projects / deliverable / linux.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Merge branch 'for-linus-4.7' of git://git.kernel.org/pub/scm/linux/kernel/git/mason...
[deliverable/linux.git] / fs / xfs / xfs_iomap.c
1 /*
2 * Copyright (c) 2000-2006 Silicon Graphics, Inc.
3 * All Rights Reserved.
4 *
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.
8 *
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.
13 *
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
17 */
18 #include "xfs.h"
19 #include "xfs_fs.h"
20 #include "xfs_shared.h"
21 #include "xfs_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_mount.h"
25 #include "xfs_inode.h"
26 #include "xfs_btree.h"
27 #include "xfs_bmap_btree.h"
28 #include "xfs_bmap.h"
29 #include "xfs_bmap_util.h"
30 #include "xfs_error.h"
31 #include "xfs_trans.h"
32 #include "xfs_trans_space.h"
33 #include "xfs_iomap.h"
34 #include "xfs_trace.h"
35 #include "xfs_icache.h"
36 #include "xfs_quota.h"
37 #include "xfs_dquot_item.h"
38 #include "xfs_dquot.h"
39
40
41 #define XFS_WRITEIO_ALIGN(mp,off) (((off) >> mp->m_writeio_log) \
42 << mp->m_writeio_log)
43 #define XFS_WRITE_IMAPS XFS_BMAP_MAX_NMAP
44
45 STATIC int
46 xfs_iomap_eof_align_last_fsb(
47 xfs_mount_t *mp,
48 xfs_inode_t *ip,
49 xfs_extlen_t extsize,
50 xfs_fileoff_t *last_fsb)
51 {
52 xfs_extlen_t align = 0;
53 int eof, error;
54
55 if (!XFS_IS_REALTIME_INODE(ip)) {
56 /*
57 * Round up the allocation request to a stripe unit
58 * (m_dalign) boundary if the file size is >= stripe unit
59 * size, and we are allocating past the allocation eof.
60 *
61 * If mounted with the "-o swalloc" option the alignment is
62 * increased from the strip unit size to the stripe width.
63 */
64 if (mp->m_swidth && (mp->m_flags & XFS_MOUNT_SWALLOC))
65 align = mp->m_swidth;
66 else if (mp->m_dalign)
67 align = mp->m_dalign;
68
69 if (align && XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, align))
70 align = 0;
71 }
72
73 /*
74 * Always round up the allocation request to an extent boundary
75 * (when file on a real-time subvolume or has di_extsize hint).
76 */
77 if (extsize) {
78 if (align)
79 align = roundup_64(align, extsize);
80 else
81 align = extsize;
82 }
83
84 if (align) {
85 xfs_fileoff_t new_last_fsb = roundup_64(*last_fsb, align);
86 error = xfs_bmap_eof(ip, new_last_fsb, XFS_DATA_FORK, &eof);
87 if (error)
88 return error;
89 if (eof)
90 *last_fsb = new_last_fsb;
91 }
92 return 0;
93 }
94
95 STATIC int
96 xfs_alert_fsblock_zero(
97 xfs_inode_t *ip,
98 xfs_bmbt_irec_t *imap)
99 {
100 xfs_alert_tag(ip->i_mount, XFS_PTAG_FSBLOCK_ZERO,
101 "Access to block zero in inode %llu "
102 "start_block: %llx start_off: %llx "
103 "blkcnt: %llx extent-state: %x",
104 (unsigned long long)ip->i_ino,
105 (unsigned long long)imap->br_startblock,
106 (unsigned long long)imap->br_startoff,
107 (unsigned long long)imap->br_blockcount,
108 imap->br_state);
109 return -EFSCORRUPTED;
110 }
111
112 int
113 xfs_iomap_write_direct(
114 xfs_inode_t *ip,
115 xfs_off_t offset,
116 size_t count,
117 xfs_bmbt_irec_t *imap,
118 int nmaps)
119 {
120 xfs_mount_t *mp = ip->i_mount;
121 xfs_fileoff_t offset_fsb;
122 xfs_fileoff_t last_fsb;
123 xfs_filblks_t count_fsb, resaligned;
124 xfs_fsblock_t firstfsb;
125 xfs_extlen_t extsz, temp;
126 int nimaps;
127 int quota_flag;
128 int rt;
129 xfs_trans_t *tp;
130 xfs_bmap_free_t free_list;
131 uint qblocks, resblks, resrtextents;
132 int error;
133 int lockmode;
134 int bmapi_flags = XFS_BMAPI_PREALLOC;
135 uint tflags = 0;
136
137 rt = XFS_IS_REALTIME_INODE(ip);
138 extsz = xfs_get_extsz_hint(ip);
139 lockmode = XFS_ILOCK_SHARED; /* locked by caller */
140
141 ASSERT(xfs_isilocked(ip, lockmode));
142
143 offset_fsb = XFS_B_TO_FSBT(mp, offset);
144 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
145 if ((offset + count) > XFS_ISIZE(ip)) {
146 /*
147 * Assert that the in-core extent list is present since this can
148 * call xfs_iread_extents() and we only have the ilock shared.
