blob: f03bf1a456fbcacf43cc4999124a5d9072d733f0 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
Nathan Scott7b718762005-11-02 14:58:39 +11002 * Copyright (c) 2000-2005 Silicon Graphics, Inc.
3 * All Rights Reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 *
Nathan Scott7b718762005-11-02 14:58:39 +11005 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License as
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 * published by the Free Software Foundation.
8 *
Nathan Scott7b718762005-11-02 14:58:39 +11009 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
Nathan Scott7b718762005-11-02 14:58:39 +110014 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070017 */
Linus Torvalds1da177e2005-04-16 15:20:36 -070018#include "xfs.h"
Christoph Hellwigdda35b82010-02-15 09:44:46 +000019#include "xfs_fs.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs_log.h"
21#include "xfs_sb.h"
Nathan Scotta844f452005-11-02 14:38:42 +110022#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include "xfs_trans.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_mount.h"
25#include "xfs_bmap_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_dinode.h"
28#include "xfs_inode.h"
Christoph Hellwigfd3200b2010-02-15 09:44:48 +000029#include "xfs_inode_item.h"
Christoph Hellwigdda35b82010-02-15 09:44:46 +000030#include "xfs_bmap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xfs_error.h"
Christoph Hellwig739bfb22007-08-29 10:58:01 +100032#include "xfs_vnodeops.h"
Christoph Hellwigf999a5b2008-11-28 14:23:32 +110033#include "xfs_da_btree.h"
Dave Chinnerda6958c2012-11-12 22:54:18 +110034#include "xfs_dir2_format.h"
35#include "xfs_dir2_priv.h"
Christoph Hellwigddcd8562008-12-03 07:55:34 -050036#include "xfs_ioctl.h"
Christoph Hellwigdda35b82010-02-15 09:44:46 +000037#include "xfs_trace.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
39#include <linux/dcache.h>
Christoph Hellwig2fe17c12011-01-14 13:07:43 +010040#include <linux/falloc.h>
Jeff Liud126d432012-08-21 17:11:57 +080041#include <linux/pagevec.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +040043static const struct vm_operations_struct xfs_file_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
Christoph Hellwigdda35b82010-02-15 09:44:46 +000045/*
Dave Chinner487f84f2011-01-12 11:37:10 +110046 * Locking primitives for read and write IO paths to ensure we consistently use
47 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
48 */
49static inline void
50xfs_rw_ilock(
51 struct xfs_inode *ip,
52 int type)
53{
54 if (type & XFS_IOLOCK_EXCL)
55 mutex_lock(&VFS_I(ip)->i_mutex);
56 xfs_ilock(ip, type);
57}
58
59static inline void
60xfs_rw_iunlock(
61 struct xfs_inode *ip,
62 int type)
63{
64 xfs_iunlock(ip, type);
65 if (type & XFS_IOLOCK_EXCL)
66 mutex_unlock(&VFS_I(ip)->i_mutex);
67}
68
69static inline void
70xfs_rw_ilock_demote(
71 struct xfs_inode *ip,
72 int type)
73{
74 xfs_ilock_demote(ip, type);
75 if (type & XFS_IOLOCK_EXCL)
76 mutex_unlock(&VFS_I(ip)->i_mutex);
77}
78
79/*
Christoph Hellwigdda35b82010-02-15 09:44:46 +000080 * xfs_iozero
81 *
82 * xfs_iozero clears the specified range of buffer supplied,
83 * and marks all the affected blocks as valid and modified. If
84 * an affected block is not allocated, it will be allocated. If
85 * an affected block is not completely overwritten, and is not
86 * valid before the operation, it will be read from disk before
87 * being partially zeroed.
88 */
Dave Chinneref9d8732012-11-29 15:26:33 +110089int
Christoph Hellwigdda35b82010-02-15 09:44:46 +000090xfs_iozero(
91 struct xfs_inode *ip, /* inode */
92 loff_t pos, /* offset in file */
93 size_t count) /* size of data to zero */
94{
95 struct page *page;
96 struct address_space *mapping;
97 int status;
98
99 mapping = VFS_I(ip)->i_mapping;
100 do {
101 unsigned offset, bytes;
102 void *fsdata;
103
104 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
105 bytes = PAGE_CACHE_SIZE - offset;
106 if (bytes > count)
107 bytes = count;
108
109 status = pagecache_write_begin(NULL, mapping, pos, bytes,
110 AOP_FLAG_UNINTERRUPTIBLE,
111 &page, &fsdata);
112 if (status)
113 break;
114
115 zero_user(page, offset, bytes);
116
117 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
118 page, fsdata);
119 WARN_ON(status <= 0); /* can't return less than zero! */
120 pos += bytes;
121 count -= bytes;
122 status = 0;
123 } while (count);
124
125 return (-status);
126}
127
Christoph Hellwig1da2f2d2011-10-02 14:25:16 +0000128/*
129 * Fsync operations on directories are much simpler than on regular files,
130 * as there is no file data to flush, and thus also no need for explicit
131 * cache flush operations, and there are no non-transaction metadata updates
132 * on directories either.
133 */
134STATIC int
135xfs_dir_fsync(
136 struct file *file,
137 loff_t start,
138 loff_t end,
139 int datasync)
140{
141 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
142 struct xfs_mount *mp = ip->i_mount;
143 xfs_lsn_t lsn = 0;
144
145 trace_xfs_dir_fsync(ip);
146
147 xfs_ilock(ip, XFS_ILOCK_SHARED);
148 if (xfs_ipincount(ip))
149 lsn = ip->i_itemp->ili_last_lsn;
150 xfs_iunlock(ip, XFS_ILOCK_SHARED);
151
152 if (!lsn)
153 return 0;
154 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
155}
156
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000157STATIC int
158xfs_file_fsync(
159 struct file *file,
Josef Bacik02c24a82011-07-16 20:44:56 -0400160 loff_t start,
161 loff_t end,
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000162 int datasync)
163{
Christoph Hellwig7ea80852010-05-26 17:53:25 +0200164 struct inode *inode = file->f_mapping->host;
165 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000166 struct xfs_mount *mp = ip->i_mount;
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000167 int error = 0;
168 int log_flushed = 0;
Christoph Hellwigb1037052011-09-19 14:55:51 +0000169 xfs_lsn_t lsn = 0;
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000170
Christoph Hellwigcca28fb2010-06-24 11:57:09 +1000171 trace_xfs_file_fsync(ip);
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000172
Josef Bacik02c24a82011-07-16 20:44:56 -0400173 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
174 if (error)
175 return error;
176
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000177 if (XFS_FORCED_SHUTDOWN(mp))
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000178 return -XFS_ERROR(EIO);
179
180 xfs_iflags_clear(ip, XFS_ITRUNCATED);
181
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000182 if (mp->m_flags & XFS_MOUNT_BARRIER) {
183 /*
184 * If we have an RT and/or log subvolume we need to make sure
185 * to flush the write cache the device used for file data
186 * first. This is to ensure newly written file data make
187 * it to disk before logging the new inode size in case of
188 * an extending write.
189 */
190 if (XFS_IS_REALTIME_INODE(ip))
191 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
192 else if (mp->m_logdev_targp != mp->m_ddev_targp)
193 xfs_blkdev_issue_flush(mp->m_ddev_targp);
194 }
195
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000196 /*
Christoph Hellwig8a9c9982012-02-29 09:53:52 +0000197 * All metadata updates are logged, which means that we just have
198 * to flush the log up to the latest LSN that touched the inode.
