Branch data Line data Source code
1 : : // SPDX-License-Identifier: GPL-2.0-only
2 : : /*
3 : : * linux/fs/buffer.c
4 : : *
5 : : * Copyright (C) 1991, 1992, 2002 Linus Torvalds
6 : : */
7 : :
8 : : /*
9 : : * Start bdflush() with kernel_thread not syscall - Paul Gortmaker, 12/95
10 : : *
11 : : * Removed a lot of unnecessary code and simplified things now that
12 : : * the buffer cache isn't our primary cache - Andrew Tridgell 12/96
13 : : *
14 : : * Speed up hash, lru, and free list operations. Use gfp() for allocating
15 : : * hash table, use SLAB cache for buffer heads. SMP threading. -DaveM
16 : : *
17 : : * Added 32k buffer block sizes - these are required older ARM systems. - RMK
18 : : *
19 : : * async buffer flushing, 1999 Andrea Arcangeli <andrea@suse.de>
20 : : */
21 : :
22 : : #include <linux/kernel.h>
23 : : #include <linux/sched/signal.h>
24 : : #include <linux/syscalls.h>
25 : : #include <linux/fs.h>
26 : : #include <linux/iomap.h>
27 : : #include <linux/mm.h>
28 : : #include <linux/percpu.h>
29 : : #include <linux/slab.h>
30 : : #include <linux/capability.h>
31 : : #include <linux/blkdev.h>
32 : : #include <linux/file.h>
33 : : #include <linux/quotaops.h>
34 : : #include <linux/highmem.h>
35 : : #include <linux/export.h>
36 : : #include <linux/backing-dev.h>
37 : : #include <linux/writeback.h>
38 : : #include <linux/hash.h>
39 : : #include <linux/suspend.h>
40 : : #include <linux/buffer_head.h>
41 : : #include <linux/task_io_accounting_ops.h>
42 : : #include <linux/bio.h>
43 : : #include <linux/cpu.h>
44 : : #include <linux/bitops.h>
45 : : #include <linux/mpage.h>
46 : : #include <linux/bit_spinlock.h>
47 : : #include <linux/pagevec.h>
48 : : #include <linux/sched/mm.h>
49 : : #include <trace/events/block.h>
50 : :
51 : : static int fsync_buffers_list(spinlock_t *lock, struct list_head *list);
52 : : static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh,
53 : : enum rw_hint hint, struct writeback_control *wbc);
54 : :
55 : : #define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers)
56 : :
57 : 0 : inline void touch_buffer(struct buffer_head *bh)
58 : : {
59 : 3 : trace_block_touch_buffer(bh);
60 : 3 : mark_page_accessed(bh->b_page);
61 : 0 : }
62 : : EXPORT_SYMBOL(touch_buffer);
63 : :
64 : 3 : void __lock_buffer(struct buffer_head *bh)
65 : : {
66 : 3 : wait_on_bit_lock_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE);
67 : 3 : }
68 : : EXPORT_SYMBOL(__lock_buffer);
69 : :
70 : 3 : void unlock_buffer(struct buffer_head *bh)
71 : : {
72 : 3 : clear_bit_unlock(BH_Lock, &bh->b_state);
73 : 3 : smp_mb__after_atomic();
74 : 3 : wake_up_bit(&bh->b_state, BH_Lock);
75 : 3 : }
76 : : EXPORT_SYMBOL(unlock_buffer);
77 : :
78 : : /*
79 : : * Returns if the page has dirty or writeback buffers. If all the buffers
80 : : * are unlocked and clean then the PageDirty information is stale. If
81 : : * any of the pages are locked, it is assumed they are locked for IO.
82 : : */
83 : 0 : void buffer_check_dirty_writeback(struct page *page,
84 : : bool *dirty, bool *writeback)
85 : : {
86 : : struct buffer_head *head, *bh;
87 : 0 : *dirty = false;
88 : 0 : *writeback = false;
89 : :
90 : 0 : BUG_ON(!PageLocked(page));
91 : :
92 : 0 : if (!page_has_buffers(page))
93 : 0 : return;
94 : :
95 : 0 : if (PageWriteback(page))
96 : 0 : *writeback = true;
97 : :
98 : 0 : head = page_buffers(page);
99 : : bh = head;
100 : : do {
101 : 0 : if (buffer_locked(bh))
102 : 0 : *writeback = true;
103 : :
104 : 0 : if (buffer_dirty(bh))
105 : 0 : *dirty = true;
106 : :
107 : 0 : bh = bh->b_this_page;
108 : 0 : } while (bh != head);
109 : : }
110 : : EXPORT_SYMBOL(buffer_check_dirty_writeback);
111 : :
112 : : /*
113 : : * Block until a buffer comes unlocked. This doesn't stop it
114 : : * from becoming locked again - you have to lock it yourself
115 : : * if you want to preserve its state.
116 : : */
117 : 3 : void __wait_on_buffer(struct buffer_head * bh)
118 : : {
119 : 3 : wait_on_bit_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE);
120 : 3 : }
121 : : EXPORT_SYMBOL(__wait_on_buffer);
122 : :
123 : : static void
124 : 3 : __clear_page_buffers(struct page *page)
125 : : {
126 : : ClearPagePrivate(page);
127 : 3 : set_page_private(page, 0);
128 : 3 : put_page(page);
129 : 3 : }
130 : :
131 : 0 : static void buffer_io_error(struct buffer_head *bh, char *msg)
132 : : {
133 : 0 : if (!test_bit(BH_Quiet, &bh->b_state))
134 : 0 : printk_ratelimited(KERN_ERR
135 : : "Buffer I/O error on dev %pg, logical block %llu%s\n",
136 : : bh->b_bdev, (unsigned long long)bh->b_blocknr, msg);
137 : 0 : }
138 : :
139 : : /*
140 : : * End-of-IO handler helper function which does not touch the bh after
141 : : * unlocking it.
142 : : * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but
143 : : * a race there is benign: unlock_buffer() only use the bh's address for
144 : : * hashing after unlocking the buffer, so it doesn't actually touch the bh
145 : : * itself.
146 : : */
147 : 3 : static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate)
148 : : {
149 : 3 : if (uptodate) {
150 : : set_buffer_uptodate(bh);
151 : : } else {
152 : : /* This happens, due to failed read-ahead attempts. */
153 : : clear_buffer_uptodate(bh);
154 : : }
155 : 3 : unlock_buffer(bh);
156 : 3 : }
157 : :
158 : : /*
159 : : * Default synchronous end-of-IO handler.. Just mark it up-to-date and
160 : : * unlock the buffer. This is what ll_rw_block uses too.
161 : : */
162 : 3 : void end_buffer_read_sync(struct buffer_head *bh, int uptodate)
163 : : {
164 : 3 : __end_buffer_read_notouch(bh, uptodate);
165 : 3 : put_bh(bh);
166 : 3 : }
167 : : EXPORT_SYMBOL(end_buffer_read_sync);
168 : :
169 : 3 : void end_buffer_write_sync(struct buffer_head *bh, int uptodate)
170 : : {
171 : 3 : if (uptodate) {
172 : : set_buffer_uptodate(bh);
173 : : } else {
174 : 0 : buffer_io_error(bh, ", lost sync page write");
175 : 0 : mark_buffer_write_io_error(bh);
176 : : clear_buffer_uptodate(bh);
177 : : }
178 : 3 : unlock_buffer(bh);
179 : 3 : put_bh(bh);
180 : 3 : }
181 : : EXPORT_SYMBOL(end_buffer_write_sync);
182 : :
183 : : /*
184 : : * Various filesystems appear to want __find_get_block to be non-blocking.
185 : : * But it's the page lock which protects the buffers. To get around this,
186 : : * we get exclusion from try_to_free_buffers with the blockdev mapping's
187 : : * private_lock.
188 : : *
189 : : * Hack idea: for the blockdev mapping, private_lock contention
190 : : * may be quite high. This code could TryLock the page, and if that
191 : : * succeeds, there is no need to take private_lock.
192 : : */
193 : : static struct buffer_head *
194 : 3 : __find_get_block_slow(struct block_device *bdev, sector_t block)
195 : : {
196 : 3 : struct inode *bd_inode = bdev->bd_inode;
197 : 3 : struct address_space *bd_mapping = bd_inode->i_mapping;
198 : : struct buffer_head *ret = NULL;
199 : : pgoff_t index;
200 : : struct buffer_head *bh;
201 : : struct buffer_head *head;
202 : : struct page *page;
203 : : int all_mapped = 1;
204 : : static DEFINE_RATELIMIT_STATE(last_warned, HZ, 1);
205 : :
206 : 3 : index = block >> (PAGE_SHIFT - bd_inode->i_blkbits);
207 : : page = find_get_page_flags(bd_mapping, index, FGP_ACCESSED);
208 : 3 : if (!page)
209 : : goto out;
210 : :
211 : : spin_lock(&bd_mapping->private_lock);
212 : 3 : if (!page_has_buffers(page))
213 : : goto out_unlock;
214 : 3 : head = page_buffers(page);
215 : : bh = head;
216 : : do {
217 : 3 : if (!buffer_mapped(bh))
218 : : all_mapped = 0;
219 : 3 : else if (bh->b_blocknr == block) {
220 : 3 : ret = bh;
221 : : get_bh(bh);
222 : : goto out_unlock;
223 : : }
224 : 3 : bh = bh->b_this_page;
225 : 3 : } while (bh != head);
226 : :
227 : : /* we might be here because some of the buffers on this page are
228 : : * not mapped. This is due to various races between
229 : : * file io on the block device and getblk. It gets dealt with
230 : : * elsewhere, don't buffer_error if we had some unmapped buffers
231 : : */
232 : : ratelimit_set_flags(&last_warned, RATELIMIT_MSG_ON_RELEASE);
233 : 0 : if (all_mapped && __ratelimit(&last_warned)) {
234 : 0 : printk("__find_get_block_slow() failed. block=%llu, "
235 : : "b_blocknr=%llu, b_state=0x%08lx, b_size=%zu, "
236 : : "device %pg blocksize: %d\n",
237 : : (unsigned long long)block,
238 : : (unsigned long long)bh->b_blocknr,
239 : : bh->b_state, bh->b_size, bdev,
240 : 0 : 1 << bd_inode->i_blkbits);
241 : : }
242 : : out_unlock:
243 : : spin_unlock(&bd_mapping->private_lock);
244 : 3 : put_page(page);
245 : : out:
246 : 3 : return ret;
247 : : }
248 : :
249 : : /*
250 : : * I/O completion handler for block_read_full_page() - pages
251 : : * which come unlocked at the end of I/O.
252 : : */
253 : 3 : static void end_buffer_async_read(struct buffer_head *bh, int uptodate)
254 : : {
255 : : unsigned long flags;
256 : : struct buffer_head *first;
257 : : struct buffer_head *tmp;
258 : : struct page *page;
259 : : int page_uptodate = 1;
260 : :
261 : 3 : BUG_ON(!buffer_async_read(bh));
262 : :
263 : 3 : page = bh->b_page;
264 : 3 : if (uptodate) {
265 : : set_buffer_uptodate(bh);
266 : : } else {
267 : : clear_buffer_uptodate(bh);
268 : 0 : buffer_io_error(bh, ", async page read");
269 : : SetPageError(page);
270 : : }
271 : :
272 : : /*
273 : : * Be _very_ careful from here on. Bad things can happen if
274 : : * two buffer heads end IO at almost the same time and both
275 : : * decide that the page is now completely done.