149 * This should be safe because the lock was held around a bmapi
150 * call in the caller and we only need it to access the in-core
151 * list.
152 */
153 ASSERT(XFS_IFORK_PTR(ip, XFS_DATA_FORK)->if_flags &
154 XFS_IFEXTENTS);
155 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
156 if (error)
157 goto out_unlock;
158 } else {
159 if (nmaps && (imap->br_startblock == HOLESTARTBLOCK))
160 last_fsb = MIN(last_fsb, (xfs_fileoff_t)
161 imap->br_blockcount +
162 imap->br_startoff);
163 }
164 count_fsb = last_fsb - offset_fsb;
165 ASSERT(count_fsb > 0);
166
167 resaligned = count_fsb;
168 if (unlikely(extsz)) {
169 if ((temp = do_mod(offset_fsb, extsz)))
170 resaligned += temp;
171 if ((temp = do_mod(resaligned, extsz)))
172 resaligned += extsz - temp;
173 }
174
175 if (unlikely(rt)) {
176 resrtextents = qblocks = resaligned;
177 resrtextents /= mp->m_sb.sb_rextsize;
178 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0);
179 quota_flag = XFS_QMOPT_RES_RTBLKS;
180 } else {
181 resrtextents = 0;
182 resblks = qblocks = XFS_DIOSTRAT_SPACE_RES(mp, resaligned);
183 quota_flag = XFS_QMOPT_RES_REGBLKS;
184 }
185
186 /*
187 * Drop the shared lock acquired by the caller, attach the dquot if
188 * necessary and move on to transaction setup.
189 */
190 xfs_iunlock(ip, lockmode);
191 error = xfs_qm_dqattach(ip, 0);
192 if (error)
193 return error;
194
195 /*
196 * For DAX, we do not allocate unwritten extents, but instead we zero
197 * the block before we commit the transaction. Ideally we'd like to do
198 * this outside the transaction context, but if we commit and then crash
199 * we may not have zeroed the blocks and this will be exposed on
200 * recovery of the allocation. Hence we must zero before commit.
201 *
202 * Further, if we are mapping unwritten extents here, we need to zero
203 * and convert them to written so that we don't need an unwritten extent
204 * callback for DAX. This also means that we need to be able to dip into
205 * the reserve block pool for bmbt block allocation if there is no space
206 * left but we need to do unwritten extent conversion.
207 */
208 if (IS_DAX(VFS_I(ip))) {
209 bmapi_flags = XFS_BMAPI_CONVERT | XFS_BMAPI_ZERO;
210 if (ISUNWRITTEN(imap)) {
211 tflags |= XFS_TRANS_RESERVE;
212 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
213 }
214 }
215 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, resrtextents,
216 tflags, &tp);
217 if (error)
218 return error;
219
220 lockmode = XFS_ILOCK_EXCL;
221 xfs_ilock(ip, lockmode);
222
223 error = xfs_trans_reserve_quota_nblks(tp, ip, qblocks, 0, quota_flag);
224 if (error)
225 goto out_trans_cancel;
226
227 xfs_trans_ijoin(tp, ip, 0);
228
229 /*
230 * From this point onwards we overwrite the imap pointer that the
231 * caller gave to us.
