blob: 0dbb9e70fe21664740bc90721f4ae7bb5d3d190b [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"
Nathan Scotta844f452005-11-02 14:38:42 +110019#include "xfs_bit.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include "xfs_log.h"
Nathan Scotta844f452005-11-02 14:38:42 +110021#include "xfs_inum.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include "xfs_sb.h"
Nathan Scotta844f452005-11-02 14:38:42 +110023#include "xfs_ag.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include "xfs_trans.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include "xfs_mount.h"
26#include "xfs_bmap_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include "xfs_dinode.h"
28#include "xfs_inode.h"
Christoph Hellwig281627d2012-03-13 08:41:05 +000029#include "xfs_inode_item.h"
Nathan Scotta844f452005-11-02 14:38:42 +110030#include "xfs_alloc.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xfs_error.h"
32#include "xfs_rw.h"
33#include "xfs_iomap.h"
Christoph Hellwig739bfb22007-08-29 10:58:01 +100034#include "xfs_vnodeops.h"
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000035#include "xfs_trace.h"
Dave Chinner3ed3a432010-03-05 02:00:42 +000036#include "xfs_bmap.h"
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090037#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <linux/mpage.h>
Christoph Hellwig10ce4442006-01-11 20:48:14 +110039#include <linux/pagevec.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include <linux/writeback.h>
41
Christoph Hellwig0b1b2132009-12-14 23:14:59 +000042void
Nathan Scottf51623b2006-03-14 13:26:27 +110043xfs_count_page_state(
44 struct page *page,
45 int *delalloc,
Nathan Scottf51623b2006-03-14 13:26:27 +110046 int *unwritten)
47{
48 struct buffer_head *bh, *head;
49
Christoph Hellwig20cb52e2010-06-24 09:46:01 +100050 *delalloc = *unwritten = 0;
Nathan Scottf51623b2006-03-14 13:26:27 +110051
52 bh = head = page_buffers(page);
53 do {
Christoph Hellwig20cb52e2010-06-24 09:46:01 +100054 if (buffer_unwritten(bh))
Nathan Scottf51623b2006-03-14 13:26:27 +110055 (*unwritten) = 1;
56 else if (buffer_delay(bh))
57 (*delalloc) = 1;
58 } while ((bh = bh->b_this_page) != head);
59}
60
Christoph Hellwig6214ed42007-09-14 15:23:17 +100061STATIC struct block_device *
62xfs_find_bdev_for_inode(
Christoph Hellwig046f1682010-04-28 12:28:52 +000063 struct inode *inode)
Christoph Hellwig6214ed42007-09-14 15:23:17 +100064{
Christoph Hellwig046f1682010-04-28 12:28:52 +000065 struct xfs_inode *ip = XFS_I(inode);
Christoph Hellwig6214ed42007-09-14 15:23:17 +100066 struct xfs_mount *mp = ip->i_mount;
67
Eric Sandeen71ddabb2007-11-23 16:29:42 +110068 if (XFS_IS_REALTIME_INODE(ip))
Christoph Hellwig6214ed42007-09-14 15:23:17 +100069 return mp->m_rtdev_targp->bt_bdev;
70 else
71 return mp->m_ddev_targp->bt_bdev;
72}
73
Christoph Hellwig0829c362005-09-02 16:58:49 +100074/*
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +110075 * We're now finished for good with this ioend structure.
76 * Update the page state via the associated buffer_heads,
77 * release holds on the inode and bio, and finally free
78 * up memory. Do not use the ioend after this.
79 */
Christoph Hellwig0829c362005-09-02 16:58:49 +100080STATIC void
81xfs_destroy_ioend(
82 xfs_ioend_t *ioend)
83{
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +110084 struct buffer_head *bh, *next;
85
86 for (bh = ioend->io_buffer_head; bh; bh = next) {
87 next = bh->b_private;
Nathan Scott7d04a332006-06-09 14:58:38 +100088 bh->b_end_io(bh, !ioend->io_error);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +110089 }
Christoph Hellwig583fa582008-12-03 12:20:38 +010090
Christoph Hellwigc859cdd2011-08-23 08:28:10 +000091 if (ioend->io_iocb) {
Christoph Hellwig04f658e2011-08-24 05:59:25 +000092 if (ioend->io_isasync) {
93 aio_complete(ioend->io_iocb, ioend->io_error ?
94 ioend->io_error : ioend->io_result, 0);
95 }
Christoph Hellwigc859cdd2011-08-23 08:28:10 +000096 inode_dio_done(ioend->io_inode);
97 }
Christoph Hellwig4a06fd22011-08-23 08:28:13 +000098
Christoph Hellwig0829c362005-09-02 16:58:49 +100099 mempool_free(ioend, xfs_ioend_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100}
101
102/*
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000103 * Fast and loose check if this write could update the on-disk inode size.
104 */
105static inline bool xfs_ioend_is_append(struct xfs_ioend *ioend)
106{
107 return ioend->io_offset + ioend->io_size >
108 XFS_I(ioend->io_inode)->i_d.di_size;
109}
110
Christoph Hellwig281627d2012-03-13 08:41:05 +0000111STATIC int
112xfs_setfilesize_trans_alloc(
113 struct xfs_ioend *ioend)
114{
115 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
116 struct xfs_trans *tp;
117 int error;
118
119 tp = xfs_trans_alloc(mp, XFS_TRANS_FSYNC_TS);
120
121 error = xfs_trans_reserve(tp, 0, XFS_FSYNC_TS_LOG_RES(mp), 0, 0, 0);
122 if (error) {
123 xfs_trans_cancel(tp, 0);
124 return error;
125 }
126
127 ioend->io_append_trans = tp;
128
129 /*
130 * We hand off the transaction to the completion thread now, so
131 * clear the flag here.
132 */
133 current_restore_flags_nested(&tp->t_pflags, PF_FSTRANS);
134 return 0;
135}
136
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000137/*
Christoph Hellwig2813d682011-12-18 20:00:12 +0000138 * Update on-disk file size now that data has been written to disk.
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000139 */
Christoph Hellwig281627d2012-03-13 08:41:05 +0000140STATIC int
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000141xfs_setfilesize(
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000142 struct xfs_ioend *ioend)
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000143{
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000144 struct xfs_inode *ip = XFS_I(ioend->io_inode);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000145 struct xfs_trans *tp = ioend->io_append_trans;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000146 xfs_fsize_t isize;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000147
Christoph Hellwig281627d2012-03-13 08:41:05 +0000148 /*
149 * The transaction was allocated in the I/O submission thread,
150 * thus we need to mark ourselves as beeing in a transaction
151 * manually.
152 */
153 current_set_flags_nested(&tp->t_pflags, PF_FSTRANS);
154
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000155 xfs_ilock(ip, XFS_ILOCK_EXCL);
Christoph Hellwig6923e682012-02-29 09:53:49 +0000156 isize = xfs_new_eof(ip, ioend->io_offset + ioend->io_size);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000157 if (!isize) {
158 xfs_iunlock(ip, XFS_ILOCK_EXCL);
159 xfs_trans_cancel(tp, 0);
160 return 0;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000161 }
162
Christoph Hellwig281627d2012-03-13 08:41:05 +0000163 trace_xfs_setfilesize(ip, ioend->io_offset, ioend->io_size);
164
165 ip->i_d.di_size = isize;
166 xfs_trans_ijoin(tp, ip, XFS_ILOCK_EXCL);
167 xfs_trans_log_inode(tp, ip, XFS_ILOG_CORE);
168
169 return xfs_trans_commit(tp, 0);
Christoph Hellwig0829c362005-09-02 16:58:49 +1000170}
171
172/*
Christoph Hellwig209fb872010-07-18 21:17:11 +0000173 * Schedule IO completion handling on the final put of an ioend.
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000174 *
175 * If there is no work to do we might as well call it a day and free the
176 * ioend right now.
Dave Chinnerc626d172009-04-06 18:42:11 +0200177 */
178STATIC void
179xfs_finish_ioend(
Christoph Hellwig209fb872010-07-18 21:17:11 +0000180 struct xfs_ioend *ioend)
Dave Chinnerc626d172009-04-06 18:42:11 +0200181{
182 if (atomic_dec_and_test(&ioend->io_remaining)) {
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000183 struct xfs_mount *mp = XFS_I(ioend->io_inode)->i_mount;
184
Christoph Hellwig209fb872010-07-18 21:17:11 +0000185 if (ioend->io_type == IO_UNWRITTEN)
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000186 queue_work(mp->m_unwritten_workqueue, &ioend->io_work);
Christoph Hellwig281627d2012-03-13 08:41:05 +0000187 else if (ioend->io_append_trans)
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000188 queue_work(mp->m_data_workqueue, &ioend->io_work);
Christoph Hellwigfc0063c2011-08-23 08:28:11 +0000189 else
190 xfs_destroy_ioend(ioend);
Dave Chinnerc626d172009-04-06 18:42:11 +0200191 }
192}
193
194/*
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000195 * IO write completion.
