blob: 3a9b7a1b8704be66439ea797dd2183c035a929e5 [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"
Dave Chinner70a98832013-10-23 10:36:05 +110019#include "xfs_shared.h"
Dave Chinner239880e2013-10-23 10:50:10 +110020#include "xfs_format.h"
21#include "xfs_log_format.h"
22#include "xfs_trans_resv.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070023#include "xfs_mount.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_inode.h"
Dave Chinner239880e2013-10-23 10:50:10 +110025#include "xfs_trans.h"
Christoph Hellwig281627d2012-03-13 08:41:05 +000026#include "xfs_inode_item.h"
Nathan Scotta844f452005-11-02 14:38:42 +110027#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include "xfs_error.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include "xfs_iomap.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000030#include "xfs_trace.h"
Dave Chinner3ed3a432010-03-05 02:00:42 +000031#include "xfs_bmap.h"
Dave Chinner68988112013-08-12 20:49:42 +100032#include "xfs_bmap_util.h"
Dave Chinnera4fbe6a2013-10-23 10:51:50 +110033#include "xfs_bmap_btree.h"
Kent Overstreeta27bb332013-05-07 16:19:08 -070034#include <linux/aio.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090035#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036#include <linux/mpage.h>
Christoph Hellwig10ce4442006-01-11 20:48:14 +110037#include <linux/pagevec.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <linux/writeback.h>
39
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000040void
Nathan Scottf51623b2006-03-14 13:26:27 +110041xfs_count_page_state(
42 struct page *page,
43 int *delalloc,
Nathan Scottf51623b2006-03-14 13:26:27 +110044 int *unwritten)
45{
46 struct buffer_head *bh, *head;
47
Christoph Hellwig20cb52e2010-06-24 09:46:01 +100048 *delalloc = *unwritten = 0;
Nathan Scottf51623b2006-03-14 13:26:27 +110049
50 bh = head = page_buffers(page);
51 do {
Christoph Hellwig20cb52e2010-06-24 09:46:01 +100052 if (buffer_unwritten(bh))
Nathan Scottf51623b2006-03-14 13:26:27 +110053 (*unwritten) = 1;
54 else if (buffer_delay(bh))
55 (*delalloc) = 1;
56 } while ((bh = bh->b_this_page) != head);
57}
58
Christoph Hellwig6214ed42007-09-14 15:23:17 +100059STATIC struct block_device *
60xfs_find_bdev_for_inode(
Christoph Hellwig046f1682010-04-28 12:28:52 +000061 struct inode *inode)
Christoph Hellwig6214ed42007-09-14 15:23:17 +100062{
Christoph Hellwig046f1682010-04-28 12:28:52 +000063 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwig6214ed42007-09-14 15:23:17 +100064 struct xfs_mount *mp = ip->i_mount;
65
Eric Sandeen71ddabb2007-11-23 16:29:42 +110066 if (XFS_IS_REALTIME_INODE(ip))
Christoph Hellwig6214ed42007-09-14 15:23:17 +100067 return mp->m_rtdev_targp->bt_bdev;
68 else
69 return mp->m_ddev_targp->bt_bdev;
70}
71
Christoph Hellwig0829c362005-09-02 16:58:49 +100072/*
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +110073 * We're now finished for good with this ioend structure.
74 * Update the page state via the associated buffer_heads,
75 * release holds on the inode and bio, and finally free
76 * up memory. Do not use the ioend after this.
77 */
Christoph Hellwig0829c362005-09-02 16:58:49 +100078STATIC void
79xfs_destroy_ioend(
80 xfs_ioend_t *ioend)
81{
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +110082 struct buffer_head *bh, *next;
83
84 for (bh = ioend->io_buffer_head; bh; bh = next) {
85 next = bh->b_private;
Nathan Scott7d04a332006-06-09 14:58:38 +100086 bh->b_end_io(bh, !ioend->io_error);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +110087 }
Christoph Hellwig583fa582008-12-03 12:20:38 +010088
Christoph Hellwig0829c362005-09-02 16:58:49 +100089 mempool_free(ioend, xfs_ioend_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -070090}
91
92/*
Christoph Hellwigfc0063c2011-08-23 08:28:11 +000093 * Fast and loose check if this write could update the on-disk inode size.
94 */
95static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend)
96{
97 return ioend->io_offset + ioend->io_size >
98 XFS_I(ioend->io_inode)->i_d.di_size;
99}
100
Christoph Hellwig281627d2012-03-13 08:41:05 +0000101STATIC int
102xfs_setfilesize_trans_alloc(
103 struct xfs_ioend *ioend)
104{
105 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
106 struct xfs_trans *tp;
107 int error;
108
109 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
110
Jie Liu3d3c8b52013-08-12 20:49:59 +1000111 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000112 if (error) {
113 xfs_trans_cancel(tp, 0);
114 return error;
115 }
116
117 ioend->io_append_trans = tp;
118
119 /*
Dave Chinner437a2552012-11-28 13:01:00 +1100120 * We may pass freeze protection with a transaction. So tell lockdep
Jan Karad9457dc2012-06-12 16:20:39 +0200121 * we released it.
122 */
123 rwsem_release(&ioend->io_inode->i_sb->s_writers.lock_map[SB_FREEZE_FS-1],
124 1, _THIS_IP_);
125 /*
Christoph Hellwig281627d2012-03-13 08:41:05 +0000126 * We hand off the transaction to the completion thread now, so
127 * clear the flag here.
128 */
129 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
130 return 0;
131}
132
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000133/*
Christoph Hellwig2813d682011-12-18 20:00:12 +0000134 * Update on-disk file size now that data has been written to disk.
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000135 */
Christoph Hellwig281627d2012-03-13 08:41:05 +0000136STATIC int
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000137xfs_setfilesize(
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100138 struct xfs_inode *ip,
139 struct xfs_trans *tp,
140 xfs_off_t offset,
141 size_t size)
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000142{
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000143 xfs_fsize_t isize;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000144
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000145 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100146 isize = xfs_new_eof(ip, offset + size);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000147 if (!isize) {
148 xfs_iunlock(ip, XFS_ILOCK_EXCL);
149 xfs_trans_cancel(tp, 0);
150 return 0;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000151 }
152
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100153 trace_xfs_setfilesize(ip, offset, size);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000154
155 ip->i_d.di_size = isize;
156 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
157 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
158
159 return xfs_trans_commit(tp, 0);
Christoph Hellwig0829c362005-09-02 16:58:49 +1000160}
161
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100162STATIC int
163xfs_setfilesize_ioend(
164 struct xfs_ioend *ioend)
165{
166 struct xfs_inode *ip = XFS_I(ioend->io_inode);
167 struct xfs_trans *tp = ioend->io_append_trans;
168
169 /*
170 * The transaction may have been allocated in the I/O submission thread,
171 * thus we need to mark ourselves as being in a transaction manually.
172 * Similarly for freeze protection.
173 */
174 current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
175 rwsem_acquire_read(&VFS_I(ip)->i_sb->s_writers.lock_map[SB_FREEZE_FS-1],
176 0, 1, _THIS_IP_);
177
178 return xfs_setfilesize(ip, tp, ioend->io_offset, ioend->io_size);
179}
180
Christoph Hellwig0829c362005-09-02 16:58:49 +1000181/*
Christoph Hellwig209fb872010-07-18 21:17:11 +0000182 * Schedule IO completion handling on the final put of an ioend.
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000183 *
184 * If there is no work to do we might as well call it a day and free the
185 * ioend right now.
Dave Chinnerc626d172009-04-06 18:42:11 +0200186 */
187STATIC void
188xfs_finish_ioend(
Christoph Hellwig209fb872010-07-18 21:17:11 +0000189 struct xfs_ioend *ioend)
Dave Chinnerc626d172009-04-06 18:42:11 +0200190{
191 if (atomic_dec_and_test(&ioend->io_remaining)) {
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000192 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
193
Alain Renaud0d882a32012-05-22 15:56:21 -0500194 if (ioend->io_type == XFS_IO_UNWRITTEN)
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000195 queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100196 else if (ioend->io_append_trans)
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000197 queue_work(mp->m_data_workqueue, &ioend->io_work);
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000198 else
199 xfs_destroy_ioend(ioend);
Dave Chinnerc626d172009-04-06 18:42:11 +0200200 }
201}
202
203/*
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000204 * IO write completion.
205 */
206STATIC void
207xfs_end_io(
208 struct work_struct *work)
209{
210 xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work);
211 struct xfs_inode *ip = XFS_I(ioend->io_inode);
Dave Chinner69418932010-03-04 00:57:09 +0000212 int error = 0;
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000213
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000214 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
Christoph Hellwig810627d2011-11-08 08:56:15 +0000215 ioend->io_error = -EIO;
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000216 goto done;
217 }
218 if (ioend->io_error)
219 goto done;
220
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000221 /*
222 * For unwritten extents we need to issue transactions to convert a
223 * range to normal written extens after the data I/O has finished.
224 */
Alain Renaud0d882a32012-05-22 15:56:21 -0500225 if (ioend->io_type == XFS_IO_UNWRITTEN) {
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000226 error = xfs_iomap_write_unwritten(ip, ioend->io_offset,
Dave Chinner437a2552012-11-28 13:01:00 +1100227 ioend->io_size);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000228 } else if (ioend->io_append_trans) {
Christoph Hellwig2ba66232015-02-02 10:02:09 +1100229 error = xfs_setfilesize_ioend(ioend);
Christoph Hellwig84803fb2012-02-29 09:53:50 +0000230 } else {
Christoph Hellwig281627d2012-03-13 08:41:05 +0000231 ASSERT(!xfs_ioend_is_append(ioend));
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000232 }
233
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000234done:
Dave Chinner437a2552012-11-28 13:01:00 +1100235 if (error)
Dave Chinner24513372014-06-25 14:58:08 +1000236 ioend->io_error = error;
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000237 xfs_destroy_ioend(ioend);
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000238}
239
240/*
Christoph Hellwig0829c362005-09-02 16:58:49 +1000241 * Allocate and initialise an IO completion structure.
242 * We need to track unwritten extent write completion here initially.
243 * We'll need to extend this for updating the ondisk inode size later
244 * (vs. incore size).
