1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
4 * Copyright (c) 2016 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));
70 struct xfs_mount *mp = ip->i_mount;
71 xfs_extlen_t align = 0;
73 if (!XFS_IS_REALTIME_INODE(ip)) {
75 * Round up the allocation request to a stripe unit
76 * (m_dalign) boundary if the file size is >= stripe unit
77 * size, and we are allocating past the allocation eof.
79 * If mounted with the "-o swalloc" option the alignment is
80 * increased from the strip unit size to the stripe width.
82 if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
84 else if (mp->m_dalign)
87 if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
92 * Always round up the allocation request to an extent boundary
93 * (when file on a real-time subvolume or has di_extsize hint).
97 align = roundup_64(align, extsize);
106 xfs_iomap_eof_align_last_fsb(
107 struct xfs_inode *ip,
108 xfs_extlen_t extsize,
109 xfs_fileoff_t *last_fsb)
111 xfs_extlen_t align = xfs_eof_alignment(ip, extsize);
114 xfs_fileoff_t new_last_fsb = roundup_64(*last_fsb, align);
117 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
121 *last_fsb = new_last_fsb;
127 xfs_alert_fsblock_zero(
129 xfs_bmbt_irec_t *imap)
131 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
132 "Access to block zero in inode %llu "
133 "start_block: %llx start_off: %llx "
134 "blkcnt: %llx extent-state: %x",
135 (unsigned long long)ip->i_ino,
136 (unsigned long long)imap->br_startblock,
137 (unsigned long long)imap->br_startoff,
138 (unsigned long long)imap->br_blockcount,
140 return -EFSCORRUPTED;
144 xfs_iomap_write_direct(
148 xfs_bmbt_irec_t *imap,
151 xfs_mount_t *mp = ip->i_mount;
152 xfs_fileoff_t offset_fsb;
153 xfs_fileoff_t last_fsb;
154 xfs_filblks_t count_fsb, resaligned;
155 xfs_fsblock_t firstfsb;
161 struct xfs_defer_ops dfops;
162 uint qblocks, resblks, resrtextents;
165 int bmapi_flags = XFS_BMAPI_PREALLOC;
168 rt = XFS_IS_REALTIME_INODE(ip);
169 extsz = xfs_get_extsz_hint(ip);
170 lockmode = XFS_ILOCK_SHARED; /* locked by caller */
172 ASSERT(xfs_isilocked(ip, lockmode));
174 offset_fsb = XFS_B_TO_FSBT(mp, offset);
175 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
176 if ((offset + count) > XFS_ISIZE(ip)) {
178 * Assert that the in-core extent list is present since this can
179 * call xfs_iread_extents() and we only have the ilock shared.
180 * This should be safe because the lock was held around a bmapi
181 * call in the caller and we only need it to access the in-core
184 ASSERT(XFS_IFORK_PTR(ip, XFS_DATA_FORK)->if_flags &
186 error = xfs_iomap_eof_align_last_fsb(ip, extsz, &last_fsb);
190 if (nmaps && (imap->br_startblock == HOLESTARTBLOCK))
191 last_fsb = MIN(last_fsb, (xfs_fileoff_t)
192 imap->br_blockcount +
195 count_fsb = last_fsb - offset_fsb;
196 ASSERT(count_fsb > 0);
197 resaligned = xfs_aligned_fsb_count(offset_fsb, count_fsb, extsz);
200 resrtextents = qblocks = resaligned;
201 resrtextents /= mp->m_sb.sb_rextsize;
202 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
203 quota_flag = XFS_QMOPT_RES_RTBLKS;
206 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
207 quota_flag = XFS_QMOPT_RES_REGBLKS;
211 * Drop the shared lock acquired by the caller, attach the dquot if
212 * necessary and move on to transaction setup.
214 xfs_iunlock(ip, lockmode);
215 error = xfs_qm_dqattach(ip);
220 * For DAX, we do not allocate unwritten extents, but instead we zero
221 * the block before we commit the transaction. Ideally we'd like to do
222 * this outside the transaction context, but if we commit and then crash
223 * we may not have zeroed the blocks and this will be exposed on
224 * recovery of the allocation. Hence we must zero before commit.
