NULL_ADDR;
51 iomap->type = IOMAP_DELALLOC;
53 iomap->addr = BBTOB(xfs_fsb_to_db(ip, imap->br_startblock));
54 if (imap->br_state == XFS_EXT_UNWRITTEN)
55 iomap->type = IOMAP_UNWRITTEN;
57 iomap->type = IOMAP_MAPPED;
59 iomap->offset = XFS_FSB_TO_B(mp, imap->br_startoff);
60 iomap->length = XFS_FSB_TO_B(mp, imap->br_blockcount);
61 iomap->bdev = xfs_find_bdev_for_inode(VFS_I(ip));
62 iomap->dax_dev = xfs_find_daxdev_for_inode(VFS_I(ip));
69 xfs_fileoff_t offset_fsb,
70 xfs_fileoff_t end_fsb)
72 iomap->addr = IOMAP_NULL_ADDR;
73 iomap->type = IOMAP_HOLE;
74 iomap->offset = XFS_FSB_TO_B(ip->i_mount, offset_fsb);
75 iomap->length = XFS_FSB_TO_B(ip->i_mount, end_fsb - offset_fsb);
76 iomap->bdev = xfs_find_bdev_for_inode(VFS_I(ip));
77 iomap->dax_dev = xfs_find_daxdev_for_inode(VFS_I(ip));
85 struct xfs_mount *mp = ip->i_mount;
86 xfs_extlen_t align = 0;
88 if (!XFS_IS_REALTIME_INODE(ip)) {
90 * Round up the allocation request to a stripe unit
91 * (m_dalign) boundary if the file size is >= stripe unit
92 * size, and we are allocating past the allocation eof.
94 * If mounted with the "-o swalloc" option the alignment is
95 * increased from the strip unit size to the stripe width.
97 if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
99 else if (mp->m_dalign)
100 align = mp->m_dalign;
102 if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
107 * Always round up the allocation request to an extent boundary
108 * (when file on a real-time subvolume or has di_extsize hint).
112 align = roundup_64(align, extsize);
121 xfs_iomap_eof_align_last_fsb(
122 struct xfs_inode *ip,
123 xfs_extlen_t extsize,
124 xfs_fileoff_t *last_fsb)
126 xfs_extlen_t align = xfs_eof_alignment(ip, extsize);
129 xfs_fileoff_t new_last_fsb = roundup_64(*last_fsb, align);
132 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
136 *last_fsb = new_last_fsb;
142 xfs_alert_fsblock_zero(
144 xfs_bmbt_irec_t *imap)
146 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
147 "Access to block zero in inode %llu "
148 "start_block: %llx start_off: %llx "
149 "blkcnt: %llx extent-state: %x",
150 (unsigned long long)ip->i_ino,
151 (unsigned long long)imap->br_startblock,
152 (unsigned long long)imap->br_startoff,
153 (unsigned long long)imap->br_blockcount,
155 return -EFSCORRUPTED;
159 xfs_iomap_write_direct(
163 xfs_bmbt_irec_t *imap,
166 xfs_mount_t *mp = ip->i_mount;
167 xfs_fileoff_t offset_fsb;
168 xfs_fileoff_t last_fsb;
169 xfs_filblks_t count_fsb, resaligned;
175 uint qblocks, resblks, resrtextents;
178 int bmapi_flags = XFS_BMAPI_PREALLOC;
181 rt = XFS_IS_REALTIME_INODE(ip);
182 extsz = xfs_get_extsz_hint(ip);
183 lockmode = XFS_ILOCK_SHARED; /* locked by caller */
185 ASSERT(xfs_isilocked(ip, lockmode));
187 offset_fsb = XFS_B_TO_FSBT(mp, offset);
188 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
189 if ((offset + count) > XFS_ISIZE(ip)) {
191 * Assert that the in-core extent list is present since this can
192 * call xfs_iread_extents() and we only have the ilock shared.
