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#include <linux/mount.h>
33
34static const struct vm_operations_struct xfs_file_vm_ops;
35
36/*
37 * Decide if the given file range is aligned to the size of the fundamental
38 * allocation unit for the file.
39 */
40static bool
41xfs_is_falloc_aligned(
42 struct xfs_inode *ip,
43 loff_t pos,
44 long long int len)
45{
46 struct xfs_mount *mp = ip->i_mount;
47 uint64_t mask;
48
49 if (XFS_IS_REALTIME_INODE(ip)) {
50 if (!is_power_of_2(mp->m_sb.sb_rextsize)) {
51 u64 rextbytes;
52 u32 mod;
53
54 rextbytes = XFS_FSB_TO_B(mp, mp->m_sb.sb_rextsize);
55 div_u64_rem(pos, rextbytes, &mod);
56 if (mod)
57 return false;
58 div_u64_rem(len, rextbytes, &mod);
59 return mod == 0;
60 }
61 mask = XFS_FSB_TO_B(mp, mp->m_sb.sb_rextsize) - 1;
62 } else {
63 mask = mp->m_sb.sb_blocksize - 1;
64 }
65
66 return !((pos | len) & mask);
67}
68
69int
70xfs_update_prealloc_flags(
71 struct xfs_inode *ip,
72 enum xfs_prealloc_flags flags)
73{
74 struct xfs_trans *tp;
75 int error;
76
77 error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_writeid,
78 0, 0, 0, &tp);
79 if (error)
80 return error;
81
82 xfs_ilock(ip, XFS_ILOCK_EXCL);
83 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
84
85 if (!(flags & XFS_PREALLOC_INVISIBLE)) {
86 VFS_I(ip)->i_mode &= ~S_ISUID;
87 if (VFS_I(ip)->i_mode & S_IXGRP)
88 VFS_I(ip)->i_mode &= ~S_ISGID;
89 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
90 }
91
92 if (flags & XFS_PREALLOC_SET)
93 ip->i_diflags |= XFS_DIFLAG_PREALLOC;
94 if (flags & XFS_PREALLOC_CLEAR)
95 ip->i_diflags &= ~XFS_DIFLAG_PREALLOC;
96
97 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
98 if (flags & XFS_PREALLOC_SYNC)
99 xfs_trans_set_sync(tp);
100 return xfs_trans_commit(tp);
101}
102
103/*
104 * Fsync operations on directories are much simpler than on regular files,
105 * as there is no file data to flush, and thus also no need for explicit
106 * cache flush operations, and there are no non-transaction metadata updates
107 * on directories either.
108 */
109STATIC int
110xfs_dir_fsync(
111 struct file *file,
112 loff_t start,
113 loff_t end,
114 int datasync)
115{
116 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
117
118 trace_xfs_dir_fsync(ip);
119 return xfs_log_force_inode(ip);
120}
121
122static xfs_csn_t
123xfs_fsync_seq(
124 struct xfs_inode *ip,
125 bool datasync)
126{
127 if (!xfs_ipincount(ip))
128 return 0;
129 if (datasync && !(ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
130 return 0;
131 return ip->i_itemp->ili_commit_seq;
132}
133
134/*
135 * All metadata updates are logged, which means that we just have to flush the
136 * log up to the latest LSN that touched the inode.
137 *
138 * If we have concurrent fsync/fdatasync() calls, we need them to all block on
139 * the log force before we clear the ili_fsync_fields field. This ensures that
140 * we don't get a racing sync operation that does not wait for the metadata to
141 * hit the journal before returning. If we race with clearing ili_fsync_fields,
142 * then all that will happen is the log force will do nothing as the lsn will
143 * already be on disk. We can't race with setting ili_fsync_fields because that
144 * is done under XFS_ILOCK_EXCL, and that can't happen because we hold the lock
145 * shared until after the ili_fsync_fields is cleared.
146 */
147static int
148xfs_fsync_flush_log(
149 struct xfs_inode *ip,
150 bool datasync,
151 int *log_flushed)
152{
153 int error = 0;
154 xfs_csn_t seq;
155
156 xfs_ilock(ip, XFS_ILOCK_SHARED);
157 seq = xfs_fsync_seq(ip, datasync);
158 if (seq) {
159 error = xfs_log_force_seq(ip->i_mount, seq, XFS_LOG_SYNC,
160 log_flushed);
161
162 spin_lock(&ip->i_itemp->ili_lock);
163 ip->i_itemp->ili_fsync_fields = 0;
164 spin_unlock(&ip->i_itemp->ili_lock);
165 }
166 xfs_iunlock(ip, XFS_ILOCK_SHARED);
167 return error;
168}
169
170STATIC int
171xfs_file_fsync(
172 struct file *file,
173 loff_t start,
174 loff_t end,
175 int datasync)
176{
177 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
178 struct xfs_mount *mp = ip->i_mount;
179 int error = 0;
180 int log_flushed = 0;
181
182 trace_xfs_file_fsync(ip);
183
184 error = file_write_and_wait_range(file, start, end);
185 if (error)
186 return error;
187
188 if (XFS_FORCED_SHUTDOWN(mp))
189 return -EIO;
190
191 xfs_iflags_clear(ip, XFS_ITRUNCATED);
192
193 /*
194 * If we have an RT and/or log subvolume we need to make sure to flush
195 * the write cache the device used for file data first. This is to
196 * ensure newly written file data make it to disk before logging the new
197 * inode size in case of an extending write.
198 */
199 if (XFS_IS_REALTIME_INODE(ip))
200 blkdev_issue_flush(mp->m_rtdev_targp->bt_bdev);
201 else if (mp->m_logdev_targp != mp->m_ddev_targp)
202 blkdev_issue_flush(mp->m_ddev_targp->bt_bdev);
203
204 /*
205 * Any inode that has dirty modifications in the log is pinned. The
206 * racy check here for a pinned inode while not catch modifications
207 * that happen concurrently to the fsync call, but fsync semantics
208 * only require to sync previously completed I/O.
209 */
210 if (xfs_ipincount(ip))
211 error = xfs_fsync_flush_log(ip, datasync, &log_flushed);
212
213 /*
214 * If we only have a single device, and the log force about was
215 * a no-op we might have to flush the data device cache here.