232 */
233 xfs_bmap_init(&free_list, &firstfsb);
234 nimaps = 1;
235 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
236 bmapi_flags, &firstfsb, resblks, imap,
237 &nimaps, &free_list);
238 if (error)
239 goto out_bmap_cancel;
240
241 /*
242 * Complete the transaction
243 */
244 error = xfs_bmap_finish(&tp, &free_list, NULL);
245 if (error)
246 goto out_bmap_cancel;
247
248 error = xfs_trans_commit(tp);
249 if (error)
250 goto out_unlock;
251
252 /*
253 * Copy any maps to caller's array and return any error.
254 */
255 if (nimaps == 0) {
256 error = -ENOSPC;
257 goto out_unlock;
258 }
259
260 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
261 error = xfs_alert_fsblock_zero(ip, imap);
262
263 out_unlock:
264 xfs_iunlock(ip, lockmode);
265 return error;
266
267 out_bmap_cancel:
268 xfs_bmap_cancel(&free_list);
269 xfs_trans_unreserve_quota_nblks(tp, ip, (long)qblocks, 0, quota_flag);
270 out_trans_cancel:
271 xfs_trans_cancel(tp);
272 goto out_unlock;
273 }
274
275 /*
276 * If the caller is doing a write at the end of the file, then extend the
277 * allocation out to the file system's write iosize. We clean up any extra
278 * space left over when the file is closed in xfs_inactive().
279 *
280 * If we find we already have delalloc preallocation beyond EOF, don't do more
281 * preallocation as it it not needed.
282 */
283 STATIC int
284 xfs_iomap_eof_want_preallocate(
285 xfs_mount_t *mp,
286 xfs_inode_t *ip,
287 xfs_off_t offset,
288 size_t count,
289 xfs_bmbt_irec_t *imap,
290 int nimaps,
291 int *prealloc)
292 {
293 xfs_fileoff_t start_fsb;
294 xfs_filblks_t count_fsb;
295 int n, error, imaps;
296 int found_delalloc = 0;
297
298 *prealloc = 0;
299 if (offset + count <= XFS_ISIZE(ip))
300 return 0;
301
302 /*
303 * If the file is smaller than the minimum prealloc and we are using
304 * dynamic preallocation, don't do any preallocation at all as it is
305 * likely this is the only write to the file that is going to be done.
306 */
307 if (!(mp->m_flags & XFS_MOUNT_DFLT_IOSIZE) &&
308 XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_writeio_blocks))
309 return 0;
310
311 /*
312 * If there are any real blocks past eof, then don't
313 * do any speculative allocation.
314 */
315 start_fsb = XFS_B_TO_FSBT(mp, ((xfs_ufsize_t)(offset + count - 1)));
316 count_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
317 while (count_fsb > 0) {
318 imaps = nimaps;
319 error = xfs_bmapi_read(ip, start_fsb, count_fsb, imap, &imaps,
320 0);
321 if (error)
322 return error;
323 for (n = 0; n < imaps; n++) {
324 if ((imap[n].br_startblock != HOLESTARTBLOCK) &&
325 (imap[n].br_startblock != DELAYSTARTBLOCK))
326 return 0;
327 start_fsb += imap[n].br_blockcount;
328 count_fsb -= imap[n].br_blockcount;
329
330 if (imap[n].br_startblock == DELAYSTARTBLOCK)
331 found_delalloc = 1;
332 }
333 }
334 if (!found_delalloc)
335 *prealloc = 1;
336 return 0;
337 }
338
339 /*
340 * Determine the initial size of the preallocation. We are beyond the current
341 * EOF here, but we need to take into account whether this is a sparse write or
342 * an extending write when determining the preallocation size. Hence we need to
343 * look up the extent that ends at the current write offset and use the result
344 * to determine the preallocation size.
345 *
346 * If the extent is a hole, then preallocation is essentially disabled.
347 * Otherwise we take the size of the preceeding data extent as the basis for the
348 * preallocation size. If the size of the extent is greater than half the
349 * maximum extent length, then use the current offset as the basis. This ensures
350 * that for large files the preallocation size always extends to MAXEXTLEN
351 * rather than falling short due to things like stripe unit/width alignment of
352 * real extents.