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000199 */
200 xfs_ilock(ip, XFS_ILOCK_SHARED);
Christoph Hellwig8f639dd2012-02-29 09:53:55 +0000201 if (xfs_ipincount(ip)) {
202 if (!datasync ||
203 (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP))
204 lsn = ip->i_itemp->ili_last_lsn;
205 }
Christoph Hellwig8a9c9982012-02-29 09:53:52 +0000206 xfs_iunlock(ip, XFS_ILOCK_SHARED);
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000207
Christoph Hellwig8a9c9982012-02-29 09:53:52 +0000208 if (lsn)
Christoph Hellwigb1037052011-09-19 14:55:51 +0000209 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
210
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000211 /*
212 * If we only have a single device, and the log force about was
213 * a no-op we might have to flush the data device cache here.
214 * This can only happen for fdatasync/O_DSYNC if we were overwriting
215 * an already allocated file and thus do not have any metadata to
216 * commit.
217 */
218 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
219 mp->m_logdev_targp == mp->m_ddev_targp &&
220 !XFS_IS_REALTIME_INODE(ip) &&
221 !log_flushed)
222 xfs_blkdev_issue_flush(mp->m_ddev_targp);
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000223
224 return -error;
225}
226
Christoph Hellwig00258e32010-02-15 09:44:47 +0000227STATIC ssize_t
228xfs_file_aio_read(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000229 struct kiocb *iocb,
230 const struct iovec *iovp,
Christoph Hellwig00258e32010-02-15 09:44:47 +0000231 unsigned long nr_segs,
232 loff_t pos)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000233{
234 struct file *file = iocb->ki_filp;
235 struct inode *inode = file->f_mapping->host;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000236 struct xfs_inode *ip = XFS_I(inode);
237 struct xfs_mount *mp = ip->i_mount;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000238 size_t size = 0;
239 ssize_t ret = 0;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000240 int ioflags = 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000241 xfs_fsize_t n;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000242
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000243 XFS_STATS_INC(xs_read_calls);
244
Christoph Hellwig00258e32010-02-15 09:44:47 +0000245 BUG_ON(iocb->ki_pos != pos);
246
247 if (unlikely(file->f_flags & O_DIRECT))
248 ioflags |= IO_ISDIRECT;
249 if (file->f_mode & FMODE_NOCMTIME)
250 ioflags |= IO_INVIS;
251
Dave Chinner52764322012-06-08 15:45:44 +1000252 ret = generic_segment_checks(iovp, &nr_segs, &size, VERIFY_WRITE);
253 if (ret < 0)
254 return ret;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000255
256 if (unlikely(ioflags & IO_ISDIRECT)) {
257 xfs_buftarg_t *target =
258 XFS_IS_REALTIME_INODE(ip) ?
259 mp->m_rtdev_targp : mp->m_ddev_targp;
Dave Chinnerfb595812012-11-12 22:53:57 +1100260 if ((pos & target->bt_smask) || (size & target->bt_smask)) {
261 if (pos == i_size_read(inode))
Christoph Hellwig00258e32010-02-15 09:44:47 +0000262 return 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000263 return -XFS_ERROR(EINVAL);
264 }
265 }
266
Dave Chinnerfb595812012-11-12 22:53:57 +1100267 n = mp->m_super->s_maxbytes - pos;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000268 if (n <= 0 || size == 0)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000269 return 0;
270
271 if (n < size)
272 size = n;
273
274 if (XFS_FORCED_SHUTDOWN(mp))
275 return -EIO;
276
Dave Chinner0c38a252011-08-25 07:17:01 +0000277 /*
278 * Locking is a bit tricky here. If we take an exclusive lock
279 * for direct IO, we effectively serialise all new concurrent
280 * read IO to this file and block it behind IO that is currently in
281 * progress because IO in progress holds the IO lock shared. We only
282 * need to hold the lock exclusive to blow away the page cache, so
283 * only take lock exclusively if the page cache needs invalidation.
284 * This allows the normal direct IO case of no page cache pages to
285 * proceeed concurrently without serialisation.
286 */
287 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
288 if ((ioflags & IO_ISDIRECT) && inode->i_mapping->nrpages) {
289 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
Dave Chinner487f84f2011-01-12 11:37:10 +1100290 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
291
Christoph Hellwig00258e32010-02-15 09:44:47 +0000292 if (inode->i_mapping->nrpages) {
Dave Chinnerfb595812012-11-12 22:53:57 +1100293 ret = -filemap_write_and_wait_range(
294 VFS_I(ip)->i_mapping,
295 pos, -1);
Dave Chinner487f84f2011-01-12 11:37:10 +1100296 if (ret) {
297 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
298 return ret;
299 }
Dave Chinnerfb595812012-11-12 22:53:57 +1100300 truncate_pagecache_range(VFS_I(ip), pos, -1);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000301 }
Dave Chinner487f84f2011-01-12 11:37:10 +1100302 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
Dave Chinner0c38a252011-08-25 07:17:01 +0000303 }
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000304
Dave Chinnerfb595812012-11-12 22:53:57 +1100305 trace_xfs_file_read(ip, size, pos, ioflags);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000306
Dave Chinnerfb595812012-11-12 22:53:57 +1100307 ret = generic_file_aio_read(iocb, iovp, nr_segs, pos);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000308 if (ret > 0)
309 XFS_STATS_ADD(xs_read_bytes, ret);
310
Dave Chinner487f84f2011-01-12 11:37:10 +1100311 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000312 return ret;
313}
314
Christoph Hellwig00258e32010-02-15 09:44:47 +0000315STATIC ssize_t
316xfs_file_splice_read(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000317 struct file *infilp,
318 loff_t *ppos,
319 struct pipe_inode_info *pipe,
320 size_t count,
Christoph Hellwig00258e32010-02-15 09:44:47 +0000321 unsigned int flags)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000322{
Christoph Hellwig00258e32010-02-15 09:44:47 +0000323 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000324 int ioflags = 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000325 ssize_t ret;
326
327 XFS_STATS_INC(xs_read_calls);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000328
329 if (infilp->f_mode & FMODE_NOCMTIME)
330 ioflags |= IO_INVIS;
331
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000332 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
333 return -EIO;
334
Dave Chinner487f84f2011-01-12 11:37:10 +1100335 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000336
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000337 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
338
339 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
340 if (ret > 0)
341 XFS_STATS_ADD(xs_read_bytes, ret);
342
Dave Chinner487f84f2011-01-12 11:37:10 +1100343 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000344 return ret;
345}
346
Dave Chinner4c5cfd12011-01-11 10:14:16 +1100347/*
Dave Chinner487f84f2011-01-12 11:37:10 +1100348 * xfs_file_splice_write() does not use xfs_rw_ilock() because
349 * generic_file_splice_write() takes the i_mutex itself. This, in theory,
350 * couuld cause lock inversions between the aio_write path and the splice path
351 * if someone is doing concurrent splice(2) based writes and write(2) based
352 * writes to the same inode. The only real way to fix this is to re-implement
353 * the generic code here with correct locking orders.