193 * This should be safe because the lock was held around a bmapi
194 * call in the caller and we only need it to access the in-core
197 ASSERT(XFS_IFORK_PTR(ip, XFS_DATA_FORK)->if_flags &
199 error = xfs_iomap_eof_align_last_fsb(ip, extsz, &last_fsb);
203 if (nmaps && (imap->br_startblock == HOLESTARTBLOCK))
204 last_fsb = min(last_fsb, (xfs_fileoff_t)
205 imap->br_blockcount +
208 count_fsb = last_fsb - offset_fsb;
209 ASSERT(count_fsb > 0);
210 resaligned = xfs_aligned_fsb_count(offset_fsb, count_fsb, extsz);
213 resrtextents = qblocks = resaligned;
214 resrtextents /= mp->m_sb.sb_rextsize;
215 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
216 quota_flag = XFS_QMOPT_RES_RTBLKS;
219 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
220 quota_flag = XFS_QMOPT_RES_REGBLKS;
224 * Drop the shared lock acquired by the caller, attach the dquot if
225 * necessary and move on to transaction setup.
227 xfs_iunlock(ip, lockmode);
228 error = xfs_qm_dqattach(ip);
233 * For DAX, we do not allocate unwritten extents, but instead we zero
234 * the block before we commit the transaction. Ideally we'd like to do
235 * this outside the transaction context, but if we commit and then crash
236 * we may not have zeroed the blocks and this will be exposed on
237 * recovery of the allocation. Hence we must zero before commit.
239 * Further, if we are mapping unwritten extents here, we need to zero
240 * and convert them to written so that we don't need an unwritten extent
241 * callback for DAX. This also means that we need to be able to dip into
242 * the reserve block pool for bmbt block allocation if there is no space
243 * left but we need to do unwritten extent conversion.
245 if (IS_DAX(VFS_I(ip))) {
246 bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO;
247 if (imap->br_state == XFS_EXT_UNWRITTEN) {
248 tflags |= XFS_TRANS_RESERVE;
249 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
252 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, resrtextents,
257 lockmode = XFS_ILOCK_EXCL;
258 xfs_ilock(ip, lockmode);
260 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
262 goto out_trans_cancel;
264 xfs_trans_ijoin(tp, ip, 0);
267 * From this point onwards we overwrite the imap pointer that the
271 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
272 bmapi_flags, resblks, imap, &nimaps);
277 * Complete the transaction
279 error = xfs_trans_commit(tp);
284 * Copy any maps to caller's array and return any error.
291 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
292 error = xfs_alert_fsblock_zero(ip, imap);
295 xfs_iunlock(ip, lockmode);
299 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
301 xfs_trans_cancel(tp);
306 xfs_quota_need_throttle(
307 struct xfs_inode *ip,
309 xfs_fsblock_t alloc_blocks)
311 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
313 if (!dq || !xfs_this_quota_on(ip->i_mount, type))
316 /* no hi watermark, no throttle */
317 if (!dq->q_prealloc_hi_wmark)
320 /* under the lo watermark, no throttle */
321 if (dq->q_res_bcount + alloc_blocks < dq->q_prealloc_lo_wmark)
328 xfs_quota_calc_throttle(
329 struct xfs_inode *ip,
331 xfs_fsblock_t *qblocks,
337 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
339 /* no dq, or over hi wmark, squash the prealloc completely */
340 if (!dq || dq->q_res_bcount >= dq->q_prealloc_hi_wmark) {
346 freesp = dq->q_prealloc_hi_wmark - dq->q_res_bcount;
347 if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
349 if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
351 if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
355 if (freesp < *qfreesp)
358 /* only overwrite the throttle values if we are more aggressive */
359 if ((freesp >> shift) < (*qblocks >> *qshift)) {
366 * If we are doing a write at the end of the file and there are no allocations
367 * past this one, then extend the allocation out to the file system's write
370 * If we don't have a user specified preallocation size, dynamically increase
371 * the preallocation size as the size of the file grows. Cap the maximum size
372 * at a single extent or less if the filesystem is near full. The closer the
373 * filesystem is to full, the smaller the maximum prealocation.
375 * As an exception we don't do any preallocation at all if the file is smaller
376 * than the minimum preallocation and we are using the default dynamic
377 * preallocation scheme, as it is likely this is the only write to the file that
378 * is going to be done.