196 */
197STATIC void
198xfs_end_io(
199 struct work_struct *work)
200{
201 xfs_ioend_t *ioend = container_of(work, xfs_ioend_t, io_work);
202 struct xfs_inode *ip = XFS_I(ioend->io_inode);
Dave Chinner69418932010-03-04 00:57:09 +0000203 int error = 0;
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000204
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000205 if (XFS_FORCED_SHUTDOWN(ip->i_mount)) {
Christoph Hellwig810627d2011-11-08 08:56:15 +0000206 ioend->io_error = -EIO;
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000207 goto done;
208 }
209 if (ioend->io_error)
210 goto done;
211
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000212 /*
213 * For unwritten extents we need to issue transactions to convert a
214 * range to normal written extens after the data I/O has finished.
215 */
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000216 if (ioend->io_type == IO_UNWRITTEN) {
Christoph Hellwig281627d2012-03-13 08:41:05 +0000217 /*
218 * For buffered I/O we never preallocate a transaction when
219 * doing the unwritten extent conversion, but for direct I/O
220 * we do not know if we are converting an unwritten extent
221 * or not at the point where we preallocate the transaction.
222 */
223 if (ioend->io_append_trans) {
224 ASSERT(ioend->io_isdirect);
225
226 current_set_flags_nested(
227 &ioend->io_append_trans->t_pflags, PF_FSTRANS);
228 xfs_trans_cancel(ioend->io_append_trans, 0);
229 }
230
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000231 error = xfs_iomap_write_unwritten(ip, ioend->io_offset,
232 ioend->io_size);
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000233 if (error) {
234 ioend->io_error = -error;
235 goto done;
236 }
Christoph Hellwig281627d2012-03-13 08:41:05 +0000237 } else if (ioend->io_append_trans) {
238 error = xfs_setfilesize(ioend);
239 if (error)
240 ioend->io_error = -error;
Christoph Hellwig84803fb2012-02-29 09:53:50 +0000241 } else {
Christoph Hellwig281627d2012-03-13 08:41:05 +0000242 ASSERT(!xfs_ioend_is_append(ioend));
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000243 }
244
Christoph Hellwig04f658e2011-08-24 05:59:25 +0000245done:
Christoph Hellwigaa6bf012012-02-29 09:53:48 +0000246 xfs_destroy_ioend(ioend);
Dave Chinner77d7a0c2010-02-17 05:36:29 +0000247}
248
249/*
Christoph Hellwig209fb872010-07-18 21:17:11 +0000250 * Call IO completion handling in caller context on the final put of an ioend.
251 */
252STATIC void
253xfs_finish_ioend_sync(
254 struct xfs_ioend *ioend)
255{
256 if (atomic_dec_and_test(&ioend->io_remaining))
257 xfs_end_io(&ioend->io_work);
258}
259
260/*
Christoph Hellwig0829c362005-09-02 16:58:49 +1000261 * Allocate and initialise an IO completion structure.
262 * We need to track unwritten extent write completion here initially.
263 * We'll need to extend this for updating the ondisk inode size later
264 * (vs. incore size).
265 */
266STATIC xfs_ioend_t *
267xfs_alloc_ioend(
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100268 struct inode *inode,
269 unsigned int type)
Christoph Hellwig0829c362005-09-02 16:58:49 +1000270{
271 xfs_ioend_t *ioend;
272
273 ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
274
275 /*
276 * Set the count to 1 initially, which will prevent an I/O
277 * completion callback from happening before we have started
278 * all the I/O from calling the completion routine too early.
279 */
280 atomic_set(&ioend->io_remaining, 1);
Christoph Hellwigc859cdd2011-08-23 08:28:10 +0000281 ioend->io_isasync = 0;
Christoph Hellwig281627d2012-03-13 08:41:05 +0000282 ioend->io_isdirect = 0;
Nathan Scott7d04a332006-06-09 14:58:38 +1000283 ioend->io_error = 0;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100284 ioend->io_list = NULL;
285 ioend->io_type = type;
Christoph Hellwigb677c212007-08-29 11:46:28 +1000286 ioend->io_inode = inode;
Christoph Hellwigc1a073b2005-09-05 08:23:35 +1000287 ioend->io_buffer_head = NULL;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100288 ioend->io_buffer_tail = NULL;
Christoph Hellwig0829c362005-09-02 16:58:49 +1000289 ioend->io_offset = 0;
290 ioend->io_size = 0;
Christoph Hellwigfb511f22010-07-18 21:17:10 +0000291 ioend->io_iocb = NULL;
292 ioend->io_result = 0;
Christoph Hellwig281627d2012-03-13 08:41:05 +0000293 ioend->io_append_trans = NULL;
Christoph Hellwig0829c362005-09-02 16:58:49 +1000294
Christoph Hellwig5ec4fab2009-10-30 09:11:47 +0000295 INIT_WORK(&ioend->io_work, xfs_end_io);
Christoph Hellwig0829c362005-09-02 16:58:49 +1000296 return ioend;
297}
298
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299STATIC int
300xfs_map_blocks(
301 struct inode *inode,
302 loff_t offset,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000303 struct xfs_bmbt_irec *imap,
Christoph Hellwiga206c812010-12-10 08:42:20 +0000304 int type,
305 int nonblocking)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306{
Christoph Hellwiga206c812010-12-10 08:42:20 +0000307 struct xfs_inode *ip = XFS_I(inode);
308 struct xfs_mount *mp = ip->i_mount;
Christoph Hellwiged1e7b72010-12-10 08:42:22 +0000309 ssize_t count = 1 << inode->i_blkbits;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000310 xfs_fileoff_t offset_fsb, end_fsb;
311 int error = 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000312 int bmapi_flags = XFS_BMAPI_ENTIRE;
313 int nimaps = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314
Christoph Hellwiga206c812010-12-10 08:42:20 +0000315 if (XFS_FORCED_SHUTDOWN(mp))
316 return -XFS_ERROR(EIO);
317
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000318 if (type == IO_UNWRITTEN)
Christoph Hellwiga206c812010-12-10 08:42:20 +0000319 bmapi_flags |= XFS_BMAPI_IGSTATE;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000320
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000321 if (!xfs_ilock_nowait(ip, XFS_ILOCK_SHARED)) {
322 if (nonblocking)
323 return -XFS_ERROR(EAGAIN);
324 xfs_ilock(ip, XFS_ILOCK_SHARED);
Christoph Hellwiga206c812010-12-10 08:42:20 +0000325 }
326
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000327 ASSERT(ip->i_d.di_format != XFS_DINODE_FMT_BTREE ||
328 (ip->i_df.if_flags & XFS_IFEXTENTS));
Christoph Hellwiga206c812010-12-10 08:42:20 +0000329 ASSERT(offset <= mp->m_maxioffset);
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000330
Christoph Hellwiga206c812010-12-10 08:42:20 +0000331 if (offset + count > mp->m_maxioffset)
332 count = mp->m_maxioffset - offset;
333 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + count);
334 offset_fsb = XFS_B_TO_FSBT(mp, offset);
Dave Chinner5c8ed202011-09-18 20:40:45 +0000335 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
336 imap, &nimaps, bmapi_flags);
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000337 xfs_iunlock(ip, XFS_ILOCK_SHARED);
338
Christoph Hellwiga206c812010-12-10 08:42:20 +0000339 if (error)
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000340 return -XFS_ERROR(error);
Christoph Hellwiga206c812010-12-10 08:42:20 +0000341
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000342 if (type == IO_DELALLOC &&
343 (!nimaps || isnullstartblock(imap->br_startblock))) {
Christoph Hellwiga206c812010-12-10 08:42:20 +0000344 error = xfs_iomap_write_allocate(ip, offset, count, imap);
345 if (!error)
346 trace_xfs_map_blocks_alloc(ip, offset, count, type, imap);
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000347 return -XFS_ERROR(error);
Christoph Hellwiga206c812010-12-10 08:42:20 +0000348 }
349
Christoph Hellwig8ff29572010-12-10 08:42:21 +0000350#ifdef DEBUG
351 if (type == IO_UNWRITTEN) {
352 ASSERT(nimaps);
353 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
354 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
355 }
356#endif
357 if (nimaps)
358 trace_xfs_map_blocks_found(ip, offset, count, type, imap);
359 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360}
361
Christoph Hellwigb8f82a42009-11-14 16:17:22 +0000362STATIC int
Christoph Hellwig558e6892010-04-28 12:28:58 +0000363xfs_imap_valid(
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000364 struct inode *inode,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000365 struct xfs_bmbt_irec *imap,
Christoph Hellwig558e6892010-04-28 12:28:58 +0000366 xfs_off_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367{
Christoph Hellwig558e6892010-04-28 12:28:58 +0000368 offset >>= inode->i_blkbits;
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000369
Christoph Hellwig558e6892010-04-28 12:28:58 +0000370 return offset >= imap->br_startoff &&
371 offset < imap->br_startoff + imap->br_blockcount;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372}
373
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100374/*
375 * BIO completion handler for buffered IO.
