blob: 52c91e1437255c60d1f56041d4eceea4601f6265 [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"
Dave Chinner70a9883c2013-10-23 10:36:05 +110020#include "xfs_shared.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110021#include "xfs_format.h"
Dave Chinner239880e2013-10-23 10:50:10 +110022#include "xfs_log_format.h"
23#include "xfs_trans_resv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_sb.h"
Nathan Scotta844f452005-11-02 14:38:42 +110025#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include "xfs_mount.h"
Dave Chinner57062782013-10-15 09:17:51 +110027#include "xfs_da_format.h"
28#include "xfs_da_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include "xfs_inode.h"
Dave Chinner239880e2013-10-23 10:50:10 +110030#include "xfs_trans.h"
Christoph Hellwigfd3200b2010-02-15 09:44:48 +000031#include "xfs_inode_item.h"
Christoph Hellwigdda35b82010-02-15 09:44:46 +000032#include "xfs_bmap.h"
Dave Chinnerc24b5df2013-08-12 20:49:45 +100033#include "xfs_bmap_util.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include "xfs_error.h"
Dave Chinner2b9ab5a2013-08-12 20:49:37 +100035#include "xfs_dir2.h"
Dave Chinnerc24b5df2013-08-12 20:49:45 +100036#include "xfs_dir2_priv.h"
Christoph Hellwigddcd8562008-12-03 07:55:34 -050037#include "xfs_ioctl.h"
Christoph Hellwigdda35b82010-02-15 09:44:46 +000038#include "xfs_trace.h"
Dave Chinner239880e2013-10-23 10:50:10 +110039#include "xfs_log.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110040#include "xfs_dinode.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Kent Overstreeta27bb332013-05-07 16:19:08 -070042#include <linux/aio.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include <linux/dcache.h>
Christoph Hellwig2fe17c12011-01-14 13:07:43 +010044#include <linux/falloc.h>
Jeff Liud126d432012-08-21 17:11:57 +080045#include <linux/pagevec.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +040047static const struct vm_operations_struct xfs_file_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
Christoph Hellwigdda35b82010-02-15 09:44:46 +000049/*
Dave Chinner487f84f2011-01-12 11:37:10 +110050 * Locking primitives for read and write IO paths to ensure we consistently use
51 * and order the inode->i_mutex, ip->i_lock and ip->i_iolock.
52 */
53static inline void
54xfs_rw_ilock(
55 struct xfs_inode *ip,
56 int type)
57{
58 if (type & XFS_IOLOCK_EXCL)
59 mutex_lock(&VFS_I(ip)->i_mutex);
60 xfs_ilock(ip, type);
61}
62
63static inline void
64xfs_rw_iunlock(
65 struct xfs_inode *ip,
66 int type)
67{
68 xfs_iunlock(ip, type);
69 if (type & XFS_IOLOCK_EXCL)
70 mutex_unlock(&VFS_I(ip)->i_mutex);
71}
72
73static inline void
74xfs_rw_ilock_demote(
75 struct xfs_inode *ip,
76 int type)
77{
78 xfs_ilock_demote(ip, type);
79 if (type & XFS_IOLOCK_EXCL)
80 mutex_unlock(&VFS_I(ip)->i_mutex);
81}
82
83/*
Christoph Hellwigdda35b82010-02-15 09:44:46 +000084 * xfs_iozero
85 *
86 * xfs_iozero clears the specified range of buffer supplied,
87 * and marks all the affected blocks as valid and modified. If
88 * an affected block is not allocated, it will be allocated. If
89 * an affected block is not completely overwritten, and is not
90 * valid before the operation, it will be read from disk before
91 * being partially zeroed.
92 */
Dave Chinneref9d8732012-11-29 15:26:33 +110093int
Christoph Hellwigdda35b82010-02-15 09:44:46 +000094xfs_iozero(
95 struct xfs_inode *ip, /* inode */
96 loff_t pos, /* offset in file */
97 size_t count) /* size of data to zero */
98{
99 struct page *page;
100 struct address_space *mapping;
101 int status;
102
103 mapping = VFS_I(ip)->i_mapping;
104 do {
105 unsigned offset, bytes;
106 void *fsdata;
107
108 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
109 bytes = PAGE_CACHE_SIZE - offset;
110 if (bytes > count)
111 bytes = count;
112
113 status = pagecache_write_begin(NULL, mapping, pos, bytes,
114 AOP_FLAG_UNINTERRUPTIBLE,
115 &page, &fsdata);
116 if (status)
117 break;
118
119 zero_user(page, offset, bytes);
120
121 status = pagecache_write_end(NULL, mapping, pos, bytes, bytes,
122 page, fsdata);
123 WARN_ON(status <= 0); /* can't return less than zero! */
124 pos += bytes;
125 count -= bytes;
126 status = 0;
127 } while (count);
128
129 return (-status);
130}
131
Christoph Hellwig1da2f2d2011-10-02 14:25:16 +0000132/*
133 * Fsync operations on directories are much simpler than on regular files,
134 * as there is no file data to flush, and thus also no need for explicit
135 * cache flush operations, and there are no non-transaction metadata updates
136 * on directories either.
137 */
138STATIC int
139xfs_dir_fsync(
140 struct file *file,
141 loff_t start,
142 loff_t end,
143 int datasync)
144{
145 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
146 struct xfs_mount *mp = ip->i_mount;
147 xfs_lsn_t lsn = 0;
148
149 trace_xfs_dir_fsync(ip);
150
151 xfs_ilock(ip, XFS_ILOCK_SHARED);
152 if (xfs_ipincount(ip))
153 lsn = ip->i_itemp->ili_last_lsn;
154 xfs_iunlock(ip, XFS_ILOCK_SHARED);
155
156 if (!lsn)
157 return 0;
158 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
159}
160
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000161STATIC int
162xfs_file_fsync(
163 struct file *file,
Josef Bacik02c24a82011-07-16 20:44:56 -0400164 loff_t start,
165 loff_t end,
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000166 int datasync)
167{
Christoph Hellwig7ea80852010-05-26 17:53:25 +0200168 struct inode *inode = file->f_mapping->host;
169 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000170 struct xfs_mount *mp = ip->i_mount;
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000171 int error = 0;
172 int log_flushed = 0;
Christoph Hellwigb1037052011-09-19 14:55:51 +0000173 xfs_lsn_t lsn = 0;
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000174
Christoph Hellwigcca28fb2010-06-24 11:57:09 +1000175 trace_xfs_file_fsync(ip);
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000176
Josef Bacik02c24a82011-07-16 20:44:56 -0400177 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
178 if (error)
179 return error;
180
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000181 if (XFS_FORCED_SHUTDOWN(mp))
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000182 return -XFS_ERROR(EIO);
183
184 xfs_iflags_clear(ip, XFS_ITRUNCATED);
185
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000186 if (mp->m_flags & XFS_MOUNT_BARRIER) {
187 /*
188 * If we have an RT and/or log subvolume we need to make sure
189 * to flush the write cache the device used for file data
190 * first. This is to ensure newly written file data make
191 * it to disk before logging the new inode size in case of
192 * an extending write.
193 */
194 if (XFS_IS_REALTIME_INODE(ip))
195 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
196 else if (mp->m_logdev_targp != mp->m_ddev_targp)
197 xfs_blkdev_issue_flush(mp->m_ddev_targp);
198 }
199
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000200 /*
Christoph Hellwig8a9c9982012-02-29 09:53:52 +0000201 * All metadata updates are logged, which means that we just have
202 * to flush the log up to the latest LSN that touched the inode.