276 : : */
277 : 3 : first = page_buffers(page);
278 : 3 : local_irq_save(flags);
279 : 3 : bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
280 : : clear_buffer_async_read(bh);
281 : 3 : unlock_buffer(bh);
282 : : tmp = bh;
283 : : do {
284 : 3 : if (!buffer_uptodate(tmp))
285 : : page_uptodate = 0;
286 : 3 : if (buffer_async_read(tmp)) {
287 : 2 : BUG_ON(!buffer_locked(tmp));
288 : : goto still_busy;
289 : : }
290 : 3 : tmp = tmp->b_this_page;
291 : 3 : } while (tmp != bh);
292 : : bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
293 : 3 : local_irq_restore(flags);
294 : :
295 : : /*
296 : : * If none of the buffers had errors and they are all
297 : : * uptodate then we can set the page uptodate.
298 : : */
299 : 3 : if (page_uptodate && !PageError(page))
300 : : SetPageUptodate(page);
301 : 3 : unlock_page(page);
302 : 3 : return;
303 : :
304 : : still_busy:
305 : : bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
306 : 2 : local_irq_restore(flags);
307 : : return;
308 : : }
309 : :
310 : : /*
311 : : * Completion handler for block_write_full_page() - pages which are unlocked
312 : : * during I/O, and which have PageWriteback cleared upon I/O completion.
313 : : */
314 : 3 : void end_buffer_async_write(struct buffer_head *bh, int uptodate)
315 : : {
316 : : unsigned long flags;
317 : : struct buffer_head *first;
318 : : struct buffer_head *tmp;
319 : : struct page *page;
320 : :
321 : 3 : BUG_ON(!buffer_async_write(bh));
322 : :
323 : 3 : page = bh->b_page;
324 : 3 : if (uptodate) {
325 : : set_buffer_uptodate(bh);
326 : : } else {
327 : 0 : buffer_io_error(bh, ", lost async page write");
328 : 0 : mark_buffer_write_io_error(bh);
329 : : clear_buffer_uptodate(bh);
330 : : SetPageError(page);
331 : : }
332 : :
333 : 3 : first = page_buffers(page);
334 : 3 : local_irq_save(flags);
335 : 3 : bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
336 : :
337 : : clear_buffer_async_write(bh);
338 : 3 : unlock_buffer(bh);
339 : 3 : tmp = bh->b_this_page;
340 : 3 : while (tmp != bh) {
341 : 3 : if (buffer_async_write(tmp)) {
342 : 0 : BUG_ON(!buffer_locked(tmp));
343 : : goto still_busy;
344 : : }
345 : 3 : tmp = tmp->b_this_page;
346 : : }
347 : : bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
348 : 3 : local_irq_restore(flags);
349 : 3 : end_page_writeback(page);
350 : 3 : return;
351 : :
352 : : still_busy:
353 : : bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
354 : 0 : local_irq_restore(flags);
355 : : return;
356 : : }
357 : : EXPORT_SYMBOL(end_buffer_async_write);
358 : :
359 : : /*
360 : : * If a page's buffers are under async readin (end_buffer_async_read
361 : : * completion) then there is a possibility that another thread of
362 : : * control could lock one of the buffers after it has completed
363 : : * but while some of the other buffers have not completed. This
364 : : * locked buffer would confuse end_buffer_async_read() into not unlocking
365 : : * the page. So the absence of BH_Async_Read tells end_buffer_async_read()
366 : : * that this buffer is not under async I/O.
367 : : *
368 : : * The page comes unlocked when it has no locked buffer_async buffers
369 : : * left.
370 : : *
371 : : * PageLocked prevents anyone starting new async I/O reads any of
372 : : * the buffers.
373 : : *
374 : : * PageWriteback is used to prevent simultaneous writeout of the same
375 : : * page.
376 : : *
377 : : * PageLocked prevents anyone from starting writeback of a page which is
378 : : * under read I/O (PageWriteback is only ever set against a locked page).
379 : : */
380 : 3 : static void mark_buffer_async_read(struct buffer_head *bh)
381 : : {
382 : 3 : bh->b_end_io = end_buffer_async_read;
383 : : set_buffer_async_read(bh);
384 : 3 : }
385 : :
386 : 3 : static void mark_buffer_async_write_endio(struct buffer_head *bh,
387 : : bh_end_io_t *handler)
388 : : {
389 : 3 : bh->b_end_io = handler;
390 : : set_buffer_async_write(bh);
391 : 3 : }
392 : :
393 : 0 : void mark_buffer_async_write(struct buffer_head *bh)
394 : : {
395 : 0 : mark_buffer_async_write_endio(bh, end_buffer_async_write);
396 : 0 : }
397 : : EXPORT_SYMBOL(mark_buffer_async_write);
398 : :
399 : :
400 : : /*
401 : : * fs/buffer.c contains helper functions for buffer-backed address space's
402 : : * fsync functions. A common requirement for buffer-based filesystems is
403 : : * that certain data from the backing blockdev needs to be written out for
404 : : * a successful fsync(). For example, ext2 indirect blocks need to be
405 : : * written back and waited upon before fsync() returns.
406 : : *
407 : : * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(),
408 : : * inode_has_buffers() and invalidate_inode_buffers() are provided for the
409 : : * management of a list of dependent buffers at ->i_mapping->private_list.
410 : : *
411 : : * Locking is a little subtle: try_to_free_buffers() will remove buffers
412 : : * from their controlling inode's queue when they are being freed. But
413 : : * try_to_free_buffers() will be operating against the *blockdev* mapping
414 : : * at the time, not against the S_ISREG file which depends on those buffers.
415 : : * So the locking for private_list is via the private_lock in the address_space
416 : : * which backs the buffers. Which is different from the address_space
417 : : * against which the buffers are listed. So for a particular address_space,
418 : : * mapping->private_lock does *not* protect mapping->private_list! In fact,
419 : : * mapping->private_list will always be protected by the backing blockdev's
420 : : * ->private_lock.
421 : : *
422 : : * Which introduces a requirement: all buffers on an address_space's
423 : : * ->private_list must be from the same address_space: the blockdev's.
424 : : *
425 : : * address_spaces which do not place buffers at ->private_list via these
426 : : * utility functions are free to use private_lock and private_list for
427 : : * whatever they want. The only requirement is that list_empty(private_list)
428 : : * be true at clear_inode() time.
429 : : *
430 : : * FIXME: clear_inode should not call invalidate_inode_buffers(). The
431 : : * filesystems should do that. invalidate_inode_buffers() should just go
432 : : * BUG_ON(!list_empty).
433 : : *
434 : : * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should
435 : : * take an address_space, not an inode. And it should be called
436 : : * mark_buffer_dirty_fsync() to clearly define why those buffers are being
437 : : * queued up.
438 : : *
439 : : * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the
440 : : * list if it is already on a list. Because if the buffer is on a list,
441 : : * it *must* already be on the right one. If not, the filesystem is being
442 : : * silly. This will save a ton of locking. But first we have to ensure
443 : : * that buffers are taken *off* the old inode's list when they are freed
444 : : * (presumably in truncate). That requires careful auditing of all
445 : : * filesystems (do it inside bforget()). It could also be done by bringing
446 : : * b_inode back.
447 : : */
448 : :
449 : : /*
450 : : * The buffer's backing address_space's private_lock must be held
451 : : */
452 : 0 : static void __remove_assoc_queue(struct buffer_head *bh)
453 : : {
454 : 0 : list_del_init(&bh->b_assoc_buffers);
455 : 0 : WARN_ON(!bh->b_assoc_map);
456 : 0 : bh->b_assoc_map = NULL;
457 : 0 : }
458 : :
459 : 3 : int inode_has_buffers(struct inode *inode)
460 : : {
461 : 3 : return !list_empty(&inode->i_data.private_list);
462 : : }
463 : :
464 : : /*
465 : : * osync is designed to support O_SYNC io. It waits synchronously for
466 : : * all already-submitted IO to complete, but does not queue any new
467 : : * writes to the disk.
468 : : *
469 : : * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as
470 : : * you dirty the buffers, and then use osync_inode_buffers to wait for
471 : : * completion. Any other dirty buffers which are not yet queued for
472 : : * write will not be flushed to disk by the osync.
473 : : */
474 : 0 : static int osync_buffers_list(spinlock_t *lock, struct list_head *list)
475 : : {
476 : : struct buffer_head *bh;
477 : : struct list_head *p;
478 : : int err = 0;
479 : :
480 : : spin_lock(lock);
481 : : repeat:
482 : 0 : list_for_each_prev(p, list) {
483 : 0 : bh = BH_ENTRY(p);
484 : 0 : if (buffer_locked(bh)) {
485 : : get_bh(bh);
486 : : spin_unlock(lock);
487 : 0 : wait_on_buffer(bh);
488 : 0 : if (!buffer_uptodate(bh))
489 : : err = -EIO;
490 : : brelse(bh);
491 : : spin_lock(lock);
492 : : goto repeat;
493 : : }
494 : : }
495 : : spin_unlock(lock);
496 : 0 : return err;
497 : : }
498 : :
499 : 0 : void emergency_thaw_bdev(struct super_block *sb)
500 : : {
501 : 0 : while (sb->s_bdev && !thaw_bdev(sb->s_bdev, sb))
502 : 0 : printk(KERN_WARNING "Emergency Thaw on %pg\n", sb->s_bdev);
503 : 0 : }
504 : :
505 : : /**
506 : : * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers
507 : : * @mapping: the mapping which wants those buffers written
508 : : *
509 : : * Starts I/O against the buffers at mapping->private_list, and waits upon
510 : : * that I/O.
511 : : *
512 : : * Basically, this is a convenience function for fsync().
513 : : * @mapping is a file or directory which needs those buffers to be written for
514 : : * a successful fsync().
515 : : */
516 : 0 : int sync_mapping_buffers(struct address_space *mapping)
517 : : {
518 : 0 : struct address_space *buffer_mapping = mapping->private_data;
519 : :
520 : 0 : if (buffer_mapping == NULL || list_empty(&mapping->private_list))
521 : : return 0;
522 : :
523 : 0 : return fsync_buffers_list(&buffer_mapping->private_lock,
524 : : &mapping->private_list);
525 : : }
526 : : EXPORT_SYMBOL(sync_mapping_buffers);
527 : :
528 : : /*
529 : : * Called when we've recently written block `bblock', and it is known that
530 : : * `bblock' was for a buffer_boundary() buffer. This means that the block at
531 : : * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's
532 : : * dirty, schedule it for IO. So that indirects merge nicely with their data.
533 : : */
534 : 0 : void write_boundary_block(struct block_device *bdev,
535 : : sector_t bblock, unsigned blocksize)
536 : : {
537 : 0 : struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize);
538 : 0 : if (bh) {
539 : 0 : if (buffer_dirty(bh))
540 : 0 : ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
541 : 0 : put_bh(bh);
542 : : }
543 : 0 : }
544 : :
545 : 0 : void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode)
546 : : {
547 : 0 : struct address_space *mapping = inode->i_mapping;
548 : 0 : struct address_space *buffer_mapping = bh->b_page->mapping;
549 : :
550 : 0 : mark_buffer_dirty(bh);
551 : 0 : if (!mapping->private_data) {
552 : 0 : mapping->private_data = buffer_mapping;
553 : : } else {
554 : 0 : BUG_ON(mapping->private_data != buffer_mapping);
555 : : }
556 : 0 : if (!bh->b_assoc_map) {
557 : : spin_lock(&buffer_mapping->private_lock);
558 : 0 : list_move_tail(&bh->b_assoc_buffers,
559 : : &mapping->private_list);
560 : 0 : bh->b_assoc_map = mapping;
561 : : spin_unlock(&buffer_mapping->private_lock);
562 : : }
563 : 0 : }
564 : : EXPORT_SYMBOL(mark_buffer_dirty_inode);
565 : :
566 : : /*
567 : : * Mark the page dirty, and set it dirty in the page cache, and mark the inode
568 : : * dirty.
