?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge with Linux 2.5.59.
[linux-2.6/linux-mips.git] / fs / xfs / linux / xfs_lrw.c
blobd08561facca09a900cdb1b14dbeabc0398888172
1 /*
2 * Copyright (c) 2000-2002 Silicon Graphics, Inc. All Rights Reserved.
3 *
4 * This program is free software; you can redistribute it and/or modify it
5 * under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it would be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
11 *
12 * Further, this software is distributed without any warranty that it is
13 * free of the rightful claim of any third person regarding infringement
14 * or the like. Any license provided herein, whether implied or
15 * otherwise, applies only to this software file. Patent licenses, if
16 * any, provided herein do not apply to combinations of this program with
17 * other software, or any other product whatsoever.
18 *
19 * You should have received a copy of the GNU General Public License along
20 * with this program; if not, write the Free Software Foundation, Inc., 59
21 * Temple Place - Suite 330, Boston MA 02111-1307, USA.
22 *
23 * Contact information: Silicon Graphics, Inc., 1600 Amphitheatre Pkwy,
24 * Mountain View, CA 94043, or:
25 *
26 * http://www.sgi.com
27 *
28 * For further information regarding this notice, see:
29 *
30 * http://oss.sgi.com/projects/GenInfo/SGIGPLNoticeExplan/
31 */
32 /*
33 * fs/xfs/linux/xfs_lrw.c (Linux Read Write stuff)
34 *
35 */
36
37 #include <xfs.h>
38 #include <linux/pagemap.h>
39 #include <linux/capability.h>
40
41
42 /*
43 * xfs_iozero
44 *
45 * xfs_iozero clears the specified range of buffer supplied,
46 * and marks all the affected blocks as valid and modified. If
47 * an affected block is not allocated, it will be allocated. If
48 * an affected block is not completely overwritten, and is not
49 * valid before the operation, it will be read from disk before
50 * being partially zeroed.
51 */
52 STATIC int
53 xfs_iozero(
54 struct inode *ip, /* inode */
55 loff_t pos, /* offset in file */
56 size_t count, /* size of data to zero */
57 loff_t end_size) /* max file size to set */
58 {
59 unsigned bytes;
60 struct page *page;
61 struct address_space *mapping;
62 char *kaddr;
63 int status;
64
65 mapping = ip->i_mapping;
66 do {
67 unsigned long index, offset;
68
69 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
70 index = pos >> PAGE_CACHE_SHIFT;
71 bytes = PAGE_CACHE_SIZE - offset;
72 if (bytes > count)
73 bytes = count;
74
75 status = -ENOMEM;
76 page = grab_cache_page(mapping, index);
77 if (!page)
78 break;
79
80 kaddr = kmap(page);
81 status = mapping->a_ops->prepare_write(NULL, page, offset,
82 offset + bytes);
83 if (status) {
84 goto unlock;
85 }
86
87 memset((void *) (kaddr + offset), 0, bytes);
88 flush_dcache_page(page);
89 status = mapping->a_ops->commit_write(NULL, page, offset,
90 offset + bytes);
91 if (!status) {
92 pos += bytes;
93 count -= bytes;
94 if (pos > ip->i_size)
95 ip->i_size = pos < end_size ? pos : end_size;
96 }
97
98 unlock:
99 kunmap(page);
100 unlock_page(page);
101 page_cache_release(page);
102 if (status)
103 break;
104 } while (count);
105
106 return (-status);
107 }
108
109 ssize_t /* bytes read, or (-) error */
110 xfs_read(
111 bhv_desc_t *bdp,
112 struct file *filp,
113 const struct iovec *iovp,
114 unsigned long segs,
115 loff_t *offp,
116 cred_t *credp)
117 {
118 size_t size = 0;
119 ssize_t ret;
120 xfs_fsize_t n;
121 xfs_inode_t *ip;
122 xfs_mount_t *mp;
123 vnode_t *vp;
124 unsigned long seg;
125 int direct = filp->f_flags & O_DIRECT;
126
127 ip = XFS_BHVTOI(bdp);
128 vp = BHV_TO_VNODE(bdp);
129 mp = ip->i_mount;
130 vn_trace_entry(vp, "xfs_read", (inst_t *)__return_address);
131
132 XFS_STATS_INC(xfsstats.xs_read_calls);
133
134 /* START copy & waste from filemap.c */
135 for (seg = 0; seg < segs; seg++) {
136 const struct iovec *iv = &iovp[seg];
137
138 /*
139 * If any segment has a negative length, or the cumulative
140 * length ever wraps negative then return -EINVAL.