216 * This can only happen for fdatasync/O_DSYNC if we were overwriting
217 * an already allocated file and thus do not have any metadata to
218 * commit.
219 */
220 if (!log_flushed && !XFS_IS_REALTIME_INODE(ip) &&
221 mp->m_logdev_targp == mp->m_ddev_targp)
222 blkdev_issue_flush(mp->m_ddev_targp->bt_bdev);
223
224 return error;
225}
226
227static int
228xfs_ilock_iocb(
229 struct kiocb *iocb,
230 unsigned int lock_mode)
231{
232 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
233
234 if (iocb->ki_flags & IOCB_NOWAIT) {
235 if (!xfs_ilock_nowait(ip, lock_mode))
236 return -EAGAIN;
237 } else {
238 xfs_ilock(ip, lock_mode);
239 }
240
241 return 0;
242}
243
244STATIC ssize_t
245xfs_file_dio_read(
246 struct kiocb *iocb,
247 struct iov_iter *to)
248{
249 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
250 ssize_t ret;
251
252 trace_xfs_file_direct_read(iocb, to);
253
254 if (!iov_iter_count(to))
255 return 0; /* skip atime */
256
257 file_accessed(iocb->ki_filp);
258
259 ret = xfs_ilock_iocb(iocb, XFS_IOLOCK_SHARED);
260 if (ret)
261 return ret;
262 ret = iomap_dio_rw(iocb, to, &xfs_read_iomap_ops, NULL, 0);
263 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
264
265 return ret;
266}
267
268static noinline ssize_t
269xfs_file_dax_read(
270 struct kiocb *iocb,
271 struct iov_iter *to)
272{
273 struct xfs_inode *ip = XFS_I(iocb->ki_filp->f_mapping->host);
274 ssize_t ret = 0;
275
276 trace_xfs_file_dax_read(iocb, to);
277
278 if (!iov_iter_count(to))
279 return 0; /* skip atime */
280
281 ret = xfs_ilock_iocb(iocb, XFS_IOLOCK_SHARED);
282 if (ret)
283 return ret;
284 ret = dax_iomap_rw(iocb, to, &xfs_read_iomap_ops);
285 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
286
287 file_accessed(iocb->ki_filp);
288 return ret;
289}
290
291STATIC ssize_t
292xfs_file_buffered_read(
293 struct kiocb *iocb,
294 struct iov_iter *to)
295{
296 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
297 ssize_t ret;
298
299 trace_xfs_file_buffered_read(iocb, to);
300
301 ret = xfs_ilock_iocb(iocb, XFS_IOLOCK_SHARED);
302 if (ret)
303 return ret;
304 ret = generic_file_read_iter(iocb, to);
305 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
306
307 return ret;
308}
309
310STATIC ssize_t
311xfs_file_read_iter(
312 struct kiocb *iocb,
313 struct iov_iter *to)
314{
315 struct inode *inode = file_inode(iocb->ki_filp);
316 struct xfs_mount *mp = XFS_I(inode)->i_mount;
317 ssize_t ret = 0;
318
319 XFS_STATS_INC(mp, xs_read_calls);
320
321 if (XFS_FORCED_SHUTDOWN(mp))
322 return -EIO;
323
324 if (IS_DAX(inode))
325 ret = xfs_file_dax_read(iocb, to);
326 else if (iocb->ki_flags & IOCB_DIRECT)
327 ret = xfs_file_dio_read(iocb, to);
328 else
329 ret = xfs_file_buffered_read(iocb, to);
330
331 if (ret > 0)
332 XFS_STATS_ADD(mp, xs_read_bytes, ret);
333 return ret;
334}
335
336/*
337 * Common pre-write limit and setup checks.
338 *
339 * Called with the iolocked held either shared and exclusive according to
340 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
341 * if called for a direct write beyond i_size.
342 */
343STATIC ssize_t
344xfs_file_write_checks(
345 struct kiocb *iocb,
346 struct iov_iter *from,
347 int *iolock)
348{
349 struct file *file = iocb->ki_filp;
350 struct inode *inode = file->f_mapping->host;
351 struct xfs_inode *ip = XFS_I(inode);
352 ssize_t error = 0;
353 size_t count = iov_iter_count(from);
354 bool drained_dio = false;
355 loff_t isize;
356
357restart:
358 error = generic_write_checks(iocb, from);
359 if (error <= 0)
360 return error;
361
362 if (iocb->ki_flags & IOCB_NOWAIT) {
363 error = break_layout(inode, false);
364 if (error == -EWOULDBLOCK)
365 error = -EAGAIN;
366 } else {
367 error = xfs_break_layouts(inode, iolock, BREAK_WRITE);
368 }
369
370 if (error)
371 return error;
372
373 /*
374 * For changing security info in file_remove_privs() we need i_rwsem
375 * exclusively.
376 */
377 if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) {
378 xfs_iunlock(ip, *iolock);
379 *iolock = XFS_IOLOCK_EXCL;
380 error = xfs_ilock_iocb(iocb, *iolock);
381 if (error) {
382 *iolock = 0;
383 return error;
384 }
385 goto restart;
386 }
387
388 /*
389 * If the offset is beyond the size of the file, we need to zero any
390 * blocks that fall between the existing EOF and the start of this
391 * write. If zeroing is needed and we are currently holding the iolock
392 * shared, we need to update it to exclusive which implies having to
393 * redo all checks before.
394 *
395 * We need to serialise against EOF updates that occur in IO completions
396 * here. We want to make sure that nobody is changing the size while we
397 * do this check until we have placed an IO barrier (i.e. hold the
398 * XFS_IOLOCK_EXCL) that prevents new IO from being dispatched. The
399 * spinlock effectively forms a memory barrier once we have the
400 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value and
401 * hence be able to correctly determine if we need to run zeroing.
402 *
403 * We can do an unlocked check here safely as IO completion can only
404 * extend EOF. Truncate is locked out at this point, so the EOF can
405 * not move backwards, only forwards. Hence we only need to take the
406 * slow path and spin locks when we are at or beyond the current EOF.