245 */
246STATIC xfs_ioend_t *
247xfs_alloc_ioend(
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100248 struct inode *inode,
249 unsigned int type)
Christoph Hellwig0829c362005-09-02 16:58:49 +1000250{
251 xfs_ioend_t *ioend;
252
253 ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
254
255 /*
256 * Set the count to 1 initially, which will prevent an I/O
257 * completion callback from happening before we have started
258 * all the I/O from calling the completion routine too early.
259 */
260 atomic_set(&ioend->io_remaining, 1);
Nathan Scott7d04a332006-06-09 14:58:38 +1000261 ioend->io_error = 0;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100262 ioend->io_list = NULL;
263 ioend->io_type = type;
Christoph Hellwigb677c212007-08-29 11:46:28 +1000264 ioend->io_inode = inode;
Christoph Hellwigc1a073b2005-09-05 08:23:35 +1000265 ioend->io_buffer_head = NULL;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100266 ioend->io_buffer_tail = NULL;
Christoph Hellwig0829c362005-09-02 16:58:49 +1000267 ioend->io_offset = 0;
268 ioend->io_size = 0;
Christoph Hellwig281627d2012-03-13 08:41:05 +0000269 ioend->io_append_trans = NULL;
Christoph Hellwig0829c362005-09-02 16:58:49 +1000270
Christoph Hellwig5ec4fab2009-10-30 09:11:47 +0000271 INIT_WORK(&ioend->io_work, xfs_end_io);
Christoph Hellwig0829c362005-09-02 16:58:49 +1000272 return ioend;
273}
274
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275STATIC int
276xfs_map_blocks(
277 struct inode *inode,
278 loff_t offset,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000279 struct xfs_bmbt_irec *imap,
Christoph Hellwiga206c812010-12-10 08:42:20 +0000280 int type,
281 int nonblocking)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282{
Christoph Hellwiga206c812010-12-10 08:42:20 +0000283 struct xfs_inode *ip = XFS_I(inode);
284 struct xfs_mount *mp = ip->i_mount;
Christoph Hellwiged1e7b72010-12-10 08:42:22 +0000285 ssize_t count = 1 << inode->i_blkbits;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000286 xfs_fileoff_t offset_fsb, end_fsb;
287 int error = 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000288 int bmapi_flags = XFS_BMAPI_ENTIRE;
289 int nimaps = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700290
Christoph Hellwiga206c812010-12-10 08:42:20 +0000291 if (XFS_FORCED_SHUTDOWN(mp))
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000292 return -EIO;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000293
Alain Renaud0d882a32012-05-22 15:56:21 -0500294 if (type == XFS_IO_UNWRITTEN)
Christoph Hellwiga206c812010-12-10 08:42:20 +0000295 bmapi_flags |= XFS_BMAPI_IGSTATE;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000296
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000297 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
298 if (nonblocking)
Eric Sandeenb474c7a2014-06-22 15:04:54 +1000299 return -EAGAIN;
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000300 xfs_ilock(ip, XFS_ILOCK_SHARED);
Christoph Hellwiga206c812010-12-10 08:42:20 +0000301 }
302
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000303 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
304 (ip->i_df.if_flags & XFS_IFEXTENTS));
Dave Chinnerd2c28192012-06-08 15:44:53 +1000305 ASSERT(offset <= mp->m_super->s_maxbytes);
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000306
Dave Chinnerd2c28192012-06-08 15:44:53 +1000307 if (offset + count > mp->m_super->s_maxbytes)
308 count = mp->m_super->s_maxbytes - offset;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000309 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
310 offset_fsb = XFS_B_TO_FSBT(mp, offset);
Dave Chinner5c8ed202011-09-18 20:40:45 +0000311 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
312 imap, &nimaps, bmapi_flags);
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000313 xfs_iunlock(ip, XFS_ILOCK_SHARED);
314
Christoph Hellwiga206c812010-12-10 08:42:20 +0000315 if (error)
Dave Chinner24513372014-06-25 14:58:08 +1000316 return error;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000317
Alain Renaud0d882a32012-05-22 15:56:21 -0500318 if (type == XFS_IO_DELALLOC &&
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000319 (!nimaps || isnullstartblock(imap->br_startblock))) {
Jie Liu0799a3e2013-09-29 18:56:04 +0800320 error = xfs_iomap_write_allocate(ip, offset, imap);
Christoph Hellwiga206c812010-12-10 08:42:20 +0000321 if (!error)
322 trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
Dave Chinner24513372014-06-25 14:58:08 +1000323 return error;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000324 }
325
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000326#ifdef DEBUG
Alain Renaud0d882a32012-05-22 15:56:21 -0500327 if (type == XFS_IO_UNWRITTEN) {
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000328 ASSERT(nimaps);
329 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
330 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
331 }
332#endif
333 if (nimaps)
334 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
335 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336}
337
Christoph Hellwigb8f82a42009-11-14 16:17:22 +0000338STATIC int
Christoph Hellwig558e6892010-04-28 12:28:58 +0000339xfs_imap_valid(
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000340 struct inode *inode,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000341 struct xfs_bmbt_irec *imap,
Christoph Hellwig558e6892010-04-28 12:28:58 +0000342 xfs_off_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343{
Christoph Hellwig558e6892010-04-28 12:28:58 +0000344 offset >>= inode->i_blkbits;
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000345
Christoph Hellwig558e6892010-04-28 12:28:58 +0000346 return offset >= imap->br_startoff &&
347 offset < imap->br_startoff + imap->br_blockcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348}
349
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100350/*
351 * BIO completion handler for buffered IO.
352 */
Al Viro782e3b32007-10-12 07:17:47 +0100353STATIC void
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100354xfs_end_bio(
355 struct bio *bio,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100356 int error)
357{
358 xfs_ioend_t *ioend = bio->bi_private;
359
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100360 ASSERT(atomic_read(&bio->bi_cnt) >= 1);
Nathan Scott7d04a332006-06-09 14:58:38 +1000361 ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100362
363 /* Toss bio and pass work off to an xfsdatad thread */
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100364 bio->bi_private = NULL;
365 bio->bi_end_io = NULL;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100366 bio_put(bio);
Nathan Scott7d04a332006-06-09 14:58:38 +1000367
Christoph Hellwig209fb872010-07-18 21:17:11 +0000368 xfs_finish_ioend(ioend);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100369}
370
371STATIC void
372xfs_submit_ioend_bio(
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000373 struct writeback_control *wbc,
374 xfs_ioend_t *ioend,
375 struct bio *bio)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100376{
377 atomic_inc(&ioend->io_remaining);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100378 bio->bi_private = ioend;
379 bio->bi_end_io = xfs_end_bio;
Jens Axboe721a9602011-03-09 11:56:30 +0100380 submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100381}
382
383STATIC struct bio *
384xfs_alloc_ioend_bio(
385 struct buffer_head *bh)
386{
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100387 int nvecs = bio_get_nr_vecs(bh->b_bdev);
Christoph Hellwig221cb252010-12-10 08:42:17 +0000388 struct bio *bio = bio_alloc(GFP_NOIO, nvecs);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100389
390 ASSERT(bio->bi_private == NULL);
Kent Overstreet4f024f32013-10-11 15:44:27 -0700391 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100392 bio->bi_bdev = bh->b_bdev;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100393 return bio;
394}
395
396STATIC void
397xfs_start_buffer_writeback(
398 struct buffer_head *bh)
399{
400 ASSERT(buffer_mapped(bh));
401 ASSERT(buffer_locked(bh));
402 ASSERT(!buffer_delay(bh));
403 ASSERT(!buffer_unwritten(bh));
404
405 mark_buffer_async_write(bh);
406 set_buffer_uptodate(bh);
407 clear_buffer_dirty(bh);
408}
409
410STATIC void
411xfs_start_page_writeback(
412 struct page *page,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100413 int clear_dirty,
414 int buffers)
415{
416 ASSERT(PageLocked(page));
417 ASSERT(!PageWriteback(page));
Dave Chinner0d085a52014-09-23 15:36:27 +1000418
419 /*
420 * if the page was not fully cleaned, we need to ensure that the higher
421 * layers come back to it correctly. That means we need to keep the page
422 * dirty, and for WB_SYNC_ALL writeback we need to ensure the
423 * PAGECACHE_TAG_TOWRITE index mark is not removed so another attempt to
424 * write this page in this writeback sweep will be made.
425 */
426 if (clear_dirty) {
David Chinner92132022006-12-21 10:24:01 +1100427 clear_page_dirty_for_io(page);
Dave Chinner0d085a52014-09-23 15:36:27 +1000428 set_page_writeback(page);
429 } else
430 set_page_writeback_keepwrite(page);
431
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100432 unlock_page(page);
Dave Chinner0d085a52014-09-23 15:36:27 +1000433
Fengguang Wu1f7decf2007-10-16 23:30:42 -0700434 /* If no buffers on the page are to be written, finish it here */
435 if (!buffers)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100436 end_page_writeback(page);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100437}
438
Zhi Yong Wuc7c1a7d2013-08-07 10:11:09 +0000439static inline int xfs_bio_add_buffer(struct bio *bio, struct buffer_head *bh)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100440{
441 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
442}
443
444/*
David Chinnerd88992f2006-01-18 13:38:12 +1100445 * Submit all of the bios for all of the ioends we have saved up, covering the
446 * initial writepage page and also any probed pages.
447 *
448 * Because we may have multiple ioends spanning a page, we need to start
449 * writeback on all the buffers before we submit them for I/O. If we mark the
450 * buffers as we got, then we can end up with a page that only has buffers
451 * marked async write and I/O complete on can occur before we mark the other
452 * buffers async write.
453 *
454 * The end result of this is that we trip a bug in end_page_writeback() because
455 * we call it twice for the one page as the code in end_buffer_async_write()
456 * assumes that all buffers on the page are started at the same time.
457 *
458 * The fix is two passes across the ioend list - one to start writeback on the
Nathan Scottc41564b2006-03-29 08:55:14 +1000459 * buffer_heads, and then submit them for I/O on the second pass.
Dave Chinner7bf7f352012-11-12 22:09:45 +1100460 *
461 * If @fail is non-zero, it means that we have a situation where some part of
462 * the submission process has failed after we have marked paged for writeback
463 * and unlocked them. In this situation, we need to fail the ioend chain rather
464 * than submit it to IO. This typically only happens on a filesystem shutdown.