376 */
Al Viro782e3b32007-10-12 07:17:47 +0100377STATIC void
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100378xfs_end_bio(
379 struct bio *bio,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100380 int error)
381{
382 xfs_ioend_t *ioend = bio->bi_private;
383
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100384 ASSERT(atomic_read(&bio->bi_cnt) >= 1);
Nathan Scott7d04a332006-06-09 14:58:38 +1000385 ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100386
387 /* Toss bio and pass work off to an xfsdatad thread */
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100388 bio->bi_private = NULL;
389 bio->bi_end_io = NULL;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100390 bio_put(bio);
Nathan Scott7d04a332006-06-09 14:58:38 +1000391
Christoph Hellwig209fb872010-07-18 21:17:11 +0000392 xfs_finish_ioend(ioend);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100393}
394
395STATIC void
396xfs_submit_ioend_bio(
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000397 struct writeback_control *wbc,
398 xfs_ioend_t *ioend,
399 struct bio *bio)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100400{
401 atomic_inc(&ioend->io_remaining);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100402 bio->bi_private = ioend;
403 bio->bi_end_io = xfs_end_bio;
Jens Axboe721a9602011-03-09 11:56:30 +0100404 submit_bio(wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE, bio);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100405}
406
407STATIC struct bio *
408xfs_alloc_ioend_bio(
409 struct buffer_head *bh)
410{
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100411 int nvecs = bio_get_nr_vecs(bh->b_bdev);
Christoph Hellwig221cb252010-12-10 08:42:17 +0000412 struct bio *bio = bio_alloc(GFP_NOIO, nvecs);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100413
414 ASSERT(bio->bi_private == NULL);
415 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
416 bio->bi_bdev = bh->b_bdev;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100417 return bio;
418}
419
420STATIC void
421xfs_start_buffer_writeback(
422 struct buffer_head *bh)
423{
424 ASSERT(buffer_mapped(bh));
425 ASSERT(buffer_locked(bh));
426 ASSERT(!buffer_delay(bh));
427 ASSERT(!buffer_unwritten(bh));
428
429 mark_buffer_async_write(bh);
430 set_buffer_uptodate(bh);
431 clear_buffer_dirty(bh);
432}
433
434STATIC void
435xfs_start_page_writeback(
436 struct page *page,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100437 int clear_dirty,
438 int buffers)
439{
440 ASSERT(PageLocked(page));
441 ASSERT(!PageWriteback(page));
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100442 if (clear_dirty)
David Chinner92132022006-12-21 10:24:01 +1100443 clear_page_dirty_for_io(page);
444 set_page_writeback(page);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100445 unlock_page(page);
Fengguang Wu1f7decf2007-10-16 23:30:42 -0700446 /* If no buffers on the page are to be written, finish it here */
447 if (!buffers)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100448 end_page_writeback(page);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100449}
450
451static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh)
452{
453 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
454}
455
456/*
David Chinnerd88992f2006-01-18 13:38:12 +1100457 * Submit all of the bios for all of the ioends we have saved up, covering the
458 * initial writepage page and also any probed pages.
459 *
460 * Because we may have multiple ioends spanning a page, we need to start
461 * writeback on all the buffers before we submit them for I/O. If we mark the
462 * buffers as we got, then we can end up with a page that only has buffers
463 * marked async write and I/O complete on can occur before we mark the other
464 * buffers async write.
465 *
466 * The end result of this is that we trip a bug in end_page_writeback() because
467 * we call it twice for the one page as the code in end_buffer_async_write()
468 * assumes that all buffers on the page are started at the same time.
469 *
470 * The fix is two passes across the ioend list - one to start writeback on the
Nathan Scottc41564b2006-03-29 08:55:14 +1000471 * buffer_heads, and then submit them for I/O on the second pass.
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100472 */
473STATIC void
474xfs_submit_ioend(
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000475 struct writeback_control *wbc,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100476 xfs_ioend_t *ioend)
477{
David Chinnerd88992f2006-01-18 13:38:12 +1100478 xfs_ioend_t *head = ioend;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100479 xfs_ioend_t *next;
480 struct buffer_head *bh;
481 struct bio *bio;
482 sector_t lastblock = 0;
483
David Chinnerd88992f2006-01-18 13:38:12 +1100484 /* Pass 1 - start writeback */
485 do {
486 next = ioend->io_list;
Christoph Hellwig221cb252010-12-10 08:42:17 +0000487 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private)
David Chinnerd88992f2006-01-18 13:38:12 +1100488 xfs_start_buffer_writeback(bh);
David Chinnerd88992f2006-01-18 13:38:12 +1100489 } while ((ioend = next) != NULL);
490
491 /* Pass 2 - submit I/O */
492 ioend = head;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100493 do {
494 next = ioend->io_list;
495 bio = NULL;
496
497 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100498
499 if (!bio) {
500 retry:
501 bio = xfs_alloc_ioend_bio(bh);
502 } else if (bh->b_blocknr != lastblock + 1) {
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000503 xfs_submit_ioend_bio(wbc, ioend, bio);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100504 goto retry;
505 }
506
507 if (bio_add_buffer(bio, bh) != bh->b_size) {
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000508 xfs_submit_ioend_bio(wbc, ioend, bio);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100509 goto retry;
510 }
511
512 lastblock = bh->b_blocknr;
513 }
514 if (bio)
Christoph Hellwig06342cf2009-10-30 09:09:15 +0000515 xfs_submit_ioend_bio(wbc, ioend, bio);
Christoph Hellwig209fb872010-07-18 21:17:11 +0000516 xfs_finish_ioend(ioend);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100517 } while ((ioend = next) != NULL);
518}
519
520/*
521 * Cancel submission of all buffer_heads so far in this endio.
522 * Toss the endio too. Only ever called for the initial page
523 * in a writepage request, so only ever one page.
524 */
525STATIC void
526xfs_cancel_ioend(
527 xfs_ioend_t *ioend)
528{
529 xfs_ioend_t *next;
530 struct buffer_head *bh, *next_bh;
531
532 do {
533 next = ioend->io_list;
534 bh = ioend->io_buffer_head;
535 do {
536 next_bh = bh->b_private;
537 clear_buffer_async_write(bh);
538 unlock_buffer(bh);
539 } while ((bh = next_bh) != NULL);
540
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100541 mempool_free(ioend, xfs_ioend_pool);
542 } while ((ioend = next) != NULL);
543}
544
545/*
546 * Test to see if we've been building up a completion structure for
547 * earlier buffers -- if so, we try to append to this ioend if we
548 * can, otherwise we finish off any current ioend and start another.
549 * Return true if we've finished the given ioend.