141 */
142 size += iv->iov_len;
143 if (unlikely((ssize_t)(size|iv->iov_len) < 0))
144 return XFS_ERROR(-EINVAL);
145 if (direct) { /* XFS specific check */
146 if ((__psint_t)iv->iov_base & BBMASK) {
147 if (*offp == ip->i_d.di_size)
148 return 0;
149 return XFS_ERROR(-EINVAL);
150 }
151 }
152 if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len))
153 continue;
154 if (seg == 0)
155 return XFS_ERROR(-EFAULT);
156 segs = seg;
157 break;
158 }
159 /* END copy & waste from filemap.c */
160
161 if (direct) {
162 if ((*offp & mp->m_blockmask) ||
163 (size & mp->m_blockmask)) {
164 if (*offp == ip->i_d.di_size) {
165 return (0);
166 }
167 return -XFS_ERROR(EINVAL);
168 }
169 }
170
171 n = XFS_MAX_FILE_OFFSET - *offp;
172 if ((n <= 0) || (size == 0))
173 return 0;
174
175 if (n < size)
176 size = n;
177
178 if (XFS_FORCED_SHUTDOWN(mp)) {
179 return -EIO;
180 }
181
182 xfs_ilock(ip, XFS_IOLOCK_SHARED);
183
184 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ) &&
185 !(filp->f_mode & FINVIS)) {
186 int error;
187 vrwlock_t locktype = VRWLOCK_READ;
188
189 error = xfs_dm_send_data_event(DM_EVENT_READ, bdp, *offp,
190 size, FILP_DELAY_FLAG(filp), &locktype);
191 if (error) {
192 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
193 return -error;
194 }
195 }
196
197 ret = generic_file_readv(filp, iovp, segs, offp);
198 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
199
200 XFS_STATS_ADD(xfsstats.xs_read_bytes, ret);
201
202 if (!(filp->f_mode & FINVIS))
203 xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
204
205 return ret;
206 }
207
208 ssize_t
209 xfs_sendfile(
210 bhv_desc_t *bdp,
211 struct file *filp,
212 loff_t *offp,
213 size_t count,
214 read_actor_t actor,
215 void *target,
216 cred_t *credp)
217 {
218 ssize_t ret;
219 xfs_fsize_t n;
220 xfs_inode_t *ip;
221 vnode_t *vp;
222 int invisible = (filp->f_mode & FINVIS);
223
224 ip = XFS_BHVTOI(bdp);
225 vp = BHV_TO_VNODE(bdp);
226 vn_trace_entry(vp, "xfs_sendfile", (inst_t *)__return_address);
227
228 XFS_STATS_INC(xfsstats.xs_read_calls);
229
230 n = XFS_MAX_FILE_OFFSET - *offp;
231 if ((n <= 0) || (count == 0))
232 return 0;
233
234 if (n < count)
235 count = n;
236
237 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
238 return -EIO;
239
240 xfs_ilock(ip, XFS_IOLOCK_SHARED);
241 if (DM_EVENT_ENABLED(vp->v_vfsp, ip, DM_EVENT_READ) && !invisible) {
242 vrwlock_t locktype = VRWLOCK_READ;
243 int error;
244
245 error = xfs_dm_send_data_event(DM_EVENT_READ, bdp, *offp,
246 count, FILP_DELAY_FLAG(filp), &locktype);
247 if (error) {
248 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
249 return -error;
250 }
251 }
252 ret = generic_file_sendfile(filp, offp, count, actor, target);
253 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
254
255 XFS_STATS_ADD(xfsstats.xs_read_bytes, ret);
256 if (!invisible)
257 xfs_ichgtime(ip, XFS_ICHGTIME_ACC);
258 return ret;
259 }
260
261 /*
262 * This routine is called to handle zeroing any space in the last
263 * block of the file that is beyond the EOF. We do this since the
264 * size is being increased without writing anything to that block
265 * and we don't want anyone to read the garbage on the disk.
