blob: c68517b0f248d77f422bf69f43a21f97d736e8b3 [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 Chinner70a98832013-10-23 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_mount.h"
Dave Chinner57062782013-10-15 09:17:51 +110025#include "xfs_da_format.h"
26#include "xfs_da_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_inode.h"
Dave Chinner239880e2013-10-23 10:50:10 +110028#include "xfs_trans.h"
Christoph Hellwigfd3200b2010-02-15 09:44:48 +000029#include "xfs_inode_item.h"
Christoph Hellwigdda35b82010-02-15 09:44:46 +000030#include "xfs_bmap.h"
Dave Chinnerc24b5df2013-08-12 20:49:45 +100031#include "xfs_bmap_util.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include "xfs_error.h"
Dave Chinner2b9ab5a2013-08-12 20:49:37 +100033#include "xfs_dir2.h"
Dave Chinnerc24b5df2013-08-12 20:49:45 +100034#include "xfs_dir2_priv.h"
Christoph Hellwigddcd8562008-12-03 07:55:34 -050035#include "xfs_ioctl.h"
Christoph Hellwigdda35b82010-02-15 09:44:46 +000036#include "xfs_trace.h"
Dave Chinner239880e2013-10-23 10:50:10 +110037#include "xfs_log.h"
Brian Fosterdc06f3982014-07-24 19:49:28 +100038#include "xfs_icache.h"
Christoph Hellwig781355c2015-02-16 11:59:50 +110039#include "xfs_pnfs.h"
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +100040#include "xfs_iomap.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
42#include <linux/dcache.h>
Christoph Hellwig2fe17c12011-01-14 13:07:43 +010043#include <linux/falloc.h>
Jeff Liud126d432012-08-21 17:11:57 +080044#include <linux/pagevec.h>
Tejun Heo66114ca2015-05-22 17:13:32 -040045#include <linux/backing-dev.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)
Al Viro59551022016-01-22 15:40:57 -050059 inode_lock(VFS_I(ip));
Dave Chinner487f84f2011-01-12 11:37:10 +110060 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)
Al Viro59551022016-01-22 15:40:57 -050070 inode_unlock(VFS_I(ip));
Dave Chinner487f84f2011-01-12 11:37:10 +110071}
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)
Al Viro59551022016-01-22 15:40:57 -050080 inode_unlock(VFS_I(ip));
Dave Chinner487f84f2011-01-12 11:37:10 +110081}
82
83/*
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +100084 * Clear the specified ranges to zero through either the pagecache or DAX.
85 * Holes and unwritten extents will be left as-is as they already are zeroed.
Christoph Hellwigdda35b82010-02-15 09:44:46 +000086 */
Dave Chinneref9d8732012-11-29 15:26:33 +110087int
Christoph Hellwig7bb41db2016-06-21 09:56:26 +100088xfs_zero_range(
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +100089 struct xfs_inode *ip,
Christoph Hellwig7bb41db2016-06-21 09:56:26 +100090 xfs_off_t pos,
91 xfs_off_t count,
92 bool *did_zero)
Christoph Hellwigdda35b82010-02-15 09:44:46 +000093{
Christoph Hellwig459f0fb2016-06-21 09:55:18 +100094 return iomap_zero_range(VFS_I(ip), pos, count, NULL, &xfs_iomap_ops);
Christoph Hellwigdda35b82010-02-15 09:44:46 +000095}
96
Christoph Hellwig8add71c2015-02-02 09:53:56 +110097int
98xfs_update_prealloc_flags(
99 struct xfs_inode *ip,
100 enum xfs_prealloc_flags flags)
101{
102 struct xfs_trans *tp;
103 int error;
104
Christoph Hellwig253f4912016-04-06 09:19:55 +1000105 error = xfs_trans_alloc(ip->i_mount, &M_RES(ip->i_mount)->tr_writeid,
106 0, 0, 0, &tp);
107 if (error)
Christoph Hellwig8add71c2015-02-02 09:53:56 +1100108 return error;
Christoph Hellwig8add71c2015-02-02 09:53:56 +1100109
110 xfs_ilock(ip, XFS_ILOCK_EXCL);
111 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
112
113 if (!(flags & XFS_PREALLOC_INVISIBLE)) {
Dave Chinnerc19b3b052016-02-09 16:54:58 +1100114 VFS_I(ip)->i_mode &= ~S_ISUID;
115 if (VFS_I(ip)->i_mode & S_IXGRP)
116 VFS_I(ip)->i_mode &= ~S_ISGID;
Christoph Hellwig8add71c2015-02-02 09:53:56 +1100117 xfs_trans_ichgtime(tp, ip, XFS_ICHGTIME_MOD | XFS_ICHGTIME_CHG);
118 }
119
120 if (flags & XFS_PREALLOC_SET)
121 ip->i_d.di_flags |= XFS_DIFLAG_PREALLOC;
122 if (flags & XFS_PREALLOC_CLEAR)
123 ip->i_d.di_flags &= ~XFS_DIFLAG_PREALLOC;
124
125 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
126 if (flags & XFS_PREALLOC_SYNC)
127 xfs_trans_set_sync(tp);
Christoph Hellwig70393312015-06-04 13:48:08 +1000128 return xfs_trans_commit(tp);
Christoph Hellwig8add71c2015-02-02 09:53:56 +1100129}
130
Christoph Hellwig1da2f2d2011-10-02 14:25:16 +0000131/*
132 * Fsync operations on directories are much simpler than on regular files,
133 * as there is no file data to flush, and thus also no need for explicit
134 * cache flush operations, and there are no non-transaction metadata updates
135 * on directories either.
136 */
137STATIC int
138xfs_dir_fsync(
139 struct file *file,
140 loff_t start,
141 loff_t end,
142 int datasync)
143{
144 struct xfs_inode *ip = XFS_I(file->f_mapping->host);
145 struct xfs_mount *mp = ip->i_mount;
146 xfs_lsn_t lsn = 0;
147
148 trace_xfs_dir_fsync(ip);
149
150 xfs_ilock(ip, XFS_ILOCK_SHARED);
151 if (xfs_ipincount(ip))
152 lsn = ip->i_itemp->ili_last_lsn;
153 xfs_iunlock(ip, XFS_ILOCK_SHARED);
154
155 if (!lsn)
156 return 0;
Dave Chinner24513372014-06-25 14:58:08 +1000157 return _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, NULL);
Christoph Hellwig1da2f2d2011-10-02 14:25:16 +0000158}
159
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000160STATIC int
161xfs_file_fsync(
162 struct file *file,
Josef Bacik02c24a82011-07-16 20:44:56 -0400163 loff_t start,
164 loff_t end,
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000165 int datasync)
166{
Christoph Hellwig7ea80852010-05-26 17:53:25 +0200167 struct inode *inode = file->f_mapping->host;
168 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000169 struct xfs_mount *mp = ip->i_mount;
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000170 int error = 0;
171 int log_flushed = 0;
Christoph Hellwigb1037052011-09-19 14:55:51 +0000172 xfs_lsn_t lsn = 0;
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000173
Christoph Hellwigcca28fb2010-06-24 11:57:09 +1000174 trace_xfs_file_fsync(ip);
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000175
Josef Bacik02c24a82011-07-16 20:44:56 -0400176 error = filemap_write_and_wait_range(inode->i_mapping, start, end);
177 if (error)
178 return error;
179
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000180 if (XFS_FORCED_SHUTDOWN(mp))
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000181 return -EIO;
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000182
183 xfs_iflags_clear(ip, XFS_ITRUNCATED);
184
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000185 if (mp->m_flags & XFS_MOUNT_BARRIER) {
186 /*
187 * If we have an RT and/or log subvolume we need to make sure
188 * to flush the write cache the device used for file data
189 * first. This is to ensure newly written file data make
190 * it to disk before logging the new inode size in case of
191 * an extending write.
192 */
193 if (XFS_IS_REALTIME_INODE(ip))
194 xfs_blkdev_issue_flush(mp->m_rtdev_targp);
195 else if (mp->m_logdev_targp != mp->m_ddev_targp)
196 xfs_blkdev_issue_flush(mp->m_ddev_targp);
197 }
198
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000199 /*
Dave Chinnerfc0561c2015-11-03 13:14:59 +1100200 * All metadata updates are logged, which means that we just have to
201 * flush the log up to the latest LSN that touched the inode. If we have
202 * concurrent fsync/fdatasync() calls, we need them to all block on the
203 * log force before we clear the ili_fsync_fields field. This ensures
204 * that we don't get a racing sync operation that does not wait for the
205 * metadata to hit the journal before returning. If we race with
206 * clearing the ili_fsync_fields, then all that will happen is the log
207 * force will do nothing as the lsn will already be on disk. We can't
208 * race with setting ili_fsync_fields because that is done under
209 * XFS_ILOCK_EXCL, and that can't happen because we hold the lock shared
210 * until after the ili_fsync_fields is cleared.
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000211 */
212 xfs_ilock(ip, XFS_ILOCK_SHARED);
Christoph Hellwig8f639dd2012-02-29 09:53:55 +0000213 if (xfs_ipincount(ip)) {
214 if (!datasync ||
Dave Chinnerfc0561c2015-11-03 13:14:59 +1100215 (ip->i_itemp->ili_fsync_fields & ~XFS_ILOG_TIMESTAMP))
Christoph Hellwig8f639dd2012-02-29 09:53:55 +0000216 lsn = ip->i_itemp->ili_last_lsn;
217 }
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000218
Dave Chinnerfc0561c2015-11-03 13:14:59 +1100219 if (lsn) {
Christoph Hellwigb1037052011-09-19 14:55:51 +0000220 error = _xfs_log_force_lsn(mp, lsn, XFS_LOG_SYNC, &log_flushed);
Dave Chinnerfc0561c2015-11-03 13:14:59 +1100221 ip->i_itemp->ili_fsync_fields = 0;
222 }
223 xfs_iunlock(ip, XFS_ILOCK_SHARED);
Christoph Hellwigb1037052011-09-19 14:55:51 +0000224
Christoph Hellwiga27a2632011-06-16 12:02:23 +0000225 /*
226 * If we only have a single device, and the log force about was
227 * a no-op we might have to flush the data device cache here.
228 * This can only happen for fdatasync/O_DSYNC if we were overwriting
229 * an already allocated file and thus do not have any metadata to
230 * commit.