380 * We clean up any extra space left over when the file is closed in
384 xfs_iomap_prealloc_size(
385 struct xfs_inode *ip,
388 struct xfs_iext_cursor *icur)
390 struct xfs_mount *mp = ip->i_mount;
391 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
392 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
393 struct xfs_bmbt_irec prev;
396 xfs_fsblock_t qblocks;
398 xfs_fsblock_t alloc_blocks = 0;
400 if (offset + count <= XFS_ISIZE(ip))
403 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) &&
404 (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_writeio_blocks)))
408 * If an explicit allocsize is set, the file is small, or we
409 * are writing behind a hole, then use the minimum prealloc:
411 if ((mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) ||
412 XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign) ||
413 !xfs_iext_peek_prev_extent(ifp, icur, &prev) ||
414 prev.br_startoff + prev.br_blockcount < offset_fsb)
415 return mp->m_writeio_blocks;
418 * Determine the initial size of the preallocation. We are beyond the
419 * current EOF here, but we need to take into account whether this is
420 * a sparse write or an extending write when determining the
421 * preallocation size. Hence we need to look up the extent that ends
422 * at the current write offset and use the result to determine the
423 * preallocation size.
425 * If the extent is a hole, then preallocation is essentially disabled.
426 * Otherwise we take the size of the preceding data extent as the basis
427 * for the preallocation size. If the size of the extent is greater than
428 * half the maximum extent length, then use the current offset as the
429 * basis. This ensures that for large files the preallocation size
430 * always extends to MAXEXTLEN rather than falling short due to things
431 * like stripe unit/width alignment of real extents.
433 if (prev.br_blockcount <= (MAXEXTLEN >> 1))
434 alloc_blocks = prev.br_blockcount << 1;
436 alloc_blocks = XFS_B_TO_FSB(mp, offset);
439 qblocks = alloc_blocks;
442 * MAXEXTLEN is not a power of two value but we round the prealloc down
443 * to the nearest power of two value after throttling. To prevent the
444 * round down from unconditionally reducing the maximum supported prealloc
445 * size, we round up first, apply appropriate throttling, round down and
446 * cap the value to MAXEXTLEN.
448 alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN),
451 freesp = percpu_counter_read_positive(&mp->m_fdblocks);
452 if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
454 if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
456 if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
458 if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
460 if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
465 * Check each quota to cap the prealloc size, provide a shift value to
466 * throttle with and adjust amount of available space.
468 if (xfs_quota_need_throttle(ip, XFS_DQ_USER, alloc_blocks))
469 xfs_quota_calc_throttle(ip, XFS_DQ_USER, &qblocks, &qshift,
471 if (xfs_quota_need_throttle(ip, XFS_DQ_GROUP, alloc_blocks))
472 xfs_quota_calc_throttle(ip, XFS_DQ_GROUP, &qblocks, &qshift,
474 if (xfs_quota_need_throttle(ip, XFS_DQ_PROJ, alloc_blocks))
475 xfs_quota_calc_throttle(ip, XFS_DQ_PROJ, &qblocks, &qshift,
479 * The final prealloc size is set to the minimum of free space available
480 * in each of the quotas and the overall filesystem.
482 * The shift throttle value is set to the maximum value as determined by
483 * the global low free space values and per-quota low free space values.
485 alloc_blocks = min(alloc_blocks, qblocks);
486 shift = max(shift, qshift);
489 alloc_blocks >>= shift;
491 * rounddown_pow_of_two() returns an undefined result if we pass in
495 alloc_blocks = rounddown_pow_of_two(alloc_blocks);
496 if (alloc_blocks > MAXEXTLEN)
497 alloc_blocks = MAXEXTLEN;
500 * If we are still trying to allocate more space than is
501 * available, squash the prealloc hard. This can happen if we
502 * have a large file on a small filesystem and the above
503 * lowspace thresholds are smaller than MAXEXTLEN.