550 */
551STATIC void
552xfs_add_to_ioend(
553 struct inode *inode,
554 struct buffer_head *bh,
Christoph Hellwig7336cea2006-01-11 20:49:16 +1100555 xfs_off_t offset,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100556 unsigned int type,
557 xfs_ioend_t **result,
558 int need_ioend)
559{
560 xfs_ioend_t *ioend = *result;
561
562 if (!ioend || need_ioend || type != ioend->io_type) {
563 xfs_ioend_t *previous = *result;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100564
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100565 ioend = xfs_alloc_ioend(inode, type);
566 ioend->io_offset = offset;
567 ioend->io_buffer_head = bh;
568 ioend->io_buffer_tail = bh;
569 if (previous)
570 previous->io_list = ioend;
571 *result = ioend;
572 } else {
573 ioend->io_buffer_tail->b_private = bh;
574 ioend->io_buffer_tail = bh;
575 }
576
577 bh->b_private = NULL;
578 ioend->io_size += bh->b_size;
579}
580
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581STATIC void
Nathan Scott87cbc492006-03-14 13:26:43 +1100582xfs_map_buffer(
Christoph Hellwig046f1682010-04-28 12:28:52 +0000583 struct inode *inode,
Nathan Scott87cbc492006-03-14 13:26:43 +1100584 struct buffer_head *bh,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000585 struct xfs_bmbt_irec *imap,
Christoph Hellwig046f1682010-04-28 12:28:52 +0000586 xfs_off_t offset)
Nathan Scott87cbc492006-03-14 13:26:43 +1100587{
588 sector_t bn;
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000589 struct xfs_mount *m = XFS_I(inode)->i_mount;
Christoph Hellwig207d0412010-04-28 12:28:56 +0000590 xfs_off_t iomap_offset = XFS_FSB_TO_B(m, imap->br_startoff);
591 xfs_daddr_t iomap_bn = xfs_fsb_to_db(XFS_I(inode), imap->br_startblock);
Nathan Scott87cbc492006-03-14 13:26:43 +1100592
Christoph Hellwig207d0412010-04-28 12:28:56 +0000593 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
594 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
Nathan Scott87cbc492006-03-14 13:26:43 +1100595
Christoph Hellwige5131822010-04-28 12:28:55 +0000596 bn = (iomap_bn >> (inode->i_blkbits - BBSHIFT)) +
Christoph Hellwig8699bb02010-04-28 12:28:54 +0000597 ((offset - iomap_offset) >> inode->i_blkbits);
Nathan Scott87cbc492006-03-14 13:26:43 +1100598
Christoph Hellwig046f1682010-04-28 12:28:52 +0000599 ASSERT(bn || XFS_IS_REALTIME_INODE(XFS_I(inode)));
Nathan Scott87cbc492006-03-14 13:26:43 +1100600
601 bh->b_blocknr = bn;
602 set_buffer_mapped(bh);
603}
604
605STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606xfs_map_at_offset(
Christoph Hellwig046f1682010-04-28 12:28:52 +0000607 struct inode *inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 struct buffer_head *bh,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000609 struct xfs_bmbt_irec *imap,
Christoph Hellwig046f1682010-04-28 12:28:52 +0000610 xfs_off_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611{
Christoph Hellwig207d0412010-04-28 12:28:56 +0000612 ASSERT(imap->br_startblock != HOLESTARTBLOCK);
613 ASSERT(imap->br_startblock != DELAYSTARTBLOCK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614
Christoph Hellwig207d0412010-04-28 12:28:56 +0000615 xfs_map_buffer(inode, bh, imap, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700616 set_buffer_mapped(bh);
617 clear_buffer_delay(bh);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100618 clear_buffer_unwritten(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619}
620
621/*
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100622 * Test if a given page is suitable for writing as part of an unwritten
623 * or delayed allocate extent.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 */
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100625STATIC int
626xfs_is_delayed_page(
627 struct page *page,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100628 unsigned int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 if (PageWriteback(page))
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100631 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632
633 if (page->mapping && page_has_buffers(page)) {
634 struct buffer_head *bh, *head;
635 int acceptable = 0;
636
637 bh = head = page_buffers(page);
638 do {
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100639 if (buffer_unwritten(bh))
Christoph Hellwig34a52c62010-04-28 12:28:57 +0000640 acceptable = (type == IO_UNWRITTEN);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100641 else if (buffer_delay(bh))
Christoph Hellwiga206c812010-12-10 08:42:20 +0000642 acceptable = (type == IO_DELALLOC);
David Chinner2ddee842006-03-22 12:47:40 +1100643 else if (buffer_dirty(bh) && buffer_mapped(bh))
Christoph Hellwiga206c812010-12-10 08:42:20 +0000644 acceptable = (type == IO_OVERWRITE);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100645 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 } while ((bh = bh->b_this_page) != head);
648
649 if (acceptable)
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100650 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 }
652
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100653 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654}
655
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656/*
657 * Allocate & map buffers for page given the extent map. Write it out.
658 * except for the original page of a writepage, this is called on
659 * delalloc/unwritten pages only, for the original page it is possible
660 * that the page has no mapping at all.
661 */
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100662STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663xfs_convert_page(
664 struct inode *inode,
665 struct page *page,
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100666 loff_t tindex,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000667 struct xfs_bmbt_irec *imap,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100668 xfs_ioend_t **ioendp,
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000669 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670{
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100671 struct buffer_head *bh, *head;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100672 xfs_off_t end_offset;
673 unsigned long p_offset;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100674 unsigned int type;
Nathan Scott24e17b52005-05-05 13:33:20 -0700675 int len, page_dirty;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100676 int count = 0, done = 0, uptodate = 1;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100677 xfs_off_t offset = page_offset(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100679 if (page->index != tindex)
680 goto fail;
Nick Piggin529ae9a2008-08-02 12:01:03 +0200681 if (!trylock_page(page))
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100682 goto fail;
683 if (PageWriteback(page))
684 goto fail_unlock_page;
685 if (page->mapping != inode->i_mapping)
686 goto fail_unlock_page;
687 if (!xfs_is_delayed_page(page, (*ioendp)->io_type))
688 goto fail_unlock_page;
689
Nathan Scott24e17b52005-05-05 13:33:20 -0700690 /*
691 * page_dirty is initially a count of buffers on the page before
Nathan Scottc41564b2006-03-29 08:55:14 +1000692 * EOF and is decremented as we move each into a cleanable state.
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100693 *
694 * Derivation:
695 *
696 * End offset is the highest offset that this page should represent.
697 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
698 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
699 * hence give us the correct page_dirty count. On any other page,
700 * it will be zero and in that case we need page_dirty to be the
701 * count of buffers on the page.
Nathan Scott24e17b52005-05-05 13:33:20 -0700702 */
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100703 end_offset = min_t(unsigned long long,
704 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
705 i_size_read(inode));
706
Nathan Scott24e17b52005-05-05 13:33:20 -0700707 len = 1 << inode->i_blkbits;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100708 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
709 PAGE_CACHE_SIZE);
710 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
711 page_dirty = p_offset / len;
Nathan Scott24e17b52005-05-05 13:33:20 -0700712
Linus Torvalds1da177e2005-04-16 15:20:36 -0700713 bh = head = page_buffers(page);
714 do {
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100715 if (offset >= end_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 break;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100717 if (!buffer_uptodate(bh))
718 uptodate = 0;
719 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
720 done = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721 continue;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100722 }
723
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000724 if (buffer_unwritten(bh) || buffer_delay(bh) ||
725 buffer_mapped(bh)) {
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100726 if (buffer_unwritten(bh))
Christoph Hellwig34a52c62010-04-28 12:28:57 +0000727 type = IO_UNWRITTEN;
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000728 else if (buffer_delay(bh))
Christoph Hellwiga206c812010-12-10 08:42:20 +0000729 type = IO_DELALLOC;
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000730 else
731 type = IO_OVERWRITE;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100732
Christoph Hellwig558e6892010-04-28 12:28:58 +0000733 if (!xfs_imap_valid(inode, imap, offset)) {
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100734 done = 1;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100735 continue;
736 }
737
Christoph Hellwigecff71e2010-12-10 08:42:25 +0000738 lock_buffer(bh);
739 if (type != IO_OVERWRITE)
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000740 xfs_map_at_offset(inode, bh, imap, offset);
Christoph Hellwig89f3b362010-06-24 09:45:48 +1000741 xfs_add_to_ioend(inode, bh, offset, type,
742 ioendp, done);
743
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100744 page_dirty--;
745 count++;
746 } else {
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000747 done = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748 }
Christoph Hellwig7336cea2006-01-11 20:49:16 +1100749 } while (offset += len, (bh = bh->b_this_page) != head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100751 if (uptodate && bh == head)
752 SetPageUptodate(page);
753
Christoph Hellwig89f3b362010-06-24 09:45:48 +1000754 if (count) {
Dave Chinnerefceab12010-08-24 11:44:56 +1000755 if (--wbc->nr_to_write <= 0 &&
756 wbc->sync_mode == WB_SYNC_NONE)
Christoph Hellwig89f3b362010-06-24 09:45:48 +1000757 done = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 }
Christoph Hellwig89f3b362010-06-24 09:45:48 +1000759 xfs_start_page_writeback(page, !page_dirty, count);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100760
761 return done;
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100762 fail_unlock_page:
763 unlock_page(page);
764 fail:
765 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766}
767
768/*
769 * Convert & write out a cluster of pages in the same extent as defined
770 * by mp and following the start page.
771 */
772STATIC void
773xfs_cluster_write(
774 struct inode *inode,
775 pgoff_t tindex,
Christoph Hellwig207d0412010-04-28 12:28:56 +0000776 struct xfs_bmbt_irec *imap,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100777 xfs_ioend_t **ioendp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 struct writeback_control *wbc,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 pgoff_t tlast)
780{
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100781 struct pagevec pvec;
782 int done = 0, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100784 pagevec_init(&pvec, 0);
785 while (!done && tindex <= tlast) {
786 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
787
788 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 break;
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100790
791 for (i = 0; i < pagevec_count(&pvec); i++) {
792 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
Christoph Hellwig2fa24f92010-12-10 08:42:23 +0000793 imap, ioendp, wbc);
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100794 if (done)
795 break;
796 }
797
798 pagevec_release(&pvec);
799 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 }
801}
802
Dave Chinner3ed3a432010-03-05 02:00:42 +0000803STATIC void
804xfs_vm_invalidatepage(
805 struct page *page,
806 unsigned long offset)
807{
808 trace_xfs_invalidatepage(page->mapping->host, page, offset);
809 block_invalidatepage(page, offset);
810}
811
812/*
813 * If the page has delalloc buffers on it, we need to punch them out before we
814 * invalidate the page. If we don't, we leave a stale delalloc mapping on the
815 * inode that can trip a BUG() in xfs_get_blocks() later on if a direct IO read
816 * is done on that same region - the delalloc extent is returned when none is
817 * supposed to be there.
818 *
819 * We prevent this by truncating away the delalloc regions on the page before
820 * invalidating it. Because they are delalloc, we can do this without needing a
821 * transaction. Indeed - if we get ENOSPC errors, we have to be able to do this
822 * truncation without a transaction as there is no space left for block
823 * reservation (typically why we see a ENOSPC in writeback).
824 *
825 * This is not a performance critical path, so for now just do the punching a
826 * buffer head at a time.