231 */
232 if ((mp->m_flags & XFS_MOUNT_BARRIER) &&
233 mp->m_logdev_targp == mp->m_ddev_targp &&
234 !XFS_IS_REALTIME_INODE(ip) &&
235 !log_flushed)
236 xfs_blkdev_issue_flush(mp->m_ddev_targp);
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000237
Dave Chinner24513372014-06-25 14:58:08 +1000238 return error;
Christoph Hellwigfd3200b2010-02-15 09:44:48 +0000239}
240
Christoph Hellwig00258e32010-02-15 09:44:47 +0000241STATIC ssize_t
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000242xfs_file_dio_aio_read(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000243 struct kiocb *iocb,
Al Virob4f5d2c2014-04-02 14:37:59 -0400244 struct iov_iter *to)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000245{
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000246 struct address_space *mapping = iocb->ki_filp->f_mapping;
247 struct inode *inode = mapping->host;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000248 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwigf1285ff2016-07-20 11:36:57 +1000249 loff_t isize = i_size_read(inode);
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000250 size_t count = iov_iter_count(to);
Christoph Hellwigf1285ff2016-07-20 11:36:57 +1000251 struct iov_iter data;
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000252 struct xfs_buftarg *target;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000253 ssize_t ret = 0;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000254
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000255 trace_xfs_file_direct_read(ip, count, iocb->ki_pos);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000256
Christoph Hellwigf1285ff2016-07-20 11:36:57 +1000257 if (!count)
258 return 0; /* skip atime */
Christoph Hellwig00258e32010-02-15 09:44:47 +0000259
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000260 if (XFS_IS_REALTIME_INODE(ip))
261 target = ip->i_mount->m_rtdev_targp;
262 else
263 target = ip->i_mount->m_ddev_targp;
264
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000265 /* DIO must be aligned to device logical sector size */
266 if ((iocb->ki_pos | count) & target->bt_logical_sectormask) {
267 if (iocb->ki_pos == isize)
268 return 0;
269 return -EINVAL;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000270 }
271
Christoph Hellwiga447d7c2016-10-03 09:47:34 +1100272 file_accessed(iocb->ki_filp);
273
Dave Chinner0c38a252011-08-25 07:17:01 +0000274 /*
Brian Foster3d751af2015-08-19 10:35:04 +1000275 * Locking is a bit tricky here. If we take an exclusive lock for direct
276 * IO, we effectively serialise all new concurrent read IO to this file
277 * and block it behind IO that is currently in progress because IO in
278 * progress holds the IO lock shared. We only need to hold the lock
279 * exclusive to blow away the page cache, so only take lock exclusively
280 * if the page cache needs invalidation. This allows the normal direct
281 * IO case of no page cache pages to proceeed concurrently without
282 * serialisation.
Dave Chinner0c38a252011-08-25 07:17:01 +0000283 */
284 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000285 if (mapping->nrpages) {
Dave Chinner0c38a252011-08-25 07:17:01 +0000286 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
Dave Chinner487f84f2011-01-12 11:37:10 +1100287 xfs_rw_ilock(ip, XFS_IOLOCK_EXCL);
288
Brian Foster3d751af2015-08-19 10:35:04 +1000289 /*
290 * The generic dio code only flushes the range of the particular
291 * I/O. Because we take an exclusive lock here, this whole
292 * sequence is considerably more expensive for us. This has a
293 * noticeable performance impact for any file with cached pages,
294 * even when outside of the range of the particular I/O.
295 *
296 * Hence, amortize the cost of the lock against a full file
297 * flush and reduce the chances of repeated iolock cycles going
298 * forward.
299 */
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000300 if (mapping->nrpages) {
301 ret = filemap_write_and_wait(mapping);
Dave Chinner487f84f2011-01-12 11:37:10 +1100302 if (ret) {
303 xfs_rw_iunlock(ip, XFS_IOLOCK_EXCL);
304 return ret;
305 }
Chris Mason85e584d2014-09-02 12:12:52 +1000306
307 /*
308 * Invalidate whole pages. This can return an error if
309 * we fail to invalidate a page, but this should never
310 * happen on XFS. Warn if it does fail.
311 */
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000312 ret = invalidate_inode_pages2(mapping);
Chris Mason85e584d2014-09-02 12:12:52 +1000313 WARN_ON_ONCE(ret);
314 ret = 0;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000315 }
Dave Chinner487f84f2011-01-12 11:37:10 +1100316 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
Dave Chinner0c38a252011-08-25 07:17:01 +0000317 }
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000318
Christoph Hellwigf1285ff2016-07-20 11:36:57 +1000319 data = *to;
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000320 ret = __blockdev_direct_IO(iocb, inode, target->bt_bdev, &data,
321 xfs_get_blocks_direct, NULL, NULL, 0);
322 if (ret > 0) {
323 iocb->ki_pos += ret;
324 iov_iter_advance(to, ret);
Christoph Hellwigfa8d9722016-07-20 11:38:01 +1000325 }
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000326 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000327
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000328 return ret;
329}
330
Arnd Bergmannf021bd02016-07-22 09:50:55 +1000331static noinline ssize_t
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000332xfs_file_dax_read(
333 struct kiocb *iocb,
334 struct iov_iter *to)
335{
Christoph Hellwig6c31f492016-09-19 11:28:38 +1000336 struct xfs_inode *ip = XFS_I(iocb->ki_filp->f_mapping->host);
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000337 size_t count = iov_iter_count(to);
338 ssize_t ret = 0;
339
340 trace_xfs_file_dax_read(ip, count, iocb->ki_pos);
341
342 if (!count)
343 return 0; /* skip atime */
344
345 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
Christoph Hellwig6c31f492016-09-19 11:28:38 +1000346 ret = iomap_dax_rw(iocb, to, &xfs_iomap_ops);
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000347 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
348
Christoph Hellwigf1285ff2016-07-20 11:36:57 +1000349 file_accessed(iocb->ki_filp);
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000350 return ret;
351}
352
353STATIC ssize_t
354xfs_file_buffered_aio_read(
355 struct kiocb *iocb,
356 struct iov_iter *to)
357{
358 struct xfs_inode *ip = XFS_I(file_inode(iocb->ki_filp));
359 ssize_t ret;
360
361 trace_xfs_file_buffered_read(ip, iov_iter_count(to), iocb->ki_pos);
362
363 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
Al Virob4f5d2c2014-04-02 14:37:59 -0400364 ret = generic_file_read_iter(iocb, to);
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000365 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
366
367 return ret;
368}
369
370STATIC ssize_t
371xfs_file_read_iter(
372 struct kiocb *iocb,
373 struct iov_iter *to)
374{
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000375 struct inode *inode = file_inode(iocb->ki_filp);
376 struct xfs_mount *mp = XFS_I(inode)->i_mount;
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000377 ssize_t ret = 0;
378
379 XFS_STATS_INC(mp, xs_read_calls);
380
381 if (XFS_FORCED_SHUTDOWN(mp))
382 return -EIO;
383
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000384 if (IS_DAX(inode))
385 ret = xfs_file_dax_read(iocb, to);
386 else if (iocb->ki_flags & IOCB_DIRECT)
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000387 ret = xfs_file_dio_aio_read(iocb, to);
388 else
389 ret = xfs_file_buffered_aio_read(iocb, to);
390
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000391 if (ret > 0)
Bill O'Donnellff6d6af2015-10-12 18:21:22 +1100392 XFS_STATS_ADD(mp, xs_read_bytes, ret);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000393 return ret;
394}
395
Christoph Hellwig00258e32010-02-15 09:44:47 +0000396STATIC ssize_t
397xfs_file_splice_read(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000398 struct file *infilp,
399 loff_t *ppos,
400 struct pipe_inode_info *pipe,
401 size_t count,
Christoph Hellwig00258e32010-02-15 09:44:47 +0000402 unsigned int flags)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000403{
Christoph Hellwig00258e32010-02-15 09:44:47 +0000404 struct xfs_inode *ip = XFS_I(infilp->f_mapping->host);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000405 ssize_t ret;
406
Bill O'Donnellff6d6af2015-10-12 18:21:22 +1100407 XFS_STATS_INC(ip->i_mount, xs_read_calls);
Christoph Hellwig00258e32010-02-15 09:44:47 +0000408
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000409 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
410 return -EIO;
411
Christoph Hellwig3176c3e2016-07-20 11:31:42 +1000412 trace_xfs_file_splice_read(ip, count, *ppos);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000413
Dave Chinnera6d76362016-01-04 16:28:25 +1100414 /*
415 * DAX inodes cannot ues the page cache for splice, so we have to push
416 * them through the VFS IO path. This means it goes through
417 * ->read_iter, which for us takes the XFS_IOLOCK_SHARED. Hence we
418 * cannot lock the splice operation at this level for DAX inodes.
419 */
420 if (IS_DAX(VFS_I(ip))) {
421 ret = default_file_splice_read(infilp, ppos, pipe, count,
422 flags);
423 goto out;
424 }
425
426 xfs_rw_ilock(ip, XFS_IOLOCK_SHARED);
427 ret = generic_file_splice_read(infilp, ppos, pipe, count, flags);
428 xfs_rw_iunlock(ip, XFS_IOLOCK_SHARED);
429out:
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000430 if (ret > 0)
Bill O'Donnellff6d6af2015-10-12 18:21:22 +1100431 XFS_STATS_ADD(ip->i_mount, xs_read_bytes, ret);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000432 return ret;
433}
434
Dave Chinner4c5cfd12011-01-11 10:14:16 +1100435/*
Christoph Hellwig193aec12012-03-27 10:34:49 -0400436 * Zero any on disk space between the current EOF and the new, larger EOF.
437 *
438 * This handles the normal case of zeroing the remainder of the last block in
439 * the file and the unusual case of zeroing blocks out beyond the size of the
440 * file. This second case only happens with fixed size extents and when the
441 * system crashes before the inode size was updated but after blocks were
442 * allocated.
443 *
444 * Expects the iolock to be held exclusive, and will take the ilock internally.