505 while (alloc_blocks && alloc_blocks >= freesp)
508 if (alloc_blocks < mp->m_writeio_blocks)
509 alloc_blocks = mp->m_writeio_blocks;
510 trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
511 mp->m_writeio_blocks);
516 xfs_file_iomap_begin_delay(
523 struct xfs_inode *ip = XFS_I(inode);
524 struct xfs_mount *mp = ip->i_mount;
525 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
526 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
527 xfs_fileoff_t maxbytes_fsb =
528 XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
529 xfs_fileoff_t end_fsb;
530 int error = 0, eof = 0;
531 struct xfs_bmbt_irec got;
532 struct xfs_iext_cursor icur;
533 xfs_fsblock_t prealloc_blocks = 0;
535 ASSERT(!XFS_IS_REALTIME_INODE(ip));
536 ASSERT(!xfs_get_extsz_hint(ip));
538 xfs_ilock(ip, XFS_ILOCK_EXCL);
540 if (unlikely(XFS_TEST_ERROR(
541 (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
542 XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
543 mp, XFS_ERRTAG_BMAPIFORMAT))) {
544 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
545 error = -EFSCORRUPTED;
549 XFS_STATS_INC(mp, xs_blk_mapw);
551 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
552 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
557 end_fsb = min(XFS_B_TO_FSB(mp, offset + count), maxbytes_fsb);
559 eof = !xfs_iext_lookup_extent(ip, ifp, offset_fsb, &icur, &got);
561 got.br_startoff = end_fsb; /* fake hole until the end */
563 if (got.br_startoff <= offset_fsb) {
565 * For reflink files we may need a delalloc reservation when
566 * overwriting shared extents. This includes zeroing of
567 * existing extents that contain data.
569 if (xfs_is_reflink_inode(ip) &&
570 ((flags & IOMAP_WRITE) ||
571 got.br_state != XFS_EXT_UNWRITTEN)) {
572 xfs_trim_extent(&got, offset_fsb, end_fsb - offset_fsb);
573 error = xfs_reflink_reserve_cow(ip, &got);
578 trace_xfs_iomap_found(ip, offset, count, 0, &got);
582 if (flags & IOMAP_ZERO) {
583 xfs_hole_to_iomap(ip, iomap, offset_fsb, got.br_startoff);
587 error = xfs_qm_dqattach_locked(ip, false);
592 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES pages
593 * to keep the chunks of work done where somewhat symmetric with the
594 * work writeback does. This is a completely arbitrary number pulled
595 * out of thin air as a best guess for initial testing.
597 * Note that the values needs to be less than 32-bits wide until
598 * the lower level functions are updated.
600 count = min_t(loff_t, count, 1024 * PAGE_SIZE);
601 end_fsb = min(XFS_B_TO_FSB(mp, offset + count), maxbytes_fsb);
604 prealloc_blocks = xfs_iomap_prealloc_size(ip, offset, count,
606 if (prealloc_blocks) {
608 xfs_off_t end_offset;
609 xfs_fileoff_t p_end_fsb;
611 end_offset = XFS_WRITEIO_ALIGN(mp, offset + count - 1);
612 p_end_fsb = XFS_B_TO_FSBT(mp, end_offset) +
615 align = xfs_eof_alignment(ip, 0);
617 p_end_fsb = roundup_64(p_end_fsb, align);
619 p_end_fsb = min(p_end_fsb, maxbytes_fsb);
620 ASSERT(p_end_fsb > offset_fsb);
621 prealloc_blocks = p_end_fsb - end_fsb;
626 error = xfs_bmapi_reserve_delalloc(ip, XFS_DATA_FORK, offset_fsb,
627 end_fsb - offset_fsb, prealloc_blocks, &got, &icur,
634 /* retry without any preallocation */
635 trace_xfs_delalloc_enospc(ip, offset, count);
636 if (prealloc_blocks) {
646 * Flag newly allocated delalloc blocks with IOMAP_F_NEW so we punch
647 * them out if the write happens to fail.
649 iomap->flags |= IOMAP_F_NEW;
650 trace_xfs_iomap_alloc(ip, offset, count, 0, &got);
652 if (isnullstartblock(got.br_startblock))
653 got.br_startblock = DELAYSTARTBLOCK;
655 if (!got.br_startblock) {
656 error = xfs_alert_fsblock_zero(ip, &got);
661 xfs_bmbt_to_iomap(ip, iomap, &got);
664 xfs_iunlock(ip, XFS_ILOCK_EXCL);
669 * Pass in a delayed allocate extent, convert it to real extents;
670 * return to the caller the extent we create which maps on top of
671 * the originating callers request.