354 */
Christoph Hellwig00258e32010-02-15 09:44:47 +0000355STATIC ssize_t
356xfs_file_splice_write(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000357 struct pipe_inode_info *pipe,
358 struct file *outfilp,
359 loff_t *ppos,
360 size_t count,
Christoph Hellwig00258e32010-02-15 09:44:47 +0000361 unsigned int flags)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000362{
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000363 struct inode *inode = outfilp->f_mapping->host;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000364 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000365 int ioflags = 0;
366 ssize_t ret;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000367
368 XFS_STATS_INC(xs_write_calls);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000369
370 if (outfilp->f_mode & FMODE_NOCMTIME)
371 ioflags |= IO_INVIS;
372
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000373 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
374 return -EIO;
375
376 xfs_ilock(ip, XFS_IOLOCK_EXCL);
377
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000378 trace_xfs_file_splice_write(ip, count, *ppos, ioflags);
379
380 ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000381 if (ret > 0)
382 XFS_STATS_ADD(xs_write_bytes, ret);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000383
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000384 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
385 return ret;
386}
387
388/*
Christoph Hellwig193aec12012-03-27 10:34:49 -0400389 * This routine is called to handle zeroing any space in the last block of the
390 * file that is beyond the EOF. We do this since the size is being increased
391 * without writing anything to that block and we don't want to read the
392 * garbage on the disk.
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000393 */
394STATIC int /* error (positive) */
395xfs_zero_last_block(
Christoph Hellwig193aec12012-03-27 10:34:49 -0400396 struct xfs_inode *ip,
397 xfs_fsize_t offset,
398 xfs_fsize_t isize)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000399{
Christoph Hellwig193aec12012-03-27 10:34:49 -0400400 struct xfs_mount *mp = ip->i_mount;
401 xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize);
402 int zero_offset = XFS_B_FSB_OFFSET(mp, isize);
403 int zero_len;
404 int nimaps = 1;
405 int error = 0;
406 struct xfs_bmbt_irec imap;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000407
Christoph Hellwig193aec12012-03-27 10:34:49 -0400408 xfs_ilock(ip, XFS_ILOCK_EXCL);
Dave Chinner5c8ed202011-09-18 20:40:45 +0000409 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
Christoph Hellwig193aec12012-03-27 10:34:49 -0400410 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Dave Chinner5c8ed202011-09-18 20:40:45 +0000411 if (error)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000412 return error;
Christoph Hellwig193aec12012-03-27 10:34:49 -0400413
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000414 ASSERT(nimaps > 0);
Christoph Hellwig193aec12012-03-27 10:34:49 -0400415
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000416 /*
417 * If the block underlying isize is just a hole, then there
418 * is nothing to zero.
419 */
Christoph Hellwig193aec12012-03-27 10:34:49 -0400420 if (imap.br_startblock == HOLESTARTBLOCK)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000421 return 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000422
423 zero_len = mp->m_sb.sb_blocksize - zero_offset;
424 if (isize + zero_len > offset)
425 zero_len = offset - isize;
Christoph Hellwig193aec12012-03-27 10:34:49 -0400426 return xfs_iozero(ip, isize, zero_len);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000427}
428
429/*
Christoph Hellwig193aec12012-03-27 10:34:49 -0400430 * Zero any on disk space between the current EOF and the new, larger EOF.
431 *
432 * This handles the normal case of zeroing the remainder of the last block in
433 * the file and the unusual case of zeroing blocks out beyond the size of the
434 * file. This second case only happens with fixed size extents and when the
435 * system crashes before the inode size was updated but after blocks were
436 * allocated.
437 *
438 * Expects the iolock to be held exclusive, and will take the ilock internally.
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000439 */
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000440int /* error (positive) */
441xfs_zero_eof(
Christoph Hellwig193aec12012-03-27 10:34:49 -0400442 struct xfs_inode *ip,
443 xfs_off_t offset, /* starting I/O offset */
444 xfs_fsize_t isize) /* current inode size */
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000445{
Christoph Hellwig193aec12012-03-27 10:34:49 -0400446 struct xfs_mount *mp = ip->i_mount;
447 xfs_fileoff_t start_zero_fsb;
448 xfs_fileoff_t end_zero_fsb;
449 xfs_fileoff_t zero_count_fsb;
450 xfs_fileoff_t last_fsb;
451 xfs_fileoff_t zero_off;
452 xfs_fsize_t zero_len;
453 int nimaps;
454 int error = 0;
455 struct xfs_bmbt_irec imap;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000456
Christoph Hellwig193aec12012-03-27 10:34:49 -0400457 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000458 ASSERT(offset > isize);
459
460 /*
461 * First handle zeroing the block on which isize resides.
Christoph Hellwig193aec12012-03-27 10:34:49 -0400462 *
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000463 * We only zero a part of that block so it is handled specially.
464 */
Christoph Hellwig193aec12012-03-27 10:34:49 -0400465 if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
466 error = xfs_zero_last_block(ip, offset, isize);
467 if (error)
468 return error;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000469 }
470
471 /*
Christoph Hellwig193aec12012-03-27 10:34:49 -0400472 * Calculate the range between the new size and the old where blocks
473 * needing to be zeroed may exist.
474 *
475 * To get the block where the last byte in the file currently resides,
476 * we need to subtract one from the size and truncate back to a block
477 * boundary. We subtract 1 in case the size is exactly on a block
478 * boundary.
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000479 */
480 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
481 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
482 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
483 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
484 if (last_fsb == end_zero_fsb) {
485 /*
486 * The size was only incremented on its last block.
487 * We took care of that above, so just return.
488 */
489 return 0;
490 }
491
492 ASSERT(start_zero_fsb <= end_zero_fsb);
493 while (start_zero_fsb <= end_zero_fsb) {
494 nimaps = 1;
495 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
Christoph Hellwig193aec12012-03-27 10:34:49 -0400496
497 xfs_ilock(ip, XFS_ILOCK_EXCL);
Dave Chinner5c8ed202011-09-18 20:40:45 +0000498 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
499 &imap, &nimaps, 0);
Christoph Hellwig193aec12012-03-27 10:34:49 -0400500 xfs_iunlock(ip, XFS_ILOCK_EXCL);
501 if (error)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000502 return error;
Christoph Hellwig193aec12012-03-27 10:34:49 -0400503
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000504 ASSERT(nimaps > 0);
505
506 if (imap.br_state == XFS_EXT_UNWRITTEN ||
507 imap.br_startblock == HOLESTARTBLOCK) {
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000508 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
509 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
510 continue;
511 }
512
513 /*
514 * There are blocks we need to zero.
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000515 */
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000516 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
517 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
518
519 if ((zero_off + zero_len) > offset)
520 zero_len = offset - zero_off;
521
522 error = xfs_iozero(ip, zero_off, zero_len);
Christoph Hellwig193aec12012-03-27 10:34:49 -0400523 if (error)
524 return error;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000525
526 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
527 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000528 }
529
530 return 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000531}
532
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100533/*
Dave Chinner4d8d1582011-01-11 10:23:42 +1100534 * Common pre-write limit and setup checks.
535 *
Christoph Hellwig5bf1f262011-12-18 20:00:13 +0000536 * Called with the iolocked held either shared and exclusive according to
537 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
538 * if called for a direct write beyond i_size.