407 */
408 if (iocb->ki_pos <= i_size_read(inode))
409 goto out;
410
411 spin_lock(&ip->i_flags_lock);
412 isize = i_size_read(inode);
413 if (iocb->ki_pos > isize) {
414 spin_unlock(&ip->i_flags_lock);
415
416 if (iocb->ki_flags & IOCB_NOWAIT)
417 return -EAGAIN;
418
419 if (!drained_dio) {
420 if (*iolock == XFS_IOLOCK_SHARED) {
421 xfs_iunlock(ip, *iolock);
422 *iolock = XFS_IOLOCK_EXCL;
423 xfs_ilock(ip, *iolock);
424 iov_iter_reexpand(from, count);
425 }
426 /*
427 * We now have an IO submission barrier in place, but
428 * AIO can do EOF updates during IO completion and hence
429 * we now need to wait for all of them to drain. Non-AIO
430 * DIO will have drained before we are given the
431 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
432 * no-op.
433 */
434 inode_dio_wait(inode);
435 drained_dio = true;
436 goto restart;
437 }
438
439 trace_xfs_zero_eof(ip, isize, iocb->ki_pos - isize);
440 error = iomap_zero_range(inode, isize, iocb->ki_pos - isize,
441 NULL, &xfs_buffered_write_iomap_ops);
442 if (error)
443 return error;
444 } else
445 spin_unlock(&ip->i_flags_lock);
446
447out:
448 return file_modified(file);
449}
450
451static int
452xfs_dio_write_end_io(
453 struct kiocb *iocb,
454 ssize_t size,
455 int error,
456 unsigned flags)
457{
458 struct inode *inode = file_inode(iocb->ki_filp);
459 struct xfs_inode *ip = XFS_I(inode);
460 loff_t offset = iocb->ki_pos;
461 unsigned int nofs_flag;
462
463 trace_xfs_end_io_direct_write(ip, offset, size);
464
465 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
466 return -EIO;
467
468 if (error)
469 return error;
470 if (!size)
471 return 0;
472
473 /*
474 * Capture amount written on completion as we can't reliably account
475 * for it on submission.
476 */
477 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, size);
478
479 /*
480 * We can allocate memory here while doing writeback on behalf of
481 * memory reclaim. To avoid memory allocation deadlocks set the
482 * task-wide nofs context for the following operations.
483 */
484 nofs_flag = memalloc_nofs_save();
485
486 if (flags & IOMAP_DIO_COW) {
487 error = xfs_reflink_end_cow(ip, offset, size);
488 if (error)
489 goto out;
490 }
491
492 /*
493 * Unwritten conversion updates the in-core isize after extent
494 * conversion but before updating the on-disk size. Updating isize any
495 * earlier allows a racing dio read to find unwritten extents before
496 * they are converted.
497 */
498 if (flags & IOMAP_DIO_UNWRITTEN) {
499 error = xfs_iomap_write_unwritten(ip, offset, size, true);
500 goto out;
501 }
502
503 /*
504 * We need to update the in-core inode size here so that we don't end up
505 * with the on-disk inode size being outside the in-core inode size. We
506 * have no other method of updating EOF for AIO, so always do it here
507 * if necessary.
508 *
509 * We need to lock the test/set EOF update as we can be racing with
510 * other IO completions here to update the EOF. Failing to serialise
511 * here can result in EOF moving backwards and Bad Things Happen when
512 * that occurs.
513 *
514 * As IO completion only ever extends EOF, we can do an unlocked check
515 * here to avoid taking the spinlock. If we land within the current EOF,
516 * then we do not need to do an extending update at all, and we don't
517 * need to take the lock to check this. If we race with an update moving
518 * EOF, then we'll either still be beyond EOF and need to take the lock,
519 * or we'll be within EOF and we don't need to take it at all.
520 */
521 if (offset + size <= i_size_read(inode))
522 goto out;
523
524 spin_lock(&ip->i_flags_lock);
525 if (offset + size > i_size_read(inode)) {
526 i_size_write(inode, offset + size);
527 spin_unlock(&ip->i_flags_lock);
528 error = xfs_setfilesize(ip, offset, size);
529 } else {
530 spin_unlock(&ip->i_flags_lock);
531 }
532
533out:
534 memalloc_nofs_restore(nofs_flag);
535 return error;
536}
537
538static const struct iomap_dio_ops xfs_dio_write_ops = {
539 .end_io = xfs_dio_write_end_io,
540};
541
542/*
543 * Handle block aligned direct I/O writes
544 */
545static noinline ssize_t
546xfs_file_dio_write_aligned(
547 struct xfs_inode *ip,
548 struct kiocb *iocb,
549 struct iov_iter *from)
550{
551 int iolock = XFS_IOLOCK_SHARED;
552 ssize_t ret;
553
554 ret = xfs_ilock_iocb(iocb, iolock);
555 if (ret)
556 return ret;
557 ret = xfs_file_write_checks(iocb, from, &iolock);
558 if (ret)
559 goto out_unlock;
560
561 /*
562 * We don't need to hold the IOLOCK exclusively across the IO, so demote
563 * the iolock back to shared if we had to take the exclusive lock in
564 * xfs_file_write_checks() for other reasons.
565 */
566 if (iolock == XFS_IOLOCK_EXCL) {
567 xfs_ilock_demote(ip, XFS_IOLOCK_EXCL);
568 iolock = XFS_IOLOCK_SHARED;
569 }
570 trace_xfs_file_direct_write(iocb, from);
571 ret = iomap_dio_rw(iocb, from, &xfs_direct_write_iomap_ops,
572 &xfs_dio_write_ops, 0);
573out_unlock:
574 if (iolock)
575 xfs_iunlock(ip, iolock);
576 return ret;
577}
578
579/*
580 * Handle block unaligned direct I/O writes
581 *
582 * In most cases direct I/O writes will be done holding IOLOCK_SHARED, allowing
583 * them to be done in parallel with reads and other direct I/O writes. However,
584 * if the I/O is not aligned to filesystem blocks, the direct I/O layer may need
585 * to do sub-block zeroing and that requires serialisation against other direct
586 * I/O to the same block. In this case we need to serialise the submission of
587 * the unaligned I/O so that we don't get racing block zeroing in the dio layer.
588 * In the case where sub-block zeroing is not required, we can do concurrent
589 * sub-block dios to the same block successfully.