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000445 */
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000446int /* error (positive) */
447xfs_zero_eof(
Christoph Hellwig193aec12012-03-27 10:34:49 -0400448 struct xfs_inode *ip,
449 xfs_off_t offset, /* starting I/O offset */
Dave Chinner5885ebd2015-02-23 22:37:08 +1100450 xfs_fsize_t isize, /* current inode size */
451 bool *did_zeroing)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000452{
Christoph Hellwig193aec12012-03-27 10:34:49 -0400453 ASSERT(xfs_isilocked(ip, XFS_IOLOCK_EXCL));
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000454 ASSERT(offset > isize);
455
Brian Foster0a50f162015-10-12 16:02:08 +1100456 trace_xfs_zero_eof(ip, isize, offset - isize);
Christoph Hellwig570b6212016-06-21 09:57:26 +1000457 return xfs_zero_range(ip, isize, offset - isize, did_zeroing);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000458}
459
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100460/*
Dave Chinner4d8d1582011-01-11 10:23:42 +1100461 * Common pre-write limit and setup checks.
462 *
Christoph Hellwig5bf1f262011-12-18 20:00:13 +0000463 * Called with the iolocked held either shared and exclusive according to
464 * @iolock, and returns with it held. Might upgrade the iolock to exclusive
465 * if called for a direct write beyond i_size.
Dave Chinner4d8d1582011-01-11 10:23:42 +1100466 */
467STATIC ssize_t
468xfs_file_aio_write_checks(
Al Viro99733fa2015-04-07 14:25:18 -0400469 struct kiocb *iocb,
470 struct iov_iter *from,
Dave Chinner4d8d1582011-01-11 10:23:42 +1100471 int *iolock)
472{
Al Viro99733fa2015-04-07 14:25:18 -0400473 struct file *file = iocb->ki_filp;
Dave Chinner4d8d1582011-01-11 10:23:42 +1100474 struct inode *inode = file->f_mapping->host;
475 struct xfs_inode *ip = XFS_I(inode);
Al Viro3309dd02015-04-09 12:55:47 -0400476 ssize_t error = 0;
Al Viro99733fa2015-04-07 14:25:18 -0400477 size_t count = iov_iter_count(from);
Brian Foster3136e8b2015-10-12 16:02:05 +1100478 bool drained_dio = false;
Dave Chinner4d8d1582011-01-11 10:23:42 +1100479
Dave Chinner7271d242011-08-25 07:17:02 +0000480restart:
Al Viro3309dd02015-04-09 12:55:47 -0400481 error = generic_write_checks(iocb, from);
482 if (error <= 0)
Dave Chinner4d8d1582011-01-11 10:23:42 +1100483 return error;
Dave Chinner4d8d1582011-01-11 10:23:42 +1100484
Christoph Hellwig21c3ea12015-04-13 11:38:29 +1000485 error = xfs_break_layouts(inode, iolock, true);
Christoph Hellwig781355c2015-02-16 11:59:50 +1100486 if (error)
487 return error;
488
Jan Karaa6de82c2015-05-21 16:05:56 +0200489 /* For changing security info in file_remove_privs() we need i_mutex */
490 if (*iolock == XFS_IOLOCK_SHARED && !IS_NOSEC(inode)) {
491 xfs_rw_iunlock(ip, *iolock);
492 *iolock = XFS_IOLOCK_EXCL;
493 xfs_rw_ilock(ip, *iolock);
494 goto restart;
495 }
Dave Chinner4d8d1582011-01-11 10:23:42 +1100496 /*
497 * If the offset is beyond the size of the file, we need to zero any
498 * blocks that fall between the existing EOF and the start of this
Christoph Hellwig2813d682011-12-18 20:00:12 +0000499 * write. If zeroing is needed and we are currently holding the
Christoph Hellwig467f7892012-03-27 10:34:47 -0400500 * iolock shared, we need to update it to exclusive which implies
501 * having to redo all checks before.
Dave Chinnerb9d59842015-04-16 22:03:07 +1000502 *
503 * We need to serialise against EOF updates that occur in IO
504 * completions here. We want to make sure that nobody is changing the
505 * size while we do this check until we have placed an IO barrier (i.e.
506 * hold the XFS_IOLOCK_EXCL) that prevents new IO from being dispatched.
507 * The spinlock effectively forms a memory barrier once we have the
508 * XFS_IOLOCK_EXCL so we are guaranteed to see the latest EOF value
509 * and hence be able to correctly determine if we need to run zeroing.
Dave Chinner4d8d1582011-01-11 10:23:42 +1100510 */
Dave Chinnerb9d59842015-04-16 22:03:07 +1000511 spin_lock(&ip->i_flags_lock);
Al Viro99733fa2015-04-07 14:25:18 -0400512 if (iocb->ki_pos > i_size_read(inode)) {
Dave Chinner5885ebd2015-02-23 22:37:08 +1100513 bool zero = false;
514
Dave Chinnerb9d59842015-04-16 22:03:07 +1000515 spin_unlock(&ip->i_flags_lock);
Brian Foster3136e8b2015-10-12 16:02:05 +1100516 if (!drained_dio) {
517 if (*iolock == XFS_IOLOCK_SHARED) {
518 xfs_rw_iunlock(ip, *iolock);
519 *iolock = XFS_IOLOCK_EXCL;
520 xfs_rw_ilock(ip, *iolock);
521 iov_iter_reexpand(from, count);
522 }
Dave Chinner40c63fb2015-04-16 22:03:17 +1000523 /*
524 * We now have an IO submission barrier in place, but
525 * AIO can do EOF updates during IO completion and hence
526 * we now need to wait for all of them to drain. Non-AIO
527 * DIO will have drained before we are given the
528 * XFS_IOLOCK_EXCL, and so for most cases this wait is a
529 * no-op.
530 */
531 inode_dio_wait(inode);
Brian Foster3136e8b2015-10-12 16:02:05 +1100532 drained_dio = true;
Dave Chinner7271d242011-08-25 07:17:02 +0000533 goto restart;
534 }
Al Viro99733fa2015-04-07 14:25:18 -0400535 error = xfs_zero_eof(ip, iocb->ki_pos, i_size_read(inode), &zero);
Christoph Hellwig467f7892012-03-27 10:34:47 -0400536 if (error)
537 return error;
Dave Chinnerb9d59842015-04-16 22:03:07 +1000538 } else
539 spin_unlock(&ip->i_flags_lock);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100540
541 /*
Christoph Hellwig8a9c9982012-02-29 09:53:52 +0000542 * Updating the timestamps will grab the ilock again from
543 * xfs_fs_dirty_inode, so we have to call it after dropping the
544 * lock above. Eventually we should look into a way to avoid
545 * the pointless lock roundtrip.
546 */
Josef Bacikc3b2da32012-03-26 09:59:21 -0400547 if (likely(!(file->f_mode & FMODE_NOCMTIME))) {
548 error = file_update_time(file);
549 if (error)
550 return error;
551 }
Christoph Hellwig8a9c9982012-02-29 09:53:52 +0000552
553 /*
Dave Chinner4d8d1582011-01-11 10:23:42 +1100554 * If we're writing the file then make sure to clear the setuid and
555 * setgid bits if the process is not being run by root. This keeps
556 * people from modifying setuid and setgid binaries.
557 */
Jan Karaa6de82c2015-05-21 16:05:56 +0200558 if (!IS_NOSEC(inode))
559 return file_remove_privs(file);
560 return 0;
Dave Chinner4d8d1582011-01-11 10:23:42 +1100561}
562
563/*
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100564 * xfs_file_dio_aio_write - handle direct IO writes
565 *
566 * Lock the inode appropriately to prepare for and issue a direct IO write.
Dave Chinnereda77982011-01-11 10:22:40 +1100567 * By separating it from the buffered write path we remove all the tricky to
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100568 * follow locking changes and looping.
569 *
Dave Chinnereda77982011-01-11 10:22:40 +1100570 * If there are cached pages or we're extending the file, we need IOLOCK_EXCL
571 * until we're sure the bytes at the new EOF have been zeroed and/or the cached
572 * pages are flushed out.
573 *
574 * In most cases the direct IO writes will be done holding IOLOCK_SHARED
575 * allowing them to be done in parallel with reads and other direct IO writes.
576 * However, if the IO is not aligned to filesystem blocks, the direct IO layer
577 * needs to do sub-block zeroing and that requires serialisation against other
578 * direct IOs to the same block. In this case we need to serialise the
579 * submission of the unaligned IOs so that we don't get racing block zeroing in
580 * the dio layer. To avoid the problem with aio, we also need to wait for
581 * outstanding IOs to complete so that unwritten extent conversion is completed
582 * before we try to map the overlapping block. This is currently implemented by
Christoph Hellwig4a06fd22011-08-23 08:28:13 +0000583 * hitting it with a big hammer (i.e. inode_dio_wait()).
Dave Chinnereda77982011-01-11 10:22:40 +1100584 *
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100585 * Returns with locks held indicated by @iolock and errors indicated by
586 * negative return values.
587 */
588STATIC ssize_t
589xfs_file_dio_aio_write(
590 struct kiocb *iocb,
Al Virob3188912014-04-02 07:06:30 -0400591 struct iov_iter *from)
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100592{
593 struct file *file = iocb->ki_filp;
594 struct address_space *mapping = file->f_mapping;
595 struct inode *inode = mapping->host;
596 struct xfs_inode *ip = XFS_I(inode);
597 struct xfs_mount *mp = ip->i_mount;
598 ssize_t ret = 0;
Dave Chinnereda77982011-01-11 10:22:40 +1100599 int unaligned_io = 0;
Christoph Hellwigd0606462011-12-18 20:00:14 +0000600 int iolock;
Al Virob3188912014-04-02 07:06:30 -0400601 size_t count = iov_iter_count(from);
Dave Chinner0cefb292015-04-16 22:03:27 +1000602 loff_t end;
603 struct iov_iter data;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100604 struct xfs_buftarg *target = XFS_IS_REALTIME_INODE(ip) ?
605 mp->m_rtdev_targp : mp->m_ddev_targp;
606
Eric Sandeen7c71ee72014-01-21 16:46:23 -0600607 /* DIO must be aligned to device logical sector size */
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000608 if ((iocb->ki_pos | count) & target->bt_logical_sectormask)
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000609 return -EINVAL;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100610
Eric Sandeen7c71ee72014-01-21 16:46:23 -0600611 /* "unaligned" here means not aligned to a filesystem block */
Christoph Hellwig13712712016-04-07 08:51:57 -0700612 if ((iocb->ki_pos & mp->m_blockmask) ||
613 ((iocb->ki_pos + count) & mp->m_blockmask))
Dave Chinnereda77982011-01-11 10:22:40 +1100614 unaligned_io = 1;
615
Dave Chinner7271d242011-08-25 07:17:02 +0000616 /*
617 * We don't need to take an exclusive lock unless there page cache needs
618 * to be invalidated or unaligned IO is being executed. We don't need to
619 * consider the EOF extension case here because
620 * xfs_file_aio_write_checks() will relock the inode as necessary for
621 * EOF zeroing cases and fill out the new inode size as appropriate.
