Loading...
1/*
2 * Copyright (c) 2000-2005 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_bit.h"
21#include "xfs_log.h"
22#include "xfs_inum.h"
23#include "xfs_sb.h"
24#include "xfs_ag.h"
25#include "xfs_trans.h"
26#include "xfs_mount.h"
27#include "xfs_bmap_btree.h"
28#include "xfs_alloc.h"
29#include "xfs_dinode.h"
30#include "xfs_inode.h"
31#include "xfs_inode_item.h"
32#include "xfs_bmap.h"
33#include "xfs_error.h"
34#include "xfs_vnodeops.h"
35#include "xfs_da_btree.h"
36#include "xfs_ioctl.h"
37#include "xfs_trace.h"
38
39#include <linux/dcache.h>
40#include <linux/falloc.h>
41
42static const struct vm_operations_struct xfs_file_vm_ops;
43
44/*
45 * Locking primitives for read and write IO paths to ensure we consistently use
46 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
47 */
48static inline void
49xfs_rw_ilock(
50 struct xfs_inode *ip,
51 int type)
52{
53 if (type & XFS_IOLOCK_EXCL)
54 mutex_lock(&VFS_I(ip)->i_mutex);
55 xfs_ilock(ip, type);
56}
57
58static inline void
59xfs_rw_iunlock(
60 struct xfs_inode *ip,
61 int type)
62{
63 xfs_iunlock(ip, type);
64 if (type & XFS_IOLOCK_EXCL)
65 mutex_unlock(&VFS_I(ip)->i_mutex);
66}
67
68static inline void
69xfs_rw_ilock_demote(
70 struct xfs_inode *ip,
71 int type)
72{
73 xfs_ilock_demote(ip, type);
74 if (type & XFS_IOLOCK_EXCL)
75 mutex_unlock(&VFS_I(ip)->i_mutex);
76}
77
78/*
79 * xfs_iozero
80 *
81 * xfs_iozero clears the specified range of buffer supplied,
82 * and marks all the affected blocks as valid and modified. If
83 * an affected block is not allocated, it will be allocated. If
84 * an affected block is not completely overwritten, and is not
85 * valid before the operation, it will be read from disk before
86 * being partially zeroed.
87 */
88STATIC int
89xfs_iozero(
90 struct xfs_inode *ip, /* inode */
91 loff_t pos, /* offset in file */
92 size_t count) /* size of data to zero */
93{
94 struct page *page;
95 struct address_space *mapping;
96 int status;
97
98 mapping = VFS_I(ip)->i_mapping;
99 do {
100 unsigned offset, bytes;
101 void *fsdata;
102
103 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
104 bytes = PAGE_CACHE_SIZE - offset;
105 if (bytes > count)
106 bytes = count;
107
108 status = pagecache_write_begin(NULL, mapping, pos, bytes,
109 AOP_FLAG_UNINTERRUPTIBLE,
110 &page, &fsdata);
111 if (status)
112 break;
113
114 zero_user(page, offset, bytes);
115
116 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
117 page, fsdata);
118 WARN_ON(status <= 0); /* can't return less than zero! */
119 pos += bytes;
120 count -= bytes;
121 status = 0;
122 } while (count);
123
124 return (-status);
125}
126
127STATIC int
128xfs_file_fsync(
129 struct file *file,
130 loff_t start,
131 loff_t end,
132 int datasync)
133{
134 struct inode *inode = file->f_mapping->host;
135 struct xfs_inode *ip = XFS_I(inode);
136 struct xfs_mount *mp = ip->i_mount;
137 struct xfs_trans *tp;
138 int error = 0;
139 int log_flushed = 0;
140
141 trace_xfs_file_fsync(ip);
142
143 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
144 if (error)
145 return error;
146
147 if (XFS_FORCED_SHUTDOWN(mp))
148 return -XFS_ERROR(EIO);
149
150 xfs_iflags_clear(ip, XFS_ITRUNCATED);
151
152 xfs_ilock(ip, XFS_IOLOCK_SHARED);
153 xfs_ioend_wait(ip);
154 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
155
156 if (mp->m_flags & XFS_MOUNT_BARRIER) {
157 /*
158 * If we have an RT and/or log subvolume we need to make sure
159 * to flush the write cache the device used for file data
160 * first. This is to ensure newly written file data make
161 * it to disk before logging the new inode size in case of
162 * an extending write.
163 */
164 if (XFS_IS_REALTIME_INODE(ip))
165 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
166 else if (mp->m_logdev_targp != mp->m_ddev_targp)
167 xfs_blkdev_issue_flush(mp->m_ddev_targp);
168 }
169
170 /*
171 * We always need to make sure that the required inode state is safe on
172 * disk. The inode might be clean but we still might need to force the
173 * log because of committed transactions that haven't hit the disk yet.
174 * Likewise, there could be unflushed non-transactional changes to the
175 * inode core that have to go to disk and this requires us to issue
176 * a synchronous transaction to capture these changes correctly.
177 *
178 * This code relies on the assumption that if the i_update_core field
179 * of the inode is clear and the inode is unpinned then it is clean
180 * and no action is required.
181 */
182 xfs_ilock(ip, XFS_ILOCK_SHARED);
183
184 /*
185 * First check if the VFS inode is marked dirty. All the dirtying
186 * of non-transactional updates no goes through mark_inode_dirty*,
187 * which allows us to distinguish beteeen pure timestamp updates
188 * and i_size updates which need to be caught for fdatasync.
189 * After that also theck for the dirty state in the XFS inode, which
190 * might gets cleared when the inode gets written out via the AIL
191 * or xfs_iflush_cluster.
192 */
193 if (((inode->i_state & I_DIRTY_DATASYNC) ||
194 ((inode->i_state & I_DIRTY_SYNC) && !datasync)) &&
195 ip->i_update_core) {
196 /*
197 * Kick off a transaction to log the inode core to get the
198 * updates. The sync transaction will also force the log.
199 */
200 xfs_iunlock(ip, XFS_ILOCK_SHARED);
201 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
202 error = xfs_trans_reserve(tp, 0,
203 XFS_FSYNC_TS_LOG_RES(mp), 0, 0, 0);
204 if (error) {
205 xfs_trans_cancel(tp, 0);
206 return -error;
207 }
208 xfs_ilock(ip, XFS_ILOCK_EXCL);
209
210 /*
211 * Note - it's possible that we might have pushed ourselves out
212 * of the way during trans_reserve which would flush the inode.
213 * But there's no guarantee that the inode buffer has actually
214 * gone out yet (it's delwri). Plus the buffer could be pinned
215 * anyway if it's part of an inode in another recent
216 * transaction. So we play it safe and fire off the
217 * transaction anyway.
218 */
219 xfs_trans_ijoin(tp, ip);
220 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
221 xfs_trans_set_sync(tp);
222 error = _xfs_trans_commit(tp, 0, &log_flushed);
223
224 xfs_iunlock(ip, XFS_ILOCK_EXCL);
225 } else {
226 /*
227 * Timestamps/size haven't changed since last inode flush or
228 * inode transaction commit. That means either nothing got
229 * written or a transaction committed which caught the updates.
230 * If the latter happened and the transaction hasn't hit the
231 * disk yet, the inode will be still be pinned. If it is,
232 * force the log.
233 */
234 if (xfs_ipincount(ip)) {
235 error = _xfs_log_force_lsn(mp,
236 ip->i_itemp->ili_last_lsn,
237 XFS_LOG_SYNC, &log_flushed);
238 }
239 xfs_iunlock(ip, XFS_ILOCK_SHARED);
240 }
241
242 /*
243 * If we only have a single device, and the log force about was
244 * a no-op we might have to flush the data device cache here.
245 * This can only happen for fdatasync/O_DSYNC if we were overwriting
246 * an already allocated file and thus do not have any metadata to
247 * commit.
248 */
249 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
250 mp->m_logdev_targp == mp->m_ddev_targp &&
251 !XFS_IS_REALTIME_INODE(ip) &&
252 !log_flushed)
253 xfs_blkdev_issue_flush(mp->m_ddev_targp);
254
255 return -error;
256}
257
258STATIC ssize_t
259xfs_file_aio_read(
260 struct kiocb *iocb,
261 const struct iovec *iovp,
262 unsigned long nr_segs,
263 loff_t pos)
264{
265 struct file *file = iocb->ki_filp;
266 struct inode *inode = file->f_mapping->host;
267 struct xfs_inode *ip = XFS_I(inode);
268 struct xfs_mount *mp = ip->i_mount;
269 size_t size = 0;
270 ssize_t ret = 0;
271 int ioflags = 0;
272 xfs_fsize_t n;
273 unsigned long seg;
274
275 XFS_STATS_INC(xs_read_calls);
276
277 BUG_ON(iocb->ki_pos != pos);
278
279 if (unlikely(file->f_flags & O_DIRECT))
280 ioflags |= IO_ISDIRECT;
281 if (file->f_mode & FMODE_NOCMTIME)
282 ioflags |= IO_INVIS;
283
284 /* START copy & waste from filemap.c */
285 for (seg = 0; seg < nr_segs; seg++) {
286 const struct iovec *iv = &iovp[seg];
287
288 /*
289 * If any segment has a negative length, or the cumulative
290 * length ever wraps negative then return -EINVAL.