590 *
591 * Optimistically submit the I/O using the shared lock first, but use the
592 * IOMAP_DIO_OVERWRITE_ONLY flag to tell the lower layers to return -EAGAIN
593 * if block allocation or partial block zeroing would be required. In that case
594 * we try again with the exclusive lock.
595 */
596static noinline ssize_t
597xfs_file_dio_write_unaligned(
598 struct xfs_inode *ip,
599 struct kiocb *iocb,
600 struct iov_iter *from)
601{
602 size_t isize = i_size_read(VFS_I(ip));
603 size_t count = iov_iter_count(from);
604 int iolock = XFS_IOLOCK_SHARED;
605 unsigned int flags = IOMAP_DIO_OVERWRITE_ONLY;
606 ssize_t ret;
607
608 /*
609 * Extending writes need exclusivity because of the sub-block zeroing
610 * that the DIO code always does for partial tail blocks beyond EOF, so
611 * don't even bother trying the fast path in this case.
612 */
613 if (iocb->ki_pos > isize || iocb->ki_pos + count >= isize) {
614retry_exclusive:
615 if (iocb->ki_flags & IOCB_NOWAIT)
616 return -EAGAIN;
617 iolock = XFS_IOLOCK_EXCL;
618 flags = IOMAP_DIO_FORCE_WAIT;
619 }
620
621 ret = xfs_ilock_iocb(iocb, iolock);
622 if (ret)
623 return ret;
624
625 /*
626 * We can't properly handle unaligned direct I/O to reflink files yet,
627 * as we can't unshare a partial block.
628 */
629 if (xfs_is_cow_inode(ip)) {
630 trace_xfs_reflink_bounce_dio_write(iocb, from);
631 ret = -ENOTBLK;
632 goto out_unlock;
633 }
634
635 ret = xfs_file_write_checks(iocb, from, &iolock);
636 if (ret)
637 goto out_unlock;
638
639 /*
640 * If we are doing exclusive unaligned I/O, this must be the only I/O
641 * in-flight. Otherwise we risk data corruption due to unwritten extent
642 * conversions from the AIO end_io handler. Wait for all other I/O to
643 * drain first.
644 */
645 if (flags & IOMAP_DIO_FORCE_WAIT)
646 inode_dio_wait(VFS_I(ip));
647
648 trace_xfs_file_direct_write(iocb, from);
649 ret = iomap_dio_rw(iocb, from, &xfs_direct_write_iomap_ops,
650 &xfs_dio_write_ops, flags);
651
652 /*
653 * Retry unaligned I/O with exclusive blocking semantics if the DIO
654 * layer rejected it for mapping or locking reasons. If we are doing
655 * nonblocking user I/O, propagate the error.
656 */
657 if (ret == -EAGAIN && !(iocb->ki_flags & IOCB_NOWAIT)) {
658 ASSERT(flags & IOMAP_DIO_OVERWRITE_ONLY);
659 xfs_iunlock(ip, iolock);
660 goto retry_exclusive;
661 }
662
663out_unlock:
664 if (iolock)
665 xfs_iunlock(ip, iolock);
666 return ret;
667}
668
669static ssize_t
670xfs_file_dio_write(
671 struct kiocb *iocb,
672 struct iov_iter *from)
673{
674 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
675 struct xfs_buftarg *target = xfs_inode_buftarg(ip);
676 size_t count = iov_iter_count(from);
677
678 /* direct I/O must be aligned to device logical sector size */
679 if ((iocb->ki_pos | count) & target->bt_logical_sectormask)
680 return -EINVAL;
681 if ((iocb->ki_pos | count) & ip->i_mount->m_blockmask)
682 return xfs_file_dio_write_unaligned(ip, iocb, from);
683 return xfs_file_dio_write_aligned(ip, iocb, from);
684}
685
686static noinline ssize_t
687xfs_file_dax_write(
688 struct kiocb *iocb,
689 struct iov_iter *from)
690{
691 struct inode *inode = iocb->ki_filp->f_mapping->host;
692 struct xfs_inode *ip = XFS_I(inode);
693 int iolock = XFS_IOLOCK_EXCL;
694 ssize_t ret, error = 0;
695 loff_t pos;
696
697 ret = xfs_ilock_iocb(iocb, iolock);
698 if (ret)
699 return ret;
700 ret = xfs_file_write_checks(iocb, from, &iolock);
701 if (ret)
702 goto out;
703
704 pos = iocb->ki_pos;
705
706 trace_xfs_file_dax_write(iocb, from);
707 ret = dax_iomap_rw(iocb, from, &xfs_direct_write_iomap_ops);
708 if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
709 i_size_write(inode, iocb->ki_pos);
710 error = xfs_setfilesize(ip, pos, ret);
711 }
712out:
713 if (iolock)
714 xfs_iunlock(ip, iolock);
715 if (error)
716 return error;
717
718 if (ret > 0) {
719 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
720
721 /* Handle various SYNC-type writes */
722 ret = generic_write_sync(iocb, ret);
723 }
724 return ret;
725}
726
727STATIC ssize_t
728xfs_file_buffered_write(
729 struct kiocb *iocb,
730 struct iov_iter *from)
731{
732 struct file *file = iocb->ki_filp;
733 struct address_space *mapping = file->f_mapping;
734 struct inode *inode = mapping->host;
735 struct xfs_inode *ip = XFS_I(inode);
736 ssize_t ret;
737 bool cleared_space = false;
738 int iolock;
739
740 if (iocb->ki_flags & IOCB_NOWAIT)
741 return -EOPNOTSUPP;
742
743write_retry:
744 iolock = XFS_IOLOCK_EXCL;
745 xfs_ilock(ip, iolock);
746
747 ret = xfs_file_write_checks(iocb, from, &iolock);
748 if (ret)
749 goto out;
750
751 /* We can write back this queue in page reclaim */
752 current->backing_dev_info = inode_to_bdi(inode);
753
754 trace_xfs_file_buffered_write(iocb, from);
755 ret = iomap_file_buffered_write(iocb, from,
756 &xfs_buffered_write_iomap_ops);
757 if (likely(ret >= 0))
758 iocb->ki_pos += ret;
759
760 /*
761 * If we hit a space limit, try to free up some lingering preallocated
762 * space before returning an error. In the case of ENOSPC, first try to
763 * write back all dirty inodes to free up some of the excess reserved
764 * metadata space. This reduces the chances that the eofblocks scan
765 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
766 * also behaves as a filter to prevent too many eofblocks scans from
767 * running at the same time. Use a synchronous scan to increase the
768 * effectiveness of the scan.
