blob: 64b48eade91d14c79408b6863f199e9181350f81 [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;
Eric Sandeen7c71ee72014-01-21 16:46:23 -0600264 /* DIO must be aligned to device logical sector size */
265 if ((pos | size) & target->bt_logical_sectormask) {
Dave Chinnerfb595812012-11-12 22:53:57 +1100266 if (pos == i_size_read(inode))
Christoph Hellwig00258e32010-02-15 09:44:47 +0000267 return 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000268 return -XFS_ERROR(EINVAL);
269 }
270 }
271
Dave Chinnerfb595812012-11-12 22:53:57 +1100272 n = mp->m_super->s_maxbytes - pos;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000273 if (n <= 0 || size == 0)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000274 return 0;
275
276 if (n < size)
277 size = n;
278
279 if (XFS_FORCED_SHUTDOWN(mp))
280 return -EIO;
281
Dave Chinner0c38a252011-08-25 07:17:01 +0000282 /*
283 * Locking is a bit tricky here. If we take an exclusive lock
284 * for direct IO, we effectively serialise all new concurrent
285 * read IO to this file and block it behind IO that is currently in
286 * progress because IO in progress holds the IO lock shared. We only
287 * need to hold the lock exclusive to blow away the page cache, so
288 * only take lock exclusively if the page cache needs invalidation.
289 * This allows the normal direct IO case of no page cache pages to
290 * proceeed concurrently without serialisation.
291 */
292 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
293 if ((ioflags & IO_ISDIRECT) && inode->i_mapping->nrpages) {
294 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
Dave Chinner487f84f2011-01-12 11:37:10 +1100295 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
296
Christoph Hellwig00258e32010-02-15 09:44:47 +0000297 if (inode->i_mapping->nrpages) {
Dave Chinnerfb595812012-11-12 22:53:57 +1100298 ret = -filemap_write_and_wait_range(
299 VFS_I(ip)->i_mapping,
300 pos, -1);
Dave Chinner487f84f2011-01-12 11:37:10 +1100301 if (ret) {
302 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
303 return ret;
304 }
Dave Chinnerfb595812012-11-12 22:53:57 +1100305 truncate_pagecache_range(VFS_I(ip), pos, -1);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000306 }
Dave Chinner487f84f2011-01-12 11:37:10 +1100307 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
Dave Chinner0c38a252011-08-25 07:17:01 +0000308 }
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000309
Dave Chinnerfb595812012-11-12 22:53:57 +1100310 trace_xfs_file_read(ip, size, pos, ioflags);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000311
Dave Chinnerfb595812012-11-12 22:53:57 +1100312 ret = generic_file_aio_read(iocb, iovp, nr_segs, pos);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000313 if (ret > 0)
314 XFS_STATS_ADD(xs_read_bytes, ret);
315
Dave Chinner487f84f2011-01-12 11:37:10 +1100316 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000317 return ret;
318}
319
Christoph Hellwig00258e32010-02-15 09:44:47 +0000320STATIC ssize_t
321xfs_file_splice_read(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000322 struct file *infilp,
323 loff_t *ppos,
324 struct pipe_inode_info *pipe,
325 size_t count,
Christoph Hellwig00258e32010-02-15 09:44:47 +0000326 unsigned int flags)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000327{
Christoph Hellwig00258e32010-02-15 09:44:47 +0000328 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000329 int ioflags = 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000330 ssize_t ret;
331
332 XFS_STATS_INC(xs_read_calls);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000333
334 if (infilp->f_mode & FMODE_NOCMTIME)
335 ioflags |= IO_INVIS;
336
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000337 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
338 return -EIO;
339
Dave Chinner487f84f2011-01-12 11:37:10 +1100340 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000341
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000342 trace_xfs_file_splice_read(ip, count, *ppos, ioflags);
343
344 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
345 if (ret > 0)
346 XFS_STATS_ADD(xs_read_bytes, ret);
347
Dave Chinner487f84f2011-01-12 11:37:10 +1100348 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000349 return ret;
350}
351
Dave Chinner4c5cfd12011-01-11 10:14:16 +1100352/*
Dave Chinner487f84f2011-01-12 11:37:10 +1100353 * xfs_file_splice_write() does not use xfs_rw_ilock() because
354 * generic_file_splice_write() takes the i_mutex itself. This, in theory,
355 * couuld cause lock inversions between the aio_write path and the splice path
356 * if someone is doing concurrent splice(2) based writes and write(2) based
357 * writes to the same inode. The only real way to fix this is to re-implement
358 * the generic code here with correct locking orders.
359 */
Christoph Hellwig00258e32010-02-15 09:44:47 +0000360STATIC ssize_t
361xfs_file_splice_write(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000362 struct pipe_inode_info *pipe,
363 struct file *outfilp,
364 loff_t *ppos,
365 size_t count,
Christoph Hellwig00258e32010-02-15 09:44:47 +0000366 unsigned int flags)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000367{
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000368 struct inode *inode = outfilp->f_mapping->host;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000369 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000370 int ioflags = 0;
371 ssize_t ret;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000372
373 XFS_STATS_INC(xs_write_calls);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000374
375 if (outfilp->f_mode & FMODE_NOCMTIME)
376 ioflags |= IO_INVIS;
377
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000378 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
379 return -EIO;
380
381 xfs_ilock(ip, XFS_IOLOCK_EXCL);
382
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000383 trace_xfs_file_splice_write(ip, count, *ppos, ioflags);
384
385 ret = generic_file_splice_write(pipe, outfilp, ppos, count, flags);
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000386 if (ret > 0)
387 XFS_STATS_ADD(xs_write_bytes, ret);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000388
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000389 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
390 return ret;
391}
392
393/*
Christoph Hellwig193aec12012-03-27 10:34:49 -0400394 * This routine is called to handle zeroing any space in the last block of the
395 * file that is beyond the EOF. We do this since the size is being increased
396 * without writing anything to that block and we don't want to read the
397 * garbage on the disk.