827 */
828STATIC void
829xfs_aops_discard_page(
830 struct page *page)
831{
832 struct inode *inode = page->mapping->host;
833 struct xfs_inode *ip = XFS_I(inode);
834 struct buffer_head *bh, *head;
835 loff_t offset = page_offset(page);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000836
Christoph Hellwiga206c812010-12-10 08:42:20 +0000837 if (!xfs_is_delayed_page(page, IO_DELALLOC))
Dave Chinner3ed3a432010-03-05 02:00:42 +0000838 goto out_invalidate;
839
Dave Chinnere8c37532010-03-15 02:36:35 +0000840 if (XFS_FORCED_SHUTDOWN(ip->i_mount))
841 goto out_invalidate;
842
Dave Chinner4f107002011-03-07 10:00:35 +1100843 xfs_alert(ip->i_mount,
Dave Chinner3ed3a432010-03-05 02:00:42 +0000844 "page discard on page %p, inode 0x%llx, offset %llu.",
845 page, ip->i_ino, offset);
846
847 xfs_ilock(ip, XFS_ILOCK_EXCL);
848 bh = head = page_buffers(page);
849 do {
Dave Chinner3ed3a432010-03-05 02:00:42 +0000850 int error;
Dave Chinnerc726de42010-11-30 15:14:39 +1100851 xfs_fileoff_t start_fsb;
Dave Chinner3ed3a432010-03-05 02:00:42 +0000852
853 if (!buffer_delay(bh))
854 goto next_buffer;
855
Dave Chinnerc726de42010-11-30 15:14:39 +1100856 start_fsb = XFS_B_TO_FSBT(ip->i_mount, offset);
857 error = xfs_bmap_punch_delalloc_range(ip, start_fsb, 1);
Dave Chinner3ed3a432010-03-05 02:00:42 +0000858 if (error) {
859 /* something screwed, just bail */
Dave Chinnere8c37532010-03-15 02:36:35 +0000860 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
Dave Chinner4f107002011-03-07 10:00:35 +1100861 xfs_alert(ip->i_mount,
Dave Chinner3ed3a432010-03-05 02:00:42 +0000862 "page discard unable to remove delalloc mapping.");
Dave Chinnere8c37532010-03-15 02:36:35 +0000863 }
Dave Chinner3ed3a432010-03-05 02:00:42 +0000864 break;
865 }
866next_buffer:
Dave Chinnerc726de42010-11-30 15:14:39 +1100867 offset += 1 << inode->i_blkbits;
Dave Chinner3ed3a432010-03-05 02:00:42 +0000868
869 } while ((bh = bh->b_this_page) != head);
870
871 xfs_iunlock(ip, XFS_ILOCK_EXCL);
872out_invalidate:
873 xfs_vm_invalidatepage(page, 0);
874 return;
875}
876
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877/*
Christoph Hellwig89f3b362010-06-24 09:45:48 +1000878 * Write out a dirty page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 *
Christoph Hellwig89f3b362010-06-24 09:45:48 +1000880 * For delalloc space on the page we need to allocate space and flush it.
881 * For unwritten space on the page we need to start the conversion to
882 * regular allocated space.
Christoph Hellwig89f3b362010-06-24 09:45:48 +1000883 * For any other dirty buffer heads on the page we should flush them.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885STATIC int
Christoph Hellwig89f3b362010-06-24 09:45:48 +1000886xfs_vm_writepage(
887 struct page *page,
888 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889{
Christoph Hellwig89f3b362010-06-24 09:45:48 +1000890 struct inode *inode = page->mapping->host;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100891 struct buffer_head *bh, *head;
Christoph Hellwig207d0412010-04-28 12:28:56 +0000892 struct xfs_bmbt_irec imap;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100893 xfs_ioend_t *ioend = NULL, *iohead = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 loff_t offset;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100895 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896 __uint64_t end_offset;
Christoph Hellwigbd1556a2010-04-28 12:29:00 +0000897 pgoff_t end_index, last_index;
Christoph Hellwiged1e7b72010-12-10 08:42:22 +0000898 ssize_t len;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000899 int err, imap_valid = 0, uptodate = 1;
Christoph Hellwig89f3b362010-06-24 09:45:48 +1000900 int count = 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +0000901 int nonblocking = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902
Christoph Hellwig89f3b362010-06-24 09:45:48 +1000903 trace_xfs_writepage(inode, page, 0);
904
Christoph Hellwig20cb52e2010-06-24 09:46:01 +1000905 ASSERT(page_has_buffers(page));
906
Christoph Hellwig89f3b362010-06-24 09:45:48 +1000907 /*
908 * Refuse to write the page out if we are called from reclaim context.
909 *
Christoph Hellwigd4f7a5c2010-06-28 10:34:44 -0400910 * This avoids stack overflows when called from deeply used stacks in
911 * random callers for direct reclaim or memcg reclaim. We explicitly
912 * allow reclaim from kswapd as the stack usage there is relatively low.
Christoph Hellwig89f3b362010-06-24 09:45:48 +1000913 *
Mel Gorman94054fa2011-10-31 17:07:45 -0700914 * This should never happen except in the case of a VM regression so
915 * warn about it.
Christoph Hellwig89f3b362010-06-24 09:45:48 +1000916 */
Mel Gorman94054fa2011-10-31 17:07:45 -0700917 if (WARN_ON_ONCE((current->flags & (PF_MEMALLOC|PF_KSWAPD)) ==
918 PF_MEMALLOC))
Christoph Hellwigb5420f22010-08-24 11:47:51 +1000919 goto redirty;
Christoph Hellwig89f3b362010-06-24 09:45:48 +1000920
921 /*
Christoph Hellwig680a6472011-07-08 14:34:05 +0200922 * Given that we do not allow direct reclaim to call us, we should
923 * never be called while in a filesystem transaction.
Christoph Hellwig89f3b362010-06-24 09:45:48 +1000924 */
Christoph Hellwig680a6472011-07-08 14:34:05 +0200925 if (WARN_ON(current->flags & PF_FSTRANS))
Christoph Hellwigb5420f22010-08-24 11:47:51 +1000926 goto redirty;
Christoph Hellwig89f3b362010-06-24 09:45:48 +1000927
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928 /* Is this page beyond the end of the file? */
929 offset = i_size_read(inode);
930 end_index = offset >> PAGE_CACHE_SHIFT;
931 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
932 if (page->index >= end_index) {
933 if ((page->index >= end_index + 1) ||
934 !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
Christoph Hellwig89f3b362010-06-24 09:45:48 +1000935 unlock_page(page);
Nathan Scott19d5bcf2005-11-02 15:14:09 +1100936 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937 }
938 }
939
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100940 end_offset = min_t(unsigned long long,
Christoph Hellwig20cb52e2010-06-24 09:46:01 +1000941 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
942 offset);
Nathan Scott24e17b52005-05-05 13:33:20 -0700943 len = 1 << inode->i_blkbits;
Nathan Scott24e17b52005-05-05 13:33:20 -0700944
Nathan Scott24e17b52005-05-05 13:33:20 -0700945 bh = head = page_buffers(page);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100946 offset = page_offset(page);
Christoph Hellwiga206c812010-12-10 08:42:20 +0000947 type = IO_OVERWRITE;
948
Christoph Hellwigdbcdde32011-07-08 14:34:14 +0200949 if (wbc->sync_mode == WB_SYNC_NONE)
Christoph Hellwiga206c812010-12-10 08:42:20 +0000950 nonblocking = 1;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100951
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 do {
Christoph Hellwig6ac72482010-12-10 08:42:18 +0000953 int new_ioend = 0;
954
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955 if (offset >= end_offset)
956 break;
957 if (!buffer_uptodate(bh))
958 uptodate = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959
Eric Sandeen3d9b02e2010-06-24 09:45:30 +1000960 /*
Christoph Hellwigece413f2010-11-10 21:39:11 +0000961 * set_page_dirty dirties all buffers in a page, independent
962 * of their state. The dirty state however is entirely
963 * meaningless for holes (!mapped && uptodate), so skip
964 * buffers covering holes here.
Eric Sandeen3d9b02e2010-06-24 09:45:30 +1000965 */
966 if (!buffer_mapped(bh) && buffer_uptodate(bh)) {
Eric Sandeen3d9b02e2010-06-24 09:45:30 +1000967 imap_valid = 0;
968 continue;
969 }
970
Christoph Hellwigaeea1b12010-12-10 08:42:24 +0000971 if (buffer_unwritten(bh)) {
972 if (type != IO_UNWRITTEN) {
973 type = IO_UNWRITTEN;
974 imap_valid = 0;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100975 }
Christoph Hellwigaeea1b12010-12-10 08:42:24 +0000976 } else if (buffer_delay(bh)) {
977 if (type != IO_DELALLOC) {
978 type = IO_DELALLOC;
979 imap_valid = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 }
Christoph Hellwig89f3b362010-06-24 09:45:48 +1000981 } else if (buffer_uptodate(bh)) {
Christoph Hellwiga206c812010-12-10 08:42:20 +0000982 if (type != IO_OVERWRITE) {
983 type = IO_OVERWRITE;
Christoph Hellwig85da94c2010-12-10 08:42:16 +0000984 imap_valid = 0;
985 }
Christoph Hellwigaeea1b12010-12-10 08:42:24 +0000986 } else {
987 if (PageUptodate(page)) {
988 ASSERT(buffer_mapped(bh));
989 imap_valid = 0;
Christoph Hellwig6c4fe192006-01-11 20:49:28 +1100990 }
Christoph Hellwigaeea1b12010-12-10 08:42:24 +0000991 continue;
992 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993
Christoph Hellwigaeea1b12010-12-10 08:42:24 +0000994 if (imap_valid)
995 imap_valid = xfs_imap_valid(inode, &imap, offset);
996 if (!imap_valid) {
997 /*
998 * If we didn't have a valid mapping then we need to
999 * put the new mapping into a separate ioend structure.