291 */
292 size += iv->iov_len;
293 if (unlikely((ssize_t)(size|iv->iov_len) < 0))
294 return XFS_ERROR(-EINVAL);
295 }
296 /* END copy & waste from filemap.c */
297
298 if (unlikely(ioflags & IO_ISDIRECT)) {
299 xfs_buftarg_t *target =
300 XFS_IS_REALTIME_INODE(ip) ?
301 mp->m_rtdev_targp : mp->m_ddev_targp;
302 if ((iocb->ki_pos & target->bt_smask) ||
303 (size & target->bt_smask)) {
304 if (iocb->ki_pos == ip->i_size)
305 return 0;
306 return -XFS_ERROR(EINVAL);
307 }
308 }
309
310 n = XFS_MAXIOFFSET(mp) - iocb->ki_pos;
311 if (n <= 0 || size == 0)
312 return 0;
313
314 if (n < size)
315 size = n;
316
317 if (XFS_FORCED_SHUTDOWN(mp))
318 return -EIO;
319
320 if (unlikely(ioflags & IO_ISDIRECT)) {
321 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
322
323 if (inode->i_mapping->nrpages) {
324 ret = -xfs_flushinval_pages(ip,
325 (iocb->ki_pos & PAGE_CACHE_MASK),
326 -1, FI_REMAPF_LOCKED);
327 if (ret) {
328 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
329 return ret;
330 }
331 }
332 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
333 } else
334 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
335
336 trace_xfs_file_read(ip, size, iocb->ki_pos, ioflags);
337
338 ret = generic_file_aio_read(iocb, iovp, nr_segs, iocb->ki_pos);
339 if (ret > 0)
340 XFS_STATS_ADD(xs_read_bytes, ret);
341
342 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
343 return ret;
344}
345
346STATIC ssize_t
347xfs_file_splice_read(
348 struct file *infilp,
349 loff_t *ppos,
350 struct pipe_inode_info *pipe,
351 size_t count,
352 unsigned int flags)
353{
354 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
355 int ioflags = 0;
356 ssize_t ret;
357
358 XFS_STATS_INC(xs_read_calls);
359
360 if (infilp->f_mode & FMODE_NOCMTIME)
361 ioflags |= IO_INVIS;
362
363 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
364 return -EIO;
365
366 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
367
368 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
369
370 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
371 if (ret > 0)
372 XFS_STATS_ADD(xs_read_bytes, ret);
373
374 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
375 return ret;
376}
377
378STATIC void
379xfs_aio_write_isize_update(
380 struct inode *inode,
381 loff_t *ppos,
382 ssize_t bytes_written)
383{
384 struct xfs_inode *ip = XFS_I(inode);
385 xfs_fsize_t isize = i_size_read(inode);
386
387 if (bytes_written > 0)
388 XFS_STATS_ADD(xs_write_bytes, bytes_written);
389
390 if (unlikely(bytes_written < 0 && bytes_written != -EFAULT &&
391 *ppos > isize))
392 *ppos = isize;
393
394 if (*ppos > ip->i_size) {
395 xfs_rw_ilock(ip, XFS_ILOCK_EXCL);
396 if (*ppos > ip->i_size)
397 ip->i_size = *ppos;
398 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL);
399 }
400}
401
402/*
403 * If this was a direct or synchronous I/O that failed (such as ENOSPC) then
404 * part of the I/O may have been written to disk before the error occurred. In
405 * this case the on-disk file size may have been adjusted beyond the in-memory
406 * file size and now needs to be truncated back.
407 */
408STATIC void
409xfs_aio_write_newsize_update(
410 struct xfs_inode *ip)
411{
412 if (ip->i_new_size) {
413 xfs_rw_ilock(ip, XFS_ILOCK_EXCL);
414 ip->i_new_size = 0;
415 if (ip->i_d.di_size > ip->i_size)
416 ip->i_d.di_size = ip->i_size;
417 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL);
418 }
419}
420
421/*
422 * xfs_file_splice_write() does not use xfs_rw_ilock() because
423 * generic_file_splice_write() takes the i_mutex itself. This, in theory,
424 * couuld cause lock inversions between the aio_write path and the splice path
425 * if someone is doing concurrent splice(2) based writes and write(2) based
426 * writes to the same inode. The only real way to fix this is to re-implement
427 * the generic code here with correct locking orders.
428 */
429STATIC ssize_t
430xfs_file_splice_write(
431 struct pipe_inode_info *pipe,
432 struct file *outfilp,
433 loff_t *ppos,
434 size_t count,
435 unsigned int flags)
436{
437 struct inode *inode = outfilp->f_mapping->host;
438 struct xfs_inode *ip = XFS_I(inode);
439 xfs_fsize_t new_size;
440 int ioflags = 0;
441 ssize_t ret;
442
443 XFS_STATS_INC(xs_write_calls);
444
445 if (outfilp->f_mode & FMODE_NOCMTIME)
446 ioflags |= IO_INVIS;
447
448 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
449 return -EIO;
450
451 xfs_ilock(ip, XFS_IOLOCK_EXCL);
452
453 new_size = *ppos + count;
454
455 xfs_ilock(ip, XFS_ILOCK_EXCL);
456 if (new_size > ip->i_size)
457 ip->i_new_size = new_size;
458 xfs_iunlock(ip, XFS_ILOCK_EXCL);
459
460 trace_xfs_file_splice_write(ip, count, *ppos, ioflags);
461
462 ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
463
464 xfs_aio_write_isize_update(inode, ppos, ret);
465 xfs_aio_write_newsize_update(ip);
466 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
467 return ret;
468}
469
470/*
471 * This routine is called to handle zeroing any space in the last
472 * block of the file that is beyond the EOF. We do this since the
473 * size is being increased without writing anything to that block
474 * and we don't want anyone to read the garbage on the disk.
475 */
476STATIC int /* error (positive) */
477xfs_zero_last_block(
478 xfs_inode_t *ip,
479 xfs_fsize_t offset,
480 xfs_fsize_t isize)
481{
482 xfs_fileoff_t last_fsb;
483 xfs_mount_t *mp = ip->i_mount;
484 int nimaps;
485 int zero_offset;
486 int zero_len;
487 int error = 0;
488 xfs_bmbt_irec_t imap;
489
490 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL));
491
492 zero_offset = XFS_B_FSB_OFFSET(mp, isize);
493 if (zero_offset == 0) {
494 /*
495 * There are no extra bytes in the last block on disk to
496 * zero, so return.
497 */
498 return 0;
499 }
500
501 last_fsb = XFS_B_TO_FSBT(mp, isize);
502 nimaps = 1;
503 error = xfs_bmapi(NULL, ip, last_fsb, 1, 0, NULL, 0, &imap,
504 &nimaps, NULL);
505 if (error) {
506 return error;
507 }
508 ASSERT(nimaps > 0);
509 /*
510 * If the block underlying isize is just a hole, then there
511 * is nothing to zero.
512 */
513 if (imap.br_startblock == HOLESTARTBLOCK) {
514 return 0;
515 }
516 /*
517 * Zero the part of the last block beyond the EOF, and write it
518 * out sync. We need to drop the ilock while we do this so we
519 * don't deadlock when the buffer cache calls back to us.
520 */
521 xfs_iunlock(ip, XFS_ILOCK_EXCL);
522
523 zero_len = mp->m_sb.sb_blocksize - zero_offset;
524 if (isize + zero_len > offset)
525 zero_len = offset - isize;
526 error = xfs_iozero(ip, isize, zero_len);
527
528 xfs_ilock(ip, XFS_ILOCK_EXCL);
529 ASSERT(error >= 0);
530 return error;
531}
532
533/*
534 * Zero any on disk space between the current EOF and the new,
535 * larger EOF. This handles the normal case of zeroing the remainder
536 * of the last block in the file and the unusual case of zeroing blocks
537 * out beyond the size of the file. This second case only happens
538 * with fixed size extents and when the system crashes before the inode
539 * size was updated but after blocks were allocated. If fill is set,
540 * then any holes in the range are filled and zeroed. If not, the holes
541 * are left alone as holes.
542 */
543
544int /* error (positive) */
545xfs_zero_eof(
546 xfs_inode_t *ip,
547 xfs_off_t offset, /* starting I/O offset */
548 xfs_fsize_t isize) /* current inode size */
549{
550 xfs_mount_t *mp = ip->i_mount;
551 xfs_fileoff_t start_zero_fsb;
552 xfs_fileoff_t end_zero_fsb;
553 xfs_fileoff_t zero_count_fsb;
554 xfs_fileoff_t last_fsb;
555 xfs_fileoff_t zero_off;
556 xfs_fsize_t zero_len;
557 int nimaps;
558 int error = 0;
559 xfs_bmbt_irec_t imap;
560
561 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
562 ASSERT(offset > isize);
563
564 /*
565 * First handle zeroing the block on which isize resides.
566 * We only zero a part of that block so it is handled specially.
567 */
568 error = xfs_zero_last_block(ip, offset, isize);
569 if (error) {
570 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
571 return error;
572 }
573
574 /*
575 * Calculate the range between the new size and the old
576 * where blocks needing to be zeroed may exist. To get the
577 * block where the last byte in the file currently resides,
578 * we need to subtract one from the size and truncate back
579 * to a block boundary. We subtract 1 in case the size is
580 * exactly on a block boundary.
581 */
582 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
583 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
584 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
585 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
586 if (last_fsb == end_zero_fsb) {
587 /*
588 * The size was only incremented on its last block.
589 * We took care of that above, so just return.