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100465 */
466STATIC void
467xfs_submit_ioend(
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000468 struct writeback_control *wbc,
Dave Chinner7bf7f352012-11-12 22:09:45 +1100469 xfs_ioend_t *ioend,
470 int fail)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100471{
David Chinnerd88992f2006-01-18 13:38:12 +1100472 xfs_ioend_t *head = ioend;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100473 xfs_ioend_t *next;
474 struct buffer_head *bh;
475 struct bio *bio;
476 sector_t lastblock = 0;
477
David Chinnerd88992f2006-01-18 13:38:12 +1100478 /* Pass 1 - start writeback */
479 do {
480 next = ioend->io_list;
Christoph Hellwig221cb252010-12-10 08:42:17 +0000481 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private)
David Chinnerd88992f2006-01-18 13:38:12 +1100482 xfs_start_buffer_writeback(bh);
David Chinnerd88992f2006-01-18 13:38:12 +1100483 } while ((ioend = next) != NULL);
484
485 /* Pass 2 - submit I/O */
486 ioend = head;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100487 do {
488 next = ioend->io_list;
489 bio = NULL;
490
Dave Chinner7bf7f352012-11-12 22:09:45 +1100491 /*
492 * If we are failing the IO now, just mark the ioend with an
493 * error and finish it. This will run IO completion immediately
494 * as there is only one reference to the ioend at this point in
495 * time.
496 */
497 if (fail) {
Dave Chinner24513372014-06-25 14:58:08 +1000498 ioend->io_error = fail;
Dave Chinner7bf7f352012-11-12 22:09:45 +1100499 xfs_finish_ioend(ioend);
500 continue;
501 }
502
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100503 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100504
505 if (!bio) {
506 retry:
507 bio = xfs_alloc_ioend_bio(bh);
508 } else if (bh->b_blocknr != lastblock + 1) {
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000509 xfs_submit_ioend_bio(wbc, ioend, bio);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100510 goto retry;
511 }
512
Zhi Yong Wuc7c1a7d2013-08-07 10:11:09 +0000513 if (xfs_bio_add_buffer(bio, bh) != bh->b_size) {
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000514 xfs_submit_ioend_bio(wbc, ioend, bio);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100515 goto retry;
516 }
517
518 lastblock = bh->b_blocknr;
519 }
520 if (bio)
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000521 xfs_submit_ioend_bio(wbc, ioend, bio);
Christoph Hellwig209fb872010-07-18 21:17:11 +0000522 xfs_finish_ioend(ioend);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100523 } while ((ioend = next) != NULL);
524}
525
526/*
527 * Cancel submission of all buffer_heads so far in this endio.
528 * Toss the endio too. Only ever called for the initial page
529 * in a writepage request, so only ever one page.
530 */
531STATIC void
532xfs_cancel_ioend(
533 xfs_ioend_t *ioend)
534{
535 xfs_ioend_t *next;
536 struct buffer_head *bh, *next_bh;
537
538 do {
539 next = ioend->io_list;
540 bh = ioend->io_buffer_head;
541 do {
542 next_bh = bh->b_private;
543 clear_buffer_async_write(bh);
Brian Foster07d08682014-10-02 09:42:06 +1000544 /*
545 * The unwritten flag is cleared when added to the
546 * ioend. We're not submitting for I/O so mark the
547 * buffer unwritten again for next time around.
548 */
549 if (ioend->io_type == XFS_IO_UNWRITTEN)
550 set_buffer_unwritten(bh);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100551 unlock_buffer(bh);
552 } while ((bh = next_bh) != NULL);
553
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100554 mempool_free(ioend, xfs_ioend_pool);
555 } while ((ioend = next) != NULL);
556}
557
558/*
559 * Test to see if we've been building up a completion structure for
560 * earlier buffers -- if so, we try to append to this ioend if we
561 * can, otherwise we finish off any current ioend and start another.
562 * Return true if we've finished the given ioend.
563 */
564STATIC void
565xfs_add_to_ioend(
566 struct inode *inode,
567 struct buffer_head *bh,
Christoph Hellwig7336cea2006-01-11 20:49:16 +1100568 xfs_off_t offset,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100569 unsigned int type,
570 xfs_ioend_t **result,
571 int need_ioend)
572{
573 xfs_ioend_t *ioend = *result;
574
575 if (!ioend || need_ioend || type != ioend->io_type) {
576 xfs_ioend_t *previous = *result;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100577
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100578 ioend = xfs_alloc_ioend(inode, type);
579 ioend->io_offset = offset;
580 ioend->io_buffer_head = bh;
581 ioend->io_buffer_tail = bh;
582 if (previous)
583 previous->io_list = ioend;
584 *result = ioend;
585 } else {
586 ioend->io_buffer_tail->b_private = bh;
587 ioend->io_buffer_tail = bh;
588 }
589
590 bh->b_private = NULL;
591 ioend->io_size += bh->b_size;
592}
593
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594STATIC void
Nathan Scott87cbc492006-03-14 13:26:43 +1100595xfs_map_buffer(
Christoph Hellwig046f1682010-04-28 12:28:52 +0000596 struct inode *inode,
Nathan Scott87cbc492006-03-14 13:26:43 +1100597 struct buffer_head *bh,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000598 struct xfs_bmbt_irec *imap,
Christoph Hellwig046f1682010-04-28 12:28:52 +0000599 xfs_off_t offset)
Nathan Scott87cbc492006-03-14 13:26:43 +1100600{
601 sector_t bn;
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000602 struct xfs_mount *m = XFS_I(inode)->i_mount;
Christoph Hellwig207d0412010-04-28 12:28:56 +0000603 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
604 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
Nathan Scott87cbc492006-03-14 13:26:43 +1100605
Christoph Hellwig207d0412010-04-28 12:28:56 +0000606 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
607 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
Nathan Scott87cbc492006-03-14 13:26:43 +1100608
Christoph Hellwige5131822010-04-28 12:28:55 +0000609 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000610 ((offset - iomap_offset) >> inode->i_blkbits);
Nathan Scott87cbc492006-03-14 13:26:43 +1100611
Christoph Hellwig046f1682010-04-28 12:28:52 +0000612 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
Nathan Scott87cbc492006-03-14 13:26:43 +1100613
614 bh->b_blocknr = bn;
615 set_buffer_mapped(bh);
616}
617
618STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619xfs_map_at_offset(
Christoph Hellwig046f1682010-04-28 12:28:52 +0000620 struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 struct buffer_head *bh,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000622 struct xfs_bmbt_irec *imap,
Christoph Hellwig046f1682010-04-28 12:28:52 +0000623 xfs_off_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624{
Christoph Hellwig207d0412010-04-28 12:28:56 +0000625 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
626 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
Christoph Hellwig207d0412010-04-28 12:28:56 +0000628 xfs_map_buffer(inode, bh, imap, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 set_buffer_mapped(bh);
630 clear_buffer_delay(bh);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100631 clear_buffer_unwritten(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632}
633
634/*
Dave Chinnera49935f2014-03-07 16:19:14 +1100635 * Test if a given page contains at least one buffer of a given @type.
636 * If @check_all_buffers is true, then we walk all the buffers in the page to
637 * try to find one of the type passed in. If it is not set, then the caller only
638 * needs to check the first buffer on the page for a match.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 */
Dave Chinnera49935f2014-03-07 16:19:14 +1100640STATIC bool
Dave Chinner6ffc4db2012-04-23 15:58:43 +1000641xfs_check_page_type(
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100642 struct page *page,
Dave Chinnera49935f2014-03-07 16:19:14 +1100643 unsigned int type,
644 bool check_all_buffers)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645{
Dave Chinnera49935f2014-03-07 16:19:14 +1100646 struct buffer_head *bh;
647 struct buffer_head *head;
648
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 if (PageWriteback(page))
Dave Chinnera49935f2014-03-07 16:19:14 +1100650 return false;
651 if (!page->mapping)
652 return false;
653 if (!page_has_buffers(page))
654 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655
Dave Chinnera49935f2014-03-07 16:19:14 +1100656 bh = head = page_buffers(page);
657 do {
658 if (buffer_unwritten(bh)) {
659 if (type == XFS_IO_UNWRITTEN)
660 return true;
661 } else if (buffer_delay(bh)) {
Dan Carpenter805eeb82014-04-04 06:56:30 +1100662 if (type == XFS_IO_DELALLOC)
Dave Chinnera49935f2014-03-07 16:19:14 +1100663 return true;
664 } else if (buffer_dirty(bh) && buffer_mapped(bh)) {
Dan Carpenter805eeb82014-04-04 06:56:30 +1100665 if (type == XFS_IO_OVERWRITE)
Dave Chinnera49935f2014-03-07 16:19:14 +1100666 return true;
667 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668
Dave Chinnera49935f2014-03-07 16:19:14 +1100669 /* If we are only checking the first buffer, we are done now. */
670 if (!check_all_buffers)
671 break;
672 } while ((bh = bh->b_this_page) != head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673
Dave Chinnera49935f2014-03-07 16:19:14 +1100674 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675}
676
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677/*
678 * Allocate & map buffers for page given the extent map. Write it out.
679 * except for the original page of a writepage, this is called on
680 * delalloc/unwritten pages only, for the original page it is possible
681 * that the page has no mapping at all.
682 */
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100683STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684xfs_convert_page(
685 struct inode *inode,
686 struct page *page,
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100687 loff_t tindex,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000688 struct xfs_bmbt_irec *imap,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100689 xfs_ioend_t **ioendp,
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000690 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691{
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100692 struct buffer_head *bh, *head;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100693 xfs_off_t end_offset;
694 unsigned long p_offset;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100695 unsigned int type;
Nathan Scott24e17b52005-05-05 13:33:20 -0700696 int len, page_dirty;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100697 int count = 0, done = 0, uptodate = 1;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100698 xfs_off_t offset = page_offset(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700699
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100700 if (page->index != tindex)
701 goto fail;
Nick Piggin529ae9a2008-08-02 12:01:03 +0200702 if (!trylock_page(page))
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100703 goto fail;
704 if (PageWriteback(page))
705 goto fail_unlock_page;
706 if (page->mapping != inode->i_mapping)
707 goto fail_unlock_page;
Dave Chinnera49935f2014-03-07 16:19:14 +1100708 if (!xfs_check_page_type(page, (*ioendp)->io_type, false))
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100709 goto fail_unlock_page;
710
Nathan Scott24e17b52005-05-05 13:33:20 -0700711 /*
712 * page_dirty is initially a count of buffers on the page before
Nathan Scottc41564b2006-03-29 08:55:14 +1000713 * EOF and is decremented as we move each into a cleanable state.