1000 * This ensures non-contiguous extents always have
1001 * separate ioends, which is particularly important
1002 * for unwritten extent conversion at I/O completion
1003 * time.
1004 */
1005 new_ioend = 1;
1006 err = xfs_map_blocks(inode, offset, &imap, type,
1007 nonblocking);
1008 if (err)
1009 goto error;
1010 imap_valid = xfs_imap_valid(inode, &imap, offset);
1011 }
1012 if (imap_valid) {
Christoph Hellwigecff71e2010-12-10 08:42:25 +00001013 lock_buffer(bh);
1014 if (type != IO_OVERWRITE)
Christoph Hellwigaeea1b12010-12-10 08:42:24 +00001015 xfs_map_at_offset(inode, bh, &imap, offset);
1016 xfs_add_to_ioend(inode, bh, offset, type, &ioend,
1017 new_ioend);
1018 count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 }
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001020
1021 if (!iohead)
1022 iohead = ioend;
1023
1024 } while (offset += len, ((bh = bh->b_this_page) != head));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025
1026 if (uptodate && bh == head)
1027 SetPageUptodate(page);
1028
Christoph Hellwig89f3b362010-06-24 09:45:48 +10001029 xfs_start_page_writeback(page, 1, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030
Christoph Hellwig558e6892010-04-28 12:28:58 +00001031 if (ioend && imap_valid) {
Christoph Hellwigbd1556a2010-04-28 12:29:00 +00001032 xfs_off_t end_index;
Christoph Hellwig8699bb02010-04-28 12:28:54 +00001033
Christoph Hellwigbd1556a2010-04-28 12:29:00 +00001034 end_index = imap.br_startoff + imap.br_blockcount;
1035
1036 /* to bytes */
1037 end_index <<= inode->i_blkbits;
1038
1039 /* to pages */
1040 end_index = (end_index - 1) >> PAGE_CACHE_SHIFT;
1041
1042 /* check against file size */
1043 if (end_index > last_index)
1044 end_index = last_index;
1045
Christoph Hellwig207d0412010-04-28 12:28:56 +00001046 xfs_cluster_write(inode, page->index + 1, &imap, &ioend,
Christoph Hellwig2fa24f92010-12-10 08:42:23 +00001047 wbc, end_index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 }
1049
Christoph Hellwig281627d2012-03-13 08:41:05 +00001050 if (iohead) {
1051 /*
1052 * Reserve log space if we might write beyond the on-disk
1053 * inode size.
1054 */
1055 if (ioend->io_type != IO_UNWRITTEN &&
1056 xfs_ioend_is_append(ioend)) {
1057 err = xfs_setfilesize_trans_alloc(ioend);
1058 if (err)
1059 goto error;
1060 }
1061
Christoph Hellwig06342cf2009-10-30 09:09:15 +00001062 xfs_submit_ioend(wbc, iohead);
Christoph Hellwig281627d2012-03-13 08:41:05 +00001063 }
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001064
Christoph Hellwig89f3b362010-06-24 09:45:48 +10001065 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066
1067error:
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001068 if (iohead)
1069 xfs_cancel_ioend(iohead);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070
Christoph Hellwigb5420f22010-08-24 11:47:51 +10001071 if (err == -EAGAIN)
1072 goto redirty;
1073
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001074 xfs_aops_discard_page(page);
Christoph Hellwig89f3b362010-06-24 09:45:48 +10001075 ClearPageUptodate(page);
1076 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 return err;
Nathan Scottf51623b2006-03-14 13:26:27 +11001078
Christoph Hellwigb5420f22010-08-24 11:47:51 +10001079redirty:
Nathan Scottf51623b2006-03-14 13:26:27 +11001080 redirty_page_for_writepage(wbc, page);
1081 unlock_page(page);
1082 return 0;
Nathan Scottf51623b2006-03-14 13:26:27 +11001083}
1084
Nathan Scott7d4fb402006-06-09 15:27:16 +10001085STATIC int
1086xfs_vm_writepages(
1087 struct address_space *mapping,
1088 struct writeback_control *wbc)
1089{
Christoph Hellwigb3aea4e2007-08-29 11:44:37 +10001090 xfs_iflags_clear(XFS_I(mapping->host), XFS_ITRUNCATED);
Nathan Scott7d4fb402006-06-09 15:27:16 +10001091 return generic_writepages(mapping, wbc);
1092}
1093
Nathan Scottf51623b2006-03-14 13:26:27 +11001094/*
1095 * Called to move a page into cleanable state - and from there
Christoph Hellwig89f3b362010-06-24 09:45:48 +10001096 * to be released. The page should already be clean. We always
Nathan Scottf51623b2006-03-14 13:26:27 +11001097 * have buffer heads in this call.
1098 *
Christoph Hellwig89f3b362010-06-24 09:45:48 +10001099 * Returns 1 if the page is ok to release, 0 otherwise.
Nathan Scottf51623b2006-03-14 13:26:27 +11001100 */
1101STATIC int
Nathan Scott238f4c52006-03-17 17:26:25 +11001102xfs_vm_releasepage(
Nathan Scottf51623b2006-03-14 13:26:27 +11001103 struct page *page,
1104 gfp_t gfp_mask)
1105{
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001106 int delalloc, unwritten;
Nathan Scottf51623b2006-03-14 13:26:27 +11001107
Christoph Hellwig89f3b362010-06-24 09:45:48 +10001108 trace_xfs_releasepage(page->mapping->host, page, 0);
Nathan Scott238f4c52006-03-17 17:26:25 +11001109
Christoph Hellwig20cb52e2010-06-24 09:46:01 +10001110 xfs_count_page_state(page, &delalloc, &unwritten);
Nathan Scottf51623b2006-03-14 13:26:27 +11001111
Christoph Hellwig89f3b362010-06-24 09:45:48 +10001112 if (WARN_ON(delalloc))
1113 return 0;
1114 if (WARN_ON(unwritten))
Nathan Scottf51623b2006-03-14 13:26:27 +11001115 return 0;
1116
Nathan Scottf51623b2006-03-14 13:26:27 +11001117 return try_to_free_buffers(page);
1118}
1119
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120STATIC int
Nathan Scottc2536662006-03-29 10:44:40 +10001121__xfs_get_blocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 struct inode *inode,
1123 sector_t iblock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 struct buffer_head *bh_result,
1125 int create,
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001126 int direct)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001127{
Christoph Hellwiga206c812010-12-10 08:42:20 +00001128 struct xfs_inode *ip = XFS_I(inode);
1129 struct xfs_mount *mp = ip->i_mount;
1130 xfs_fileoff_t offset_fsb, end_fsb;
1131 int error = 0;
1132 int lockmode = 0;
Christoph Hellwig207d0412010-04-28 12:28:56 +00001133 struct xfs_bmbt_irec imap;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001134 int nimaps = 1;
Nathan Scottfdc7ed72005-11-02 15:13:13 +11001135 xfs_off_t offset;
1136 ssize_t size;
Christoph Hellwig207d0412010-04-28 12:28:56 +00001137 int new = 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001138
1139 if (XFS_FORCED_SHUTDOWN(mp))
1140 return -XFS_ERROR(EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141
Nathan Scottfdc7ed72005-11-02 15:13:13 +11001142 offset = (xfs_off_t)iblock << inode->i_blkbits;
Nathan Scottc2536662006-03-29 10:44:40 +10001143 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1144 size = bh_result->b_size;
Lachlan McIlroy364f3582008-09-17 16:50:14 +10001145
1146 if (!create && direct && offset >= i_size_read(inode))
1147 return 0;
1148
Christoph Hellwiga206c812010-12-10 08:42:20 +00001149 if (create) {
1150 lockmode = XFS_ILOCK_EXCL;
1151 xfs_ilock(ip, lockmode);
1152 } else {
1153 lockmode = xfs_ilock_map_shared(ip);
1154 }
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001155
Christoph Hellwiga206c812010-12-10 08:42:20 +00001156 ASSERT(offset <= mp->m_maxioffset);
1157 if (offset + size > mp->m_maxioffset)
1158 size = mp->m_maxioffset - offset;
1159 end_fsb = XFS_B_TO_FSB(mp, (xfs_ufsize_t)offset + size);
1160 offset_fsb = XFS_B_TO_FSBT(mp, offset);
1161
Dave Chinner5c8ed202011-09-18 20:40:45 +00001162 error = xfs_bmapi_read(ip, offset_fsb, end_fsb - offset_fsb,
1163 &imap, &nimaps, XFS_BMAPI_ENTIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 if (error)
Christoph Hellwiga206c812010-12-10 08:42:20 +00001165 goto out_unlock;
1166
1167 if (create &&
1168 (!nimaps ||
1169 (imap.br_startblock == HOLESTARTBLOCK ||
1170 imap.br_startblock == DELAYSTARTBLOCK))) {
1171 if (direct) {
1172 error = xfs_iomap_write_direct(ip, offset, size,
1173 &imap, nimaps);
1174 } else {
1175 error = xfs_iomap_write_delay(ip, offset, size, &imap);
1176 }
1177 if (error)
1178 goto out_unlock;
1179
1180 trace_xfs_get_blocks_alloc(ip, offset, size, 0, &imap);
1181 } else if (nimaps) {
1182 trace_xfs_get_blocks_found(ip, offset, size, 0, &imap);
1183 } else {
1184 trace_xfs_get_blocks_notfound(ip, offset, size);
1185 goto out_unlock;
1186 }
1187 xfs_iunlock(ip, lockmode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188
Christoph Hellwig207d0412010-04-28 12:28:56 +00001189 if (imap.br_startblock != HOLESTARTBLOCK &&
1190 imap.br_startblock != DELAYSTARTBLOCK) {
Nathan Scott87cbc492006-03-14 13:26:43 +11001191 /*
1192 * For unwritten extents do not report a disk address on
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193 * the read case (treat as if we're reading into a hole).