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000398 */
399STATIC int /* error (positive) */
400xfs_zero_last_block(
Christoph Hellwig193aec12012-03-27 10:34:49 -0400401 struct xfs_inode *ip,
402 xfs_fsize_t offset,
403 xfs_fsize_t isize)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000404{
Christoph Hellwig193aec12012-03-27 10:34:49 -0400405 struct xfs_mount *mp = ip->i_mount;
406 xfs_fileoff_t last_fsb = XFS_B_TO_FSBT(mp, isize);
407 int zero_offset = XFS_B_FSB_OFFSET(mp, isize);
408 int zero_len;
409 int nimaps = 1;
410 int error = 0;
411 struct xfs_bmbt_irec imap;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000412
Christoph Hellwig193aec12012-03-27 10:34:49 -0400413 xfs_ilock(ip, XFS_ILOCK_EXCL);
Dave Chinner5c8ed202011-09-18 20:40:45 +0000414 error = xfs_bmapi_read(ip, last_fsb, 1, &imap, &nimaps, 0);
Christoph Hellwig193aec12012-03-27 10:34:49 -0400415 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Dave Chinner5c8ed202011-09-18 20:40:45 +0000416 if (error)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000417 return error;
Christoph Hellwig193aec12012-03-27 10:34:49 -0400418
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000419 ASSERT(nimaps > 0);
Christoph Hellwig193aec12012-03-27 10:34:49 -0400420
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000421 /*
422 * If the block underlying isize is just a hole, then there
423 * is nothing to zero.
424 */
Christoph Hellwig193aec12012-03-27 10:34:49 -0400425 if (imap.br_startblock == HOLESTARTBLOCK)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000426 return 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000427
428 zero_len = mp->m_sb.sb_blocksize - zero_offset;
429 if (isize + zero_len > offset)
430 zero_len = offset - isize;
Christoph Hellwig193aec12012-03-27 10:34:49 -0400431 return xfs_iozero(ip, isize, zero_len);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000432}
433
434/*
Christoph Hellwig193aec12012-03-27 10:34:49 -0400435 * Zero any on disk space between the current EOF and the new, larger EOF.
436 *
437 * This handles the normal case of zeroing the remainder of the last block in
438 * the file and the unusual case of zeroing blocks out beyond the size of the
439 * file. This second case only happens with fixed size extents and when the
440 * system crashes before the inode size was updated but after blocks were
441 * allocated.
442 *
443 * Expects the iolock to be held exclusive, and will take the ilock internally.
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000444 */
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000445int /* error (positive) */
446xfs_zero_eof(
Christoph Hellwig193aec12012-03-27 10:34:49 -0400447 struct xfs_inode *ip,
448 xfs_off_t offset, /* starting I/O offset */
449 xfs_fsize_t isize) /* current inode size */
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000450{
Christoph Hellwig193aec12012-03-27 10:34:49 -0400451 struct xfs_mount *mp = ip->i_mount;
452 xfs_fileoff_t start_zero_fsb;
453 xfs_fileoff_t end_zero_fsb;
454 xfs_fileoff_t zero_count_fsb;
455 xfs_fileoff_t last_fsb;
456 xfs_fileoff_t zero_off;
457 xfs_fsize_t zero_len;
458 int nimaps;
459 int error = 0;
460 struct xfs_bmbt_irec imap;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000461
Christoph Hellwig193aec12012-03-27 10:34:49 -0400462 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000463 ASSERT(offset > isize);
464
465 /*
466 * First handle zeroing the block on which isize resides.
Christoph Hellwig193aec12012-03-27 10:34:49 -0400467 *
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000468 * We only zero a part of that block so it is handled specially.
469 */
Christoph Hellwig193aec12012-03-27 10:34:49 -0400470 if (XFS_B_FSB_OFFSET(mp, isize) != 0) {
471 error = xfs_zero_last_block(ip, offset, isize);
472 if (error)
473 return error;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000474 }
475
476 /*
Christoph Hellwig193aec12012-03-27 10:34:49 -0400477 * Calculate the range between the new size and the old where blocks
478 * needing to be zeroed may exist.
479 *
480 * To get the block where the last byte in the file currently resides,
481 * we need to subtract one from the size and truncate back to a block
482 * boundary. We subtract 1 in case the size is exactly on a block
483 * boundary.
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000484 */
485 last_fsb = isize ? XFS_B_TO_FSBT(mp, isize - 1) : (xfs_fileoff_t)-1;
486 start_zero_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)isize);
487 end_zero_fsb = XFS_B_TO_FSBT(mp, offset - 1);
488 ASSERT((xfs_sfiloff_t)last_fsb < (xfs_sfiloff_t)start_zero_fsb);
489 if (last_fsb == end_zero_fsb) {
490 /*
491 * The size was only incremented on its last block.
492 * We took care of that above, so just return.
493 */
494 return 0;
495 }
496
497 ASSERT(start_zero_fsb <= end_zero_fsb);
498 while (start_zero_fsb <= end_zero_fsb) {
499 nimaps = 1;
500 zero_count_fsb = end_zero_fsb - start_zero_fsb + 1;
Christoph Hellwig193aec12012-03-27 10:34:49 -0400501
502 xfs_ilock(ip, XFS_ILOCK_EXCL);
Dave Chinner5c8ed202011-09-18 20:40:45 +0000503 error = xfs_bmapi_read(ip, start_zero_fsb, zero_count_fsb,
504 &imap, &nimaps, 0);
Christoph Hellwig193aec12012-03-27 10:34:49 -0400505 xfs_iunlock(ip, XFS_ILOCK_EXCL);
506 if (error)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000507 return error;
Christoph Hellwig193aec12012-03-27 10:34:49 -0400508
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000509 ASSERT(nimaps > 0);
510
511 if (imap.br_state == XFS_EXT_UNWRITTEN ||
512 imap.br_startblock == HOLESTARTBLOCK) {
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000513 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
514 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
515 continue;
516 }
517
518 /*
519 * There are blocks we need to zero.
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000520 */
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000521 zero_off = XFS_FSB_TO_B(mp, start_zero_fsb);
522 zero_len = XFS_FSB_TO_B(mp, imap.br_blockcount);
523
524 if ((zero_off + zero_len) > offset)
525 zero_len = offset - zero_off;
526
527 error = xfs_iozero(ip, zero_off, zero_len);
Christoph Hellwig193aec12012-03-27 10:34:49 -0400528 if (error)
529 return error;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000530
531 start_zero_fsb = imap.br_startoff + imap.br_blockcount;
532 ASSERT(start_zero_fsb <= (end_zero_fsb + 1));
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000533 }
534
535 return 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000536}
537
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100538/*
Dave Chinner4d8d1582011-01-11 10:23:42 +1100539 * Common pre-write limit and setup checks.
540 *
Christoph Hellwig5bf1f262011-12-18 20:00:13 +0000541 * Called with the iolocked held either shared and exclusive according to
542 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
543 * if called for a direct write beyond i_size.
