blob: 354d68a32d4a4e4aa9e701c327bdd95d9139be1c [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_dir2.h"
25#include "xfs_trans.h"
26#include "xfs_dmapi.h"
27#include "xfs_mount.h"
28#include "xfs_bmap_btree.h"
29#include "xfs_alloc_btree.h"
30#include "xfs_ialloc_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include "xfs_dir2_sf.h"
Nathan Scotta844f452005-11-02 14:38:42 +110032#include "xfs_attr_sf.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include "xfs_dinode.h"
34#include "xfs_inode.h"
Nathan Scotta844f452005-11-02 14:38:42 +110035#include "xfs_alloc.h"
36#include "xfs_btree.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include "xfs_error.h"
38#include "xfs_rw.h"
39#include "xfs_iomap.h"
40#include <linux/mpage.h>
Christoph Hellwig10ce4442006-01-11 20:48:14 +110041#include <linux/pagevec.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include <linux/writeback.h>
43
Nathan Scottf51623b2006-03-14 13:26:27 +110044STATIC void
45xfs_count_page_state(
46 struct page *page,
47 int *delalloc,
48 int *unmapped,
49 int *unwritten)
50{
51 struct buffer_head *bh, *head;
52
53 *delalloc = *unmapped = *unwritten = 0;
54
55 bh = head = page_buffers(page);
56 do {
57 if (buffer_uptodate(bh) && !buffer_mapped(bh))
58 (*unmapped) = 1;
Nathan Scottf51623b2006-03-14 13:26:27 +110059 else if (buffer_unwritten(bh))
60 (*unwritten) = 1;
61 else if (buffer_delay(bh))
62 (*delalloc) = 1;
63 } while ((bh = bh->b_this_page) != head);
64}
65
Linus Torvalds1da177e2005-04-16 15:20:36 -070066#if defined(XFS_RW_TRACE)
67void
68xfs_page_trace(
69 int tag,
70 struct inode *inode,
71 struct page *page,
Nathan Scotted9d88f2006-09-28 10:56:43 +100072 unsigned long pgoff)
Linus Torvalds1da177e2005-04-16 15:20:36 -070073{
74 xfs_inode_t *ip;
Nathan Scott67fcaa72006-06-09 17:00:52 +100075 bhv_vnode_t *vp = vn_from_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -070076 loff_t isize = i_size_read(inode);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +110077 loff_t offset = page_offset(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 int delalloc = -1, unmapped = -1, unwritten = -1;
79
80 if (page_has_buffers(page))
81 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
82
Christoph Hellwig75e17b32006-01-11 20:58:44 +110083 ip = xfs_vtoi(vp);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 if (!ip->i_rwtrace)
85 return;
86
87 ktrace_enter(ip->i_rwtrace,
88 (void *)((unsigned long)tag),
89 (void *)ip,
90 (void *)inode,
91 (void *)page,
Nathan Scotted9d88f2006-09-28 10:56:43 +100092 (void *)pgoff,
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 (void *)((unsigned long)((ip->i_d.di_size >> 32) & 0xffffffff)),
94 (void *)((unsigned long)(ip->i_d.di_size & 0xffffffff)),
95 (void *)((unsigned long)((isize >> 32) & 0xffffffff)),
96 (void *)((unsigned long)(isize & 0xffffffff)),
97 (void *)((unsigned long)((offset >> 32) & 0xffffffff)),
98 (void *)((unsigned long)(offset & 0xffffffff)),
99 (void *)((unsigned long)delalloc),
100 (void *)((unsigned long)unmapped),
101 (void *)((unsigned long)unwritten),
Yingping Luf1fdc842006-03-22 12:44:15 +1100102 (void *)((unsigned long)current_pid()),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 (void *)NULL);
104}
105#else
Nathan Scotted9d88f2006-09-28 10:56:43 +1000106#define xfs_page_trace(tag, inode, page, pgoff)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700107#endif
108
Christoph Hellwig0829c362005-09-02 16:58:49 +1000109/*
110 * Schedule IO completion handling on a xfsdatad if this was
David Chinnere927af92007-06-05 16:24:36 +1000111 * the final hold on this ioend. If we are asked to wait,
112 * flush the workqueue.
Christoph Hellwig0829c362005-09-02 16:58:49 +1000113 */
114STATIC void
115xfs_finish_ioend(
David Chinnere927af92007-06-05 16:24:36 +1000116 xfs_ioend_t *ioend,
117 int wait)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118{
David Chinnere927af92007-06-05 16:24:36 +1000119 if (atomic_dec_and_test(&ioend->io_remaining)) {
Christoph Hellwig0829c362005-09-02 16:58:49 +1000120 queue_work(xfsdatad_workqueue, &ioend->io_work);
David Chinnere927af92007-06-05 16:24:36 +1000121 if (wait)
122 flush_workqueue(xfsdatad_workqueue);
123 }
Christoph Hellwig0829c362005-09-02 16:58:49 +1000124}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100126/*
127 * We're now finished for good with this ioend structure.
128 * Update the page state via the associated buffer_heads,
129 * release holds on the inode and bio, and finally free
130 * up memory. Do not use the ioend after this.
131 */
Christoph Hellwig0829c362005-09-02 16:58:49 +1000132STATIC void
133xfs_destroy_ioend(
134 xfs_ioend_t *ioend)
135{
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100136 struct buffer_head *bh, *next;
137
138 for (bh = ioend->io_buffer_head; bh; bh = next) {
139 next = bh->b_private;
Nathan Scott7d04a332006-06-09 14:58:38 +1000140 bh->b_end_io(bh, !ioend->io_error);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100141 }
Nathan Scott7d04a332006-06-09 14:58:38 +1000142 if (unlikely(ioend->io_error))
143 vn_ioerror(ioend->io_vnode, ioend->io_error, __FILE__,__LINE__);
Christoph Hellwig0829c362005-09-02 16:58:49 +1000144 vn_iowake(ioend->io_vnode);
145 mempool_free(ioend, xfs_ioend_pool);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146}
147
148/*
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000149 * Update on-disk file size now that data has been written to disk.
150 * The current in-memory file size is i_size. If a write is beyond
151 * eof io_new_size will be the intended file size until i_size is
152 * updated. If this write does not extend all the way to the valid
153 * file size then restrict this update to the end of the write.
154 */
155STATIC void
156xfs_setfilesize(
157 xfs_ioend_t *ioend)
158{
159 xfs_inode_t *ip;
160 xfs_fsize_t isize;
161 xfs_fsize_t bsize;
162
163 ip = xfs_vtoi(ioend->io_vnode);
David Chinnerb2826132007-06-05 16:24:44 +1000164 if (!ip)
165 return;
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000166
167 ASSERT((ip->i_d.di_mode & S_IFMT) == S_IFREG);
168 ASSERT(ioend->io_type != IOMAP_READ);
169
170 if (unlikely(ioend->io_error))
171 return;
172
173 bsize = ioend->io_offset + ioend->io_size;
174
175 xfs_ilock(ip, XFS_ILOCK_EXCL);
176
177 isize = MAX(ip->i_size, ip->i_iocore.io_new_size);
178 isize = MIN(isize, bsize);
179
180 if (ip->i_d.di_size < isize) {
181 ip->i_d.di_size = isize;
182 ip->i_update_core = 1;
183 ip->i_update_size = 1;
Lachlan McIlroy776a75fa2007-09-14 15:22:50 +1000184 mark_inode_dirty_sync(vn_to_inode(ioend->io_vnode));
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000185 }
186
187 xfs_iunlock(ip, XFS_ILOCK_EXCL);
188}
189
190/*
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100191 * Buffered IO write completion for delayed allocate extents.
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100192 */
193STATIC void
194xfs_end_bio_delalloc(
David Howellsc4028952006-11-22 14:57:56 +0000195 struct work_struct *work)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100196{
David Howellsc4028952006-11-22 14:57:56 +0000197 xfs_ioend_t *ioend =
198 container_of(work, xfs_ioend_t, io_work);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100199
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000200 xfs_setfilesize(ioend);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100201 xfs_destroy_ioend(ioend);
202}
203
204/*
205 * Buffered IO write completion for regular, written extents.
206 */
207STATIC void
208xfs_end_bio_written(
David Howellsc4028952006-11-22 14:57:56 +0000209 struct work_struct *work)
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100210{
David Howellsc4028952006-11-22 14:57:56 +0000211 xfs_ioend_t *ioend =
212 container_of(work, xfs_ioend_t, io_work);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100213
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000214 xfs_setfilesize(ioend);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100215 xfs_destroy_ioend(ioend);
216}
217
218/*
219 * IO write completion for unwritten extents.