1194 */
Christoph Hellwig207d0412010-04-28 12:28:56 +00001195 if (create || !ISUNWRITTEN(&imap))
1196 xfs_map_buffer(inode, bh_result, &imap, offset);
1197 if (create && ISUNWRITTEN(&imap)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198 if (direct)
1199 bh_result->b_private = inode;
1200 set_buffer_unwritten(bh_result);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 }
1202 }
1203
Nathan Scottc2536662006-03-29 10:44:40 +10001204 /*
1205 * If this is a realtime file, data may be on a different device.
1206 * to that pointed to from the buffer_head b_bdev currently.
1207 */
Christoph Hellwig046f1682010-04-28 12:28:52 +00001208 bh_result->b_bdev = xfs_find_bdev_for_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209
Nathan Scottc2536662006-03-29 10:44:40 +10001210 /*
David Chinner549054a2007-02-10 18:36:35 +11001211 * If we previously allocated a block out beyond eof and we are now
1212 * coming back to use it then we will need to flag it as new even if it
1213 * has a disk address.
1214 *
1215 * With sub-block writes into unwritten extents we also need to mark
1216 * the buffer as new so that the unwritten parts of the buffer gets
1217 * correctly zeroed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218 */
1219 if (create &&
1220 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
David Chinner549054a2007-02-10 18:36:35 +11001221 (offset >= i_size_read(inode)) ||
Christoph Hellwig207d0412010-04-28 12:28:56 +00001222 (new || ISUNWRITTEN(&imap))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 set_buffer_new(bh_result);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224
Christoph Hellwig207d0412010-04-28 12:28:56 +00001225 if (imap.br_startblock == DELAYSTARTBLOCK) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226 BUG_ON(direct);
1227 if (create) {
1228 set_buffer_uptodate(bh_result);
1229 set_buffer_mapped(bh_result);
1230 set_buffer_delay(bh_result);
1231 }
1232 }
1233
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001234 /*
1235 * If this is O_DIRECT or the mpage code calling tell them how large
1236 * the mapping is, so that we can avoid repeated get_blocks calls.
1237 */
Nathan Scottc2536662006-03-29 10:44:40 +10001238 if (direct || size > (1 << inode->i_blkbits)) {
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001239 xfs_off_t mapping_size;
Christoph Hellwig9563b3d2010-04-28 12:28:53 +00001240
Christoph Hellwig2b8f12b2010-04-28 12:28:59 +00001241 mapping_size = imap.br_startoff + imap.br_blockcount - iblock;
1242 mapping_size <<= inode->i_blkbits;
1243
1244 ASSERT(mapping_size > 0);
1245 if (mapping_size > size)
1246 mapping_size = size;
1247 if (mapping_size > LONG_MAX)
1248 mapping_size = LONG_MAX;
1249
1250 bh_result->b_size = mapping_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 }
1252
1253 return 0;
Christoph Hellwiga206c812010-12-10 08:42:20 +00001254
1255out_unlock:
1256 xfs_iunlock(ip, lockmode);
1257 return -error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001258}
1259
1260int
Nathan Scottc2536662006-03-29 10:44:40 +10001261xfs_get_blocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 struct inode *inode,
1263 sector_t iblock,
1264 struct buffer_head *bh_result,
1265 int create)
1266{
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001267 return __xfs_get_blocks(inode, iblock, bh_result, create, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268}
1269
1270STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001271xfs_get_blocks_direct(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 struct inode *inode,
1273 sector_t iblock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 struct buffer_head *bh_result,
1275 int create)
1276{
Christoph Hellwigf2bde9b2010-06-24 11:44:35 +10001277 return __xfs_get_blocks(inode, iblock, bh_result, create, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278}
1279
Christoph Hellwig209fb872010-07-18 21:17:11 +00001280/*
1281 * Complete a direct I/O write request.
1282 *
1283 * If the private argument is non-NULL __xfs_get_blocks signals us that we
1284 * need to issue a transaction to convert the range from unwritten to written
1285 * extents. In case this is regular synchronous I/O we just call xfs_end_io
Lucas De Marchi25985ed2011-03-30 22:57:33 -03001286 * to do this and we are done. But in case this was a successful AIO
Christoph Hellwig209fb872010-07-18 21:17:11 +00001287 * request this handler is called from interrupt context, from which we
1288 * can't start transactions. In that case offload the I/O completion to
1289 * the workqueues we also use for buffered I/O completion.
1290 */
Christoph Hellwigf0973862005-09-05 08:22:52 +10001291STATIC void
Christoph Hellwig209fb872010-07-18 21:17:11 +00001292xfs_end_io_direct_write(
1293 struct kiocb *iocb,
1294 loff_t offset,
1295 ssize_t size,
1296 void *private,
1297 int ret,
1298 bool is_async)
Christoph Hellwigf0973862005-09-05 08:22:52 +10001299{
Christoph Hellwig209fb872010-07-18 21:17:11 +00001300 struct xfs_ioend *ioend = iocb->private;
Christoph Hellwigf0973862005-09-05 08:22:52 +10001301
1302 /*
Christoph Hellwig2813d682011-12-18 20:00:12 +00001303 * While the generic direct I/O code updates the inode size, it does
1304 * so only after the end_io handler is called, which means our
1305 * end_io handler thinks the on-disk size is outside the in-core
1306 * size. To prevent this just update it a little bit earlier here.
1307 */
1308 if (offset + size > i_size_read(ioend->io_inode))
1309 i_size_write(ioend->io_inode, offset + size);
1310
1311 /*
Nathan Scottc41564b2006-03-29 08:55:14 +10001312 * blockdev_direct_IO can return an error even after the I/O
Christoph Hellwigf0973862005-09-05 08:22:52 +10001313 * completion handler was called. Thus we need to protect
1314 * against double-freeing.
1315 */
1316 iocb->private = NULL;
Christoph Hellwig40e2e972010-07-18 21:17:09 +00001317
Christoph Hellwig209fb872010-07-18 21:17:11 +00001318 ioend->io_offset = offset;
1319 ioend->io_size = size;
Christoph Hellwigc859cdd2011-08-23 08:28:10 +00001320 ioend->io_iocb = iocb;
1321 ioend->io_result = ret;
Christoph Hellwig209fb872010-07-18 21:17:11 +00001322 if (private && size > 0)
1323 ioend->io_type = IO_UNWRITTEN;
1324
1325 if (is_async) {
Christoph Hellwigc859cdd2011-08-23 08:28:10 +00001326 ioend->io_isasync = 1;
Christoph Hellwig209fb872010-07-18 21:17:11 +00001327 xfs_finish_ioend(ioend);
1328 } else {
1329 xfs_finish_ioend_sync(ioend);
1330 }
Christoph Hellwigf0973862005-09-05 08:22:52 +10001331}
1332
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333STATIC ssize_t
Nathan Scotte4c573b2006-03-14 13:54:26 +11001334xfs_vm_direct_IO(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 int rw,
1336 struct kiocb *iocb,
1337 const struct iovec *iov,
1338 loff_t offset,
1339 unsigned long nr_segs)
1340{
Christoph Hellwig209fb872010-07-18 21:17:11 +00001341 struct inode *inode = iocb->ki_filp->f_mapping->host;
1342 struct block_device *bdev = xfs_find_bdev_for_inode(inode);
Christoph Hellwig281627d2012-03-13 08:41:05 +00001343 struct xfs_ioend *ioend = NULL;
Christoph Hellwig209fb872010-07-18 21:17:11 +00001344 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345
Christoph Hellwig209fb872010-07-18 21:17:11 +00001346 if (rw & WRITE) {
Christoph Hellwig281627d2012-03-13 08:41:05 +00001347 size_t size = iov_length(iov, nr_segs);
1348
1349 /*
1350 * We need to preallocate a transaction for a size update
1351 * here. In the case that this write both updates the size
1352 * and converts at least on unwritten extent we will cancel
1353 * the still clean transaction after the I/O has finished.