590 */
591 return 0;
592 }
593
594 ASSERT(start_zero_fsb <= end_zero_fsb);
595 while (start_zero_fsb <= end_zero_fsb) {
596 nimaps = 1;
597 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
598 error = xfs_bmapi(NULL, ip, start_zero_fsb, zero_count_fsb,
599 0, NULL, 0, &imap, &nimaps, NULL);
600 if (error) {
601 ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_IOLOCK_EXCL));
602 return error;
603 }
604 ASSERT(nimaps > 0);
605
606 if (imap.br_state == XFS_EXT_UNWRITTEN ||
607 imap.br_startblock == HOLESTARTBLOCK) {
608 /*
609 * This loop handles initializing pages that were
610 * partially initialized by the code below this
611 * loop. It basically zeroes the part of the page
612 * that sits on a hole and sets the page as P_HOLE
613 * and calls remapf if it is a mapped file.
614 */
615 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
616 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
617 continue;
618 }
619
620 /*
621 * There are blocks we need to zero.
622 * Drop the inode lock while we're doing the I/O.
623 * We'll still have the iolock to protect us.
624 */
625 xfs_iunlock(ip, XFS_ILOCK_EXCL);
626
627 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
628 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
629
630 if ((zero_off + zero_len) > offset)
631 zero_len = offset - zero_off;
632
633 error = xfs_iozero(ip, zero_off, zero_len);
634 if (error) {
635 goto out_lock;
636 }
637
638 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
639 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
640
641 xfs_ilock(ip, XFS_ILOCK_EXCL);
642 }
643
644 return 0;
645
646out_lock:
647 xfs_ilock(ip, XFS_ILOCK_EXCL);
648 ASSERT(error >= 0);
649 return error;
650}
651
652/*
653 * Common pre-write limit and setup checks.
654 *
655 * Returns with iolock held according to @iolock.
656 */
657STATIC ssize_t
658xfs_file_aio_write_checks(
659 struct file *file,
660 loff_t *pos,
661 size_t *count,
662 int *iolock)
663{
664 struct inode *inode = file->f_mapping->host;
665 struct xfs_inode *ip = XFS_I(inode);
666 xfs_fsize_t new_size;
667 int error = 0;
668
669 error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
670 if (error) {
671 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL | *iolock);
672 *iolock = 0;
673 return error;
674 }
675
676 new_size = *pos + *count;
677 if (new_size > ip->i_size)
678 ip->i_new_size = new_size;
679
680 if (likely(!(file->f_mode & FMODE_NOCMTIME)))
681 file_update_time(file);
682
683 /*
684 * If the offset is beyond the size of the file, we need to zero any
685 * blocks that fall between the existing EOF and the start of this
686 * write.
687 */
688 if (*pos > ip->i_size)
689 error = -xfs_zero_eof(ip, *pos, ip->i_size);
690
691 xfs_rw_iunlock(ip, XFS_ILOCK_EXCL);
692 if (error)
693 return error;
694
695 /*
696 * If we're writing the file then make sure to clear the setuid and
697 * setgid bits if the process is not being run by root. This keeps
698 * people from modifying setuid and setgid binaries.
699 */
700 return file_remove_suid(file);
701
702}
703
704/*
705 * xfs_file_dio_aio_write - handle direct IO writes
706 *
707 * Lock the inode appropriately to prepare for and issue a direct IO write.
708 * By separating it from the buffered write path we remove all the tricky to
709 * follow locking changes and looping.
710 *
711 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
712 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
713 * pages are flushed out.
714 *
715 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
716 * allowing them to be done in parallel with reads and other direct IO writes.
717 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
718 * needs to do sub-block zeroing and that requires serialisation against other
719 * direct IOs to the same block. In this case we need to serialise the
720 * submission of the unaligned IOs so that we don't get racing block zeroing in
721 * the dio layer. To avoid the problem with aio, we also need to wait for
722 * outstanding IOs to complete so that unwritten extent conversion is completed
723 * before we try to map the overlapping block. This is currently implemented by
724 * hitting it with a big hammer (i.e. xfs_ioend_wait()).
725 *
726 * Returns with locks held indicated by @iolock and errors indicated by
727 * negative return values.
728 */
729STATIC ssize_t
730xfs_file_dio_aio_write(
731 struct kiocb *iocb,
732 const struct iovec *iovp,
733 unsigned long nr_segs,
734 loff_t pos,
735 size_t ocount,
736 int *iolock)
737{
738 struct file *file = iocb->ki_filp;
739 struct address_space *mapping = file->f_mapping;
740 struct inode *inode = mapping->host;
741 struct xfs_inode *ip = XFS_I(inode);
742 struct xfs_mount *mp = ip->i_mount;
743 ssize_t ret = 0;
744 size_t count = ocount;
745 int unaligned_io = 0;
746 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
747 mp->m_rtdev_targp : mp->m_ddev_targp;
748
749 *iolock = 0;
750 if ((pos & target->bt_smask) || (count & target->bt_smask))
751 return -XFS_ERROR(EINVAL);
752
753 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
754 unaligned_io = 1;
755
756 if (unaligned_io || mapping->nrpages || pos > ip->i_size)
757 *iolock = XFS_IOLOCK_EXCL;
758 else
759 *iolock = XFS_IOLOCK_SHARED;
760 xfs_rw_ilock(ip, XFS_ILOCK_EXCL | *iolock);
761
762 ret = xfs_file_aio_write_checks(file, &pos, &count, iolock);
763 if (ret)
764 return ret;
765
766 if (mapping->nrpages) {
767 WARN_ON(*iolock != XFS_IOLOCK_EXCL);
768 ret = -xfs_flushinval_pages(ip, (pos & PAGE_CACHE_MASK), -1,
769 FI_REMAPF_LOCKED);
770 if (ret)
771 return ret;
772 }
773
774 /*
775 * If we are doing unaligned IO, wait for all other IO to drain,
776 * otherwise demote the lock if we had to flush cached pages
777 */
778 if (unaligned_io)
779 xfs_ioend_wait(ip);
780 else if (*iolock == XFS_IOLOCK_EXCL) {
781 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
782 *iolock = XFS_IOLOCK_SHARED;
783 }
784
785 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
786 ret = generic_file_direct_write(iocb, iovp,
787 &nr_segs, pos, &iocb->ki_pos, count, ocount);
788
789 /* No fallback to buffered IO on errors for XFS. */
790 ASSERT(ret < 0 || ret == count);
791 return ret;
792}
793
794STATIC ssize_t
795xfs_file_buffered_aio_write(
796 struct kiocb *iocb,
797 const struct iovec *iovp,
798 unsigned long nr_segs,
799 loff_t pos,
800 size_t ocount,
801 int *iolock)
802{
803 struct file *file = iocb->ki_filp;
804 struct address_space *mapping = file->f_mapping;
805 struct inode *inode = mapping->host;
806 struct xfs_inode *ip = XFS_I(inode);
807 ssize_t ret;
808 int enospc = 0;
809 size_t count = ocount;
810
811 *iolock = XFS_IOLOCK_EXCL;
812 xfs_rw_ilock(ip, XFS_ILOCK_EXCL | *iolock);
813
814 ret = xfs_file_aio_write_checks(file, &pos, &count, iolock);
815 if (ret)
816 return ret;
817
818 /* We can write back this queue in page reclaim */
819 current->backing_dev_info = mapping->backing_dev_info;
820
821write_retry:
822 trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
823 ret = generic_file_buffered_write(iocb, iovp, nr_segs,
824 pos, &iocb->ki_pos, count, ret);
825 /*
826 * if we just got an ENOSPC, flush the inode now we aren't holding any
827 * page locks and retry *once*
828 */
829 if (ret == -ENOSPC && !enospc) {
830 ret = -xfs_flush_pages(ip, 0, -1, 0, FI_NONE);
831 if (ret)
832 return ret;
833 enospc = 1;
834 goto write_retry;
835 }
836 current->backing_dev_info = NULL;
837 return ret;
838}
839
840STATIC ssize_t
841xfs_file_aio_write(
842 struct kiocb *iocb,
843 const struct iovec *iovp,
844 unsigned long nr_segs,
845 loff_t pos)
846{
847 struct file *file = iocb->ki_filp;
848 struct address_space *mapping = file->f_mapping;
849 struct inode *inode = mapping->host;
850 struct xfs_inode *ip = XFS_I(inode);
851 ssize_t ret;
852 int iolock;
853 size_t ocount = 0;
854
855 XFS_STATS_INC(xs_write_calls);
856
857 BUG_ON(iocb->ki_pos != pos);
858
859 ret = generic_segment_checks(iovp, &nr_segs, &ocount, VERIFY_READ);
860 if (ret)
861 return ret;
862
863 if (ocount == 0)
864 return 0;
865
866 xfs_wait_for_freeze(ip->i_mount, SB_FREEZE_WRITE);
867
868 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
869 return -EIO;
870
871 if (unlikely(file->f_flags & O_DIRECT))
872 ret = xfs_file_dio_aio_write(iocb, iovp, nr_segs, pos,
873 ocount, &iolock);
874 else
875 ret = xfs_file_buffered_aio_write(iocb, iovp, nr_segs, pos,
876 ocount, &iolock);
877
878 xfs_aio_write_isize_update(inode, &iocb->ki_pos, ret);
879
880 if (ret <= 0)
881 goto out_unlock;
882
883 /* Handle various SYNC-type writes */
884 if ((file->f_flags & O_DSYNC) || IS_SYNC(inode)) {
885 loff_t end = pos + ret - 1;
886 int error;
887
888 xfs_rw_iunlock(ip, iolock);
889 error = xfs_file_fsync(file, pos, end,
890 (file->f_flags & __O_SYNC) ? 0 : 1);
891 xfs_rw_ilock(ip, iolock);
892 if (error)
893 ret = error;
894 }
895
896out_unlock:
897 xfs_aio_write_newsize_update(ip);
898 xfs_rw_iunlock(ip, iolock);
899 return ret;
900}
901
902STATIC long
903xfs_file_fallocate(
904 struct file *file,
905 int mode,
906 loff_t offset,
907 loff_t len)
908{
909 struct inode *inode = file->f_path.dentry->d_inode;
910 long error;
911 loff_t new_size = 0;
912 xfs_flock64_t bf;
913 xfs_inode_t *ip = XFS_I(inode);
914 int cmd = XFS_IOC_RESVSP;
915 int attr_flags = XFS_ATTR_NOLOCK;
916
917 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
918 return -EOPNOTSUPP;
919
920 bf.l_whence = 0;
921 bf.l_start = offset;
922 bf.l_len = len;
923
924 xfs_ilock(ip, XFS_IOLOCK_EXCL);
925
926 if (mode & FALLOC_FL_PUNCH_HOLE)
927 cmd = XFS_IOC_UNRESVSP;
928
929 /* check the new inode size is valid before allocating */
930 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
931 offset + len > i_size_read(inode)) {
932 new_size = offset + len;
933 error = inode_newsize_ok(inode, new_size);
934 if (error)
935 goto out_unlock;
936 }
937
938 if (file->f_flags & O_DSYNC)
939 attr_flags |= XFS_ATTR_SYNC;
940
941 error = -xfs_change_file_space(ip, cmd, &bf, 0, attr_flags);
942 if (error)
943 goto out_unlock;
944
945 /* Change file size if needed */
946 if (new_size) {
947 struct iattr iattr;
948
949 iattr.ia_valid = ATTR_SIZE;
950 iattr.ia_size = new_size;
951 error = -xfs_setattr_size(ip, &iattr, XFS_ATTR_NOLOCK);
952 }
953
954out_unlock:
955 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
956 return error;
957}
958
959
960STATIC int
961xfs_file_open(
962 struct inode *inode,
963 struct file *file)
964{
965 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
966 return -EFBIG;
967 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
968 return -EIO;
969 return 0;
970}
971
972STATIC int
973xfs_dir_open(
974 struct inode *inode,
975 struct file *file)
976{
977 struct xfs_inode *ip = XFS_I(inode);
978 int mode;
979 int error;
980
981 error = xfs_file_open(inode, file);
982 if (error)
983 return error;
984
985 /*
986 * If there are any blocks, read-ahead block 0 as we're almost
987 * certain to have the next operation be a read there.