220 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 * Issue transactions to convert a buffer range from unwritten
Christoph Hellwigf0973862005-09-05 08:22:52 +1000222 * to written extents.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 */
224STATIC void
Christoph Hellwig0829c362005-09-02 16:58:49 +1000225xfs_end_bio_unwritten(
David Howellsc4028952006-11-22 14:57:56 +0000226 struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227{
David Howellsc4028952006-11-22 14:57:56 +0000228 xfs_ioend_t *ioend =
229 container_of(work, xfs_ioend_t, io_work);
Nathan Scott67fcaa72006-06-09 17:00:52 +1000230 bhv_vnode_t *vp = ioend->io_vnode;
Christoph Hellwig0829c362005-09-02 16:58:49 +1000231 xfs_off_t offset = ioend->io_offset;
232 size_t size = ioend->io_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000234 if (likely(!ioend->io_error)) {
Nathan Scott67fcaa72006-06-09 17:00:52 +1000235 bhv_vop_bmap(vp, offset, size, BMAPI_UNWRITTEN, NULL, NULL);
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000236 xfs_setfilesize(ioend);
237 }
238 xfs_destroy_ioend(ioend);
239}
240
241/*
242 * IO read completion for regular, written extents.
243 */
244STATIC void
245xfs_end_bio_read(
246 struct work_struct *work)
247{
248 xfs_ioend_t *ioend =
249 container_of(work, xfs_ioend_t, io_work);
250
Christoph Hellwig0829c362005-09-02 16:58:49 +1000251 xfs_destroy_ioend(ioend);
252}
253
254/*
255 * Allocate and initialise an IO completion structure.
256 * We need to track unwritten extent write completion here initially.
257 * We'll need to extend this for updating the ondisk inode size later
258 * (vs. incore size).
259 */
260STATIC xfs_ioend_t *
261xfs_alloc_ioend(
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100262 struct inode *inode,
263 unsigned int type)
Christoph Hellwig0829c362005-09-02 16:58:49 +1000264{
265 xfs_ioend_t *ioend;
266
267 ioend = mempool_alloc(xfs_ioend_pool, GFP_NOFS);
268
269 /*
270 * Set the count to 1 initially, which will prevent an I/O
271 * completion callback from happening before we have started
272 * all the I/O from calling the completion routine too early.
273 */
274 atomic_set(&ioend->io_remaining, 1);
Nathan Scott7d04a332006-06-09 14:58:38 +1000275 ioend->io_error = 0;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100276 ioend->io_list = NULL;
277 ioend->io_type = type;
Nathan Scottec86dc02006-03-17 17:25:36 +1100278 ioend->io_vnode = vn_from_inode(inode);
Christoph Hellwigc1a073b2005-09-05 08:23:35 +1000279 ioend->io_buffer_head = NULL;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100280 ioend->io_buffer_tail = NULL;
Christoph Hellwig0829c362005-09-02 16:58:49 +1000281 atomic_inc(&ioend->io_vnode->v_iocount);
282 ioend->io_offset = 0;
283 ioend->io_size = 0;
284
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100285 if (type == IOMAP_UNWRITTEN)
David Howellsc4028952006-11-22 14:57:56 +0000286 INIT_WORK(&ioend->io_work, xfs_end_bio_unwritten);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100287 else if (type == IOMAP_DELAY)
David Howellsc4028952006-11-22 14:57:56 +0000288 INIT_WORK(&ioend->io_work, xfs_end_bio_delalloc);
Lachlan McIlroyba87ea62007-05-08 13:49:46 +1000289 else if (type == IOMAP_READ)
290 INIT_WORK(&ioend->io_work, xfs_end_bio_read);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100291 else
David Howellsc4028952006-11-22 14:57:56 +0000292 INIT_WORK(&ioend->io_work, xfs_end_bio_written);
Christoph Hellwig0829c362005-09-02 16:58:49 +1000293
294 return ioend;
295}
296
Linus Torvalds1da177e2005-04-16 15:20:36 -0700297STATIC int
298xfs_map_blocks(
299 struct inode *inode,
300 loff_t offset,
301 ssize_t count,
302 xfs_iomap_t *mapp,
303 int flags)
304{
Nathan Scott67fcaa72006-06-09 17:00:52 +1000305 bhv_vnode_t *vp = vn_from_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 int error, nmaps = 1;
307
Nathan Scott67fcaa72006-06-09 17:00:52 +1000308 error = bhv_vop_bmap(vp, offset, count, flags, mapp, &nmaps);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 if (!error && (flags & (BMAPI_WRITE|BMAPI_ALLOCATE)))
310 VMODIFY(vp);
311 return -error;
312}
313
David Chinner7989cb82007-02-10 18:34:56 +1100314STATIC_INLINE int
Christoph Hellwig1defeac2006-01-11 20:48:33 +1100315xfs_iomap_valid(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 xfs_iomap_t *iomapp,
Christoph Hellwig1defeac2006-01-11 20:48:33 +1100317 loff_t offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318{
Christoph Hellwig1defeac2006-01-11 20:48:33 +1100319 return offset >= iomapp->iomap_offset &&
320 offset < iomapp->iomap_offset + iomapp->iomap_bsize;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321}
322
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100323/*
324 * BIO completion handler for buffered IO.
325 */
Al Viro782e3b32007-10-12 07:17:47 +0100326STATIC void
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100327xfs_end_bio(
328 struct bio *bio,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100329 int error)
330{
331 xfs_ioend_t *ioend = bio->bi_private;
332
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100333 ASSERT(atomic_read(&bio->bi_cnt) >= 1);
Nathan Scott7d04a332006-06-09 14:58:38 +1000334 ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100335
336 /* Toss bio and pass work off to an xfsdatad thread */
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100337 bio->bi_private = NULL;
338 bio->bi_end_io = NULL;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100339 bio_put(bio);
Nathan Scott7d04a332006-06-09 14:58:38 +1000340
David Chinnere927af92007-06-05 16:24:36 +1000341 xfs_finish_ioend(ioend, 0);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100342}
343
344STATIC void
345xfs_submit_ioend_bio(
346 xfs_ioend_t *ioend,
347 struct bio *bio)
348{
349 atomic_inc(&ioend->io_remaining);
350
351 bio->bi_private = ioend;
352 bio->bi_end_io = xfs_end_bio;
353
354 submit_bio(WRITE, bio);
355 ASSERT(!bio_flagged(bio, BIO_EOPNOTSUPP));
356 bio_put(bio);
357}
358
359STATIC struct bio *
360xfs_alloc_ioend_bio(
361 struct buffer_head *bh)
362{
363 struct bio *bio;
364 int nvecs = bio_get_nr_vecs(bh->b_bdev);
365
366 do {
367 bio = bio_alloc(GFP_NOIO, nvecs);
368 nvecs >>= 1;
369 } while (!bio);
370
371 ASSERT(bio->bi_private == NULL);
372 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
373 bio->bi_bdev = bh->b_bdev;
374 bio_get(bio);
375 return bio;
376}
377
378STATIC void
379xfs_start_buffer_writeback(
380 struct buffer_head *bh)
381{
382 ASSERT(buffer_mapped(bh));
383 ASSERT(buffer_locked(bh));
384 ASSERT(!buffer_delay(bh));
385 ASSERT(!buffer_unwritten(bh));
386
387 mark_buffer_async_write(bh);
388 set_buffer_uptodate(bh);
389 clear_buffer_dirty(bh);
390}
391
392STATIC void
393xfs_start_page_writeback(
394 struct page *page,
395 struct writeback_control *wbc,
396 int clear_dirty,
397 int buffers)
398{
399 ASSERT(PageLocked(page));
400 ASSERT(!PageWriteback(page));
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100401 if (clear_dirty)
David Chinner92132022006-12-21 10:24:01 +1100402 clear_page_dirty_for_io(page);
403 set_page_writeback(page);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100404 unlock_page(page);
405 if (!buffers) {
406 end_page_writeback(page);
407 wbc->pages_skipped++; /* We didn't write this page */
408 }
409}
410
411static inline int bio_add_buffer(struct bio *bio, struct buffer_head *bh)
412{
413 return bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
414}
415
416/*
David Chinnerd88992f2006-01-18 13:38:12 +1100417 * Submit all of the bios for all of the ioends we have saved up, covering the
418 * initial writepage page and also any probed pages.
419 *
420 * Because we may have multiple ioends spanning a page, we need to start
421 * writeback on all the buffers before we submit them for I/O. If we mark the
422 * buffers as we got, then we can end up with a page that only has buffers
423 * marked async write and I/O complete on can occur before we mark the other
424 * buffers async write.