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000203 */
204 xfs_ilock(ip, XFS_ILOCK_SHARED);
Christoph Hellwig8f639dd2012-02-29 09:53:55 +0000205 if (xfs_ipincount(ip)) {
206 if (!datasync ||
207 (ip->i_itemp->ili_fields & ~XFS_ILOG_TIMESTAMP))
208 lsn = ip->i_itemp->ili_last_lsn;
209 }
Christoph Hellwig8a9c9982012-02-29 09:53:52 +0000210 xfs_iunlock(ip, XFS_ILOCK_SHARED);
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000211
Christoph Hellwig8a9c9982012-02-29 09:53:52 +0000212 if (lsn)
Christoph Hellwigb1037052011-09-19 14:55:51 +0000213 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
214
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000215 /*
216 * If we only have a single device, and the log force about was
217 * a no-op we might have to flush the data device cache here.
218 * This can only happen for fdatasync/O_DSYNC if we were overwriting
219 * an already allocated file and thus do not have any metadata to
220 * commit.
221 */
222 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
223 mp->m_logdev_targp == mp->m_ddev_targp &&
224 !XFS_IS_REALTIME_INODE(ip) &&
225 !log_flushed)
226 xfs_blkdev_issue_flush(mp->m_ddev_targp);
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000227
228 return -error;
229}
230
Christoph Hellwig00258e32010-02-15 09:44:47 +0000231STATIC ssize_t
232xfs_file_aio_read(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000233 struct kiocb *iocb,
234 const struct iovec *iovp,
Christoph Hellwig00258e32010-02-15 09:44:47 +0000235 unsigned long nr_segs,
236 loff_t pos)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000237{
238 struct file *file = iocb->ki_filp;
239 struct inode *inode = file->f_mapping->host;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000240 struct xfs_inode *ip = XFS_I(inode);
241 struct xfs_mount *mp = ip->i_mount;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000242 size_t size = 0;
243 ssize_t ret = 0;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000244 int ioflags = 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000245 xfs_fsize_t n;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000246
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000247 XFS_STATS_INC(xs_read_calls);
248
Christoph Hellwig00258e32010-02-15 09:44:47 +0000249 BUG_ON(iocb->ki_pos != pos);
250
251 if (unlikely(file->f_flags & O_DIRECT))
252 ioflags |= IO_ISDIRECT;
253 if (file->f_mode & FMODE_NOCMTIME)
254 ioflags |= IO_INVIS;
255
Dave Chinner52764322012-06-08 15:45:44 +1000256 ret = generic_segment_checks(iovp, &nr_segs, &size, VERIFY_WRITE);
257 if (ret < 0)
258 return ret;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000259
260 if (unlikely(ioflags & IO_ISDIRECT)) {
261 xfs_buftarg_t *target =
262 XFS_IS_REALTIME_INODE(ip) ?
263 mp->m_rtdev_targp : mp->m_ddev_targp;
Dave Chinnerfb595812012-11-12 22:53:57 +1100264 if ((pos & target->bt_smask) || (size & target->bt_smask)) {
265 if (pos == i_size_read(inode))
Christoph Hellwig00258e32010-02-15 09:44:47 +0000266 return 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000267 return -XFS_ERROR(EINVAL);
268 }
269 }
270
Dave Chinnerfb595812012-11-12 22:53:57 +1100271 n = mp->m_super->s_maxbytes - pos;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000272 if (n <= 0 || size == 0)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000273 return 0;
274
275 if (n < size)
276 size = n;
277
278 if (XFS_FORCED_SHUTDOWN(mp))
279 return -EIO;
280
Dave Chinner0c38a252011-08-25 07:17:01 +0000281 /*
282 * Locking is a bit tricky here. If we take an exclusive lock
283 * for direct IO, we effectively serialise all new concurrent
284 * read IO to this file and block it behind IO that is currently in
285 * progress because IO in progress holds the IO lock shared. We only
286 * need to hold the lock exclusive to blow away the page cache, so
287 * only take lock exclusively if the page cache needs invalidation.
288 * This allows the normal direct IO case of no page cache pages to
289 * proceeed concurrently without serialisation.
290 */
291 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
292 if ((ioflags & IO_ISDIRECT) && inode->i_mapping->nrpages) {
293 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
Dave Chinner487f84f2011-01-12 11:37:10 +1100294 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
295
Christoph Hellwig00258e32010-02-15 09:44:47 +0000296 if (inode->i_mapping->nrpages) {
Dave Chinnerfb595812012-11-12 22:53:57 +1100297 ret = -filemap_write_and_wait_range(
298 VFS_I(ip)->i_mapping,
299 pos, -1);
Dave Chinner487f84f2011-01-12 11:37:10 +1100300 if (ret) {
301 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
302 return ret;
303 }
Dave Chinnerfb595812012-11-12 22:53:57 +1100304 truncate_pagecache_range(VFS_I(ip), pos, -1);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000305 }
Dave Chinner487f84f2011-01-12 11:37:10 +1100306 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
Dave Chinner0c38a252011-08-25 07:17:01 +0000307 }
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000308
Dave Chinnerfb595812012-11-12 22:53:57 +1100309 trace_xfs_file_read(ip, size, pos, ioflags);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000310
Dave Chinnerfb595812012-11-12 22:53:57 +1100311 ret = generic_file_aio_read(iocb, iovp, nr_segs, pos);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000312 if (ret > 0)
313 XFS_STATS_ADD(xs_read_bytes, ret);
314
Dave Chinner487f84f2011-01-12 11:37:10 +1100315 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000316 return ret;
317}
318
Christoph Hellwig00258e32010-02-15 09:44:47 +0000319STATIC ssize_t
320xfs_file_splice_read(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000321 struct file *infilp,
322 loff_t *ppos,
323 struct pipe_inode_info *pipe,
324 size_t count,
Christoph Hellwig00258e32010-02-15 09:44:47 +0000325 unsigned int flags)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000326{
Christoph Hellwig00258e32010-02-15 09:44:47 +0000327 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000328 int ioflags = 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000329 ssize_t ret;
330
331 XFS_STATS_INC(xs_read_calls);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000332
333 if (infilp->f_mode & FMODE_NOCMTIME)
334 ioflags |= IO_INVIS;
335
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000336 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
337 return -EIO;
338
Dave Chinner487f84f2011-01-12 11:37:10 +1100339 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000340
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000341 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
342
343 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
344 if (ret > 0)
345 XFS_STATS_ADD(xs_read_bytes, ret);
346
Dave Chinner487f84f2011-01-12 11:37:10 +1100347 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000348 return ret;
349}
350
Dave Chinner4c5cfd12011-01-11 10:14:16 +1100351/*
Dave Chinner487f84f2011-01-12 11:37:10 +1100352 * xfs_file_splice_write() does not use xfs_rw_ilock() because
353 * generic_file_splice_write() takes the i_mutex itself. This, in theory,
354 * couuld cause lock inversions between the aio_write path and the splice path
355 * if someone is doing concurrent splice(2) based writes and write(2) based
356 * writes to the same inode. The only real way to fix this is to re-implement
357 * the generic code here with correct locking orders.