988 */
989 mode = xfs_ilock_map_shared(ip);
990 if (ip->i_d.di_nextents > 0)
991 xfs_da_reada_buf(NULL, ip, 0, XFS_DATA_FORK);
992 xfs_iunlock(ip, mode);
993 return 0;
994}
995
996STATIC int
997xfs_file_release(
998 struct inode *inode,
999 struct file *filp)
1000{
1001 return -xfs_release(XFS_I(inode));
1002}
1003
1004STATIC int
1005xfs_file_readdir(
1006 struct file *filp,
1007 void *dirent,
1008 filldir_t filldir)
1009{
1010 struct inode *inode = filp->f_path.dentry->d_inode;
1011 xfs_inode_t *ip = XFS_I(inode);
1012 int error;
1013 size_t bufsize;
1014
1015 /*
1016 * The Linux API doesn't pass down the total size of the buffer
1017 * we read into down to the filesystem. With the filldir concept
1018 * it's not needed for correct information, but the XFS dir2 leaf
1019 * code wants an estimate of the buffer size to calculate it's
1020 * readahead window and size the buffers used for mapping to
1021 * physical blocks.
1022 *
1023 * Try to give it an estimate that's good enough, maybe at some
1024 * point we can change the ->readdir prototype to include the
1025 * buffer size. For now we use the current glibc buffer size.
1026 */
1027 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
1028
1029 error = xfs_readdir(ip, dirent, bufsize,
1030 (xfs_off_t *)&filp->f_pos, filldir);
1031 if (error)
1032 return -error;
1033 return 0;
1034}
1035
1036STATIC int
1037xfs_file_mmap(
1038 struct file *filp,
1039 struct vm_area_struct *vma)
1040{
1041 vma->vm_ops = &xfs_file_vm_ops;
1042 vma->vm_flags |= VM_CAN_NONLINEAR;
1043
1044 file_accessed(filp);
1045 return 0;
1046}
1047
1048/*
1049 * mmap()d file has taken write protection fault and is being made
1050 * writable. We can set the page state up correctly for a writable
1051 * page, which means we can do correct delalloc accounting (ENOSPC
1052 * checking!) and unwritten extent mapping.
1053 */
1054STATIC int
1055xfs_vm_page_mkwrite(
1056 struct vm_area_struct *vma,
1057 struct vm_fault *vmf)
1058{
1059 return block_page_mkwrite(vma, vmf, xfs_get_blocks);
1060}
1061
1062const struct file_operations xfs_file_operations = {
1063 .llseek = generic_file_llseek,
1064 .read = do_sync_read,
1065 .write = do_sync_write,
1066 .aio_read = xfs_file_aio_read,
1067 .aio_write = xfs_file_aio_write,
1068 .splice_read = xfs_file_splice_read,
1069 .splice_write = xfs_file_splice_write,
1070 .unlocked_ioctl = xfs_file_ioctl,
1071#ifdef CONFIG_COMPAT
1072 .compat_ioctl = xfs_file_compat_ioctl,
1073#endif
1074 .mmap = xfs_file_mmap,
1075 .open = xfs_file_open,
1076 .release = xfs_file_release,
1077 .fsync = xfs_file_fsync,
1078 .fallocate = xfs_file_fallocate,
1079};
1080
1081const struct file_operations xfs_dir_file_operations = {
1082 .open = xfs_dir_open,
1083 .read = generic_read_dir,
1084 .readdir = xfs_file_readdir,
1085 .llseek = generic_file_llseek,
1086 .unlocked_ioctl = xfs_file_ioctl,
1087#ifdef CONFIG_COMPAT
1088 .compat_ioctl = xfs_file_compat_ioctl,
1089#endif
1090 .fsync = xfs_file_fsync,
1091};
1092
1093static const struct vm_operations_struct xfs_file_vm_ops = {
1094 .fault = filemap_fault,
1095 .page_mkwrite = xfs_vm_page_mkwrite,
1096};
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
4 * All Rights Reserved.
5 */
6#include "xfs.h"
7#include "xfs_fs.h"
8#include "xfs_shared.h"
9#include "xfs_format.h"
10#include "xfs_log_format.h"
11#include "xfs_trans_resv.h"
12#include "xfs_mount.h"
13#include "xfs_inode.h"
14#include "xfs_trans.h"
15#include "xfs_inode_item.h"
16#include "xfs_bmap.h"
17#include "xfs_bmap_util.h"
18#include "xfs_dir2.h"
19#include "xfs_dir2_priv.h"
20#include "xfs_ioctl.h"
21#include "xfs_trace.h"
22#include "xfs_log.h"
23#include "xfs_icache.h"
24#include "xfs_pnfs.h"
25#include "xfs_iomap.h"
26#include "xfs_reflink.h"
27
28#include <linux/falloc.h>
29#include <linux/backing-dev.h>
30#include <linux/mman.h>
31#include <linux/fadvise.h>
32
33static const struct vm_operations_struct xfs_file_vm_ops;
34
35int
36xfs_update_prealloc_flags(
37 struct xfs_inode *ip,
38 enum xfs_prealloc_flags flags)
39{
40 struct xfs_trans *tp;
41 int error;
42
43 error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_writeid,
44 0, 0, 0, &tp);
45 if (error)
46 return error;
47
48 xfs_ilock(ip, XFS_ILOCK_EXCL);
49 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
50
51 if (!(flags & XFS_PREALLOC_INVISIBLE)) {
52 VFS_I(ip)->i_mode &= ~S_ISUID;
53 if (VFS_I(ip)->i_mode & S_IXGRP)
54 VFS_I(ip)->i_mode &= ~S_ISGID;
55 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
56 }
57
58 if (flags & XFS_PREALLOC_SET)
59 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
60 if (flags & XFS_PREALLOC_CLEAR)
61 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
62
63 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
64 if (flags & XFS_PREALLOC_SYNC)
65 xfs_trans_set_sync(tp);
66 return xfs_trans_commit(tp);
67}
68
69/*
70 * Fsync operations on directories are much simpler than on regular files,
71 * as there is no file data to flush, and thus also no need for explicit
72 * cache flush operations, and there are no non-transaction metadata updates
73 * on directories either.
74 */
75STATIC int
76xfs_dir_fsync(
77 struct file *file,
78 loff_t start,
79 loff_t end,
80 int datasync)
81{
82 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
83 struct xfs_mount *mp = ip->i_mount;
84 xfs_lsn_t lsn = 0;
85
86 trace_xfs_dir_fsync(ip);
87
88 xfs_ilock(ip, XFS_ILOCK_SHARED);
89 if (xfs_ipincount(ip))
90 lsn = ip->i_itemp->ili_last_lsn;
91 xfs_iunlock(ip, XFS_ILOCK_SHARED);
92
93 if (!lsn)
94 return 0;
95 return xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
96}
97
98STATIC int
99xfs_file_fsync(
100 struct file *file,
101 loff_t start,
102 loff_t end,
103 int datasync)
104{
105 struct inode *inode = file->f_mapping->host;
106 struct xfs_inode *ip = XFS_I(inode);
107 struct xfs_mount *mp = ip->i_mount;
108 int error = 0;
109 int log_flushed = 0;
110 xfs_lsn_t lsn = 0;
111
112 trace_xfs_file_fsync(ip);
113
114 error = file_write_and_wait_range(file, start, end);
115 if (error)
116 return error;
117
118 if (XFS_FORCED_SHUTDOWN(mp))
119 return -EIO;
120
121 xfs_iflags_clear(ip, XFS_ITRUNCATED);
122
123 /*
124 * If we have an RT and/or log subvolume we need to make sure to flush
125 * the write cache the device used for file data first. This is to
126 * ensure newly written file data make it to disk before logging the new
127 * inode size in case of an extending write.