425 *
426 * The end result of this is that we trip a bug in end_page_writeback() because
427 * we call it twice for the one page as the code in end_buffer_async_write()
428 * assumes that all buffers on the page are started at the same time.
429 *
430 * The fix is two passes across the ioend list - one to start writeback on the
Nathan Scottc41564b2006-03-29 08:55:14 +1000431 * buffer_heads, and then submit them for I/O on the second pass.
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100432 */
433STATIC void
434xfs_submit_ioend(
435 xfs_ioend_t *ioend)
436{
David Chinnerd88992f2006-01-18 13:38:12 +1100437 xfs_ioend_t *head = ioend;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100438 xfs_ioend_t *next;
439 struct buffer_head *bh;
440 struct bio *bio;
441 sector_t lastblock = 0;
442
David Chinnerd88992f2006-01-18 13:38:12 +1100443 /* Pass 1 - start writeback */
444 do {
445 next = ioend->io_list;
446 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
447 xfs_start_buffer_writeback(bh);
448 }
449 } while ((ioend = next) != NULL);
450
451 /* Pass 2 - submit I/O */
452 ioend = head;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100453 do {
454 next = ioend->io_list;
455 bio = NULL;
456
457 for (bh = ioend->io_buffer_head; bh; bh = bh->b_private) {
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100458
459 if (!bio) {
460 retry:
461 bio = xfs_alloc_ioend_bio(bh);
462 } else if (bh->b_blocknr != lastblock + 1) {
463 xfs_submit_ioend_bio(ioend, bio);
464 goto retry;
465 }
466
467 if (bio_add_buffer(bio, bh) != bh->b_size) {
468 xfs_submit_ioend_bio(ioend, bio);
469 goto retry;
470 }
471
472 lastblock = bh->b_blocknr;
473 }
474 if (bio)
475 xfs_submit_ioend_bio(ioend, bio);
David Chinnere927af92007-06-05 16:24:36 +1000476 xfs_finish_ioend(ioend, 0);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100477 } while ((ioend = next) != NULL);
478}
479
480/*
481 * Cancel submission of all buffer_heads so far in this endio.
482 * Toss the endio too. Only ever called for the initial page
483 * in a writepage request, so only ever one page.
484 */
485STATIC void
486xfs_cancel_ioend(
487 xfs_ioend_t *ioend)
488{
489 xfs_ioend_t *next;
490 struct buffer_head *bh, *next_bh;
491
492 do {
493 next = ioend->io_list;
494 bh = ioend->io_buffer_head;
495 do {
496 next_bh = bh->b_private;
497 clear_buffer_async_write(bh);
498 unlock_buffer(bh);
499 } while ((bh = next_bh) != NULL);
500
501 vn_iowake(ioend->io_vnode);
502 mempool_free(ioend, xfs_ioend_pool);
503 } while ((ioend = next) != NULL);
504}
505
506/*
507 * Test to see if we've been building up a completion structure for
508 * earlier buffers -- if so, we try to append to this ioend if we
509 * can, otherwise we finish off any current ioend and start another.
510 * Return true if we've finished the given ioend.
511 */
512STATIC void
513xfs_add_to_ioend(
514 struct inode *inode,
515 struct buffer_head *bh,
Christoph Hellwig7336cea2006-01-11 20:49:16 +1100516 xfs_off_t offset,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100517 unsigned int type,
518 xfs_ioend_t **result,
519 int need_ioend)
520{
521 xfs_ioend_t *ioend = *result;
522
523 if (!ioend || need_ioend || type != ioend->io_type) {
524 xfs_ioend_t *previous = *result;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100525
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100526 ioend = xfs_alloc_ioend(inode, type);
527 ioend->io_offset = offset;
528 ioend->io_buffer_head = bh;
529 ioend->io_buffer_tail = bh;
530 if (previous)
531 previous->io_list = ioend;
532 *result = ioend;
533 } else {
534 ioend->io_buffer_tail->b_private = bh;
535 ioend->io_buffer_tail = bh;
536 }
537
538 bh->b_private = NULL;
539 ioend->io_size += bh->b_size;
540}
541
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542STATIC void
Nathan Scott87cbc492006-03-14 13:26:43 +1100543xfs_map_buffer(
544 struct buffer_head *bh,
545 xfs_iomap_t *mp,
546 xfs_off_t offset,
547 uint block_bits)
548{
549 sector_t bn;
550
551 ASSERT(mp->iomap_bn != IOMAP_DADDR_NULL);
552
553 bn = (mp->iomap_bn >> (block_bits - BBSHIFT)) +
554 ((offset - mp->iomap_offset) >> block_bits);
555
556 ASSERT(bn || (mp->iomap_flags & IOMAP_REALTIME));
557
558 bh->b_blocknr = bn;
559 set_buffer_mapped(bh);
560}
561
562STATIC void
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563xfs_map_at_offset(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 struct buffer_head *bh,
Christoph Hellwig1defeac2006-01-11 20:48:33 +1100565 loff_t offset,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 int block_bits,
Christoph Hellwig1defeac2006-01-11 20:48:33 +1100567 xfs_iomap_t *iomapp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 ASSERT(!(iomapp->iomap_flags & IOMAP_HOLE));
570 ASSERT(!(iomapp->iomap_flags & IOMAP_DELAY));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571
572 lock_buffer(bh);
Nathan Scott87cbc492006-03-14 13:26:43 +1100573 xfs_map_buffer(bh, iomapp, offset, block_bits);
Nathan Scottce8e9222006-01-11 15:39:08 +1100574 bh->b_bdev = iomapp->iomap_target->bt_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 set_buffer_mapped(bh);
576 clear_buffer_delay(bh);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100577 clear_buffer_unwritten(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578}
579
580/*
Christoph Hellwig6c4fe192006-01-11 20:49:28 +1100581 * Look for a page at index that is suitable for clustering.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 */
583STATIC unsigned int
Christoph Hellwig6c4fe192006-01-11 20:49:28 +1100584xfs_probe_page(
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100585 struct page *page,
Christoph Hellwig6c4fe192006-01-11 20:49:28 +1100586 unsigned int pg_offset,
587 int mapped)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 int ret = 0;
590
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 if (PageWriteback(page))
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100592 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593
594 if (page->mapping && PageDirty(page)) {
595 if (page_has_buffers(page)) {
596 struct buffer_head *bh, *head;
597
598 bh = head = page_buffers(page);
599 do {
Christoph Hellwig6c4fe192006-01-11 20:49:28 +1100600 if (!buffer_uptodate(bh))
601 break;
602 if (mapped != buffer_mapped(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603 break;
604 ret += bh->b_size;
605 if (ret >= pg_offset)
606 break;
607 } while ((bh = bh->b_this_page) != head);
608 } else
Christoph Hellwig6c4fe192006-01-11 20:49:28 +1100609 ret = mapped ? 0 : PAGE_CACHE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 }
611
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 return ret;
613}
614
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100615STATIC size_t
Christoph Hellwig6c4fe192006-01-11 20:49:28 +1100616xfs_probe_cluster(
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 struct inode *inode,
618 struct page *startpage,
619 struct buffer_head *bh,
Christoph Hellwig6c4fe192006-01-11 20:49:28 +1100620 struct buffer_head *head,
621 int mapped)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622{
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100623 struct pagevec pvec;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 pgoff_t tindex, tlast, tloff;
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100625 size_t total = 0;
626 int done = 0, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627
628 /* First sum forwards in this page */
629 do {
Eric Sandeen2353e8e2006-02-28 12:30:30 +1100630 if (!buffer_uptodate(bh) || (mapped != buffer_mapped(bh)))
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100631 return total;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 total += bh->b_size;
633 } while ((bh = bh->b_this_page) != head);
634
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100635 /* if we reached the end of the page, sum forwards in following pages */
636 tlast = i_size_read(inode) >> PAGE_CACHE_SHIFT;
637 tindex = startpage->index + 1;
638
639 /* Prune this back to avoid pathological behavior */
640 tloff = min(tlast, startpage->index + 64);
641
642 pagevec_init(&pvec, 0);
643 while (!done && tindex <= tloff) {
644 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
645
646 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
647 break;
648
649 for (i = 0; i < pagevec_count(&pvec); i++) {
650 struct page *page = pvec.pages[i];
Christoph Hellwig265c1fa2007-08-16 15:38:19 +1000651 size_t pg_offset, pg_len = 0;
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100652
653 if (tindex == tlast) {
654 pg_offset =
655 i_size_read(inode) & (PAGE_CACHE_SIZE - 1);
Christoph Hellwig1defeac2006-01-11 20:48:33 +1100656 if (!pg_offset) {
657 done = 1;
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100658 break;
Christoph Hellwig1defeac2006-01-11 20:48:33 +1100659 }
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100660 } else
661 pg_offset = PAGE_CACHE_SIZE;
662
663 if (page->index == tindex && !TestSetPageLocked(page)) {
Christoph Hellwig265c1fa2007-08-16 15:38:19 +1000664 pg_len = xfs_probe_page(page, pg_offset, mapped);
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100665 unlock_page(page);
666 }
667
Christoph Hellwig265c1fa2007-08-16 15:38:19 +1000668 if (!pg_len) {
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100669 done = 1;
670 break;
671 }
672
Christoph Hellwig265c1fa2007-08-16 15:38:19 +1000673 total += pg_len;
Christoph Hellwig1defeac2006-01-11 20:48:33 +1100674 tindex++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 }
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100676
677 pagevec_release(&pvec);
678 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679 }
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100680
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 return total;
682}
683
684/*
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100685 * Test if a given page is suitable for writing as part of an unwritten
686 * or delayed allocate extent.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 */
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100688STATIC int
689xfs_is_delayed_page(
690 struct page *page,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100691 unsigned int type)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 if (PageWriteback(page))
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100694 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695
696 if (page->mapping && page_has_buffers(page)) {
697 struct buffer_head *bh, *head;
698 int acceptable = 0;
699
700 bh = head = page_buffers(page);
701 do {
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100702 if (buffer_unwritten(bh))
703 acceptable = (type == IOMAP_UNWRITTEN);
704 else if (buffer_delay(bh))
705 acceptable = (type == IOMAP_DELAY);
David Chinner2ddee842006-03-22 12:47:40 +1100706 else if (buffer_dirty(bh) && buffer_mapped(bh))
David Chinnerdf3c7242007-05-24 15:27:03 +1000707 acceptable = (type == IOMAP_NEW);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100708 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 } while ((bh = bh->b_this_page) != head);
711
712 if (acceptable)
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100713 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714 }
715
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100716 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717}
718
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719/*
720 * Allocate & map buffers for page given the extent map. Write it out.