1354 */
1355 iocb->private = ioend = xfs_alloc_ioend(inode, IO_DIRECT);
1356 if (offset + size > XFS_I(inode)->i_d.di_size) {
1357 ret = xfs_setfilesize_trans_alloc(ioend);
1358 if (ret)
1359 goto out_destroy_ioend;
1360 ioend->io_isdirect = 1;
1361 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +02001363 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov,
1364 offset, nr_segs,
1365 xfs_get_blocks_direct,
1366 xfs_end_io_direct_write, NULL, 0);
Christoph Hellwig209fb872010-07-18 21:17:11 +00001367 if (ret != -EIOCBQUEUED && iocb->private)
Christoph Hellwig281627d2012-03-13 08:41:05 +00001368 goto out_trans_cancel;
Christoph Hellwig209fb872010-07-18 21:17:11 +00001369 } else {
Christoph Hellwigeafdc7d2010-06-04 11:29:53 +02001370 ret = __blockdev_direct_IO(rw, iocb, inode, bdev, iov,
1371 offset, nr_segs,
1372 xfs_get_blocks_direct,
1373 NULL, NULL, 0);
Christoph Hellwig209fb872010-07-18 21:17:11 +00001374 }
Christoph Hellwig5fe878a2009-12-15 16:47:50 -08001375
Christoph Hellwigf0973862005-09-05 08:22:52 +10001376 return ret;
Christoph Hellwig281627d2012-03-13 08:41:05 +00001377
1378out_trans_cancel:
1379 if (ioend->io_append_trans) {
1380 current_set_flags_nested(&ioend->io_append_trans->t_pflags,
1381 PF_FSTRANS);
1382 xfs_trans_cancel(ioend->io_append_trans, 0);
1383 }
1384out_destroy_ioend:
1385 xfs_destroy_ioend(ioend);
1386 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387}
1388
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001389STATIC void
1390xfs_vm_write_failed(
1391 struct address_space *mapping,
1392 loff_t to)
1393{
1394 struct inode *inode = mapping->host;
1395
1396 if (to > inode->i_size) {
Dave Chinnerc726de42010-11-30 15:14:39 +11001397 /*
Christoph Hellwig2813d682011-12-18 20:00:12 +00001398 * Punch out the delalloc blocks we have already allocated.
1399 *
1400 * Don't bother with xfs_setattr given that nothing can have
1401 * made it to disk yet as the page is still locked at this
1402 * point.
Dave Chinnerc726de42010-11-30 15:14:39 +11001403 */
1404 struct xfs_inode *ip = XFS_I(inode);
1405 xfs_fileoff_t start_fsb;
1406 xfs_fileoff_t end_fsb;
1407 int error;
1408
1409 truncate_pagecache(inode, to, inode->i_size);
1410
1411 /*
1412 * Check if there are any blocks that are outside of i_size
1413 * that need to be trimmed back.
1414 */
1415 start_fsb = XFS_B_TO_FSB(ip->i_mount, inode->i_size) + 1;
1416 end_fsb = XFS_B_TO_FSB(ip->i_mount, to);
1417 if (end_fsb <= start_fsb)
1418 return;
1419
1420 xfs_ilock(ip, XFS_ILOCK_EXCL);
1421 error = xfs_bmap_punch_delalloc_range(ip, start_fsb,
1422 end_fsb - start_fsb);
1423 if (error) {
1424 /* something screwed, just bail */
1425 if (!XFS_FORCED_SHUTDOWN(ip->i_mount)) {
Dave Chinner4f107002011-03-07 10:00:35 +11001426 xfs_alert(ip->i_mount,
Dave Chinnerc726de42010-11-30 15:14:39 +11001427 "xfs_vm_write_failed: unable to clean up ino %lld",
1428 ip->i_ino);
1429 }
1430 }
1431 xfs_iunlock(ip, XFS_ILOCK_EXCL);
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001432 }
1433}
1434
Nathan Scottf51623b2006-03-14 13:26:27 +11001435STATIC int
Nick Piggind79689c2007-10-16 01:25:06 -07001436xfs_vm_write_begin(
Nathan Scottf51623b2006-03-14 13:26:27 +11001437 struct file *file,
Nick Piggind79689c2007-10-16 01:25:06 -07001438 struct address_space *mapping,
1439 loff_t pos,
1440 unsigned len,
1441 unsigned flags,
1442 struct page **pagep,
1443 void **fsdata)
Nathan Scottf51623b2006-03-14 13:26:27 +11001444{
Christoph Hellwig155130a2010-06-04 11:29:58 +02001445 int ret;
1446
1447 ret = block_write_begin(mapping, pos, len, flags | AOP_FLAG_NOFS,
1448 pagep, xfs_get_blocks);
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001449 if (unlikely(ret))
1450 xfs_vm_write_failed(mapping, pos + len);
1451 return ret;
1452}
Christoph Hellwig155130a2010-06-04 11:29:58 +02001453
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001454STATIC int
1455xfs_vm_write_end(
1456 struct file *file,
1457 struct address_space *mapping,
1458 loff_t pos,
1459 unsigned len,
1460 unsigned copied,
1461 struct page *page,
1462 void *fsdata)
1463{
1464 int ret;
1465
1466 ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata);
1467 if (unlikely(ret < len))
1468 xfs_vm_write_failed(mapping, pos + len);
Christoph Hellwig155130a2010-06-04 11:29:58 +02001469 return ret;
Nathan Scottf51623b2006-03-14 13:26:27 +11001470}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471
1472STATIC sector_t
Nathan Scotte4c573b2006-03-14 13:54:26 +11001473xfs_vm_bmap(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 struct address_space *mapping,
1475 sector_t block)
1476{
1477 struct inode *inode = (struct inode *)mapping->host;
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001478 struct xfs_inode *ip = XFS_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479
Christoph Hellwigcca28fb2010-06-24 11:57:09 +10001480 trace_xfs_vm_bmap(XFS_I(inode));
Christoph Hellwig126468b2008-03-06 13:44:57 +11001481 xfs_ilock(ip, XFS_IOLOCK_SHARED);
Christoph Hellwig739bfb22007-08-29 10:58:01 +10001482 xfs_flush_pages(ip, (xfs_off_t)0, -1, 0, FI_REMAPF);
Christoph Hellwig126468b2008-03-06 13:44:57 +11001483 xfs_iunlock(ip, XFS_IOLOCK_SHARED);
Nathan Scottc2536662006-03-29 10:44:40 +10001484 return generic_block_bmap(mapping, block, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001485}
1486
1487STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001488xfs_vm_readpage(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 struct file *unused,
1490 struct page *page)
1491{
Nathan Scottc2536662006-03-29 10:44:40 +10001492 return mpage_readpage(page, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493}
1494
1495STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001496xfs_vm_readpages(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 struct file *unused,
1498 struct address_space *mapping,
1499 struct list_head *pages,
1500 unsigned nr_pages)
1501{
Nathan Scottc2536662006-03-29 10:44:40 +10001502 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503}
1504
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001505const struct address_space_operations xfs_address_space_operations = {
Nathan Scotte4c573b2006-03-14 13:54:26 +11001506 .readpage = xfs_vm_readpage,
1507 .readpages = xfs_vm_readpages,
1508 .writepage = xfs_vm_writepage,
Nathan Scott7d4fb402006-06-09 15:27:16 +10001509 .writepages = xfs_vm_writepages,
Nathan Scott238f4c52006-03-17 17:26:25 +11001510 .releasepage = xfs_vm_releasepage,
1511 .invalidatepage = xfs_vm_invalidatepage,
Nick Piggind79689c2007-10-16 01:25:06 -07001512 .write_begin = xfs_vm_write_begin,
Christoph Hellwigfa9b2272010-06-14 05:17:31 -04001513 .write_end = xfs_vm_write_end,
Nathan Scotte4c573b2006-03-14 13:54:26 +11001514 .bmap = xfs_vm_bmap,
1515 .direct_IO = xfs_vm_direct_IO,
Christoph Lametere965f962006-02-01 03:05:41 -08001516 .migratepage = buffer_migrate_page,
Hisashi Hifumibddaafa2009-03-29 09:53:38 +02001517 .is_partially_uptodate = block_is_partially_uptodate,
Andi Kleenaa261f52009-09-16 11:50:16 +02001518 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519};