Dave Chinner4d8d1582011-01-11 10:23:42 +1100539 */
540STATIC ssize_t
541xfs_file_aio_write_checks(
542 struct file *file,
543 loff_t *pos,
544 size_t *count,
545 int *iolock)
546{
547 struct inode *inode = file->f_mapping->host;
548 struct xfs_inode *ip = XFS_I(inode);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100549 int error = 0;
550
Dave Chinner7271d242011-08-25 07:17:02 +0000551restart:
Dave Chinner4d8d1582011-01-11 10:23:42 +1100552 error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
Christoph Hellwig467f7892012-03-27 10:34:47 -0400553 if (error)
Dave Chinner4d8d1582011-01-11 10:23:42 +1100554 return error;
Dave Chinner4d8d1582011-01-11 10:23:42 +1100555
Dave Chinner4d8d1582011-01-11 10:23:42 +1100556 /*
557 * If the offset is beyond the size of the file, we need to zero any
558 * blocks that fall between the existing EOF and the start of this
Christoph Hellwig2813d682011-12-18 20:00:12 +0000559 * write. If zeroing is needed and we are currently holding the
Christoph Hellwig467f7892012-03-27 10:34:47 -0400560 * iolock shared, we need to update it to exclusive which implies
561 * having to redo all checks before.
Dave Chinner4d8d1582011-01-11 10:23:42 +1100562 */
Christoph Hellwig2813d682011-12-18 20:00:12 +0000563 if (*pos > i_size_read(inode)) {
Dave Chinner7271d242011-08-25 07:17:02 +0000564 if (*iolock == XFS_IOLOCK_SHARED) {
Christoph Hellwig467f7892012-03-27 10:34:47 -0400565 xfs_rw_iunlock(ip, *iolock);
Dave Chinner7271d242011-08-25 07:17:02 +0000566 *iolock = XFS_IOLOCK_EXCL;
Christoph Hellwig467f7892012-03-27 10:34:47 -0400567 xfs_rw_ilock(ip, *iolock);
Dave Chinner7271d242011-08-25 07:17:02 +0000568 goto restart;
569 }
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000570 error = -xfs_zero_eof(ip, *pos, i_size_read(inode));
Christoph Hellwig467f7892012-03-27 10:34:47 -0400571 if (error)
572 return error;
Dave Chinner7271d242011-08-25 07:17:02 +0000573 }
Dave Chinner4d8d1582011-01-11 10:23:42 +1100574
575 /*
Christoph Hellwig8a9c9982012-02-29 09:53:52 +0000576 * Updating the timestamps will grab the ilock again from
577 * xfs_fs_dirty_inode, so we have to call it after dropping the
578 * lock above. Eventually we should look into a way to avoid
579 * the pointless lock roundtrip.
580 */
Josef Bacikc3b2da32012-03-26 09:59:21 -0400581 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
582 error = file_update_time(file);
583 if (error)
584 return error;
585 }
Christoph Hellwig8a9c9982012-02-29 09:53:52 +0000586
587 /*
Dave Chinner4d8d1582011-01-11 10:23:42 +1100588 * If we're writing the file then make sure to clear the setuid and
589 * setgid bits if the process is not being run by root. This keeps
590 * people from modifying setuid and setgid binaries.
591 */
592 return file_remove_suid(file);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100593}
594
595/*
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100596 * xfs_file_dio_aio_write - handle direct IO writes
597 *
598 * Lock the inode appropriately to prepare for and issue a direct IO write.
Dave Chinnereda77982011-01-11 10:22:40 +1100599 * By separating it from the buffered write path we remove all the tricky to
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100600 * follow locking changes and looping.
601 *
Dave Chinnereda77982011-01-11 10:22:40 +1100602 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
603 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
604 * pages are flushed out.
605 *
606 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
607 * allowing them to be done in parallel with reads and other direct IO writes.
608 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
609 * needs to do sub-block zeroing and that requires serialisation against other
610 * direct IOs to the same block. In this case we need to serialise the
611 * submission of the unaligned IOs so that we don't get racing block zeroing in
612 * the dio layer. To avoid the problem with aio, we also need to wait for
613 * outstanding IOs to complete so that unwritten extent conversion is completed
614 * before we try to map the overlapping block. This is currently implemented by
Christoph Hellwig4a06fd22011-08-23 08:28:13 +0000615 * hitting it with a big hammer (i.e. inode_dio_wait()).
Dave Chinnereda77982011-01-11 10:22:40 +1100616 *
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100617 * Returns with locks held indicated by @iolock and errors indicated by
618 * negative return values.
619 */
620STATIC ssize_t
621xfs_file_dio_aio_write(
622 struct kiocb *iocb,
623 const struct iovec *iovp,
624 unsigned long nr_segs,
625 loff_t pos,
Christoph Hellwigd0606462011-12-18 20:00:14 +0000626 size_t ocount)
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100627{
628 struct file *file = iocb->ki_filp;
629 struct address_space *mapping = file->f_mapping;
630 struct inode *inode = mapping->host;
631 struct xfs_inode *ip = XFS_I(inode);
632 struct xfs_mount *mp = ip->i_mount;
633 ssize_t ret = 0;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100634 size_t count = ocount;
Dave Chinnereda77982011-01-11 10:22:40 +1100635 int unaligned_io = 0;
Christoph Hellwigd0606462011-12-18 20:00:14 +0000636 int iolock;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100637 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
638 mp->m_rtdev_targp : mp->m_ddev_targp;
639
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100640 if ((pos & target->bt_smask) || (count & target->bt_smask))
641 return -XFS_ERROR(EINVAL);
642
Dave Chinnereda77982011-01-11 10:22:40 +1100643 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
644 unaligned_io = 1;
645
Dave Chinner7271d242011-08-25 07:17:02 +0000646 /*
647 * We don't need to take an exclusive lock unless there page cache needs
648 * to be invalidated or unaligned IO is being executed. We don't need to
649 * consider the EOF extension case here because
650 * xfs_file_aio_write_checks() will relock the inode as necessary for
651 * EOF zeroing cases and fill out the new inode size as appropriate.
652 */
653 if (unaligned_io || mapping->nrpages)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000654 iolock = XFS_IOLOCK_EXCL;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100655 else
Christoph Hellwigd0606462011-12-18 20:00:14 +0000656 iolock = XFS_IOLOCK_SHARED;
657 xfs_rw_ilock(ip, iolock);
Christoph Hellwigc58cb162011-08-27 14:42:53 +0000658
659 /*
660 * Recheck if there are cached pages that need invalidate after we got
661 * the iolock to protect against other threads adding new pages while
662 * we were waiting for the iolock.