622 */
623 if (unaligned_io || mapping->nrpages)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000624 iolock = XFS_IOLOCK_EXCL;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100625 else
Christoph Hellwigd0606462011-12-18 20:00:14 +0000626 iolock = XFS_IOLOCK_SHARED;
627 xfs_rw_ilock(ip, iolock);
Christoph Hellwigc58cb162011-08-27 14:42:53 +0000628
629 /*
630 * Recheck if there are cached pages that need invalidate after we got
631 * the iolock to protect against other threads adding new pages while
632 * we were waiting for the iolock.
633 */
Christoph Hellwigd0606462011-12-18 20:00:14 +0000634 if (mapping->nrpages && iolock == XFS_IOLOCK_SHARED) {
635 xfs_rw_iunlock(ip, iolock);
636 iolock = XFS_IOLOCK_EXCL;
637 xfs_rw_ilock(ip, iolock);
Christoph Hellwigc58cb162011-08-27 14:42:53 +0000638 }
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100639
Al Viro99733fa2015-04-07 14:25:18 -0400640 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100641 if (ret)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000642 goto out;
Al Viro99733fa2015-04-07 14:25:18 -0400643 count = iov_iter_count(from);
Christoph Hellwig13712712016-04-07 08:51:57 -0700644 end = iocb->ki_pos + count - 1;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100645
Brian Foster3d751af2015-08-19 10:35:04 +1000646 /*
Christoph Hellwigbbc5a742016-07-20 11:35:42 +1000647 * See xfs_file_dio_aio_read() for why we do a full-file flush here.
Brian Foster3d751af2015-08-19 10:35:04 +1000648 */
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100649 if (mapping->nrpages) {
Brian Foster3d751af2015-08-19 10:35:04 +1000650 ret = filemap_write_and_wait(VFS_I(ip)->i_mapping);
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100651 if (ret)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000652 goto out;
Dave Chinner834ffca2014-09-02 12:12:52 +1000653 /*
Brian Foster3d751af2015-08-19 10:35:04 +1000654 * Invalidate whole pages. This can return an error if we fail
655 * to invalidate a page, but this should never happen on XFS.
656 * Warn if it does fail.
Dave Chinner834ffca2014-09-02 12:12:52 +1000657 */
Brian Foster3d751af2015-08-19 10:35:04 +1000658 ret = invalidate_inode_pages2(VFS_I(ip)->i_mapping);
Dave Chinner834ffca2014-09-02 12:12:52 +1000659 WARN_ON_ONCE(ret);
660 ret = 0;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100661 }
662
Dave Chinnereda77982011-01-11 10:22:40 +1100663 /*
664 * If we are doing unaligned IO, wait for all other IO to drain,
665 * otherwise demote the lock if we had to flush cached pages
666 */
667 if (unaligned_io)
Christoph Hellwig4a06fd22011-08-23 08:28:13 +0000668 inode_dio_wait(inode);
Christoph Hellwigd0606462011-12-18 20:00:14 +0000669 else if (iolock == XFS_IOLOCK_EXCL) {
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100670 xfs_rw_ilock_demote(ip, XFS_IOLOCK_EXCL);
Christoph Hellwigd0606462011-12-18 20:00:14 +0000671 iolock = XFS_IOLOCK_SHARED;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100672 }
673
Christoph Hellwig3176c3e2016-07-20 11:31:42 +1000674 trace_xfs_file_direct_write(ip, count, iocb->ki_pos);
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100675
Dave Chinner0cefb292015-04-16 22:03:27 +1000676 data = *from;
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000677 ret = __blockdev_direct_IO(iocb, inode, target->bt_bdev, &data,
678 xfs_get_blocks_direct, xfs_end_io_direct_write,
679 NULL, DIO_ASYNC_EXTEND);
Dave Chinner0cefb292015-04-16 22:03:27 +1000680
681 /* see generic_file_direct_write() for why this is necessary */
682 if (mapping->nrpages) {
683 invalidate_inode_pages2_range(mapping,
Christoph Hellwig13712712016-04-07 08:51:57 -0700684 iocb->ki_pos >> PAGE_SHIFT,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300685 end >> PAGE_SHIFT);
Dave Chinner0cefb292015-04-16 22:03:27 +1000686 }
687
688 if (ret > 0) {
Christoph Hellwig13712712016-04-07 08:51:57 -0700689 iocb->ki_pos += ret;
Dave Chinner0cefb292015-04-16 22:03:27 +1000690 iov_iter_advance(from, ret);
Dave Chinner0cefb292015-04-16 22:03:27 +1000691 }
Christoph Hellwigd0606462011-12-18 20:00:14 +0000692out:
693 xfs_rw_iunlock(ip, iolock);
694
Dave Chinner6b698ed2015-06-04 09:18:53 +1000695 /*
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000696 * No fallback to buffered IO on errors for XFS, direct IO will either
697 * complete fully or fail.
Dave Chinner6b698ed2015-06-04 09:18:53 +1000698 */
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000699 ASSERT(ret < 0 || ret == count);
700 return ret;
701}
702
Arnd Bergmannf021bd02016-07-22 09:50:55 +1000703static noinline ssize_t
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000704xfs_file_dax_write(
705 struct kiocb *iocb,
706 struct iov_iter *from)
707{
Christoph Hellwig6c31f492016-09-19 11:28:38 +1000708 struct inode *inode = iocb->ki_filp->f_mapping->host;
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000709 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwig17879e82016-09-19 11:24:50 +1000710 int iolock = XFS_IOLOCK_EXCL;
Christoph Hellwig6c31f492016-09-19 11:28:38 +1000711 ssize_t ret, error = 0;
712 size_t count;
713 loff_t pos;
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000714
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000715 xfs_rw_ilock(ip, iolock);
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000716 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
717 if (ret)
718 goto out;
719
Christoph Hellwig6c31f492016-09-19 11:28:38 +1000720 pos = iocb->ki_pos;
721 count = iov_iter_count(from);
Dave Chinner8b2180b2016-08-17 08:31:33 +1000722
Christoph Hellwig6c31f492016-09-19 11:28:38 +1000723 trace_xfs_file_dax_write(ip, count, pos);
724
725 ret = iomap_dax_rw(iocb, from, &xfs_iomap_ops);
726 if (ret > 0 && iocb->ki_pos > i_size_read(inode)) {
727 i_size_write(inode, iocb->ki_pos);
728 error = xfs_setfilesize(ip, pos, ret);
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000729 }
730
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000731out:
732 xfs_rw_iunlock(ip, iolock);
Christoph Hellwig6c31f492016-09-19 11:28:38 +1000733 return error ? error : ret;
Dave Chinnerf0d26e82011-01-11 10:15:36 +1100734}
735
Christoph Hellwig00258e32010-02-15 09:44:47 +0000736STATIC ssize_t
Dave Chinner637bbc72011-01-11 10:17:30 +1100737xfs_file_buffered_aio_write(
738 struct kiocb *iocb,
Al Virob3188912014-04-02 07:06:30 -0400739 struct iov_iter *from)
Dave Chinner637bbc72011-01-11 10:17:30 +1100740{
741 struct file *file = iocb->ki_filp;
742 struct address_space *mapping = file->f_mapping;
743 struct inode *inode = mapping->host;
744 struct xfs_inode *ip = XFS_I(inode);
745 ssize_t ret;
746 int enospc = 0;
Christoph Hellwigd0606462011-12-18 20:00:14 +0000747 int iolock = XFS_IOLOCK_EXCL;
Dave Chinner637bbc72011-01-11 10:17:30 +1100748
Christoph Hellwigd0606462011-12-18 20:00:14 +0000749 xfs_rw_ilock(ip, iolock);
Dave Chinner637bbc72011-01-11 10:17:30 +1100750
Al Viro99733fa2015-04-07 14:25:18 -0400751 ret = xfs_file_aio_write_checks(iocb, from, &iolock);
Dave Chinner4d8d1582011-01-11 10:23:42 +1100752 if (ret)
Christoph Hellwigd0606462011-12-18 20:00:14 +0000753 goto out;
Dave Chinner637bbc72011-01-11 10:17:30 +1100754
755 /* We can write back this queue in page reclaim */
Christoph Hellwigde1414a2015-01-14 10:42:36 +0100756 current->backing_dev_info = inode_to_bdi(inode);
Dave Chinner637bbc72011-01-11 10:17:30 +1100757
758write_retry:
Christoph Hellwig3176c3e2016-07-20 11:31:42 +1000759 trace_xfs_file_buffered_write(ip, iov_iter_count(from), iocb->ki_pos);
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +1000760 ret = iomap_file_buffered_write(iocb, from, &xfs_iomap_ops);
Al Viro0a64bc22014-02-11 22:25:22 -0500761 if (likely(ret >= 0))
Al Viro99733fa2015-04-07 14:25:18 -0400762 iocb->ki_pos += ret;
Brian Fosterdc06f3982014-07-24 19:49:28 +1000763
Dave Chinner637bbc72011-01-11 10:17:30 +1100764 /*
Brian Fosterdc06f3982014-07-24 19:49:28 +1000765 * If we hit a space limit, try to free up some lingering preallocated
766 * space before returning an error. In the case of ENOSPC, first try to
767 * write back all dirty inodes to free up some of the excess reserved
768 * metadata space. This reduces the chances that the eofblocks scan
769 * waits on dirty mappings. Since xfs_flush_inodes() is serialized, this
770 * also behaves as a filter to prevent too many eofblocks scans from
771 * running at the same time.