769 */
770 if (ret == -EDQUOT && !cleared_space) {
771 xfs_iunlock(ip, iolock);
772 xfs_blockgc_free_quota(ip, XFS_ICWALK_FLAG_SYNC);
773 cleared_space = true;
774 goto write_retry;
775 } else if (ret == -ENOSPC && !cleared_space) {
776 struct xfs_icwalk icw = {0};
777
778 cleared_space = true;
779 xfs_flush_inodes(ip->i_mount);
780
781 xfs_iunlock(ip, iolock);
782 icw.icw_flags = XFS_ICWALK_FLAG_SYNC;
783 xfs_blockgc_free_space(ip->i_mount, &icw);
784 goto write_retry;
785 }
786
787 current->backing_dev_info = NULL;
788out:
789 if (iolock)
790 xfs_iunlock(ip, iolock);
791
792 if (ret > 0) {
793 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
794 /* Handle various SYNC-type writes */
795 ret = generic_write_sync(iocb, ret);
796 }
797 return ret;
798}
799
800STATIC ssize_t
801xfs_file_write_iter(
802 struct kiocb *iocb,
803 struct iov_iter *from)
804{
805 struct file *file = iocb->ki_filp;
806 struct address_space *mapping = file->f_mapping;
807 struct inode *inode = mapping->host;
808 struct xfs_inode *ip = XFS_I(inode);
809 ssize_t ret;
810 size_t ocount = iov_iter_count(from);
811
812 XFS_STATS_INC(ip->i_mount, xs_write_calls);
813
814 if (ocount == 0)
815 return 0;
816
817 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
818 return -EIO;
819
820 if (IS_DAX(inode))
821 return xfs_file_dax_write(iocb, from);
822
823 if (iocb->ki_flags & IOCB_DIRECT) {
824 /*
825 * Allow a directio write to fall back to a buffered
826 * write *only* in the case that we're doing a reflink
827 * CoW. In all other directio scenarios we do not
828 * allow an operation to fall back to buffered mode.
829 */
830 ret = xfs_file_dio_write(iocb, from);
831 if (ret != -ENOTBLK)
832 return ret;
833 }
834
835 return xfs_file_buffered_write(iocb, from);
836}
837
838static void
839xfs_wait_dax_page(
840 struct inode *inode)
841{
842 struct xfs_inode *ip = XFS_I(inode);
843
844 xfs_iunlock(ip, XFS_MMAPLOCK_EXCL);
845 schedule();
846 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
847}
848
849static int
850xfs_break_dax_layouts(
851 struct inode *inode,
852 bool *retry)
853{
854 struct page *page;
855
856 ASSERT(xfs_isilocked(XFS_I(inode), XFS_MMAPLOCK_EXCL));
857
858 page = dax_layout_busy_page(inode->i_mapping);
859 if (!page)
860 return 0;
861
862 *retry = true;
863 return ___wait_var_event(&page->_refcount,
864 atomic_read(&page->_refcount) == 1, TASK_INTERRUPTIBLE,
865 0, 0, xfs_wait_dax_page(inode));
866}
867
868int
869xfs_break_layouts(
870 struct inode *inode,
871 uint *iolock,
872 enum layout_break_reason reason)
873{
874 bool retry;
875 int error;
876
877 ASSERT(xfs_isilocked(XFS_I(inode), XFS_IOLOCK_SHARED|XFS_IOLOCK_EXCL));
878
879 do {
880 retry = false;
881 switch (reason) {
882 case BREAK_UNMAP:
883 error = xfs_break_dax_layouts(inode, &retry);
884 if (error || retry)
885 break;
886 fallthrough;
887 case BREAK_WRITE:
888 error = xfs_break_leased_layouts(inode, iolock, &retry);
889 break;
890 default:
891 WARN_ON_ONCE(1);
892 error = -EINVAL;
893 }
894 } while (error == 0 && retry);
895
896 return error;
897}
898
899#define XFS_FALLOC_FL_SUPPORTED \
900 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
901 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
902 FALLOC_FL_INSERT_RANGE | FALLOC_FL_UNSHARE_RANGE)
903
904STATIC long
905xfs_file_fallocate(
906 struct file *file,
907 int mode,
908 loff_t offset,
909 loff_t len)
910{
911 struct inode *inode = file_inode(file);
912 struct xfs_inode *ip = XFS_I(inode);
913 long error;
914 enum xfs_prealloc_flags flags = 0;
915 uint iolock = XFS_IOLOCK_EXCL | XFS_MMAPLOCK_EXCL;
916 loff_t new_size = 0;
917 bool do_file_insert = false;
918
919 if (!S_ISREG(inode->i_mode))
920 return -EINVAL;
921 if (mode & ~XFS_FALLOC_FL_SUPPORTED)
922 return -EOPNOTSUPP;
923
924 xfs_ilock(ip, iolock);
925 error = xfs_break_layouts(inode, &iolock, BREAK_UNMAP);
926 if (error)
927 goto out_unlock;
928
929 /*
930 * Must wait for all AIO to complete before we continue as AIO can
931 * change the file size on completion without holding any locks we
932 * currently hold. We must do this first because AIO can update both
933 * the on disk and in memory inode sizes, and the operations that follow
934 * require the in-memory size to be fully up-to-date.
935 */
936 inode_dio_wait(inode);
937
938 /*
939 * Now AIO and DIO has drained we flush and (if necessary) invalidate
940 * the cached range over the first operation we are about to run.
941 *
942 * We care about zero and collapse here because they both run a hole
943 * punch over the range first. Because that can zero data, and the range
944 * of invalidation for the shift operations is much larger, we still do
945 * the required flush for collapse in xfs_prepare_shift().
946 *
947 * Insert has the same range requirements as collapse, and we extend the
948 * file first which can zero data. Hence insert has the same
949 * flush/invalidate requirements as collapse and so they are both
950 * handled at the right time by xfs_prepare_shift().