569 : : *
570 : : * If warn is true, then emit a warning if the page is not uptodate and has
571 : : * not been truncated.
572 : : *
573 : : * The caller must hold lock_page_memcg().
574 : : */
575 : 3 : void __set_page_dirty(struct page *page, struct address_space *mapping,
576 : : int warn)
577 : : {
578 : : unsigned long flags;
579 : :
580 : 3 : xa_lock_irqsave(&mapping->i_pages, flags);
581 : 3 : if (page->mapping) { /* Race with truncate? */
582 : 3 : WARN_ON_ONCE(warn && !PageUptodate(page));
583 : 3 : account_page_dirtied(page, mapping);
584 : 3 : __xa_set_mark(&mapping->i_pages, page_index(page),
585 : : PAGECACHE_TAG_DIRTY);
586 : : }
587 : : xa_unlock_irqrestore(&mapping->i_pages, flags);
588 : 3 : }
589 : : EXPORT_SYMBOL_GPL(__set_page_dirty);
590 : :
591 : : /*
592 : : * Add a page to the dirty page list.
593 : : *
594 : : * It is a sad fact of life that this function is called from several places
595 : : * deeply under spinlocking. It may not sleep.
596 : : *
597 : : * If the page has buffers, the uptodate buffers are set dirty, to preserve
598 : : * dirty-state coherency between the page and the buffers. It the page does
599 : : * not have buffers then when they are later attached they will all be set
600 : : * dirty.
601 : : *
602 : : * The buffers are dirtied before the page is dirtied. There's a small race
603 : : * window in which a writepage caller may see the page cleanness but not the
604 : : * buffer dirtiness. That's fine. If this code were to set the page dirty
605 : : * before the buffers, a concurrent writepage caller could clear the page dirty
606 : : * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean
607 : : * page on the dirty page list.
608 : : *
609 : : * We use private_lock to lock against try_to_free_buffers while using the
610 : : * page's buffer list. Also use this to protect against clean buffers being
611 : : * added to the page after it was set dirty.
612 : : *
613 : : * FIXME: may need to call ->reservepage here as well. That's rather up to the
614 : : * address_space though.
615 : : */
616 : 3 : int __set_page_dirty_buffers(struct page *page)
617 : : {
618 : : int newly_dirty;
619 : 3 : struct address_space *mapping = page_mapping(page);
620 : :
621 : 3 : if (unlikely(!mapping))
622 : 0 : return !TestSetPageDirty(page);
623 : :
624 : : spin_lock(&mapping->private_lock);
625 : 3 : if (page_has_buffers(page)) {
626 : 3 : struct buffer_head *head = page_buffers(page);
627 : : struct buffer_head *bh = head;
628 : :
629 : : do {
630 : : set_buffer_dirty(bh);
631 : 3 : bh = bh->b_this_page;
632 : 3 : } while (bh != head);
633 : : }
634 : : /*
635 : : * Lock out page->mem_cgroup migration to keep PageDirty
636 : : * synchronized with per-memcg dirty page counters.
637 : : */
638 : 3 : lock_page_memcg(page);
639 : 3 : newly_dirty = !TestSetPageDirty(page);
640 : : spin_unlock(&mapping->private_lock);
641 : :
642 : 3 : if (newly_dirty)
643 : 0 : __set_page_dirty(page, mapping, 1);
644 : :
645 : 3 : unlock_page_memcg(page);
646 : :
647 : 3 : if (newly_dirty)
648 : 0 : __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
649 : :
650 : 3 : return newly_dirty;
651 : : }
652 : : EXPORT_SYMBOL(__set_page_dirty_buffers);
653 : :
654 : : /*
655 : : * Write out and wait upon a list of buffers.
656 : : *
657 : : * We have conflicting pressures: we want to make sure that all
658 : : * initially dirty buffers get waited on, but that any subsequently
659 : : * dirtied buffers don't. After all, we don't want fsync to last
660 : : * forever if somebody is actively writing to the file.
661 : : *
662 : : * Do this in two main stages: first we copy dirty buffers to a
663 : : * temporary inode list, queueing the writes as we go. Then we clean
664 : : * up, waiting for those writes to complete.
665 : : *
666 : : * During this second stage, any subsequent updates to the file may end
667 : : * up refiling the buffer on the original inode's dirty list again, so
668 : : * there is a chance we will end up with a buffer queued for write but
669 : : * not yet completed on that list. So, as a final cleanup we go through
670 : : * the osync code to catch these locked, dirty buffers without requeuing
671 : : * any newly dirty buffers for write.
672 : : */
673 : 0 : static int fsync_buffers_list(spinlock_t *lock, struct list_head *list)
674 : : {
675 : : struct buffer_head *bh;
676 : : struct list_head tmp;
677 : : struct address_space *mapping;
678 : : int err = 0, err2;
679 : : struct blk_plug plug;
680 : :
681 : : INIT_LIST_HEAD(&tmp);
682 : 0 : blk_start_plug(&plug);
683 : :
684 : : spin_lock(lock);
685 : 0 : while (!list_empty(list)) {
686 : 0 : bh = BH_ENTRY(list->next);
687 : 0 : mapping = bh->b_assoc_map;
688 : 0 : __remove_assoc_queue(bh);
689 : : /* Avoid race with mark_buffer_dirty_inode() which does
690 : : * a lockless check and we rely on seeing the dirty bit */
691 : 0 : smp_mb();
692 : 0 : if (buffer_dirty(bh) || buffer_locked(bh)) {
693 : 0 : list_add(&bh->b_assoc_buffers, &tmp);
694 : 0 : bh->b_assoc_map = mapping;
695 : 0 : if (buffer_dirty(bh)) {
696 : : get_bh(bh);
697 : : spin_unlock(lock);
698 : : /*
699 : : * Ensure any pending I/O completes so that
700 : : * write_dirty_buffer() actually writes the
701 : : * current contents - it is a noop if I/O is
702 : : * still in flight on potentially older
703 : : * contents.
704 : : */
705 : 0 : write_dirty_buffer(bh, REQ_SYNC);
706 : :
707 : : /*
708 : : * Kick off IO for the previous mapping. Note
709 : : * that we will not run the very last mapping,
710 : : * wait_on_buffer() will do that for us
711 : : * through sync_buffer().
712 : : */
713 : : brelse(bh);
714 : : spin_lock(lock);
715 : : }
716 : : }
717 : : }
718 : :
719 : : spin_unlock(lock);
720 : 0 : blk_finish_plug(&plug);
721 : : spin_lock(lock);
722 : :
723 : 0 : while (!list_empty(&tmp)) {
724 : 0 : bh = BH_ENTRY(tmp.prev);
725 : : get_bh(bh);
726 : 0 : mapping = bh->b_assoc_map;
727 : 0 : __remove_assoc_queue(bh);
728 : : /* Avoid race with mark_buffer_dirty_inode() which does
729 : : * a lockless check and we rely on seeing the dirty bit */
730 : 0 : smp_mb();
731 : 0 : if (buffer_dirty(bh)) {
732 : 0 : list_add(&bh->b_assoc_buffers,
733 : : &mapping->private_list);
734 : 0 : bh->b_assoc_map = mapping;
735 : : }
736 : : spin_unlock(lock);
737 : 0 : wait_on_buffer(bh);
738 : 0 : if (!buffer_uptodate(bh))
739 : : err = -EIO;
740 : : brelse(bh);
741 : : spin_lock(lock);
742 : : }
743 : :
744 : : spin_unlock(lock);
745 : 0 : err2 = osync_buffers_list(lock, list);
746 : 0 : if (err)
747 : 0 : return err;
748 : : else
749 : : return err2;
750 : : }
751 : :
752 : : /*
753 : : * Invalidate any and all dirty buffers on a given inode. We are
754 : : * probably unmounting the fs, but that doesn't mean we have already
755 : : * done a sync(). Just drop the buffers from the inode list.
756 : : *
757 : : * NOTE: we take the inode's blockdev's mapping's private_lock. Which
758 : : * assumes that all the buffers are against the blockdev. Not true
759 : : * for reiserfs.
760 : : */
761 : 3 : void invalidate_inode_buffers(struct inode *inode)
762 : : {
763 : 3 : if (inode_has_buffers(inode)) {
764 : : struct address_space *mapping = &inode->i_data;
765 : : struct list_head *list = &mapping->private_list;
766 : 0 : struct address_space *buffer_mapping = mapping->private_data;
767 : :
768 : : spin_lock(&buffer_mapping->private_lock);
769 : 0 : while (!list_empty(list))
770 : 0 : __remove_assoc_queue(BH_ENTRY(list->next));
771 : : spin_unlock(&buffer_mapping->private_lock);
772 : : }
773 : 3 : }
774 : : EXPORT_SYMBOL(invalidate_inode_buffers);
775 : :
776 : : /*
777 : : * Remove any clean buffers from the inode's buffer list. This is called
778 : : * when we're trying to free the inode itself. Those buffers can pin it.
779 : : *
780 : : * Returns true if all buffers were removed.
781 : : */
782 : 0 : int remove_inode_buffers(struct inode *inode)
783 : : {
784 : : int ret = 1;
785 : :
786 : 0 : if (inode_has_buffers(inode)) {
787 : : struct address_space *mapping = &inode->i_data;
788 : : struct list_head *list = &mapping->private_list;
789 : 0 : struct address_space *buffer_mapping = mapping->private_data;
790 : :
791 : : spin_lock(&buffer_mapping->private_lock);
792 : 0 : while (!list_empty(list)) {
793 : 0 : struct buffer_head *bh = BH_ENTRY(list->next);
794 : 0 : if (buffer_dirty(bh)) {
795 : : ret = 0;
796 : : break;
797 : : }
798 : 0 : __remove_assoc_queue(bh);
799 : : }
800 : : spin_unlock(&buffer_mapping->private_lock);
801 : : }
802 : 0 : return ret;
803 : : }
804 : :
805 : : /*
806 : : * Create the appropriate buffers when given a page for data area and
807 : : * the size of each buffer.. Use the bh->b_this_page linked list to
808 : : * follow the buffers created. Return NULL if unable to create more
809 : : * buffers.
810 : : *
811 : : * The retry flag is used to differentiate async IO (paging, swapping)
812 : : * which may not fail from ordinary buffer allocations.