Dave Chinner4d8d1582011-01-11 10:23:42 +1100544 */
545STATIC ssize_t
546xfs_file_aio_write_checks(
547 struct file *file,
548 loff_t *pos,
549 size_t *count,
550 int *iolock)
551{
552 struct inode *inode = file->f_mapping->host;
553 struct xfs_inode *ip = XFS_I(inode);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100554 int error = 0;
555
Dave Chinner7271d242011-08-25 07:17:02 +0000556restart:
Dave Chinner4d8d1582011-01-11 10:23:42 +1100557 error = generic_write_checks(file, pos, count, S_ISBLK(inode->i_mode));
Christoph Hellwig467f7892012-03-27 10:34:47 -0400558 if (error)
Dave Chinner4d8d1582011-01-11 10:23:42 +1100559 return error;
Dave Chinner4d8d1582011-01-11 10:23:42 +1100560
Dave Chinner4d8d1582011-01-11 10:23:42 +1100561 /*
562 * If the offset is beyond the size of the file, we need to zero any
563 * blocks that fall between the existing EOF and the start of this
Christoph Hellwig2813d682011-12-18 20:00:12 +0000564 * write. If zeroing is needed and we are currently holding the
Christoph Hellwig467f7892012-03-27 10:34:47 -0400565 * iolock shared, we need to update it to exclusive which implies
566 * having to redo all checks before.
Dave Chinner4d8d1582011-01-11 10:23:42 +1100567 */
Christoph Hellwig2813d682011-12-18 20:00:12 +0000568 if (*pos > i_size_read(inode)) {
Dave Chinner7271d242011-08-25 07:17:02 +0000569 if (*iolock == XFS_IOLOCK_SHARED) {
Christoph Hellwig467f7892012-03-27 10:34:47 -0400570 xfs_rw_iunlock(ip, *iolock);
Dave Chinner7271d242011-08-25 07:17:02 +0000571 *iolock = XFS_IOLOCK_EXCL;
Christoph Hellwig467f7892012-03-27 10:34:47 -0400572 xfs_rw_ilock(ip, *iolock);
Dave Chinner7271d242011-08-25 07:17:02 +0000573 goto restart;
574 }
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000575 error = -xfs_zero_eof(ip, *pos, i_size_read(inode));
Christoph Hellwig467f7892012-03-27 10:34:47 -0400576 if (error)
577 return error;
Dave Chinner7271d242011-08-25 07:17:02 +0000578 }
Dave Chinner4d8d1582011-01-11 10:23:42 +1100579
580 /*
Christoph Hellwig8a9c9982012-02-29 09:53:52 +0000581 * Updating the timestamps will grab the ilock again from
582 * xfs_fs_dirty_inode, so we have to call it after dropping the
583 * lock above. Eventually we should look into a way to avoid
584 * the pointless lock roundtrip.
585 */
Josef Bacikc3b2da32012-03-26 09:59:21 -0400586 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
587 error = file_update_time(file);
588 if (error)
589 return error;
590 }
Christoph Hellwig8a9c9982012-02-29 09:53:52 +0000591
592 /*
Dave Chinner4d8d1582011-01-11 10:23:42 +1100593 * If we're writing the file then make sure to clear the setuid and
594 * setgid bits if the process is not being run by root. This keeps
595 * people from modifying setuid and setgid binaries.
596 */
597 return file_remove_suid(file);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100598}
599
600/*
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100601 * xfs_file_dio_aio_write - handle direct IO writes
602 *
603 * Lock the inode appropriately to prepare for and issue a direct IO write.
Dave Chinnereda77982011-01-11 10:22:40 +1100604 * By separating it from the buffered write path we remove all the tricky to
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100605 * follow locking changes and looping.
606 *
Dave Chinnereda77982011-01-11 10:22:40 +1100607 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
608 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
609 * pages are flushed out.
610 *
611 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
612 * allowing them to be done in parallel with reads and other direct IO writes.
613 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
614 * needs to do sub-block zeroing and that requires serialisation against other
615 * direct IOs to the same block. In this case we need to serialise the
616 * submission of the unaligned IOs so that we don't get racing block zeroing in
617 * the dio layer. To avoid the problem with aio, we also need to wait for
618 * outstanding IOs to complete so that unwritten extent conversion is completed
619 * before we try to map the overlapping block. This is currently implemented by
Christoph Hellwig4a06fd22011-08-23 08:28:13 +0000620 * hitting it with a big hammer (i.e. inode_dio_wait()).
Dave Chinnereda77982011-01-11 10:22:40 +1100621 *
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100622 * Returns with locks held indicated by @iolock and errors indicated by
623 * negative return values.
624 */
625STATIC ssize_t
626xfs_file_dio_aio_write(
627 struct kiocb *iocb,
628 const struct iovec *iovp,
629 unsigned long nr_segs,
630 loff_t pos,
Christoph Hellwigd0606462011-12-18 20:00:14 +0000631 size_t ocount)
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100632{
633 struct file *file = iocb->ki_filp;
634 struct address_space *mapping = file->f_mapping;
635 struct inode *inode = mapping->host;
636 struct xfs_inode *ip = XFS_I(inode);
637 struct xfs_mount *mp = ip->i_mount;
638 ssize_t ret = 0;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100639 size_t count = ocount;
Dave Chinnereda77982011-01-11 10:22:40 +1100640 int unaligned_io = 0;
Christoph Hellwigd0606462011-12-18 20:00:14 +0000641 int iolock;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100642 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
643 mp->m_rtdev_targp : mp->m_ddev_targp;
644
Eric Sandeen7c71ee72014-01-21 16:46:23 -0600645 /* DIO must be aligned to device logical sector size */
646 if ((pos | count) & target->bt_logical_sectormask)
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100647 return -XFS_ERROR(EINVAL);
648
Eric Sandeen7c71ee72014-01-21 16:46:23 -0600649 /* "unaligned" here means not aligned to a filesystem block */
Dave Chinnereda77982011-01-11 10:22:40 +1100650 if ((pos & mp->m_blockmask) || ((pos + count) & mp->m_blockmask))
651 unaligned_io = 1;
652
Dave Chinner7271d242011-08-25 07:17:02 +0000653 /*
654 * We don't need to take an exclusive lock unless there page cache needs
655 * to be invalidated or unaligned IO is being executed. We don't need to
656 * consider the EOF extension case here because
657 * xfs_file_aio_write_checks() will relock the inode as necessary for
658 * EOF zeroing cases and fill out the new inode size as appropriate.