226 * Further, if we are mapping unwritten extents here, we need to zero
227 * and convert them to written so that we don't need an unwritten extent
228 * callback for DAX. This also means that we need to be able to dip into
229 * the reserve block pool for bmbt block allocation if there is no space
230 * left but we need to do unwritten extent conversion.
232 if (IS_DAX(VFS_I(ip))) {
233 bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO;
234 if (imap->br_state == XFS_EXT_UNWRITTEN) {
235 tflags |= XFS_TRANS_RESERVE;
236 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
239 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, resrtextents,
244 lockmode = XFS_ILOCK_EXCL;
245 xfs_ilock(ip, lockmode);
247 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
249 goto out_trans_cancel;
251 xfs_trans_ijoin(tp, ip, 0);
254 * From this point onwards we overwrite the imap pointer that the
257 xfs_defer_init(&dfops, &firstfsb);
259 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
260 bmapi_flags, &firstfsb, resblks, imap,
263 goto out_bmap_cancel;
266 * Complete the transaction
268 error = xfs_defer_finish(&tp, &dfops);
270 goto out_bmap_cancel;
272 error = xfs_trans_commit(tp);
277 * Copy any maps to caller's array and return any error.
284 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
285 error = xfs_alert_fsblock_zero(ip, imap);
288 xfs_iunlock(ip, lockmode);
292 xfs_defer_cancel(&dfops);
293 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
295 xfs_trans_cancel(tp);
300 xfs_quota_need_throttle(
301 struct xfs_inode *ip,
303 xfs_fsblock_t alloc_blocks)
305 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
307 if (!dq || !xfs_this_quota_on(ip->i_mount, type))
310 /* no hi watermark, no throttle */
311 if (!dq->q_prealloc_hi_wmark)
314 /* under the lo watermark, no throttle */
315 if (dq->q_res_bcount + alloc_blocks < dq->q_prealloc_lo_wmark)
322 xfs_quota_calc_throttle(
323 struct xfs_inode *ip,
325 xfs_fsblock_t *qblocks,
331 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
333 /* no dq, or over hi wmark, squash the prealloc completely */
334 if (!dq || dq->q_res_bcount >= dq->q_prealloc_hi_wmark) {
340 freesp = dq->q_prealloc_hi_wmark - dq->q_res_bcount;
341 if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
343 if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
345 if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
349 if (freesp < *qfreesp)
352 /* only overwrite the throttle values if we are more aggressive */
353 if ((freesp >> shift) < (*qblocks >> *qshift)) {
360 * If we are doing a write at the end of the file and there are no allocations
361 * past this one, then extend the allocation out to the file system's write
364 * If we don't have a user specified preallocation size, dynamically increase
365 * the preallocation size as the size of the file grows. Cap the maximum size
366 * at a single extent or less if the filesystem is near full. The closer the
367 * filesystem is to full, the smaller the maximum prealocation.
369 * As an exception we don't do any preallocation at all if the file is smaller
370 * than the minimum preallocation and we are using the default dynamic
371 * preallocation scheme, as it is likely this is the only write to the file that
372 * is going to be done.
374 * We clean up any extra space left over when the file is closed in
378 xfs_iomap_prealloc_size(
379 struct xfs_inode *ip,
382 struct xfs_iext_cursor *icur)
384 struct xfs_mount *mp = ip->i_mount;
385 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
386 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
387 struct xfs_bmbt_irec prev;
390 xfs_fsblock_t qblocks;
392 xfs_fsblock_t alloc_blocks = 0;
394 if (offset + count <= XFS_ISIZE(ip))
397 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) &&
398 (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_writeio_blocks)))
402 * If an explicit allocsize is set, the file is small, or we
403 * are writing behind a hole, then use the minimum prealloc:
405 if ((mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) ||
406 XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign) ||
407 !xfs_iext_peek_prev_extent(ifp, icur, &prev) ||
408 prev.br_startoff + prev.br_blockcount < offset_fsb)
409 return mp->m_writeio_blocks;
412 * Determine the initial size of the preallocation. We are beyond the
413 * current EOF here, but we need to take into account whether this is
414 * a sparse write or an extending write when determining the
415 * preallocation size. Hence we need to look up the extent that ends
416 * at the current write offset and use the result to determine the
417 * preallocation size.