266 */
267 STATIC int /* error (positive) */
268 xfs_zero_last_block(
269 struct inode *ip,
270 xfs_iocore_t *io,
271 xfs_off_t offset,
272 xfs_fsize_t isize,
273 xfs_fsize_t end_size)
274 {
275 xfs_fileoff_t last_fsb;
276 xfs_mount_t *mp;
277 int nimaps;
278 int zero_offset;
279 int zero_len;
280 int isize_fsb_offset;
281 int error = 0;
282 xfs_bmbt_irec_t imap;
283 loff_t loff;
284 size_t lsize;
285
286 ASSERT(ismrlocked(io->io_lock, MR_UPDATE) != 0);
287 ASSERT(offset > isize);
288
289 mp = io->io_mount;
290
291 isize_fsb_offset = XFS_B_FSB_OFFSET(mp, isize);
292 if (isize_fsb_offset == 0) {
293 /*
294 * There are no extra bytes in the last block on disk to
295 * zero, so return.
296 */
297 return 0;
298 }
299
300 last_fsb = XFS_B_TO_FSBT(mp, isize);
301 nimaps = 1;
302 error = XFS_BMAPI(mp, NULL, io, last_fsb, 1, 0, NULL, 0, &imap,
303 &nimaps, NULL);
304 if (error) {
305 return error;
306 }
307 ASSERT(nimaps > 0);
308 /*
309 * If the block underlying isize is just a hole, then there
310 * is nothing to zero.
311 */
312 if (imap.br_startblock == HOLESTARTBLOCK) {
313 return 0;
314 }
315 /*
316 * Zero the part of the last block beyond the EOF, and write it
317 * out sync. We need to drop the ilock while we do this so we
318 * don't deadlock when the buffer cache calls back to us.
319 */
320 XFS_IUNLOCK(mp, io, XFS_ILOCK_EXCL| XFS_EXTSIZE_RD);
321 loff = XFS_FSB_TO_B(mp, last_fsb);
322 lsize = XFS_FSB_TO_B(mp, 1);
323
324 zero_offset = isize_fsb_offset;
325 zero_len = mp->m_sb.sb_blocksize - isize_fsb_offset;
326
327 error = xfs_iozero(ip, loff + zero_offset, zero_len, end_size);
328
329 XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
330 ASSERT(error >= 0);
331 return error;
332 }
333
334 /*
335 * Zero any on disk space between the current EOF and the new,
336 * larger EOF. This handles the normal case of zeroing the remainder
337 * of the last block in the file and the unusual case of zeroing blocks
338 * out beyond the size of the file. This second case only happens
339 * with fixed size extents and when the system crashes before the inode
340 * size was updated but after blocks were allocated. If fill is set,
341 * then any holes in the range are filled and zeroed. If not, the holes
342 * are left alone as holes.