358 */
Christoph Hellwig00258e32010-02-15 09:44:47 +0000359STATIC ssize_t
360xfs_file_splice_write(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000361 struct pipe_inode_info *pipe,
362 struct file *outfilp,
363 loff_t *ppos,
364 size_t count,
Christoph Hellwig00258e32010-02-15 09:44:47 +0000365 unsigned int flags)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000366{
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000367 struct inode *inode = outfilp->f_mapping->host;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000368 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000369 int ioflags = 0;
370 ssize_t ret;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000371
372 XFS_STATS_INC(xs_write_calls);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000373
374 if (outfilp->f_mode & FMODE_NOCMTIME)
375 ioflags |= IO_INVIS;
376
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000377 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
378 return -EIO;
379
380 xfs_ilock(ip, XFS_IOLOCK_EXCL);
381
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000382 trace_xfs_file_splice_write(ip, count, *ppos, ioflags);
383
384 ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000385 if (ret > 0)
386 XFS_STATS_ADD(xs_write_bytes, ret);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000387
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000388 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
389 return ret;
390}
391
392/*
Christoph Hellwig193aec12012-03-27 10:34:49 -0400393 * This routine is called to handle zeroing any space in the last block of the
394 * file that is beyond the EOF. We do this since the size is being increased
395 * without writing anything to that block and we don't want to read the
396 * garbage on the disk.
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000397 */
398STATIC int /* error (positive) */
399xfs_zero_last_block(
Christoph Hellwig193aec12012-03-27 10:34:49 -0400400 struct xfs_inode *ip,
401 xfs_fsize_t offset,
402 xfs_fsize_t isize)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000403{
Christoph Hellwig193aec12012-03-27 10:34:49 -0400404 struct xfs_mount *mp = ip->i_mount;
405 xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize);
406 int zero_offset = XFS_B_FSB_OFFSET(mp, isize);
407 int zero_len;
408 int nimaps = 1;
409 int error = 0;
410 struct xfs_bmbt_irec imap;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000411
Christoph Hellwig193aec12012-03-27 10:34:49 -0400412 xfs_ilock(ip, XFS_ILOCK_EXCL);
Dave Chinner5c8ed202011-09-18 20:40:45 +0000413 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
Christoph Hellwig193aec12012-03-27 10:34:49 -0400414 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Dave Chinner5c8ed202011-09-18 20:40:45 +0000415 if (error)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000416 return error;
Christoph Hellwig193aec12012-03-27 10:34:49 -0400417
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000418 ASSERT(nimaps > 0);
Christoph Hellwig193aec12012-03-27 10:34:49 -0400419
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000420 /*
421 * If the block underlying isize is just a hole, then there
422 * is nothing to zero.
423 */
Christoph Hellwig193aec12012-03-27 10:34:49 -0400424 if (imap.br_startblock == HOLESTARTBLOCK)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000425 return 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000426
427 zero_len = mp->m_sb.sb_blocksize - zero_offset;
428 if (isize + zero_len > offset)
429 zero_len = offset - isize;
Christoph Hellwig193aec12012-03-27 10:34:49 -0400430 return xfs_iozero(ip, isize, zero_len);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000431}
432
433/*
Christoph Hellwig193aec12012-03-27 10:34:49 -0400434 * Zero any on disk space between the current EOF and the new, larger EOF.
435 *
436 * This handles the normal case of zeroing the remainder of the last block in
437 * the file and the unusual case of zeroing blocks out beyond the size of the
438 * file. This second case only happens with fixed size extents and when the
439 * system crashes before the inode size was updated but after blocks were
440 * allocated.
441 *
442 * Expects the iolock to be held exclusive, and will take the ilock internally.
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000443 */
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000444int /* error (positive) */
445xfs_zero_eof(
Christoph Hellwig193aec12012-03-27 10:34:49 -0400446 struct xfs_inode *ip,
447 xfs_off_t offset, /* starting I/O offset */
448 xfs_fsize_t isize) /* current inode size */
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000449{
Christoph Hellwig193aec12012-03-27 10:34:49 -0400450 struct xfs_mount *mp = ip->i_mount;
451 xfs_fileoff_t start_zero_fsb;
452 xfs_fileoff_t end_zero_fsb;
453 xfs_fileoff_t zero_count_fsb;
454 xfs_fileoff_t last_fsb;
455 xfs_fileoff_t zero_off;
456 xfs_fsize_t zero_len;
457 int nimaps;
458 int error = 0;
459 struct xfs_bmbt_irec imap;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000460
Christoph Hellwig193aec12012-03-27 10:34:49 -0400461 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000462 ASSERT(offset > isize);
463
464 /*
465 * First handle zeroing the block on which isize resides.
Christoph Hellwig193aec12012-03-27 10:34:49 -0400466 *
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000467 * We only zero a part of that block so it is handled specially.
468 */
Christoph Hellwig193aec12012-03-27 10:34:49 -0400469 if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
470 error = xfs_zero_last_block(ip, offset, isize);
471 if (error)
472 return error;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000473 }
474
475 /*
Christoph Hellwig193aec12012-03-27 10:34:49 -0400476 * Calculate the range between the new size and the old where blocks
477 * needing to be zeroed may exist.
478 *
479 * To get the block where the last byte in the file currently resides,
480 * we need to subtract one from the size and truncate back to a block
481 * boundary. We subtract 1 in case the size is exactly on a block
482 * boundary.
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000483 */
484 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
485 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
486 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
487 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
488 if (last_fsb == end_zero_fsb) {
489 /*
490 * The size was only incremented on its last block.
491 * We took care of that above, so just return.
492 */
493 return 0;
494 }
495
496 ASSERT(start_zero_fsb <= end_zero_fsb);
497 while (start_zero_fsb <= end_zero_fsb) {
498 nimaps = 1;
499 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
Christoph Hellwig193aec12012-03-27 10:34:49 -0400500
501 xfs_ilock(ip, XFS_ILOCK_EXCL);
Dave Chinner5c8ed202011-09-18 20:40:45 +0000502 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
503 &imap, &nimaps, 0);
Christoph Hellwig193aec12012-03-27 10:34:49 -0400504 xfs_iunlock(ip, XFS_ILOCK_EXCL);
505 if (error)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000506 return error;
Christoph Hellwig193aec12012-03-27 10:34:49 -0400507
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000508 ASSERT(nimaps > 0);
509
510 if (imap.br_state == XFS_EXT_UNWRITTEN ||
511 imap.br_startblock == HOLESTARTBLOCK) {
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000512 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
513 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
514 continue;
515 }
516
517 /*
518 * There are blocks we need to zero.
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000519 */
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000520 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
521 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
522
523 if ((zero_off + zero_len) > offset)
524 zero_len = offset - zero_off;
525
526 error = xfs_iozero(ip, zero_off, zero_len);
Christoph Hellwig193aec12012-03-27 10:34:49 -0400527 if (error)
528 return error;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000529
530 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
531 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000532 }
533
534 return 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000535}
536
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100537/*
Dave Chinner4d8d1582011-01-11 10:23:42 +1100538 * Common pre-write limit and setup checks.
539 *
Christoph Hellwig5bf1f262011-12-18 20:00:13 +0000540 * Called with the iolocked held either shared and exclusive according to
541 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
542 * if called for a direct write beyond i_size.