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100714 *
715 * Derivation:
716 *
717 * End offset is the highest offset that this page should represent.
718 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
719 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
720 * hence give us the correct page_dirty count. On any other page,
721 * it will be zero and in that case we need page_dirty to be the
722 * count of buffers on the page.
Nathan Scott24e17b52005-05-05 13:33:20 -0700723 */
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100724 end_offset = min_t(unsigned long long,
725 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
726 i_size_read(inode));
727
Dave Chinner480d7462013-05-20 09:51:08 +1000728 /*
729 * If the current map does not span the entire page we are about to try
730 * to write, then give up. The only way we can write a page that spans
731 * multiple mappings in a single writeback iteration is via the
732 * xfs_vm_writepage() function. Data integrity writeback requires the
733 * entire page to be written in a single attempt, otherwise the part of
734 * the page we don't write here doesn't get written as part of the data
735 * integrity sync.
736 *
737 * For normal writeback, we also don't attempt to write partial pages
738 * here as it simply means that write_cache_pages() will see it under
739 * writeback and ignore the page until some point in the future, at
740 * which time this will be the only page in the file that needs
741 * writeback. Hence for more optimal IO patterns, we should always
742 * avoid partial page writeback due to multiple mappings on a page here.
743 */
744 if (!xfs_imap_valid(inode, imap, end_offset))
745 goto fail_unlock_page;
746
Nathan Scott24e17b52005-05-05 13:33:20 -0700747 len = 1 << inode->i_blkbits;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100748 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
749 PAGE_CACHE_SIZE);
750 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
751 page_dirty = p_offset / len;
Nathan Scott24e17b52005-05-05 13:33:20 -0700752
Dave Chinnera49935f2014-03-07 16:19:14 +1100753 /*
754 * The moment we find a buffer that doesn't match our current type
755 * specification or can't be written, abort the loop and start
756 * writeback. As per the above xfs_imap_valid() check, only
757 * xfs_vm_writepage() can handle partial page writeback fully - we are
758 * limited here to the buffers that are contiguous with the current
759 * ioend, and hence a buffer we can't write breaks that contiguity and
760 * we have to defer the rest of the IO to xfs_vm_writepage().
761 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 bh = head = page_buffers(page);
763 do {
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100764 if (offset >= end_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 break;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100766 if (!buffer_uptodate(bh))
767 uptodate = 0;
768 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
769 done = 1;
Dave Chinnera49935f2014-03-07 16:19:14 +1100770 break;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100771 }
772
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000773 if (buffer_unwritten(bh) || buffer_delay(bh) ||
774 buffer_mapped(bh)) {
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100775 if (buffer_unwritten(bh))
Alain Renaud0d882a32012-05-22 15:56:21 -0500776 type = XFS_IO_UNWRITTEN;
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000777 else if (buffer_delay(bh))
Alain Renaud0d882a32012-05-22 15:56:21 -0500778 type = XFS_IO_DELALLOC;
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000779 else
Alain Renaud0d882a32012-05-22 15:56:21 -0500780 type = XFS_IO_OVERWRITE;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100781
Dave Chinnera49935f2014-03-07 16:19:14 +1100782 /*
783 * imap should always be valid because of the above
784 * partial page end_offset check on the imap.
785 */
786 ASSERT(xfs_imap_valid(inode, imap, offset));
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100787
Christoph Hellwigecff71e2010-12-10 08:42:25 +0000788 lock_buffer(bh);
Alain Renaud0d882a32012-05-22 15:56:21 -0500789 if (type != XFS_IO_OVERWRITE)
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000790 xfs_map_at_offset(inode, bh, imap, offset);
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000791 xfs_add_to_ioend(inode, bh, offset, type,
792 ioendp, done);
793
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100794 page_dirty--;
795 count++;
796 } else {
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000797 done = 1;
Dave Chinnera49935f2014-03-07 16:19:14 +1100798 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799 }
Christoph Hellwig7336cea2006-01-11 20:49:16 +1100800 } while (offset += len, (bh = bh->b_this_page) != head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100802 if (uptodate && bh == head)
803 SetPageUptodate(page);
804
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000805 if (count) {
Dave Chinnerefceab12010-08-24 11:44:56 +1000806 if (--wbc->nr_to_write <= 0 &&
807 wbc->sync_mode == WB_SYNC_NONE)
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000808 done = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809 }
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000810 xfs_start_page_writeback(page, !page_dirty, count);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100811
812 return done;
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100813 fail_unlock_page:
814 unlock_page(page);
815 fail:
816 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817}
818
819/*
820 * Convert & write out a cluster of pages in the same extent as defined
821 * by mp and following the start page.
822 */
823STATIC void
824xfs_cluster_write(
825 struct inode *inode,
826 pgoff_t tindex,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000827 struct xfs_bmbt_irec *imap,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100828 xfs_ioend_t **ioendp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 struct writeback_control *wbc,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 pgoff_t tlast)
831{
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100832 struct pagevec pvec;
833 int done = 0, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100835 pagevec_init(&pvec, 0);
836 while (!done && tindex <= tlast) {
837 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
838
839 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 break;
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100841
842 for (i = 0; i < pagevec_count(&pvec); i++) {
843 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000844 imap, ioendp, wbc);
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100845 if (done)
846 break;
847 }
848
849 pagevec_release(&pvec);
850 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700851 }
852}
853
Dave Chinner3ed3a432010-03-05 02:00:42 +0000854STATIC void
855xfs_vm_invalidatepage(
856 struct page *page,
Lukas Czernerd47992f2013-05-21 23:17:23 -0400857 unsigned int offset,
858 unsigned int length)
Dave Chinner3ed3a432010-03-05 02:00:42 +0000859{
Lukas Czerner34097df2013-05-21 23:58:01 -0400860 trace_xfs_invalidatepage(page->mapping->host, page, offset,
861 length);
862 block_invalidatepage(page, offset, length);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000863}
864
865/*
866 * If the page has delalloc buffers on it, we need to punch them out before we
867 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
868 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
869 * is done on that same region - the delalloc extent is returned when none is
870 * supposed to be there.
871 *
872 * We prevent this by truncating away the delalloc regions on the page before
873 * invalidating it. Because they are delalloc, we can do this without needing a
874 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
875 * truncation without a transaction as there is no space left for block
876 * reservation (typically why we see a ENOSPC in writeback).
877 *
878 * This is not a performance critical path, so for now just do the punching a
879 * buffer head at a time.
880 */
881STATIC void
882xfs_aops_discard_page(
883 struct page *page)
884{
885 struct inode *inode = page->mapping->host;
886 struct xfs_inode *ip = XFS_I(inode);
887 struct buffer_head *bh, *head;
888 loff_t offset = page_offset(page);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000889
Dave Chinnera49935f2014-03-07 16:19:14 +1100890 if (!xfs_check_page_type(page, XFS_IO_DELALLOC, true))
Dave Chinner3ed3a432010-03-05 02:00:42 +0000891 goto out_invalidate;
892
Dave Chinnere8c37532010-03-15 02:36:35 +0000893 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
894 goto out_invalidate;
895
Dave Chinner4f107002011-03-07 10:00:35 +1100896 xfs_alert(ip->i_mount,
Dave Chinner3ed3a432010-03-05 02:00:42 +0000897 "page discard on page %p, inode 0x%llx, offset %llu.",
898 page, ip->i_ino, offset);
899
900 xfs_ilock(ip, XFS_ILOCK_EXCL);
901 bh = head = page_buffers(page);
902 do {
Dave Chinner3ed3a432010-03-05 02:00:42 +0000903 int error;
Dave Chinnerc726de42010-11-30 15:14:39 +1100904 xfs_fileoff_t start_fsb;
Dave Chinner3ed3a432010-03-05 02:00:42 +0000905
906 if (!buffer_delay(bh))
907 goto next_buffer;
908
Dave Chinnerc726de42010-11-30 15:14:39 +1100909 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
910 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000911 if (error) {
912 /* something screwed, just bail */
Dave Chinnere8c37532010-03-15 02:36:35 +0000913 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
Dave Chinner4f107002011-03-07 10:00:35 +1100914 xfs_alert(ip->i_mount,
Dave Chinner3ed3a432010-03-05 02:00:42 +0000915 "page discard unable to remove delalloc mapping.");
Dave Chinnere8c37532010-03-15 02:36:35 +0000916 }
Dave Chinner3ed3a432010-03-05 02:00:42 +0000917 break;
918 }
919next_buffer:
Dave Chinnerc726de42010-11-30 15:14:39 +1100920 offset += 1 << inode->i_blkbits;
Dave Chinner3ed3a432010-03-05 02:00:42 +0000921
922 } while ((bh = bh->b_this_page) != head);
923
924 xfs_iunlock(ip, XFS_ILOCK_EXCL);
925out_invalidate:
Lukas Czernerd47992f2013-05-21 23:17:23 -0400926 xfs_vm_invalidatepage(page, 0, PAGE_CACHE_SIZE);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000927 return;
928}
929
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930/*
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000931 * Write out a dirty page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932 *
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000933 * For delalloc space on the page we need to allocate space and flush it.
934 * For unwritten space on the page we need to start the conversion to
935 * regular allocated space.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000936 * For any other dirty buffer heads on the page we should flush them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938STATIC int
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000939xfs_vm_writepage(
940 struct page *page,
941 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942{
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000943 struct inode *inode = page->mapping->host;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100944 struct buffer_head *bh, *head;
Christoph Hellwig207d0412010-04-28 12:28:56 +0000945 struct xfs_bmbt_irec imap;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100946 xfs_ioend_t *ioend = NULL, *iohead = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947 loff_t offset;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100948 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949 __uint64_t end_offset;
Christoph Hellwigbd1556a2010-04-28 12:29:00 +0000950 pgoff_t end_index, last_index;
Christoph Hellwiged1e7b72010-12-10 08:42:22 +0000951 ssize_t len;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000952 int err, imap_valid = 0, uptodate = 1;
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000953 int count = 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000954 int nonblocking = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955
Lukas Czerner34097df2013-05-21 23:58:01 -0400956 trace_xfs_writepage(inode, page, 0, 0);
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000957
Christoph Hellwig20cb52e2010-06-24 09:46:01 +1000958 ASSERT(page_has_buffers(page));
959
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000960 /*
961 * Refuse to write the page out if we are called from reclaim context.