673 * Called without a lock on the inode.
675 * We no longer bother to look at the incoming map - all we have to
676 * guarantee is that whatever we allocate fills the required range.
679 xfs_iomap_write_allocate(
683 xfs_bmbt_irec_t *imap,
684 unsigned int *cow_seq)
686 xfs_mount_t *mp = ip->i_mount;
687 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, whichfork);
688 xfs_fileoff_t offset_fsb, last_block;
689 xfs_fileoff_t end_fsb, map_start_fsb;
690 xfs_filblks_t count_fsb;
694 int flags = XFS_BMAPI_DELALLOC;
697 if (whichfork == XFS_COW_FORK)
698 flags |= XFS_BMAPI_COWFORK | XFS_BMAPI_PREALLOC;
701 * Make sure that the dquots are there.
703 error = xfs_qm_dqattach(ip);
707 offset_fsb = XFS_B_TO_FSBT(mp, offset);
708 count_fsb = imap->br_blockcount;
709 map_start_fsb = imap->br_startoff;
711 XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
713 while (count_fsb != 0) {
715 * Set up a transaction with which to allocate the
716 * backing store for the file. Do allocations in a
717 * loop until we get some space in the range we are
718 * interested in. The other space that might be allocated
719 * is in the delayed allocation extent on which we sit
720 * but before our buffer starts.
723 while (nimaps == 0) {
724 nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
726 * We have already reserved space for the extent and any
727 * indirect blocks when creating the delalloc extent,
728 * there is no need to reserve space in this transaction
731 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0,
732 0, XFS_TRANS_RESERVE, &tp);
736 xfs_ilock(ip, XFS_ILOCK_EXCL);
737 xfs_trans_ijoin(tp, ip, 0);
740 * it is possible that the extents have changed since
741 * we did the read call as we dropped the ilock for a
742 * while. We have to be careful about truncates or hole
743 * punchs here - we are not allowed to allocate
744 * non-delalloc blocks here.
746 * The only protection against truncation is the pages
747 * for the range we are being asked to convert are
748 * locked and hence a truncate will block on them
751 * As a result, if we go beyond the range we really
752 * need and hit an delalloc extent boundary followed by
753 * a hole while we have excess blocks in the map, we
754 * will fill the hole incorrectly and overrun the
755 * transaction reservation.
757 * Using a single map prevents this as we are forced to
758 * check each map we look for overlap with the desired
759 * range and abort as soon as we find it. Also, given
760 * that we only return a single map, having one beyond
761 * what we can return is probably a bit silly.
763 * We also need to check that we don't go beyond EOF;
764 * this is a truncate optimisation as a truncate sets
765 * the new file size before block on the pages we
766 * currently have locked under writeback. Because they
767 * are about to be tossed, we don't need to write them
771 end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
772 error = xfs_bmap_last_offset(ip, &last_block,
777 last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
778 if ((map_start_fsb + count_fsb) > last_block) {
779 count_fsb = last_block - map_start_fsb;
780 if (count_fsb == 0) {
787 * From this point onwards we overwrite the imap
788 * pointer that the caller gave to us.
790 error = xfs_bmapi_write(tp, ip, map_start_fsb,
791 count_fsb, flags, nres, imap,
796 error = xfs_trans_commit(tp);
800 if (whichfork == XFS_COW_FORK)
801 *cow_seq = READ_ONCE(ifp->if_seq);
802 xfs_iunlock(ip, XFS_ILOCK_EXCL);
806 * See if we were able to allocate an extent that
807 * covers at least part of the callers request
809 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
810 return xfs_alert_fsblock_zero(ip, imap);
812 if ((offset_fsb >= imap->br_startoff) &&
813 (offset_fsb < (imap->br_startoff +
814 imap->br_blockcount))) {
815 XFS_STATS_INC(mp, xs_xstrat_quick);
820 * So far we have not mapped the requested part of the
821 * file, just surrounding data, try again.