Dave Chinner4d8d1582011-01-11 10:23:42 +1100543 */
544STATIC ssize_t
545xfs_file_aio_write_checks(
546 struct file *file,
547 loff_t *pos,
548 size_t *count,
549 int *iolock)
550{
551 struct inode *inode = file->f_mapping->host;
552 struct xfs_inode *ip = XFS_I(inode);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100553 int error = 0;
554
Dave Chinner7271d242011-08-25 07:17:02 +0000555restart:
Dave Chinner4d8d1582011-01-11 10:23:42 +1100556 error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
Christoph Hellwig467f7892012-03-27 10:34:47 -0400557 if (error)
Dave Chinner4d8d1582011-01-11 10:23:42 +1100558 return error;
Dave Chinner4d8d1582011-01-11 10:23:42 +1100559
Dave Chinner4d8d1582011-01-11 10:23:42 +1100560 /*
561 * If the offset is beyond the size of the file, we need to zero any
562 * blocks that fall between the existing EOF and the start of this
Christoph Hellwig2813d682011-12-18 20:00:12 +0000563 * write. If zeroing is needed and we are currently holding the
Christoph Hellwig467f7892012-03-27 10:34:47 -0400564 * iolock shared, we need to update it to exclusive which implies
565 * having to redo all checks before.
Dave Chinner4d8d1582011-01-11 10:23:42 +1100566 */
Christoph Hellwig2813d682011-12-18 20:00:12 +0000567 if (*pos > i_size_read(inode)) {
Dave Chinner7271d242011-08-25 07:17:02 +0000568 if (*iolock == XFS_IOLOCK_SHARED) {
Christoph Hellwig467f7892012-03-27 10:34:47 -0400569 xfs_rw_iunlock(ip, *iolock);
Dave Chinner7271d242011-08-25 07:17:02 +0000570 *iolock = XFS_IOLOCK_EXCL;
Christoph Hellwig467f7892012-03-27 10:34:47 -0400571 xfs_rw_ilock(ip, *iolock);
Dave Chinner7271d242011-08-25 07:17:02 +0000572 goto restart;
573 }
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000574 error = -xfs_zero_eof(ip, *pos, i_size_read(inode));
Christoph Hellwig467f7892012-03-27 10:34:47 -0400575 if (error)
576 return error;
Dave Chinner7271d242011-08-25 07:17:02 +0000577 }
Dave Chinner4d8d1582011-01-11 10:23:42 +1100578
579 /*
Christoph Hellwig8a9c9982012-02-29 09:53:52 +0000580 * Updating the timestamps will grab the ilock again from
581 * xfs_fs_dirty_inode, so we have to call it after dropping the
582 * lock above. Eventually we should look into a way to avoid
583 * the pointless lock roundtrip.
584 */
Josef Bacikc3b2da32012-03-26 09:59:21 -0400585 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
586 error = file_update_time(file);
587 if (error)
588 return error;
589 }
Christoph Hellwig8a9c9982012-02-29 09:53:52 +0000590
591 /*
Dave Chinner4d8d1582011-01-11 10:23:42 +1100592 * If we're writing the file then make sure to clear the setuid and
593 * setgid bits if the process is not being run by root. This keeps
594 * people from modifying setuid and setgid binaries.
595 */
596 return file_remove_suid(file);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100597}
598
599/*
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100600 * xfs_file_dio_aio_write - handle direct IO writes
601 *
602 * Lock the inode appropriately to prepare for and issue a direct IO write.
Dave Chinnereda77982011-01-11 10:22:40 +1100603 * By separating it from the buffered write path we remove all the tricky to
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100604 * follow locking changes and looping.
605 *
Dave Chinnereda77982011-01-11 10:22:40 +1100606 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
607 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
608 * pages are flushed out.
609 *
610 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
611 * allowing them to be done in parallel with reads and other direct IO writes.
612 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
613 * needs to do sub-block zeroing and that requires serialisation against other
614 * direct IOs to the same block. In this case we need to serialise the
615 * submission of the unaligned IOs so that we don't get racing block zeroing in
616 * the dio layer. To avoid the problem with aio, we also need to wait for
617 * outstanding IOs to complete so that unwritten extent conversion is completed
618 * before we try to map the overlapping block. This is currently implemented by
Christoph Hellwig4a06fd22011-08-23 08:28:13 +0000619 * hitting it with a big hammer (i.e. inode_dio_wait()).
Dave Chinnereda77982011-01-11 10:22:40 +1100620 *
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100621 * Returns with locks held indicated by @iolock and errors indicated by
622 * negative return values.
623 */
624STATIC ssize_t
625xfs_file_dio_aio_write(
626 struct kiocb *iocb,
627 const struct iovec *iovp,
628 unsigned long nr_segs,
629 loff_t pos,
Christoph Hellwigd0606462011-12-18 20:00:14 +0000630 size_t ocount)
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100631{
632 struct file *file = iocb->ki_filp;
633 struct address_space *mapping = file->f_mapping;
634 struct inode *inode = mapping->host;
635 struct xfs_inode *ip = XFS_I(inode);
636 struct xfs_mount *mp = ip->i_mount;
637 ssize_t ret = 0;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100638 size_t count = ocount;
Dave Chinnereda77982011-01-11 10:22:40 +1100639 int unaligned_io = 0;
Christoph Hellwigd0606462011-12-18 20:00:14 +0000640 int iolock;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100641 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
642 mp->m_rtdev_targp : mp->m_ddev_targp;
643
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100644 if ((pos & target->bt_smask) || (count & target->bt_smask))
645 return -XFS_ERROR(EINVAL);
646
Dave Chinnereda77982011-01-11 10:22:40 +1100647 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
648 unaligned_io = 1;
649
Dave Chinner7271d242011-08-25 07:17:02 +0000650 /*
651 * We don't need to take an exclusive lock unless there page cache needs
652 * to be invalidated or unaligned IO is being executed. We don't need to
653 * consider the EOF extension case here because
654 * xfs_file_aio_write_checks() will relock the inode as necessary for
655 * EOF zeroing cases and fill out the new inode size as appropriate.
656 */
657 if (unaligned_io || mapping->nrpages)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000658 iolock = XFS_IOLOCK_EXCL;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100659 else
Christoph Hellwigd0606462011-12-18 20:00:14 +0000660 iolock = XFS_IOLOCK_SHARED;
661 xfs_rw_ilock(ip, iolock);
Christoph Hellwigc58cb162011-08-27 14:42:53 +0000662
663 /*
664 * Recheck if there are cached pages that need invalidate after we got
665 * the iolock to protect against other threads adding new pages while
666 * we were waiting for the iolock.