951 */
952 if (mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE |
953 FALLOC_FL_COLLAPSE_RANGE)) {
954 error = xfs_flush_unmap_range(ip, offset, len);
955 if (error)
956 goto out_unlock;
957 }
958
959 if (mode & FALLOC_FL_PUNCH_HOLE) {
960 error = xfs_free_file_space(ip, offset, len);
961 if (error)
962 goto out_unlock;
963 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
964 if (!xfs_is_falloc_aligned(ip, offset, len)) {
965 error = -EINVAL;
966 goto out_unlock;
967 }
968
969 /*
970 * There is no need to overlap collapse range with EOF,
971 * in which case it is effectively a truncate operation
972 */
973 if (offset + len >= i_size_read(inode)) {
974 error = -EINVAL;
975 goto out_unlock;
976 }
977
978 new_size = i_size_read(inode) - len;
979
980 error = xfs_collapse_file_space(ip, offset, len);
981 if (error)
982 goto out_unlock;
983 } else if (mode & FALLOC_FL_INSERT_RANGE) {
984 loff_t isize = i_size_read(inode);
985
986 if (!xfs_is_falloc_aligned(ip, offset, len)) {
987 error = -EINVAL;
988 goto out_unlock;
989 }
990
991 /*
992 * New inode size must not exceed ->s_maxbytes, accounting for
993 * possible signed overflow.
994 */
995 if (inode->i_sb->s_maxbytes - isize < len) {
996 error = -EFBIG;
997 goto out_unlock;
998 }
999 new_size = isize + len;
1000
1001 /* Offset should be less than i_size */
1002 if (offset >= isize) {
1003 error = -EINVAL;
1004 goto out_unlock;
1005 }
1006 do_file_insert = true;
1007 } else {
1008 flags |= XFS_PREALLOC_SET;
1009
1010 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
1011 offset + len > i_size_read(inode)) {
1012 new_size = offset + len;
1013 error = inode_newsize_ok(inode, new_size);
1014 if (error)
1015 goto out_unlock;
1016 }
1017
1018 if (mode & FALLOC_FL_ZERO_RANGE) {
1019 /*
1020 * Punch a hole and prealloc the range. We use a hole
1021 * punch rather than unwritten extent conversion for two
1022 * reasons:
1023 *
1024 * 1.) Hole punch handles partial block zeroing for us.
1025 * 2.) If prealloc returns ENOSPC, the file range is
1026 * still zero-valued by virtue of the hole punch.
1027 */
1028 unsigned int blksize = i_blocksize(inode);
1029
1030 trace_xfs_zero_file_space(ip);
1031
1032 error = xfs_free_file_space(ip, offset, len);
1033 if (error)
1034 goto out_unlock;
1035
1036 len = round_up(offset + len, blksize) -
1037 round_down(offset, blksize);
1038 offset = round_down(offset, blksize);
1039 } else if (mode & FALLOC_FL_UNSHARE_RANGE) {
1040 error = xfs_reflink_unshare(ip, offset, len);
1041 if (error)
1042 goto out_unlock;
1043 } else {
1044 /*
1045 * If always_cow mode we can't use preallocations and
1046 * thus should not create them.
1047 */
1048 if (xfs_is_always_cow_inode(ip)) {
1049 error = -EOPNOTSUPP;
1050 goto out_unlock;
1051 }
1052 }
1053
1054 if (!xfs_is_always_cow_inode(ip)) {
1055 error = xfs_alloc_file_space(ip, offset, len,
1056 XFS_BMAPI_PREALLOC);
1057 if (error)
1058 goto out_unlock;
1059 }
1060 }
1061
1062 if (file->f_flags & O_DSYNC)
1063 flags |= XFS_PREALLOC_SYNC;
1064
1065 error = xfs_update_prealloc_flags(ip, flags);
1066 if (error)
1067 goto out_unlock;
1068
1069 /* Change file size if needed */
1070 if (new_size) {
1071 struct iattr iattr;
1072
1073 iattr.ia_valid = ATTR_SIZE;
1074 iattr.ia_size = new_size;
1075 error = xfs_vn_setattr_size(file_mnt_user_ns(file),
1076 file_dentry(file), &iattr);
1077 if (error)
1078 goto out_unlock;
1079 }
1080
1081 /*
1082 * Perform hole insertion now that the file size has been
1083 * updated so that if we crash during the operation we don't
1084 * leave shifted extents past EOF and hence losing access to
1085 * the data that is contained within them.
1086 */
1087 if (do_file_insert)
1088 error = xfs_insert_file_space(ip, offset, len);
1089
1090out_unlock:
1091 xfs_iunlock(ip, iolock);
1092 return error;
1093}
1094
1095STATIC int
1096xfs_file_fadvise(
1097 struct file *file,
1098 loff_t start,
1099 loff_t end,
1100 int advice)
1101{
1102 struct xfs_inode *ip = XFS_I(file_inode(file));
1103 int ret;
1104 int lockflags = 0;
1105
1106 /*
1107 * Operations creating pages in page cache need protection from hole
1108 * punching and similar ops
1109 */
1110 if (advice == POSIX_FADV_WILLNEED) {
1111 lockflags = XFS_IOLOCK_SHARED;
1112 xfs_ilock(ip, lockflags);
1113 }
1114 ret = generic_fadvise(file, start, end, advice);
1115 if (lockflags)
1116 xfs_iunlock(ip, lockflags);
1117 return ret;
1118}
1119
1120/* Does this file, inode, or mount want synchronous writes? */
1121static inline bool xfs_file_sync_writes(struct file *filp)
1122{
1123 struct xfs_inode *ip = XFS_I(file_inode(filp));
1124
1125 if (ip->i_mount->m_flags & XFS_MOUNT_WSYNC)
1126 return true;
1127 if (filp->f_flags & (__O_SYNC | O_DSYNC))
1128 return true;
1129 if (IS_SYNC(file_inode(filp)))
1130 return true;
1131
1132 return false;
1133}
1134
1135STATIC loff_t
1136xfs_file_remap_range(
1137 struct file *file_in,
1138 loff_t pos_in,
1139 struct file *file_out,
1140 loff_t pos_out,
1141 loff_t len,
1142 unsigned int remap_flags)
1143{
1144 struct inode *inode_in = file_inode(file_in);
1145 struct xfs_inode *src = XFS_I(inode_in);
1146 struct inode *inode_out = file_inode(file_out);
1147 struct xfs_inode *dest = XFS_I(inode_out);
1148 struct xfs_mount *mp = src->i_mount;
1149 loff_t remapped = 0;
1150 xfs_extlen_t cowextsize;
1151 int ret;
1152
1153 if (remap_flags & ~(REMAP_FILE_DEDUP | REMAP_FILE_ADVISORY))
1154 return -EINVAL;
1155
1156 if (!xfs_sb_version_hasreflink(&mp->m_sb))
1157 return -EOPNOTSUPP;
1158
1159 if (XFS_FORCED_SHUTDOWN(mp))
1160 return -EIO;
1161
1162 /* Prepare and then clone file data. */
1163 ret = xfs_reflink_remap_prep(file_in, pos_in, file_out, pos_out,
1164 &len, remap_flags);
1165 if (ret || len == 0)
1166 return ret;
1167
1168 trace_xfs_reflink_remap_range(src, pos_in, len, dest, pos_out);
1169
1170 ret = xfs_reflink_remap_blocks(src, pos_in, dest, pos_out, len,
1171 &remapped);
1172 if (ret)
1173 goto out_unlock;
1174
1175 /*
1176 * Carry the cowextsize hint from src to dest if we're sharing the
1177 * entire source file to the entire destination file, the source file
1178 * has a cowextsize hint, and the destination file does not.