419 * If the extent is a hole, then preallocation is essentially disabled.
420 * Otherwise we take the size of the preceding data extent as the basis
421 * for the preallocation size. If the size of the extent is greater than
422 * half the maximum extent length, then use the current offset as the
423 * basis. This ensures that for large files the preallocation size
424 * always extends to MAXEXTLEN rather than falling short due to things
425 * like stripe unit/width alignment of real extents.
427 if (prev.br_blockcount <= (MAXEXTLEN >> 1))
428 alloc_blocks = prev.br_blockcount << 1;
430 alloc_blocks = XFS_B_TO_FSB(mp, offset);
433 qblocks = alloc_blocks;
436 * MAXEXTLEN is not a power of two value but we round the prealloc down
437 * to the nearest power of two value after throttling. To prevent the
438 * round down from unconditionally reducing the maximum supported prealloc
439 * size, we round up first, apply appropriate throttling, round down and
440 * cap the value to MAXEXTLEN.
442 alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN),
445 freesp = percpu_counter_read_positive(&mp->m_fdblocks);
446 if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
448 if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
450 if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
452 if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
454 if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
459 * Check each quota to cap the prealloc size, provide a shift value to
460 * throttle with and adjust amount of available space.
462 if (xfs_quota_need_throttle(ip, XFS_DQ_USER, alloc_blocks))
463 xfs_quota_calc_throttle(ip, XFS_DQ_USER, &qblocks, &qshift,
465 if (xfs_quota_need_throttle(ip, XFS_DQ_GROUP, alloc_blocks))
466 xfs_quota_calc_throttle(ip, XFS_DQ_GROUP, &qblocks, &qshift,
468 if (xfs_quota_need_throttle(ip, XFS_DQ_PROJ, alloc_blocks))
469 xfs_quota_calc_throttle(ip, XFS_DQ_PROJ, &qblocks, &qshift,
473 * The final prealloc size is set to the minimum of free space available
474 * in each of the quotas and the overall filesystem.
476 * The shift throttle value is set to the maximum value as determined by
477 * the global low free space values and per-quota low free space values.
479 alloc_blocks = MIN(alloc_blocks, qblocks);
480 shift = MAX(shift, qshift);
483 alloc_blocks >>= shift;
485 * rounddown_pow_of_two() returns an undefined result if we pass in
489 alloc_blocks = rounddown_pow_of_two(alloc_blocks);
490 if (alloc_blocks > MAXEXTLEN)
491 alloc_blocks = MAXEXTLEN;
494 * If we are still trying to allocate more space than is
495 * available, squash the prealloc hard. This can happen if we
496 * have a large file on a small filesystem and the above
497 * lowspace thresholds are smaller than MAXEXTLEN.