667 */
Christoph Hellwigd0606462011-12-18 20:00:14 +0000668 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
669 xfs_rw_iunlock(ip, iolock);
670 iolock = XFS_IOLOCK_EXCL;
671 xfs_rw_ilock(ip, iolock);
Christoph Hellwigc58cb162011-08-27 14:42:53 +0000672 }
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100673
Christoph Hellwigd0606462011-12-18 20:00:14 +0000674 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100675 if (ret)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000676 goto out;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100677
678 if (mapping->nrpages) {
Dave Chinnerfb595812012-11-12 22:53:57 +1100679 ret = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
680 pos, -1);
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100681 if (ret)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000682 goto out;
Dave Chinnerfb595812012-11-12 22:53:57 +1100683 truncate_pagecache_range(VFS_I(ip), pos, -1);
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100684 }
685
Dave Chinnereda77982011-01-11 10:22:40 +1100686 /*
687 * If we are doing unaligned IO, wait for all other IO to drain,
688 * otherwise demote the lock if we had to flush cached pages
689 */
690 if (unaligned_io)
Christoph Hellwig4a06fd22011-08-23 08:28:13 +0000691 inode_dio_wait(inode);
Christoph Hellwigd0606462011-12-18 20:00:14 +0000692 else if (iolock == XFS_IOLOCK_EXCL) {
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100693 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
Christoph Hellwigd0606462011-12-18 20:00:14 +0000694 iolock = XFS_IOLOCK_SHARED;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100695 }
696
697 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
698 ret = generic_file_direct_write(iocb, iovp,
699 &nr_segs, pos, &iocb->ki_pos, count, ocount);
700
Christoph Hellwigd0606462011-12-18 20:00:14 +0000701out:
702 xfs_rw_iunlock(ip, iolock);
703
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100704 /* No fallback to buffered IO on errors for XFS. */
705 ASSERT(ret < 0 || ret == count);
706 return ret;
707}
708
Christoph Hellwig00258e32010-02-15 09:44:47 +0000709STATIC ssize_t
Dave Chinner637bbc72011-01-11 10:17:30 +1100710xfs_file_buffered_aio_write(
711 struct kiocb *iocb,
712 const struct iovec *iovp,
713 unsigned long nr_segs,
714 loff_t pos,
Christoph Hellwigd0606462011-12-18 20:00:14 +0000715 size_t ocount)
Dave Chinner637bbc72011-01-11 10:17:30 +1100716{
717 struct file *file = iocb->ki_filp;
718 struct address_space *mapping = file->f_mapping;
719 struct inode *inode = mapping->host;
720 struct xfs_inode *ip = XFS_I(inode);
721 ssize_t ret;
722 int enospc = 0;
Christoph Hellwigd0606462011-12-18 20:00:14 +0000723 int iolock = XFS_IOLOCK_EXCL;
Dave Chinner637bbc72011-01-11 10:17:30 +1100724 size_t count = ocount;
725
Christoph Hellwigd0606462011-12-18 20:00:14 +0000726 xfs_rw_ilock(ip, iolock);
Dave Chinner637bbc72011-01-11 10:17:30 +1100727
Christoph Hellwigd0606462011-12-18 20:00:14 +0000728 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100729 if (ret)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000730 goto out;
Dave Chinner637bbc72011-01-11 10:17:30 +1100731
732 /* We can write back this queue in page reclaim */
733 current->backing_dev_info = mapping->backing_dev_info;
734
735write_retry:
736 trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
737 ret = generic_file_buffered_write(iocb, iovp, nr_segs,
Dave Chinner9aa05002012-10-08 21:56:04 +1100738 pos, &iocb->ki_pos, count, 0);
739
Dave Chinner637bbc72011-01-11 10:17:30 +1100740 /*
Dave Chinner9aa05002012-10-08 21:56:04 +1100741 * If we just got an ENOSPC, try to write back all dirty inodes to
742 * convert delalloc space to free up some of the excess reserved
743 * metadata space.
Dave Chinner637bbc72011-01-11 10:17:30 +1100744 */
745 if (ret == -ENOSPC && !enospc) {
Dave Chinner637bbc72011-01-11 10:17:30 +1100746 enospc = 1;
Dave Chinner9aa05002012-10-08 21:56:04 +1100747 xfs_flush_inodes(ip->i_mount);
748 goto write_retry;
Dave Chinner637bbc72011-01-11 10:17:30 +1100749 }
Christoph Hellwigd0606462011-12-18 20:00:14 +0000750
Dave Chinner637bbc72011-01-11 10:17:30 +1100751 current->backing_dev_info = NULL;
Christoph Hellwigd0606462011-12-18 20:00:14 +0000752out:
753 xfs_rw_iunlock(ip, iolock);
Dave Chinner637bbc72011-01-11 10:17:30 +1100754 return ret;
755}
756
757STATIC ssize_t
Christoph Hellwig00258e32010-02-15 09:44:47 +0000758xfs_file_aio_write(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000759 struct kiocb *iocb,
760 const struct iovec *iovp,
Christoph Hellwig00258e32010-02-15 09:44:47 +0000761 unsigned long nr_segs,
762 loff_t pos)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000763{
764 struct file *file = iocb->ki_filp;
765 struct address_space *mapping = file->f_mapping;
766 struct inode *inode = mapping->host;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000767 struct xfs_inode *ip = XFS_I(inode);
Dave Chinner637bbc72011-01-11 10:17:30 +1100768 ssize_t ret;
Dave Chinner637bbc72011-01-11 10:17:30 +1100769 size_t ocount = 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000770
771 XFS_STATS_INC(xs_write_calls);
772
Christoph Hellwig00258e32010-02-15 09:44:47 +0000773 BUG_ON(iocb->ki_pos != pos);
774
Dave Chinnera363f0c2011-01-11 10:13:53 +1100775 ret = generic_segment_checks(iovp, &nr_segs, &ocount, VERIFY_READ);
776 if (ret)
777 return ret;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000778
Dave Chinner637bbc72011-01-11 10:17:30 +1100779 if (ocount == 0)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000780 return 0;
781
Jan Karad9457dc2012-06-12 16:20:39 +0200782 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
783 ret = -EIO;
784 goto out;
785 }
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000786
Dave Chinner637bbc72011-01-11 10:17:30 +1100787 if (unlikely(file->f_flags & O_DIRECT))
Christoph Hellwigd0606462011-12-18 20:00:14 +0000788 ret = xfs_file_dio_aio_write(iocb, iovp, nr_segs, pos, ocount);
Dave Chinner637bbc72011-01-11 10:17:30 +1100789 else
790 ret = xfs_file_buffered_aio_write(iocb, iovp, nr_segs, pos,
Christoph Hellwigd0606462011-12-18 20:00:14 +0000791 ocount);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000792
Christoph Hellwigd0606462011-12-18 20:00:14 +0000793 if (ret > 0) {
794 ssize_t err;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000795
Christoph Hellwigd0606462011-12-18 20:00:14 +0000796 XFS_STATS_ADD(xs_write_bytes, ret);
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000797
Christoph Hellwigd0606462011-12-18 20:00:14 +0000798 /* Handle various SYNC-type writes */
799 err = generic_write_sync(file, pos, ret);
800 if (err < 0)
801 ret = err;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000802 }
803
Jan Karad9457dc2012-06-12 16:20:39 +0200804out:
Dave Chinnera363f0c2011-01-11 10:13:53 +1100805 return ret;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000806}
807
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100808STATIC long
809xfs_file_fallocate(
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700810 struct file *file,
811 int mode,
812 loff_t offset,
813 loff_t len)
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100814{
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700815 struct inode *inode = file_inode(file);
816 struct xfs_inode *ip = XFS_I(inode);
817 struct xfs_trans *tp;
818 long error;
819 loff_t new_size = 0;
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100820
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700821 if (!S_ISREG(inode->i_mode))
822 return -EINVAL;
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100823 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
824 return -EOPNOTSUPP;
825
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100826 xfs_ilock(ip, XFS_IOLOCK_EXCL);
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700827 if (mode & FALLOC_FL_PUNCH_HOLE) {
828 error = xfs_free_file_space(ip, offset, len);
829 if (error)
830 goto out_unlock;
831 } else {
832 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
833 offset + len > i_size_read(inode)) {
834 new_size = offset + len;
835 error = -inode_newsize_ok(inode, new_size);
836 if (error)
837 goto out_unlock;
838 }
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100839
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700840 error = xfs_alloc_file_space(ip, offset, len,
841 XFS_BMAPI_PREALLOC);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100842 if (error)
843 goto out_unlock;
844 }
845
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700846 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID);
847 error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0);
848 if (error) {
849 xfs_trans_cancel(tp, 0);
850 goto out_unlock;
851 }
Dave Chinner82878892011-03-26 09:13:08 +1100852
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700853 xfs_ilock(ip, XFS_ILOCK_EXCL);
854 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
855 ip->i_d.di_mode &= ~S_ISUID;
856 if (ip->i_d.di_mode & S_IXGRP)
857 ip->i_d.di_mode &= ~S_ISGID;
858
859 if (!(mode & FALLOC_FL_PUNCH_HOLE))
860 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
861
862 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
863 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
864
865 if (file->f_flags & O_DSYNC)
866 xfs_trans_set_sync(tp);
867 error = xfs_trans_commit(tp, 0);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100868 if (error)
869 goto out_unlock;
870
871 /* Change file size if needed */
872 if (new_size) {
873 struct iattr iattr;
874
875 iattr.ia_valid = ATTR_SIZE;
876 iattr.ia_size = new_size;
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700877 error = xfs_setattr_size(ip, &iattr);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100878 }
879
880out_unlock:
881 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700882 return -error;
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100883}
884
885
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +1100887xfs_file_open(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 struct inode *inode,
Christoph Hellwigf999a5b2008-11-28 14:23:32 +1100889 struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890{
Christoph Hellwigf999a5b2008-11-28 14:23:32 +1100891 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892 return -EFBIG;
Christoph Hellwigf999a5b2008-11-28 14:23:32 +1100893 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
894 return -EIO;
895 return 0;
896}
897
898STATIC int
899xfs_dir_open(
900 struct inode *inode,
901 struct file *file)
902{
903 struct xfs_inode *ip = XFS_I(inode);
904 int mode;
905 int error;
906
907 error = xfs_file_open(inode, file);
908 if (error)
909 return error;
910
911 /*
912 * If there are any blocks, read-ahead block 0 as we're almost
913 * certain to have the next operation be a read there.