353 */
354 STATIC xfs_fsblock_t
355 xfs_iomap_eof_prealloc_initial_size(
356 struct xfs_mount *mp,
357 struct xfs_inode *ip,
358 xfs_off_t offset,
359 xfs_bmbt_irec_t *imap,
360 int nimaps)
361 {
362 xfs_fileoff_t start_fsb;
363 int imaps = 1;
364 int error;
365
366 ASSERT(nimaps >= imaps);
367
368 /* if we are using a specific prealloc size, return now */
369 if (mp->m_flags & XFS_MOUNT_DFLT_IOSIZE)
370 return 0;
371
372 /* If the file is small, then use the minimum prealloc */
373 if (XFS_ISIZE(ip) < XFS_FSB_TO_B(mp, mp->m_dalign))
374 return 0;
375
376 /*
377 * As we write multiple pages, the offset will always align to the
378 * start of a page and hence point to a hole at EOF. i.e. if the size is
379 * 4096 bytes, we only have one block at FSB 0, but XFS_B_TO_FSB(4096)
380 * will return FSB 1. Hence if there are blocks in the file, we want to
381 * point to the block prior to the EOF block and not the hole that maps
382 * directly at @offset.
383 */
384 start_fsb = XFS_B_TO_FSB(mp, offset);
385 if (start_fsb)
386 start_fsb--;
387 error = xfs_bmapi_read(ip, start_fsb, 1, imap, &imaps, XFS_BMAPI_ENTIRE);
388 if (error)
389 return 0;
390
391 ASSERT(imaps == 1);
392 if (imap[0].br_startblock == HOLESTARTBLOCK)
393 return 0;
394 if (imap[0].br_blockcount <= (MAXEXTLEN >> 1))
395 return imap[0].br_blockcount << 1;
396 return XFS_B_TO_FSB(mp, offset);
397 }
398
399 STATIC bool
400 xfs_quota_need_throttle(
401 struct xfs_inode *ip,
402 int type,
403 xfs_fsblock_t alloc_blocks)
404 {
405 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
406
407 if (!dq || !xfs_this_quota_on(ip->i_mount, type))
408 return false;
409
410 /* no hi watermark, no throttle */
411 if (!dq->q_prealloc_hi_wmark)
412 return false;
413
414 /* under the lo watermark, no throttle */
415 if (dq->q_res_bcount + alloc_blocks < dq->q_prealloc_lo_wmark)
416 return false;
417
418 return true;
419 }
420
421 STATIC void
422 xfs_quota_calc_throttle(
423 struct xfs_inode *ip,
424 int type,
425 xfs_fsblock_t *qblocks,
426 int *qshift,
427 int64_t *qfreesp)
428 {
429 int64_t freesp;
430 int shift = 0;
431 struct xfs_dquot *dq = xfs_inode_dquot(ip, type);
432
433 /* no dq, or over hi wmark, squash the prealloc completely */
434 if (!dq || dq->q_res_bcount >= dq->q_prealloc_hi_wmark) {
435 *qblocks = 0;
436 *qfreesp = 0;
437 return;
438 }
439
440 freesp = dq->q_prealloc_hi_wmark - dq->q_res_bcount;
441 if (freesp < dq->q_low_space[XFS_QLOWSP_5_PCNT]) {
442 shift = 2;
443 if (freesp < dq->q_low_space[XFS_QLOWSP_3_PCNT])
444 shift += 2;
445 if (freesp < dq->q_low_space[XFS_QLOWSP_1_PCNT])
446 shift += 2;
447 }
448
449 if (freesp < *qfreesp)
450 *qfreesp = freesp;
451
452 /* only overwrite the throttle values if we are more aggressive */
453 if ((freesp >> shift) < (*qblocks >> *qshift)) {
454 *qblocks = freesp;
455 *qshift = shift;
456 }
457 }
458
459 /*
460 * If we don't have a user specified preallocation size, dynamically increase
461 * the preallocation size as the size of the file grows. Cap the maximum size
462 * at a single extent or less if the filesystem is near full. The closer the
463 * filesystem is to full, the smaller the maximum prealocation.