659 */
660 if (unaligned_io || mapping->nrpages)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000661 iolock = XFS_IOLOCK_EXCL;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100662 else
Christoph Hellwigd0606462011-12-18 20:00:14 +0000663 iolock = XFS_IOLOCK_SHARED;
664 xfs_rw_ilock(ip, iolock);
Christoph Hellwigc58cb162011-08-27 14:42:53 +0000665
666 /*
667 * Recheck if there are cached pages that need invalidate after we got
668 * the iolock to protect against other threads adding new pages while
669 * we were waiting for the iolock.
670 */
Christoph Hellwigd0606462011-12-18 20:00:14 +0000671 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
672 xfs_rw_iunlock(ip, iolock);
673 iolock = XFS_IOLOCK_EXCL;
674 xfs_rw_ilock(ip, iolock);
Christoph Hellwigc58cb162011-08-27 14:42:53 +0000675 }
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100676
Christoph Hellwigd0606462011-12-18 20:00:14 +0000677 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100678 if (ret)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000679 goto out;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100680
681 if (mapping->nrpages) {
Dave Chinnerfb595812012-11-12 22:53:57 +1100682 ret = -filemap_write_and_wait_range(VFS_I(ip)->i_mapping,
683 pos, -1);
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100684 if (ret)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000685 goto out;
Dave Chinnerfb595812012-11-12 22:53:57 +1100686 truncate_pagecache_range(VFS_I(ip), pos, -1);
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100687 }
688
Dave Chinnereda77982011-01-11 10:22:40 +1100689 /*
690 * If we are doing unaligned IO, wait for all other IO to drain,
691 * otherwise demote the lock if we had to flush cached pages
692 */
693 if (unaligned_io)
Christoph Hellwig4a06fd22011-08-23 08:28:13 +0000694 inode_dio_wait(inode);
Christoph Hellwigd0606462011-12-18 20:00:14 +0000695 else if (iolock == XFS_IOLOCK_EXCL) {
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100696 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
Christoph Hellwigd0606462011-12-18 20:00:14 +0000697 iolock = XFS_IOLOCK_SHARED;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100698 }
699
700 trace_xfs_file_direct_write(ip, count, iocb->ki_pos, 0);
701 ret = generic_file_direct_write(iocb, iovp,
702 &nr_segs, pos, &iocb->ki_pos, count, ocount);
703
Christoph Hellwigd0606462011-12-18 20:00:14 +0000704out:
705 xfs_rw_iunlock(ip, iolock);
706
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100707 /* No fallback to buffered IO on errors for XFS. */
708 ASSERT(ret < 0 || ret == count);
709 return ret;
710}
711
Christoph Hellwig00258e32010-02-15 09:44:47 +0000712STATIC ssize_t
Dave Chinner637bbc72011-01-11 10:17:30 +1100713xfs_file_buffered_aio_write(
714 struct kiocb *iocb,
715 const struct iovec *iovp,
716 unsigned long nr_segs,
717 loff_t pos,
Christoph Hellwigd0606462011-12-18 20:00:14 +0000718 size_t ocount)
Dave Chinner637bbc72011-01-11 10:17:30 +1100719{
720 struct file *file = iocb->ki_filp;
721 struct address_space *mapping = file->f_mapping;
722 struct inode *inode = mapping->host;
723 struct xfs_inode *ip = XFS_I(inode);
724 ssize_t ret;
725 int enospc = 0;
Christoph Hellwigd0606462011-12-18 20:00:14 +0000726 int iolock = XFS_IOLOCK_EXCL;
Dave Chinner637bbc72011-01-11 10:17:30 +1100727 size_t count = ocount;
728
Christoph Hellwigd0606462011-12-18 20:00:14 +0000729 xfs_rw_ilock(ip, iolock);
Dave Chinner637bbc72011-01-11 10:17:30 +1100730
Christoph Hellwigd0606462011-12-18 20:00:14 +0000731 ret = xfs_file_aio_write_checks(file, &pos, &count, &iolock);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100732 if (ret)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000733 goto out;
Dave Chinner637bbc72011-01-11 10:17:30 +1100734
735 /* We can write back this queue in page reclaim */
736 current->backing_dev_info = mapping->backing_dev_info;
737
738write_retry:
739 trace_xfs_file_buffered_write(ip, count, iocb->ki_pos, 0);
740 ret = generic_file_buffered_write(iocb, iovp, nr_segs,
Dave Chinner9aa05002012-10-08 21:56:04 +1100741 pos, &iocb->ki_pos, count, 0);
742
Dave Chinner637bbc72011-01-11 10:17:30 +1100743 /*
Dave Chinner9aa05002012-10-08 21:56:04 +1100744 * If we just got an ENOSPC, try to write back all dirty inodes to
745 * convert delalloc space to free up some of the excess reserved
746 * metadata space.
Dave Chinner637bbc72011-01-11 10:17:30 +1100747 */
748 if (ret == -ENOSPC && !enospc) {
Dave Chinner637bbc72011-01-11 10:17:30 +1100749 enospc = 1;
Dave Chinner9aa05002012-10-08 21:56:04 +1100750 xfs_flush_inodes(ip->i_mount);
751 goto write_retry;
Dave Chinner637bbc72011-01-11 10:17:30 +1100752 }
Christoph Hellwigd0606462011-12-18 20:00:14 +0000753
Dave Chinner637bbc72011-01-11 10:17:30 +1100754 current->backing_dev_info = NULL;
Christoph Hellwigd0606462011-12-18 20:00:14 +0000755out:
756 xfs_rw_iunlock(ip, iolock);
Dave Chinner637bbc72011-01-11 10:17:30 +1100757 return ret;
758}
759
760STATIC ssize_t
Christoph Hellwig00258e32010-02-15 09:44:47 +0000761xfs_file_aio_write(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000762 struct kiocb *iocb,
763 const struct iovec *iovp,
Christoph Hellwig00258e32010-02-15 09:44:47 +0000764 unsigned long nr_segs,
765 loff_t pos)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000766{
767 struct file *file = iocb->ki_filp;
768 struct address_space *mapping = file->f_mapping;
769 struct inode *inode = mapping->host;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000770 struct xfs_inode *ip = XFS_I(inode);
Dave Chinner637bbc72011-01-11 10:17:30 +1100771 ssize_t ret;
Dave Chinner637bbc72011-01-11 10:17:30 +1100772 size_t ocount = 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000773
774 XFS_STATS_INC(xs_write_calls);
775
Christoph Hellwig00258e32010-02-15 09:44:47 +0000776 BUG_ON(iocb->ki_pos != pos);
777
Dave Chinnera363f0c2011-01-11 10:13:53 +1100778 ret = generic_segment_checks(iovp, &nr_segs, &ocount, VERIFY_READ);
779 if (ret)
780 return ret;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000781
Dave Chinner637bbc72011-01-11 10:17:30 +1100782 if (ocount == 0)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000783 return 0;
784
Jan Karad9457dc2012-06-12 16:20:39 +0200785 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
786 ret = -EIO;
787 goto out;
788 }
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000789
Dave Chinner637bbc72011-01-11 10:17:30 +1100790 if (unlikely(file->f_flags & O_DIRECT))
Christoph Hellwigd0606462011-12-18 20:00:14 +0000791 ret = xfs_file_dio_aio_write(iocb, iovp, nr_segs, pos, ocount);