663 */
Christoph Hellwigd0606462011-12-18 20:00:14 +0000664 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
665 xfs_rw_iunlock(ip, iolock);
666 iolock = XFS_IOLOCK_EXCL;
667 xfs_rw_ilock(ip, iolock);
Christoph Hellwigc58cb162011-08-27 14:42:53 +0000668 }
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100669
Christoph Hellwigd0606462011-12-18 20:00:14 +0000670 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100671 if (ret)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000672 goto out;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100673
674 if (mapping->nrpages) {
Dave Chinnerfb595812012-11-12 22:53:57 +1100675 ret = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
676 pos, -1);
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100677 if (ret)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000678 goto out;
Dave Chinnerfb595812012-11-12 22:53:57 +1100679 truncate_pagecache_range(VFS_I(ip), pos, -1);
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100680 }
681
Dave Chinnereda77982011-01-11 10:22:40 +1100682 /*
683 * If we are doing unaligned IO, wait for all other IO to drain,
684 * otherwise demote the lock if we had to flush cached pages
685 */
686 if (unaligned_io)
Christoph Hellwig4a06fd22011-08-23 08:28:13 +0000687 inode_dio_wait(inode);
Christoph Hellwigd0606462011-12-18 20:00:14 +0000688 else if (iolock == XFS_IOLOCK_EXCL) {
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100689 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
Christoph Hellwigd0606462011-12-18 20:00:14 +0000690 iolock = XFS_IOLOCK_SHARED;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100691 }
692
693 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
694 ret = generic_file_direct_write(iocb, iovp,
695 &nr_segs, pos, &iocb->ki_pos, count, ocount);
696
Christoph Hellwigd0606462011-12-18 20:00:14 +0000697out:
698 xfs_rw_iunlock(ip, iolock);
699
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100700 /* No fallback to buffered IO on errors for XFS. */
701 ASSERT(ret < 0 || ret == count);
702 return ret;
703}
704
Christoph Hellwig00258e32010-02-15 09:44:47 +0000705STATIC ssize_t
Dave Chinner637bbc72011-01-11 10:17:30 +1100706xfs_file_buffered_aio_write(
707 struct kiocb *iocb,
708 const struct iovec *iovp,
709 unsigned long nr_segs,
710 loff_t pos,
Christoph Hellwigd0606462011-12-18 20:00:14 +0000711 size_t ocount)
Dave Chinner637bbc72011-01-11 10:17:30 +1100712{
713 struct file *file = iocb->ki_filp;
714 struct address_space *mapping = file->f_mapping;
715 struct inode *inode = mapping->host;
716 struct xfs_inode *ip = XFS_I(inode);
717 ssize_t ret;
718 int enospc = 0;
Christoph Hellwigd0606462011-12-18 20:00:14 +0000719 int iolock = XFS_IOLOCK_EXCL;
Dave Chinner637bbc72011-01-11 10:17:30 +1100720 size_t count = ocount;
721
Christoph Hellwigd0606462011-12-18 20:00:14 +0000722 xfs_rw_ilock(ip, iolock);
Dave Chinner637bbc72011-01-11 10:17:30 +1100723
Christoph Hellwigd0606462011-12-18 20:00:14 +0000724 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100725 if (ret)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000726 goto out;
Dave Chinner637bbc72011-01-11 10:17:30 +1100727
728 /* We can write back this queue in page reclaim */
729 current->backing_dev_info = mapping->backing_dev_info;
730
731write_retry:
732 trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
733 ret = generic_file_buffered_write(iocb, iovp, nr_segs,
Dave Chinner9aa05002012-10-08 21:56:04 +1100734 pos, &iocb->ki_pos, count, 0);
735
Dave Chinner637bbc72011-01-11 10:17:30 +1100736 /*
Dave Chinner9aa05002012-10-08 21:56:04 +1100737 * If we just got an ENOSPC, try to write back all dirty inodes to
738 * convert delalloc space to free up some of the excess reserved
739 * metadata space.
Dave Chinner637bbc72011-01-11 10:17:30 +1100740 */
741 if (ret == -ENOSPC && !enospc) {
Dave Chinner637bbc72011-01-11 10:17:30 +1100742 enospc = 1;
Dave Chinner9aa05002012-10-08 21:56:04 +1100743 xfs_flush_inodes(ip->i_mount);
744 goto write_retry;
Dave Chinner637bbc72011-01-11 10:17:30 +1100745 }
Christoph Hellwigd0606462011-12-18 20:00:14 +0000746
Dave Chinner637bbc72011-01-11 10:17:30 +1100747 current->backing_dev_info = NULL;
Christoph Hellwigd0606462011-12-18 20:00:14 +0000748out:
749 xfs_rw_iunlock(ip, iolock);
Dave Chinner637bbc72011-01-11 10:17:30 +1100750 return ret;
751}
752
753STATIC ssize_t
Christoph Hellwig00258e32010-02-15 09:44:47 +0000754xfs_file_aio_write(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000755 struct kiocb *iocb,
756 const struct iovec *iovp,
Christoph Hellwig00258e32010-02-15 09:44:47 +0000757 unsigned long nr_segs,
758 loff_t pos)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000759{
760 struct file *file = iocb->ki_filp;
761 struct address_space *mapping = file->f_mapping;
762 struct inode *inode = mapping->host;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000763 struct xfs_inode *ip = XFS_I(inode);
Dave Chinner637bbc72011-01-11 10:17:30 +1100764 ssize_t ret;
Dave Chinner637bbc72011-01-11 10:17:30 +1100765 size_t ocount = 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000766
767 XFS_STATS_INC(xs_write_calls);
768
Christoph Hellwig00258e32010-02-15 09:44:47 +0000769 BUG_ON(iocb->ki_pos != pos);
770
Dave Chinnera363f0c2011-01-11 10:13:53 +1100771 ret = generic_segment_checks(iovp, &nr_segs, &ocount, VERIFY_READ);
772 if (ret)
773 return ret;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000774
Dave Chinner637bbc72011-01-11 10:17:30 +1100775 if (ocount == 0)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000776 return 0;
777
Jan Karad9457dc2012-06-12 16:20:39 +0200778 sb_start_write(inode->i_sb);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000779
Jan Karad9457dc2012-06-12 16:20:39 +0200780 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
781 ret = -EIO;
782 goto out;
783 }
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000784
Dave Chinner637bbc72011-01-11 10:17:30 +1100785 if (unlikely(file->f_flags & O_DIRECT))
Christoph Hellwigd0606462011-12-18 20:00:14 +0000786 ret = xfs_file_dio_aio_write(iocb, iovp, nr_segs, pos, ocount);
Dave Chinner637bbc72011-01-11 10:17:30 +1100787 else
788 ret = xfs_file_buffered_aio_write(iocb, iovp, nr_segs, pos,
Christoph Hellwigd0606462011-12-18 20:00:14 +0000789 ocount);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000790
Christoph Hellwigd0606462011-12-18 20:00:14 +0000791 if (ret > 0) {
792 ssize_t err;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000793
Christoph Hellwigd0606462011-12-18 20:00:14 +0000794 XFS_STATS_ADD(xs_write_bytes, ret);
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000795
Christoph Hellwigd0606462011-12-18 20:00:14 +0000796 /* Handle various SYNC-type writes */
797 err = generic_write_sync(file, pos, ret);
798 if (err < 0)
799 ret = err;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000800 }
801
Jan Karad9457dc2012-06-12 16:20:39 +0200802out:
803 sb_end_write(inode->i_sb);
Dave Chinnera363f0c2011-01-11 10:13:53 +1100804 return ret;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000805}
806
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100807STATIC long
808xfs_file_fallocate(
809 struct file *file,
810 int mode,
811 loff_t offset,
812 loff_t len)
813{
Al Viro496ad9a2013-01-23 17:07:38 -0500814 struct inode *inode = file_inode(file);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100815 long error;
816 loff_t new_size = 0;
817 xfs_flock64_t bf;
818 xfs_inode_t *ip = XFS_I(inode);
819 int cmd = XFS_IOC_RESVSP;
Dave Chinner82878892011-03-26 09:13:08 +1100820 int attr_flags = XFS_ATTR_NOLOCK;
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100821
822 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
823 return -EOPNOTSUPP;
824
825 bf.l_whence = 0;
826 bf.l_start = offset;
827 bf.l_len = len;
828
829 xfs_ilock(ip, XFS_IOLOCK_EXCL);
830
831 if (mode & FALLOC_FL_PUNCH_HOLE)
832 cmd = XFS_IOC_UNRESVSP;
833
834 /* check the new inode size is valid before allocating */
835 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
836 offset + len > i_size_read(inode)) {
837 new_size = offset + len;
838 error = inode_newsize_ok(inode, new_size);
839 if (error)
840 goto out_unlock;
841 }
842
Dave Chinner82878892011-03-26 09:13:08 +1100843 if (file->f_flags & O_DSYNC)
844 attr_flags |= XFS_ATTR_SYNC;
845
846 error = -xfs_change_file_space(ip, cmd, &bf, 0, attr_flags);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100847 if (error)
848 goto out_unlock;
849
850 /* Change file size if needed */
851 if (new_size) {
852 struct iattr iattr;
853
854 iattr.ia_valid = ATTR_SIZE;
855 iattr.ia_size = new_size;
Christoph Hellwigc4ed4242011-07-08 14:34:23 +0200856 error = -xfs_setattr_size(ip, &iattr, XFS_ATTR_NOLOCK);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100857 }
858
859out_unlock:
860 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
861 return error;
862}
863
864
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +1100866xfs_file_open(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700867 struct inode *inode,
Christoph Hellwigf999a5b2008-11-28 14:23:32 +1100868 struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700869{
Christoph Hellwigf999a5b2008-11-28 14:23:32 +1100870 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 return -EFBIG;
Christoph Hellwigf999a5b2008-11-28 14:23:32 +1100872 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
873 return -EIO;
874 return 0;
875}
876
877STATIC int
878xfs_dir_open(
879 struct inode *inode,
880 struct file *file)
881{
882 struct xfs_inode *ip = XFS_I(inode);
883 int mode;
884 int error;
885
886 error = xfs_file_open(inode, file);
887 if (error)
888 return error;
889
890 /*
891 * If there are any blocks, read-ahead block 0 as we're almost
892 * certain to have the next operation be a read there.