464 */
465 STATIC xfs_fsblock_t
466 xfs_iomap_prealloc_size(
467 struct xfs_mount *mp,
468 struct xfs_inode *ip,
469 xfs_off_t offset,
470 struct xfs_bmbt_irec *imap,
471 int nimaps)
472 {
473 xfs_fsblock_t alloc_blocks = 0;
474 int shift = 0;
475 int64_t freesp;
476 xfs_fsblock_t qblocks;
477 int qshift = 0;
478
479 alloc_blocks = xfs_iomap_eof_prealloc_initial_size(mp, ip, offset,
480 imap, nimaps);
481 if (!alloc_blocks)
482 goto check_writeio;
483 qblocks = alloc_blocks;
484
485 /*
486 * MAXEXTLEN is not a power of two value but we round the prealloc down
487 * to the nearest power of two value after throttling. To prevent the
488 * round down from unconditionally reducing the maximum supported prealloc
489 * size, we round up first, apply appropriate throttling, round down and
490 * cap the value to MAXEXTLEN.
491 */
492 alloc_blocks = XFS_FILEOFF_MIN(roundup_pow_of_two(MAXEXTLEN),
493 alloc_blocks);
494
495 freesp = percpu_counter_read_positive(&mp->m_fdblocks);
496 if (freesp < mp->m_low_space[XFS_LOWSP_5_PCNT]) {
497 shift = 2;
498 if (freesp < mp->m_low_space[XFS_LOWSP_4_PCNT])
499 shift++;
500 if (freesp < mp->m_low_space[XFS_LOWSP_3_PCNT])
501 shift++;
502 if (freesp < mp->m_low_space[XFS_LOWSP_2_PCNT])
503 shift++;
504 if (freesp < mp->m_low_space[XFS_LOWSP_1_PCNT])
505 shift++;
506 }
507
508 /*
509 * Check each quota to cap the prealloc size, provide a shift value to
510 * throttle with and adjust amount of available space.
511 */
512 if (xfs_quota_need_throttle(ip, XFS_DQ_USER, alloc_blocks))
513 xfs_quota_calc_throttle(ip, XFS_DQ_USER, &qblocks, &qshift,
514 &freesp);
515 if (xfs_quota_need_throttle(ip, XFS_DQ_GROUP, alloc_blocks))
516 xfs_quota_calc_throttle(ip, XFS_DQ_GROUP, &qblocks, &qshift,
517 &freesp);
518 if (xfs_quota_need_throttle(ip, XFS_DQ_PROJ, alloc_blocks))
519 xfs_quota_calc_throttle(ip, XFS_DQ_PROJ, &qblocks, &qshift,
520 &freesp);
521
522 /*
523 * The final prealloc size is set to the minimum of free space available
524 * in each of the quotas and the overall filesystem.
525 *
526 * The shift throttle value is set to the maximum value as determined by
527 * the global low free space values and per-quota low free space values.
528 */
529 alloc_blocks = MIN(alloc_blocks, qblocks);
530 shift = MAX(shift, qshift);
531
532 if (shift)
533 alloc_blocks >>= shift;
534 /*
535 * rounddown_pow_of_two() returns an undefined result if we pass in
536 * alloc_blocks = 0.
537 */
538 if (alloc_blocks)
539 alloc_blocks = rounddown_pow_of_two(alloc_blocks);
540 if (alloc_blocks > MAXEXTLEN)
541 alloc_blocks = MAXEXTLEN;
542
543 /*
544 * If we are still trying to allocate more space than is
545 * available, squash the prealloc hard. This can happen if we
546 * have a large file on a small filesystem and the above
547 * lowspace thresholds are smaller than MAXEXTLEN.
548 */
549 while (alloc_blocks && alloc_blocks >= freesp)
550 alloc_blocks >>= 4;
551
552 check_writeio:
553 if (alloc_blocks < mp->m_writeio_blocks)
554 alloc_blocks = mp->m_writeio_blocks;
555
556 trace_xfs_iomap_prealloc_size(ip, alloc_blocks, shift,
557 mp->m_writeio_blocks);
558
559 return alloc_blocks;
560 }
561
562 int
563 xfs_iomap_write_delay(
564 xfs_inode_t *ip,
565 xfs_off_t offset,
566 size_t count,
567 xfs_bmbt_irec_t *ret_imap)
568 {
569 xfs_mount_t *mp = ip->i_mount;
570 xfs_fileoff_t offset_fsb;
571 xfs_fileoff_t last_fsb;
572 xfs_off_t aligned_offset;
573 xfs_fileoff_t ioalign;
574 xfs_extlen_t extsz;
575 int nimaps;
576 xfs_bmbt_irec_t imap[XFS_WRITE_IMAPS];
577 int prealloc;
578 int error;
579
580 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
581
582 /*
583 * Make sure that the dquots are there. This doesn't hold
584 * the ilock across a disk read.