128 */
129 if (XFS_IS_REALTIME_INODE(ip))
130 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
131 else if (mp->m_logdev_targp != mp->m_ddev_targp)
132 xfs_blkdev_issue_flush(mp->m_ddev_targp);
133
134 /*
135 * All metadata updates are logged, which means that we just have to
136 * flush the log up to the latest LSN that touched the inode. If we have
137 * concurrent fsync/fdatasync() calls, we need them to all block on the
138 * log force before we clear the ili_fsync_fields field. This ensures
139 * that we don't get a racing sync operation that does not wait for the
140 * metadata to hit the journal before returning. If we race with
141 * clearing the ili_fsync_fields, then all that will happen is the log
142 * force will do nothing as the lsn will already be on disk. We can't
143 * race with setting ili_fsync_fields because that is done under
144 * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared
145 * until after the ili_fsync_fields is cleared.
146 */
147 xfs_ilock(ip, XFS_ILOCK_SHARED);
148 if (xfs_ipincount(ip)) {
149 if (!datasync ||
150 (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
151 lsn = ip->i_itemp->ili_last_lsn;
152 }
153
154 if (lsn) {
155 error = xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
156 ip->i_itemp->ili_fsync_fields = 0;
157 }
158 xfs_iunlock(ip, XFS_ILOCK_SHARED);
159
160 /*
161 * If we only have a single device, and the log force about was
162 * a no-op we might have to flush the data device cache here.
163 * This can only happen for fdatasync/O_DSYNC if we were overwriting
164 * an already allocated file and thus do not have any metadata to
165 * commit.
166 */
167 if (!log_flushed && !XFS_IS_REALTIME_INODE(ip) &&
168 mp->m_logdev_targp == mp->m_ddev_targp)
169 xfs_blkdev_issue_flush(mp->m_ddev_targp);
170
171 return error;
172}
173
174STATIC ssize_t
175xfs_file_dio_aio_read(
176 struct kiocb *iocb,
177 struct iov_iter *to)
178{
179 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
180 size_t count = iov_iter_count(to);
181 ssize_t ret;
182
183 trace_xfs_file_direct_read(ip, count, iocb->ki_pos);
184
185 if (!count)
186 return 0; /* skip atime */
187
188 file_accessed(iocb->ki_filp);
189
190 xfs_ilock(ip, XFS_IOLOCK_SHARED);
191 ret = iomap_dio_rw(iocb, to, &xfs_iomap_ops, NULL);
192 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
193
194 return ret;
195}
196
197static noinline ssize_t
198xfs_file_dax_read(
199 struct kiocb *iocb,
200 struct iov_iter *to)
201{
202 struct xfs_inode *ip = XFS_I(iocb->ki_filp->f_mapping->host);
203 size_t count = iov_iter_count(to);
204 ssize_t ret = 0;
205
206 trace_xfs_file_dax_read(ip, count, iocb->ki_pos);
207
208 if (!count)
209 return 0; /* skip atime */
210
211 if (iocb->ki_flags & IOCB_NOWAIT) {
212 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED))
213 return -EAGAIN;
214 } else {
215 xfs_ilock(ip, XFS_IOLOCK_SHARED);
216 }
217
218 ret = dax_iomap_rw(iocb, to, &xfs_iomap_ops);
219 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
220
221 file_accessed(iocb->ki_filp);
222 return ret;
223}
224
225STATIC ssize_t
226xfs_file_buffered_aio_read(
227 struct kiocb *iocb,
228 struct iov_iter *to)
229{
230 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
231 ssize_t ret;
232
233 trace_xfs_file_buffered_read(ip, iov_iter_count(to), iocb->ki_pos);
234
235 if (iocb->ki_flags & IOCB_NOWAIT) {
236 if (!xfs_ilock_nowait(ip, XFS_IOLOCK_SHARED))
237 return -EAGAIN;
238 } else {
239 xfs_ilock(ip, XFS_IOLOCK_SHARED);
240 }
241 ret = generic_file_read_iter(iocb, to);
242 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
243
244 return ret;
245}
246
247STATIC ssize_t
248xfs_file_read_iter(
249 struct kiocb *iocb,
250 struct iov_iter *to)
251{
252 struct inode *inode = file_inode(iocb->ki_filp);
253 struct xfs_mount *mp = XFS_I(inode)->i_mount;
254 ssize_t ret = 0;
255
256 XFS_STATS_INC(mp, xs_read_calls);
257
258 if (XFS_FORCED_SHUTDOWN(mp))
259 return -EIO;
260
261 if (IS_DAX(inode))
262 ret = xfs_file_dax_read(iocb, to);
263 else if (iocb->ki_flags & IOCB_DIRECT)
264 ret = xfs_file_dio_aio_read(iocb, to);
265 else
266 ret = xfs_file_buffered_aio_read(iocb, to);
267
268 if (ret > 0)
269 XFS_STATS_ADD(mp, xs_read_bytes, ret);
270 return ret;
271}
272
273/*
274 * Common pre-write limit and setup checks.
275 *
276 * Called with the iolocked held either shared and exclusive according to
277 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
278 * if called for a direct write beyond i_size.
279 */
280STATIC ssize_t
281xfs_file_aio_write_checks(
282 struct kiocb *iocb,
283 struct iov_iter *from,
284 int *iolock)
285{
286 struct file *file = iocb->ki_filp;
287 struct inode *inode = file->f_mapping->host;
288 struct xfs_inode *ip = XFS_I(inode);
289 ssize_t error = 0;
290 size_t count = iov_iter_count(from);
291 bool drained_dio = false;
292 loff_t isize;
293
294restart:
295 error = generic_write_checks(iocb, from);
296 if (error <= 0)
297 return error;
298
299 error = xfs_break_layouts(inode, iolock, BREAK_WRITE);
300 if (error)
301 return error;
302
303 /*
304 * For changing security info in file_remove_privs() we need i_rwsem
305 * exclusively.
306 */
307 if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) {
308 xfs_iunlock(ip, *iolock);
309 *iolock = XFS_IOLOCK_EXCL;
310 xfs_ilock(ip, *iolock);
311 goto restart;
312 }
313 /*
314 * If the offset is beyond the size of the file, we need to zero any
315 * blocks that fall between the existing EOF and the start of this
316 * write. If zeroing is needed and we are currently holding the
317 * iolock shared, we need to update it to exclusive which implies
318 * having to redo all checks before.
319 *
320 * We need to serialise against EOF updates that occur in IO
321 * completions here. We want to make sure that nobody is changing the
322 * size while we do this check until we have placed an IO barrier (i.e.
323 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
324 * The spinlock effectively forms a memory barrier once we have the
325 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
326 * and hence be able to correctly determine if we need to run zeroing.
327 */
328 spin_lock(&ip->i_flags_lock);
329 isize = i_size_read(inode);
330 if (iocb->ki_pos > isize) {
331 spin_unlock(&ip->i_flags_lock);
332 if (!drained_dio) {
333 if (*iolock == XFS_IOLOCK_SHARED) {
334 xfs_iunlock(ip, *iolock);
335 *iolock = XFS_IOLOCK_EXCL;
336 xfs_ilock(ip, *iolock);
337 iov_iter_reexpand(from, count);
338 }
339 /*
340 * We now have an IO submission barrier in place, but
341 * AIO can do EOF updates during IO completion and hence
342 * we now need to wait for all of them to drain. Non-AIO
343 * DIO will have drained before we are given the
344 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
345 * no-op.
346 */
347 inode_dio_wait(inode);
348 drained_dio = true;
349 goto restart;
350 }
351
352 trace_xfs_zero_eof(ip, isize, iocb->ki_pos - isize);
353 error = iomap_zero_range(inode, isize, iocb->ki_pos - isize,
354 NULL, &xfs_iomap_ops);
355 if (error)
356 return error;
357 } else
358 spin_unlock(&ip->i_flags_lock);
359
360 /*
361 * Updating the timestamps will grab the ilock again from
362 * xfs_fs_dirty_inode, so we have to call it after dropping the
363 * lock above. Eventually we should look into a way to avoid
364 * the pointless lock roundtrip.
365 */
366 return file_modified(file);
367}
368
369static int
370xfs_dio_write_end_io(
371 struct kiocb *iocb,
372 ssize_t size,
373 int error,
374 unsigned flags)
375{
376 struct inode *inode = file_inode(iocb->ki_filp);
377 struct xfs_inode *ip = XFS_I(inode);
378 loff_t offset = iocb->ki_pos;
379 unsigned int nofs_flag;
380
381 trace_xfs_end_io_direct_write(ip, offset, size);
382
383 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
384 return -EIO;
385
386 if (error)
387 return error;
388 if (!size)
389 return 0;
390
391 /*
392 * Capture amount written on completion as we can't reliably account
393 * for it on submission.
394 */
395 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, size);
396
397 /*
398 * We can allocate memory here while doing writeback on behalf of
399 * memory reclaim. To avoid memory allocation deadlocks set the
400 * task-wide nofs context for the following operations.
401 */
402 nofs_flag = memalloc_nofs_save();
403
404 if (flags & IOMAP_DIO_COW) {
405 error = xfs_reflink_end_cow(ip, offset, size);
406 if (error)
407 goto out;
408 }
409
410 /*
411 * Unwritten conversion updates the in-core isize after extent
412 * conversion but before updating the on-disk size. Updating isize any
413 * earlier allows a racing dio read to find unwritten extents before
414 * they are converted.