962 *
Christoph Hellwigd4f7a5c2010-06-28 10:34:44 -0400963 * This avoids stack overflows when called from deeply used stacks in
964 * random callers for direct reclaim or memcg reclaim. We explicitly
965 * allow reclaim from kswapd as the stack usage there is relatively low.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000966 *
Mel Gorman94054fa2011-10-31 17:07:45 -0700967 * This should never happen except in the case of a VM regression so
968 * warn about it.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000969 */
Mel Gorman94054fa2011-10-31 17:07:45 -0700970 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
971 PF_MEMALLOC))
Christoph Hellwigb5420f22010-08-24 11:47:51 +1000972 goto redirty;
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000973
974 /*
Christoph Hellwig680a6472011-07-08 14:34:05 +0200975 * Given that we do not allow direct reclaim to call us, we should
976 * never be called while in a filesystem transaction.
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000977 */
Christoph Hellwig448011e2014-06-06 16:05:15 +1000978 if (WARN_ON_ONCE(current->flags & PF_FSTRANS))
Christoph Hellwigb5420f22010-08-24 11:47:51 +1000979 goto redirty;
Christoph Hellwig89f3b3632010-06-24 09:45:48 +1000980
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981 /* Is this page beyond the end of the file? */
982 offset = i_size_read(inode);
983 end_index = offset >> PAGE_CACHE_SHIFT;
984 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
Jie Liu8695d272014-05-20 08:24:26 +1000985
986 /*
987 * The page index is less than the end_index, adjust the end_offset
988 * to the highest offset that this page should represent.
989 * -----------------------------------------------------
990 * | file mapping | <EOF> |
991 * -----------------------------------------------------
992 * | Page ... | Page N-2 | Page N-1 | Page N | |
993 * ^--------------------------------^----------|--------
994 * | desired writeback range | see else |
995 * ---------------------------------^------------------|
996 */
997 if (page->index < end_index)
998 end_offset = (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT;
999 else {
1000 /*
1001 * Check whether the page to write out is beyond or straddles
1002 * i_size or not.
1003 * -------------------------------------------------------
1004 * | file mapping | <EOF> |
1005 * -------------------------------------------------------
1006 * | Page ... | Page N-2 | Page N-1 | Page N | Beyond |
1007 * ^--------------------------------^-----------|---------
1008 * | | Straddles |
1009 * ---------------------------------^-----------|--------|
1010 */
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -04001011 unsigned offset_into_page = offset & (PAGE_CACHE_SIZE - 1);
1012
1013 /*
Jan Karaff9a28f2013-03-14 14:30:54 +01001014 * Skip the page if it is fully outside i_size, e.g. due to a
1015 * truncate operation that is in progress. We must redirty the
1016 * page so that reclaim stops reclaiming it. Otherwise
1017 * xfs_vm_releasepage() is called on it and gets confused.
Jie Liu8695d272014-05-20 08:24:26 +10001018 *
1019 * Note that the end_index is unsigned long, it would overflow
1020 * if the given offset is greater than 16TB on 32-bit system
1021 * and if we do check the page is fully outside i_size or not
1022 * via "if (page->index >= end_index + 1)" as "end_index + 1"
1023 * will be evaluated to 0. Hence this page will be redirtied
1024 * and be written out repeatedly which would result in an
1025 * infinite loop, the user program that perform this operation
1026 * will hang. Instead, we can verify this situation by checking
1027 * if the page to write is totally beyond the i_size or if it's
1028 * offset is just equal to the EOF.
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -04001029 */
Jie Liu8695d272014-05-20 08:24:26 +10001030 if (page->index > end_index ||
1031 (page->index == end_index && offset_into_page == 0))
Jan Karaff9a28f2013-03-14 14:30:54 +01001032 goto redirty;
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -04001033
1034 /*
1035 * The page straddles i_size. It must be zeroed out on each
1036 * and every writepage invocation because it may be mmapped.
1037 * "A file is mapped in multiples of the page size. For a file
Jie Liu8695d272014-05-20 08:24:26 +10001038 * that is not a multiple of the page size, the remaining
Christoph Hellwig6b7a03f2012-07-03 12:20:00 -04001039 * memory is zeroed when mapped, and writes to that region are
1040 * not written out to the file."
1041 */
1042 zero_user_segment(page, offset_into_page, PAGE_CACHE_SIZE);
Jie Liu8695d272014-05-20 08:24:26 +10001043
1044 /* Adjust the end_offset to the end of file */
1045 end_offset = offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 }
1047
Nathan Scott24e17b52005-05-05 13:33:20 -07001048 len = 1 << inode->i_blkbits;
Nathan Scott24e17b52005-05-05 13:33:20 -07001049
Nathan Scott24e17b52005-05-05 13:33:20 -07001050 bh = head = page_buffers(page);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001051 offset = page_offset(page);
Alain Renaud0d882a32012-05-22 15:56:21 -05001052 type = XFS_IO_OVERWRITE;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001053
Christoph Hellwigdbcdde32011-07-08 14:34:14 +02001054 if (wbc->sync_mode == WB_SYNC_NONE)
Christoph Hellwiga206c812010-12-10 08:42:20 +00001055 nonblocking = 1;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001056
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 do {
Christoph Hellwig6ac72482010-12-10 08:42:18 +00001058 int new_ioend = 0;
1059
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 if (offset >= end_offset)
1061 break;
1062 if (!buffer_uptodate(bh))
1063 uptodate = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064
Eric Sandeen3d9b02e2010-06-24 09:45:30 +10001065 /*
Christoph Hellwigece413f2010-11-10 21:39:11 +00001066 * set_page_dirty dirties all buffers in a page, independent
1067 * of their state. The dirty state however is entirely
1068 * meaningless for holes (!mapped && uptodate), so skip
1069 * buffers covering holes here.
Eric Sandeen3d9b02e2010-06-24 09:45:30 +10001070 */
1071 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
Eric Sandeen3d9b02e2010-06-24 09:45:30 +10001072 imap_valid = 0;
1073 continue;
1074 }
1075
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001076 if (buffer_unwritten(bh)) {
Alain Renaud0d882a32012-05-22 15:56:21 -05001077 if (type != XFS_IO_UNWRITTEN) {
1078 type = XFS_IO_UNWRITTEN;
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001079 imap_valid = 0;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001080 }
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001081 } else if (buffer_delay(bh)) {
Alain Renaud0d882a32012-05-22 15:56:21 -05001082 if (type != XFS_IO_DELALLOC) {
1083 type = XFS_IO_DELALLOC;
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001084 imap_valid = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 }
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001086 } else if (buffer_uptodate(bh)) {
Alain Renaud0d882a32012-05-22 15:56:21 -05001087 if (type != XFS_IO_OVERWRITE) {
1088 type = XFS_IO_OVERWRITE;
Christoph Hellwig85da94c2010-12-10 08:42:16 +00001089 imap_valid = 0;
1090 }
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001091 } else {
Alain Renaud7d0fa3e2012-06-08 15:34:46 -04001092 if (PageUptodate(page))
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001093 ASSERT(buffer_mapped(bh));
Alain Renaud7d0fa3e2012-06-08 15:34:46 -04001094 /*
1095 * This buffer is not uptodate and will not be
1096 * written to disk. Ensure that we will put any
1097 * subsequent writeable buffers into a new
1098 * ioend.
1099 */
1100 imap_valid = 0;
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001101 continue;
1102 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001104 if (imap_valid)
1105 imap_valid = xfs_imap_valid(inode, &imap, offset);
1106 if (!imap_valid) {
1107 /*
1108 * If we didn't have a valid mapping then we need to
1109 * put the new mapping into a separate ioend structure.
1110 * This ensures non-contiguous extents always have
1111 * separate ioends, which is particularly important
1112 * for unwritten extent conversion at I/O completion
1113 * time.
1114 */
1115 new_ioend = 1;
1116 err = xfs_map_blocks(inode, offset, &imap, type,
1117 nonblocking);
1118 if (err)
1119 goto error;
1120 imap_valid = xfs_imap_valid(inode, &imap, offset);
1121 }
1122 if (imap_valid) {
Christoph Hellwigecff71e2010-12-10 08:42:25 +00001123 lock_buffer(bh);
Alain Renaud0d882a32012-05-22 15:56:21 -05001124 if (type != XFS_IO_OVERWRITE)
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001125 xfs_map_at_offset(inode, bh, &imap, offset);
1126 xfs_add_to_ioend(inode, bh, offset, type, &ioend,
1127 new_ioend);
1128 count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 }
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001130
1131 if (!iohead)
1132 iohead = ioend;
1133
1134 } while (offset += len, ((bh = bh->b_this_page) != head));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135
1136 if (uptodate && bh == head)
1137 SetPageUptodate(page);
1138
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001139 xfs_start_page_writeback(page, 1, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140
Dave Chinner7bf7f352012-11-12 22:09:45 +11001141 /* if there is no IO to be submitted for this page, we are done */
1142 if (!ioend)
1143 return 0;
1144
1145 ASSERT(iohead);
1146
1147 /*
1148 * Any errors from this point onwards need tobe reported through the IO
1149 * completion path as we have marked the initial page as under writeback
1150 * and unlocked it.
1151 */
1152 if (imap_valid) {
Christoph Hellwigbd1556a2010-04-28 12:29:00 +00001153 xfs_off_t end_index;
Christoph Hellwig8699bb02010-04-28 12:28:54 +00001154
Christoph Hellwigbd1556a2010-04-28 12:29:00 +00001155 end_index = imap.br_startoff + imap.br_blockcount;
1156
1157 /* to bytes */
1158 end_index <<= inode->i_blkbits;
1159
1160 /* to pages */
1161 end_index = (end_index - 1) >> PAGE_CACHE_SHIFT;
1162
1163 /* check against file size */
1164 if (end_index > last_index)
1165 end_index = last_index;
1166
Christoph Hellwig207d0412010-04-28 12:28:56 +00001167 xfs_cluster_write(inode, page->index + 1, &imap, &ioend,
Christoph Hellwig2fa24f92010-12-10 08:42:23 +00001168 wbc, end_index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 }
1170
Christoph Hellwig281627d2012-03-13 08:41:05 +00001171
Dave Chinner7bf7f352012-11-12 22:09:45 +11001172 /*
1173 * Reserve log space if we might write beyond the on-disk inode size.