813 : : */
814 : 3 : struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size,
815 : : bool retry)
816 : : {
817 : : struct buffer_head *bh, *head;
818 : : gfp_t gfp = GFP_NOFS | __GFP_ACCOUNT;
819 : : long offset;
820 : : struct mem_cgroup *memcg;
821 : :
822 : 3 : if (retry)
823 : : gfp |= __GFP_NOFAIL;
824 : :
825 : 3 : memcg = get_mem_cgroup_from_page(page);
826 : 3 : memalloc_use_memcg(memcg);
827 : :
828 : : head = NULL;
829 : : offset = PAGE_SIZE;
830 : 3 : while ((offset -= size) >= 0) {
831 : 3 : bh = alloc_buffer_head(gfp);
832 : 3 : if (!bh)
833 : : goto no_grow;
834 : :
835 : 3 : bh->b_this_page = head;
836 : 3 : bh->b_blocknr = -1;
837 : : head = bh;
838 : :
839 : 3 : bh->b_size = size;
840 : :
841 : : /* Link the buffer to its page */
842 : 3 : set_bh_page(bh, page, offset);
843 : : }
844 : : out:
845 : : memalloc_unuse_memcg();
846 : 3 : mem_cgroup_put(memcg);
847 : 3 : return head;
848 : : /*
849 : : * In case anything failed, we just free everything we got.
850 : : */
851 : : no_grow:
852 : 0 : if (head) {
853 : : do {
854 : : bh = head;
855 : 0 : head = head->b_this_page;
856 : 0 : free_buffer_head(bh);
857 : 0 : } while (head);
858 : : }
859 : :
860 : : goto out;
861 : : }
862 : : EXPORT_SYMBOL_GPL(alloc_page_buffers);
863 : :
864 : : static inline void
865 : 3 : link_dev_buffers(struct page *page, struct buffer_head *head)
866 : : {
867 : : struct buffer_head *bh, *tail;
868 : :
869 : : bh = head;
870 : : do {
871 : : tail = bh;
872 : 3 : bh = bh->b_this_page;
873 : 3 : } while (bh);
874 : 3 : tail->b_this_page = head;
875 : : attach_page_buffers(page, head);
876 : 3 : }
877 : :
878 : 3 : static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size)
879 : : {
880 : : sector_t retval = ~((sector_t)0);
881 : 3 : loff_t sz = i_size_read(bdev->bd_inode);
882 : :
883 : 3 : if (sz) {
884 : : unsigned int sizebits = blksize_bits(size);
885 : 3 : retval = (sz >> sizebits);
886 : : }
887 : 3 : return retval;
888 : : }
889 : :
890 : : /*
891 : : * Initialise the state of a blockdev page's buffers.
892 : : */
893 : : static sector_t
894 : 3 : init_page_buffers(struct page *page, struct block_device *bdev,
895 : : sector_t block, int size)
896 : : {
897 : 3 : struct buffer_head *head = page_buffers(page);
898 : : struct buffer_head *bh = head;
899 : : int uptodate = PageUptodate(page);
900 : 3 : sector_t end_block = blkdev_max_block(I_BDEV(bdev->bd_inode), size);
901 : :
902 : : do {
903 : 3 : if (!buffer_mapped(bh)) {
904 : 3 : bh->b_end_io = NULL;
905 : 3 : bh->b_private = NULL;
906 : 3 : bh->b_bdev = bdev;
907 : 3 : bh->b_blocknr = block;
908 : 3 : if (uptodate)
909 : : set_buffer_uptodate(bh);
910 : 3 : if (block < end_block)
911 : : set_buffer_mapped(bh);
912 : : }
913 : 3 : block++;
914 : 3 : bh = bh->b_this_page;
915 : 3 : } while (bh != head);
916 : :
917 : : /*
918 : : * Caller needs to validate requested block against end of device.
919 : : */
920 : 3 : return end_block;
921 : : }
922 : :
923 : : /*
924 : : * Create the page-cache page that contains the requested block.
925 : : *
926 : : * This is used purely for blockdev mappings.
927 : : */
928 : : static int
929 : 3 : grow_dev_page(struct block_device *bdev, sector_t block,
930 : : pgoff_t index, int size, int sizebits, gfp_t gfp)
931 : : {
932 : 3 : struct inode *inode = bdev->bd_inode;
933 : : struct page *page;
934 : : struct buffer_head *bh;
935 : : sector_t end_block;
936 : : int ret = 0; /* Will call free_more_memory() */
937 : : gfp_t gfp_mask;
938 : :
939 : 3 : gfp_mask = mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS) | gfp;
940 : :
941 : : /*
942 : : * XXX: __getblk_slow() can not really deal with failure and
943 : : * will endlessly loop on improvised global reclaim. Prefer
944 : : * looping in the allocator rather than here, at least that
945 : : * code knows what it's doing.
946 : : */
947 : 3 : gfp_mask |= __GFP_NOFAIL;
948 : :
949 : : page = find_or_create_page(inode->i_mapping, index, gfp_mask);
950 : :
951 : 3 : BUG_ON(!PageLocked(page));
952 : :
953 : 3 : if (page_has_buffers(page)) {
954 : 3 : bh = page_buffers(page);
955 : 3 : if (bh->b_size == size) {
956 : 3 : end_block = init_page_buffers(page, bdev,
957 : 3 : (sector_t)index << sizebits,
958 : : size);
959 : 3 : goto done;
960 : : }
961 : 0 : if (!try_to_free_buffers(page))
962 : : goto failed;
963 : : }
964 : :
965 : : /*
966 : : * Allocate some buffers for this page
967 : : */
968 : 3 : bh = alloc_page_buffers(page, size, true);
969 : :
970 : : /*
971 : : * Link the page to the buffers and initialise them. Take the
972 : : * lock to be atomic wrt __find_get_block(), which does not
973 : : * run under the page lock.
974 : : */
975 : 3 : spin_lock(&inode->i_mapping->private_lock);
976 : 3 : link_dev_buffers(page, bh);
977 : 3 : end_block = init_page_buffers(page, bdev, (sector_t)index << sizebits,
978 : : size);
979 : 3 : spin_unlock(&inode->i_mapping->private_lock);
980 : : done:
981 : 3 : ret = (block < end_block) ? 1 : -ENXIO;
982 : : failed:
983 : 3 : unlock_page(page);
984 : 3 : put_page(page);
985 : 3 : return ret;
986 : : }
987 : :
988 : : /*
989 : : * Create buffers for the specified block device block's page. If
990 : : * that page was dirty, the buffers are set dirty also.
991 : : */
992 : : static int
993 : 3 : grow_buffers(struct block_device *bdev, sector_t block, int size, gfp_t gfp)
994 : : {
995 : : pgoff_t index;
996 : : int sizebits;
997 : :
998 : : sizebits = -1;
999 : : do {
1000 : 3 : sizebits++;
1001 : 3 : } while ((size << sizebits) < PAGE_SIZE);
1002 : :
1003 : 3 : index = block >> sizebits;
1004 : :
1005 : : /*
1006 : : * Check for a block which wants to lie outside our maximum possible
1007 : : * pagecache index. (this comparison is done using sector_t types).
1008 : : */
1009 : 3 : if (unlikely(index != block >> sizebits)) {
1010 : 0 : printk(KERN_ERR "%s: requested out-of-range block %llu for "
1011 : : "device %pg\n",
1012 : : __func__, (unsigned long long)block,
1013 : : bdev);
1014 : 0 : return -EIO;
1015 : : }
1016 : :
1017 : : /* Create a page with the proper size buffers.. */
1018 : 3 : return grow_dev_page(bdev, block, index, size, sizebits, gfp);
1019 : : }
1020 : :
1021 : : static struct buffer_head *
1022 : 3 : __getblk_slow(struct block_device *bdev, sector_t block,
1023 : : unsigned size, gfp_t gfp)
1024 : : {
1025 : : /* Size must be multiple of hard sectorsize */
1026 : 3 : if (unlikely(size & (bdev_logical_block_size(bdev)-1) ||
1027 : : (size < 512 || size > PAGE_SIZE))) {
1028 : 3 : printk(KERN_ERR "getblk(): invalid block size %d requested\n",
1029 : : size);
1030 : 0 : printk(KERN_ERR "logical block size: %d\n",
1031 : : bdev_logical_block_size(bdev));
1032 : :
1033 : 0 : dump_stack();
1034 : 0 : return NULL;
1035 : : }
1036 : :
1037 : : for (;;) {
1038 : : struct buffer_head *bh;
1039 : : int ret;
1040 : :
1041 : 3 : bh = __find_get_block(bdev, block, size);
1042 : 3 : if (bh)
1043 : 3 : return bh;
1044 : :
1045 : 3 : ret = grow_buffers(bdev, block, size, gfp);
1046 : 3 : if (ret < 0)
1047 : : return NULL;
1048 : : }
1049 : : }
1050 : :
1051 : : /*
1052 : : * The relationship between dirty buffers and dirty pages:
1053 : : *
1054 : : * Whenever a page has any dirty buffers, the page's dirty bit is set, and
1055 : : * the page is tagged dirty in the page cache.
1056 : : *
1057 : : * At all times, the dirtiness of the buffers represents the dirtiness of
1058 : : * subsections of the page. If the page has buffers, the page dirty bit is
1059 : : * merely a hint about the true dirty state.
1060 : : *
1061 : : * When a page is set dirty in its entirety, all its buffers are marked dirty
1062 : : * (if the page has buffers).
1063 : : *
1064 : : * When a buffer is marked dirty, its page is dirtied, but the page's other
1065 : : * buffers are not.
1066 : : *
1067 : : * Also. When blockdev buffers are explicitly read with bread(), they
1068 : : * individually become uptodate. But their backing page remains not
1069 : : * uptodate - even if all of its buffers are uptodate. A subsequent
1070 : : * block_read_full_page() against that page will discover all the uptodate
1071 : : * buffers, will set the page uptodate and will perform no I/O.
1072 : : */
1073 : :
1074 : : /**
1075 : : * mark_buffer_dirty - mark a buffer_head as needing writeout
1076 : : * @bh: the buffer_head to mark dirty
1077 : : *
1078 : : * mark_buffer_dirty() will set the dirty bit against the buffer, then set
1079 : : * its backing page dirty, then tag the page as dirty in the page cache
1080 : : * and then attach the address_space's inode to its superblock's dirty
1081 : : * inode list.
1082 : : *
1083 : : * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock,
1084 : : * i_pages lock and mapping->host->i_lock.
1085 : : */
1086 : 3 : void mark_buffer_dirty(struct buffer_head *bh)
1087 : : {
1088 : 3 : WARN_ON_ONCE(!buffer_uptodate(bh));
1089 : :
1090 : 3 : trace_block_dirty_buffer(bh);
1091 : :
1092 : : /*
1093 : : * Very *carefully* optimize the it-is-already-dirty case.
1094 : : *
1095 : : * Don't let the final "is it dirty" escape to before we
1096 : : * perhaps modified the buffer.