893 */
894 mode = xfs_ilock_map_shared(ip);
895 if (ip->i_d.di_nextents > 0)
Dave Chinnerda6958c2012-11-12 22:54:18 +1100896 xfs_dir2_data_readahead(NULL, ip, 0, -1);
Christoph Hellwigf999a5b2008-11-28 14:23:32 +1100897 xfs_iunlock(ip, mode);
898 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700899}
900
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +1100902xfs_file_release(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 struct inode *inode,
904 struct file *filp)
905{
Christoph Hellwig739bfb22007-08-29 10:58:01 +1000906 return -xfs_release(XFS_I(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700907}
908
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +1100910xfs_file_readdir(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 struct file *filp,
912 void *dirent,
913 filldir_t filldir)
914{
Al Viro496ad9a2013-01-23 17:07:38 -0500915 struct inode *inode = file_inode(filp);
Christoph Hellwig739bfb22007-08-29 10:58:01 +1000916 xfs_inode_t *ip = XFS_I(inode);
Christoph Hellwig051e7cd2007-08-28 13:58:24 +1000917 int error;
918 size_t bufsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700919
Christoph Hellwig051e7cd2007-08-28 13:58:24 +1000920 /*
921 * The Linux API doesn't pass down the total size of the buffer
922 * we read into down to the filesystem. With the filldir concept
923 * it's not needed for correct information, but the XFS dir2 leaf
924 * code wants an estimate of the buffer size to calculate it's
925 * readahead window and size the buffers used for mapping to
926 * physical blocks.
927 *
928 * Try to give it an estimate that's good enough, maybe at some
929 * point we can change the ->readdir prototype to include the
Eric Sandeena9cc7992010-02-03 17:50:13 +0000930 * buffer size. For now we use the current glibc buffer size.
Christoph Hellwig051e7cd2007-08-28 13:58:24 +1000931 */
Eric Sandeena9cc7992010-02-03 17:50:13 +0000932 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933
Christoph Hellwig739bfb22007-08-29 10:58:01 +1000934 error = xfs_readdir(ip, dirent, bufsize,
Christoph Hellwig051e7cd2007-08-28 13:58:24 +1000935 (xfs_off_t *)&filp->f_pos, filldir);
936 if (error)
937 return -error;
938 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939}
940
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +1100942xfs_file_mmap(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 struct file *filp,
944 struct vm_area_struct *vma)
945{
Nathan Scott3562fd42006-03-14 14:00:35 +1100946 vma->vm_ops = &xfs_file_vm_ops;
Dean Roehrich6fac0cb2005-06-21 14:07:45 +1000947
Nathan Scottfbc14622006-06-09 14:52:13 +1000948 file_accessed(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 return 0;
950}
951
David Chinner4f57dbc2007-07-19 16:28:17 +1000952/*
953 * mmap()d file has taken write protection fault and is being made
954 * writable. We can set the page state up correctly for a writable
955 * page, which means we can do correct delalloc accounting (ENOSPC
956 * checking!) and unwritten extent mapping.
957 */
958STATIC int
959xfs_vm_page_mkwrite(
960 struct vm_area_struct *vma,
Nick Pigginc2ec1752009-03-31 15:23:21 -0700961 struct vm_fault *vmf)
David Chinner4f57dbc2007-07-19 16:28:17 +1000962{
Nick Pigginc2ec1752009-03-31 15:23:21 -0700963 return block_page_mkwrite(vma, vmf, xfs_get_blocks);
David Chinner4f57dbc2007-07-19 16:28:17 +1000964}
965
Jeff Liud126d432012-08-21 17:11:57 +0800966/*
967 * This type is designed to indicate the type of offset we would like
968 * to search from page cache for either xfs_seek_data() or xfs_seek_hole().
969 */
970enum {
971 HOLE_OFF = 0,
972 DATA_OFF,
973};
974
975/*
976 * Lookup the desired type of offset from the given page.
977 *
978 * On success, return true and the offset argument will point to the
979 * start of the region that was found. Otherwise this function will
980 * return false and keep the offset argument unchanged.
981 */
982STATIC bool
983xfs_lookup_buffer_offset(
984 struct page *page,
985 loff_t *offset,
986 unsigned int type)
987{
988 loff_t lastoff = page_offset(page);
989 bool found = false;
990 struct buffer_head *bh, *head;
991
992 bh = head = page_buffers(page);
993 do {
994 /*
995 * Unwritten extents that have data in the page
996 * cache covering them can be identified by the
997 * BH_Unwritten state flag. Pages with multiple
998 * buffers might have a mix of holes, data and
999 * unwritten extents - any buffer with valid
1000 * data in it should have BH_Uptodate flag set
1001 * on it.
1002 */
1003 if (buffer_unwritten(bh) ||
1004 buffer_uptodate(bh)) {
1005 if (type == DATA_OFF)
1006 found = true;
1007 } else {
1008 if (type == HOLE_OFF)
1009 found = true;
1010 }
1011
1012 if (found) {
1013 *offset = lastoff;
1014 break;
1015 }
1016 lastoff += bh->b_size;
1017 } while ((bh = bh->b_this_page) != head);
1018
1019 return found;
1020}
1021
1022/*
1023 * This routine is called to find out and return a data or hole offset
1024 * from the page cache for unwritten extents according to the desired
1025 * type for xfs_seek_data() or xfs_seek_hole().
1026 *
1027 * The argument offset is used to tell where we start to search from the
1028 * page cache. Map is used to figure out the end points of the range to
1029 * lookup pages.
1030 *
1031 * Return true if the desired type of offset was found, and the argument
1032 * offset is filled with that address. Otherwise, return false and keep
1033 * offset unchanged.