415 */
416 if (flags & IOMAP_DIO_UNWRITTEN) {
417 error = xfs_iomap_write_unwritten(ip, offset, size, true);
418 goto out;
419 }
420
421 /*
422 * We need to update the in-core inode size here so that we don't end up
423 * with the on-disk inode size being outside the in-core inode size. We
424 * have no other method of updating EOF for AIO, so always do it here
425 * if necessary.
426 *
427 * We need to lock the test/set EOF update as we can be racing with
428 * other IO completions here to update the EOF. Failing to serialise
429 * here can result in EOF moving backwards and Bad Things Happen when
430 * that occurs.
431 */
432 spin_lock(&ip->i_flags_lock);
433 if (offset + size > i_size_read(inode)) {
434 i_size_write(inode, offset + size);
435 spin_unlock(&ip->i_flags_lock);
436 error = xfs_setfilesize(ip, offset, size);
437 } else {
438 spin_unlock(&ip->i_flags_lock);
439 }
440
441out:
442 memalloc_nofs_restore(nofs_flag);
443 return error;
444}
445
446static const struct iomap_dio_ops xfs_dio_write_ops = {
447 .end_io = xfs_dio_write_end_io,
448};
449
450/*
451 * xfs_file_dio_aio_write - handle direct IO writes
452 *
453 * Lock the inode appropriately to prepare for and issue a direct IO write.
454 * By separating it from the buffered write path we remove all the tricky to
455 * follow locking changes and looping.
456 *
457 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
458 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
459 * pages are flushed out.
460 *
461 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
462 * allowing them to be done in parallel with reads and other direct IO writes.
463 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
464 * needs to do sub-block zeroing and that requires serialisation against other
465 * direct IOs to the same block. In this case we need to serialise the
466 * submission of the unaligned IOs so that we don't get racing block zeroing in
467 * the dio layer. To avoid the problem with aio, we also need to wait for
468 * outstanding IOs to complete so that unwritten extent conversion is completed
469 * before we try to map the overlapping block. This is currently implemented by
470 * hitting it with a big hammer (i.e. inode_dio_wait()).
471 *
472 * Returns with locks held indicated by @iolock and errors indicated by
473 * negative return values.
474 */
475STATIC ssize_t
476xfs_file_dio_aio_write(
477 struct kiocb *iocb,
478 struct iov_iter *from)
479{
480 struct file *file = iocb->ki_filp;
481 struct address_space *mapping = file->f_mapping;
482 struct inode *inode = mapping->host;
483 struct xfs_inode *ip = XFS_I(inode);
484 struct xfs_mount *mp = ip->i_mount;
485 ssize_t ret = 0;
486 int unaligned_io = 0;
487 int iolock;
488 size_t count = iov_iter_count(from);
489 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
490 mp->m_rtdev_targp : mp->m_ddev_targp;
491
492 /* DIO must be aligned to device logical sector size */
493 if ((iocb->ki_pos | count) & target->bt_logical_sectormask)
494 return -EINVAL;
495
496 /*
497 * Don't take the exclusive iolock here unless the I/O is unaligned to
498 * the file system block size. We don't need to consider the EOF
499 * extension case here because xfs_file_aio_write_checks() will relock
500 * the inode as necessary for EOF zeroing cases and fill out the new
501 * inode size as appropriate.
502 */
503 if ((iocb->ki_pos & mp->m_blockmask) ||
504 ((iocb->ki_pos + count) & mp->m_blockmask)) {
505 unaligned_io = 1;
506
507 /*
508 * We can't properly handle unaligned direct I/O to reflink
509 * files yet, as we can't unshare a partial block.
510 */
511 if (xfs_is_cow_inode(ip)) {
512 trace_xfs_reflink_bounce_dio_write(ip, iocb->ki_pos, count);
513 return -EREMCHG;
514 }
515 iolock = XFS_IOLOCK_EXCL;
516 } else {
517 iolock = XFS_IOLOCK_SHARED;
518 }
519
520 if (iocb->ki_flags & IOCB_NOWAIT) {
521 /* unaligned dio always waits, bail */
522 if (unaligned_io)
523 return -EAGAIN;
524 if (!xfs_ilock_nowait(ip, iolock))
525 return -EAGAIN;
526 } else {
527 xfs_ilock(ip, iolock);
528 }
529
530 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
531 if (ret)
532 goto out;
533 count = iov_iter_count(from);
534
535 /*
536 * If we are doing unaligned IO, we can't allow any other overlapping IO
537 * in-flight at the same time or we risk data corruption. Wait for all
538 * other IO to drain before we submit. If the IO is aligned, demote the
539 * iolock if we had to take the exclusive lock in
540 * xfs_file_aio_write_checks() for other reasons.
541 */
542 if (unaligned_io) {
543 inode_dio_wait(inode);
544 } else if (iolock == XFS_IOLOCK_EXCL) {
545 xfs_ilock_demote(ip, XFS_IOLOCK_EXCL);
546 iolock = XFS_IOLOCK_SHARED;
547 }
548
549 trace_xfs_file_direct_write(ip, count, iocb->ki_pos);
550 ret = iomap_dio_rw(iocb, from, &xfs_iomap_ops, &xfs_dio_write_ops);
551
552 /*
553 * If unaligned, this is the only IO in-flight. If it has not yet
554 * completed, wait on it before we release the iolock to prevent
555 * subsequent overlapping IO.
556 */
557 if (ret == -EIOCBQUEUED && unaligned_io)
558 inode_dio_wait(inode);
559out:
560 xfs_iunlock(ip, iolock);
561
562 /*
563 * No fallback to buffered IO on errors for XFS, direct IO will either
564 * complete fully or fail.
565 */
566 ASSERT(ret < 0 || ret == count);
567 return ret;
568}
569
570static noinline ssize_t
571xfs_file_dax_write(
572 struct kiocb *iocb,
573 struct iov_iter *from)
574{
575 struct inode *inode = iocb->ki_filp->f_mapping->host;
576 struct xfs_inode *ip = XFS_I(inode);
577 int iolock = XFS_IOLOCK_EXCL;
578 ssize_t ret, error = 0;
579 size_t count;
580 loff_t pos;
581
582 if (iocb->ki_flags & IOCB_NOWAIT) {
583 if (!xfs_ilock_nowait(ip, iolock))
584 return -EAGAIN;
585 } else {
586 xfs_ilock(ip, iolock);
587 }
588
589 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
590 if (ret)
591 goto out;
592
593 pos = iocb->ki_pos;
594 count = iov_iter_count(from);
595
596 trace_xfs_file_dax_write(ip, count, pos);
597 ret = dax_iomap_rw(iocb, from, &xfs_iomap_ops);
598 if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
599 i_size_write(inode, iocb->ki_pos);
600 error = xfs_setfilesize(ip, pos, ret);
601 }
602out:
603 xfs_iunlock(ip, iolock);
604 if (error)
605 return error;
606
607 if (ret > 0) {
608 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
609
610 /* Handle various SYNC-type writes */
611 ret = generic_write_sync(iocb, ret);
612 }
613 return ret;
614}
615
616STATIC ssize_t
617xfs_file_buffered_aio_write(
618 struct kiocb *iocb,
619 struct iov_iter *from)
620{
621 struct file *file = iocb->ki_filp;
622 struct address_space *mapping = file->f_mapping;
623 struct inode *inode = mapping->host;
624 struct xfs_inode *ip = XFS_I(inode);
625 ssize_t ret;
626 int enospc = 0;
627 int iolock;
628
629 if (iocb->ki_flags & IOCB_NOWAIT)
630 return -EOPNOTSUPP;
631
632write_retry:
633 iolock = XFS_IOLOCK_EXCL;
634 xfs_ilock(ip, iolock);
635
636 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
637 if (ret)
638 goto out;
639
640 /* We can write back this queue in page reclaim */
641 current->backing_dev_info = inode_to_bdi(inode);
642
643 trace_xfs_file_buffered_write(ip, iov_iter_count(from), iocb->ki_pos);
644 ret = iomap_file_buffered_write(iocb, from, &xfs_iomap_ops);
645 if (likely(ret >= 0))
646 iocb->ki_pos += ret;
647
648 /*
649 * If we hit a space limit, try to free up some lingering preallocated
650 * space before returning an error. In the case of ENOSPC, first try to
651 * write back all dirty inodes to free up some of the excess reserved
652 * metadata space. This reduces the chances that the eofblocks scan
653 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
654 * also behaves as a filter to prevent too many eofblocks scans from
655 * running at the same time.