1097 : : */
1098 : 3 : if (buffer_dirty(bh)) {
1099 : 3 : smp_mb();
1100 : 3 : if (buffer_dirty(bh))
1101 : 3 : return;
1102 : : }
1103 : :
1104 : 3 : if (!test_set_buffer_dirty(bh)) {
1105 : 3 : struct page *page = bh->b_page;
1106 : : struct address_space *mapping = NULL;
1107 : :
1108 : 3 : lock_page_memcg(page);
1109 : 3 : if (!TestSetPageDirty(page)) {
1110 : 3 : mapping = page_mapping(page);
1111 : 3 : if (mapping)
1112 : 3 : __set_page_dirty(page, mapping, 0);
1113 : : }
1114 : 3 : unlock_page_memcg(page);
1115 : 3 : if (mapping)
1116 : 3 : __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1117 : : }
1118 : : }
1119 : : EXPORT_SYMBOL(mark_buffer_dirty);
1120 : :
1121 : 0 : void mark_buffer_write_io_error(struct buffer_head *bh)
1122 : : {
1123 : : set_buffer_write_io_error(bh);
1124 : : /* FIXME: do we need to set this in both places? */
1125 : 0 : if (bh->b_page && bh->b_page->mapping)
1126 : 0 : mapping_set_error(bh->b_page->mapping, -EIO);
1127 : 0 : if (bh->b_assoc_map)
1128 : 0 : mapping_set_error(bh->b_assoc_map, -EIO);
1129 : 0 : }
1130 : : EXPORT_SYMBOL(mark_buffer_write_io_error);
1131 : :
1132 : : /*
1133 : : * Decrement a buffer_head's reference count. If all buffers against a page
1134 : : * have zero reference count, are clean and unlocked, and if the page is clean
1135 : : * and unlocked then try_to_free_buffers() may strip the buffers from the page
1136 : : * in preparation for freeing it (sometimes, rarely, buffers are removed from
1137 : : * a page but it ends up not being freed, and buffers may later be reattached).
1138 : : */
1139 : 3 : void __brelse(struct buffer_head * buf)
1140 : : {
1141 : 3 : if (atomic_read(&buf->b_count)) {
1142 : 3 : put_bh(buf);
1143 : 3 : return;
1144 : : }
1145 : 0 : WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n");
1146 : : }
1147 : : EXPORT_SYMBOL(__brelse);
1148 : :
1149 : : /*
1150 : : * bforget() is like brelse(), except it discards any
1151 : : * potentially dirty data.
1152 : : */
1153 : 3 : void __bforget(struct buffer_head *bh)
1154 : : {
1155 : : clear_buffer_dirty(bh);
1156 : 3 : if (bh->b_assoc_map) {
1157 : 0 : struct address_space *buffer_mapping = bh->b_page->mapping;
1158 : :
1159 : : spin_lock(&buffer_mapping->private_lock);
1160 : 0 : list_del_init(&bh->b_assoc_buffers);
1161 : 0 : bh->b_assoc_map = NULL;
1162 : : spin_unlock(&buffer_mapping->private_lock);
1163 : : }
1164 : 3 : __brelse(bh);
1165 : 3 : }
1166 : : EXPORT_SYMBOL(__bforget);
1167 : :
1168 : 3 : static struct buffer_head *__bread_slow(struct buffer_head *bh)
1169 : : {
1170 : 3 : lock_buffer(bh);
1171 : 3 : if (buffer_uptodate(bh)) {
1172 : 3 : unlock_buffer(bh);
1173 : 3 : return bh;
1174 : : } else {
1175 : : get_bh(bh);
1176 : 3 : bh->b_end_io = end_buffer_read_sync;
1177 : : submit_bh(REQ_OP_READ, 0, bh);
1178 : 3 : wait_on_buffer(bh);
1179 : 3 : if (buffer_uptodate(bh))
1180 : : return bh;
1181 : : }
1182 : : brelse(bh);
1183 : : return NULL;
1184 : : }
1185 : :
1186 : : /*
1187 : : * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block().
1188 : : * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their
1189 : : * refcount elevated by one when they're in an LRU. A buffer can only appear
1190 : : * once in a particular CPU's LRU. A single buffer can be present in multiple
1191 : : * CPU's LRUs at the same time.
1192 : : *
1193 : : * This is a transparent caching front-end to sb_bread(), sb_getblk() and
1194 : : * sb_find_get_block().
1195 : : *
1196 : : * The LRUs themselves only need locking against invalidate_bh_lrus. We use
1197 : : * a local interrupt disable for that.
1198 : : */
1199 : :
1200 : : #define BH_LRU_SIZE 16
1201 : :
1202 : : struct bh_lru {
1203 : : struct buffer_head *bhs[BH_LRU_SIZE];
1204 : : };
1205 : :
1206 : : static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }};
1207 : :
1208 : : #ifdef CONFIG_SMP
1209 : : #define bh_lru_lock() local_irq_disable()
1210 : : #define bh_lru_unlock() local_irq_enable()
1211 : : #else
1212 : : #define bh_lru_lock() preempt_disable()
1213 : : #define bh_lru_unlock() preempt_enable()
1214 : : #endif
1215 : :
1216 : 3 : static inline void check_irqs_on(void)
1217 : : {
1218 : : #ifdef irqs_disabled
1219 : 3 : BUG_ON(irqs_disabled());
1220 : : #endif
1221 : 3 : }
1222 : :
1223 : : /*
1224 : : * Install a buffer_head into this cpu's LRU. If not already in the LRU, it is
1225 : : * inserted at the front, and the buffer_head at the back if any is evicted.
1226 : : * Or, if already in the LRU it is moved to the front.
1227 : : */
1228 : 3 : static void bh_lru_install(struct buffer_head *bh)
1229 : : {
1230 : : struct buffer_head *evictee = bh;
1231 : : struct bh_lru *b;
1232 : : int i;
1233 : :
1234 : 3 : check_irqs_on();
1235 : 3 : bh_lru_lock();
1236 : :
1237 : 3 : b = this_cpu_ptr(&bh_lrus);
1238 : 3 : for (i = 0; i < BH_LRU_SIZE; i++) {
1239 : 3 : swap(evictee, b->bhs[i]);
1240 : 3 : if (evictee == bh) {
1241 : 1 : bh_lru_unlock();
1242 : 3 : return;
1243 : : }
1244 : : }
1245 : :
1246 : : get_bh(bh);
1247 : 3 : bh_lru_unlock();
1248 : : brelse(evictee);
1249 : : }
1250 : :
1251 : : /*
1252 : : * Look up the bh in this cpu's LRU. If it's there, move it to the head.
1253 : : */
1254 : : static struct buffer_head *
1255 : 3 : lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size)
1256 : : {
1257 : : struct buffer_head *ret = NULL;
1258 : : unsigned int i;
1259 : :
1260 : 3 : check_irqs_on();
1261 : 3 : bh_lru_lock();
1262 : 3 : for (i = 0; i < BH_LRU_SIZE; i++) {
1263 : 3 : struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]);
1264 : :
1265 : 3 : if (bh && bh->b_blocknr == block && bh->b_bdev == bdev &&
1266 : 3 : bh->b_size == size) {
1267 : 3 : if (i) {
1268 : 3 : while (i) {
1269 : 3 : __this_cpu_write(bh_lrus.bhs[i],
1270 : : __this_cpu_read(bh_lrus.bhs[i - 1]));
1271 : : i--;
1272 : : }
1273 : 3 : __this_cpu_write(bh_lrus.bhs[0], bh);
1274 : : }
1275 : : get_bh(bh);
1276 : 3 : ret = bh;
1277 : 3 : break;
1278 : : }
1279 : : }
1280 : 3 : bh_lru_unlock();
1281 : 3 : return ret;
1282 : : }
1283 : :
1284 : : /*
1285 : : * Perform a pagecache lookup for the matching buffer. If it's there, refresh
1286 : : * it in the LRU and mark it as accessed. If it is not present then return
1287 : : * NULL
1288 : : */
1289 : : struct buffer_head *
1290 : 3 : __find_get_block(struct block_device *bdev, sector_t block, unsigned size)
1291 : : {
1292 : 3 : struct buffer_head *bh = lookup_bh_lru(bdev, block, size);
1293 : :
1294 : 3 : if (bh == NULL) {
1295 : : /* __find_get_block_slow will mark the page accessed */
1296 : 3 : bh = __find_get_block_slow(bdev, block);
1297 : 3 : if (bh)
1298 : 3 : bh_lru_install(bh);
1299 : : } else
1300 : : touch_buffer(bh);
1301 : :
1302 : 3 : return bh;
1303 : : }
1304 : : EXPORT_SYMBOL(__find_get_block);
1305 : :
1306 : : /*
1307 : : * __getblk_gfp() will locate (and, if necessary, create) the buffer_head
1308 : : * which corresponds to the passed block_device, block and size. The
1309 : : * returned buffer has its reference count incremented.
1310 : : *
1311 : : * __getblk_gfp() will lock up the machine if grow_dev_page's
1312 : : * try_to_free_buffers() attempt is failing. FIXME, perhaps?
1313 : : */
1314 : : struct buffer_head *
1315 : 3 : __getblk_gfp(struct block_device *bdev, sector_t block,
1316 : : unsigned size, gfp_t gfp)
1317 : : {
1318 : 3 : struct buffer_head *bh = __find_get_block(bdev, block, size);
1319 : :
1320 : 3 : might_sleep();
1321 : 3 : if (bh == NULL)
1322 : 3 : bh = __getblk_slow(bdev, block, size, gfp);
1323 : 3 : return bh;
1324 : : }
1325 : : EXPORT_SYMBOL(__getblk_gfp);
1326 : :
1327 : : /*
1328 : : * Do async read-ahead on a buffer..
1329 : : */
1330 : 3 : void __breadahead(struct block_device *bdev, sector_t block, unsigned size)
1331 : : {
1332 : 3 : struct buffer_head *bh = __getblk(bdev, block, size);
1333 : 3 : if (likely(bh)) {
1334 : 3 : ll_rw_block(REQ_OP_READ, REQ_RAHEAD, 1, &bh);
1335 : 3 : brelse(bh);
1336 : : }
1337 : 3 : }
1338 : : EXPORT_SYMBOL(__breadahead);
1339 : :
1340 : 3 : void __breadahead_gfp(struct block_device *bdev, sector_t block, unsigned size,
1341 : : gfp_t gfp)
1342 : : {
1343 : 3 : struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp);
1344 : 3 : if (likely(bh)) {
1345 : 3 : ll_rw_block(REQ_OP_READ, REQ_RAHEAD, 1, &bh);
1346 : 3 : brelse(bh);
1347 : : }
1348 : 3 : }
1349 : : EXPORT_SYMBOL(__breadahead_gfp);
1350 : :
1351 : : /**
1352 : : * __bread_gfp() - reads a specified block and returns the bh
1353 : : * @bdev: the block_device to read from
1354 : : * @block: number of block
1355 : : * @size: size (in bytes) to read
1356 : : * @gfp: page allocation flag
1357 : : *
1358 : : * Reads a specified block, and returns buffer head that contains it.
1359 : : * The page cache can be allocated from non-movable area
1360 : : * not to prevent page migration if you set gfp to zero.
1361 : : * It returns NULL if the block was unreadable.
1362 : : */
1363 : : struct buffer_head *
1364 : 3 : __bread_gfp(struct block_device *bdev, sector_t block,
1365 : : unsigned size, gfp_t gfp)
1366 : : {
1367 : 3 : struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp);
1368 : :
1369 : 3 : if (likely(bh) && !buffer_uptodate(bh))
1370 : 3 : bh = __bread_slow(bh);
1371 : 3 : return bh;
1372 : : }
1373 : : EXPORT_SYMBOL(__bread_gfp);
1374 : :
1375 : : /*
1376 : : * invalidate_bh_lrus() is called rarely - but not only at unmount.
1377 : : * This doesn't race because it runs in each cpu either in irq
1378 : : * or with preempt disabled.