914 */
915 mode = xfs_ilock_map_shared(ip);
916 if (ip->i_d.di_nextents > 0)
Dave Chinner33363fe2013-04-03 16:11:22 +1100917 xfs_dir3_data_readahead(NULL, ip, 0, -1);
Christoph Hellwigf999a5b2008-11-28 14:23:32 +1100918 xfs_iunlock(ip, mode);
919 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920}
921
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +1100923xfs_file_release(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 struct inode *inode,
925 struct file *filp)
926{
Christoph Hellwig739bfb22007-08-29 10:58:01 +1000927 return -xfs_release(XFS_I(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928}
929
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +1100931xfs_file_readdir(
Al Virob8227552013-05-22 17:07:56 -0400932 struct file *file,
933 struct dir_context *ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934{
Al Virob8227552013-05-22 17:07:56 -0400935 struct inode *inode = file_inode(file);
Christoph Hellwig739bfb22007-08-29 10:58:01 +1000936 xfs_inode_t *ip = XFS_I(inode);
Christoph Hellwig051e7cd2007-08-28 13:58:24 +1000937 int error;
938 size_t bufsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939
Christoph Hellwig051e7cd2007-08-28 13:58:24 +1000940 /*
941 * The Linux API doesn't pass down the total size of the buffer
942 * we read into down to the filesystem. With the filldir concept
943 * it's not needed for correct information, but the XFS dir2 leaf
944 * code wants an estimate of the buffer size to calculate it's
945 * readahead window and size the buffers used for mapping to
946 * physical blocks.
947 *
948 * Try to give it an estimate that's good enough, maybe at some
949 * point we can change the ->readdir prototype to include the
Eric Sandeena9cc7992010-02-03 17:50:13 +0000950 * buffer size. For now we use the current glibc buffer size.
Christoph Hellwig051e7cd2007-08-28 13:58:24 +1000951 */
Eric Sandeena9cc7992010-02-03 17:50:13 +0000952 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953
Al Virob8227552013-05-22 17:07:56 -0400954 error = xfs_readdir(ip, ctx, bufsize);
Christoph Hellwig051e7cd2007-08-28 13:58:24 +1000955 if (error)
956 return -error;
957 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958}
959
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +1100961xfs_file_mmap(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 struct file *filp,
963 struct vm_area_struct *vma)
964{
Nathan Scott3562fd42006-03-14 14:00:35 +1100965 vma->vm_ops = &xfs_file_vm_ops;
Dean Roehrich6fac0cb2005-06-21 14:07:45 +1000966
Nathan Scottfbc14622006-06-09 14:52:13 +1000967 file_accessed(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 return 0;
969}
970
David Chinner4f57dbc2007-07-19 16:28:17 +1000971/*
972 * mmap()d file has taken write protection fault and is being made
973 * writable. We can set the page state up correctly for a writable
974 * page, which means we can do correct delalloc accounting (ENOSPC
975 * checking!) and unwritten extent mapping.
976 */
977STATIC int
978xfs_vm_page_mkwrite(
979 struct vm_area_struct *vma,
Nick Pigginc2ec1752009-03-31 15:23:21 -0700980 struct vm_fault *vmf)
David Chinner4f57dbc2007-07-19 16:28:17 +1000981{
Nick Pigginc2ec1752009-03-31 15:23:21 -0700982 return block_page_mkwrite(vma, vmf, xfs_get_blocks);
David Chinner4f57dbc2007-07-19 16:28:17 +1000983}
984
Jeff Liud126d432012-08-21 17:11:57 +0800985/*
986 * This type is designed to indicate the type of offset we would like
987 * to search from page cache for either xfs_seek_data() or xfs_seek_hole().
988 */
989enum {
990 HOLE_OFF = 0,
991 DATA_OFF,
992};
993
994/*
995 * Lookup the desired type of offset from the given page.
996 *
997 * On success, return true and the offset argument will point to the
998 * start of the region that was found. Otherwise this function will
999 * return false and keep the offset argument unchanged.
1000 */
1001STATIC bool
1002xfs_lookup_buffer_offset(
1003 struct page *page,
1004 loff_t *offset,
1005 unsigned int type)
1006{
1007 loff_t lastoff = page_offset(page);
1008 bool found = false;
1009 struct buffer_head *bh, *head;
1010
1011 bh = head = page_buffers(page);
1012 do {
1013 /*
1014 * Unwritten extents that have data in the page
1015 * cache covering them can be identified by the
1016 * BH_Unwritten state flag. Pages with multiple
1017 * buffers might have a mix of holes, data and
1018 * unwritten extents - any buffer with valid
1019 * data in it should have BH_Uptodate flag set
1020 * on it.
1021 */
1022 if (buffer_unwritten(bh) ||
1023 buffer_uptodate(bh)) {
1024 if (type == DATA_OFF)
1025 found = true;
1026 } else {
1027 if (type == HOLE_OFF)
1028 found = true;
1029 }
1030
1031 if (found) {
1032 *offset = lastoff;
1033 break;
1034 }
1035 lastoff += bh->b_size;
1036 } while ((bh = bh->b_this_page) != head);
1037
1038 return found;
1039}
1040
1041/*
1042 * This routine is called to find out and return a data or hole offset
1043 * from the page cache for unwritten extents according to the desired
1044 * type for xfs_seek_data() or xfs_seek_hole().
1045 *
1046 * The argument offset is used to tell where we start to search from the
1047 * page cache. Map is used to figure out the end points of the range to
1048 * lookup pages.
1049 *
1050 * Return true if the desired type of offset was found, and the argument
1051 * offset is filled with that address. Otherwise, return false and keep
1052 * offset unchanged.