499 while (alloc_blocks && alloc_blocks >= freesp)
502 if (alloc_blocks < mp->m_writeio_blocks)
503 alloc_blocks = mp->m_writeio_blocks;
504 trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
505 mp->m_writeio_blocks);
510 xfs_file_iomap_begin_delay(
516 struct xfs_inode *ip = XFS_I(inode);
517 struct xfs_mount *mp = ip->i_mount;
518 struct xfs_ifork *ifp = XFS_IFORK_PTR(ip, XFS_DATA_FORK);
519 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
520 xfs_fileoff_t maxbytes_fsb =
521 XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
522 xfs_fileoff_t end_fsb;
523 int error = 0, eof = 0;
524 struct xfs_bmbt_irec got;
525 struct xfs_iext_cursor icur;
526 xfs_fsblock_t prealloc_blocks = 0;
528 ASSERT(!XFS_IS_REALTIME_INODE(ip));
529 ASSERT(!xfs_get_extsz_hint(ip));
531 xfs_ilock(ip, XFS_ILOCK_EXCL);
533 if (unlikely(XFS_TEST_ERROR(
534 (XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_EXTENTS &&
535 XFS_IFORK_FORMAT(ip, XFS_DATA_FORK) != XFS_DINODE_FMT_BTREE),
536 mp, XFS_ERRTAG_BMAPIFORMAT))) {
537 XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_LOW, mp);
538 error = -EFSCORRUPTED;
542 XFS_STATS_INC(mp, xs_blk_mapw);
544 if (!(ifp->if_flags & XFS_IFEXTENTS)) {
545 error = xfs_iread_extents(NULL, ip, XFS_DATA_FORK);
550 eof = !xfs_iext_lookup_extent(ip, ifp, offset_fsb, &icur, &got);
551 if (!eof && got.br_startoff <= offset_fsb) {
552 if (xfs_is_reflink_inode(ip)) {
555 end_fsb = min(XFS_B_TO_FSB(mp, offset + count),
557 xfs_trim_extent(&got, offset_fsb, end_fsb - offset_fsb);
558 error = xfs_reflink_reserve_cow(ip, &got, &shared);
563 trace_xfs_iomap_found(ip, offset, count, 0, &got);
567 error = xfs_qm_dqattach_locked(ip, false);
572 * We cap the maximum length we map here to MAX_WRITEBACK_PAGES pages
573 * to keep the chunks of work done where somewhat symmetric with the
574 * work writeback does. This is a completely arbitrary number pulled
575 * out of thin air as a best guess for initial testing.
577 * Note that the values needs to be less than 32-bits wide until
578 * the lower level functions are updated.
580 count = min_t(loff_t, count, 1024 * PAGE_SIZE);
581 end_fsb = min(XFS_B_TO_FSB(mp, offset + count), maxbytes_fsb);
584 prealloc_blocks = xfs_iomap_prealloc_size(ip, offset, count,
586 if (prealloc_blocks) {
588 xfs_off_t end_offset;
589 xfs_fileoff_t p_end_fsb;
591 end_offset = XFS_WRITEIO_ALIGN(mp, offset + count - 1);
592 p_end_fsb = XFS_B_TO_FSBT(mp, end_offset) +
595 align = xfs_eof_alignment(ip, 0);
597 p_end_fsb = roundup_64(p_end_fsb, align);
599 p_end_fsb = min(p_end_fsb, maxbytes_fsb);
600 ASSERT(p_end_fsb > offset_fsb);
601 prealloc_blocks = p_end_fsb - end_fsb;
606 error = xfs_bmapi_reserve_delalloc(ip, XFS_DATA_FORK, offset_fsb,
607 end_fsb - offset_fsb, prealloc_blocks, &got, &icur,
614 /* retry without any preallocation */
615 trace_xfs_delalloc_enospc(ip, offset, count);
616 if (prealloc_blocks) {
626 * Flag newly allocated delalloc blocks with IOMAP_F_NEW so we punch
627 * them out if the write happens to fail.
629 iomap->flags = IOMAP_F_NEW;
630 trace_xfs_iomap_alloc(ip, offset, count, 0, &got);
632 if (isnullstartblock(got.br_startblock))
633 got.br_startblock = DELAYSTARTBLOCK;
635 if (!got.br_startblock) {
636 error = xfs_alert_fsblock_zero(ip, &got);
641 xfs_bmbt_to_iomap(ip, iomap, &got);
644 xfs_iunlock(ip, XFS_ILOCK_EXCL);
649 * Pass in a delayed allocate extent, convert it to real extents;
650 * return to the caller the extent we create which maps on top of
651 * the originating callers request.
653 * Called without a lock on the inode.
655 * We no longer bother to look at the incoming map - all we have to
656 * guarantee is that whatever we allocate fills the required range.
659 xfs_iomap_write_allocate(
663 xfs_bmbt_irec_t *imap)
665 xfs_mount_t *mp = ip->i_mount;
666 xfs_fileoff_t offset_fsb, last_block;
667 xfs_fileoff_t end_fsb, map_start_fsb;
668 xfs_fsblock_t first_block;
669 struct xfs_defer_ops dfops;
670 xfs_filblks_t count_fsb;
674 int flags = XFS_BMAPI_DELALLOC;
677 if (whichfork == XFS_COW_FORK)
678 flags |= XFS_BMAPI_COWFORK | XFS_BMAPI_PREALLOC;
681 * Make sure that the dquots are there.