Dave Chinner637bbc72011-01-11 10:17:30 +1100772 */
Brian Fosterdc06f3982014-07-24 19:49:28 +1000773 if (ret == -EDQUOT && !enospc) {
774 enospc = xfs_inode_free_quota_eofblocks(ip);
775 if (enospc)
776 goto write_retry;
777 } else if (ret == -ENOSPC && !enospc) {
778 struct xfs_eofblocks eofb = {0};
779
Dave Chinner637bbc72011-01-11 10:17:30 +1100780 enospc = 1;
Dave Chinner9aa05002012-10-08 21:56:04 +1100781 xfs_flush_inodes(ip->i_mount);
Brian Fosterdc06f3982014-07-24 19:49:28 +1000782 eofb.eof_scan_owner = ip->i_ino; /* for locking */
783 eofb.eof_flags = XFS_EOF_FLAGS_SYNC;
784 xfs_icache_free_eofblocks(ip->i_mount, &eofb);
Dave Chinner9aa05002012-10-08 21:56:04 +1100785 goto write_retry;
Dave Chinner637bbc72011-01-11 10:17:30 +1100786 }
Christoph Hellwigd0606462011-12-18 20:00:14 +0000787
Dave Chinner637bbc72011-01-11 10:17:30 +1100788 current->backing_dev_info = NULL;
Christoph Hellwigd0606462011-12-18 20:00:14 +0000789out:
790 xfs_rw_iunlock(ip, iolock);
Dave Chinner637bbc72011-01-11 10:17:30 +1100791 return ret;
792}
793
794STATIC ssize_t
Al Virobf97f3bc2014-04-03 14:20:23 -0400795xfs_file_write_iter(
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000796 struct kiocb *iocb,
Al Virobf97f3bc2014-04-03 14:20:23 -0400797 struct iov_iter *from)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000798{
799 struct file *file = iocb->ki_filp;
800 struct address_space *mapping = file->f_mapping;
801 struct inode *inode = mapping->host;
Christoph Hellwig00258e32010-02-15 09:44:47 +0000802 struct xfs_inode *ip = XFS_I(inode);
Dave Chinner637bbc72011-01-11 10:17:30 +1100803 ssize_t ret;
Al Virobf97f3bc2014-04-03 14:20:23 -0400804 size_t ocount = iov_iter_count(from);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000805
Bill O'Donnellff6d6af2015-10-12 18:21:22 +1100806 XFS_STATS_INC(ip->i_mount, xs_write_calls);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000807
Dave Chinner637bbc72011-01-11 10:17:30 +1100808 if (ocount == 0)
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000809 return 0;
810
Al Virobf97f3bc2014-04-03 14:20:23 -0400811 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
812 return -EIO;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000813
Christoph Hellwig16d4d432016-07-20 11:38:55 +1000814 if (IS_DAX(inode))
815 ret = xfs_file_dax_write(iocb, from);
816 else if (iocb->ki_flags & IOCB_DIRECT)
Al Virobf97f3bc2014-04-03 14:20:23 -0400817 ret = xfs_file_dio_aio_write(iocb, from);
Dave Chinner637bbc72011-01-11 10:17:30 +1100818 else
Al Virobf97f3bc2014-04-03 14:20:23 -0400819 ret = xfs_file_buffered_aio_write(iocb, from);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000820
Christoph Hellwigd0606462011-12-18 20:00:14 +0000821 if (ret > 0) {
Bill O'Donnellff6d6af2015-10-12 18:21:22 +1100822 XFS_STATS_ADD(ip->i_mount, xs_write_bytes, ret);
Christoph Hellwigce7ae1512011-12-18 20:00:11 +0000823
Christoph Hellwigd0606462011-12-18 20:00:14 +0000824 /* Handle various SYNC-type writes */
Christoph Hellwige2592212016-04-07 08:52:01 -0700825 ret = generic_write_sync(iocb, ret);
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000826 }
Dave Chinnera363f0c2011-01-11 10:13:53 +1100827 return ret;
Christoph Hellwigdda35b82010-02-15 09:44:46 +0000828}
829
Namjae Jeona904b1c2015-03-25 15:08:56 +1100830#define XFS_FALLOC_FL_SUPPORTED \
831 (FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | \
832 FALLOC_FL_COLLAPSE_RANGE | FALLOC_FL_ZERO_RANGE | \
833 FALLOC_FL_INSERT_RANGE)
834
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100835STATIC long
836xfs_file_fallocate(
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700837 struct file *file,
838 int mode,
839 loff_t offset,
840 loff_t len)
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100841{
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700842 struct inode *inode = file_inode(file);
843 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700844 long error;
Christoph Hellwig8add71c2015-02-02 09:53:56 +1100845 enum xfs_prealloc_flags flags = 0;
Christoph Hellwig781355c2015-02-16 11:59:50 +1100846 uint iolock = XFS_IOLOCK_EXCL;
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700847 loff_t new_size = 0;
Namjae Jeona904b1c2015-03-25 15:08:56 +1100848 bool do_file_insert = 0;
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100849
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700850 if (!S_ISREG(inode->i_mode))
851 return -EINVAL;
Namjae Jeona904b1c2015-03-25 15:08:56 +1100852 if (mode & ~XFS_FALLOC_FL_SUPPORTED)
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100853 return -EOPNOTSUPP;
854
Christoph Hellwig781355c2015-02-16 11:59:50 +1100855 xfs_ilock(ip, iolock);
Christoph Hellwig21c3ea12015-04-13 11:38:29 +1000856 error = xfs_break_layouts(inode, &iolock, false);
Christoph Hellwig781355c2015-02-16 11:59:50 +1100857 if (error)
858 goto out_unlock;
859
Dave Chinnere8e9ad42015-02-23 21:45:32 +1100860 xfs_ilock(ip, XFS_MMAPLOCK_EXCL);
861 iolock |= XFS_MMAPLOCK_EXCL;
862
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700863 if (mode & FALLOC_FL_PUNCH_HOLE) {
864 error = xfs_free_file_space(ip, offset, len);
865 if (error)
866 goto out_unlock;
Namjae Jeone1d8fb82014-02-24 10:58:19 +1100867 } else if (mode & FALLOC_FL_COLLAPSE_RANGE) {
868 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
869
870 if (offset & blksize_mask || len & blksize_mask) {
Dave Chinner24513372014-06-25 14:58:08 +1000871 error = -EINVAL;
Namjae Jeone1d8fb82014-02-24 10:58:19 +1100872 goto out_unlock;
873 }
874
Lukas Czerner23fffa92014-04-12 09:56:41 -0400875 /*
876 * There is no need to overlap collapse range with EOF,
877 * in which case it is effectively a truncate operation
878 */
879 if (offset + len >= i_size_read(inode)) {
Dave Chinner24513372014-06-25 14:58:08 +1000880 error = -EINVAL;
Lukas Czerner23fffa92014-04-12 09:56:41 -0400881 goto out_unlock;
882 }
883
Namjae Jeone1d8fb82014-02-24 10:58:19 +1100884 new_size = i_size_read(inode) - len;
885
886 error = xfs_collapse_file_space(ip, offset, len);
887 if (error)
888 goto out_unlock;
Namjae Jeona904b1c2015-03-25 15:08:56 +1100889 } else if (mode & FALLOC_FL_INSERT_RANGE) {
890 unsigned blksize_mask = (1 << inode->i_blkbits) - 1;
891
892 new_size = i_size_read(inode) + len;
893 if (offset & blksize_mask || len & blksize_mask) {
894 error = -EINVAL;
895 goto out_unlock;
896 }
897
898 /* check the new inode size does not wrap through zero */
899 if (new_size > inode->i_sb->s_maxbytes) {
900 error = -EFBIG;
901 goto out_unlock;
902 }
903
904 /* Offset should be less than i_size */
905 if (offset >= i_size_read(inode)) {
906 error = -EINVAL;
907 goto out_unlock;
908 }
909 do_file_insert = 1;
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700910 } else {
Christoph Hellwig8add71c2015-02-02 09:53:56 +1100911 flags |= XFS_PREALLOC_SET;
912
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700913 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
914 offset + len > i_size_read(inode)) {
915 new_size = offset + len;
Dave Chinner24513372014-06-25 14:58:08 +1000916 error = inode_newsize_ok(inode, new_size);
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700917 if (error)
918 goto out_unlock;
919 }
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100920
Lukas Czerner376ba312014-03-13 19:07:58 +1100921 if (mode & FALLOC_FL_ZERO_RANGE)
922 error = xfs_zero_file_space(ip, offset, len);
923 else
924 error = xfs_alloc_file_space(ip, offset, len,
925 XFS_BMAPI_PREALLOC);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100926 if (error)
927 goto out_unlock;
928 }
929
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700930 if (file->f_flags & O_DSYNC)
Christoph Hellwig8add71c2015-02-02 09:53:56 +1100931 flags |= XFS_PREALLOC_SYNC;
932
933 error = xfs_update_prealloc_flags(ip, flags);
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100934 if (error)
935 goto out_unlock;
936
937 /* Change file size if needed */
938 if (new_size) {
939 struct iattr iattr;
940
941 iattr.ia_valid = ATTR_SIZE;
942 iattr.ia_size = new_size;
Christoph Hellwig83aee9e2013-10-12 00:55:07 -0700943 error = xfs_setattr_size(ip, &iattr);
Namjae Jeona904b1c2015-03-25 15:08:56 +1100944 if (error)
945 goto out_unlock;
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100946 }
947
Namjae Jeona904b1c2015-03-25 15:08:56 +1100948 /*
949 * Perform hole insertion now that the file size has been
950 * updated so that if we crash during the operation we don't
951 * leave shifted extents past EOF and hence losing access to
952 * the data that is contained within them.
953 */
954 if (do_file_insert)
955 error = xfs_insert_file_space(ip, offset, len);
956
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100957out_unlock:
Christoph Hellwig781355c2015-02-16 11:59:50 +1100958 xfs_iunlock(ip, iolock);
Dave Chinner24513372014-06-25 14:58:08 +1000959 return error;
Christoph Hellwig2fe17c12011-01-14 13:07:43 +0100960}
961
962
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +1100964xfs_file_open(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965 struct inode *inode,
Christoph Hellwigf999a5b2008-11-28 14:23:32 +1100966 struct file *file)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967{
Christoph Hellwigf999a5b2008-11-28 14:23:32 +1100968 if (!(file->f_flags & O_LARGEFILE) && i_size_read(inode) > MAX_NON_LFS)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 return -EFBIG;
Christoph Hellwigf999a5b2008-11-28 14:23:32 +1100970 if (XFS_FORCED_SHUTDOWN(XFS_M(inode->i_sb)))
971 return -EIO;
972 return 0;
973}
974
975STATIC int
976xfs_dir_open(
977 struct inode *inode,
978 struct file *file)
979{
980 struct xfs_inode *ip = XFS_I(inode);
981 int mode;
982 int error;
983
984 error = xfs_file_open(inode, file);
985 if (error)
986 return error;
987
988 /*
989 * If there are any blocks, read-ahead block 0 as we're almost
990 * certain to have the next operation be a read there.