1053 */
1054STATIC bool
1055xfs_find_get_desired_pgoff(
1056 struct inode *inode,
1057 struct xfs_bmbt_irec *map,
1058 unsigned int type,
1059 loff_t *offset)
1060{
1061 struct xfs_inode *ip = XFS_I(inode);
1062 struct xfs_mount *mp = ip->i_mount;
1063 struct pagevec pvec;
1064 pgoff_t index;
1065 pgoff_t end;
1066 loff_t endoff;
1067 loff_t startoff = *offset;
1068 loff_t lastoff = startoff;
1069 bool found = false;
1070
1071 pagevec_init(&pvec, 0);
1072
1073 index = startoff >> PAGE_CACHE_SHIFT;
1074 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
1075 end = endoff >> PAGE_CACHE_SHIFT;
1076 do {
1077 int want;
1078 unsigned nr_pages;
1079 unsigned int i;
1080
1081 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1082 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1083 want);
1084 /*
1085 * No page mapped into given range. If we are searching holes
1086 * and if this is the first time we got into the loop, it means
1087 * that the given offset is landed in a hole, return it.
1088 *
1089 * If we have already stepped through some block buffers to find
1090 * holes but they all contains data. In this case, the last
1091 * offset is already updated and pointed to the end of the last
1092 * mapped page, if it does not reach the endpoint to search,
1093 * that means there should be a hole between them.
1094 */
1095 if (nr_pages == 0) {
1096 /* Data search found nothing */
1097 if (type == DATA_OFF)
1098 break;
1099
1100 ASSERT(type == HOLE_OFF);
1101 if (lastoff == startoff || lastoff < endoff) {
1102 found = true;
1103 *offset = lastoff;
1104 }
1105 break;
1106 }
1107
1108 /*
1109 * At lease we found one page. If this is the first time we
1110 * step into the loop, and if the first page index offset is
1111 * greater than the given search offset, a hole was found.
1112 */
1113 if (type == HOLE_OFF && lastoff == startoff &&
1114 lastoff < page_offset(pvec.pages[0])) {
1115 found = true;
1116 break;
1117 }
1118
1119 for (i = 0; i < nr_pages; i++) {
1120 struct page *page = pvec.pages[i];
1121 loff_t b_offset;
1122
1123 /*
1124 * At this point, the page may be truncated or
1125 * invalidated (changing page->mapping to NULL),
1126 * or even swizzled back from swapper_space to tmpfs
1127 * file mapping. However, page->index will not change
1128 * because we have a reference on the page.
1129 *
1130 * Searching done if the page index is out of range.
1131 * If the current offset is not reaches the end of
1132 * the specified search range, there should be a hole
1133 * between them.
1134 */
1135 if (page->index > end) {
1136 if (type == HOLE_OFF && lastoff < endoff) {
1137 *offset = lastoff;
1138 found = true;
1139 }
1140 goto out;
1141 }
1142
1143 lock_page(page);
1144 /*
1145 * Page truncated or invalidated(page->mapping == NULL).
1146 * We can freely skip it and proceed to check the next
1147 * page.
1148 */
1149 if (unlikely(page->mapping != inode->i_mapping)) {
1150 unlock_page(page);
1151 continue;
1152 }
1153
1154 if (!page_has_buffers(page)) {
1155 unlock_page(page);
1156 continue;
1157 }
1158
1159 found = xfs_lookup_buffer_offset(page, &b_offset, type);
1160 if (found) {
1161 /*
1162 * The found offset may be less than the start
1163 * point to search if this is the first time to
1164 * come here.
1165 */
1166 *offset = max_t(loff_t, startoff, b_offset);
1167 unlock_page(page);
1168 goto out;
1169 }
1170
1171 /*
1172 * We either searching data but nothing was found, or
1173 * searching hole but found a data buffer. In either
1174 * case, probably the next page contains the desired
1175 * things, update the last offset to it so.
1176 */
1177 lastoff = page_offset(page) + PAGE_SIZE;
1178 unlock_page(page);
1179 }
1180
1181 /*
1182 * The number of returned pages less than our desired, search
1183 * done. In this case, nothing was found for searching data,
1184 * but we found a hole behind the last offset.
1185 */
1186 if (nr_pages < want) {
1187 if (type == HOLE_OFF) {
1188 *offset = lastoff;
1189 found = true;
1190 }
1191 break;
1192 }
1193
1194 index = pvec.pages[i - 1]->index + 1;
1195 pagevec_release(&pvec);
1196 } while (index <= end);
1197
1198out:
1199 pagevec_release(&pvec);
1200 return found;
1201}
1202
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001203STATIC loff_t
1204xfs_seek_data(
1205 struct file *file,
Jeff Liu834ab122012-08-21 17:11:45 +08001206 loff_t start)
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001207{
1208 struct inode *inode = file->f_mapping->host;
1209 struct xfs_inode *ip = XFS_I(inode);
1210 struct xfs_mount *mp = ip->i_mount;
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001211 loff_t uninitialized_var(offset);
1212 xfs_fsize_t isize;
1213 xfs_fileoff_t fsbno;
1214 xfs_filblks_t end;
1215 uint lock;
1216 int error;
1217
1218 lock = xfs_ilock_map_shared(ip);
1219
1220 isize = i_size_read(inode);
1221 if (start >= isize) {
1222 error = ENXIO;
1223 goto out_unlock;
1224 }
1225
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001226 /*
1227 * Try to read extents from the first block indicated
1228 * by fsbno to the end block of the file.
1229 */
Jeff Liu52f1acc2012-08-21 17:12:07 +08001230 fsbno = XFS_B_TO_FSBT(mp, start);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001231 end = XFS_B_TO_FSB(mp, isize);
Jeff Liu52f1acc2012-08-21 17:12:07 +08001232 for (;;) {
1233 struct xfs_bmbt_irec map[2];
1234 int nmap = 2;
1235 unsigned int i;
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001236
Jeff Liu52f1acc2012-08-21 17:12:07 +08001237 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1238 XFS_BMAPI_ENTIRE);
1239 if (error)
1240 goto out_unlock;
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001241
Jeff Liu52f1acc2012-08-21 17:12:07 +08001242 /* No extents at given offset, must be beyond EOF */
1243 if (nmap == 0) {
1244 error = ENXIO;
1245 goto out_unlock;
1246 }
1247
1248 for (i = 0; i < nmap; i++) {
1249 offset = max_t(loff_t, start,
1250 XFS_FSB_TO_B(mp, map[i].br_startoff));
1251
1252 /* Landed in a data extent */
1253 if (map[i].br_startblock == DELAYSTARTBLOCK ||
1254 (map[i].br_state == XFS_EXT_NORM &&
1255 !isnullstartblock(map[i].br_startblock)))
1256 goto out;
1257
1258 /*
1259 * Landed in an unwritten extent, try to search data
1260 * from page cache.
1261 */
1262 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1263 if (xfs_find_get_desired_pgoff(inode, &map[i],
1264 DATA_OFF, &offset))
1265 goto out;
1266 }
1267 }
1268
1269 /*
1270 * map[0] is hole or its an unwritten extent but
1271 * without data in page cache. Probably means that
1272 * we are reading after EOF if nothing in map[1].
1273 */
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001274 if (nmap == 1) {
1275 error = ENXIO;
1276 goto out_unlock;
1277 }
1278
Jeff Liu52f1acc2012-08-21 17:12:07 +08001279 ASSERT(i > 1);
1280
1281 /*
1282 * Nothing was found, proceed to the next round of search
1283 * if reading offset not beyond or hit EOF.