Dave Chinner637bbc72011-01-11 10:17:30 +1100792 else
793 ret = xfs_file_buffered_aio_write(iocb, iovp, nr_segs, pos,
Christoph Hellwigd0606462011-12-18 20:00:14 +0000794 ocount);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000795
Christoph Hellwigd0606462011-12-18 20:00:14 +0000796 if (ret > 0) {
797 ssize_t err;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000798
Christoph Hellwigd0606462011-12-18 20:00:14 +0000799 XFS_STATS_ADD(xs_write_bytes, ret);
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000800
Christoph Hellwigd0606462011-12-18 20:00:14 +0000801 /* Handle various SYNC-type writes */
Al Virod311d792014-02-09 15:18:09 -0500802 err = generic_write_sync(file, iocb->ki_pos - ret, ret);
Christoph Hellwigd0606462011-12-18 20:00:14 +0000803 if (err < 0)
804 ret = err;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000805 }
806
Jan Karad9457dc2012-06-12 16:20:39 +0200807out:
Dave Chinnera363f0c2011-01-11 10:13:53 +1100808 return ret;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000809}
810
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100811STATIC long
812xfs_file_fallocate(
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700813 struct file *file,
814 int mode,
815 loff_t offset,
816 loff_t len)
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100817{
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700818 struct inode *inode = file_inode(file);
819 struct xfs_inode *ip = XFS_I(inode);
820 struct xfs_trans *tp;
821 long error;
822 loff_t new_size = 0;
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100823
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700824 if (!S_ISREG(inode->i_mode))
825 return -EINVAL;
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100826 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
827 return -EOPNOTSUPP;
828
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100829 xfs_ilock(ip, XFS_IOLOCK_EXCL);
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700830 if (mode & FALLOC_FL_PUNCH_HOLE) {
831 error = xfs_free_file_space(ip, offset, len);
832 if (error)
833 goto out_unlock;
834 } else {
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 }
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100842
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700843 error = xfs_alloc_file_space(ip, offset, len,
844 XFS_BMAPI_PREALLOC);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100845 if (error)
846 goto out_unlock;
847 }
848
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700849 tp = xfs_trans_alloc(ip->i_mount, XFS_TRANS_WRITEID);
850 error = xfs_trans_reserve(tp, &M_RES(ip->i_mount)->tr_writeid, 0, 0);
851 if (error) {
852 xfs_trans_cancel(tp, 0);
853 goto out_unlock;
854 }
Dave Chinner82878892011-03-26 09:13:08 +1100855
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700856 xfs_ilock(ip, XFS_ILOCK_EXCL);
857 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
858 ip->i_d.di_mode &= ~S_ISUID;
859 if (ip->i_d.di_mode & S_IXGRP)
860 ip->i_d.di_mode &= ~S_ISGID;
861
862 if (!(mode & FALLOC_FL_PUNCH_HOLE))
863 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
864
865 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
866 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
867
868 if (file->f_flags & O_DSYNC)
869 xfs_trans_set_sync(tp);
870 error = xfs_trans_commit(tp, 0);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100871 if (error)
872 goto out_unlock;
873
874 /* Change file size if needed */
875 if (new_size) {
876 struct iattr iattr;
877
878 iattr.ia_valid = ATTR_SIZE;
879 iattr.ia_size = new_size;
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700880 error = xfs_setattr_size(ip, &iattr);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100881 }
882
883out_unlock:
884 xfs_iunlock(ip, XFS_IOLOCK_EXCL);
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700885 return -error;
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100886}
887
888
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +1100890xfs_file_open(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891 struct inode *inode,
Christoph Hellwigf999a5b2008-11-28 14:23:32 +1100892 struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700893{
Christoph Hellwigf999a5b2008-11-28 14:23:32 +1100894 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895 return -EFBIG;
Christoph Hellwigf999a5b2008-11-28 14:23:32 +1100896 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
897 return -EIO;
898 return 0;
899}
900
901STATIC int
902xfs_dir_open(
903 struct inode *inode,
904 struct file *file)
905{
906 struct xfs_inode *ip = XFS_I(inode);
907 int mode;
908 int error;
909
910 error = xfs_file_open(inode, file);
911 if (error)
912 return error;
913
914 /*
915 * If there are any blocks, read-ahead block 0 as we're almost
916 * certain to have the next operation be a read there.
917 */
Christoph Hellwig309ecac2013-12-06 12:30:09 -0800918 mode = xfs_ilock_data_map_shared(ip);
Christoph Hellwigf999a5b2008-11-28 14:23:32 +1100919 if (ip->i_d.di_nextents > 0)
Dave Chinner33363fe2013-04-03 16:11:22 +1100920 xfs_dir3_data_readahead(NULL, ip, 0, -1);
Christoph Hellwigf999a5b2008-11-28 14:23:32 +1100921 xfs_iunlock(ip, mode);
922 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923}
924
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +1100926xfs_file_release(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 struct inode *inode,
928 struct file *filp)
929{
Christoph Hellwig739bfb22007-08-29 10:58:01 +1000930 return -xfs_release(XFS_I(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700931}
932
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +1100934xfs_file_readdir(
Al Virob8227552013-05-22 17:07:56 -0400935 struct file *file,
936 struct dir_context *ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937{
Al Virob8227552013-05-22 17:07:56 -0400938 struct inode *inode = file_inode(file);
Christoph Hellwig739bfb22007-08-29 10:58:01 +1000939 xfs_inode_t *ip = XFS_I(inode);
Christoph Hellwig051e7cd2007-08-28 13:58:24 +1000940 int error;
941 size_t bufsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942
Christoph Hellwig051e7cd2007-08-28 13:58:24 +1000943 /*
944 * The Linux API doesn't pass down the total size of the buffer
945 * we read into down to the filesystem. With the filldir concept
946 * it's not needed for correct information, but the XFS dir2 leaf
947 * code wants an estimate of the buffer size to calculate it's
948 * readahead window and size the buffers used for mapping to
949 * physical blocks.