343 */
344
345 int /* error (positive) */
346 xfs_zero_eof(
347 vnode_t *vp,
348 xfs_iocore_t *io,
349 xfs_off_t offset, /* starting I/O offset */
350 xfs_fsize_t isize, /* current inode size */
351 xfs_fsize_t end_size) /* terminal inode size */
352 {
353 struct inode *ip = LINVFS_GET_IP(vp);
354 xfs_fileoff_t start_zero_fsb;
355 xfs_fileoff_t end_zero_fsb;
356 xfs_fileoff_t prev_zero_fsb;
357 xfs_fileoff_t zero_count_fsb;
358 xfs_fileoff_t last_fsb;
359 xfs_extlen_t buf_len_fsb;
360 xfs_extlen_t prev_zero_count;
361 xfs_mount_t *mp;
362 int nimaps;
363 int error = 0;
364 xfs_bmbt_irec_t imap;
365 loff_t loff;
366 size_t lsize;
367
368 ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
369 ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
370
371 mp = io->io_mount;
372
373 /*
374 * First handle zeroing the block on which isize resides.
375 * We only zero a part of that block so it is handled specially.
376 */
377 error = xfs_zero_last_block(ip, io, offset, isize, end_size);
378 if (error) {
379 ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
380 ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
381 return error;
382 }
383
384 /*
385 * Calculate the range between the new size and the old
386 * where blocks needing to be zeroed may exist. To get the
387 * block where the last byte in the file currently resides,
388 * we need to subtract one from the size and truncate back
389 * to a block boundary. We subtract 1 in case the size is
390 * exactly on a block boundary.
391 */
392 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
393 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
394 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
395 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
396 if (last_fsb == end_zero_fsb) {
397 /*
398 * The size was only incremented on its last block.
399 * We took care of that above, so just return.
400 */
401 return 0;
402 }
403
404 ASSERT(start_zero_fsb <= end_zero_fsb);
405 prev_zero_fsb = NULLFILEOFF;
406 prev_zero_count = 0;
407 while (start_zero_fsb <= end_zero_fsb) {
408 nimaps = 1;
409 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
410 error = XFS_BMAPI(mp, NULL, io, start_zero_fsb, zero_count_fsb,
411 0, NULL, 0, &imap, &nimaps, NULL);
412 if (error) {
413 ASSERT(ismrlocked(io->io_lock, MR_UPDATE));
414 ASSERT(ismrlocked(io->io_iolock, MR_UPDATE));
415 return error;
416 }
417 ASSERT(nimaps > 0);
418
419 if (imap.br_startblock == HOLESTARTBLOCK) {
420 /*
421 * This loop handles initializing pages that were
422 * partially initialized by the code below this
423 * loop. It basically zeroes the part of the page
424 * that sits on a hole and sets the page as P_HOLE
425 * and calls remapf if it is a mapped file.
426 */
427 prev_zero_fsb = NULLFILEOFF;
428 prev_zero_count = 0;
429 start_zero_fsb = imap.br_startoff +
430 imap.br_blockcount;
431 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
432 continue;
433 }
434
435 /*
436 * There are blocks in the range requested.
437 * Zero them a single write at a time. We actually
438 * don't zero the entire range returned if it is
439 * too big and simply loop around to get the rest.
440 * That is not the most efficient thing to do, but it
441 * is simple and this path should not be exercised often.
442 */
443 buf_len_fsb = XFS_FILBLKS_MIN(imap.br_blockcount,
444 mp->m_writeio_blocks << 8);
445 /*
446 * Drop the inode lock while we're doing the I/O.
447 * We'll still have the iolock to protect us.