721 * except for the original page of a writepage, this is called on
722 * delalloc/unwritten pages only, for the original page it is possible
723 * that the page has no mapping at all.
724 */
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100725STATIC int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726xfs_convert_page(
727 struct inode *inode,
728 struct page *page,
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100729 loff_t tindex,
Christoph Hellwig1defeac2006-01-11 20:48:33 +1100730 xfs_iomap_t *mp,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100731 xfs_ioend_t **ioendp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732 struct writeback_control *wbc,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 int startio,
734 int all_bh)
735{
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100736 struct buffer_head *bh, *head;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100737 xfs_off_t end_offset;
738 unsigned long p_offset;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100739 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700740 int bbits = inode->i_blkbits;
Nathan Scott24e17b52005-05-05 13:33:20 -0700741 int len, page_dirty;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100742 int count = 0, done = 0, uptodate = 1;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100743 xfs_off_t offset = page_offset(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100745 if (page->index != tindex)
746 goto fail;
747 if (TestSetPageLocked(page))
748 goto fail;
749 if (PageWriteback(page))
750 goto fail_unlock_page;
751 if (page->mapping != inode->i_mapping)
752 goto fail_unlock_page;
753 if (!xfs_is_delayed_page(page, (*ioendp)->io_type))
754 goto fail_unlock_page;
755
Nathan Scott24e17b52005-05-05 13:33:20 -0700756 /*
757 * page_dirty is initially a count of buffers on the page before
Nathan Scottc41564b2006-03-29 08:55:14 +1000758 * EOF and is decremented as we move each into a cleanable state.
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100759 *
760 * Derivation:
761 *
762 * End offset is the highest offset that this page should represent.
763 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
764 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
765 * hence give us the correct page_dirty count. On any other page,
766 * it will be zero and in that case we need page_dirty to be the
767 * count of buffers on the page.
Nathan Scott24e17b52005-05-05 13:33:20 -0700768 */
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100769 end_offset = min_t(unsigned long long,
770 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT,
771 i_size_read(inode));
772
Nathan Scott24e17b52005-05-05 13:33:20 -0700773 len = 1 << inode->i_blkbits;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100774 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
775 PAGE_CACHE_SIZE);
776 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
777 page_dirty = p_offset / len;
Nathan Scott24e17b52005-05-05 13:33:20 -0700778
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779 bh = head = page_buffers(page);
780 do {
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100781 if (offset >= end_offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700782 break;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100783 if (!buffer_uptodate(bh))
784 uptodate = 0;
785 if (!(PageUptodate(page) || buffer_uptodate(bh))) {
786 done = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 continue;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100788 }
789
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100790 if (buffer_unwritten(bh) || buffer_delay(bh)) {
791 if (buffer_unwritten(bh))
792 type = IOMAP_UNWRITTEN;
793 else
794 type = IOMAP_DELAY;
795
796 if (!xfs_iomap_valid(mp, offset)) {
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100797 done = 1;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100798 continue;
799 }
800
801 ASSERT(!(mp->iomap_flags & IOMAP_HOLE));
802 ASSERT(!(mp->iomap_flags & IOMAP_DELAY));
803
804 xfs_map_at_offset(bh, offset, bbits, mp);
805 if (startio) {
Christoph Hellwig7336cea2006-01-11 20:49:16 +1100806 xfs_add_to_ioend(inode, bh, offset,
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100807 type, ioendp, done);
808 } else {
809 set_buffer_dirty(bh);
810 unlock_buffer(bh);
811 mark_buffer_dirty(bh);
812 }
813 page_dirty--;
814 count++;
815 } else {
David Chinnerdf3c7242007-05-24 15:27:03 +1000816 type = IOMAP_NEW;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100817 if (buffer_mapped(bh) && all_bh && startio) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818 lock_buffer(bh);
Christoph Hellwig7336cea2006-01-11 20:49:16 +1100819 xfs_add_to_ioend(inode, bh, offset,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100820 type, ioendp, done);
821 count++;
Nathan Scott24e17b52005-05-05 13:33:20 -0700822 page_dirty--;
Christoph Hellwig9260dc62006-01-11 20:48:47 +1100823 } else {
824 done = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 }
Christoph Hellwig7336cea2006-01-11 20:49:16 +1100827 } while (offset += len, (bh = bh->b_this_page) != head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100829 if (uptodate && bh == head)
830 SetPageUptodate(page);
831
832 if (startio) {
Christoph Hellwigf5e596b2006-01-11 20:49:42 +1100833 if (count) {
834 struct backing_dev_info *bdi;
835
836 bdi = inode->i_mapping->backing_dev_info;
David Chinner9fddaca2006-02-07 20:27:24 +1100837 wbc->nr_to_write--;
Christoph Hellwigf5e596b2006-01-11 20:49:42 +1100838 if (bdi_write_congested(bdi)) {
839 wbc->encountered_congestion = 1;
840 done = 1;
David Chinner9fddaca2006-02-07 20:27:24 +1100841 } else if (wbc->nr_to_write <= 0) {
Christoph Hellwigf5e596b2006-01-11 20:49:42 +1100842 done = 1;
843 }
844 }
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100845 xfs_start_page_writeback(page, wbc, !page_dirty, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700846 }
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100847
848 return done;
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100849 fail_unlock_page:
850 unlock_page(page);
851 fail:
852 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853}
854
855/*
856 * Convert & write out a cluster of pages in the same extent as defined
857 * by mp and following the start page.
858 */
859STATIC void
860xfs_cluster_write(
861 struct inode *inode,
862 pgoff_t tindex,
863 xfs_iomap_t *iomapp,
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100864 xfs_ioend_t **ioendp,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700865 struct writeback_control *wbc,
866 int startio,
867 int all_bh,
868 pgoff_t tlast)
869{
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100870 struct pagevec pvec;
871 int done = 0, i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100873 pagevec_init(&pvec, 0);
874 while (!done && tindex <= tlast) {
875 unsigned len = min_t(pgoff_t, PAGEVEC_SIZE, tlast - tindex + 1);
876
877 if (!pagevec_lookup(&pvec, inode->i_mapping, tindex, len))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 break;
Christoph Hellwig10ce4442006-01-11 20:48:14 +1100879
880 for (i = 0; i < pagevec_count(&pvec); i++) {
881 done = xfs_convert_page(inode, pvec.pages[i], tindex++,
882 iomapp, ioendp, wbc, startio, all_bh);
883 if (done)
884 break;
885 }
886
887 pagevec_release(&pvec);
888 cond_resched();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 }
890}
891
892/*
893 * Calling this without startio set means we are being asked to make a dirty
894 * page ready for freeing it's buffers. When called with startio set then
895 * we are coming from writepage.