991 */
Christoph Hellwig309ecac2013-12-06 12:30:09 -0800992 mode = xfs_ilock_data_map_shared(ip);
Christoph Hellwigf999a5b2008-11-28 14:23:32 +1100993 if (ip->i_d.di_nextents > 0)
Eric Sandeen9df2dd02014-04-14 19:01:59 +1000994 xfs_dir3_data_readahead(ip, 0, -1);
Christoph Hellwigf999a5b2008-11-28 14:23:32 +1100995 xfs_iunlock(ip, mode);
996 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997}
998
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +11001000xfs_file_release(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 struct inode *inode,
1002 struct file *filp)
1003{
Dave Chinner24513372014-06-25 14:58:08 +10001004 return xfs_release(XFS_I(inode));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005}
1006
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007STATIC int
Nathan Scott3562fd42006-03-14 14:00:35 +11001008xfs_file_readdir(
Al Virob8227552013-05-22 17:07:56 -04001009 struct file *file,
1010 struct dir_context *ctx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011{
Al Virob8227552013-05-22 17:07:56 -04001012 struct inode *inode = file_inode(file);
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001013 xfs_inode_t *ip = XFS_I(inode);
Christoph Hellwig051e7cd2007-08-28 13:58:24 +10001014 size_t bufsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015
Christoph Hellwig051e7cd2007-08-28 13:58:24 +10001016 /*
1017 * The Linux API doesn't pass down the total size of the buffer
1018 * we read into down to the filesystem. With the filldir concept
1019 * it's not needed for correct information, but the XFS dir2 leaf
1020 * code wants an estimate of the buffer size to calculate it's
1021 * readahead window and size the buffers used for mapping to
1022 * physical blocks.
1023 *
1024 * Try to give it an estimate that's good enough, maybe at some
1025 * point we can change the ->readdir prototype to include the
Eric Sandeena9cc7992010-02-03 17:50:13 +00001026 * buffer size. For now we use the current glibc buffer size.
Christoph Hellwig051e7cd2007-08-28 13:58:24 +10001027 */
Eric Sandeena9cc7992010-02-03 17:50:13 +00001028 bufsize = (size_t)min_t(loff_t, 32768, ip->i_d.di_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001029
kbuild test robot83004752014-12-01 08:25:28 +11001030 return xfs_readdir(ip, ctx, bufsize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031}
1032
David Chinner4f57dbc2007-07-19 16:28:17 +10001033/*
Jeff Liud126d432012-08-21 17:11:57 +08001034 * This type is designed to indicate the type of offset we would like
Eric Sandeen49c69592014-09-09 11:56:48 +10001035 * to search from page cache for xfs_seek_hole_data().
Jeff Liud126d432012-08-21 17:11:57 +08001036 */
1037enum {
1038 HOLE_OFF = 0,
1039 DATA_OFF,
1040};
1041
1042/*
1043 * Lookup the desired type of offset from the given page.
1044 *
1045 * On success, return true and the offset argument will point to the
1046 * start of the region that was found. Otherwise this function will
1047 * return false and keep the offset argument unchanged.
1048 */
1049STATIC bool
1050xfs_lookup_buffer_offset(
1051 struct page *page,
1052 loff_t *offset,
1053 unsigned int type)
1054{
1055 loff_t lastoff = page_offset(page);
1056 bool found = false;
1057 struct buffer_head *bh, *head;
1058
1059 bh = head = page_buffers(page);
1060 do {
1061 /*
1062 * Unwritten extents that have data in the page
1063 * cache covering them can be identified by the
1064 * BH_Unwritten state flag. Pages with multiple
1065 * buffers might have a mix of holes, data and
1066 * unwritten extents - any buffer with valid
1067 * data in it should have BH_Uptodate flag set
1068 * on it.
1069 */
1070 if (buffer_unwritten(bh) ||
1071 buffer_uptodate(bh)) {
1072 if (type == DATA_OFF)
1073 found = true;
1074 } else {
1075 if (type == HOLE_OFF)
1076 found = true;
1077 }
1078
1079 if (found) {
1080 *offset = lastoff;
1081 break;
1082 }
1083 lastoff += bh->b_size;
1084 } while ((bh = bh->b_this_page) != head);
1085
1086 return found;
1087}
1088
1089/*
1090 * This routine is called to find out and return a data or hole offset
1091 * from the page cache for unwritten extents according to the desired
Eric Sandeen49c69592014-09-09 11:56:48 +10001092 * type for xfs_seek_hole_data().
Jeff Liud126d432012-08-21 17:11:57 +08001093 *
1094 * The argument offset is used to tell where we start to search from the
1095 * page cache. Map is used to figure out the end points of the range to
1096 * lookup pages.
1097 *
1098 * Return true if the desired type of offset was found, and the argument
1099 * offset is filled with that address. Otherwise, return false and keep
1100 * offset unchanged.
1101 */
1102STATIC bool
1103xfs_find_get_desired_pgoff(
1104 struct inode *inode,
1105 struct xfs_bmbt_irec *map,
1106 unsigned int type,
1107 loff_t *offset)
1108{
1109 struct xfs_inode *ip = XFS_I(inode);
1110 struct xfs_mount *mp = ip->i_mount;
1111 struct pagevec pvec;
1112 pgoff_t index;
1113 pgoff_t end;
1114 loff_t endoff;
1115 loff_t startoff = *offset;
1116 loff_t lastoff = startoff;
1117 bool found = false;
1118
1119 pagevec_init(&pvec, 0);
1120
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001121 index = startoff >> PAGE_SHIFT;
Jeff Liud126d432012-08-21 17:11:57 +08001122 endoff = XFS_FSB_TO_B(mp, map->br_startoff + map->br_blockcount);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001123 end = endoff >> PAGE_SHIFT;
Jeff Liud126d432012-08-21 17:11:57 +08001124 do {
1125 int want;
1126 unsigned nr_pages;
1127 unsigned int i;
1128
1129 want = min_t(pgoff_t, end - index, PAGEVEC_SIZE);
1130 nr_pages = pagevec_lookup(&pvec, inode->i_mapping, index,
1131 want);
1132 /*
1133 * No page mapped into given range. If we are searching holes
1134 * and if this is the first time we got into the loop, it means
1135 * that the given offset is landed in a hole, return it.
1136 *
1137 * If we have already stepped through some block buffers to find
1138 * holes but they all contains data. In this case, the last
1139 * offset is already updated and pointed to the end of the last
1140 * mapped page, if it does not reach the endpoint to search,
1141 * that means there should be a hole between them.
1142 */
1143 if (nr_pages == 0) {
1144 /* Data search found nothing */
1145 if (type == DATA_OFF)
1146 break;
1147
1148 ASSERT(type == HOLE_OFF);
1149 if (lastoff == startoff || lastoff < endoff) {
1150 found = true;
1151 *offset = lastoff;
1152 }
1153 break;
1154 }
1155
1156 /*
1157 * At lease we found one page. If this is the first time we
1158 * step into the loop, and if the first page index offset is
1159 * greater than the given search offset, a hole was found.
1160 */
1161 if (type == HOLE_OFF && lastoff == startoff &&
1162 lastoff < page_offset(pvec.pages[0])) {
1163 found = true;
1164 break;
1165 }
1166
1167 for (i = 0; i < nr_pages; i++) {
1168 struct page *page = pvec.pages[i];
1169 loff_t b_offset;
1170
1171 /*
1172 * At this point, the page may be truncated or
1173 * invalidated (changing page->mapping to NULL),
1174 * or even swizzled back from swapper_space to tmpfs
1175 * file mapping. However, page->index will not change
1176 * because we have a reference on the page.
1177 *
1178 * Searching done if the page index is out of range.
1179 * If the current offset is not reaches the end of
1180 * the specified search range, there should be a hole
1181 * between them.
1182 */
1183 if (page->index > end) {
1184 if (type == HOLE_OFF && lastoff < endoff) {
1185 *offset = lastoff;
1186 found = true;
1187 }
1188 goto out;
1189 }
1190
1191 lock_page(page);
1192 /*
1193 * Page truncated or invalidated(page->mapping == NULL).
1194 * We can freely skip it and proceed to check the next
1195 * page.
1196 */
1197 if (unlikely(page->mapping != inode->i_mapping)) {
1198 unlock_page(page);
1199 continue;
1200 }
1201
1202 if (!page_has_buffers(page)) {
1203 unlock_page(page);
1204 continue;
1205 }
1206
1207 found = xfs_lookup_buffer_offset(page, &b_offset, type);
1208 if (found) {
1209 /*
1210 * The found offset may be less than the start
1211 * point to search if this is the first time to
1212 * come here.
1213 */
1214 *offset = max_t(loff_t, startoff, b_offset);
1215 unlock_page(page);
1216 goto out;
1217 }
1218
1219 /*
1220 * We either searching data but nothing was found, or
1221 * searching hole but found a data buffer. In either
1222 * case, probably the next page contains the desired
1223 * things, update the last offset to it so.
1224 */
1225 lastoff = page_offset(page) + PAGE_SIZE;
1226 unlock_page(page);
1227 }
1228
1229 /*
1230 * The number of returned pages less than our desired, search
1231 * done. In this case, nothing was found for searching data,
1232 * but we found a hole behind the last offset.
1233 */
1234 if (nr_pages < want) {
1235 if (type == HOLE_OFF) {
1236 *offset = lastoff;
1237 found = true;
1238 }
1239 break;
1240 }
1241
1242 index = pvec.pages[i - 1]->index + 1;
1243 pagevec_release(&pvec);
1244 } while (index <= end);
1245
1246out:
1247 pagevec_release(&pvec);
1248 return found;
1249}
1250
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001251/*
1252 * caller must lock inode with xfs_ilock_data_map_shared,
1253 * can we craft an appropriate ASSERT?