683 error = xfs_qm_dqattach(ip);
687 offset_fsb = XFS_B_TO_FSBT(mp, offset);
688 count_fsb = imap->br_blockcount;
689 map_start_fsb = imap->br_startoff;
691 XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
693 while (count_fsb != 0) {
695 * Set up a transaction with which to allocate the
696 * backing store for the file. Do allocations in a
697 * loop until we get some space in the range we are
698 * interested in. The other space that might be allocated
699 * is in the delayed allocation extent on which we sit
700 * but before our buffer starts.
703 while (nimaps == 0) {
704 nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
706 * We have already reserved space for the extent and any
707 * indirect blocks when creating the delalloc extent,
708 * there is no need to reserve space in this transaction
711 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, 0,
712 0, XFS_TRANS_RESERVE, &tp);
716 xfs_ilock(ip, XFS_ILOCK_EXCL);
717 xfs_trans_ijoin(tp, ip, 0);
719 xfs_defer_init(&dfops, &first_block);
722 * it is possible that the extents have changed since
723 * we did the read call as we dropped the ilock for a
724 * while. We have to be careful about truncates or hole
725 * punchs here - we are not allowed to allocate
726 * non-delalloc blocks here.
728 * The only protection against truncation is the pages
729 * for the range we are being asked to convert are
730 * locked and hence a truncate will block on them
733 * As a result, if we go beyond the range we really
734 * need and hit an delalloc extent boundary followed by
735 * a hole while we have excess blocks in the map, we
736 * will fill the hole incorrectly and overrun the
737 * transaction reservation.
739 * Using a single map prevents this as we are forced to
740 * check each map we look for overlap with the desired
741 * range and abort as soon as we find it. Also, given
742 * that we only return a single map, having one beyond
743 * what we can return is probably a bit silly.
745 * We also need to check that we don't go beyond EOF;
746 * this is a truncate optimisation as a truncate sets
747 * the new file size before block on the pages we
748 * currently have locked under writeback. Because they
749 * are about to be tossed, we don't need to write them
753 end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
754 error = xfs_bmap_last_offset(ip, &last_block,
759 last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
760 if ((map_start_fsb + count_fsb) > last_block) {
761 count_fsb = last_block - map_start_fsb;
762 if (count_fsb == 0) {
769 * From this point onwards we overwrite the imap
770 * pointer that the caller gave to us.
772 error = xfs_bmapi_write(tp, ip, map_start_fsb,
773 count_fsb, flags, &first_block,
779 error = xfs_defer_finish(&tp, &dfops);
783 error = xfs_trans_commit(tp);
787 xfs_iunlock(ip, XFS_ILOCK_EXCL);
791 * See if we were able to allocate an extent that
792 * covers at least part of the callers request
794 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
795 return xfs_alert_fsblock_zero(ip, imap);
797 if ((offset_fsb >= imap->br_startoff) &&
798 (offset_fsb < (imap->br_startoff +
799 imap->br_blockcount))) {
800 XFS_STATS_INC(mp, xs_xstrat_quick);
805 * So far we have not mapped the requested part of the
806 * file, just surrounding data, try again.
808 count_fsb -= imap->br_blockcount;
809 map_start_fsb = imap->br_startoff + imap->br_blockcount;
813 xfs_defer_cancel(&dfops);
814 xfs_trans_cancel(tp);
816 xfs_iunlock(ip, XFS_ILOCK_EXCL);
821 xfs_iomap_write_unwritten(
827 xfs_mount_t *mp = ip->i_mount;
828 xfs_fileoff_t offset_fsb;
829 xfs_filblks_t count_fsb;
830 xfs_filblks_t numblks_fsb;
831 xfs_fsblock_t firstfsb;
834 xfs_bmbt_irec_t imap;
835 struct xfs_defer_ops dfops;
836 struct inode *inode = VFS_I(ip);
841 trace_xfs_unwritten_convert(ip, offset, count);
843 offset_fsb = XFS_B_TO_FSBT(mp, offset);
844 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
845 count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
848 * Reserve enough blocks in this transaction for two complete extent
849 * btree splits. We may be converting the middle part of an unwritten
850 * extent and in this case we will insert two new extents in the btree
851 * each of which could cause a full split.