656 */
657 if (ret == -EDQUOT && !enospc) {
658 xfs_iunlock(ip, iolock);
659 enospc = xfs_inode_free_quota_eofblocks(ip);
660 if (enospc)
661 goto write_retry;
662 enospc = xfs_inode_free_quota_cowblocks(ip);
663 if (enospc)
664 goto write_retry;
665 iolock = 0;
666 } else if (ret == -ENOSPC && !enospc) {
667 struct xfs_eofblocks eofb = {0};
668
669 enospc = 1;
670 xfs_flush_inodes(ip->i_mount);
671
672 xfs_iunlock(ip, iolock);
673 eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
674 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
675 xfs_icache_free_cowblocks(ip->i_mount, &eofb);
676 goto write_retry;
677 }
678
679 current->backing_dev_info = NULL;
680out:
681 if (iolock)
682 xfs_iunlock(ip, iolock);
683
684 if (ret > 0) {
685 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
686 /* Handle various SYNC-type writes */
687 ret = generic_write_sync(iocb, ret);
688 }
689 return ret;
690}
691
692STATIC ssize_t
693xfs_file_write_iter(
694 struct kiocb *iocb,
695 struct iov_iter *from)
696{
697 struct file *file = iocb->ki_filp;
698 struct address_space *mapping = file->f_mapping;
699 struct inode *inode = mapping->host;
700 struct xfs_inode *ip = XFS_I(inode);
701 ssize_t ret;
702 size_t ocount = iov_iter_count(from);
703
704 XFS_STATS_INC(ip->i_mount, xs_write_calls);
705
706 if (ocount == 0)
707 return 0;
708
709 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
710 return -EIO;
711
712 if (IS_DAX(inode))
713 return xfs_file_dax_write(iocb, from);
714
715 if (iocb->ki_flags & IOCB_DIRECT) {
716 /*
717 * Allow a directio write to fall back to a buffered
718 * write *only* in the case that we're doing a reflink
719 * CoW. In all other directio scenarios we do not
720 * allow an operation to fall back to buffered mode.
721 */
722 ret = xfs_file_dio_aio_write(iocb, from);
723 if (ret != -EREMCHG)
724 return ret;
725 }
726
727 return xfs_file_buffered_aio_write(iocb, from);
728}
729
730static void
731xfs_wait_dax_page(
732 struct inode *inode)
733{
734 struct xfs_inode *ip = XFS_I(inode);
735
736 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
737 schedule();
738 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
739}
740
741static int
742xfs_break_dax_layouts(
743 struct inode *inode,
744 bool *retry)
745{
746 struct page *page;
747
748 ASSERT(xfs_isilocked(XFS_I(inode), XFS_MMAPLOCK_EXCL));
749
750 page = dax_layout_busy_page(inode->i_mapping);
751 if (!page)
752 return 0;
753
754 *retry = true;
755 return ___wait_var_event(&page->_refcount,
756 atomic_read(&page->_refcount) == 1, TASK_INTERRUPTIBLE,
757 0, 0, xfs_wait_dax_page(inode));
758}
759
760int
761xfs_break_layouts(
762 struct inode *inode,
763 uint *iolock,
764 enum layout_break_reason reason)
765{
766 bool retry;
767 int error;
768
769 ASSERT(xfs_isilocked(XFS_I(inode), XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL));
770
771 do {
772 retry = false;
773 switch (reason) {
774 case BREAK_UNMAP:
775 error = xfs_break_dax_layouts(inode, &retry);
776 if (error || retry)
777 break;
778 /* fall through */
779 case BREAK_WRITE:
780 error = xfs_break_leased_layouts(inode, iolock, &retry);
781 break;
782 default:
783 WARN_ON_ONCE(1);
784 error = -EINVAL;
785 }
786 } while (error == 0 && retry);
787
788 return error;
789}
790
791#define XFS_FALLOC_FL_SUPPORTED \
792 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
793 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
794 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
795
796STATIC long
797xfs_file_fallocate(
798 struct file *file,
799 int mode,
800 loff_t offset,
801 loff_t len)
802{
803 struct inode *inode = file_inode(file);
804 struct xfs_inode *ip = XFS_I(inode);
805 long error;
806 enum xfs_prealloc_flags flags = 0;
807 uint iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
808 loff_t new_size = 0;
809 bool do_file_insert = false;
810
811 if (!S_ISREG(inode->i_mode))
812 return -EINVAL;
813 if (mode & ~XFS_FALLOC_FL_SUPPORTED)
814 return -EOPNOTSUPP;
815
816 xfs_ilock(ip, iolock);
817 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
818 if (error)
819 goto out_unlock;
820
821 if (mode & FALLOC_FL_PUNCH_HOLE) {
822 error = xfs_free_file_space(ip, offset, len);
823 if (error)
824 goto out_unlock;
825 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
826 unsigned int blksize_mask = i_blocksize(inode) - 1;
827
828 if (offset & blksize_mask || len & blksize_mask) {
829 error = -EINVAL;
830 goto out_unlock;
831 }
832
833 /*
834 * There is no need to overlap collapse range with EOF,
835 * in which case it is effectively a truncate operation
836 */
837 if (offset + len >= i_size_read(inode)) {
838 error = -EINVAL;
839 goto out_unlock;
840 }
841
842 new_size = i_size_read(inode) - len;
843
844 error = xfs_collapse_file_space(ip, offset, len);
845 if (error)
846 goto out_unlock;
847 } else if (mode & FALLOC_FL_INSERT_RANGE) {
848 unsigned int blksize_mask = i_blocksize(inode) - 1;
849 loff_t isize = i_size_read(inode);
850
851 if (offset & blksize_mask || len & blksize_mask) {
852 error = -EINVAL;
853 goto out_unlock;
854 }
855
856 /*
857 * New inode size must not exceed ->s_maxbytes, accounting for
858 * possible signed overflow.
859 */
860 if (inode->i_sb->s_maxbytes - isize < len) {
861 error = -EFBIG;
862 goto out_unlock;
863 }
864 new_size = isize + len;
865
866 /* Offset should be less than i_size */
867 if (offset >= isize) {
868 error = -EINVAL;
869 goto out_unlock;
870 }
871 do_file_insert = true;
872 } else {
873 flags |= XFS_PREALLOC_SET;
874
875 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
876 offset + len > i_size_read(inode)) {
877 new_size = offset + len;
878 error = inode_newsize_ok(inode, new_size);
879 if (error)
880 goto out_unlock;
881 }
882
883 if (mode & FALLOC_FL_ZERO_RANGE) {
884 error = xfs_zero_file_space(ip, offset, len);
885 } else if (mode & FALLOC_FL_UNSHARE_RANGE) {
886 error = xfs_reflink_unshare(ip, offset, len);
887 if (error)
888 goto out_unlock;
889
890 if (!xfs_is_always_cow_inode(ip)) {
891 error = xfs_alloc_file_space(ip, offset, len,
892 XFS_BMAPI_PREALLOC);
893 }
894 } else {
895 /*
896 * If always_cow mode we can't use preallocations and
897 * thus should not create them.
898 */
899 if (xfs_is_always_cow_inode(ip)) {
900 error = -EOPNOTSUPP;
901 goto out_unlock;
902 }
903
904 error = xfs_alloc_file_space(ip, offset, len,
905 XFS_BMAPI_PREALLOC);
906 }
907 if (error)
908 goto out_unlock;
909 }
910
911 if (file->f_flags & O_DSYNC)
912 flags |= XFS_PREALLOC_SYNC;
913
914 error = xfs_update_prealloc_flags(ip, flags);
915 if (error)
916 goto out_unlock;
917
918 /* Change file size if needed */
919 if (new_size) {
920 struct iattr iattr;
921
922 iattr.ia_valid = ATTR_SIZE;
923 iattr.ia_size = new_size;
924 error = xfs_vn_setattr_size(file_dentry(file), &iattr);
925 if (error)
926 goto out_unlock;
927 }
928
929 /*
930 * Perform hole insertion now that the file size has been
931 * updated so that if we crash during the operation we don't
932 * leave shifted extents past EOF and hence losing access to
933 * the data that is contained within them.
934 */
935 if (do_file_insert)
936 error = xfs_insert_file_space(ip, offset, len);
937
938out_unlock:
939 xfs_iunlock(ip, iolock);
940 return error;
941}
942
943STATIC int
944xfs_file_fadvise(
945 struct file *file,
946 loff_t start,
947 loff_t end,
948 int advice)
949{
950 struct xfs_inode *ip = XFS_I(file_inode(file));
951 int ret;
952 int lockflags = 0;
953
954 /*
955 * Operations creating pages in page cache need protection from hole
956 * punching and similar ops
957 */
958 if (advice == POSIX_FADV_WILLNEED) {
959 lockflags = XFS_IOLOCK_SHARED;
960 xfs_ilock(ip, lockflags);
961 }
962 ret = generic_fadvise(file, start, end, advice);
963 if (lockflags)
964 xfs_iunlock(ip, lockflags);
965 return ret;
966}
967
968STATIC loff_t
969xfs_file_remap_range(
970 struct file *file_in,
971 loff_t pos_in,
972 struct file *file_out,
973 loff_t pos_out,
974 loff_t len,
975 unsigned int remap_flags)
976{
977 struct inode *inode_in = file_inode(file_in);
978 struct xfs_inode *src = XFS_I(inode_in);
979 struct inode *inode_out = file_inode(file_out);
980 struct xfs_inode *dest = XFS_I(inode_out);
981 struct xfs_mount *mp = src->i_mount;
982 loff_t remapped = 0;
983 xfs_extlen_t cowextsize;
984 int ret;
985
986 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
987 return -EINVAL;
988
989 if (!xfs_sb_version_hasreflink(&mp->m_sb))
990 return -EOPNOTSUPP;
991
992 if (XFS_FORCED_SHUTDOWN(mp))
993 return -EIO;
994
995 /* Prepare and then clone file data. */
996 ret = xfs_reflink_remap_prep(file_in, pos_in, file_out, pos_out,
997 &len, remap_flags);
998 if (ret < 0 || len == 0)
999 return ret;
1000
1001 trace_xfs_reflink_remap_range(src, pos_in, len, dest, pos_out);
1002
1003 ret = xfs_reflink_remap_blocks(src, pos_in, dest, pos_out, len,
1004 &remapped);
1005 if (ret)
1006 goto out_unlock;
1007
1008 /*
1009 * Carry the cowextsize hint from src to dest if we're sharing the
1010 * entire source file to the entire destination file, the source file
1011 * has a cowextsize hint, and the destination file does not.