1034 */
1035STATIC bool
1036xfs_find_get_desired_pgoff(
1037 struct inode *inode,
1038 struct xfs_bmbt_irec *map,
1039 unsigned int type,
1040 loff_t *offset)
1041{
1042 struct xfs_inode *ip = XFS_I(inode);
1043 struct xfs_mount *mp = ip->i_mount;
1044 struct pagevec pvec;
1045 pgoff_t index;
1046 pgoff_t end;
1047 loff_t endoff;
1048 loff_t startoff = *offset;
1049 loff_t lastoff = startoff;
1050 bool found = false;
1051
1052 pagevec_init(&pvec, 0);
1053
1054 index = startoff >> PAGE_CACHE_SHIFT;
1055 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
1056 end = endoff >> PAGE_CACHE_SHIFT;
1057 do {
1058 int want;
1059 unsigned nr_pages;
1060 unsigned int i;
1061
1062 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1063 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1064 want);
1065 /*
1066 * No page mapped into given range. If we are searching holes
1067 * and if this is the first time we got into the loop, it means
1068 * that the given offset is landed in a hole, return it.
1069 *
1070 * If we have already stepped through some block buffers to find
1071 * holes but they all contains data. In this case, the last
1072 * offset is already updated and pointed to the end of the last
1073 * mapped page, if it does not reach the endpoint to search,
1074 * that means there should be a hole between them.
1075 */
1076 if (nr_pages == 0) {
1077 /* Data search found nothing */
1078 if (type == DATA_OFF)
1079 break;
1080
1081 ASSERT(type == HOLE_OFF);
1082 if (lastoff == startoff || lastoff < endoff) {
1083 found = true;
1084 *offset = lastoff;
1085 }
1086 break;
1087 }
1088
1089 /*
1090 * At lease we found one page. If this is the first time we
1091 * step into the loop, and if the first page index offset is
1092 * greater than the given search offset, a hole was found.
1093 */
1094 if (type == HOLE_OFF && lastoff == startoff &&
1095 lastoff < page_offset(pvec.pages[0])) {
1096 found = true;
1097 break;
1098 }
1099
1100 for (i = 0; i < nr_pages; i++) {
1101 struct page *page = pvec.pages[i];
1102 loff_t b_offset;
1103
1104 /*
1105 * At this point, the page may be truncated or
1106 * invalidated (changing page->mapping to NULL),
1107 * or even swizzled back from swapper_space to tmpfs
1108 * file mapping. However, page->index will not change
1109 * because we have a reference on the page.
1110 *
1111 * Searching done if the page index is out of range.
1112 * If the current offset is not reaches the end of
1113 * the specified search range, there should be a hole
1114 * between them.
1115 */
1116 if (page->index > end) {
1117 if (type == HOLE_OFF && lastoff < endoff) {
1118 *offset = lastoff;
1119 found = true;
1120 }
1121 goto out;
1122 }
1123
1124 lock_page(page);
1125 /*
1126 * Page truncated or invalidated(page->mapping == NULL).
1127 * We can freely skip it and proceed to check the next
1128 * page.
1129 */
1130 if (unlikely(page->mapping != inode->i_mapping)) {
1131 unlock_page(page);
1132 continue;
1133 }
1134
1135 if (!page_has_buffers(page)) {
1136 unlock_page(page);
1137 continue;
1138 }
1139
1140 found = xfs_lookup_buffer_offset(page, &b_offset, type);
1141 if (found) {
1142 /*
1143 * The found offset may be less than the start
1144 * point to search if this is the first time to
1145 * come here.
1146 */
1147 *offset = max_t(loff_t, startoff, b_offset);
1148 unlock_page(page);
1149 goto out;
1150 }
1151
1152 /*
1153 * We either searching data but nothing was found, or
1154 * searching hole but found a data buffer. In either
1155 * case, probably the next page contains the desired
1156 * things, update the last offset to it so.
1157 */
1158 lastoff = page_offset(page) + PAGE_SIZE;
1159 unlock_page(page);
1160 }
1161
1162 /*
1163 * The number of returned pages less than our desired, search
1164 * done. In this case, nothing was found for searching data,
1165 * but we found a hole behind the last offset.
1166 */
1167 if (nr_pages < want) {
1168 if (type == HOLE_OFF) {
1169 *offset = lastoff;
1170 found = true;
1171 }
1172 break;
1173 }
1174
1175 index = pvec.pages[i - 1]->index + 1;
1176 pagevec_release(&pvec);
1177 } while (index <= end);
1178
1179out:
1180 pagevec_release(&pvec);
1181 return found;
1182}
1183
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001184STATIC loff_t
1185xfs_seek_data(
1186 struct file *file,
Jeff Liu834ab122012-08-21 17:11:45 +08001187 loff_t start)
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001188{
1189 struct inode *inode = file->f_mapping->host;
1190 struct xfs_inode *ip = XFS_I(inode);
1191 struct xfs_mount *mp = ip->i_mount;
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001192 loff_t uninitialized_var(offset);
1193 xfs_fsize_t isize;
1194 xfs_fileoff_t fsbno;
1195 xfs_filblks_t end;
1196 uint lock;
1197 int error;
1198
1199 lock = xfs_ilock_map_shared(ip);
1200
1201 isize = i_size_read(inode);
1202 if (start >= isize) {
1203 error = ENXIO;
1204 goto out_unlock;
1205 }
1206
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001207 /*
1208 * Try to read extents from the first block indicated
1209 * by fsbno to the end block of the file.
1210 */
Jeff Liu52f1acc2012-08-21 17:12:07 +08001211 fsbno = XFS_B_TO_FSBT(mp, start);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001212 end = XFS_B_TO_FSB(mp, isize);
Jeff Liu52f1acc2012-08-21 17:12:07 +08001213 for (;;) {
1214 struct xfs_bmbt_irec map[2];
1215 int nmap = 2;
1216 unsigned int i;
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001217
Jeff Liu52f1acc2012-08-21 17:12:07 +08001218 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1219 XFS_BMAPI_ENTIRE);
1220 if (error)
1221 goto out_unlock;
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001222
Jeff Liu52f1acc2012-08-21 17:12:07 +08001223 /* No extents at given offset, must be beyond EOF */
1224 if (nmap == 0) {
1225 error = ENXIO;
1226 goto out_unlock;
1227 }
1228
1229 for (i = 0; i < nmap; i++) {
1230 offset = max_t(loff_t, start,
1231 XFS_FSB_TO_B(mp, map[i].br_startoff));
1232
1233 /* Landed in a data extent */
1234 if (map[i].br_startblock == DELAYSTARTBLOCK ||
1235 (map[i].br_state == XFS_EXT_NORM &&
1236 !isnullstartblock(map[i].br_startblock)))
1237 goto out;
1238
1239 /*
1240 * Landed in an unwritten extent, try to search data
1241 * from page cache.
1242 */
1243 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1244 if (xfs_find_get_desired_pgoff(inode, &map[i],
1245 DATA_OFF, &offset))
1246 goto out;
1247 }
1248 }
1249
1250 /*
1251 * map[0] is hole or its an unwritten extent but
1252 * without data in page cache. Probably means that
1253 * we are reading after EOF if nothing in map[1].
1254 */
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001255 if (nmap == 1) {
1256 error = ENXIO;
1257 goto out_unlock;
1258 }
1259
Jeff Liu52f1acc2012-08-21 17:12:07 +08001260 ASSERT(i > 1);
1261
1262 /*
1263 * Nothing was found, proceed to the next round of search
1264 * if reading offset not beyond or hit EOF.