1174 */
1175 err = 0;
1176 if (ioend->io_type != XFS_IO_UNWRITTEN && xfs_ioend_is_append(ioend))
1177 err = xfs_setfilesize_trans_alloc(ioend);
1178
1179 xfs_submit_ioend(wbc, iohead, err);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001180
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001181 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182
1183error:
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001184 if (iohead)
1185 xfs_cancel_ioend(iohead);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186
Christoph Hellwigb5420f22010-08-24 11:47:51 +10001187 if (err == -EAGAIN)
1188 goto redirty;
1189
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001190 xfs_aops_discard_page(page);
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001191 ClearPageUptodate(page);
1192 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 return err;
Nathan Scottf51623b2006-03-14 13:26:27 +11001194
Christoph Hellwigb5420f22010-08-24 11:47:51 +10001195redirty:
Nathan Scottf51623b2006-03-14 13:26:27 +11001196 redirty_page_for_writepage(wbc, page);
1197 unlock_page(page);
1198 return 0;
Nathan Scottf51623b2006-03-14 13:26:27 +11001199}
1200
Nathan Scott7d4fb402006-06-09 15:27:16 +10001201STATIC int
1202xfs_vm_writepages(
1203 struct address_space *mapping,
1204 struct writeback_control *wbc)
1205{
Christoph Hellwigb3aea4e2007-08-29 11:44:37 +10001206 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
Nathan Scott7d4fb402006-06-09 15:27:16 +10001207 return generic_writepages(mapping, wbc);
1208}
1209
Nathan Scottf51623b2006-03-14 13:26:27 +11001210/*
1211 * Called to move a page into cleanable state - and from there
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001212 * to be released. The page should already be clean. We always
Nathan Scottf51623b2006-03-14 13:26:27 +11001213 * have buffer heads in this call.
1214 *
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001215 * Returns 1 if the page is ok to release, 0 otherwise.
Nathan Scottf51623b2006-03-14 13:26:27 +11001216 */
1217STATIC int
Nathan Scott238f4c52006-03-17 17:26:25 +11001218xfs_vm_releasepage(
Nathan Scottf51623b2006-03-14 13:26:27 +11001219 struct page *page,
1220 gfp_t gfp_mask)
1221{
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001222 int delalloc, unwritten;
Nathan Scottf51623b2006-03-14 13:26:27 +11001223
Lukas Czerner34097df2013-05-21 23:58:01 -04001224 trace_xfs_releasepage(page->mapping->host, page, 0, 0);
Nathan Scott238f4c52006-03-17 17:26:25 +11001225
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001226 xfs_count_page_state(page, &delalloc, &unwritten);
Nathan Scottf51623b2006-03-14 13:26:27 +11001227
Christoph Hellwig448011e2014-06-06 16:05:15 +10001228 if (WARN_ON_ONCE(delalloc))
Christoph Hellwig89f3b3632010-06-24 09:45:48 +10001229 return 0;
Christoph Hellwig448011e2014-06-06 16:05:15 +10001230 if (WARN_ON_ONCE(unwritten))
Nathan Scottf51623b2006-03-14 13:26:27 +11001231 return 0;
1232
Nathan Scottf51623b2006-03-14 13:26:27 +11001233 return try_to_free_buffers(page);
1234}
1235
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236STATIC int
Nathan Scottc2536662006-03-29 10:44:40 +10001237__xfs_get_blocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238 struct inode *inode,
1239 sector_t iblock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 struct buffer_head *bh_result,
1241 int create,
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001242 int direct)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001243{
Christoph Hellwiga206c812010-12-10 08:42:20 +00001244 struct xfs_inode *ip = XFS_I(inode);
1245 struct xfs_mount *mp = ip->i_mount;
1246 xfs_fileoff_t offset_fsb, end_fsb;
1247 int error = 0;
1248 int lockmode = 0;
Christoph Hellwig207d0412010-04-28 12:28:56 +00001249 struct xfs_bmbt_irec imap;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001250 int nimaps = 1;
Nathan Scottfdc7ed72005-11-02 15:13:13 +11001251 xfs_off_t offset;
1252 ssize_t size;
Christoph Hellwig207d0412010-04-28 12:28:56 +00001253 int new = 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001254
1255 if (XFS_FORCED_SHUTDOWN(mp))
Eric Sandeenb474c7a2014-06-22 15:04:54 +10001256 return -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001257
Nathan Scottfdc7ed72005-11-02 15:13:13 +11001258 offset = (xfs_off_t)iblock << inode->i_blkbits;
Nathan Scottc2536662006-03-29 10:44:40 +10001259 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1260 size = bh_result->b_size;
Lachlan McIlroy364f358a2008-09-17 16:50:14 +10001261
1262 if (!create && direct && offset >= i_size_read(inode))
1263 return 0;
1264
Dave Chinner507630b2012-03-27 10:34:50 -04001265 /*
1266 * Direct I/O is usually done on preallocated files, so try getting
1267 * a block mapping without an exclusive lock first. For buffered
1268 * writes we already have the exclusive iolock anyway, so avoiding
1269 * a lock roundtrip here by taking the ilock exclusive from the
1270 * beginning is a useful micro optimization.
1271 */
1272 if (create && !direct) {
Christoph Hellwiga206c812010-12-10 08:42:20 +00001273 lockmode = XFS_ILOCK_EXCL;
1274 xfs_ilock(ip, lockmode);
1275 } else {
Christoph Hellwig309ecac82013-12-06 12:30:09 -08001276 lockmode = xfs_ilock_data_map_shared(ip);
Christoph Hellwiga206c812010-12-10 08:42:20 +00001277 }
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001278
Dave Chinnerd2c28192012-06-08 15:44:53 +10001279 ASSERT(offset <= mp->m_super->s_maxbytes);
1280 if (offset + size > mp->m_super->s_maxbytes)
1281 size = mp->m_super->s_maxbytes - offset;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001282 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1283 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1284
Dave Chinner5c8ed202011-09-18 20:40:45 +00001285 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
1286 &imap, &nimaps, XFS_BMAPI_ENTIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 if (error)
Christoph Hellwiga206c812010-12-10 08:42:20 +00001288 goto out_unlock;
1289
1290 if (create &&
1291 (!nimaps ||
1292 (imap.br_startblock == HOLESTARTBLOCK ||
1293 imap.br_startblock == DELAYSTARTBLOCK))) {
Dave Chinneraff3a9e2012-04-23 15:58:44 +10001294 if (direct || xfs_get_extsz_hint(ip)) {
Dave Chinner507630b2012-03-27 10:34:50 -04001295 /*
1296 * Drop the ilock in preparation for starting the block
1297 * allocation transaction. It will be retaken
1298 * exclusively inside xfs_iomap_write_direct for the
1299 * actual allocation.
1300 */
1301 xfs_iunlock(ip, lockmode);
Christoph Hellwiga206c812010-12-10 08:42:20 +00001302 error = xfs_iomap_write_direct(ip, offset, size,
1303 &imap, nimaps);
Dave Chinner507630b2012-03-27 10:34:50 -04001304 if (error)
Dave Chinner24513372014-06-25 14:58:08 +10001305 return error;
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001306 new = 1;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001307 } else {
Dave Chinner507630b2012-03-27 10:34:50 -04001308 /*
1309 * Delalloc reservations do not require a transaction,
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001310 * we can go on without dropping the lock here. If we
1311 * are allocating a new delalloc block, make sure that
1312 * we set the new flag so that we mark the buffer new so
1313 * that we know that it is newly allocated if the write
1314 * fails.
Dave Chinner507630b2012-03-27 10:34:50 -04001315 */
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001316 if (nimaps && imap.br_startblock == HOLESTARTBLOCK)
1317 new = 1;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001318 error = xfs_iomap_write_delay(ip, offset, size, &imap);
Dave Chinner507630b2012-03-27 10:34:50 -04001319 if (error)
1320 goto out_unlock;
1321
1322 xfs_iunlock(ip, lockmode);
Christoph Hellwiga206c812010-12-10 08:42:20 +00001323 }
Christoph Hellwiga206c812010-12-10 08:42:20 +00001324
1325 trace_xfs_get_blocks_alloc(ip, offset, size, 0, &imap);
1326 } else if (nimaps) {
1327 trace_xfs_get_blocks_found(ip, offset, size, 0, &imap);
Dave Chinner507630b2012-03-27 10:34:50 -04001328 xfs_iunlock(ip, lockmode);
Christoph Hellwiga206c812010-12-10 08:42:20 +00001329 } else {
1330 trace_xfs_get_blocks_notfound(ip, offset, size);
1331 goto out_unlock;
1332 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333
Christoph Hellwig207d0412010-04-28 12:28:56 +00001334 if (imap.br_startblock != HOLESTARTBLOCK &&
1335 imap.br_startblock != DELAYSTARTBLOCK) {
Nathan Scott87cbc492006-03-14 13:26:43 +11001336 /*
1337 * For unwritten extents do not report a disk address on
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 * the read case (treat as if we're reading into a hole).
1339 */
Christoph Hellwig207d0412010-04-28 12:28:56 +00001340 if (create || !ISUNWRITTEN(&imap))
1341 xfs_map_buffer(inode, bh_result, &imap, offset);
1342 if (create && ISUNWRITTEN(&imap)) {
Christoph Hellwig7b7a8662013-09-04 15:04:39 +02001343 if (direct) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 bh_result->b_private = inode;
Christoph Hellwig7b7a8662013-09-04 15:04:39 +02001345 set_buffer_defer_completion(bh_result);
1346 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 set_buffer_unwritten(bh_result);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348 }
1349 }
1350
Nathan Scottc2536662006-03-29 10:44:40 +10001351 /*
1352 * If this is a realtime file, data may be on a different device.
1353 * to that pointed to from the buffer_head b_bdev currently.
1354 */
Christoph Hellwig046f1682010-04-28 12:28:52 +00001355 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356
Nathan Scottc2536662006-03-29 10:44:40 +10001357 /*
David Chinner549054a2007-02-10 18:36:35 +11001358 * If we previously allocated a block out beyond eof and we are now
1359 * coming back to use it then we will need to flag it as new even if it
1360 * has a disk address.