1179 */
1180 cowextsize = 0;
1181 if (pos_in == 0 && len == i_size_read(inode_in) &&
1182 (src->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE) &&
1183 pos_out == 0 && len >= i_size_read(inode_out) &&
1184 !(dest->i_diflags2 & XFS_DIFLAG2_COWEXTSIZE))
1185 cowextsize = src->i_cowextsize;
1186
1187 ret = xfs_reflink_update_dest(dest, pos_out + len, cowextsize,
1188 remap_flags);
1189 if (ret)
1190 goto out_unlock;
1191
1192 if (xfs_file_sync_writes(file_in) || xfs_file_sync_writes(file_out))
1193 xfs_log_force_inode(dest);
1194out_unlock:
1195 xfs_iunlock2_io_mmap(src, dest);
1196 if (ret)
1197 trace_xfs_reflink_remap_range_error(dest, ret, _RET_IP_);
1198 return remapped > 0 ? remapped : ret;
1199}
1200
1201STATIC int
1202xfs_file_open(
1203 struct inode *inode,
1204 struct file *file)
1205{
1206 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
1207 return -EFBIG;
1208 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
1209 return -EIO;
1210 file->f_mode |= FMODE_NOWAIT | FMODE_BUF_RASYNC;
1211 return 0;
1212}
1213
1214STATIC int
1215xfs_dir_open(
1216 struct inode *inode,
1217 struct file *file)
1218{
1219 struct xfs_inode *ip = XFS_I(inode);
1220 int mode;
1221 int error;
1222
1223 error = xfs_file_open(inode, file);
1224 if (error)
1225 return error;
1226
1227 /*
1228 * If there are any blocks, read-ahead block 0 as we're almost
1229 * certain to have the next operation be a read there.
1230 */
1231 mode = xfs_ilock_data_map_shared(ip);
1232 if (ip->i_df.if_nextents > 0)
1233 error = xfs_dir3_data_readahead(ip, 0, 0);
1234 xfs_iunlock(ip, mode);
1235 return error;
1236}
1237
1238STATIC int
1239xfs_file_release(
1240 struct inode *inode,
1241 struct file *filp)
1242{
1243 return xfs_release(XFS_I(inode));
1244}
1245
1246STATIC int
1247xfs_file_readdir(
1248 struct file *file,
1249 struct dir_context *ctx)
1250{
1251 struct inode *inode = file_inode(file);
1252 xfs_inode_t *ip = XFS_I(inode);
1253 size_t bufsize;
1254
1255 /*
1256 * The Linux API doesn't pass down the total size of the buffer
1257 * we read into down to the filesystem. With the filldir concept
1258 * it's not needed for correct information, but the XFS dir2 leaf
1259 * code wants an estimate of the buffer size to calculate it's
1260 * readahead window and size the buffers used for mapping to
1261 * physical blocks.
1262 *
1263 * Try to give it an estimate that's good enough, maybe at some
1264 * point we can change the ->readdir prototype to include the
1265 * buffer size. For now we use the current glibc buffer size.
1266 */
1267 bufsize = (size_t)min_t(loff_t, XFS_READDIR_BUFSIZE, ip->i_disk_size);
1268
1269 return xfs_readdir(NULL, ip, ctx, bufsize);
1270}
1271
1272STATIC loff_t
1273xfs_file_llseek(
1274 struct file *file,
1275 loff_t offset,
1276 int whence)
1277{
1278 struct inode *inode = file->f_mapping->host;
1279
1280 if (XFS_FORCED_SHUTDOWN(XFS_I(inode)->i_mount))
1281 return -EIO;
1282
1283 switch (whence) {
1284 default:
1285 return generic_file_llseek(file, offset, whence);
1286 case SEEK_HOLE:
1287 offset = iomap_seek_hole(inode, offset, &xfs_seek_iomap_ops);
1288 break;
1289 case SEEK_DATA:
1290 offset = iomap_seek_data(inode, offset, &xfs_seek_iomap_ops);
1291 break;
1292 }
1293
1294 if (offset < 0)
1295 return offset;
1296 return vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
1297}
1298
1299/*
1300 * Locking for serialisation of IO during page faults. This results in a lock
1301 * ordering of:
1302 *
1303 * mmap_lock (MM)
1304 * sb_start_pagefault(vfs, freeze)
1305 * i_mmaplock (XFS - truncate serialisation)
1306 * page_lock (MM)
1307 * i_lock (XFS - extent map serialisation)
1308 */
1309static vm_fault_t
1310__xfs_filemap_fault(
1311 struct vm_fault *vmf,
1312 enum page_entry_size pe_size,
1313 bool write_fault)
1314{
1315 struct inode *inode = file_inode(vmf->vma->vm_file);
1316 struct xfs_inode *ip = XFS_I(inode);
1317 vm_fault_t ret;
1318
1319 trace_xfs_filemap_fault(ip, pe_size, write_fault);
1320
1321 if (write_fault) {
1322 sb_start_pagefault(inode->i_sb);
1323 file_update_time(vmf->vma->vm_file);
1324 }
1325
1326 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1327 if (IS_DAX(inode)) {
1328 pfn_t pfn;
1329
1330 ret = dax_iomap_fault(vmf, pe_size, &pfn, NULL,
1331 (write_fault && !vmf->cow_page) ?