853 * This reservation amount will be used in the first call to
854 * xfs_bmbt_split() to select an AG with enough space to satisfy the
855 * rest of the operation.
857 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
861 * Set up a transaction to convert the range of extents
862 * from unwritten to real. Do allocations in a loop until
863 * we have covered the range passed in.
865 * Note that we can't risk to recursing back into the filesystem
866 * here as we might be asked to write out the same inode that we
867 * complete here and might deadlock on the iolock.
869 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0,
870 XFS_TRANS_RESERVE | XFS_TRANS_NOFS, &tp);
874 xfs_ilock(ip, XFS_ILOCK_EXCL);
875 xfs_trans_ijoin(tp, ip, 0);
878 * Modify the unwritten extent state of the buffer.
880 xfs_defer_init(&dfops, &firstfsb);
882 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
883 XFS_BMAPI_CONVERT, &firstfsb, resblks,
884 &imap, &nimaps, &dfops);
886 goto error_on_bmapi_transaction;
889 * Log the updated inode size as we go. We have to be careful
890 * to only log it up to the actual write offset if it is
891 * halfway into a block.
893 i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
894 if (i_size > offset + count)
895 i_size = offset + count;
896 if (update_isize && i_size > i_size_read(inode))
897 i_size_write(inode, i_size);
898 i_size = xfs_new_eof(ip, i_size);
900 ip->i_d.di_size = i_size;
901 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
904 error = xfs_defer_finish(&tp, &dfops);
906 goto error_on_bmapi_transaction;
908 error = xfs_trans_commit(tp);
909 xfs_iunlock(ip, XFS_ILOCK_EXCL);
913 if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
914 return xfs_alert_fsblock_zero(ip, &imap);
916 if ((numblks_fsb = imap.br_blockcount) == 0) {
918 * The numblks_fsb value should always get
919 * smaller, otherwise the loop is stuck.
921 ASSERT(imap.br_blockcount);
924 offset_fsb += numblks_fsb;
925 count_fsb -= numblks_fsb;
926 } while (count_fsb > 0);
930 error_on_bmapi_transaction:
931 xfs_defer_cancel(&dfops);
932 xfs_trans_cancel(tp);
933 xfs_iunlock(ip, XFS_ILOCK_EXCL);
940 struct xfs_bmbt_irec *imap,
944 imap->br_startblock == HOLESTARTBLOCK ||
945 imap->br_startblock == DELAYSTARTBLOCK ||
946 (IS_DAX(inode) && imap->br_state == XFS_EXT_UNWRITTEN);
950 needs_cow_for_zeroing(
951 struct xfs_bmbt_irec *imap,
955 imap->br_startblock != HOLESTARTBLOCK &&
956 imap->br_state != XFS_EXT_UNWRITTEN;
961 struct xfs_inode *ip,
965 unsigned mode = XFS_ILOCK_SHARED;
968 * COW writes may allocate delalloc space or convert unwritten COW
969 * extents, so we need to make sure to take the lock exclusively here.
971 if (xfs_is_reflink_inode(ip) && (flags & (IOMAP_WRITE | IOMAP_ZERO))) {
973 * FIXME: It could still overwrite on unshared extents and not
976 if (flags & IOMAP_NOWAIT)
978 mode = XFS_ILOCK_EXCL;
982 * Extents not yet cached requires exclusive access, don't block. This
983 * is an opencoded xfs_ilock_data_map_shared() call but with
984 * non-blocking behaviour.