1361 *
1362 * With sub-block writes into unwritten extents we also need to mark
1363 * the buffer as new so that the unwritten parts of the buffer gets
1364 * correctly zeroed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 */
1366 if (create &&
1367 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
David Chinner549054a2007-02-10 18:36:35 +11001368 (offset >= i_size_read(inode)) ||
Christoph Hellwig207d0412010-04-28 12:28:56 +00001369 (new || ISUNWRITTEN(&imap))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 set_buffer_new(bh_result);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371
Christoph Hellwig207d0412010-04-28 12:28:56 +00001372 if (imap.br_startblock == DELAYSTARTBLOCK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 BUG_ON(direct);
1374 if (create) {
1375 set_buffer_uptodate(bh_result);
1376 set_buffer_mapped(bh_result);
1377 set_buffer_delay(bh_result);
1378 }
1379 }
1380
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001381 /*
1382 * If this is O_DIRECT or the mpage code calling tell them how large
1383 * the mapping is, so that we can avoid repeated get_blocks calls.
Dave Chinner0e1f7892014-04-17 08:15:19 +10001384 *
1385 * If the mapping spans EOF, then we have to break the mapping up as the
1386 * mapping for blocks beyond EOF must be marked new so that sub block
1387 * regions can be correctly zeroed. We can't do this for mappings within
1388 * EOF unless the mapping was just allocated or is unwritten, otherwise
1389 * the callers would overwrite existing data with zeros. Hence we have
1390 * to split the mapping into a range up to and including EOF, and a
1391 * second mapping for beyond EOF.
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001392 */
Nathan Scottc2536662006-03-29 10:44:40 +10001393 if (direct || size > (1 << inode->i_blkbits)) {
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001394 xfs_off_t mapping_size;
Christoph Hellwig9563b3d2010-04-28 12:28:53 +00001395
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001396 mapping_size = imap.br_startoff + imap.br_blockcount - iblock;
1397 mapping_size <<= inode->i_blkbits;
1398
1399 ASSERT(mapping_size > 0);
1400 if (mapping_size > size)
1401 mapping_size = size;
Dave Chinner0e1f7892014-04-17 08:15:19 +10001402 if (offset < i_size_read(inode) &&
1403 offset + mapping_size >= i_size_read(inode)) {
1404 /* limit mapping to block that spans EOF */
1405 mapping_size = roundup_64(i_size_read(inode) - offset,
1406 1 << inode->i_blkbits);
1407 }
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001408 if (mapping_size > LONG_MAX)
1409 mapping_size = LONG_MAX;
1410
1411 bh_result->b_size = mapping_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 }
1413
1414 return 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001415
1416out_unlock:
1417 xfs_iunlock(ip, lockmode);
Dave Chinner24513372014-06-25 14:58:08 +10001418 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419}
1420
1421int
Nathan Scottc2536662006-03-29 10:44:40 +10001422xfs_get_blocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 struct inode *inode,
1424 sector_t iblock,
1425 struct buffer_head *bh_result,
1426 int create)
1427{
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001428 return __xfs_get_blocks(inode, iblock, bh_result, create, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429}
1430
1431STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001432xfs_get_blocks_direct(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 struct inode *inode,
1434 sector_t iblock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435 struct buffer_head *bh_result,
1436 int create)
1437{
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001438 return __xfs_get_blocks(inode, iblock, bh_result, create, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439}
1440
Christoph Hellwig209fb872010-07-18 21:17:11 +00001441/*
1442 * Complete a direct I/O write request.
1443 *
1444 * If the private argument is non-NULL __xfs_get_blocks signals us that we
1445 * need to issue a transaction to convert the range from unwritten to written
Christoph Hellwig2ba66232015-02-02 10:02:09 +11001446 * extents.
Christoph Hellwig209fb872010-07-18 21:17:11 +00001447 */
Christoph Hellwigf0973862005-09-05 08:22:52 +10001448STATIC void
Christoph Hellwig209fb872010-07-18 21:17:11 +00001449xfs_end_io_direct_write(
1450 struct kiocb *iocb,
1451 loff_t offset,
1452 ssize_t size,
Christoph Hellwig7b7a8662013-09-04 15:04:39 +02001453 void *private)
Christoph Hellwigf0973862005-09-05 08:22:52 +10001454{
Christoph Hellwig2ba66232015-02-02 10:02:09 +11001455 struct inode *inode = file_inode(iocb->ki_filp);
1456 struct xfs_inode *ip = XFS_I(inode);
1457 struct xfs_mount *mp = ip->i_mount;
1458
1459 if (XFS_FORCED_SHUTDOWN(mp))
1460 return;
Christoph Hellwigf0973862005-09-05 08:22:52 +10001461
1462 /*
Christoph Hellwig2813d682011-12-18 20:00:12 +00001463 * While the generic direct I/O code updates the inode size, it does
1464 * so only after the end_io handler is called, which means our
1465 * end_io handler thinks the on-disk size is outside the in-core
1466 * size. To prevent this just update it a little bit earlier here.
1467 */
Christoph Hellwig2ba66232015-02-02 10:02:09 +11001468 if (offset + size > i_size_read(inode))
1469 i_size_write(inode, offset + size);
Christoph Hellwig2813d682011-12-18 20:00:12 +00001470
1471 /*
Christoph Hellwig2ba66232015-02-02 10:02:09 +11001472 * For direct I/O we do not know if we need to allocate blocks or not,
1473 * so we can't preallocate an append transaction, as that results in
1474 * nested reservations and log space deadlocks. Hence allocate the
1475 * transaction here. While this is sub-optimal and can block IO
1476 * completion for some time, we're stuck with doing it this way until
1477 * we can pass the ioend to the direct IO allocation callbacks and
1478 * avoid nesting that way.
Christoph Hellwigf0973862005-09-05 08:22:52 +10001479 */
Christoph Hellwig2ba66232015-02-02 10:02:09 +11001480 if (private && size > 0) {
1481 xfs_iomap_write_unwritten(ip, offset, size);
1482 } else if (offset + size > ip->i_d.di_size) {
1483 struct xfs_trans *tp;
1484 int error;
Christoph Hellwig40e2e972010-07-18 21:17:09 +00001485
Christoph Hellwig2ba66232015-02-02 10:02:09 +11001486 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
1487 error = xfs_trans_reserve(tp, &M_RES(mp)->tr_fsyncts, 0, 0);
1488 if (error) {
1489 xfs_trans_cancel(tp, 0);
1490 return;
1491 }
Christoph Hellwig209fb872010-07-18 21:17:11 +00001492
Christoph Hellwig2ba66232015-02-02 10:02:09 +11001493 xfs_setfilesize(ip, tp, offset, size);
1494 }
Christoph Hellwigf0973862005-09-05 08:22:52 +10001495}
1496
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497STATIC ssize_t
Nathan Scotte4c573b2006-03-14 13:54:26 +11001498xfs_vm_direct_IO(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 int rw,
1500 struct kiocb *iocb,
Al Virod8d3d942014-03-04 21:27:34 -05001501 struct iov_iter *iter,
1502 loff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503{
Christoph Hellwig209fb872010-07-18 21:17:11 +00001504 struct inode *inode = iocb->ki_filp->f_mapping->host;
1505 struct block_device *bdev = xfs_find_bdev_for_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001506
Christoph Hellwig209fb872010-07-18 21:17:11 +00001507 if (rw & WRITE) {
Christoph Hellwig2ba66232015-02-02 10:02:09 +11001508 return __blockdev_direct_IO(rw, iocb, inode, bdev, iter,
Al Viro31b14032014-03-05 01:33:16 -05001509 offset, xfs_get_blocks_direct,
Christoph Hellwig9862f622014-02-10 10:28:04 +11001510 xfs_end_io_direct_write, NULL,
1511 DIO_ASYNC_EXTEND);
Christoph Hellwig209fb872010-07-18 21:17:11 +00001512 }
Christoph Hellwig2ba66232015-02-02 10:02:09 +11001513 return __blockdev_direct_IO(rw, iocb, inode, bdev, iter,
1514 offset, xfs_get_blocks_direct,
1515 NULL, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516}
1517
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001518/*
1519 * Punch out the delalloc blocks we have already allocated.
1520 *
1521 * Don't bother with xfs_setattr given that nothing can have made it to disk yet
1522 * as the page is still locked at this point.
1523 */
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001524STATIC void
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001525xfs_vm_kill_delalloc_range(
1526 struct inode *inode,
1527 loff_t start,
1528 loff_t end)
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001529{
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001530 struct xfs_inode *ip = XFS_I(inode);
1531 xfs_fileoff_t start_fsb;
1532 xfs_fileoff_t end_fsb;
1533 int error;
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001534
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001535 start_fsb = XFS_B_TO_FSB(ip->i_mount, start);
1536 end_fsb = XFS_B_TO_FSB(ip->i_mount, end);
1537 if (end_fsb <= start_fsb)
1538 return;
Dave Chinnerc726de42010-11-30 15:14:39 +11001539
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001540 xfs_ilock(ip, XFS_ILOCK_EXCL);
1541 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1542 end_fsb - start_fsb);
1543 if (error) {
1544 /* something screwed, just bail */
1545 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
1546 xfs_alert(ip->i_mount,
1547 "xfs_vm_write_failed: unable to clean up ino %lld",
1548 ip->i_ino);
Dave Chinnerc726de42010-11-30 15:14:39 +11001549 }
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001550 }
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001551 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001552}
1553
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001554STATIC void
1555xfs_vm_write_failed(
1556 struct inode *inode,
1557 struct page *page,
1558 loff_t pos,
1559 unsigned len)
1560{
Jie Liu58e59852013-07-16 13:11:16 +08001561 loff_t block_offset;
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001562 loff_t block_start;
1563 loff_t block_end;
1564 loff_t from = pos & (PAGE_CACHE_SIZE - 1);
1565 loff_t to = from + len;
1566 struct buffer_head *bh, *head;
1567
Jie Liu58e59852013-07-16 13:11:16 +08001568 /*
1569 * The request pos offset might be 32 or 64 bit, this is all fine
1570 * on 64-bit platform. However, for 64-bit pos request on 32-bit
1571 * platform, the high 32-bit will be masked off if we evaluate the
1572 * block_offset via (pos & PAGE_MASK) because the PAGE_MASK is
1573 * 0xfffff000 as an unsigned long, hence the result is incorrect
1574 * which could cause the following ASSERT failed in most cases.
1575 * In order to avoid this, we can evaluate the block_offset of the
1576 * start of the page by using shifts rather than masks the mismatch
1577 * problem.