896 *
897 * When called with startio set it is important that we write the WHOLE
898 * page if possible.
899 * The bh->b_state's cannot know if any of the blocks or which block for
900 * that matter are dirty due to mmap writes, and therefore bh uptodate is
Nathan Scottc41564b2006-03-29 08:55:14 +1000901 * only valid if the page itself isn't completely uptodate. Some layers
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 * may clear the page dirty flag prior to calling write page, under the
903 * assumption the entire page will be written out; by not writing out the
904 * whole page the page can be reused before all valid dirty data is
905 * written out. Note: in the case of a page that has been dirty'd by
906 * mapwrite and but partially setup by block_prepare_write the
907 * bh->b_states's will not agree and only ones setup by BPW/BCW will have
908 * valid state, thus the whole page must be written out thing.
909 */
910
911STATIC int
912xfs_page_state_convert(
913 struct inode *inode,
914 struct page *page,
915 struct writeback_control *wbc,
916 int startio,
917 int unmapped) /* also implies page uptodate */
918{
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100919 struct buffer_head *bh, *head;
Christoph Hellwig1defeac2006-01-11 20:48:33 +1100920 xfs_iomap_t iomap;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100921 xfs_ioend_t *ioend = NULL, *iohead = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922 loff_t offset;
923 unsigned long p_offset = 0;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100924 unsigned int type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 __uint64_t end_offset;
926 pgoff_t end_index, last_index, tlast;
Christoph Hellwigd5cb48a2006-01-11 20:49:02 +1100927 ssize_t size, len;
928 int flags, err, iomap_valid = 0, uptodate = 1;
Nathan Scott82721452006-04-11 15:10:55 +1000929 int page_dirty, count = 0;
930 int trylock = 0;
Christoph Hellwig6c4fe192006-01-11 20:49:28 +1100931 int all_bh = unmapped;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932
Nathan Scott82721452006-04-11 15:10:55 +1000933 if (startio) {
934 if (wbc->sync_mode == WB_SYNC_NONE && wbc->nonblocking)
935 trylock |= BMAPI_TRYLOCK;
936 }
Daniel Moore3ba08152005-05-05 13:31:34 -0700937
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938 /* Is this page beyond the end of the file? */
939 offset = i_size_read(inode);
940 end_index = offset >> PAGE_CACHE_SHIFT;
941 last_index = (offset - 1) >> PAGE_CACHE_SHIFT;
942 if (page->index >= end_index) {
943 if ((page->index >= end_index + 1) ||
944 !(i_size_read(inode) & (PAGE_CACHE_SIZE - 1))) {
Nathan Scott19d5bcf2005-11-02 15:14:09 +1100945 if (startio)
946 unlock_page(page);
947 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 }
949 }
950
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 /*
Nathan Scott24e17b52005-05-05 13:33:20 -0700952 * page_dirty is initially a count of buffers on the page before
Nathan Scottc41564b2006-03-29 08:55:14 +1000953 * EOF and is decremented as we move each into a cleanable state.
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100954 *
955 * Derivation:
956 *
957 * End offset is the highest offset that this page should represent.
958 * If we are on the last page, (end_offset & (PAGE_CACHE_SIZE - 1))
959 * will evaluate non-zero and be less than PAGE_CACHE_SIZE and
960 * hence give us the correct page_dirty count. On any other page,
961 * it will be zero and in that case we need page_dirty to be the
962 * count of buffers on the page.
963 */
964 end_offset = min_t(unsigned long long,
965 (xfs_off_t)(page->index + 1) << PAGE_CACHE_SHIFT, offset);
Nathan Scott24e17b52005-05-05 13:33:20 -0700966 len = 1 << inode->i_blkbits;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100967 p_offset = min_t(unsigned long, end_offset & (PAGE_CACHE_SIZE - 1),
968 PAGE_CACHE_SIZE);
969 p_offset = p_offset ? roundup(p_offset, len) : PAGE_CACHE_SIZE;
Nathan Scott24e17b52005-05-05 13:33:20 -0700970 page_dirty = p_offset / len;
971
Nathan Scott24e17b52005-05-05 13:33:20 -0700972 bh = head = page_buffers(page);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100973 offset = page_offset(page);
David Chinnerdf3c7242007-05-24 15:27:03 +1000974 flags = BMAPI_READ;
975 type = IOMAP_NEW;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100976
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100977 /* TODO: cleanup count and page_dirty */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978
979 do {
980 if (offset >= end_offset)
981 break;
982 if (!buffer_uptodate(bh))
983 uptodate = 0;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100984 if (!(PageUptodate(page) || buffer_uptodate(bh)) && !startio) {
Christoph Hellwig1defeac2006-01-11 20:48:33 +1100985 /*
986 * the iomap is actually still valid, but the ioend
987 * isn't. shouldn't happen too often.
988 */
989 iomap_valid = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 continue;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100991 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992
Christoph Hellwig1defeac2006-01-11 20:48:33 +1100993 if (iomap_valid)
994 iomap_valid = xfs_iomap_valid(&iomap, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995
996 /*
997 * First case, map an unwritten extent and prepare for
998 * extent state conversion transaction on completion.
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +1100999 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 * Second case, allocate space for a delalloc buffer.
1001 * We can return EAGAIN here in the release page case.
Christoph Hellwigd5cb48a2006-01-11 20:49:02 +11001002 *
1003 * Third case, an unmapped buffer was found, and we are
1004 * in a path where we need to write the whole page out.
David Chinnerdf3c7242007-05-24 15:27:03 +10001005 */
Christoph Hellwigd5cb48a2006-01-11 20:49:02 +11001006 if (buffer_unwritten(bh) || buffer_delay(bh) ||
1007 ((buffer_uptodate(bh) || PageUptodate(page)) &&
1008 !buffer_mapped(bh) && (unmapped || startio))) {
David Chinnereffd1202007-06-18 16:49:58 +10001009 int new_ioend = 0;
1010
David Chinnerdf3c7242007-05-24 15:27:03 +10001011 /*
Christoph Hellwig6c4fe192006-01-11 20:49:28 +11001012 * Make sure we don't use a read-only iomap
1013 */
David Chinnerdf3c7242007-05-24 15:27:03 +10001014 if (flags == BMAPI_READ)
Christoph Hellwig6c4fe192006-01-11 20:49:28 +11001015 iomap_valid = 0;
1016
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001017 if (buffer_unwritten(bh)) {
1018 type = IOMAP_UNWRITTEN;
Nathan Scott82721452006-04-11 15:10:55 +10001019 flags = BMAPI_WRITE | BMAPI_IGNSTATE;
Christoph Hellwigd5cb48a2006-01-11 20:49:02 +11001020 } else if (buffer_delay(bh)) {
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001021 type = IOMAP_DELAY;
Nathan Scott82721452006-04-11 15:10:55 +10001022 flags = BMAPI_ALLOCATE | trylock;
Christoph Hellwigd5cb48a2006-01-11 20:49:02 +11001023 } else {
Christoph Hellwig6c4fe192006-01-11 20:49:28 +11001024 type = IOMAP_NEW;
Nathan Scott82721452006-04-11 15:10:55 +10001025 flags = BMAPI_WRITE | BMAPI_MMAP;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001026 }
1027
Christoph Hellwig1defeac2006-01-11 20:48:33 +11001028 if (!iomap_valid) {
David Chinnereffd1202007-06-18 16:49:58 +10001029 /*
1030 * if we didn't have a valid mapping then we
1031 * need to ensure that we put the new mapping
1032 * in a new ioend structure. This needs to be
1033 * done to ensure that the ioends correctly
1034 * reflect the block mappings at io completion
1035 * for unwritten extent conversion.
1036 */
1037 new_ioend = 1;
Christoph Hellwig6c4fe192006-01-11 20:49:28 +11001038 if (type == IOMAP_NEW) {
1039 size = xfs_probe_cluster(inode,
1040 page, bh, head, 0);
Christoph Hellwigd5cb48a2006-01-11 20:49:02 +11001041 } else {
1042 size = len;
1043 }
1044
1045 err = xfs_map_blocks(inode, offset, size,
1046 &iomap, flags);
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001047 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048 goto error;
Christoph Hellwig1defeac2006-01-11 20:48:33 +11001049 iomap_valid = xfs_iomap_valid(&iomap, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 }
Christoph Hellwig1defeac2006-01-11 20:48:33 +11001051 if (iomap_valid) {
1052 xfs_map_at_offset(bh, offset,
1053 inode->i_blkbits, &iomap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 if (startio) {
Christoph Hellwig7336cea2006-01-11 20:49:16 +11001055 xfs_add_to_ioend(inode, bh, offset,
Christoph Hellwig1defeac2006-01-11 20:48:33 +11001056 type, &ioend,
David Chinnereffd1202007-06-18 16:49:58 +10001057 new_ioend);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 } else {
1059 set_buffer_dirty(bh);
1060 unlock_buffer(bh);
1061 mark_buffer_dirty(bh);
1062 }
1063 page_dirty--;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001064 count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 }
Christoph Hellwigd5cb48a2006-01-11 20:49:02 +11001066 } else if (buffer_uptodate(bh) && startio) {
Christoph Hellwig6c4fe192006-01-11 20:49:28 +11001067 /*
1068 * we got here because the buffer is already mapped.