986 if (!(ip->i_df.if_flags & XFS_IFEXTENTS)) {
987 if (flags & IOMAP_NOWAIT)
989 mode = XFS_ILOCK_EXCL;
992 if (flags & IOMAP_NOWAIT) {
993 if (!xfs_ilock_nowait(ip, mode))
1004 xfs_file_iomap_begin(
1005 struct inode *inode,
1009 struct iomap *iomap)
1011 struct xfs_inode *ip = XFS_I(inode);
1012 struct xfs_mount *mp = ip->i_mount;
1013 struct xfs_bmbt_irec imap;
1014 xfs_fileoff_t offset_fsb, end_fsb;
1015 int nimaps = 1, error = 0;
1016 bool shared = false, trimmed = false;
1019 if (XFS_FORCED_SHUTDOWN(mp))
1022 if (((flags & (IOMAP_WRITE | IOMAP_DIRECT)) == IOMAP_WRITE) &&
1023 !IS_DAX(inode) && !xfs_get_extsz_hint(ip)) {
1024 /* Reserve delalloc blocks for regular writeback. */
1025 return xfs_file_iomap_begin_delay(inode, offset, length, iomap);
1029 * Lock the inode in the manner required for the specified operation and
1030 * check for as many conditions that would result in blocking as
1031 * possible. This removes most of the non-blocking checks from the
1032 * mapping code below.
1034 error = xfs_ilock_for_iomap(ip, flags, &lockmode);
1038 ASSERT(offset <= mp->m_super->s_maxbytes);
1039 if (offset > mp->m_super->s_maxbytes - length)
1040 length = mp->m_super->s_maxbytes - offset;
1041 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1042 end_fsb = XFS_B_TO_FSB(mp, offset + length);
1044 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1049 if (flags & IOMAP_REPORT) {
1050 /* Trim the mapping to the nearest shared extent boundary. */
1051 error = xfs_reflink_trim_around_shared(ip, &imap, &shared,
1057 /* Non-modifying mapping requested, so we are done */
1058 if (!(flags & (IOMAP_WRITE | IOMAP_ZERO)))
1062 * Break shared extents if necessary. Checks for non-blocking IO have
1063 * been done up front, so we don't need to do them here.
1065 if (xfs_is_reflink_inode(ip)) {
1066 /* if zeroing doesn't need COW allocation, then we are done. */
1067 if ((flags & IOMAP_ZERO) &&
1068 !needs_cow_for_zeroing(&imap, nimaps))
1071 if (flags & IOMAP_DIRECT) {
1072 /* may drop and re-acquire the ilock */
1073 error = xfs_reflink_allocate_cow(ip, &imap, &shared,
1078 error = xfs_reflink_reserve_cow(ip, &imap, &shared);
1083 end_fsb = imap.br_startoff + imap.br_blockcount;
1084 length = XFS_FSB_TO_B(mp, end_fsb) - offset;
1087 /* Don't need to allocate over holes when doing zeroing operations. */
1088 if (flags & IOMAP_ZERO)
1091 if (!imap_needs_alloc(inode, &imap, nimaps))
1094 /* If nowait is set bail since we are going to make allocations. */
1095 if (flags & IOMAP_NOWAIT) {
1101 * We cap the maximum length we map to a sane size to keep the chunks
1102 * of work done where somewhat symmetric with the work writeback does.
1103 * This is a completely arbitrary number pulled out of thin air as a
1104 * best guess for initial testing.
1106 * Note that the values needs to be less than 32-bits wide until the
1107 * lower level functions are updated.