585 */
586 error = xfs_qm_dqattach_locked(ip, 0);
587 if (error)
588 return error;
589
590 extsz = xfs_get_extsz_hint(ip);
591 offset_fsb = XFS_B_TO_FSBT(mp, offset);
592
593 error = xfs_iomap_eof_want_preallocate(mp, ip, offset, count,
594 imap, XFS_WRITE_IMAPS, &prealloc);
595 if (error)
596 return error;
597
598 retry:
599 if (prealloc) {
600 xfs_fsblock_t alloc_blocks;
601
602 alloc_blocks = xfs_iomap_prealloc_size(mp, ip, offset, imap,
603 XFS_WRITE_IMAPS);
604
605 aligned_offset = XFS_WRITEIO_ALIGN(mp, (offset + count - 1));
606 ioalign = XFS_B_TO_FSBT(mp, aligned_offset);
607 last_fsb = ioalign + alloc_blocks;
608 } else {
609 last_fsb = XFS_B_TO_FSB(mp, ((xfs_ufsize_t)(offset + count)));
610 }
611
612 if (prealloc || extsz) {
613 error = xfs_iomap_eof_align_last_fsb(mp, ip, extsz, &last_fsb);
614 if (error)
615 return error;
616 }
617
618 /*
619 * Make sure preallocation does not create extents beyond the range we
620 * actually support in this filesystem.
621 */
622 if (last_fsb > XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes))
623 last_fsb = XFS_B_TO_FSB(mp, mp->m_super->s_maxbytes);
624
625 ASSERT(last_fsb > offset_fsb);
626
627 nimaps = XFS_WRITE_IMAPS;
628 error = xfs_bmapi_delay(ip, offset_fsb, last_fsb - offset_fsb,
629 imap, &nimaps, XFS_BMAPI_ENTIRE);
630 switch (error) {
631 case 0:
632 case -ENOSPC:
633 case -EDQUOT:
634 break;
635 default:
636 return error;
637 }
638
639 /*
640 * If bmapi returned us nothing, we got either ENOSPC or EDQUOT. Retry
641 * without EOF preallocation.
642 */
643 if (nimaps == 0) {
644 trace_xfs_delalloc_enospc(ip, offset, count);
645 if (prealloc) {
646 prealloc = 0;
647 error = 0;
648 goto retry;
649 }
650 return error ? error : -ENOSPC;
651 }
652
653 if (!(imap[0].br_startblock || XFS_IS_REALTIME_INODE(ip)))
654 return xfs_alert_fsblock_zero(ip, &imap[0]);
655
656 /*
657 * Tag the inode as speculatively preallocated so we can reclaim this
658 * space on demand, if necessary.
659 */
660 if (prealloc)
661 xfs_inode_set_eofblocks_tag(ip);
662
663 *ret_imap = imap[0];
664 return 0;
665 }
666
667 /*
668 * Pass in a delayed allocate extent, convert it to real extents;
669 * return to the caller the extent we create which maps on top of
670 * the originating callers request.
671 *
672 * Called without a lock on the inode.
673 *
674 * We no longer bother to look at the incoming map - all we have to
675 * guarantee is that whatever we allocate fills the required range.
676 */
677 int
678 xfs_iomap_write_allocate(
679 xfs_inode_t *ip,
680 xfs_off_t offset,
681 xfs_bmbt_irec_t *imap)
682 {
683 xfs_mount_t *mp = ip->i_mount;
684 xfs_fileoff_t offset_fsb, last_block;
685 xfs_fileoff_t end_fsb, map_start_fsb;
686 xfs_fsblock_t first_block;
687 xfs_bmap_free_t free_list;
688 xfs_filblks_t count_fsb;
689 xfs_trans_t *tp;
690 int nimaps;
691 int error = 0;
692 int nres;
693
694 /*
695 * Make sure that the dquots are there.