1284 */
1285 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1286 start = XFS_FSB_TO_B(mp, fsbno);
1287 if (start >= isize) {
1288 error = ENXIO;
1289 goto out_unlock;
1290 }
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001291 }
1292
Jeff Liu52f1acc2012-08-21 17:12:07 +08001293out:
Jie Liu46a1c2c2013-06-25 12:02:13 +08001294 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001295
1296out_unlock:
1297 xfs_iunlock_map_shared(ip, lock);
1298
1299 if (error)
1300 return -error;
1301 return offset;
1302}
1303
1304STATIC loff_t
1305xfs_seek_hole(
1306 struct file *file,
Jeff Liu834ab122012-08-21 17:11:45 +08001307 loff_t start)
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001308{
1309 struct inode *inode = file->f_mapping->host;
1310 struct xfs_inode *ip = XFS_I(inode);
1311 struct xfs_mount *mp = ip->i_mount;
1312 loff_t uninitialized_var(offset);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001313 xfs_fsize_t isize;
1314 xfs_fileoff_t fsbno;
Jeff Liub686d1f2012-08-21 17:12:18 +08001315 xfs_filblks_t end;
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001316 uint lock;
1317 int error;
1318
1319 if (XFS_FORCED_SHUTDOWN(mp))
1320 return -XFS_ERROR(EIO);
1321
1322 lock = xfs_ilock_map_shared(ip);
1323
1324 isize = i_size_read(inode);
1325 if (start >= isize) {
1326 error = ENXIO;
1327 goto out_unlock;
1328 }
1329
1330 fsbno = XFS_B_TO_FSBT(mp, start);
Jeff Liub686d1f2012-08-21 17:12:18 +08001331 end = XFS_B_TO_FSB(mp, isize);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001332
Jeff Liub686d1f2012-08-21 17:12:18 +08001333 for (;;) {
1334 struct xfs_bmbt_irec map[2];
1335 int nmap = 2;
1336 unsigned int i;
1337
1338 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1339 XFS_BMAPI_ENTIRE);
1340 if (error)
1341 goto out_unlock;
1342
1343 /* No extents at given offset, must be beyond EOF */
1344 if (nmap == 0) {
1345 error = ENXIO;
1346 goto out_unlock;
1347 }
1348
1349 for (i = 0; i < nmap; i++) {
1350 offset = max_t(loff_t, start,
1351 XFS_FSB_TO_B(mp, map[i].br_startoff));
1352
1353 /* Landed in a hole */
1354 if (map[i].br_startblock == HOLESTARTBLOCK)
1355 goto out;
1356
1357 /*
1358 * Landed in an unwritten extent, try to search hole
1359 * from page cache.
1360 */
1361 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1362 if (xfs_find_get_desired_pgoff(inode, &map[i],
1363 HOLE_OFF, &offset))
1364 goto out;
1365 }
1366 }
1367
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001368 /*
Jeff Liub686d1f2012-08-21 17:12:18 +08001369 * map[0] contains data or its unwritten but contains
1370 * data in page cache, probably means that we are
1371 * reading after EOF. We should fix offset to point
1372 * to the end of the file(i.e., there is an implicit
1373 * hole at the end of any file).
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001374 */
Jeff Liub686d1f2012-08-21 17:12:18 +08001375 if (nmap == 1) {
1376 offset = isize;
1377 break;
1378 }
1379
1380 ASSERT(i > 1);
1381
1382 /*
1383 * Both mappings contains data, proceed to the next round of
1384 * search if the current reading offset not beyond or hit EOF.
1385 */
1386 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1387 start = XFS_FSB_TO_B(mp, fsbno);
1388 if (start >= isize) {
1389 offset = isize;
1390 break;
1391 }
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001392 }
1393
Jeff Liub686d1f2012-08-21 17:12:18 +08001394out:
1395 /*
1396 * At this point, we must have found a hole. However, the returned
1397 * offset may be bigger than the file size as it may be aligned to
1398 * page boundary for unwritten extents, we need to deal with this
1399 * situation in particular.
1400 */
1401 offset = min_t(loff_t, offset, isize);
Jie Liu46a1c2c2013-06-25 12:02:13 +08001402 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001403
1404out_unlock:
1405 xfs_iunlock_map_shared(ip, lock);
1406
1407 if (error)
1408 return -error;
1409 return offset;
1410}
1411
1412STATIC loff_t
1413xfs_file_llseek(
1414 struct file *file,
1415 loff_t offset,
1416 int origin)
1417{
1418 switch (origin) {
1419 case SEEK_END:
1420 case SEEK_CUR:
1421 case SEEK_SET:
1422 return generic_file_llseek(file, offset, origin);
1423 case SEEK_DATA:
Jeff Liu834ab122012-08-21 17:11:45 +08001424 return xfs_seek_data(file, offset);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001425 case SEEK_HOLE:
Jeff Liu834ab122012-08-21 17:11:45 +08001426 return xfs_seek_hole(file, offset);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001427 default:
1428 return -EINVAL;
1429 }
1430}
1431
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001432const struct file_operations xfs_file_operations = {
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001433 .llseek = xfs_file_llseek,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 .read = do_sync_read,
Dean Roehrichbb3f7242005-09-02 15:43:05 +10001435 .write = do_sync_write,
Nathan Scott3562fd42006-03-14 14:00:35 +11001436 .aio_read = xfs_file_aio_read,
1437 .aio_write = xfs_file_aio_write,
Nathan Scott1b895842006-03-31 13:08:59 +10001438 .splice_read = xfs_file_splice_read,
1439 .splice_write = xfs_file_splice_write,
Nathan Scott3562fd42006-03-14 14:00:35 +11001440 .unlocked_ioctl = xfs_file_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441#ifdef CONFIG_COMPAT
Nathan Scott3562fd42006-03-14 14:00:35 +11001442 .compat_ioctl = xfs_file_compat_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443#endif
Nathan Scott3562fd42006-03-14 14:00:35 +11001444 .mmap = xfs_file_mmap,
1445 .open = xfs_file_open,
1446 .release = xfs_file_release,
1447 .fsync = xfs_file_fsync,
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001448 .fallocate = xfs_file_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449};
1450
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001451const struct file_operations xfs_dir_file_operations = {
Christoph Hellwigf999a5b2008-11-28 14:23:32 +11001452 .open = xfs_dir_open,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 .read = generic_read_dir,
Al Virob8227552013-05-22 17:07:56 -04001454 .iterate = xfs_file_readdir,
Al Viro59af1582008-08-24 07:24:41 -04001455 .llseek = generic_file_llseek,
Nathan Scott3562fd42006-03-14 14:00:35 +11001456 .unlocked_ioctl = xfs_file_ioctl,
Nathan Scottd3870392005-05-06 06:44:46 -07001457#ifdef CONFIG_COMPAT
Nathan Scott3562fd42006-03-14 14:00:35 +11001458 .compat_ioctl = xfs_file_compat_ioctl,
Nathan Scottd3870392005-05-06 06:44:46 -07001459#endif
Christoph Hellwig1da2f2d2011-10-02 14:25:16 +00001460 .fsync = xfs_dir_fsync,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461};
1462
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001463static const struct vm_operations_struct xfs_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07001464 .fault = filemap_fault,
David Chinner4f57dbc2007-07-19 16:28:17 +10001465 .page_mkwrite = xfs_vm_page_mkwrite,
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -07001466 .remap_pages = generic_file_remap_pages,
Dean Roehrich6fac0cb2005-06-21 14:07:45 +10001467};