2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
7 * published by the Free Software Foundation.
9 * This program is distributed in the hope that it would be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write the Free Software Foundation,
16 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
22 #include "xfs_trans.h"
25 #include "xfs_alloc.h"
26 #include "xfs_quota.h"
27 #include "xfs_mount.h"
28 #include "xfs_bmap_btree.h"
29 #include "xfs_alloc_btree.h"
30 #include "xfs_ialloc_btree.h"
31 #include "xfs_dinode.h"
32 #include "xfs_inode.h"
33 #include "xfs_inode_item.h"
34 #include "xfs_btree.h"
36 #include "xfs_rtalloc.h"
37 #include "xfs_error.h"
38 #include "xfs_itable.h"
41 #include "xfs_buf_item.h"
42 #include "xfs_trans_space.h"
43 #include "xfs_utils.h"
44 #include "xfs_iomap.h"
45 #include "xfs_trace.h"
48 #define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \
50 #define XFS_WRITE_IMAPS XFS_BMAP_MAX_NMAP
53 xfs_iomap_eof_align_last_fsb(
57 xfs_fileoff_t *last_fsb)
59 xfs_fileoff_t new_last_fsb = 0;
60 xfs_extlen_t align = 0;
63 if (!XFS_IS_REALTIME_INODE(ip)) {
65 * Round up the allocation request to a stripe unit
66 * (m_dalign) boundary if the file size is >= stripe unit
67 * size, and we are allocating past the allocation eof.
69 * If mounted with the "-o swalloc" option the alignment is
70 * increased from the strip unit size to the stripe width.
72 if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
74 else if (mp->m_dalign)
77 if (align && XFS_ISIZE(ip) >= XFS_FSB_TO_B(mp, align))
78 new_last_fsb = roundup_64(*last_fsb, align);
82 * Always round up the allocation request to an extent boundary
83 * (when file on a real-time subvolume or has di_extsize hint).
87 align = roundup_64(new_last_fsb, extsize);
90 new_last_fsb = roundup_64(*last_fsb, align);
94 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
98 *last_fsb = new_last_fsb;
104 xfs_alert_fsblock_zero(
106 xfs_bmbt_irec_t *imap)
108 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
109 "Access to block zero in inode %llu "
110 "start_block: %llx start_off: %llx "
111 "blkcnt: %llx extent-state: %x\n",
112 (unsigned long long)ip->i_ino,
113 (unsigned long long)imap->br_startblock,
114 (unsigned long long)imap->br_startoff,
115 (unsigned long long)imap->br_blockcount,
121 xfs_iomap_write_direct(
125 xfs_bmbt_irec_t *imap,
128 xfs_mount_t *mp = ip->i_mount;
129 xfs_fileoff_t offset_fsb;
130 xfs_fileoff_t last_fsb;
131 xfs_filblks_t count_fsb, resaligned;
132 xfs_fsblock_t firstfsb;
133 xfs_extlen_t extsz, temp;
139 xfs_bmap_free_t free_list;
140 uint qblocks, resblks, resrtextents;
144 error = xfs_qm_dqattach(ip, 0);
146 return XFS_ERROR(error);
148 rt = XFS_IS_REALTIME_INODE(ip);
149 extsz = xfs_get_extsz_hint(ip);
151 offset_fsb = XFS_B_TO_FSBT(mp, offset);
152 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
153 if ((offset + count) > XFS_ISIZE(ip)) {
154 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
156 return XFS_ERROR(error);
158 if (nmaps && (imap->br_startblock == HOLESTARTBLOCK))
159 last_fsb = MIN(last_fsb, (xfs_fileoff_t)
160 imap->br_blockcount +
163 count_fsb = last_fsb - offset_fsb;
164 ASSERT(count_fsb > 0);
166 resaligned = count_fsb;
167 if (unlikely(extsz)) {
168 if ((temp = do_mod(offset_fsb, extsz)))
170 if ((temp = do_mod(resaligned, extsz)))
171 resaligned += extsz - temp;
175 resrtextents = qblocks = resaligned;
176 resrtextents /= mp->m_sb.sb_rextsize;
177 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
178 quota_flag = XFS_QMOPT_RES_RTBLKS;
181 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
182 quota_flag = XFS_QMOPT_RES_REGBLKS;
186 * Allocate and setup the transaction
188 tp = xfs_trans_alloc(mp, XFS_TRANS_DIOSTRAT);
189 error = xfs_trans_reserve(tp, resblks,
190 XFS_WRITE_LOG_RES(mp), resrtextents,
191 XFS_TRANS_PERM_LOG_RES,
192 XFS_WRITE_LOG_COUNT);
194 * Check for running out of space, note: need lock to return
197 xfs_trans_cancel(tp, 0);
198 return XFS_ERROR(error);
201 xfs_ilock(ip, XFS_ILOCK_EXCL);
203 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
205 goto out_trans_cancel;
207 xfs_trans_ijoin(tp, ip, 0);
210 if (offset < XFS_ISIZE(ip) || extsz)
211 bmapi_flag |= XFS_BMAPI_PREALLOC;
214 * From this point onwards we overwrite the imap pointer that the
217 xfs_bmap_init(&free_list, &firstfsb);
219 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb, bmapi_flag,
220 &firstfsb, 0, imap, &nimaps, &free_list);
222 goto out_bmap_cancel;
225 * Complete the transaction
227 error = xfs_bmap_finish(&tp, &free_list, &committed);
229 goto out_bmap_cancel;
230 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
235 * Copy any maps to caller's array and return any error.