696 */
697 error = xfs_qm_dqattach(ip, 0);
698 if (error)
699 return error;
700
701 offset_fsb = XFS_B_TO_FSBT(mp, offset);
702 count_fsb = imap->br_blockcount;
703 map_start_fsb = imap->br_startoff;
704
705 XFS_STATS_ADD(mp, xs_xstrat_bytes, XFS_FSB_TO_B(mp, count_fsb));
706
707 while (count_fsb != 0) {
708 /*
709 * Set up a transaction with which to allocate the
710 * backing store for the file. Do allocations in a
711 * loop until we get some space in the range we are
712 * interested in. The other space that might be allocated
713 * is in the delayed allocation extent on which we sit
714 * but before our buffer starts.
715 */
716
717 nimaps = 0;
718 while (nimaps == 0) {
719 nres = XFS_EXTENTADD_SPACE_RES(mp, XFS_DATA_FORK);
720
721 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, nres,
722 0, XFS_TRANS_RESERVE, &tp);
723 if (error)
724 return error;
725
726 xfs_ilock(ip, XFS_ILOCK_EXCL);
727 xfs_trans_ijoin(tp, ip, 0);
728
729 xfs_bmap_init(&free_list, &first_block);
730
731 /*
732 * it is possible that the extents have changed since
733 * we did the read call as we dropped the ilock for a
734 * while. We have to be careful about truncates or hole
735 * punchs here - we are not allowed to allocate
736 * non-delalloc blocks here.
737 *
738 * The only protection against truncation is the pages
739 * for the range we are being asked to convert are
740 * locked and hence a truncate will block on them
741 * first.
742 *
743 * As a result, if we go beyond the range we really
744 * need and hit an delalloc extent boundary followed by
745 * a hole while we have excess blocks in the map, we
746 * will fill the hole incorrectly and overrun the
747 * transaction reservation.
748 *
749 * Using a single map prevents this as we are forced to
750 * check each map we look for overlap with the desired
751 * range and abort as soon as we find it. Also, given
752 * that we only return a single map, having one beyond
753 * what we can return is probably a bit silly.
754 *
755 * We also need to check that we don't go beyond EOF;
756 * this is a truncate optimisation as a truncate sets
757 * the new file size before block on the pages we
758 * currently have locked under writeback. Because they
759 * are about to be tossed, we don't need to write them
760 * back....
761 */
762 nimaps = 1;
763 end_fsb = XFS_B_TO_FSB(mp, XFS_ISIZE(ip));
764 error = xfs_bmap_last_offset(ip, &last_block,
765 XFS_DATA_FORK);
766 if (error)
767 goto trans_cancel;
768
769 last_block = XFS_FILEOFF_MAX(last_block, end_fsb);
770 if ((map_start_fsb + count_fsb) > last_block) {
771 count_fsb = last_block - map_start_fsb;
772 if (count_fsb == 0) {
773 error = -EAGAIN;
774 goto trans_cancel;
775 }
776 }
777
778 /*
779 * From this point onwards we overwrite the imap
780 * pointer that the caller gave to us.
781 */
782 error = xfs_bmapi_write(tp, ip, map_start_fsb,
783 count_fsb, 0, &first_block,
784 nres, imap, &nimaps,
785 &free_list);
786 if (error)
787 goto trans_cancel;
788
789 error = xfs_bmap_finish(&tp, &free_list, NULL);
790 if (error)
791 goto trans_cancel;
792
793 error = xfs_trans_commit(tp);
794 if (error)
795 goto error0;
796
797 xfs_iunlock(ip, XFS_ILOCK_EXCL);
798 }
799
800 /*
801 * See if we were able to allocate an extent that
802 * covers at least part of the callers request
803 */
804 if (!(imap->br_startblock || XFS_IS_REALTIME_INODE(ip)))
805 return xfs_alert_fsblock_zero(ip, imap);
806
807 if ((offset_fsb >= imap->br_startoff) &&
808 (offset_fsb < (imap->br_startoff +
809 imap->br_blockcount))) {
810 XFS_STATS_INC(mp, xs_xstrat_quick);
811 return 0;
812 }
813
814 /*
815 * So far we have not mapped the requested part of the
816 * file, just surrounding data, try again.