1254 *
1255 * end is because the VFS-level lseek interface is defined such that any
1256 * offset past i_size shall return -ENXIO, but we use this for quota code
1257 * which does not maintain i_size, and we want to SEEK_DATA past i_size.
1258 */
1259loff_t
1260__xfs_seek_hole_data(
1261 struct inode *inode,
Eric Sandeen49c69592014-09-09 11:56:48 +10001262 loff_t start,
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001263 loff_t end,
Eric Sandeen49c69592014-09-09 11:56:48 +10001264 int whence)
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001265{
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001266 struct xfs_inode *ip = XFS_I(inode);
1267 struct xfs_mount *mp = ip->i_mount;
1268 loff_t uninitialized_var(offset);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001269 xfs_fileoff_t fsbno;
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001270 xfs_filblks_t lastbno;
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001271 int error;
1272
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001273 if (start >= end) {
Dave Chinner24513372014-06-25 14:58:08 +10001274 error = -ENXIO;
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001275 goto out_error;
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001276 }
1277
Eric Sandeen49c69592014-09-09 11:56:48 +10001278 /*
1279 * Try to read extents from the first block indicated
1280 * by fsbno to the end block of the file.
1281 */
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001282 fsbno = XFS_B_TO_FSBT(mp, start);
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001283 lastbno = XFS_B_TO_FSB(mp, end);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001284
Jeff Liub686d1f2012-08-21 17:12:18 +08001285 for (;;) {
1286 struct xfs_bmbt_irec map[2];
1287 int nmap = 2;
1288 unsigned int i;
1289
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001290 error = xfs_bmapi_read(ip, fsbno, lastbno - fsbno, map, &nmap,
Jeff Liub686d1f2012-08-21 17:12:18 +08001291 XFS_BMAPI_ENTIRE);
1292 if (error)
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001293 goto out_error;
Jeff Liub686d1f2012-08-21 17:12:18 +08001294
1295 /* No extents at given offset, must be beyond EOF */
1296 if (nmap == 0) {
Dave Chinner24513372014-06-25 14:58:08 +10001297 error = -ENXIO;
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001298 goto out_error;
Jeff Liub686d1f2012-08-21 17:12:18 +08001299 }
1300
1301 for (i = 0; i < nmap; i++) {
1302 offset = max_t(loff_t, start,
1303 XFS_FSB_TO_B(mp, map[i].br_startoff));
1304
Eric Sandeen49c69592014-09-09 11:56:48 +10001305 /* Landed in the hole we wanted? */
1306 if (whence == SEEK_HOLE &&
1307 map[i].br_startblock == HOLESTARTBLOCK)
1308 goto out;
1309
1310 /* Landed in the data extent we wanted? */
1311 if (whence == SEEK_DATA &&
1312 (map[i].br_startblock == DELAYSTARTBLOCK ||
1313 (map[i].br_state == XFS_EXT_NORM &&
1314 !isnullstartblock(map[i].br_startblock))))
Jeff Liub686d1f2012-08-21 17:12:18 +08001315 goto out;
1316
1317 /*
Eric Sandeen49c69592014-09-09 11:56:48 +10001318 * Landed in an unwritten extent, try to search
1319 * for hole or data from page cache.
Jeff Liub686d1f2012-08-21 17:12:18 +08001320 */
1321 if (map[i].br_state == XFS_EXT_UNWRITTEN) {
1322 if (xfs_find_get_desired_pgoff(inode, &map[i],
Eric Sandeen49c69592014-09-09 11:56:48 +10001323 whence == SEEK_HOLE ? HOLE_OFF : DATA_OFF,
1324 &offset))
Jeff Liub686d1f2012-08-21 17:12:18 +08001325 goto out;
1326 }
1327 }
1328
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001329 /*
Eric Sandeen49c69592014-09-09 11:56:48 +10001330 * We only received one extent out of the two requested. This
1331 * means we've hit EOF and didn't find what we are looking for.
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001332 */
Jeff Liub686d1f2012-08-21 17:12:18 +08001333 if (nmap == 1) {
Eric Sandeen49c69592014-09-09 11:56:48 +10001334 /*
1335 * If we were looking for a hole, set offset to
1336 * the end of the file (i.e., there is an implicit
1337 * hole at the end of any file).
1338 */
1339 if (whence == SEEK_HOLE) {
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001340 offset = end;
Eric Sandeen49c69592014-09-09 11:56:48 +10001341 break;
1342 }
1343 /*
1344 * If we were looking for data, it's nowhere to be found
1345 */
1346 ASSERT(whence == SEEK_DATA);
1347 error = -ENXIO;
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001348 goto out_error;
Jeff Liub686d1f2012-08-21 17:12:18 +08001349 }
1350
1351 ASSERT(i > 1);
1352
1353 /*
Eric Sandeen49c69592014-09-09 11:56:48 +10001354 * Nothing was found, proceed to the next round of search
1355 * if the next reading offset is not at or beyond EOF.
Jeff Liub686d1f2012-08-21 17:12:18 +08001356 */
1357 fsbno = map[i - 1].br_startoff + map[i - 1].br_blockcount;
1358 start = XFS_FSB_TO_B(mp, fsbno);
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001359 if (start >= end) {
Eric Sandeen49c69592014-09-09 11:56:48 +10001360 if (whence == SEEK_HOLE) {
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001361 offset = end;
Eric Sandeen49c69592014-09-09 11:56:48 +10001362 break;
1363 }
1364 ASSERT(whence == SEEK_DATA);
1365 error = -ENXIO;
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001366 goto out_error;
Jeff Liub686d1f2012-08-21 17:12:18 +08001367 }
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001368 }
1369
Jeff Liub686d1f2012-08-21 17:12:18 +08001370out:
1371 /*
Eric Sandeen49c69592014-09-09 11:56:48 +10001372 * If at this point we have found the hole we wanted, the returned
Jeff Liub686d1f2012-08-21 17:12:18 +08001373 * offset may be bigger than the file size as it may be aligned to
Eric Sandeen49c69592014-09-09 11:56:48 +10001374 * page boundary for unwritten extents. We need to deal with this
Jeff Liub686d1f2012-08-21 17:12:18 +08001375 * situation in particular.
1376 */
Eric Sandeen49c69592014-09-09 11:56:48 +10001377 if (whence == SEEK_HOLE)
Eric Sandeen8aa7d372016-02-08 11:25:16 +11001378 offset = min_t(loff_t, offset, end);
1379
1380 return offset;
1381
1382out_error:
1383 return error;
1384}
1385
1386STATIC loff_t
1387xfs_seek_hole_data(
1388 struct file *file,
1389 loff_t start,
1390 int whence)
1391{
1392 struct inode *inode = file->f_mapping->host;
1393 struct xfs_inode *ip = XFS_I(inode);
1394 struct xfs_mount *mp = ip->i_mount;
1395 uint lock;
1396 loff_t offset, end;
1397 int error = 0;
1398
1399 if (XFS_FORCED_SHUTDOWN(mp))
1400 return -EIO;
1401
1402 lock = xfs_ilock_data_map_shared(ip);
1403
1404 end = i_size_read(inode);
1405 offset = __xfs_seek_hole_data(inode, start, end, whence);
1406 if (offset < 0) {
1407 error = offset;
1408 goto out_unlock;
1409 }
1410
Jie Liu46a1c2c2013-06-25 12:02:13 +08001411 offset = vfs_setpos(file, offset, inode->i_sb->s_maxbytes);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001412
1413out_unlock:
Christoph Hellwig01f4f322013-12-06 12:30:08 -08001414 xfs_iunlock(ip, lock);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001415
1416 if (error)
Dave Chinner24513372014-06-25 14:58:08 +10001417 return error;
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001418 return offset;
1419}
1420
1421STATIC loff_t
1422xfs_file_llseek(
1423 struct file *file,
1424 loff_t offset,
Eric Sandeen59f9c002014-09-09 11:57:10 +10001425 int whence)
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001426{
Eric Sandeen59f9c002014-09-09 11:57:10 +10001427 switch (whence) {
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001428 case SEEK_END:
1429 case SEEK_CUR:
1430 case SEEK_SET:
Eric Sandeen59f9c002014-09-09 11:57:10 +10001431 return generic_file_llseek(file, offset, whence);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001432 case SEEK_HOLE:
Eric Sandeen49c69592014-09-09 11:56:48 +10001433 case SEEK_DATA:
Eric Sandeen59f9c002014-09-09 11:57:10 +10001434 return xfs_seek_hole_data(file, offset, whence);
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001435 default:
1436 return -EINVAL;
1437 }
1438}
1439
Dave Chinnerde0e8c22015-02-23 21:44:19 +11001440/*
1441 * Locking for serialisation of IO during page faults. This results in a lock
1442 * ordering of:
1443 *
1444 * mmap_sem (MM)
Dave Chinner6b698ed2015-06-04 09:18:53 +10001445 * sb_start_pagefault(vfs, freeze)
Dave Chinner13ad4fe2015-11-03 12:37:02 +11001446 * i_mmaplock (XFS - truncate serialisation)
Dave Chinner6b698ed2015-06-04 09:18:53 +10001447 * page_lock (MM)
1448 * i_lock (XFS - extent map serialisation)
Dave Chinnerde0e8c22015-02-23 21:44:19 +11001449 */
Dave Chinnerde0e8c22015-02-23 21:44:19 +11001450
Dave Chinner075a9242015-02-23 21:44:54 +11001451/*
1452 * mmap()d file has taken write protection fault and is being made writable. We
1453 * can set the page state up correctly for a writable page, which means we can
1454 * do correct delalloc accounting (ENOSPC checking!) and unwritten extent
1455 * mapping.