448 */
449 XFS_IUNLOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
450
451 loff = XFS_FSB_TO_B(mp, start_zero_fsb);
452 lsize = XFS_FSB_TO_B(mp, buf_len_fsb);
453
454 error = xfs_iozero(ip, loff, lsize, end_size);
455
456 if (error) {
457 goto out_lock;
458 }
459
460 prev_zero_fsb = start_zero_fsb;
461 prev_zero_count = buf_len_fsb;
462 start_zero_fsb = imap.br_startoff + buf_len_fsb;
463 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
464
465 XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
466 }
467
468 return 0;
469
470 out_lock:
471
472 XFS_ILOCK(mp, io, XFS_ILOCK_EXCL|XFS_EXTSIZE_RD);
473 ASSERT(error >= 0);
474 return error;
475 }
476
477 ssize_t /* bytes written, or (-) error */
478 xfs_write(
479 bhv_desc_t *bdp,
480 struct file *file,
481 const struct iovec *iovp,
482 unsigned long segs,
483 loff_t *offset,
484 cred_t *credp)
485 {
486 size_t size = 0;
487 xfs_inode_t *xip;
488 xfs_mount_t *mp;
489 ssize_t ret;
490 int error = 0;
491 xfs_fsize_t isize, new_size;
492 xfs_fsize_t n, limit = XFS_MAX_FILE_OFFSET;
493 xfs_iocore_t *io;
494 vnode_t *vp;
495 unsigned long seg;
496 int iolock;
497 int direct = file->f_flags & O_DIRECT;
498 int eventsent = 0;
499 vrwlock_t locktype;
500
501 XFS_STATS_INC(xfsstats.xs_write_calls);
502
503 vp = BHV_TO_VNODE(bdp);
504 vn_trace_entry(vp, "xfs_write", (inst_t *)__return_address);
505 xip = XFS_BHVTOI(bdp);
506
507 /* START copy & waste from filemap.c */
508 for (seg = 0; seg < segs; seg++) {
509 const struct iovec *iv = &iovp[seg];
510
511 /*
512 * If any segment has a negative length, or the cumulative
513 * length ever wraps negative then return -EINVAL.
514 */
515 size += iv->iov_len;
516 if (unlikely((ssize_t)(size|iv->iov_len) < 0))
517 return XFS_ERROR(-EINVAL);
518 if (direct) { /* XFS specific check */
519 if ((__psint_t)iv->iov_base & BBMASK)
520 return XFS_ERROR(-EINVAL);
521 }
522 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len))
523 continue;
524 if (seg == 0)
525 return XFS_ERROR(-EFAULT);
526 segs = seg;
527 break;
528 }
529 /* END copy & waste from filemap.c */
530
531 if (size == 0)
532 return 0;
533
534 io = &(xip->i_iocore);
535 mp = io->io_mount;
536
537 xfs_check_frozen(mp, bdp, XFS_FREEZE_WRITE);
538
539 if (XFS_FORCED_SHUTDOWN(xip->i_mount)) {
540 return -EIO;
541 }
542
543 if (direct) {
544 if ((*offset & mp->m_blockmask) ||
545 (size & mp->m_blockmask)) {
546 return XFS_ERROR(-EINVAL);
547 }
548 iolock = XFS_IOLOCK_SHARED;
549 locktype = VRWLOCK_WRITE_DIRECT;
550 } else {
551 iolock = XFS_IOLOCK_EXCL;
552 locktype = VRWLOCK_WRITE;
553 }
554
555 xfs_ilock(xip, XFS_ILOCK_EXCL|iolock);
556 isize = xip->i_d.di_size;
557
558 if (file->f_flags & O_APPEND)
559 *offset = isize;
560
561 start:
562 n = limit - *offset;
563 if (n <= 0) {
564 xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
565 return -EFBIG;
566 }
567
568 if (n < size)
569 size = n;
570
571 new_size = *offset + size;
572 if (new_size > isize) {
573 io->io_new_size = new_size;
574 }
575
576 if ((DM_EVENT_ENABLED(vp->v_vfsp, xip, DM_EVENT_WRITE) &&
577 !(file->f_mode & FINVIS) && !eventsent)) {
578 loff_t savedsize = *offset;
579
580 xfs_iunlock(xip, XFS_ILOCK_EXCL);
581 error = xfs_dm_send_data_event(DM_EVENT_WRITE, bdp,
582 *offset, size,
583 FILP_DELAY_FLAG(file), &locktype);
584 if (error) {
585 xfs_iunlock(xip, iolock);
586 return -error;
587 }
588 xfs_ilock(xip, XFS_ILOCK_EXCL);
589 eventsent = 1;
590
591 /*
592 * The iolock was dropped and reaquired in
593 * xfs_dm_send_data_event so we have to recheck the size
594 * when appending. We will only "goto start;" once,
595 * since having sent the event prevents another call
596 * to xfs_dm_send_data_event, which is what
597 * allows the size to change in the first place.