1265 */
1266 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1267 start = XFS_FSB_TO_B(mp, fsbno);
1268 if (start >= isize) {
1269 error = ENXIO;
1270 goto out_unlock;
1271 }
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001272 }
1273
Jeff Liu52f1acc2012-08-21 17:12:07 +08001274out:
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001275 if (offset != file->f_pos)
1276 file->f_pos = offset;
1277
1278out_unlock:
1279 xfs_iunlock_map_shared(ip, lock);
1280
1281 if (error)
1282 return -error;
1283 return offset;
1284}
1285
1286STATIC loff_t
1287xfs_seek_hole(
1288 struct file *file,
Jeff Liu834ab122012-08-21 17:11:45 +08001289 loff_t start)
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001290{
1291 struct inode *inode = file->f_mapping->host;
1292 struct xfs_inode *ip = XFS_I(inode);
1293 struct xfs_mount *mp = ip->i_mount;
1294 loff_t uninitialized_var(offset);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001295 xfs_fsize_t isize;
1296 xfs_fileoff_t fsbno;
Jeff Liub686d1f2012-08-21 17:12:18 +08001297 xfs_filblks_t end;
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001298 uint lock;
1299 int error;
1300
1301 if (XFS_FORCED_SHUTDOWN(mp))
1302 return -XFS_ERROR(EIO);
1303
1304 lock = xfs_ilock_map_shared(ip);
1305
1306 isize = i_size_read(inode);
1307 if (start >= isize) {
1308 error = ENXIO;
1309 goto out_unlock;
1310 }
1311
1312 fsbno = XFS_B_TO_FSBT(mp, start);
Jeff Liub686d1f2012-08-21 17:12:18 +08001313 end = XFS_B_TO_FSB(mp, isize);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001314
Jeff Liub686d1f2012-08-21 17:12:18 +08001315 for (;;) {
1316 struct xfs_bmbt_irec map[2];
1317 int nmap = 2;
1318 unsigned int i;
1319
1320 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1321 XFS_BMAPI_ENTIRE);
1322 if (error)
1323 goto out_unlock;
1324
1325 /* No extents at given offset, must be beyond EOF */
1326 if (nmap == 0) {
1327 error = ENXIO;
1328 goto out_unlock;
1329 }
1330
1331 for (i = 0; i < nmap; i++) {
1332 offset = max_t(loff_t, start,
1333 XFS_FSB_TO_B(mp, map[i].br_startoff));
1334
1335 /* Landed in a hole */
1336 if (map[i].br_startblock == HOLESTARTBLOCK)
1337 goto out;
1338
1339 /*
1340 * Landed in an unwritten extent, try to search hole
1341 * from page cache.
1342 */
1343 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1344 if (xfs_find_get_desired_pgoff(inode, &map[i],
1345 HOLE_OFF, &offset))
1346 goto out;
1347 }
1348 }
1349
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001350 /*
Jeff Liub686d1f2012-08-21 17:12:18 +08001351 * map[0] contains data or its unwritten but contains
1352 * data in page cache, probably means that we are
1353 * reading after EOF. We should fix offset to point
1354 * to the end of the file(i.e., there is an implicit
1355 * hole at the end of any file).
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001356 */
Jeff Liub686d1f2012-08-21 17:12:18 +08001357 if (nmap == 1) {
1358 offset = isize;
1359 break;
1360 }
1361
1362 ASSERT(i > 1);
1363
1364 /*
1365 * Both mappings contains data, proceed to the next round of
1366 * search if the current reading offset not beyond or hit EOF.
1367 */
1368 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1369 start = XFS_FSB_TO_B(mp, fsbno);
1370 if (start >= isize) {
1371 offset = isize;
1372 break;
1373 }
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001374 }
1375
Jeff Liub686d1f2012-08-21 17:12:18 +08001376out:
1377 /*
1378 * At this point, we must have found a hole. However, the returned
1379 * offset may be bigger than the file size as it may be aligned to
1380 * page boundary for unwritten extents, we need to deal with this
1381 * situation in particular.
1382 */
1383 offset = min_t(loff_t, offset, isize);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001384 if (offset != file->f_pos)
1385 file->f_pos = offset;
1386
1387out_unlock:
1388 xfs_iunlock_map_shared(ip, lock);
1389
1390 if (error)
1391 return -error;
1392 return offset;
1393}
1394
1395STATIC loff_t
1396xfs_file_llseek(
1397 struct file *file,
1398 loff_t offset,
1399 int origin)
1400{
1401 switch (origin) {
1402 case SEEK_END:
1403 case SEEK_CUR:
1404 case SEEK_SET:
1405 return generic_file_llseek(file, offset, origin);
1406 case SEEK_DATA:
Jeff Liu834ab122012-08-21 17:11:45 +08001407 return xfs_seek_data(file, offset);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001408 case SEEK_HOLE:
Jeff Liu834ab122012-08-21 17:11:45 +08001409 return xfs_seek_hole(file, offset);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001410 default:
1411 return -EINVAL;
1412 }
1413}
1414
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001415const struct file_operations xfs_file_operations = {
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001416 .llseek = xfs_file_llseek,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001417 .read = do_sync_read,
Dean Roehrichbb3f7242005-09-02 15:43:05 +10001418 .write = do_sync_write,
Nathan Scott3562fd42006-03-14 14:00:35 +11001419 .aio_read = xfs_file_aio_read,
1420 .aio_write = xfs_file_aio_write,
Nathan Scott1b895842006-03-31 13:08:59 +10001421 .splice_read = xfs_file_splice_read,
1422 .splice_write = xfs_file_splice_write,
Nathan Scott3562fd42006-03-14 14:00:35 +11001423 .unlocked_ioctl = xfs_file_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424#ifdef CONFIG_COMPAT
Nathan Scott3562fd42006-03-14 14:00:35 +11001425 .compat_ioctl = xfs_file_compat_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426#endif
Nathan Scott3562fd42006-03-14 14:00:35 +11001427 .mmap = xfs_file_mmap,
1428 .open = xfs_file_open,
1429 .release = xfs_file_release,
1430 .fsync = xfs_file_fsync,
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001431 .fallocate = xfs_file_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432};
1433
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001434const struct file_operations xfs_dir_file_operations = {
Christoph Hellwigf999a5b2008-11-28 14:23:32 +11001435 .open = xfs_dir_open,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 .read = generic_read_dir,
Nathan Scott3562fd42006-03-14 14:00:35 +11001437 .readdir = xfs_file_readdir,
Al Viro59af1582008-08-24 07:24:41 -04001438 .llseek = generic_file_llseek,
Nathan Scott3562fd42006-03-14 14:00:35 +11001439 .unlocked_ioctl = xfs_file_ioctl,
Nathan Scottd3870392005-05-06 06:44:46 -07001440#ifdef CONFIG_COMPAT
Nathan Scott3562fd42006-03-14 14:00:35 +11001441 .compat_ioctl = xfs_file_compat_ioctl,
Nathan Scottd3870392005-05-06 06:44:46 -07001442#endif
Christoph Hellwig1da2f2d2011-10-02 14:25:16 +00001443 .fsync = xfs_dir_fsync,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444};
1445
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001446static const struct vm_operations_struct xfs_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07001447 .fault = filemap_fault,
David Chinner4f57dbc2007-07-19 16:28:17 +10001448 .page_mkwrite = xfs_vm_page_mkwrite,
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -07001449 .remap_pages = generic_file_remap_pages,
Dean Roehrich6fac0cb2005-06-21 14:07:45 +10001450};