823 count_fsb -= imap->br_blockcount;
824 map_start_fsb = imap->br_startoff + imap->br_blockcount;
828 xfs_trans_cancel(tp);
830 xfs_iunlock(ip, XFS_ILOCK_EXCL);
835 xfs_iomap_write_unwritten(
841 xfs_mount_t *mp = ip->i_mount;
842 xfs_fileoff_t offset_fsb;
843 xfs_filblks_t count_fsb;
844 xfs_filblks_t numblks_fsb;
847 xfs_bmbt_irec_t imap;
848 struct inode *inode = VFS_I(ip);
853 trace_xfs_unwritten_convert(ip, offset, count);
855 offset_fsb = XFS_B_TO_FSBT(mp, offset);
856 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
857 count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
860 * Reserve enough blocks in this transaction for two complete extent
861 * btree splits. We may be converting the middle part of an unwritten
862 * extent and in this case we will insert two new extents in the btree
863 * each of which could cause a full split.
865 * This reservation amount will be used in the first call to
866 * xfs_bmbt_split() to select an AG with enough space to satisfy the
867 * rest of the operation.
869 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
873 * Set up a transaction to convert the range of extents
874 * from unwritten to real. Do allocations in a loop until
875 * we have covered the range passed in.
877 * Note that we can't risk to recursing back into the filesystem
878 * here as we might be asked to write out the same inode that we
879 * complete here and might deadlock on the iolock.
881 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0,
882 XFS_TRANS_RESERVE | XFS_TRANS_NOFS, &tp);
886 xfs_ilock(ip, XFS_ILOCK_EXCL);
887 xfs_trans_ijoin(tp, ip, 0);
890 * Modify the unwritten extent state of the buffer.
893 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
894 XFS_BMAPI_CONVERT, resblks, &imap,
897 goto error_on_bmapi_transaction;
900 * Log the updated inode size as we go. We have to be careful
901 * to only log it up to the actual write offset if it is
902 * halfway into a block.
904 i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
905 if (i_size > offset + count)
906 i_size = offset + count;
907 if (update_isize && i_size > i_size_read(inode))
908 i_size_write(inode, i_size);
909 i_size = xfs_new_eof(ip, i_size);
911 ip->i_d.di_size = i_size;
912 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
915 error = xfs_trans_commit(tp);
916 xfs_iunlock(ip, XFS_ILOCK_EXCL);
920 if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
921 return xfs_alert_fsblock_zero(ip, &imap);
923 if ((numblks_fsb = imap.br_blockcount) == 0) {
925 * The numblks_fsb value should always get
926 * smaller, otherwise the loop is stuck.
928 ASSERT(imap.br_blockcount);
931 offset_fsb += numblks_fsb;
932 count_fsb -= numblks_fsb;
933 } while (count_fsb > 0);
937 error_on_bmapi_transaction:
938 xfs_trans_cancel(tp);
939 xfs_iunlock(ip, XFS_ILOCK_EXCL);
946 struct xfs_bmbt_irec *imap,
950 imap->br_startblock == HOLESTARTBLOCK ||
951 imap->br_startblock == DELAYSTARTBLOCK ||
952 (IS_DAX(inode) && imap->br_state == XFS_EXT_UNWRITTEN);
956 needs_cow_for_zeroing(
957 struct xfs_bmbt_irec *imap,
961 imap->br_startblock != HOLESTARTBLOCK &&
962 imap->br_state != XFS_EXT_UNWRITTEN;
967 struct xfs_inode *ip,
971 unsigned mode = XFS_ILOCK_SHARED;
972 bool is_write = flags & (IOMAP_WRITE | IOMAP_ZERO);
975 * COW writes may allocate delalloc space or convert unwritten COW
976 * extents, so we need to make sure to take the lock exclusively here.
978 if (xfs_is_reflink_inode(ip) && is_write) {
980 * FIXME: It could still overwrite on unshared extents and not
983 if (flags & IOMAP_NOWAIT)
985 mode = XFS_ILOCK_EXCL;
989 * Extents not yet cached requires exclusive access, don't block. This
990 * is an opencoded xfs_ilock_data_map_shared() call but with
991 * non-blocking behaviour.