238 error = XFS_ERROR(ENOSPC);
242 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
243 error = xfs_alert_fsblock_zero(ip, imap);
246 xfs_iunlock(ip, XFS_ILOCK_EXCL);
250 xfs_bmap_cancel(&free_list);
251 xfs_trans_unreserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
253 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
258 * If the caller is doing a write at the end of the file, then extend the
259 * allocation out to the file system's write iosize. We clean up any extra
260 * space left over when the file is closed in xfs_inactive().
262 * If we find we already have delalloc preallocation beyond EOF, don't do more
263 * preallocation as it it not needed.
266 xfs_iomap_eof_want_preallocate(
271 xfs_bmbt_irec_t *imap,
275 xfs_fileoff_t start_fsb;
276 xfs_filblks_t count_fsb;
277 xfs_fsblock_t firstblock;
279 int found_delalloc = 0;
282 if (offset + count <= XFS_ISIZE(ip))
286 * If there are any real blocks past eof, then don't
287 * do any speculative allocation.
289 start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
290 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)XFS_MAXIOFFSET(mp));
291 while (count_fsb > 0) {
293 firstblock = NULLFSBLOCK;
294 error = xfs_bmapi_read(ip, start_fsb, count_fsb, imap, &imaps,
298 for (n = 0; n < imaps; n++) {
299 if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
300 (imap[n].br_startblock != DELAYSTARTBLOCK))
302 start_fsb += imap[n].br_blockcount;
303 count_fsb -= imap[n].br_blockcount;
305 if (imap[n].br_startblock == DELAYSTARTBLOCK)
315 * If we don't have a user specified preallocation size, dynamically increase
316 * the preallocation size as the size of the file grows. Cap the maximum size
317 * at a single extent or less if the filesystem is near full. The closer the
318 * filesystem is to full, the smaller the maximum prealocation.
321 xfs_iomap_prealloc_size(
322 struct xfs_mount *mp,
323 struct xfs_inode *ip)
325 xfs_fsblock_t alloc_blocks = 0;
327 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)) {
332 * rounddown_pow_of_two() returns an undefined result
333 * if we pass in alloc_blocks = 0. Hence the "+ 1" to
334 * ensure we always pass in a non-zero value.
336 alloc_blocks = XFS_B_TO_FSB(mp, XFS_ISIZE(ip)) + 1;
337 alloc_blocks = XFS_FILEOFF_MIN(MAXEXTLEN,
338 rounddown_pow_of_two(alloc_blocks));
340 xfs_icsb_sync_counters(mp, XFS_ICSB_LAZY_COUNT);
341 freesp = mp->m_sb.sb_fdblocks;
342 if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
344 if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
346 if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
348 if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
350 if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
354 alloc_blocks >>= shift;
357 if (alloc_blocks < mp->m_writeio_blocks)
358 alloc_blocks = mp->m_writeio_blocks;
364 xfs_iomap_write_delay(
368 xfs_bmbt_irec_t *ret_imap)
370 xfs_mount_t *mp = ip->i_mount;
371 xfs_fileoff_t offset_fsb;
372 xfs_fileoff_t last_fsb;
373 xfs_off_t aligned_offset;
374 xfs_fileoff_t ioalign;
377 xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
378 int prealloc, flushed = 0;
381 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
384 * Make sure that the dquots are there. This doesn't hold
385 * the ilock across a disk read.
387 error = xfs_qm_dqattach_locked(ip, 0);
389 return XFS_ERROR(error);
391 extsz = xfs_get_extsz_hint(ip);
392 offset_fsb = XFS_B_TO_FSBT(mp, offset);
395 error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count,
396 imap, XFS_WRITE_IMAPS, &prealloc);
402 xfs_fsblock_t alloc_blocks = xfs_iomap_prealloc_size(mp, ip);
404 aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
405 ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
406 last_fsb = ioalign + alloc_blocks;
408 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
411 if (prealloc || extsz) {
412 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
418 * Make sure preallocation does not create extents beyond the range we
419 * actually support in this filesystem.