598 */
599 if ((file->f_flags & O_APPEND) &&
600 savedsize != xip->i_d.di_size) {
601 *offset = isize = xip->i_d.di_size;
602 goto start;
603 }
604 }
605
606 /*
607 * On Linux, generic_file_write updates the times even if
608 * no data is copied in so long as the write had a size.
609 *
610 * We must update xfs' times since revalidate will overcopy xfs.
611 */
612 if (size) {
613 if (!(file->f_mode & FINVIS))
614 xfs_ichgtime(xip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
615 }
616
617 /*
618 * If the offset is beyond the size of the file, we have a couple
619 * of things to do. First, if there is already space allocated
620 * we need to either create holes or zero the disk or ...
621 *
622 * If there is a page where the previous size lands, we need
623 * to zero it out up to the new size.
624 */
625
626 if (!direct && (*offset > isize && isize)) {
627 error = xfs_zero_eof(BHV_TO_VNODE(bdp), io, *offset,
628 isize, *offset + size);
629 if (error) {
630 xfs_iunlock(xip, XFS_ILOCK_EXCL|iolock);
631 return(-error);
632 }
633 }
634 xfs_iunlock(xip, XFS_ILOCK_EXCL);
635
636 /*
637 * If we're writing the file then make sure to clear the
638 * setuid and setgid bits if the process is not being run
639 * by root. This keeps people from modifying setuid and
640 * setgid binaries.
641 */
642
643 if (((xip->i_d.di_mode & ISUID) ||
644 ((xip->i_d.di_mode & (ISGID | (IEXEC >> 3))) ==
645 (ISGID | (IEXEC >> 3)))) &&
646 !capable(CAP_FSETID)) {
647 error = xfs_write_clear_setuid(xip);
648 if (error) {
649 xfs_iunlock(xip, iolock);
650 return -error;
651 }
652 }
653
654 retry:
655 if (direct) {
656 xfs_inval_cached_pages(vp, &xip->i_iocore, *offset, 1, 1);
657 }
658
659 ret = generic_file_write_nolock(file, iovp, segs, offset);
660
661 if ((ret == -ENOSPC) &&
662 DM_EVENT_ENABLED(vp->v_vfsp, xip, DM_EVENT_NOSPACE) &&
663 !(file->f_mode & FINVIS)) {
664
665 xfs_rwunlock(bdp, locktype);
666 error = dm_send_namesp_event(DM_EVENT_NOSPACE, bdp,
667 DM_RIGHT_NULL, bdp, DM_RIGHT_NULL, NULL, NULL,
668 0, 0, 0); /* Delay flag intentionally unused */
669 if (error)
670 return -error;
671 xfs_rwlock(bdp, locktype);
672 *offset = xip->i_d.di_size;
673 goto retry;
674
675 }
676
677 if (ret <= 0) {
678 xfs_rwunlock(bdp, locktype);
679 return ret;
680 }
681
682 XFS_STATS_ADD(xfsstats.xs_write_bytes, ret);
683
684 if (*offset > xip->i_d.di_size) {
685 xfs_ilock(xip, XFS_ILOCK_EXCL);
686 if (*offset > xip->i_d.di_size) {
687 struct inode *inode = LINVFS_GET_IP(vp);
688
689 inode->i_size = xip->i_d.di_size = *offset;
690 xip->i_update_core = 1;
691 xip->i_update_size = 1;
692 }
693 xfs_iunlock(xip, XFS_ILOCK_EXCL);
694 }
695
696 /* Handle various SYNC-type writes */
697 if ((file->f_flags & O_SYNC) || IS_SYNC(file->f_dentry->d_inode)) {
698
699 /*
700 * If we're treating this as O_DSYNC and we have not updated the
701 * size, force the log.