817 */
818 count_fsb -= imap->br_blockcount;
819 map_start_fsb = imap->br_startoff + imap->br_blockcount;
820 }
821
822 trans_cancel:
823 xfs_bmap_cancel(&free_list);
824 xfs_trans_cancel(tp);
825 error0:
826 xfs_iunlock(ip, XFS_ILOCK_EXCL);
827 return error;
828 }
829
830 int
831 xfs_iomap_write_unwritten(
832 xfs_inode_t *ip,
833 xfs_off_t offset,
834 xfs_off_t count)
835 {
836 xfs_mount_t *mp = ip->i_mount;
837 xfs_fileoff_t offset_fsb;
838 xfs_filblks_t count_fsb;
839 xfs_filblks_t numblks_fsb;
840 xfs_fsblock_t firstfsb;
841 int nimaps;
842 xfs_trans_t *tp;
843 xfs_bmbt_irec_t imap;
844 xfs_bmap_free_t free_list;
845 xfs_fsize_t i_size;
846 uint resblks;
847 int error;
848
849 trace_xfs_unwritten_convert(ip, offset, count);
850
851 offset_fsb = XFS_B_TO_FSBT(mp, offset);
852 count_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
853 count_fsb = (xfs_filblks_t)(count_fsb - offset_fsb);
854
855 /*
856 * Reserve enough blocks in this transaction for two complete extent
857 * btree splits. We may be converting the middle part of an unwritten
858 * extent and in this case we will insert two new extents in the btree
859 * each of which could cause a full split.
860 *
861 * This reservation amount will be used in the first call to
862 * xfs_bmbt_split() to select an AG with enough space to satisfy the
863 * rest of the operation.
864 */
865 resblks = XFS_DIOSTRAT_SPACE_RES(mp, 0) << 1;
866
867 do {
868 /*
869 * Set up a transaction to convert the range of extents
870 * from unwritten to real. Do allocations in a loop until
871 * we have covered the range passed in.
872 *
873 * Note that we can't risk to recursing back into the filesystem
874 * here as we might be asked to write out the same inode that we
875 * complete here and might deadlock on the iolock.
876 */
877 error = xfs_trans_alloc(mp, &M_RES(mp)->tr_write, resblks, 0,
878 XFS_TRANS_RESERVE | XFS_TRANS_NOFS, &tp);
879 if (error)
880 return error;
881
882 xfs_ilock(ip, XFS_ILOCK_EXCL);
883 xfs_trans_ijoin(tp, ip, 0);
884
885 /*
886 * Modify the unwritten extent state of the buffer.
887 */
888 xfs_bmap_init(&free_list, &firstfsb);
889 nimaps = 1;
890 error = xfs_bmapi_write(tp, ip, offset_fsb, count_fsb,
891 XFS_BMAPI_CONVERT, &firstfsb, resblks,
892 &imap, &nimaps, &free_list);
893 if (error)
894 goto error_on_bmapi_transaction;
895
896 /*
897 * Log the updated inode size as we go. We have to be careful
898 * to only log it up to the actual write offset if it is
899 * halfway into a block.
900 */
901 i_size = XFS_FSB_TO_B(mp, offset_fsb + count_fsb);
902 if (i_size > offset + count)
903 i_size = offset + count;
904
905 i_size = xfs_new_eof(ip, i_size);
906 if (i_size) {
907 ip->i_d.di_size = i_size;
908 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
909 }
910
911 error = xfs_bmap_finish(&tp, &free_list, NULL);
912 if (error)
913 goto error_on_bmapi_transaction;
914
915 error = xfs_trans_commit(tp);
916 xfs_iunlock(ip, XFS_ILOCK_EXCL);
917 if (error)
918 return error;
919
920 if (!(imap.br_startblock || XFS_IS_REALTIME_INODE(ip)))
921 return xfs_alert_fsblock_zero(ip, &imap);
922
923 if ((numblks_fsb = imap.br_blockcount) == 0) {
924 /*
925 * The numblks_fsb value should always get
926 * smaller, otherwise the loop is stuck.
927 */
928 ASSERT(imap.br_blockcount);
929 break;
930 }
931 offset_fsb += numblks_fsb;
932 count_fsb -= numblks_fsb;
933 } while (count_fsb > 0);
934
935 return 0;
936
937 error_on_bmapi_transaction:
938 xfs_bmap_cancel(&free_list);
939 xfs_trans_cancel(tp);
940 xfs_iunlock(ip, XFS_ILOCK_EXCL);
941 return error;
942 }
Linux kernel - Deliverables
This page took 0.13975 seconds and 5 git commands to generate.