421 if (last_fsb > XFS_B_TO_FSB(mp, mp->m_maxioffset))
422 last_fsb = XFS_B_TO_FSB(mp, mp->m_maxioffset);
424 ASSERT(last_fsb > offset_fsb);
426 nimaps = XFS_WRITE_IMAPS;
427 error = xfs_bmapi_delay(ip, offset_fsb, last_fsb - offset_fsb,
428 imap, &nimaps, XFS_BMAPI_ENTIRE);
435 return XFS_ERROR(error);
439 * If bmapi returned us nothing, we got either ENOSPC or EDQUOT. For
440 * ENOSPC, * flush all other inodes with delalloc blocks to free up
441 * some of the excess reserved metadata space. For both cases, retry
442 * without EOF preallocation.
445 trace_xfs_delalloc_enospc(ip, offset, count);
447 return XFS_ERROR(error ? error : ENOSPC);
449 if (error == ENOSPC) {
450 xfs_iunlock(ip, XFS_ILOCK_EXCL);
451 xfs_flush_inodes(ip);
452 xfs_ilock(ip, XFS_ILOCK_EXCL);
461 if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip)))
462 return xfs_alert_fsblock_zero(ip, &imap[0]);
469 * Pass in a delayed allocate extent, convert it to real extents;
470 * return to the caller the extent we create which maps on top of
471 * the originating callers request.
473 * Called without a lock on the inode.
475 * We no longer bother to look at the incoming map - all we have to
476 * guarantee is that whatever we allocate fills the required range.
479 xfs_iomap_write_allocate(
483 xfs_bmbt_irec_t *imap)
485 xfs_mount_t *mp = ip->i_mount;
486 xfs_fileoff_t offset_fsb, last_block;
487 xfs_fileoff_t end_fsb, map_start_fsb;
488 xfs_fsblock_t first_block;
489 xfs_bmap_free_t free_list;
490 xfs_filblks_t count_fsb;
492 int nimaps, committed;
497 * Make sure that the dquots are there.
499 error = xfs_qm_dqattach(ip, 0);
501 return XFS_ERROR(error);
503 offset_fsb = XFS_B_TO_FSBT(mp, offset);
504 count_fsb = imap->br_blockcount;
505 map_start_fsb = imap->br_startoff;
507 XFS_STATS_ADD(xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
509 while (count_fsb != 0) {
511 * Set up a transaction with which to allocate the
512 * backing store for the file. Do allocations in a
513 * loop until we get some space in the range we are
514 * interested in. The other space that might be allocated
515 * is in the delayed allocation extent on which we sit
516 * but before our buffer starts.
520 while (nimaps == 0) {
521 tp = xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE);
522 tp->t_flags |= XFS_TRANS_RESERVE;
523 nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
524 error = xfs_trans_reserve(tp, nres,
525 XFS_WRITE_LOG_RES(mp),
526 0, XFS_TRANS_PERM_LOG_RES,
527 XFS_WRITE_LOG_COUNT);
529 xfs_trans_cancel(tp, 0);
530 return XFS_ERROR(error);
532 xfs_ilock(ip, XFS_ILOCK_EXCL);
533 xfs_trans_ijoin(tp, ip, 0);
535 xfs_bmap_init(&free_list, &first_block);
538 * it is possible that the extents have changed since
539 * we did the read call as we dropped the ilock for a
540 * while. We have to be careful about truncates or hole
541 * punchs here - we are not allowed to allocate
542 * non-delalloc blocks here.
544 * The only protection against truncation is the pages
545 * for the range we are being asked to convert are
546 * locked and hence a truncate will block on them
549 * As a result, if we go beyond the range we really
550 * need and hit an delalloc extent boundary followed by
551 * a hole while we have excess blocks in the map, we
552 * will fill the hole incorrectly and overrun the
553 * transaction reservation.
555 * Using a single map prevents this as we are forced to
556 * check each map we look for overlap with the desired
557 * range and abort as soon as we find it. Also, given
558 * that we only return a single map, having one beyond
559 * what we can return is probably a bit silly.
561 * We also need to check that we don't go beyond EOF;
562 * this is a truncate optimisation as a truncate sets
563 * the new file size before block on the pages we
564 * currently have locked under writeback. Because they
565 * are about to be tossed, we don't need to write them
569 end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
570 error = xfs_bmap_last_offset(NULL, ip, &last_block,
575 last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
576 if ((map_start_fsb + count_fsb) > last_block) {
577 count_fsb = last_block - map_start_fsb;
578 if (count_fsb == 0) {
585 * From this point onwards we overwrite the imap
586 * pointer that the caller gave to us.