702 */
703
704 if (!(mp->m_flags & XFS_MOUNT_OSYNCISOSYNC)
705 && !(xip->i_update_size)) {
706 /*
707 * If an allocation transaction occurred
708 * without extending the size, then we have to force
709 * the log up the proper point to ensure that the
710 * allocation is permanent. We can't count on
711 * the fact that buffered writes lock out direct I/O
712 * writes - the direct I/O write could have extended
713 * the size nontransactionally, then finished before
714 * we started. xfs_write_file will think that the file
715 * didn't grow but the update isn't safe unless the
716 * size change is logged.
717 *
718 * Force the log if we've committed a transaction
719 * against the inode or if someone else has and
720 * the commit record hasn't gone to disk (e.g.
721 * the inode is pinned). This guarantees that
722 * all changes affecting the inode are permanent
723 * when we return.
724 */
725
726 xfs_inode_log_item_t *iip;
727 xfs_lsn_t lsn;
728
729 iip = xip->i_itemp;
730 if (iip && iip->ili_last_lsn) {
731 lsn = iip->ili_last_lsn;
732 xfs_log_force(mp, lsn,
733 XFS_LOG_FORCE | XFS_LOG_SYNC);
734 } else if (xfs_ipincount(xip) > 0) {
735 xfs_log_force(mp, (xfs_lsn_t)0,
736 XFS_LOG_FORCE | XFS_LOG_SYNC);
737 }
738
739 } else {
740 xfs_trans_t *tp;
741
742 /*
743 * O_SYNC or O_DSYNC _with_ a size update are handled
744 * the same way.
745 *
746 * If the write was synchronous then we need to make
747 * sure that the inode modification time is permanent.
748 * We'll have updated the timestamp above, so here
749 * we use a synchronous transaction to log the inode.
750 * It's not fast, but it's necessary.
751 *
752 * If this a dsync write and the size got changed
753 * non-transactionally, then we need to ensure that
754 * the size change gets logged in a synchronous
755 * transaction.
756 */
757
758 tp = xfs_trans_alloc(mp, XFS_TRANS_WRITE_SYNC);
759 if ((error = xfs_trans_reserve(tp, 0,
760 XFS_SWRITE_LOG_RES(mp),
761 0, 0, 0))) {
762 /* Transaction reserve failed */
763 xfs_trans_cancel(tp, 0);
764 } else {
765 /* Transaction reserve successful */
766 xfs_ilock(xip, XFS_ILOCK_EXCL);
767 xfs_trans_ijoin(tp, xip, XFS_ILOCK_EXCL);
768 xfs_trans_ihold(tp, xip);
769 xfs_trans_log_inode(tp, xip, XFS_ILOG_CORE);
770 xfs_trans_set_sync(tp);
771 error = xfs_trans_commit(tp, 0, (xfs_lsn_t)0);
772 xfs_iunlock(xip, XFS_ILOCK_EXCL);
773 }
774 }
775 } /* (ioflags & O_SYNC) */
776
777 xfs_rwunlock(bdp, locktype);
778 return(ret);
779 }
780
781 /*
782 * All xfs metadata buffers except log state machine buffers
783 * get this attached as their b_bdstrat callback function.
784 * This is so that we can catch a buffer
785 * after prematurely unpinning it to forcibly shutdown the filesystem.
786 */
787 int
788 xfs_bdstrat_cb(struct xfs_buf *bp)
789 {
790 xfs_mount_t *mp;
791
792 mp = XFS_BUF_FSPRIVATE3(bp, xfs_mount_t *);
793 if (!XFS_FORCED_SHUTDOWN(mp)) {
794 pagebuf_iorequest(bp);
795 return 0;
796 } else {
797 xfs_buftrace("XFS__BDSTRAT IOERROR", bp);
798 /*
799 * Metadata write that didn't get logged but
800 * written delayed anyway. These aren't associated
801 * with a transaction, and can be ignored.