1456 */
1457STATIC int
1458xfs_filemap_page_mkwrite(
1459 struct vm_area_struct *vma,
1460 struct vm_fault *vmf)
1461{
Dave Chinner6b698ed2015-06-04 09:18:53 +10001462 struct inode *inode = file_inode(vma->vm_file);
Dave Chinnerec56b1f2015-06-04 09:18:18 +10001463 int ret;
Dave Chinner075a9242015-02-23 21:44:54 +11001464
Dave Chinner6b698ed2015-06-04 09:18:53 +10001465 trace_xfs_filemap_page_mkwrite(XFS_I(inode));
Dave Chinner075a9242015-02-23 21:44:54 +11001466
Dave Chinner6b698ed2015-06-04 09:18:53 +10001467 sb_start_pagefault(inode->i_sb);
Dave Chinnerec56b1f2015-06-04 09:18:18 +10001468 file_update_time(vma->vm_file);
Dave Chinner6b698ed2015-06-04 09:18:53 +10001469 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1470
1471 if (IS_DAX(inode)) {
Christoph Hellwig6c31f492016-09-19 11:28:38 +10001472 ret = iomap_dax_fault(vma, vmf, &xfs_iomap_ops);
Dave Chinner6b698ed2015-06-04 09:18:53 +10001473 } else {
Christoph Hellwig68a9f5e2016-06-21 09:53:44 +10001474 ret = iomap_page_mkwrite(vma, vmf, &xfs_iomap_ops);
Dave Chinner6b698ed2015-06-04 09:18:53 +10001475 ret = block_page_mkwrite_return(ret);
1476 }
1477
1478 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1479 sb_end_pagefault(inode->i_sb);
1480
1481 return ret;
1482}
1483
1484STATIC int
1485xfs_filemap_fault(
1486 struct vm_area_struct *vma,
1487 struct vm_fault *vmf)
1488{
Dave Chinnerb2442c52015-07-29 11:48:00 +10001489 struct inode *inode = file_inode(vma->vm_file);
Dave Chinner6b698ed2015-06-04 09:18:53 +10001490 int ret;
1491
Dave Chinnerb2442c52015-07-29 11:48:00 +10001492 trace_xfs_filemap_fault(XFS_I(inode));
Dave Chinner6b698ed2015-06-04 09:18:53 +10001493
1494 /* DAX can shortcut the normal fault path on write faults! */
Dave Chinnerb2442c52015-07-29 11:48:00 +10001495 if ((vmf->flags & FAULT_FLAG_WRITE) && IS_DAX(inode))
Dave Chinner6b698ed2015-06-04 09:18:53 +10001496 return xfs_filemap_page_mkwrite(vma, vmf);
Dave Chinner075a9242015-02-23 21:44:54 +11001497
Dave Chinnerb2442c52015-07-29 11:48:00 +10001498 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
1499 if (IS_DAX(inode)) {
1500 /*
1501 * we do not want to trigger unwritten extent conversion on read
1502 * faults - that is unnecessary overhead and would also require
1503 * changes to xfs_get_blocks_direct() to map unwritten extent
1504 * ioend for conversion on read-only mappings.
1505 */
Christoph Hellwig6c31f492016-09-19 11:28:38 +10001506 ret = iomap_dax_fault(vma, vmf, &xfs_iomap_ops);
Dave Chinnerb2442c52015-07-29 11:48:00 +10001507 } else
1508 ret = filemap_fault(vma, vmf);
1509 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
Dave Chinner075a9242015-02-23 21:44:54 +11001510
Dave Chinner6b698ed2015-06-04 09:18:53 +10001511 return ret;
1512}
1513
Dave Chinner13ad4fe2015-11-03 12:37:02 +11001514/*
1515 * Similar to xfs_filemap_fault(), the DAX fault path can call into here on
1516 * both read and write faults. Hence we need to handle both cases. There is no
1517 * ->pmd_mkwrite callout for huge pages, so we have a single function here to
1518 * handle both cases here. @flags carries the information on the type of fault
1519 * occuring.
1520 */
Matthew Wilcoxacd76e72015-09-08 14:59:06 -07001521STATIC int
1522xfs_filemap_pmd_fault(
1523 struct vm_area_struct *vma,
1524 unsigned long addr,
1525 pmd_t *pmd,
1526 unsigned int flags)
1527{
1528 struct inode *inode = file_inode(vma->vm_file);
1529 struct xfs_inode *ip = XFS_I(inode);
1530 int ret;
1531
1532 if (!IS_DAX(inode))
1533 return VM_FAULT_FALLBACK;
1534
1535 trace_xfs_filemap_pmd_fault(ip);
1536
Dave Chinner13ad4fe2015-11-03 12:37:02 +11001537 if (flags & FAULT_FLAG_WRITE) {
1538 sb_start_pagefault(inode->i_sb);
1539 file_update_time(vma->vm_file);
1540 }
1541
Matthew Wilcoxacd76e72015-09-08 14:59:06 -07001542 xfs_ilock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
Ross Zwisler6b524992016-07-26 15:21:05 -07001543 ret = dax_pmd_fault(vma, addr, pmd, flags, xfs_get_blocks_dax_fault);
Matthew Wilcoxacd76e72015-09-08 14:59:06 -07001544 xfs_iunlock(XFS_I(inode), XFS_MMAPLOCK_SHARED);
Dave Chinner13ad4fe2015-11-03 12:37:02 +11001545
1546 if (flags & FAULT_FLAG_WRITE)
1547 sb_end_pagefault(inode->i_sb);
Matthew Wilcoxacd76e72015-09-08 14:59:06 -07001548
1549 return ret;
1550}
1551
Dave Chinner3af49282015-11-03 12:37:02 +11001552/*
1553 * pfn_mkwrite was originally inteneded to ensure we capture time stamp
1554 * updates on write faults. In reality, it's need to serialise against
Ross Zwisler5eb88dc2016-01-22 15:10:56 -08001555 * truncate similar to page_mkwrite. Hence we cycle the XFS_MMAPLOCK_SHARED
1556 * to ensure we serialise the fault barrier in place.
Dave Chinner3af49282015-11-03 12:37:02 +11001557 */
1558static int
1559xfs_filemap_pfn_mkwrite(
1560 struct vm_area_struct *vma,
1561 struct vm_fault *vmf)
1562{
1563
1564 struct inode *inode = file_inode(vma->vm_file);
1565 struct xfs_inode *ip = XFS_I(inode);
1566 int ret = VM_FAULT_NOPAGE;
1567 loff_t size;
1568
1569 trace_xfs_filemap_pfn_mkwrite(ip);
1570
1571 sb_start_pagefault(inode->i_sb);
1572 file_update_time(vma->vm_file);
1573
1574 /* check if the faulting page hasn't raced with truncate */
1575 xfs_ilock(ip, XFS_MMAPLOCK_SHARED);
1576 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1577 if (vmf->pgoff >= size)
1578 ret = VM_FAULT_SIGBUS;
Ross Zwisler5eb88dc2016-01-22 15:10:56 -08001579 else if (IS_DAX(inode))
1580 ret = dax_pfn_mkwrite(vma, vmf);
Dave Chinner3af49282015-11-03 12:37:02 +11001581 xfs_iunlock(ip, XFS_MMAPLOCK_SHARED);
1582 sb_end_pagefault(inode->i_sb);
1583 return ret;
1584
1585}
1586
Dave Chinner6b698ed2015-06-04 09:18:53 +10001587static const struct vm_operations_struct xfs_file_vm_ops = {
1588 .fault = xfs_filemap_fault,
Matthew Wilcoxacd76e72015-09-08 14:59:06 -07001589 .pmd_fault = xfs_filemap_pmd_fault,
Dave Chinner6b698ed2015-06-04 09:18:53 +10001590 .map_pages = filemap_map_pages,
1591 .page_mkwrite = xfs_filemap_page_mkwrite,
Dave Chinner3af49282015-11-03 12:37:02 +11001592 .pfn_mkwrite = xfs_filemap_pfn_mkwrite,
Dave Chinner6b698ed2015-06-04 09:18:53 +10001593};
1594
1595STATIC int
1596xfs_file_mmap(
1597 struct file *filp,
1598 struct vm_area_struct *vma)
1599{
1600 file_accessed(filp);
1601 vma->vm_ops = &xfs_file_vm_ops;
1602 if (IS_DAX(file_inode(filp)))
Matthew Wilcoxacd76e72015-09-08 14:59:06 -07001603 vma->vm_flags |= VM_MIXEDMAP | VM_HUGEPAGE;
Dave Chinner6b698ed2015-06-04 09:18:53 +10001604 return 0;
Dave Chinner075a9242015-02-23 21:44:54 +11001605}
1606
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001607const struct file_operations xfs_file_operations = {
Jeff Liu3fe3e6b2012-05-10 21:29:17 +08001608 .llseek = xfs_file_llseek,
Al Virob4f5d2c2014-04-02 14:37:59 -04001609 .read_iter = xfs_file_read_iter,
Al Virobf97f3bc2014-04-03 14:20:23 -04001610 .write_iter = xfs_file_write_iter,
Nathan Scott1b895842006-03-31 13:08:59 +10001611 .splice_read = xfs_file_splice_read,
Al Viro8d020762014-04-05 04:27:08 -04001612 .splice_write = iter_file_splice_write,
Nathan Scott3562fd42006-03-14 14:00:35 +11001613 .unlocked_ioctl = xfs_file_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614#ifdef CONFIG_COMPAT
Nathan Scott3562fd42006-03-14 14:00:35 +11001615 .compat_ioctl = xfs_file_compat_ioctl,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616#endif
Nathan Scott3562fd42006-03-14 14:00:35 +11001617 .mmap = xfs_file_mmap,
1618 .open = xfs_file_open,
1619 .release = xfs_file_release,
1620 .fsync = xfs_file_fsync,
Christoph Hellwig2fe17c12011-01-14 13:07:43 +01001621 .fallocate = xfs_file_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001622};
1623
Arjan van de Ven4b6f5d22006-03-28 01:56:42 -08001624const struct file_operations xfs_dir_file_operations = {
Christoph Hellwigf999a5b2008-11-28 14:23:32 +11001625 .open = xfs_dir_open,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 .read = generic_read_dir,
Al Viro3b0a3c12016-04-20 23:42:46 -04001627 .iterate_shared = xfs_file_readdir,
Al Viro59af1582008-08-24 07:24:41 -04001628 .llseek = generic_file_llseek,
Nathan Scott3562fd42006-03-14 14:00:35 +11001629 .unlocked_ioctl = xfs_file_ioctl,
Nathan Scottd3870392005-05-06 06:44:46 -07001630#ifdef CONFIG_COMPAT
Nathan Scott3562fd42006-03-14 14:00:35 +11001631 .compat_ioctl = xfs_file_compat_ioctl,
Nathan Scottd3870392005-05-06 06:44:46 -07001632#endif
Christoph Hellwig1da2f2d2011-10-02 14:25:16 +00001633 .fsync = xfs_dir_fsync,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634};