993 if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) {
994 if (flags & IOMAP_NOWAIT)
996 mode = XFS_ILOCK_EXCL;
1000 if (flags & IOMAP_NOWAIT) {
1001 if (!xfs_ilock_nowait(ip, mode))
1004 xfs_ilock(ip, mode);
1008 * The reflink iflag could have changed since the earlier unlocked
1009 * check, so if we got ILOCK_SHARED for a write and but we're now a
1010 * reflink inode we have to switch to ILOCK_EXCL and relock.
1012 if (mode == XFS_ILOCK_SHARED && is_write && xfs_is_reflink_inode(ip)) {
1013 xfs_iunlock(ip, mode);
1014 mode = XFS_ILOCK_EXCL;
1023 xfs_file_iomap_begin(
1024 struct inode *inode,
1028 struct iomap *iomap)
1030 struct xfs_inode *ip = XFS_I(inode);
1031 struct xfs_mount *mp = ip->i_mount;
1032 struct xfs_bmbt_irec imap;
1033 xfs_fileoff_t offset_fsb, end_fsb;
1034 int nimaps = 1, error = 0;
1035 bool shared = false;
1038 if (XFS_FORCED_SHUTDOWN(mp))
1041 if ((flags & (IOMAP_WRITE | IOMAP_ZERO)) && !(flags & IOMAP_DIRECT) &&
1042 !IS_DAX(inode) && !xfs_get_extsz_hint(ip)) {
1043 /* Reserve delalloc blocks for regular writeback. */
1044 return xfs_file_iomap_begin_delay(inode, offset, length, flags,
1049 * Lock the inode in the manner required for the specified operation and
1050 * check for as many conditions that would result in blocking as
1051 * possible. This removes most of the non-blocking checks from the
1052 * mapping code below.
1054 error = xfs_ilock_for_iomap(ip, flags, &lockmode);
1058 ASSERT(offset <= mp->m_super->s_maxbytes);
1059 if (offset > mp->m_super->s_maxbytes - length)
1060 length = mp->m_super->s_maxbytes - offset;
1061 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1062 end_fsb = XFS_B_TO_FSB(mp, offset + length);
1064 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1069 if (flags & IOMAP_REPORT) {
1070 /* Trim the mapping to the nearest shared extent boundary. */
1071 error = xfs_reflink_trim_around_shared(ip, &imap, &shared);
1076 /* Non-modifying mapping requested, so we are done */
1077 if (!(flags & (IOMAP_WRITE | IOMAP_ZERO)))
1081 * Break shared extents if necessary. Checks for non-blocking IO have
1082 * been done up front, so we don't need to do them here.
1084 if (xfs_is_reflink_inode(ip)) {
1085 /* if zeroing doesn't need COW allocation, then we are done. */
1086 if ((flags & IOMAP_ZERO) &&
1087 !needs_cow_for_zeroing(&imap, nimaps))
1090 if (flags & IOMAP_DIRECT) {
1091 /* may drop and re-acquire the ilock */
1092 error = xfs_reflink_allocate_cow(ip, &imap, &shared,
1097 error = xfs_reflink_reserve_cow(ip, &imap);
1102 end_fsb = imap.br_startoff + imap.br_blockcount;
1103 length = XFS_FSB_TO_B(mp, end_fsb) - offset;
1106 /* Don't need to allocate over holes when doing zeroing operations. */
1107 if (flags & IOMAP_ZERO)
1110 if (!imap_needs_alloc(inode, &imap, nimaps))
1113 /* If nowait is set bail since we are going to make allocations. */
1114 if (flags & IOMAP_NOWAIT) {
1120 * We cap the maximum length we map to a sane size to keep the chunks
1121 * of work done where somewhat symmetric with the work writeback does.
1122 * This is a completely arbitrary number pulled out of thin air as a
1123 * best guess for initial testing.
1125 * Note that the values needs to be less than 32-bits wide until the
1126 * lower level functions are updated.