1109 length = min_t(loff_t, length, 1024 * PAGE_SIZE);
1112 * xfs_iomap_write_direct() expects the shared lock. It is unlocked on
1115 if (lockmode == XFS_ILOCK_EXCL)
1116 xfs_ilock_demote(ip, lockmode);
1117 error = xfs_iomap_write_direct(ip, offset, length, &imap,
1122 iomap->flags = IOMAP_F_NEW;
1123 trace_xfs_iomap_alloc(ip, offset, length, 0, &imap);
1126 if (xfs_ipincount(ip) && (ip->i_itemp->ili_fsync_fields
1127 & ~XFS_ILOG_TIMESTAMP))
1128 iomap->flags |= IOMAP_F_DIRTY;
1130 xfs_bmbt_to_iomap(ip, iomap, &imap);
1133 iomap->flags |= IOMAP_F_SHARED;
1138 xfs_iunlock(ip, lockmode);
1139 trace_xfs_iomap_found(ip, offset, length, 0, &imap);
1143 xfs_iunlock(ip, lockmode);
1148 xfs_file_iomap_end_delalloc(
1149 struct xfs_inode *ip,
1153 struct iomap *iomap)
1155 struct xfs_mount *mp = ip->i_mount;
1156 xfs_fileoff_t start_fsb;
1157 xfs_fileoff_t end_fsb;
1161 * Behave as if the write failed if drop writes is enabled. Set the NEW
1162 * flag to force delalloc cleanup.
1164 if (XFS_TEST_ERROR(false, mp, XFS_ERRTAG_DROP_WRITES)) {
1165 iomap->flags |= IOMAP_F_NEW;
1170 * start_fsb refers to the first unused block after a short write. If
1171 * nothing was written, round offset down to point at the first block in
1174 if (unlikely(!written))
1175 start_fsb = XFS_B_TO_FSBT(mp, offset);
1177 start_fsb = XFS_B_TO_FSB(mp, offset + written);
1178 end_fsb = XFS_B_TO_FSB(mp, offset + length);
1181 * Trim delalloc blocks if they were allocated by this write and we
1182 * didn't manage to write the whole range.
1184 * We don't need to care about racing delalloc as we hold i_mutex
1185 * across the reserve/allocate/unreserve calls. If there are delalloc
1186 * blocks in the range, they are ours.
1188 if ((iomap->flags & IOMAP_F_NEW) && start_fsb < end_fsb) {
1189 truncate_pagecache_range(VFS_I(ip), XFS_FSB_TO_B(mp, start_fsb),
1190 XFS_FSB_TO_B(mp, end_fsb) - 1);
1192 xfs_ilock(ip, XFS_ILOCK_EXCL);
1193 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1194 end_fsb - start_fsb);
1195 xfs_iunlock(ip, XFS_ILOCK_EXCL);
1197 if (error && !XFS_FORCED_SHUTDOWN(mp)) {
1198 xfs_alert(mp, "%s: unable to clean up ino %lld",
1199 __func__, ip->i_ino);
1209 struct inode *inode,
1214 struct iomap *iomap)
1216 if ((flags & IOMAP_WRITE) && iomap->type == IOMAP_DELALLOC)
1217 return xfs_file_iomap_end_delalloc(XFS_I(inode), offset,
1218 length, written, iomap);
1222 const struct iomap_ops xfs_iomap_ops = {
1223 .iomap_begin = xfs_file_iomap_begin,
1224 .iomap_end = xfs_file_iomap_end,
1228 xfs_xattr_iomap_begin(
1229 struct inode *inode,
1233 struct iomap *iomap)
1235 struct xfs_inode *ip = XFS_I(inode);
1236 struct xfs_mount *mp = ip->i_mount;
1237 xfs_fileoff_t offset_fsb = XFS_B_TO_FSBT(mp, offset);
1238 xfs_fileoff_t end_fsb = XFS_B_TO_FSB(mp, offset + length);
1239 struct xfs_bmbt_irec imap;
1240 int nimaps = 1, error = 0;
1243 if (XFS_FORCED_SHUTDOWN(mp))
1246 lockmode = xfs_ilock_attr_map_shared(ip);
1248 /* if there are no attribute fork or extents, return ENOENT */
1249 if (!XFS_IFORK_Q(ip) || !ip->i_d.di_anextents) {
1254 ASSERT(ip->i_d.di_aformat != XFS_DINODE_FMT_LOCAL);
1255 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb, &imap,
1256 &nimaps, XFS_BMAPI_ATTRFORK);
1258 xfs_iunlock(ip, lockmode);
1262 xfs_bmbt_to_iomap(ip, iomap, &imap);
1268 const struct iomap_ops xfs_xattr_iomap_ops = {
1269 .iomap_begin = xfs_xattr_iomap_begin,