588 error = xfs_bmapi_write(tp, ip, map_start_fsb,
589 count_fsb, 0, &first_block, 1,
590 imap, &nimaps, &free_list);
594 error = xfs_bmap_finish(&tp, &free_list, &committed);
598 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
602 xfs_iunlock(ip, XFS_ILOCK_EXCL);
606 * See if we were able to allocate an extent that
607 * covers at least part of the callers request
609 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
610 return xfs_alert_fsblock_zero(ip, imap);
612 if ((offset_fsb >= imap->br_startoff) &&
613 (offset_fsb < (imap->br_startoff +
614 imap->br_blockcount))) {
615 XFS_STATS_INC(xs_xstrat_quick);
620 * So far we have not mapped the requested part of the
621 * file, just surrounding data, try again.
623 count_fsb -= imap->br_blockcount;
624 map_start_fsb = imap->br_startoff + imap->br_blockcount;
628 xfs_bmap_cancel(&free_list);
629 xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT);
631 xfs_iunlock(ip, XFS_ILOCK_EXCL);
632 return XFS_ERROR(error);
636 xfs_iomap_write_unwritten(
641 xfs_mount_t *mp = ip->i_mount;
642 xfs_fileoff_t offset_fsb;
643 xfs_filblks_t count_fsb;
644 xfs_filblks_t numblks_fsb;
645 xfs_fsblock_t firstfsb;
648 xfs_bmbt_irec_t imap;
649 xfs_bmap_free_t free_list;
655 trace_xfs_unwritten_convert(ip, offset, count);
657 offset_fsb = XFS_B_TO_FSBT(mp, offset);
658 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
659 count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
662 * Reserve enough blocks in this transaction for two complete extent
663 * btree splits. We may be converting the middle part of an unwritten
664 * extent and in this case we will insert two new extents in the btree
665 * each of which could cause a full split.
667 * This reservation amount will be used in the first call to
668 * xfs_bmbt_split() to select an AG with enough space to satisfy the
669 * rest of the operation.
671 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
675 * set up a transaction to convert the range of extents
676 * from unwritten to real. Do allocations in a loop until
677 * we have covered the range passed in.
679 * Note that we open code the transaction allocation here
680 * to pass KM_NOFS--we can't risk to recursing back into
681 * the filesystem here as we might be asked to write out
682 * the same inode that we complete here and might deadlock
685 xfs_wait_for_freeze(mp, SB_FREEZE_TRANS);
686 tp = _xfs_trans_alloc(mp, XFS_TRANS_STRAT_WRITE, KM_NOFS);
687 tp->t_flags |= XFS_TRANS_RESERVE;
688 error = xfs_trans_reserve(tp, resblks,
689 XFS_WRITE_LOG_RES(mp), 0,
690 XFS_TRANS_PERM_LOG_RES,
691 XFS_WRITE_LOG_COUNT);
693 xfs_trans_cancel(tp, 0);
694 return XFS_ERROR(error);
697 xfs_ilock(ip, XFS_ILOCK_EXCL);
698 xfs_trans_ijoin(tp, ip, 0);
701 * Modify the unwritten extent state of the buffer.
703 xfs_bmap_init(&free_list, &firstfsb);
705 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
706 XFS_BMAPI_CONVERT, &firstfsb,
707 1, &imap, &nimaps, &free_list);
709 goto error_on_bmapi_transaction;
712 * Log the updated inode size as we go. We have to be careful
713 * to only log it up to the actual write offset if it is
714 * halfway into a block.
716 i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
717 if (i_size > offset + count)
718 i_size = offset + count;
720 i_size = xfs_new_eof(ip, i_size);
722 ip->i_d.di_size = i_size;
723 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
726 error = xfs_bmap_finish(&tp, &free_list, &committed);
728 goto error_on_bmapi_transaction;
730 error = xfs_trans_commit(tp, XFS_TRANS_RELEASE_LOG_RES);
731 xfs_iunlock(ip, XFS_ILOCK_EXCL);
733 return XFS_ERROR(error);
735 if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
736 return xfs_alert_fsblock_zero(ip, &imap);
738 if ((numblks_fsb = imap.br_blockcount) == 0) {
740 * The numblks_fsb value should always get
741 * smaller, otherwise the loop is stuck.
743 ASSERT(imap.br_blockcount);
746 offset_fsb += numblks_fsb;
747 count_fsb -= numblks_fsb;
748 } while (count_fsb > 0);
752 error_on_bmapi_transaction:
753 xfs_bmap_cancel(&free_list);
754 xfs_trans_cancel(tp, (XFS_TRANS_RELEASE_LOG_RES | XFS_TRANS_ABORT));
755 xfs_iunlock(ip, XFS_ILOCK_EXCL);
756 return XFS_ERROR(error);