950 *
951 * Try to give it an estimate that's good enough, maybe at some
952 * point we can change the ->readdir prototype to include the
Eric Sandeena9cc7992010-02-03 17:50:13 +0000953 * buffer size. For now we use the current glibc buffer size.
Christoph Hellwig051e7cd2007-08-28 13:58:24 +1000954 */
Eric Sandeena9cc7992010-02-03 17:50:13 +0000955 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956
Al Virob8227552013-05-22 17:07:56 -0400957 error = xfs_readdir(ip, ctx, bufsize);
Christoph Hellwig051e7cd2007-08-28 13:58:24 +1000958 if (error)
959 return -error;
960 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961}
962
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +1100964xfs_file_mmap(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 struct file *filp,
966 struct vm_area_struct *vma)
967{
Nathan Scott3562fd42006-03-14 14:00:35 +1100968 vma->vm_ops = &xfs_file_vm_ops;
Dean Roehrich6fac0cb2005-06-21 14:07:45 +1000969
Nathan Scottfbc14622006-06-09 14:52:13 +1000970 file_accessed(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 return 0;
972}
973
David Chinner4f57dbc2007-07-19 16:28:17 +1000974/*
975 * mmap()d file has taken write protection fault and is being made
976 * writable. We can set the page state up correctly for a writable
977 * page, which means we can do correct delalloc accounting (ENOSPC
978 * checking!) and unwritten extent mapping.
979 */
980STATIC int
981xfs_vm_page_mkwrite(
982 struct vm_area_struct *vma,
Nick Pigginc2ec1752009-03-31 15:23:21 -0700983 struct vm_fault *vmf)
David Chinner4f57dbc2007-07-19 16:28:17 +1000984{
Nick Pigginc2ec1752009-03-31 15:23:21 -0700985 return block_page_mkwrite(vma, vmf, xfs_get_blocks);
David Chinner4f57dbc2007-07-19 16:28:17 +1000986}
987
Jeff Liud126d432012-08-21 17:11:57 +0800988/*
989 * This type is designed to indicate the type of offset we would like
990 * to search from page cache for either xfs_seek_data() or xfs_seek_hole().
991 */
992enum {
993 HOLE_OFF = 0,
994 DATA_OFF,
995};
996
997/*
998 * Lookup the desired type of offset from the given page.
999 *
1000 * On success, return true and the offset argument will point to the
1001 * start of the region that was found. Otherwise this function will
1002 * return false and keep the offset argument unchanged.
1003 */
1004STATIC bool
1005xfs_lookup_buffer_offset(
1006 struct page *page,
1007 loff_t *offset,
1008 unsigned int type)
1009{
1010 loff_t lastoff = page_offset(page);
1011 bool found = false;
1012 struct buffer_head *bh, *head;
1013
1014 bh = head = page_buffers(page);
1015 do {
1016 /*
1017 * Unwritten extents that have data in the page
1018 * cache covering them can be identified by the
1019 * BH_Unwritten state flag. Pages with multiple
1020 * buffers might have a mix of holes, data and
1021 * unwritten extents - any buffer with valid
1022 * data in it should have BH_Uptodate flag set
1023 * on it.
1024 */
1025 if (buffer_unwritten(bh) ||
1026 buffer_uptodate(bh)) {
1027 if (type == DATA_OFF)
1028 found = true;
1029 } else {
1030 if (type == HOLE_OFF)
1031 found = true;
1032 }
1033
1034 if (found) {
1035 *offset = lastoff;
1036 break;
1037 }
1038 lastoff += bh->b_size;
1039 } while ((bh = bh->b_this_page) != head);
1040
1041 return found;
1042}
1043
1044/*
1045 * This routine is called to find out and return a data or hole offset
1046 * from the page cache for unwritten extents according to the desired
1047 * type for xfs_seek_data() or xfs_seek_hole().
1048 *
1049 * The argument offset is used to tell where we start to search from the
1050 * page cache. Map is used to figure out the end points of the range to
1051 * lookup pages.
1052 *
1053 * Return true if the desired type of offset was found, and the argument
1054 * offset is filled with that address. Otherwise, return false and keep
1055 * offset unchanged.
1056 */
1057STATIC bool
1058xfs_find_get_desired_pgoff(
1059 struct inode *inode,
1060 struct xfs_bmbt_irec *map,
1061 unsigned int type,
1062 loff_t *offset)
1063{
1064 struct xfs_inode *ip = XFS_I(inode);
1065 struct xfs_mount *mp = ip->i_mount;
1066 struct pagevec pvec;
1067 pgoff_t index;
1068 pgoff_t end;
1069 loff_t endoff;
1070 loff_t startoff = *offset;
1071 loff_t lastoff = startoff;
1072 bool found = false;
1073
1074 pagevec_init(&pvec, 0);
1075
1076 index = startoff >> PAGE_CACHE_SHIFT;
1077 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
1078 end = endoff >> PAGE_CACHE_SHIFT;
1079 do {
1080 int want;
1081 unsigned nr_pages;
1082 unsigned int i;
1083
1084 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1085 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1086 want);
1087 /*
1088 * No page mapped into given range. If we are searching holes
1089 * and if this is the first time we got into the loop, it means
1090 * that the given offset is landed in a hole, return it.
1091 *
1092 * If we have already stepped through some block buffers to find
1093 * holes but they all contains data. In this case, the last
1094 * offset is already updated and pointed to the end of the last
1095 * mapped page, if it does not reach the endpoint to search,
1096 * that means there should be a hole between them.
1097 */
1098 if (nr_pages == 0) {
1099 /* Data search found nothing */
1100 if (type == DATA_OFF)
1101 break;
1102
1103 ASSERT(type == HOLE_OFF);
1104 if (lastoff == startoff || lastoff < endoff) {
1105 found = true;
1106 *offset = lastoff;
1107 }
1108 break;
1109 }
1110
1111 /*
1112 * At lease we found one page. If this is the first time we
1113 * step into the loop, and if the first page index offset is
1114 * greater than the given search offset, a hole was found.
1115 */
1116 if (type == HOLE_OFF && lastoff == startoff &&
1117 lastoff < page_offset(pvec.pages[0])) {
1118 found = true;
1119 break;
1120 }
1121
1122 for (i = 0; i < nr_pages; i++) {
1123 struct page *page = pvec.pages[i];
1124 loff_t b_offset;
1125
1126 /*
1127 * At this point, the page may be truncated or
1128 * invalidated (changing page->mapping to NULL),
1129 * or even swizzled back from swapper_space to tmpfs
1130 * file mapping. However, page->index will not change
1131 * because we have a reference on the page.