1379 : : */
1380 : 3 : static void invalidate_bh_lru(void *arg)
1381 : : {
1382 : 3 : struct bh_lru *b = &get_cpu_var(bh_lrus);
1383 : : int i;
1384 : :
1385 : 3 : for (i = 0; i < BH_LRU_SIZE; i++) {
1386 : 3 : brelse(b->bhs[i]);
1387 : 3 : b->bhs[i] = NULL;
1388 : : }
1389 : 3 : put_cpu_var(bh_lrus);
1390 : 3 : }
1391 : :
1392 : 3 : static bool has_bh_in_lru(int cpu, void *dummy)
1393 : : {
1394 : 3 : struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu);
1395 : : int i;
1396 : :
1397 : 3 : for (i = 0; i < BH_LRU_SIZE; i++) {
1398 : 3 : if (b->bhs[i])
1399 : : return 1;
1400 : : }
1401 : :
1402 : : return 0;
1403 : : }
1404 : :
1405 : 3 : void invalidate_bh_lrus(void)
1406 : : {
1407 : 3 : on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1, GFP_KERNEL);
1408 : 3 : }
1409 : : EXPORT_SYMBOL_GPL(invalidate_bh_lrus);
1410 : :
1411 : 3 : void set_bh_page(struct buffer_head *bh,
1412 : : struct page *page, unsigned long offset)
1413 : : {
1414 : 3 : bh->b_page = page;
1415 : 3 : BUG_ON(offset >= PAGE_SIZE);
1416 : : if (PageHighMem(page))
1417 : : /*
1418 : : * This catches illegal uses and preserves the offset:
1419 : : */
1420 : : bh->b_data = (char *)(0 + offset);
1421 : : else
1422 : 3 : bh->b_data = page_address(page) + offset;
1423 : 3 : }
1424 : : EXPORT_SYMBOL(set_bh_page);
1425 : :
1426 : : /*
1427 : : * Called when truncating a buffer on a page completely.
1428 : : */
1429 : :
1430 : : /* Bits that are cleared during an invalidate */
1431 : : #define BUFFER_FLAGS_DISCARD \
1432 : : (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \
1433 : : 1 << BH_Delay | 1 << BH_Unwritten)
1434 : :
1435 : 3 : static void discard_buffer(struct buffer_head * bh)
1436 : : {
1437 : : unsigned long b_state, b_state_old;
1438 : :
1439 : 3 : lock_buffer(bh);
1440 : : clear_buffer_dirty(bh);
1441 : 3 : bh->b_bdev = NULL;
1442 : 3 : b_state = bh->b_state;
1443 : : for (;;) {
1444 : 3 : b_state_old = cmpxchg(&bh->b_state, b_state,
1445 : : (b_state & ~BUFFER_FLAGS_DISCARD));
1446 : 3 : if (b_state_old == b_state)
1447 : : break;
1448 : : b_state = b_state_old;
1449 : : }
1450 : 3 : unlock_buffer(bh);
1451 : 3 : }
1452 : :
1453 : : /**
1454 : : * block_invalidatepage - invalidate part or all of a buffer-backed page
1455 : : *
1456 : : * @page: the page which is affected
1457 : : * @offset: start of the range to invalidate
1458 : : * @length: length of the range to invalidate
1459 : : *
1460 : : * block_invalidatepage() is called when all or part of the page has become
1461 : : * invalidated by a truncate operation.
1462 : : *
1463 : : * block_invalidatepage() does not have to release all buffers, but it must
1464 : : * ensure that no dirty buffer is left outside @offset and that no I/O
1465 : : * is underway against any of the blocks which are outside the truncation
1466 : : * point. Because the caller is about to free (and possibly reuse) those
1467 : : * blocks on-disk.
1468 : : */
1469 : 3 : void block_invalidatepage(struct page *page, unsigned int offset,
1470 : : unsigned int length)
1471 : : {
1472 : : struct buffer_head *head, *bh, *next;
1473 : : unsigned int curr_off = 0;
1474 : 3 : unsigned int stop = length + offset;
1475 : :
1476 : 3 : BUG_ON(!PageLocked(page));
1477 : 3 : if (!page_has_buffers(page))
1478 : : goto out;
1479 : :
1480 : : /*
1481 : : * Check for overflow
1482 : : */
1483 : 3 : BUG_ON(stop > PAGE_SIZE || stop < length);
1484 : :
1485 : 3 : head = page_buffers(page);
1486 : : bh = head;
1487 : : do {
1488 : 3 : unsigned int next_off = curr_off + bh->b_size;
1489 : 3 : next = bh->b_this_page;
1490 : :
1491 : : /*
1492 : : * Are we still fully in range ?
1493 : : */
1494 : 3 : if (next_off > stop)
1495 : : goto out;
1496 : :
1497 : : /*
1498 : : * is this block fully invalidated?
1499 : : */
1500 : 3 : if (offset <= curr_off)
1501 : 3 : discard_buffer(bh);
1502 : : curr_off = next_off;
1503 : : bh = next;
1504 : 3 : } while (bh != head);
1505 : :
1506 : : /*
1507 : : * We release buffers only if the entire page is being invalidated.
1508 : : * The get_block cached value has been unconditionally invalidated,
1509 : : * so real IO is not possible anymore.
1510 : : */
1511 : 3 : if (length == PAGE_SIZE)
1512 : 3 : try_to_release_page(page, 0);
1513 : : out:
1514 : 3 : return;
1515 : : }
1516 : : EXPORT_SYMBOL(block_invalidatepage);
1517 : :
1518 : :
1519 : : /*
1520 : : * We attach and possibly dirty the buffers atomically wrt
1521 : : * __set_page_dirty_buffers() via private_lock. try_to_free_buffers
1522 : : * is already excluded via the page lock.
1523 : : */
1524 : 3 : void create_empty_buffers(struct page *page,
1525 : : unsigned long blocksize, unsigned long b_state)
1526 : : {
1527 : : struct buffer_head *bh, *head, *tail;
1528 : :
1529 : 3 : head = alloc_page_buffers(page, blocksize, true);
1530 : : bh = head;
1531 : : do {
1532 : 3 : bh->b_state |= b_state;
1533 : : tail = bh;
1534 : 3 : bh = bh->b_this_page;
1535 : 3 : } while (bh);
1536 : 3 : tail->b_this_page = head;
1537 : :
1538 : 3 : spin_lock(&page->mapping->private_lock);
1539 : 3 : if (PageUptodate(page) || PageDirty(page)) {
1540 : : bh = head;
1541 : : do {
1542 : 3 : if (PageDirty(page))
1543 : : set_buffer_dirty(bh);
1544 : 3 : if (PageUptodate(page))
1545 : : set_buffer_uptodate(bh);
1546 : 3 : bh = bh->b_this_page;
1547 : 3 : } while (bh != head);
1548 : : }
1549 : : attach_page_buffers(page, head);
1550 : 3 : spin_unlock(&page->mapping->private_lock);
1551 : 3 : }
1552 : : EXPORT_SYMBOL(create_empty_buffers);
1553 : :
1554 : : /**
1555 : : * clean_bdev_aliases: clean a range of buffers in block device
1556 : : * @bdev: Block device to clean buffers in
1557 : : * @block: Start of a range of blocks to clean
1558 : : * @len: Number of blocks to clean
1559 : : *
1560 : : * We are taking a range of blocks for data and we don't want writeback of any
1561 : : * buffer-cache aliases starting from return from this function and until the
1562 : : * moment when something will explicitly mark the buffer dirty (hopefully that
1563 : : * will not happen until we will free that block ;-) We don't even need to mark
1564 : : * it not-uptodate - nobody can expect anything from a newly allocated buffer
1565 : : * anyway. We used to use unmap_buffer() for such invalidation, but that was
1566 : : * wrong. We definitely don't want to mark the alias unmapped, for example - it
1567 : : * would confuse anyone who might pick it with bread() afterwards...
1568 : : *
1569 : : * Also.. Note that bforget() doesn't lock the buffer. So there can be
1570 : : * writeout I/O going on against recently-freed buffers. We don't wait on that
1571 : : * I/O in bforget() - it's more efficient to wait on the I/O only if we really
1572 : : * need to. That happens here.
1573 : : */
1574 : 1 : void clean_bdev_aliases(struct block_device *bdev, sector_t block, sector_t len)
1575 : : {
1576 : 1 : struct inode *bd_inode = bdev->bd_inode;
1577 : 1 : struct address_space *bd_mapping = bd_inode->i_mapping;
1578 : : struct pagevec pvec;
1579 : 1 : pgoff_t index = block >> (PAGE_SHIFT - bd_inode->i_blkbits);
1580 : : pgoff_t end;
1581 : : int i, count;
1582 : : struct buffer_head *bh;
1583 : : struct buffer_head *head;
1584 : :
1585 : 1 : end = (block + len - 1) >> (PAGE_SHIFT - bd_inode->i_blkbits);
1586 : : pagevec_init(&pvec);
1587 : 1 : while (pagevec_lookup_range(&pvec, bd_mapping, &index, end)) {
1588 : 0 : count = pagevec_count(&pvec);
1589 : 0 : for (i = 0; i < count; i++) {
1590 : 0 : struct page *page = pvec.pages[i];
1591 : :
1592 : 0 : if (!page_has_buffers(page))
1593 : 0 : continue;
1594 : : /*
1595 : : * We use page lock instead of bd_mapping->private_lock
1596 : : * to pin buffers here since we can afford to sleep and
1597 : : * it scales better than a global spinlock lock.
1598 : : */
1599 : 0 : lock_page(page);
1600 : : /* Recheck when the page is locked which pins bhs */
1601 : 0 : if (!page_has_buffers(page))
1602 : : goto unlock_page;
1603 : 0 : head = page_buffers(page);
1604 : : bh = head;
1605 : : do {
1606 : 0 : if (!buffer_mapped(bh) || (bh->b_blocknr < block))
1607 : : goto next;
1608 : 0 : if (bh->b_blocknr >= block + len)
1609 : : break;
1610 : : clear_buffer_dirty(bh);
1611 : 0 : wait_on_buffer(bh);
1612 : : clear_buffer_req(bh);
1613 : : next:
1614 : 0 : bh = bh->b_this_page;
1615 : 0 : } while (bh != head);
1616 : : unlock_page:
1617 : 0 : unlock_page(page);
1618 : : }
1619 : : pagevec_release(&pvec);
1620 : 0 : cond_resched();
1621 : : /* End of range already reached? */
1622 : 0 : if (index > end || !index)
1623 : : break;
1624 : : }
1625 : 1 : }
1626 : : EXPORT_SYMBOL(clean_bdev_aliases);
1627 : :
1628 : : /*
1629 : : * Size is a power-of-two in the range 512..PAGE_SIZE,
1630 : : * and the case we care about most is PAGE_SIZE.
1631 : : *
1632 : : * So this *could* possibly be written with those
1633 : : * constraints in mind (relevant mostly if some
1634 : : * architecture has a slow bit-scan instruction)
1635 : : */
1636 : 3 : static inline int block_size_bits(unsigned int blocksize)
1637 : : {
1638 : 3 : return ilog2(blocksize);
1639 : : }
1640 : :
1641 : 3 : static struct buffer_head *create_page_buffers(struct page *page, struct inode *inode, unsigned int b_state)
1642 : : {
1643 : 3 : BUG_ON(!PageLocked(page));
1644 : :
1645 : 3 : if (!page_has_buffers(page))
1646 : 3 : create_empty_buffers(page, 1 << READ_ONCE(inode->i_blkbits),
1647 : : b_state);
1648 : 3 : return page_buffers(page);
1649 : : }
1650 : :
1651 : : /*
1652 : : * NOTE! All mapped/uptodate combinations are valid:
1653 : : *
1654 : : * Mapped Uptodate Meaning
1655 : : *
1656 : : * No No "unknown" - must do get_block()
1657 : : * No Yes "hole" - zero-filled
1658 : : * Yes No "allocated" - allocated on disk, not read in
1659 : : * Yes Yes "valid" - allocated and up-to-date in memory.