1332 &xfs_direct_write_iomap_ops :
1333 &xfs_read_iomap_ops);
1334 if (ret & VM_FAULT_NEEDDSYNC)
1335 ret = dax_finish_sync_fault(vmf, pe_size, pfn);
1336 } else {
1337 if (write_fault)
1338 ret = iomap_page_mkwrite(vmf,
1339 &xfs_buffered_write_iomap_ops);
1340 else
1341 ret = filemap_fault(vmf);
1342 }
1343 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1344
1345 if (write_fault)
1346 sb_end_pagefault(inode->i_sb);
1347 return ret;
1348}
1349
1350static inline bool
1351xfs_is_write_fault(
1352 struct vm_fault *vmf)
1353{
1354 return (vmf->flags & FAULT_FLAG_WRITE) &&
1355 (vmf->vma->vm_flags & VM_SHARED);
1356}
1357
1358static vm_fault_t
1359xfs_filemap_fault(
1360 struct vm_fault *vmf)
1361{
1362 /* DAX can shortcut the normal fault path on write faults! */
1363 return __xfs_filemap_fault(vmf, PE_SIZE_PTE,
1364 IS_DAX(file_inode(vmf->vma->vm_file)) &&
1365 xfs_is_write_fault(vmf));
1366}
1367
1368static vm_fault_t
1369xfs_filemap_huge_fault(
1370 struct vm_fault *vmf,
1371 enum page_entry_size pe_size)
1372{
1373 if (!IS_DAX(file_inode(vmf->vma->vm_file)))
1374 return VM_FAULT_FALLBACK;
1375
1376 /* DAX can shortcut the normal fault path on write faults! */
1377 return __xfs_filemap_fault(vmf, pe_size,
1378 xfs_is_write_fault(vmf));
1379}
1380
1381static vm_fault_t
1382xfs_filemap_page_mkwrite(
1383 struct vm_fault *vmf)
1384{
1385 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1386}
1387
1388/*
1389 * pfn_mkwrite was originally intended to ensure we capture time stamp updates
1390 * on write faults. In reality, it needs to serialise against truncate and
1391 * prepare memory for writing so handle is as standard write fault.
1392 */
1393static vm_fault_t
1394xfs_filemap_pfn_mkwrite(
1395 struct vm_fault *vmf)
1396{
1397
1398 return __xfs_filemap_fault(vmf, PE_SIZE_PTE, true);
1399}
1400
1401static vm_fault_t
1402xfs_filemap_map_pages(
1403 struct vm_fault *vmf,
1404 pgoff_t start_pgoff,
1405 pgoff_t end_pgoff)
1406{
1407 struct inode *inode = file_inode(vmf->vma->vm_file);
1408 vm_fault_t ret;
1409
1410 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1411 ret = filemap_map_pages(vmf, start_pgoff, end_pgoff);
1412 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1413 return ret;
1414}
1415
1416static const struct vm_operations_struct xfs_file_vm_ops = {
1417 .fault = xfs_filemap_fault,
1418 .huge_fault = xfs_filemap_huge_fault,
1419 .map_pages = xfs_filemap_map_pages,
1420 .page_mkwrite = xfs_filemap_page_mkwrite,
1421 .pfn_mkwrite = xfs_filemap_pfn_mkwrite,
1422};
1423
1424STATIC int
1425xfs_file_mmap(
1426 struct file *file,
1427 struct vm_area_struct *vma)
1428{
1429 struct inode *inode = file_inode(file);
1430 struct xfs_buftarg *target = xfs_inode_buftarg(XFS_I(inode));
1431
1432 /*
1433 * We don't support synchronous mappings for non-DAX files and
1434 * for DAX files if underneath dax_device is not synchronous.
1435 */
1436 if (!daxdev_mapping_supported(vma, target->bt_daxdev))
1437 return -EOPNOTSUPP;
1438
1439 file_accessed(file);
1440 vma->vm_ops = &xfs_file_vm_ops;
1441 if (IS_DAX(inode))
1442 vma->vm_flags |= VM_HUGEPAGE;
1443 return 0;
1444}
1445
1446const struct file_operations xfs_file_operations = {
1447 .llseek = xfs_file_llseek,
1448 .read_iter = xfs_file_read_iter,
1449 .write_iter = xfs_file_write_iter,
1450 .splice_read = generic_file_splice_read,
1451 .splice_write = iter_file_splice_write,
1452 .iopoll = iomap_dio_iopoll,
1453 .unlocked_ioctl = xfs_file_ioctl,
1454#ifdef CONFIG_COMPAT
1455 .compat_ioctl = xfs_file_compat_ioctl,
1456#endif
1457 .mmap = xfs_file_mmap,
1458 .mmap_supported_flags = MAP_SYNC,
1459 .open = xfs_file_open,
1460 .release = xfs_file_release,
1461 .fsync = xfs_file_fsync,
1462 .get_unmapped_area = thp_get_unmapped_area,
1463 .fallocate = xfs_file_fallocate,
1464 .fadvise = xfs_file_fadvise,
1465 .remap_file_range = xfs_file_remap_range,
1466};
1467
1468const struct file_operations xfs_dir_file_operations = {
1469 .open = xfs_dir_open,
1470 .read = generic_read_dir,
1471 .iterate_shared = xfs_file_readdir,
1472 .llseek = generic_file_llseek,
1473 .unlocked_ioctl = xfs_file_ioctl,
1474#ifdef CONFIG_COMPAT
1475 .compat_ioctl = xfs_file_compat_ioctl,
1476#endif
1477 .fsync = xfs_dir_fsync,
1478};