1069 * That means it must already have extents allocated
1070 * underneath it. Map the extent by reading it.
1071 */
David Chinnerdf3c7242007-05-24 15:27:03 +10001072 if (!iomap_valid || flags != BMAPI_READ) {
Christoph Hellwig6c4fe192006-01-11 20:49:28 +11001073 flags = BMAPI_READ;
1074 size = xfs_probe_cluster(inode, page, bh,
1075 head, 1);
1076 err = xfs_map_blocks(inode, offset, size,
1077 &iomap, flags);
1078 if (err)
1079 goto error;
1080 iomap_valid = xfs_iomap_valid(&iomap, offset);
1081 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082
David Chinnerdf3c7242007-05-24 15:27:03 +10001083 /*
1084 * We set the type to IOMAP_NEW in case we are doing a
1085 * small write at EOF that is extending the file but
1086 * without needing an allocation. We need to update the
1087 * file size on I/O completion in this case so it is
1088 * the same case as having just allocated a new extent
1089 * that we are writing into for the first time.
1090 */
1091 type = IOMAP_NEW;
Christoph Hellwigd5cb48a2006-01-11 20:49:02 +11001092 if (!test_and_set_bit(BH_Lock, &bh->b_state)) {
1093 ASSERT(buffer_mapped(bh));
Christoph Hellwig6c4fe192006-01-11 20:49:28 +11001094 if (iomap_valid)
1095 all_bh = 1;
Christoph Hellwig7336cea2006-01-11 20:49:16 +11001096 xfs_add_to_ioend(inode, bh, offset, type,
Christoph Hellwigd5cb48a2006-01-11 20:49:02 +11001097 &ioend, !iomap_valid);
1098 page_dirty--;
1099 count++;
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001100 } else {
Christoph Hellwig1defeac2006-01-11 20:48:33 +11001101 iomap_valid = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102 }
Christoph Hellwigd5cb48a2006-01-11 20:49:02 +11001103 } else if ((buffer_uptodate(bh) || PageUptodate(page)) &&
1104 (unmapped || startio)) {
1105 iomap_valid = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 }
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001107
1108 if (!iohead)
1109 iohead = ioend;
1110
1111 } while (offset += len, ((bh = bh->b_this_page) != head));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112
1113 if (uptodate && bh == head)
1114 SetPageUptodate(page);
1115
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001116 if (startio)
1117 xfs_start_page_writeback(page, wbc, 1, count);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118
Christoph Hellwig1defeac2006-01-11 20:48:33 +11001119 if (ioend && iomap_valid) {
1120 offset = (iomap.iomap_offset + iomap.iomap_bsize - 1) >>
Linus Torvalds1da177e2005-04-16 15:20:36 -07001121 PAGE_CACHE_SHIFT;
Nathan Scott775bf6c2005-05-05 13:33:01 -07001122 tlast = min_t(pgoff_t, offset, last_index);
Christoph Hellwig1defeac2006-01-11 20:48:33 +11001123 xfs_cluster_write(inode, page->index + 1, &iomap, &ioend,
Christoph Hellwig6c4fe192006-01-11 20:49:28 +11001124 wbc, startio, all_bh, tlast);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 }
1126
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001127 if (iohead)
1128 xfs_submit_ioend(iohead);
1129
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 return page_dirty;
1131
1132error:
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001133 if (iohead)
1134 xfs_cancel_ioend(iohead);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135
1136 /*
1137 * If it's delalloc and we have nowhere to put it,
1138 * throw it away, unless the lower layers told
1139 * us to try again.
1140 */
1141 if (err != -EAGAIN) {
Christoph Hellwigf6d6d4f2006-01-11 15:40:13 +11001142 if (!unmapped)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 block_invalidatepage(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 ClearPageUptodate(page);
1145 }
1146 return err;
1147}
1148
Nathan Scottf51623b2006-03-14 13:26:27 +11001149/*
1150 * writepage: Called from one of two places:
1151 *
1152 * 1. we are flushing a delalloc buffer head.
1153 *
1154 * 2. we are writing out a dirty page. Typically the page dirty
1155 * state is cleared before we get here. In this case is it
1156 * conceivable we have no buffer heads.
1157 *
1158 * For delalloc space on the page we need to allocate space and
1159 * flush it. For unmapped buffer heads on the page we should
1160 * allocate space if the page is uptodate. For any other dirty
1161 * buffer heads on the page we should flush them.
1162 *
1163 * If we detect that a transaction would be required to flush
1164 * the page, we have to check the process flags first, if we
1165 * are already in a transaction or disk I/O during allocations
1166 * is off, we need to fail the writepage and redirty the page.
1167 */
1168
1169STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001170xfs_vm_writepage(
Nathan Scottf51623b2006-03-14 13:26:27 +11001171 struct page *page,
1172 struct writeback_control *wbc)
1173{
1174 int error;
1175 int need_trans;
1176 int delalloc, unmapped, unwritten;
1177 struct inode *inode = page->mapping->host;
1178
1179 xfs_page_trace(XFS_WRITEPAGE_ENTER, inode, page, 0);
1180
1181 /*
1182 * We need a transaction if:
1183 * 1. There are delalloc buffers on the page
1184 * 2. The page is uptodate and we have unmapped buffers
1185 * 3. The page is uptodate and we have no buffers
1186 * 4. There are unwritten buffers on the page
1187 */
1188
1189 if (!page_has_buffers(page)) {
1190 unmapped = 1;
1191 need_trans = 1;
1192 } else {
1193 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1194 if (!PageUptodate(page))
1195 unmapped = 0;
1196 need_trans = delalloc + unmapped + unwritten;
1197 }
1198
1199 /*
1200 * If we need a transaction and the process flags say
1201 * we are already in a transaction, or no IO is allowed
1202 * then mark the page dirty again and leave the page
1203 * as is.
1204 */
Nathan Scott59c1b082006-06-09 14:59:13 +10001205 if (current_test_flags(PF_FSTRANS) && need_trans)
Nathan Scottf51623b2006-03-14 13:26:27 +11001206 goto out_fail;
1207
1208 /*
1209 * Delay hooking up buffer heads until we have
1210 * made our go/no-go decision.
1211 */
1212 if (!page_has_buffers(page))
1213 create_empty_buffers(page, 1 << inode->i_blkbits, 0);
1214
1215 /*
1216 * Convert delayed allocate, unwritten or unmapped space
1217 * to real space and flush out to disk.
1218 */
1219 error = xfs_page_state_convert(inode, page, wbc, 1, unmapped);
1220 if (error == -EAGAIN)
1221 goto out_fail;
1222 if (unlikely(error < 0))
1223 goto out_unlock;
1224
1225 return 0;
1226
1227out_fail:
1228 redirty_page_for_writepage(wbc, page);
1229 unlock_page(page);
1230 return 0;
1231out_unlock:
1232 unlock_page(page);
1233 return error;
1234}
1235
Nathan Scott7d4fb402006-06-09 15:27:16 +10001236STATIC int
1237xfs_vm_writepages(
1238 struct address_space *mapping,
1239 struct writeback_control *wbc)
1240{
Nathan Scott67fcaa72006-06-09 17:00:52 +10001241 struct bhv_vnode *vp = vn_from_inode(mapping->host);
Nathan Scott7d4fb402006-06-09 15:27:16 +10001242
1243 if (VN_TRUNC(vp))
1244 VUNTRUNCATE(vp);
1245 return generic_writepages(mapping, wbc);
1246}
1247
Nathan Scottf51623b2006-03-14 13:26:27 +11001248/*
1249 * Called to move a page into cleanable state - and from there
1250 * to be released. Possibly the page is already clean. We always
1251 * have buffer heads in this call.
1252 *
1253 * Returns 0 if the page is ok to release, 1 otherwise.