1660 : : *
1661 : : * "Dirty" is valid only with the last case (mapped+uptodate).
1662 : : */
1663 : :
1664 : : /*
1665 : : * While block_write_full_page is writing back the dirty buffers under
1666 : : * the page lock, whoever dirtied the buffers may decide to clean them
1667 : : * again at any time. We handle that by only looking at the buffer
1668 : : * state inside lock_buffer().
1669 : : *
1670 : : * If block_write_full_page() is called for regular writeback
1671 : : * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1672 : : * locked buffer. This only can happen if someone has written the buffer
1673 : : * directly, with submit_bh(). At the address_space level PageWriteback
1674 : : * prevents this contention from occurring.
1675 : : *
1676 : : * If block_write_full_page() is called with wbc->sync_mode ==
1677 : : * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this
1678 : : * causes the writes to be flagged as synchronous writes.
1679 : : */
1680 : 3 : int __block_write_full_page(struct inode *inode, struct page *page,
1681 : : get_block_t *get_block, struct writeback_control *wbc,
1682 : : bh_end_io_t *handler)
1683 : : {
1684 : : int err;
1685 : : sector_t block;
1686 : : sector_t last_block;
1687 : : struct buffer_head *bh, *head;
1688 : : unsigned int blocksize, bbits;
1689 : : int nr_underway = 0;
1690 : : int write_flags = wbc_to_write_flags(wbc);
1691 : :
1692 : 3 : head = create_page_buffers(page, inode,
1693 : : (1 << BH_Dirty)|(1 << BH_Uptodate));
1694 : :
1695 : : /*
1696 : : * Be very careful. We have no exclusion from __set_page_dirty_buffers
1697 : : * here, and the (potentially unmapped) buffers may become dirty at
1698 : : * any time. If a buffer becomes dirty here after we've inspected it
1699 : : * then we just miss that fact, and the page stays dirty.
1700 : : *
1701 : : * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
1702 : : * handle that here by just cleaning them.
1703 : : */
1704 : :
1705 : : bh = head;
1706 : 3 : blocksize = bh->b_size;
1707 : 3 : bbits = block_size_bits(blocksize);
1708 : :
1709 : 3 : block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1710 : 3 : last_block = (i_size_read(inode) - 1) >> bbits;
1711 : :
1712 : : /*
1713 : : * Get all the dirty buffers mapped to disk addresses and
1714 : : * handle any aliases from the underlying blockdev's mapping.
1715 : : */
1716 : : do {
1717 : 3 : if (block > last_block) {
1718 : : /*
1719 : : * mapped buffers outside i_size will occur, because
1720 : : * this page can be outside i_size when there is a
1721 : : * truncate in progress.
1722 : : */
1723 : : /*
1724 : : * The buffer was zeroed by block_write_full_page()
1725 : : */
1726 : : clear_buffer_dirty(bh);
1727 : : set_buffer_uptodate(bh);
1728 : 3 : } else if ((!buffer_mapped(bh) || buffer_delay(bh)) &&
1729 : : buffer_dirty(bh)) {
1730 : 0 : WARN_ON(bh->b_size != blocksize);
1731 : 0 : err = get_block(inode, block, bh, 1);
1732 : 0 : if (err)
1733 : : goto recover;
1734 : : clear_buffer_delay(bh);
1735 : 0 : if (buffer_new(bh)) {
1736 : : /* blockdev mappings never come here */
1737 : : clear_buffer_new(bh);
1738 : 0 : clean_bdev_bh_alias(bh);
1739 : : }
1740 : : }
1741 : 3 : bh = bh->b_this_page;
1742 : 3 : block++;
1743 : 3 : } while (bh != head);
1744 : :
1745 : : do {
1746 : 3 : if (!buffer_mapped(bh))
1747 : 0 : continue;
1748 : : /*
1749 : : * If it's a fully non-blocking write attempt and we cannot
1750 : : * lock the buffer then redirty the page. Note that this can
1751 : : * potentially cause a busy-wait loop from writeback threads
1752 : : * and kswapd activity, but those code paths have their own
1753 : : * higher-level throttling.
1754 : : */
1755 : 3 : if (wbc->sync_mode != WB_SYNC_NONE) {
1756 : 3 : lock_buffer(bh);
1757 : 3 : } else if (!trylock_buffer(bh)) {
1758 : 0 : redirty_page_for_writepage(wbc, page);
1759 : 0 : continue;
1760 : : }
1761 : 3 : if (test_clear_buffer_dirty(bh)) {
1762 : 3 : mark_buffer_async_write_endio(bh, handler);
1763 : : } else {
1764 : 3 : unlock_buffer(bh);
1765 : : }
1766 : 3 : } while ((bh = bh->b_this_page) != head);
1767 : :
1768 : : /*
1769 : : * The page and its buffers are protected by PageWriteback(), so we can
1770 : : * drop the bh refcounts early.
1771 : : */
1772 : 3 : BUG_ON(PageWriteback(page));
1773 : : set_page_writeback(page);
1774 : :
1775 : : do {
1776 : 3 : struct buffer_head *next = bh->b_this_page;
1777 : 3 : if (buffer_async_write(bh)) {
1778 : 3 : submit_bh_wbc(REQ_OP_WRITE, write_flags, bh,
1779 : 3 : inode->i_write_hint, wbc);
1780 : 3 : nr_underway++;
1781 : : }
1782 : : bh = next;
1783 : 3 : } while (bh != head);
1784 : 3 : unlock_page(page);
1785 : :
1786 : : err = 0;
1787 : : done:
1788 : 3 : if (nr_underway == 0) {
1789 : : /*
1790 : : * The page was marked dirty, but the buffers were
1791 : : * clean. Someone wrote them back by hand with
1792 : : * ll_rw_block/submit_bh. A rare case.
1793 : : */
1794 : 3 : end_page_writeback(page);
1795 : :
1796 : : /*
1797 : : * The page and buffer_heads can be released at any time from
1798 : : * here on.
1799 : : */
1800 : : }
1801 : 3 : return err;
1802 : :
1803 : : recover:
1804 : : /*
1805 : : * ENOSPC, or some other error. We may already have added some
1806 : : * blocks to the file, so we need to write these out to avoid
1807 : : * exposing stale data.
1808 : : * The page is currently locked and not marked for writeback
1809 : : */
1810 : : bh = head;
1811 : : /* Recovery: lock and submit the mapped buffers */
1812 : : do {
1813 : 0 : if (buffer_mapped(bh) && buffer_dirty(bh) &&
1814 : : !buffer_delay(bh)) {
1815 : 0 : lock_buffer(bh);
1816 : 0 : mark_buffer_async_write_endio(bh, handler);
1817 : : } else {
1818 : : /*
1819 : : * The buffer may have been set dirty during
1820 : : * attachment to a dirty page.
1821 : : */
1822 : : clear_buffer_dirty(bh);
1823 : : }
1824 : 0 : } while ((bh = bh->b_this_page) != head);
1825 : : SetPageError(page);
1826 : 0 : BUG_ON(PageWriteback(page));
1827 : 0 : mapping_set_error(page->mapping, err);
1828 : : set_page_writeback(page);
1829 : : do {
1830 : 0 : struct buffer_head *next = bh->b_this_page;
1831 : 0 : if (buffer_async_write(bh)) {
1832 : : clear_buffer_dirty(bh);
1833 : 0 : submit_bh_wbc(REQ_OP_WRITE, write_flags, bh,
1834 : 0 : inode->i_write_hint, wbc);
1835 : 0 : nr_underway++;
1836 : : }
1837 : : bh = next;
1838 : 0 : } while (bh != head);
1839 : 0 : unlock_page(page);
1840 : 0 : goto done;
1841 : : }
1842 : : EXPORT_SYMBOL(__block_write_full_page);
1843 : :
1844 : : /*
1845 : : * If a page has any new buffers, zero them out here, and mark them uptodate
1846 : : * and dirty so they'll be written out (in order to prevent uninitialised
1847 : : * block data from leaking). And clear the new bit.
1848 : : */
1849 : 0 : void page_zero_new_buffers(struct page *page, unsigned from, unsigned to)
1850 : : {
1851 : : unsigned int block_start, block_end;
1852 : : struct buffer_head *head, *bh;
1853 : :
1854 : 0 : BUG_ON(!PageLocked(page));
1855 : 0 : if (!page_has_buffers(page))
1856 : 0 : return;
1857 : :
1858 : 0 : bh = head = page_buffers(page);
1859 : : block_start = 0;
1860 : : do {
1861 : 0 : block_end = block_start + bh->b_size;
1862 : :
1863 : 0 : if (buffer_new(bh)) {
1864 : 0 : if (block_end > from && block_start < to) {
1865 : 0 : if (!PageUptodate(page)) {
1866 : : unsigned start, size;
1867 : :
1868 : 0 : start = max(from, block_start);
1869 : 0 : size = min(to, block_end) - start;
1870 : :
1871 : : zero_user(page, start, size);
1872 : : set_buffer_uptodate(bh);
1873 : : }
1874 : :
1875 : : clear_buffer_new(bh);
1876 : 0 : mark_buffer_dirty(bh);
1877 : : }
1878 : : }
1879 : :
1880 : : block_start = block_end;
1881 : 0 : bh = bh->b_this_page;
1882 : 0 : } while (bh != head);
1883 : : }
1884 : : EXPORT_SYMBOL(page_zero_new_buffers);
1885 : :
1886 : : static void
1887 : 0 : iomap_to_bh(struct inode *inode, sector_t block, struct buffer_head *bh,
1888 : : struct iomap *iomap)
1889 : : {
1890 : 0 : loff_t offset = block << inode->i_blkbits;
1891 : :
1892 : 0 : bh->b_bdev = iomap->bdev;
1893 : :
1894 : : /*
1895 : : * Block points to offset in file we need to map, iomap contains
1896 : : * the offset at which the map starts. If the map ends before the
1897 : : * current block, then do not map the buffer and let the caller
1898 : : * handle it.
1899 : : */
1900 : 0 : BUG_ON(offset >= iomap->offset + iomap->length);
1901 : :
1902 : 0 : switch (iomap->type) {
1903 : : case IOMAP_HOLE:
1904 : : /*
1905 : : * If the buffer is not up to date or beyond the current EOF,
1906 : : * we need to mark it as new to ensure sub-block zeroing is
1907 : : * executed if necessary.
1908 : : */
1909 : 0 : if (!buffer_uptodate(bh) ||
1910 : : (offset >= i_size_read(inode)))
1911 : : set_buffer_new(bh);
1912 : : break;
1913 : : case IOMAP_DELALLOC:
1914 : 0 : if (!buffer_uptodate(bh) ||
1915 : : (offset >= i_size_read(inode)))
1916 : : set_buffer_new(bh);
1917 : : set_buffer_uptodate(bh);
1918 : : set_buffer_mapped(bh);
1919 : : set_buffer_delay(bh);
1920 : : break;
1921 : : case IOMAP_UNWRITTEN:
1922 : : /*
1923 : : * For unwritten regions, we always need to ensure that regions
1924 : : * in the block we are not writing to are zeroed. Mark the
1925 : : * buffer as new to ensure this.