1132 *
1133 * Searching done if the page index is out of range.
1134 * If the current offset is not reaches the end of
1135 * the specified search range, there should be a hole
1136 * between them.
1137 */
1138 if (page->index > end) {
1139 if (type == HOLE_OFF && lastoff < endoff) {
1140 *offset = lastoff;
1141 found = true;
1142 }
1143 goto out;
1144 }
1145
1146 lock_page(page);
1147 /*
1148 * Page truncated or invalidated(page->mapping == NULL).
1149 * We can freely skip it and proceed to check the next
1150 * page.
1151 */
1152 if (unlikely(page->mapping != inode->i_mapping)) {
1153 unlock_page(page);
1154 continue;
1155 }
1156
1157 if (!page_has_buffers(page)) {
1158 unlock_page(page);
1159 continue;
1160 }
1161
1162 found = xfs_lookup_buffer_offset(page, &b_offset, type);
1163 if (found) {
1164 /*
1165 * The found offset may be less than the start
1166 * point to search if this is the first time to
1167 * come here.
1168 */
1169 *offset = max_t(loff_t, startoff, b_offset);
1170 unlock_page(page);
1171 goto out;
1172 }
1173
1174 /*
1175 * We either searching data but nothing was found, or
1176 * searching hole but found a data buffer. In either
1177 * case, probably the next page contains the desired
1178 * things, update the last offset to it so.
1179 */
1180 lastoff = page_offset(page) + PAGE_SIZE;
1181 unlock_page(page);
1182 }
1183
1184 /*
1185 * The number of returned pages less than our desired, search
1186 * done. In this case, nothing was found for searching data,
1187 * but we found a hole behind the last offset.
1188 */
1189 if (nr_pages < want) {
1190 if (type == HOLE_OFF) {
1191 *offset = lastoff;
1192 found = true;
1193 }
1194 break;
1195 }
1196
1197 index = pvec.pages[i - 1]->index + 1;
1198 pagevec_release(&pvec);
1199 } while (index <= end);
1200
1201out:
1202 pagevec_release(&pvec);
1203 return found;
1204}
1205
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001206STATIC loff_t
1207xfs_seek_data(
1208 struct file *file,
Jeff Liu834ab122012-08-21 17:11:45 +08001209 loff_t start)
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001210{
1211 struct inode *inode = file->f_mapping->host;
1212 struct xfs_inode *ip = XFS_I(inode);
1213 struct xfs_mount *mp = ip->i_mount;
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001214 loff_t uninitialized_var(offset);
1215 xfs_fsize_t isize;
1216 xfs_fileoff_t fsbno;
1217 xfs_filblks_t end;
1218 uint lock;
1219 int error;
1220
Christoph Hellwig309ecac2013-12-06 12:30:09 -08001221 lock = xfs_ilock_data_map_shared(ip);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001222
1223 isize = i_size_read(inode);
1224 if (start >= isize) {
1225 error = ENXIO;
1226 goto out_unlock;
1227 }
1228
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001229 /*
1230 * Try to read extents from the first block indicated
1231 * by fsbno to the end block of the file.
1232 */
Jeff Liu52f1acc2012-08-21 17:12:07 +08001233 fsbno = XFS_B_TO_FSBT(mp, start);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001234 end = XFS_B_TO_FSB(mp, isize);
Jeff Liu52f1acc2012-08-21 17:12:07 +08001235 for (;;) {
1236 struct xfs_bmbt_irec map[2];
1237 int nmap = 2;
1238 unsigned int i;
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001239
Jeff Liu52f1acc2012-08-21 17:12:07 +08001240 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1241 XFS_BMAPI_ENTIRE);
1242 if (error)
1243 goto out_unlock;
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001244
Jeff Liu52f1acc2012-08-21 17:12:07 +08001245 /* No extents at given offset, must be beyond EOF */
1246 if (nmap == 0) {
1247 error = ENXIO;
1248 goto out_unlock;
1249 }
1250
1251 for (i = 0; i < nmap; i++) {
1252 offset = max_t(loff_t, start,
1253 XFS_FSB_TO_B(mp, map[i].br_startoff));
1254
1255 /* Landed in a data extent */
1256 if (map[i].br_startblock == DELAYSTARTBLOCK ||
1257 (map[i].br_state == XFS_EXT_NORM &&
1258 !isnullstartblock(map[i].br_startblock)))
1259 goto out;
1260
1261 /*
1262 * Landed in an unwritten extent, try to search data
1263 * from page cache.
1264 */
1265 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1266 if (xfs_find_get_desired_pgoff(inode, &map[i],
1267 DATA_OFF, &offset))
1268 goto out;
1269 }
1270 }
1271
1272 /*
1273 * map[0] is hole or its an unwritten extent but
1274 * without data in page cache. Probably means that
1275 * we are reading after EOF if nothing in map[1].
1276 */
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001277 if (nmap == 1) {
1278 error = ENXIO;
1279 goto out_unlock;
1280 }
1281
Jeff Liu52f1acc2012-08-21 17:12:07 +08001282 ASSERT(i > 1);
1283
1284 /*
1285 * Nothing was found, proceed to the next round of search
1286 * if reading offset not beyond or hit EOF.