1012 */
1013 cowextsize = 0;
1014 if (pos_in == 0 && len == i_size_read(inode_in) &&
1015 (src->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE) &&
1016 pos_out == 0 && len >= i_size_read(inode_out) &&
1017 !(dest->i_d.di_flags2 & XFS_DIFLAG2_COWEXTSIZE))
1018 cowextsize = src->i_d.di_cowextsize;
1019
1020 ret = xfs_reflink_update_dest(dest, pos_out + len, cowextsize,
1021 remap_flags);
1022
1023out_unlock:
1024 xfs_reflink_remap_unlock(file_in, file_out);
1025 if (ret)
1026 trace_xfs_reflink_remap_range_error(dest, ret, _RET_IP_);
1027 return remapped > 0 ? remapped : ret;
1028}
1029
1030STATIC int
1031xfs_file_open(
1032 struct inode *inode,
1033 struct file *file)
1034{
1035 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1036 return -EFBIG;
1037 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
1038 return -EIO;
1039 file->f_mode |= FMODE_NOWAIT;
1040 return 0;
1041}
1042
1043STATIC int
1044xfs_dir_open(
1045 struct inode *inode,
1046 struct file *file)
1047{
1048 struct xfs_inode *ip = XFS_I(inode);
1049 int mode;
1050 int error;
1051
1052 error = xfs_file_open(inode, file);
1053 if (error)
1054 return error;
1055
1056 /*
1057 * If there are any blocks, read-ahead block 0 as we're almost
1058 * certain to have the next operation be a read there.
1059 */
1060 mode = xfs_ilock_data_map_shared(ip);
1061 if (ip->i_d.di_nextents > 0)
1062 error = xfs_dir3_data_readahead(ip, 0, -1);
1063 xfs_iunlock(ip, mode);
1064 return error;
1065}
1066
1067STATIC int
1068xfs_file_release(
1069 struct inode *inode,
1070 struct file *filp)
1071{
1072 return xfs_release(XFS_I(inode));
1073}
1074
1075STATIC int
1076xfs_file_readdir(
1077 struct file *file,
1078 struct dir_context *ctx)
1079{
1080 struct inode *inode = file_inode(file);
1081 xfs_inode_t *ip = XFS_I(inode);
1082 size_t bufsize;
1083
1084 /*
1085 * The Linux API doesn't pass down the total size of the buffer
1086 * we read into down to the filesystem. With the filldir concept
1087 * it's not needed for correct information, but the XFS dir2 leaf
1088 * code wants an estimate of the buffer size to calculate it's
1089 * readahead window and size the buffers used for mapping to
1090 * physical blocks.
1091 *
1092 * Try to give it an estimate that's good enough, maybe at some
1093 * point we can change the ->readdir prototype to include the
1094 * buffer size. For now we use the current glibc buffer size.
1095 */
1096 bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE, ip->i_d.di_size);
1097
1098 return xfs_readdir(NULL, ip, ctx, bufsize);
1099}
1100
1101STATIC loff_t
1102xfs_file_llseek(
1103 struct file *file,
1104 loff_t offset,
1105 int whence)
1106{
1107 struct inode *inode = file->f_mapping->host;
1108
1109 if (XFS_FORCED_SHUTDOWN(XFS_I(inode)->i_mount))
1110 return -EIO;
1111
1112 switch (whence) {
1113 default:
1114 return generic_file_llseek(file, offset, whence);
1115 case SEEK_HOLE:
1116 offset = iomap_seek_hole(inode, offset, &xfs_seek_iomap_ops);
1117 break;
1118 case SEEK_DATA:
1119 offset = iomap_seek_data(inode, offset, &xfs_seek_iomap_ops);
1120 break;
1121 }
1122
1123 if (offset < 0)
1124 return offset;
1125 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1126}
1127
1128/*
1129 * Locking for serialisation of IO during page faults. This results in a lock
1130 * ordering of:
1131 *
1132 * mmap_sem (MM)
1133 * sb_start_pagefault(vfs, freeze)
1134 * i_mmaplock (XFS - truncate serialisation)
1135 * page_lock (MM)
1136 * i_lock (XFS - extent map serialisation)
1137 */
1138static vm_fault_t
1139__xfs_filemap_fault(
1140 struct vm_fault *vmf,
1141 enum page_entry_size pe_size,
1142 bool write_fault)
1143{
1144 struct inode *inode = file_inode(vmf->vma->vm_file);
1145 struct xfs_inode *ip = XFS_I(inode);
1146 vm_fault_t ret;
1147
1148 trace_xfs_filemap_fault(ip, pe_size, write_fault);
1149
1150 if (write_fault) {
1151 sb_start_pagefault(inode->i_sb);
1152 file_update_time(vmf->vma->vm_file);
1153 }
1154
1155 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1156 if (IS_DAX(inode)) {
1157 pfn_t pfn;
1158
1159 ret = dax_iomap_fault(vmf, pe_size, &pfn, NULL, &xfs_iomap_ops);
1160 if (ret & VM_FAULT_NEEDDSYNC)
1161 ret = dax_finish_sync_fault(vmf, pe_size, pfn);
1162 } else {
1163 if (write_fault)
1164 ret = iomap_page_mkwrite(vmf, &xfs_iomap_ops);
1165 else
1166 ret = filemap_fault(vmf);
1167 }
1168 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1169
1170 if (write_fault)
1171 sb_end_pagefault(inode->i_sb);
1172 return ret;
1173}
1174
1175static vm_fault_t
1176xfs_filemap_fault(
1177 struct vm_fault *vmf)
1178{
1179 /* DAX can shortcut the normal fault path on write faults! */
1180 return __xfs_filemap_fault(vmf, PE_SIZE_PTE,
1181 IS_DAX(file_inode(vmf->vma->vm_file)) &&
1182 (vmf->flags & FAULT_FLAG_WRITE));
1183}
1184
1185static vm_fault_t
1186xfs_filemap_huge_fault(
1187 struct vm_fault *vmf,
1188 enum page_entry_size pe_size)
1189{
1190 if (!IS_DAX(file_inode(vmf->vma->vm_file)))
1191 return VM_FAULT_FALLBACK;
1192
1193 /* DAX can shortcut the normal fault path on write faults! */
1194 return __xfs_filemap_fault(vmf, pe_size,
1195 (vmf->flags & FAULT_FLAG_WRITE));
1196}
1197
1198static vm_fault_t
1199xfs_filemap_page_mkwrite(
1200 struct vm_fault *vmf)
1201{
1202 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1203}
1204
1205/*
1206 * pfn_mkwrite was originally intended to ensure we capture time stamp updates
1207 * on write faults. In reality, it needs to serialise against truncate and
1208 * prepare memory for writing so handle is as standard write fault.
1209 */
1210static vm_fault_t
1211xfs_filemap_pfn_mkwrite(
1212 struct vm_fault *vmf)
1213{
1214
1215 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1216}
1217
1218static const struct vm_operations_struct xfs_file_vm_ops = {
1219 .fault = xfs_filemap_fault,
1220 .huge_fault = xfs_filemap_huge_fault,
1221 .map_pages = filemap_map_pages,
1222 .page_mkwrite = xfs_filemap_page_mkwrite,
1223 .pfn_mkwrite = xfs_filemap_pfn_mkwrite,
1224};
1225
1226STATIC int
1227xfs_file_mmap(
1228 struct file *filp,
1229 struct vm_area_struct *vma)
1230{
1231 struct dax_device *dax_dev;
1232
1233 dax_dev = xfs_find_daxdev_for_inode(file_inode(filp));
1234 /*
1235 * We don't support synchronous mappings for non-DAX files and
1236 * for DAX files if underneath dax_device is not synchronous.
1237 */
1238 if (!daxdev_mapping_supported(vma, dax_dev))
1239 return -EOPNOTSUPP;
1240
1241 file_accessed(filp);
1242 vma->vm_ops = &xfs_file_vm_ops;
1243 if (IS_DAX(file_inode(filp)))
1244 vma->vm_flags |= VM_HUGEPAGE;
1245 return 0;
1246}
1247
1248const struct file_operations xfs_file_operations = {
1249 .llseek = xfs_file_llseek,
1250 .read_iter = xfs_file_read_iter,
1251 .write_iter = xfs_file_write_iter,
1252 .splice_read = generic_file_splice_read,
1253 .splice_write = iter_file_splice_write,
1254 .iopoll = iomap_dio_iopoll,
1255 .unlocked_ioctl = xfs_file_ioctl,
1256#ifdef CONFIG_COMPAT
1257 .compat_ioctl = xfs_file_compat_ioctl,
1258#endif
1259 .mmap = xfs_file_mmap,
1260 .mmap_supported_flags = MAP_SYNC,
1261 .open = xfs_file_open,
1262 .release = xfs_file_release,
1263 .fsync = xfs_file_fsync,
1264 .get_unmapped_area = thp_get_unmapped_area,
1265 .fallocate = xfs_file_fallocate,
1266 .fadvise = xfs_file_fadvise,
1267 .remap_file_range = xfs_file_remap_range,
1268};
1269
1270const struct file_operations xfs_dir_file_operations = {
1271 .open = xfs_dir_open,
1272 .read = generic_read_dir,
1273 .iterate_shared = xfs_file_readdir,
1274 .llseek = generic_file_llseek,
1275 .unlocked_ioctl = xfs_file_ioctl,
1276#ifdef CONFIG_COMPAT
1277 .compat_ioctl = xfs_file_compat_ioctl,
1278#endif
1279 .fsync = xfs_dir_fsync,
1280};