1128 length = min_t(loff_t, length, 1024 * PAGE_SIZE);
1131 * xfs_iomap_write_direct() expects the shared lock. It is unlocked on
1134 if (lockmode == XFS_ILOCK_EXCL)
1135 xfs_ilock_demote(ip, lockmode);
1136 error = xfs_iomap_write_direct(ip, offset, length, &imap,
1141 iomap->flags |= IOMAP_F_NEW;
1142 trace_xfs_iomap_alloc(ip, offset, length, 0, &imap);
1145 if (xfs_ipincount(ip) && (ip->i_itemp->ili_fsync_fields
1146 & ~XFS_ILOG_TIMESTAMP))
1147 iomap->flags |= IOMAP_F_DIRTY;
1149 xfs_bmbt_to_iomap(ip, iomap, &imap);
1152 iomap->flags |= IOMAP_F_SHARED;
1157 xfs_iunlock(ip, lockmode);
1158 trace_xfs_iomap_found(ip, offset, length, 0, &imap);
1162 xfs_iunlock(ip, lockmode);
1167 xfs_file_iomap_end_delalloc(
1168 struct xfs_inode *ip,
1172 struct iomap *iomap)
1174 struct xfs_mount *mp = ip->i_mount;
1175 xfs_fileoff_t start_fsb;
1176 xfs_fileoff_t end_fsb;
1180 * Behave as if the write failed if drop writes is enabled. Set the NEW
1181 * flag to force delalloc cleanup.
1183 if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_DROP_WRITES)) {
1184 iomap->flags |= IOMAP_F_NEW;
1189 * start_fsb refers to the first unused block after a short write. If
1190 * nothing was written, round offset down to point at the first block in
1193 if (unlikely(!written))
1194 start_fsb = XFS_B_TO_FSBT(mp, offset);
1196 start_fsb = XFS_B_TO_FSB(mp, offset + written);
1197 end_fsb = XFS_B_TO_FSB(mp, offset + length);
1200 * Trim delalloc blocks if they were allocated by this write and we
1201 * didn't manage to write the whole range.
1203 * We don't need to care about racing delalloc as we hold i_mutex
1204 * across the reserve/allocate/unreserve calls. If there are delalloc
1205 * blocks in the range, they are ours.
1207 if ((iomap->flags & IOMAP_F_NEW) && start_fsb < end_fsb) {
1208 truncate_pagecache_range(VFS_I(ip), XFS_FSB_TO_B(mp, start_fsb),
1209 XFS_FSB_TO_B(mp, end_fsb) - 1);
1211 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1212 end_fsb - start_fsb);
1213 if (error && !XFS_FORCED_SHUTDOWN(mp)) {
1214 xfs_alert(mp, "%s: unable to clean up ino %lld",
1215 __func__, ip->i_ino);
1225 struct inode *inode,
1230 struct iomap *iomap)
1232 if ((flags & IOMAP_WRITE) && iomap->type == IOMAP_DELALLOC)
1233 return xfs_file_iomap_end_delalloc(XFS_I(inode), offset,
1234 length, written, iomap);
1238 const struct iomap_ops xfs_iomap_ops = {
1239 .iomap_begin = xfs_file_iomap_begin,
1240 .iomap_end = xfs_file_iomap_end,
1244 xfs_xattr_iomap_begin(
1245 struct inode *inode,
1249 struct iomap *iomap)
1251 struct xfs_inode *ip = XFS_I(inode);
1252 struct xfs_mount *mp = ip->i_mount;
1253 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
1254 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + length);
1255 struct xfs_bmbt_irec imap;
1256 int nimaps = 1, error = 0;
1259 if (XFS_FORCED_SHUTDOWN(mp))
1262 lockmode = xfs_ilock_attr_map_shared(ip);
1264 /* if there are no attribute fork or extents, return ENOENT */
1265 if (!XFS_IFORK_Q(ip) || !ip->i_d.di_anextents) {
1270 ASSERT(ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL);
1271 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1272 &nimaps, XFS_BMAPI_ATTRFORK);
1274 xfs_iunlock(ip, lockmode);
1278 xfs_bmbt_to_iomap(ip, iomap, &imap);
1284 const struct iomap_ops xfs_xattr_iomap_ops = {
1285 .iomap_begin = xfs_xattr_iomap_begin,