1287 */
1288 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1289 start = XFS_FSB_TO_B(mp, fsbno);
1290 if (start >= isize) {
1291 error = ENXIO;
1292 goto out_unlock;
1293 }
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001294 }
1295
Jeff Liu52f1acc2012-08-21 17:12:07 +08001296out:
Jie Liu46a1c2c2013-06-25 12:02:13 +08001297 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001298
1299out_unlock:
Christoph Hellwig01f4f322013-12-06 12:30:08 -08001300 xfs_iunlock(ip, lock);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001301
1302 if (error)
1303 return -error;
1304 return offset;
1305}
1306
1307STATIC loff_t
1308xfs_seek_hole(
1309 struct file *file,
Jeff Liu834ab122012-08-21 17:11:45 +08001310 loff_t start)
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001311{
1312 struct inode *inode = file->f_mapping->host;
1313 struct xfs_inode *ip = XFS_I(inode);
1314 struct xfs_mount *mp = ip->i_mount;
1315 loff_t uninitialized_var(offset);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001316 xfs_fsize_t isize;
1317 xfs_fileoff_t fsbno;
Jeff Liub686d1f2012-08-21 17:12:18 +08001318 xfs_filblks_t end;
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001319 uint lock;
1320 int error;
1321
1322 if (XFS_FORCED_SHUTDOWN(mp))
1323 return -XFS_ERROR(EIO);
1324
Christoph Hellwig309ecac2013-12-06 12:30:09 -08001325 lock = xfs_ilock_data_map_shared(ip);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001326
1327 isize = i_size_read(inode);
1328 if (start >= isize) {
1329 error = ENXIO;
1330 goto out_unlock;
1331 }
1332
1333 fsbno = XFS_B_TO_FSBT(mp, start);
Jeff Liub686d1f2012-08-21 17:12:18 +08001334 end = XFS_B_TO_FSB(mp, isize);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001335
Jeff Liub686d1f2012-08-21 17:12:18 +08001336 for (;;) {
1337 struct xfs_bmbt_irec map[2];
1338 int nmap = 2;
1339 unsigned int i;
1340
1341 error = xfs_bmapi_read(ip, fsbno, end - fsbno, map, &nmap,
1342 XFS_BMAPI_ENTIRE);
1343 if (error)
1344 goto out_unlock;
1345
1346 /* No extents at given offset, must be beyond EOF */
1347 if (nmap == 0) {
1348 error = ENXIO;
1349 goto out_unlock;
1350 }
1351
1352 for (i = 0; i < nmap; i++) {
1353 offset = max_t(loff_t, start,
1354 XFS_FSB_TO_B(mp, map[i].br_startoff));
1355
1356 /* Landed in a hole */
1357 if (map[i].br_startblock == HOLESTARTBLOCK)
1358 goto out;
1359
1360 /*
1361 * Landed in an unwritten extent, try to search hole
1362 * from page cache.
1363 */
1364 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1365 if (xfs_find_get_desired_pgoff(inode, &map[i],
1366 HOLE_OFF, &offset))
1367 goto out;
1368 }
1369 }
1370
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001371 /*
Jeff Liub686d1f2012-08-21 17:12:18 +08001372 * map[0] contains data or its unwritten but contains
1373 * data in page cache, probably means that we are
1374 * reading after EOF. We should fix offset to point
1375 * to the end of the file(i.e., there is an implicit
1376 * hole at the end of any file).
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001377 */
Jeff Liub686d1f2012-08-21 17:12:18 +08001378 if (nmap == 1) {
1379 offset = isize;
1380 break;
1381 }
1382
1383 ASSERT(i > 1);
1384
1385 /*
1386 * Both mappings contains data, proceed to the next round of
1387 * search if the current reading offset not beyond or hit EOF.
1388 */
1389 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1390 start = XFS_FSB_TO_B(mp, fsbno);
1391 if (start >= isize) {
1392 offset = isize;
1393 break;
1394 }
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001395 }
1396
Jeff Liub686d1f2012-08-21 17:12:18 +08001397out:
1398 /*
1399 * At this point, we must have found a hole. However, the returned
1400 * offset may be bigger than the file size as it may be aligned to
1401 * page boundary for unwritten extents, we need to deal with this
1402 * situation in particular.
1403 */
1404 offset = min_t(loff_t, offset, isize);
Jie Liu46a1c2c2013-06-25 12:02:13 +08001405 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001406
1407out_unlock:
Christoph Hellwig01f4f322013-12-06 12:30:08 -08001408 xfs_iunlock(ip, lock);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001409
1410 if (error)
1411 return -error;
1412 return offset;
1413}
1414
1415STATIC loff_t
1416xfs_file_llseek(
1417 struct file *file,
1418 loff_t offset,
1419 int origin)
1420{
1421 switch (origin) {
1422 case SEEK_END:
1423 case SEEK_CUR:
1424 case SEEK_SET:
1425 return generic_file_llseek(file, offset, origin);
1426 case SEEK_DATA:
Jeff Liu834ab122012-08-21 17:11:45 +08001427 return xfs_seek_data(file, offset);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001428 case SEEK_HOLE:
Jeff Liu834ab122012-08-21 17:11:45 +08001429 return xfs_seek_hole(file, offset);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001430 default:
1431 return -EINVAL;
1432 }
1433}
1434
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001435const struct file_operations xfs_file_operations = {
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001436 .llseek = xfs_file_llseek,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 .read = do_sync_read,
Dean Roehrichbb3f7242005-09-02 15:43:05 +10001438 .write = do_sync_write,
Nathan Scott3562fd42006-03-14 14:00:35 +11001439 .aio_read = xfs_file_aio_read,
1440 .aio_write = xfs_file_aio_write,
Nathan Scott1b895842006-03-31 13:08:59 +10001441 .splice_read = xfs_file_splice_read,
1442 .splice_write = xfs_file_splice_write,
Nathan Scott3562fd42006-03-14 14:00:35 +11001443 .unlocked_ioctl = xfs_file_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001444#ifdef CONFIG_COMPAT
Nathan Scott3562fd42006-03-14 14:00:35 +11001445 .compat_ioctl = xfs_file_compat_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446#endif
Nathan Scott3562fd42006-03-14 14:00:35 +11001447 .mmap = xfs_file_mmap,
1448 .open = xfs_file_open,
1449 .release = xfs_file_release,
1450 .fsync = xfs_file_fsync,
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001451 .fallocate = xfs_file_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452};
1453
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001454const struct file_operations xfs_dir_file_operations = {
Christoph Hellwigf999a5b2008-11-28 14:23:32 +11001455 .open = xfs_dir_open,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456 .read = generic_read_dir,
Al Virob8227552013-05-22 17:07:56 -04001457 .iterate = xfs_file_readdir,
Al Viro59af1582008-08-24 07:24:41 -04001458 .llseek = generic_file_llseek,
Nathan Scott3562fd42006-03-14 14:00:35 +11001459 .unlocked_ioctl = xfs_file_ioctl,
Nathan Scottd3870392005-05-06 06:44:46 -07001460#ifdef CONFIG_COMPAT
Nathan Scott3562fd42006-03-14 14:00:35 +11001461 .compat_ioctl = xfs_file_compat_ioctl,
Nathan Scottd3870392005-05-06 06:44:46 -07001462#endif
Christoph Hellwig1da2f2d2011-10-02 14:25:16 +00001463 .fsync = xfs_dir_fsync,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464};
1465
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04001466static const struct vm_operations_struct xfs_file_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07001467 .fault = filemap_fault,
David Chinner4f57dbc2007-07-19 16:28:17 +10001468 .page_mkwrite = xfs_vm_page_mkwrite,
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -07001469 .remap_pages = generic_file_remap_pages,
Dean Roehrich6fac0cb2005-06-21 14:07:45 +10001470};