802 */
803 if (XFS_BUF_IODONE_FUNC(bp) == NULL &&
804 (XFS_BUF_ISREAD(bp)) == 0)
805 return (xfs_bioerror_relse(bp));
806 else
807 return (xfs_bioerror(bp));
808 }
809 }
810
811
812 int
813 xfs_bmap(bhv_desc_t *bdp,
814 xfs_off_t offset,
815 ssize_t count,
816 int flags,
817 page_buf_bmap_t *pbmapp,
818 int *npbmaps)
819 {
820 xfs_inode_t *ip = XFS_BHVTOI(bdp);
821 xfs_iocore_t *io = &ip->i_iocore;
822
823 ASSERT((ip->i_d.di_mode & IFMT) == IFREG);
824 ASSERT(((ip->i_d.di_flags & XFS_DIFLAG_REALTIME) != 0) ==
825 ((ip->i_iocore.io_flags & XFS_IOCORE_RT) != 0));
826
827 return xfs_iomap(io, offset, count, flags, pbmapp, npbmaps);
828 }
829
830 /*
831 * Wrapper around bdstrat so that we can stop data
832 * from going to disk in case we are shutting down the filesystem.
833 * Typically user data goes thru this path; one of the exceptions
834 * is the superblock.
835 */
836 int
837 xfsbdstrat(
838 struct xfs_mount *mp,
839 struct xfs_buf *bp)
840 {
841 ASSERT(mp);
842 if (!XFS_FORCED_SHUTDOWN(mp)) {
843 /* Grio redirection would go here
844 * if (XFS_BUF_IS_GRIO(bp)) {
845 */
846
847 pagebuf_iorequest(bp);
848 return 0;
849 }
850
851 xfs_buftrace("XFSBDSTRAT IOERROR", bp);
852 return (xfs_bioerror_relse(bp));
853 }
854
855
856 void
857 XFS_bflush(xfs_buftarg_t *target)
858 {
859 pagebuf_delwri_flush(target, PBDF_WAIT, NULL);
860 }
861
862
863 /* Push all fs state out to disk
864 */
865
866 void
867 XFS_log_write_unmount_ro(bhv_desc_t *bdp)
868 {
869 xfs_mount_t *mp;
870 int pincount = 0;
871 int count = 0;
872 int error;
873
874 mp = XFS_BHVTOM(bdp);
875 pagebuf_delwri_flush(mp->m_ddev_targp, PBDF_WAIT, &pincount);
876 xfs_finish_reclaim_all(mp);
877
878 do {
879 VFS_SYNC(XFS_MTOVFS(mp), SYNC_ATTR|SYNC_WAIT, NULL, error);
880 pagebuf_delwri_flush(mp->m_ddev_targp, PBDF_WAIT, &pincount);
881 if (pincount == 0) {delay(50); count++;}
882 } while (count < 2);
883
884 /* Ok now write out an unmount record */
885 xfs_log_unmount_write(mp);
886 xfs_unmountfs_writesb(mp);
887 }
888
889 /*
890 * If the underlying (log or data) device is readonly, there are some
891 * operations that cannot proceed.
892 */
893 int
894 xfs_dev_is_read_only(xfs_mount_t *mp, char *message)
895 {
896 if (bdev_read_only(mp->m_ddev_targp->pbr_bdev) ||
897 bdev_read_only(mp->m_logdev_targp->pbr_bdev) ||
898 (mp->m_rtdev_targp && bdev_read_only(mp->m_rtdev_targp->pbr_bdev))) {
899 cmn_err(CE_NOTE,
900 "XFS: %s required on read-only device.", message);
901 cmn_err(CE_NOTE,
902 "XFS: write access unavailable, cannot proceed.");
903 return EROFS;
904 }
905
906 return 0;
907 }
908
Mirror of the linux-mips.org kernel tree