1578 */
1579 block_offset = (pos >> PAGE_CACHE_SHIFT) << PAGE_CACHE_SHIFT;
1580
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001581 ASSERT(block_offset + from == pos);
1582
1583 head = page_buffers(page);
1584 block_start = 0;
1585 for (bh = head; bh != head || !block_start;
1586 bh = bh->b_this_page, block_start = block_end,
1587 block_offset += bh->b_size) {
1588 block_end = block_start + bh->b_size;
1589
1590 /* skip buffers before the write */
1591 if (block_end <= from)
1592 continue;
1593
1594 /* if the buffer is after the write, we're done */
1595 if (block_start >= to)
1596 break;
1597
1598 if (!buffer_delay(bh))
1599 continue;
1600
1601 if (!buffer_new(bh) && block_offset < i_size_read(inode))
1602 continue;
1603
1604 xfs_vm_kill_delalloc_range(inode, block_offset,
1605 block_offset + bh->b_size);
Dave Chinner4ab9ed52014-04-14 18:11:58 +10001606
1607 /*
1608 * This buffer does not contain data anymore. make sure anyone
1609 * who finds it knows that for certain.
1610 */
1611 clear_buffer_delay(bh);
1612 clear_buffer_uptodate(bh);
1613 clear_buffer_mapped(bh);
1614 clear_buffer_new(bh);
1615 clear_buffer_dirty(bh);
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001616 }
1617
1618}
1619
1620/*
1621 * This used to call block_write_begin(), but it unlocks and releases the page
1622 * on error, and we need that page to be able to punch stale delalloc blocks out
1623 * on failure. hence we copy-n-waste it here and call xfs_vm_write_failed() at
1624 * the appropriate point.
1625 */
Nathan Scottf51623b2006-03-14 13:26:27 +11001626STATIC int
Nick Piggind79689c2007-10-16 01:25:06 -07001627xfs_vm_write_begin(
Nathan Scottf51623b2006-03-14 13:26:27 +11001628 struct file *file,
Nick Piggind79689c2007-10-16 01:25:06 -07001629 struct address_space *mapping,
1630 loff_t pos,
1631 unsigned len,
1632 unsigned flags,
1633 struct page **pagep,
1634 void **fsdata)
Nathan Scottf51623b2006-03-14 13:26:27 +11001635{
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001636 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1637 struct page *page;
1638 int status;
Christoph Hellwig155130a2010-06-04 11:29:58 +02001639
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001640 ASSERT(len <= PAGE_CACHE_SIZE);
1641
Dave Chinnerad22c7a2013-10-29 22:11:57 +11001642 page = grab_cache_page_write_begin(mapping, index, flags);
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001643 if (!page)
1644 return -ENOMEM;
1645
1646 status = __block_write_begin(page, pos, len, xfs_get_blocks);
1647 if (unlikely(status)) {
1648 struct inode *inode = mapping->host;
Dave Chinner72ab70a2014-04-14 18:13:29 +10001649 size_t isize = i_size_read(inode);
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001650
1651 xfs_vm_write_failed(inode, page, pos, len);
1652 unlock_page(page);
1653
Dave Chinner72ab70a2014-04-14 18:13:29 +10001654 /*
1655 * If the write is beyond EOF, we only want to kill blocks
1656 * allocated in this write, not blocks that were previously
1657 * written successfully.
1658 */
1659 if (pos + len > isize) {
1660 ssize_t start = max_t(ssize_t, pos, isize);
1661
1662 truncate_pagecache_range(inode, start, pos + len);
1663 }
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001664
1665 page_cache_release(page);
1666 page = NULL;
1667 }
1668
1669 *pagep = page;
1670 return status;
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001671}
Christoph Hellwig155130a2010-06-04 11:29:58 +02001672
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001673/*
Dave Chinneraad3f372014-04-14 18:14:11 +10001674 * On failure, we only need to kill delalloc blocks beyond EOF in the range of
1675 * this specific write because they will never be written. Previous writes
1676 * beyond EOF where block allocation succeeded do not need to be trashed, so
1677 * only new blocks from this write should be trashed. For blocks within
1678 * EOF, generic_write_end() zeros them so they are safe to leave alone and be
1679 * written with all the other valid data.
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001680 */
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001681STATIC int
1682xfs_vm_write_end(
1683 struct file *file,
1684 struct address_space *mapping,
1685 loff_t pos,
1686 unsigned len,
1687 unsigned copied,
1688 struct page *page,
1689 void *fsdata)
1690{
1691 int ret;
1692
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001693 ASSERT(len <= PAGE_CACHE_SIZE);
1694
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001695 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001696 if (unlikely(ret < len)) {
1697 struct inode *inode = mapping->host;
1698 size_t isize = i_size_read(inode);
1699 loff_t to = pos + len;
1700
1701 if (to > isize) {
Dave Chinneraad3f372014-04-14 18:14:11 +10001702 /* only kill blocks in this write beyond EOF */
1703 if (pos > isize)
1704 isize = pos;
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001705 xfs_vm_kill_delalloc_range(inode, isize, to);
Dave Chinneraad3f372014-04-14 18:14:11 +10001706 truncate_pagecache_range(inode, isize, to);
Dave Chinnerd3bc8152012-04-27 19:45:21 +10001707 }
1708 }
Christoph Hellwig155130a2010-06-04 11:29:58 +02001709 return ret;
Nathan Scottf51623b2006-03-14 13:26:27 +11001710}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711
1712STATIC sector_t
Nathan Scotte4c573b2006-03-14 13:54:26 +11001713xfs_vm_bmap(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 struct address_space *mapping,
1715 sector_t block)
1716{
1717 struct inode *inode = (struct inode *)mapping->host;
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001718 struct xfs_inode *ip = XFS_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10001720 trace_xfs_vm_bmap(XFS_I(inode));
Christoph Hellwig126468b2008-03-06 13:44:57 +11001721 xfs_ilock(ip, XFS_IOLOCK_SHARED);
Dave Chinner4bc1ea62012-11-12 22:53:56 +11001722 filemap_write_and_wait(mapping);
Christoph Hellwig126468b2008-03-06 13:44:57 +11001723 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
Nathan Scottc2536662006-03-29 10:44:40 +10001724 return generic_block_bmap(mapping, block, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725}
1726
1727STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001728xfs_vm_readpage(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 struct file *unused,
1730 struct page *page)
1731{
Nathan Scottc2536662006-03-29 10:44:40 +10001732 return mpage_readpage(page, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733}
1734
1735STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001736xfs_vm_readpages(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737 struct file *unused,
1738 struct address_space *mapping,
1739 struct list_head *pages,
1740 unsigned nr_pages)
1741{
Nathan Scottc2536662006-03-29 10:44:40 +10001742 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743}
1744
Dave Chinner22e757a2014-09-02 12:12:51 +10001745/*
1746 * This is basically a copy of __set_page_dirty_buffers() with one
1747 * small tweak: buffers beyond EOF do not get marked dirty. If we mark them
1748 * dirty, we'll never be able to clean them because we don't write buffers
1749 * beyond EOF, and that means we can't invalidate pages that span EOF
1750 * that have been marked dirty. Further, the dirty state can leak into
1751 * the file interior if the file is extended, resulting in all sorts of
1752 * bad things happening as the state does not match the underlying data.
1753 *
1754 * XXX: this really indicates that bufferheads in XFS need to die. Warts like
1755 * this only exist because of bufferheads and how the generic code manages them.
1756 */
1757STATIC int
1758xfs_vm_set_page_dirty(
1759 struct page *page)
1760{
1761 struct address_space *mapping = page->mapping;
1762 struct inode *inode = mapping->host;
1763 loff_t end_offset;
1764 loff_t offset;
1765 int newly_dirty;
1766
1767 if (unlikely(!mapping))
1768 return !TestSetPageDirty(page);
1769
1770 end_offset = i_size_read(inode);
1771 offset = page_offset(page);
1772
1773 spin_lock(&mapping->private_lock);
1774 if (page_has_buffers(page)) {
1775 struct buffer_head *head = page_buffers(page);
1776 struct buffer_head *bh = head;
1777
1778 do {
1779 if (offset < end_offset)
1780 set_buffer_dirty(bh);
1781 bh = bh->b_this_page;
1782 offset += 1 << inode->i_blkbits;
1783 } while (bh != head);
1784 }
1785 newly_dirty = !TestSetPageDirty(page);
1786 spin_unlock(&mapping->private_lock);
1787
1788 if (newly_dirty) {
1789 /* sigh - __set_page_dirty() is static, so copy it here, too */
1790 unsigned long flags;
1791
1792 spin_lock_irqsave(&mapping->tree_lock, flags);
1793 if (page->mapping) { /* Race with truncate? */
1794 WARN_ON_ONCE(!PageUptodate(page));
1795 account_page_dirtied(page, mapping);
1796 radix_tree_tag_set(&mapping->page_tree,
1797 page_index(page), PAGECACHE_TAG_DIRTY);
1798 }
1799 spin_unlock_irqrestore(&mapping->tree_lock, flags);
1800 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1801 }
1802 return newly_dirty;
1803}
1804
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001805const struct address_space_operations xfs_address_space_operations = {
Nathan Scotte4c573b2006-03-14 13:54:26 +11001806 .readpage = xfs_vm_readpage,
1807 .readpages = xfs_vm_readpages,
1808 .writepage = xfs_vm_writepage,
Nathan Scott7d4fb402006-06-09 15:27:16 +10001809 .writepages = xfs_vm_writepages,
Dave Chinner22e757a2014-09-02 12:12:51 +10001810 .set_page_dirty = xfs_vm_set_page_dirty,
Nathan Scott238f4c52006-03-17 17:26:25 +11001811 .releasepage = xfs_vm_releasepage,
1812 .invalidatepage = xfs_vm_invalidatepage,
Nick Piggind79689c2007-10-16 01:25:06 -07001813 .write_begin = xfs_vm_write_begin,
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001814 .write_end = xfs_vm_write_end,
Nathan Scotte4c573b2006-03-14 13:54:26 +11001815 .bmap = xfs_vm_bmap,
1816 .direct_IO = xfs_vm_direct_IO,
Christoph Lametere965f962006-02-01 03:05:41 -08001817 .migratepage = buffer_migrate_page,
Hisashi Hifumibddaafa2009-03-29 09:53:38 +02001818 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02001819 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820};