1254 *
1255 * Possible scenarios are:
1256 *
1257 * 1. We are being called to release a page which has been written
1258 * to via regular I/O. buffer heads will be dirty and possibly
1259 * delalloc. If no delalloc buffer heads in this case then we
1260 * can just return zero.
1261 *
1262 * 2. We are called to release a page which has been written via
1263 * mmap, all we need to do is ensure there is no delalloc
1264 * state in the buffer heads, if not we can let the caller
1265 * free them and we should come back later via writepage.
1266 */
1267STATIC int
Nathan Scott238f4c52006-03-17 17:26:25 +11001268xfs_vm_releasepage(
Nathan Scottf51623b2006-03-14 13:26:27 +11001269 struct page *page,
1270 gfp_t gfp_mask)
1271{
1272 struct inode *inode = page->mapping->host;
1273 int dirty, delalloc, unmapped, unwritten;
1274 struct writeback_control wbc = {
1275 .sync_mode = WB_SYNC_ALL,
1276 .nr_to_write = 1,
1277 };
1278
Nathan Scotted9d88f2006-09-28 10:56:43 +10001279 xfs_page_trace(XFS_RELEASEPAGE_ENTER, inode, page, 0);
Nathan Scottf51623b2006-03-14 13:26:27 +11001280
Nathan Scott238f4c52006-03-17 17:26:25 +11001281 if (!page_has_buffers(page))
1282 return 0;
1283
Nathan Scottf51623b2006-03-14 13:26:27 +11001284 xfs_count_page_state(page, &delalloc, &unmapped, &unwritten);
1285 if (!delalloc && !unwritten)
1286 goto free_buffers;
1287
1288 if (!(gfp_mask & __GFP_FS))
1289 return 0;
1290
1291 /* If we are already inside a transaction or the thread cannot
1292 * do I/O, we cannot release this page.
1293 */
Nathan Scott59c1b082006-06-09 14:59:13 +10001294 if (current_test_flags(PF_FSTRANS))
Nathan Scottf51623b2006-03-14 13:26:27 +11001295 return 0;
1296
1297 /*
1298 * Convert delalloc space to real space, do not flush the
1299 * data out to disk, that will be done by the caller.
1300 * Never need to allocate space here - we will always
1301 * come back to writepage in that case.
1302 */
1303 dirty = xfs_page_state_convert(inode, page, &wbc, 0, 0);
1304 if (dirty == 0 && !unwritten)
1305 goto free_buffers;
1306 return 0;
1307
1308free_buffers:
1309 return try_to_free_buffers(page);
1310}
1311
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312STATIC int
Nathan Scottc2536662006-03-29 10:44:40 +10001313__xfs_get_blocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 struct inode *inode,
1315 sector_t iblock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 struct buffer_head *bh_result,
1317 int create,
1318 int direct,
1319 bmapi_flags_t flags)
1320{
Nathan Scott67fcaa72006-06-09 17:00:52 +10001321 bhv_vnode_t *vp = vn_from_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 xfs_iomap_t iomap;
Nathan Scottfdc7ed72005-11-02 15:13:13 +11001323 xfs_off_t offset;
1324 ssize_t size;
Nathan Scottc2536662006-03-29 10:44:40 +10001325 int niomap = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327
Nathan Scottfdc7ed72005-11-02 15:13:13 +11001328 offset = (xfs_off_t)iblock << inode->i_blkbits;
Nathan Scottc2536662006-03-29 10:44:40 +10001329 ASSERT(bh_result->b_size >= (1 << inode->i_blkbits));
1330 size = bh_result->b_size;
Nathan Scott67fcaa72006-06-09 17:00:52 +10001331 error = bhv_vop_bmap(vp, offset, size,
1332 create ? flags : BMAPI_READ, &iomap, &niomap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333 if (error)
1334 return -error;
Nathan Scottc2536662006-03-29 10:44:40 +10001335 if (niomap == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 return 0;
1337
1338 if (iomap.iomap_bn != IOMAP_DADDR_NULL) {
Nathan Scott87cbc492006-03-14 13:26:43 +11001339 /*
1340 * For unwritten extents do not report a disk address on
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 * the read case (treat as if we're reading into a hole).
1342 */
1343 if (create || !(iomap.iomap_flags & IOMAP_UNWRITTEN)) {
Nathan Scott87cbc492006-03-14 13:26:43 +11001344 xfs_map_buffer(bh_result, &iomap, offset,
1345 inode->i_blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 }
1347 if (create && (iomap.iomap_flags & IOMAP_UNWRITTEN)) {
1348 if (direct)
1349 bh_result->b_private = inode;
1350 set_buffer_unwritten(bh_result);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 }
1352 }
1353
Nathan Scottc2536662006-03-29 10:44:40 +10001354 /*
1355 * If this is a realtime file, data may be on a different device.
1356 * to that pointed to from the buffer_head b_bdev currently.
1357 */
Nathan Scottce8e9222006-01-11 15:39:08 +11001358 bh_result->b_bdev = iomap.iomap_target->bt_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359
Nathan Scottc2536662006-03-29 10:44:40 +10001360 /*
David Chinner549054a2007-02-10 18:36:35 +11001361 * If we previously allocated a block out beyond eof and we are now
1362 * coming back to use it then we will need to flag it as new even if it
1363 * has a disk address.
1364 *
1365 * With sub-block writes into unwritten extents we also need to mark
1366 * the buffer as new so that the unwritten parts of the buffer gets
1367 * correctly zeroed.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001368 */
1369 if (create &&
1370 ((!buffer_mapped(bh_result) && !buffer_uptodate(bh_result)) ||
David Chinner549054a2007-02-10 18:36:35 +11001371 (offset >= i_size_read(inode)) ||
1372 (iomap.iomap_flags & (IOMAP_NEW|IOMAP_UNWRITTEN))))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 set_buffer_new(bh_result);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374
1375 if (iomap.iomap_flags & IOMAP_DELAY) {
1376 BUG_ON(direct);
1377 if (create) {
1378 set_buffer_uptodate(bh_result);
1379 set_buffer_mapped(bh_result);
1380 set_buffer_delay(bh_result);
1381 }
1382 }
1383
Nathan Scottc2536662006-03-29 10:44:40 +10001384 if (direct || size > (1 << inode->i_blkbits)) {
Nathan Scottfdc7ed72005-11-02 15:13:13 +11001385 ASSERT(iomap.iomap_bsize - iomap.iomap_delta > 0);
1386 offset = min_t(xfs_off_t,
Nathan Scottc2536662006-03-29 10:44:40 +10001387 iomap.iomap_bsize - iomap.iomap_delta, size);
1388 bh_result->b_size = (ssize_t)min_t(xfs_off_t, LONG_MAX, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 }
1390
1391 return 0;
1392}
1393
1394int
Nathan Scottc2536662006-03-29 10:44:40 +10001395xfs_get_blocks(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 struct inode *inode,
1397 sector_t iblock,
1398 struct buffer_head *bh_result,
1399 int create)
1400{
Nathan Scottc2536662006-03-29 10:44:40 +10001401 return __xfs_get_blocks(inode, iblock,
Badari Pulavartyfa30bd02006-03-26 01:38:01 -08001402 bh_result, create, 0, BMAPI_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403}
1404
1405STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001406xfs_get_blocks_direct(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 struct inode *inode,
1408 sector_t iblock,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 struct buffer_head *bh_result,
1410 int create)
1411{
Nathan Scottc2536662006-03-29 10:44:40 +10001412 return __xfs_get_blocks(inode, iblock,
Badari Pulavarty1d8fa7a2006-03-26 01:38:02 -08001413 bh_result, create, 1, BMAPI_WRITE|BMAPI_DIRECT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414}
1415
Christoph Hellwigf0973862005-09-05 08:22:52 +10001416STATIC void
Nathan Scotte4c573b2006-03-14 13:54:26 +11001417xfs_end_io_direct(
Christoph Hellwigf0973862005-09-05 08:22:52 +10001418 struct kiocb *iocb,
1419 loff_t offset,
1420 ssize_t size,
1421 void *private)
1422{
1423 xfs_ioend_t *ioend = iocb->private;
1424
1425 /*
1426 * Non-NULL private data means we need to issue a transaction to
1427 * convert a range from unwritten to written extents. This needs
Nathan Scottc41564b2006-03-29 08:55:14 +10001428 * to happen from process context but aio+dio I/O completion
Christoph Hellwigf0973862005-09-05 08:22:52 +10001429 * happens from irq context so we need to defer it to a workqueue.
Nathan Scottc41564b2006-03-29 08:55:14 +10001430 * This is not necessary for synchronous direct I/O, but we do
Christoph Hellwigf0973862005-09-05 08:22:52 +10001431 * it anyway to keep the code uniform and simpler.
1432 *
David Chinnere927af92007-06-05 16:24:36 +10001433 * Well, if only it were that simple. Because synchronous direct I/O
1434 * requires extent conversion to occur *before* we return to userspace,
1435 * we have to wait for extent conversion to complete. Look at the
1436 * iocb that has been passed to us to determine if this is AIO or
1437 * not. If it is synchronous, tell xfs_finish_ioend() to kick the
1438 * workqueue and wait for it to complete.
1439 *
Christoph Hellwigf0973862005-09-05 08:22:52 +10001440 * The core direct I/O code might be changed to always call the
1441 * completion handler in the future, in which case all this can
1442 * go away.
1443 */
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001444 ioend->io_offset = offset;
1445 ioend->io_size = size;
1446 if (ioend->io_type == IOMAP_READ) {
David Chinnere927af92007-06-05 16:24:36 +10001447 xfs_finish_ioend(ioend, 0);
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001448 } else if (private && size > 0) {
David Chinnere927af92007-06-05 16:24:36 +10001449 xfs_finish_ioend(ioend, is_sync_kiocb(iocb));
Christoph Hellwigf0973862005-09-05 08:22:52 +10001450 } else {
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001451 /*
1452 * A direct I/O write ioend starts it's life in unwritten
1453 * state in case they map an unwritten extent. This write
1454 * didn't map an unwritten extent so switch it's completion
1455 * handler.
1456 */
1457 INIT_WORK(&ioend->io_work, xfs_end_bio_written);
David Chinnere927af92007-06-05 16:24:36 +10001458 xfs_finish_ioend(ioend, 0);
Christoph Hellwigf0973862005-09-05 08:22:52 +10001459 }
1460
1461 /*
Nathan Scottc41564b2006-03-29 08:55:14 +10001462 * blockdev_direct_IO can return an error even after the I/O
Christoph Hellwigf0973862005-09-05 08:22:52 +10001463 * completion handler was called. Thus we need to protect
1464 * against double-freeing.
1465 */
1466 iocb->private = NULL;
1467}
1468
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469STATIC ssize_t
Nathan Scotte4c573b2006-03-14 13:54:26 +11001470xfs_vm_direct_IO(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 int rw,
1472 struct kiocb *iocb,
1473 const struct iovec *iov,
1474 loff_t offset,
1475 unsigned long nr_segs)
1476{
1477 struct file *file = iocb->ki_filp;
1478 struct inode *inode = file->f_mapping->host;
Nathan Scott67fcaa72006-06-09 17:00:52 +10001479 bhv_vnode_t *vp = vn_from_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 xfs_iomap_t iomap;
1481 int maps = 1;
1482 int error;
Christoph Hellwigf0973862005-09-05 08:22:52 +10001483 ssize_t ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484
Nathan Scott67fcaa72006-06-09 17:00:52 +10001485 error = bhv_vop_bmap(vp, offset, 0, BMAPI_DEVICE, &iomap, &maps);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 if (error)
1487 return -error;
1488
Lachlan McIlroy721259b2006-09-07 14:27:05 +10001489 if (rw == WRITE) {
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001490 iocb->private = xfs_alloc_ioend(inode, IOMAP_UNWRITTEN);
Lachlan McIlroy721259b2006-09-07 14:27:05 +10001491 ret = blockdev_direct_IO_own_locking(rw, iocb, inode,
1492 iomap.iomap_target->bt_bdev,
1493 iov, offset, nr_segs,
1494 xfs_get_blocks_direct,
1495 xfs_end_io_direct);
1496 } else {
Lachlan McIlroyba87ea62007-05-08 13:49:46 +10001497 iocb->private = xfs_alloc_ioend(inode, IOMAP_READ);
Lachlan McIlroy721259b2006-09-07 14:27:05 +10001498 ret = blockdev_direct_IO_no_locking(rw, iocb, inode,
1499 iomap.iomap_target->bt_bdev,
1500 iov, offset, nr_segs,
1501 xfs_get_blocks_direct,
1502 xfs_end_io_direct);
1503 }
Christoph Hellwigf0973862005-09-05 08:22:52 +10001504
Zach Brown8459d862006-12-10 02:21:05 -08001505 if (unlikely(ret != -EIOCBQUEUED && iocb->private))
Christoph Hellwigf0973862005-09-05 08:22:52 +10001506 xfs_destroy_ioend(iocb->private);
1507 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508}
1509
Nathan Scottf51623b2006-03-14 13:26:27 +11001510STATIC int
Nick Piggind79689c2007-10-16 01:25:06 -07001511xfs_vm_write_begin(
Nathan Scottf51623b2006-03-14 13:26:27 +11001512 struct file *file,
Nick Piggind79689c2007-10-16 01:25:06 -07001513 struct address_space *mapping,
1514 loff_t pos,
1515 unsigned len,
1516 unsigned flags,
1517 struct page **pagep,
1518 void **fsdata)
Nathan Scottf51623b2006-03-14 13:26:27 +11001519{
Nick Piggind79689c2007-10-16 01:25:06 -07001520 *pagep = NULL;
1521 return block_write_begin(file, mapping, pos, len, flags, pagep, fsdata,
1522 xfs_get_blocks);
Nathan Scottf51623b2006-03-14 13:26:27 +11001523}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524
1525STATIC sector_t
Nathan Scotte4c573b2006-03-14 13:54:26 +11001526xfs_vm_bmap(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 struct address_space *mapping,
1528 sector_t block)
1529{
1530 struct inode *inode = (struct inode *)mapping->host;
Nathan Scott67fcaa72006-06-09 17:00:52 +10001531 bhv_vnode_t *vp = vn_from_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532
Nathan Scotte4c573b2006-03-14 13:54:26 +11001533 vn_trace_entry(vp, __FUNCTION__, (inst_t *)__return_address);
Nathan Scott67fcaa72006-06-09 17:00:52 +10001534 bhv_vop_rwlock(vp, VRWLOCK_READ);
1535 bhv_vop_flush_pages(vp, (xfs_off_t)0, -1, 0, FI_REMAPF);
1536 bhv_vop_rwunlock(vp, VRWLOCK_READ);
Nathan Scottc2536662006-03-29 10:44:40 +10001537 return generic_block_bmap(mapping, block, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538}
1539
1540STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001541xfs_vm_readpage(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542 struct file *unused,
1543 struct page *page)
1544{
Nathan Scottc2536662006-03-29 10:44:40 +10001545 return mpage_readpage(page, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546}
1547
1548STATIC int
Nathan Scotte4c573b2006-03-14 13:54:26 +11001549xfs_vm_readpages(
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550 struct file *unused,
1551 struct address_space *mapping,
1552 struct list_head *pages,
1553 unsigned nr_pages)
1554{
Nathan Scottc2536662006-03-29 10:44:40 +10001555 return mpage_readpages(mapping, pages, nr_pages, xfs_get_blocks);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556}
1557
NeilBrown2ff28e22006-03-26 01:37:18 -08001558STATIC void
Nathan Scott238f4c52006-03-17 17:26:25 +11001559xfs_vm_invalidatepage(
Nathan Scottbcec2b72005-09-02 16:40:17 +10001560 struct page *page,
1561 unsigned long offset)
1562{
1563 xfs_page_trace(XFS_INVALIDPAGE_ENTER,
1564 page->mapping->host, page, offset);
NeilBrown2ff28e22006-03-26 01:37:18 -08001565 block_invalidatepage(page, offset);
Nathan Scottbcec2b72005-09-02 16:40:17 +10001566}
1567
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07001568const struct address_space_operations xfs_address_space_operations = {
Nathan Scotte4c573b2006-03-14 13:54:26 +11001569 .readpage = xfs_vm_readpage,
1570 .readpages = xfs_vm_readpages,
1571 .writepage = xfs_vm_writepage,
Nathan Scott7d4fb402006-06-09 15:27:16 +10001572 .writepages = xfs_vm_writepages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573 .sync_page = block_sync_page,
Nathan Scott238f4c52006-03-17 17:26:25 +11001574 .releasepage = xfs_vm_releasepage,
1575 .invalidatepage = xfs_vm_invalidatepage,
Nick Piggind79689c2007-10-16 01:25:06 -07001576 .write_begin = xfs_vm_write_begin,
1577 .write_end = generic_write_end,
Nathan Scotte4c573b2006-03-14 13:54:26 +11001578 .bmap = xfs_vm_bmap,
1579 .direct_IO = xfs_vm_direct_IO,
Christoph Lametere965f962006-02-01 03:05:41 -08001580 .migratepage = buffer_migrate_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581};