1926 : : */
1927 : : set_buffer_new(bh);
1928 : : set_buffer_unwritten(bh);
1929 : : /* FALLTHRU */
1930 : : case IOMAP_MAPPED:
1931 : 0 : if ((iomap->flags & IOMAP_F_NEW) ||
1932 : : offset >= i_size_read(inode))
1933 : : set_buffer_new(bh);
1934 : 0 : bh->b_blocknr = (iomap->addr + offset - iomap->offset) >>
1935 : 0 : inode->i_blkbits;
1936 : : set_buffer_mapped(bh);
1937 : : break;
1938 : : }
1939 : 0 : }
1940 : :
1941 : 3 : int __block_write_begin_int(struct page *page, loff_t pos, unsigned len,
1942 : : get_block_t *get_block, struct iomap *iomap)
1943 : : {
1944 : 3 : unsigned from = pos & (PAGE_SIZE - 1);
1945 : 3 : unsigned to = from + len;
1946 : 3 : struct inode *inode = page->mapping->host;
1947 : : unsigned block_start, block_end;
1948 : : sector_t block;
1949 : : int err = 0;
1950 : : unsigned blocksize, bbits;
1951 : : struct buffer_head *bh, *head, *wait[2], **wait_bh=wait;
1952 : :
1953 : 3 : BUG_ON(!PageLocked(page));
1954 : : BUG_ON(from > PAGE_SIZE);
1955 : 3 : BUG_ON(to > PAGE_SIZE);
1956 : 3 : BUG_ON(from > to);
1957 : :
1958 : 3 : head = create_page_buffers(page, inode, 0);
1959 : 3 : blocksize = head->b_size;
1960 : 3 : bbits = block_size_bits(blocksize);
1961 : :
1962 : 3 : block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1963 : :
1964 : 3 : for(bh = head, block_start = 0; bh != head || !block_start;
1965 : 3 : block++, block_start=block_end, bh = bh->b_this_page) {
1966 : 3 : block_end = block_start + blocksize;
1967 : 3 : if (block_end <= from || block_start >= to) {
1968 : 0 : if (PageUptodate(page)) {
1969 : 0 : if (!buffer_uptodate(bh))
1970 : 0 : set_buffer_uptodate(bh);
1971 : : }
1972 : 0 : continue;
1973 : : }
1974 : 3 : if (buffer_new(bh))
1975 : 0 : clear_buffer_new(bh);
1976 : 3 : if (!buffer_mapped(bh)) {
1977 : 1 : WARN_ON(bh->b_size != blocksize);
1978 : 1 : if (get_block) {
1979 : 1 : err = get_block(inode, block, bh, 1);
1980 : 1 : if (err)
1981 : : break;
1982 : : } else {
1983 : 0 : iomap_to_bh(inode, block, bh, iomap);
1984 : : }
1985 : :
1986 : 1 : if (buffer_new(bh)) {
1987 : 1 : clean_bdev_bh_alias(bh);
1988 : 1 : if (PageUptodate(page)) {
1989 : 1 : clear_buffer_new(bh);
1990 : 1 : set_buffer_uptodate(bh);
1991 : 1 : mark_buffer_dirty(bh);
1992 : 1 : continue;
1993 : : }
1994 : 0 : if (block_end > to || block_start < from)
1995 : 0 : zero_user_segments(page,
1996 : : to, block_end,
1997 : : block_start, from);
1998 : 0 : continue;
1999 : : }
2000 : : }
2001 : 2 : if (PageUptodate(page)) {
2002 : 2 : if (!buffer_uptodate(bh))
2003 : 0 : set_buffer_uptodate(bh);
2004 : 2 : continue;
2005 : : }
2006 : 0 : if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
2007 : 0 : !buffer_unwritten(bh) &&
2008 : 0 : (block_start < from || block_end > to)) {
2009 : 0 : ll_rw_block(REQ_OP_READ, 0, 1, &bh);
2010 : 0 : *wait_bh++=bh;
2011 : : }
2012 : : }
2013 : : /*
2014 : : * If we issued read requests - let them complete.
2015 : : */
2016 : 3 : while(wait_bh > wait) {
2017 : 0 : wait_on_buffer(*--wait_bh);
2018 : 0 : if (!buffer_uptodate(*wait_bh))
2019 : : err = -EIO;
2020 : : }
2021 : 3 : if (unlikely(err))
2022 : 0 : page_zero_new_buffers(page, from, to);
2023 : 3 : return err;
2024 : : }
2025 : :
2026 : 0 : int __block_write_begin(struct page *page, loff_t pos, unsigned len,
2027 : : get_block_t *get_block)
2028 : : {
2029 : 3 : return __block_write_begin_int(page, pos, len, get_block, NULL);
2030 : : }
2031 : : EXPORT_SYMBOL(__block_write_begin);
2032 : :
2033 : 3 : static int __block_commit_write(struct inode *inode, struct page *page,
2034 : : unsigned from, unsigned to)
2035 : : {
2036 : : unsigned block_start, block_end;
2037 : : int partial = 0;
2038 : : unsigned blocksize;
2039 : : struct buffer_head *bh, *head;
2040 : :
2041 : 3 : bh = head = page_buffers(page);
2042 : 3 : blocksize = bh->b_size;
2043 : :
2044 : : block_start = 0;
2045 : : do {
2046 : 3 : block_end = block_start + blocksize;
2047 : 3 : if (block_end <= from || block_start >= to) {
2048 : 0 : if (!buffer_uptodate(bh))
2049 : : partial = 1;
2050 : : } else {
2051 : : set_buffer_uptodate(bh);
2052 : 3 : mark_buffer_dirty(bh);
2053 : : }
2054 : : clear_buffer_new(bh);
2055 : :
2056 : : block_start = block_end;
2057 : 3 : bh = bh->b_this_page;
2058 : 3 : } while (bh != head);
2059 : :
2060 : : /*
2061 : : * If this is a partial write which happened to make all buffers
2062 : : * uptodate then we can optimize away a bogus readpage() for
2063 : : * the next read(). Here we 'discover' whether the page went
2064 : : * uptodate as a result of this (potentially partial) write.
2065 : : */
2066 : 3 : if (!partial)
2067 : : SetPageUptodate(page);
2068 : 3 : return 0;
2069 : : }
2070 : :
2071 : : /*
2072 : : * block_write_begin takes care of the basic task of block allocation and
2073 : : * bringing partial write blocks uptodate first.
2074 : : *
2075 : : * The filesystem needs to handle block truncation upon failure.
2076 : : */
2077 : 0 : int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,
2078 : : unsigned flags, struct page **pagep, get_block_t *get_block)
2079 : : {
2080 : 0 : pgoff_t index = pos >> PAGE_SHIFT;
2081 : : struct page *page;
2082 : : int status;
2083 : :
2084 : 0 : page = grab_cache_page_write_begin(mapping, index, flags);
2085 : 0 : if (!page)
2086 : : return -ENOMEM;
2087 : :
2088 : : status = __block_write_begin(page, pos, len, get_block);
2089 : 0 : if (unlikely(status)) {
2090 : 0 : unlock_page(page);
2091 : 0 : put_page(page);
2092 : : page = NULL;
2093 : : }
2094 : :
2095 : 0 : *pagep = page;
2096 : 0 : return status;
2097 : : }
2098 : : EXPORT_SYMBOL(block_write_begin);
2099 : :
2100 : 3 : int block_write_end(struct file *file, struct address_space *mapping,
2101 : : loff_t pos, unsigned len, unsigned copied,
2102 : : struct page *page, void *fsdata)
2103 : : {
2104 : 3 : struct inode *inode = mapping->host;
2105 : : unsigned start;
2106 : :
2107 : 3 : start = pos & (PAGE_SIZE - 1);
2108 : :
2109 : 3 : if (unlikely(copied < len)) {
2110 : : /*
2111 : : * The buffers that were written will now be uptodate, so we
2112 : : * don't have to worry about a readpage reading them and
2113 : : * overwriting a partial write. However if we have encountered
2114 : : * a short write and only partially written into a buffer, it
2115 : : * will not be marked uptodate, so a readpage might come in and
2116 : : * destroy our partial write.
2117 : : *
2118 : : * Do the simplest thing, and just treat any short write to a
2119 : : * non uptodate page as a zero-length write, and force the
2120 : : * caller to redo the whole thing.
2121 : : */
2122 : 0 : if (!PageUptodate(page))
2123 : : copied = 0;
2124 : :
2125 : 0 : page_zero_new_buffers(page, start+copied, start+len);
2126 : : }
2127 : 3 : flush_dcache_page(page);
2128 : :
2129 : : /* This could be a short (even 0-length) commit */
2130 : 3 : __block_commit_write(inode, page, start, start+copied);
2131 : :
2132 : 3 : return copied;
2133 : : }
2134 : : EXPORT_SYMBOL(block_write_end);
2135 : :
2136 : 3 : int generic_write_end(struct file *file, struct address_space *mapping,
2137 : : loff_t pos, unsigned len, unsigned copied,
2138 : : struct page *page, void *fsdata)
2139 : : {
2140 : 3 : struct inode *inode = mapping->host;
2141 : 3 : loff_t old_size = inode->i_size;
2142 : : bool i_size_changed = false;
2143 : :
2144 : 3 : copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
2145 : :
2146 : : /*
2147 : : * No need to use i_size_read() here, the i_size cannot change under us
2148 : : * because we hold i_rwsem.
2149 : : *
2150 : : * But it's important to update i_size while still holding page lock:
2151 : : * page writeout could otherwise come in and zero beyond i_size.
2152 : : */
2153 : 3 : if (pos + copied > inode->i_size) {
2154 : : i_size_write(inode, pos + copied);
2155 : : i_size_changed = true;
2156 : : }
2157 : :
2158 : 3 : unlock_page(page);
2159 : 3 : put_page(page);
2160 : :
2161 : 3 : if (old_size < pos)
2162 : 0 : pagecache_isize_extended(inode, old_size, pos);
2163 : : /*
2164 : : * Don't mark the inode dirty under page lock. First, it unnecessarily
2165 : : * makes the holding time of page lock longer. Second, it forces lock
2166 : : * ordering of page lock and transaction start for journaling
2167 : : * filesystems.
2168 : : */
2169 : 3 : if (i_size_changed)
2170 : : mark_inode_dirty(inode);
2171 : 3 : return copied;
2172 : : }
2173 : : EXPORT_SYMBOL(generic_write_end);
2174 : :
2175 : : /*
2176 : : * block_is_partially_uptodate checks whether buffers within a page are
2177 : : * uptodate or not.
2178 : : *
2179 : : * Returns true if all buffers which correspond to a file portion
2180 : : * we want to read are uptodate.
2181 : : */
2182 : 0 : int block_is_partially_uptodate(struct page *page, unsigned long from,
2183 : : unsigned long count)
2184 : : {
2185 : : unsigned block_start, block_end, blocksize;
2186 : : unsigned to;
2187 : : struct buffer_head *bh, *head;
2188 : : int ret = 1;
2189 : :
2190 : 0 : if (!page_has_buffers(page))
2191 : : return 0;
2192 : :
2193 : 0 : head = page_buffers(page);
2194 : 0 : blocksize = head->b_size;
2195 : 0 : to = min_t(unsigned, PAGE_SIZE - from, count);
2196 : 0 : to = from + to;
2197 : 0 : if (from < blocksize && to > PAGE_SIZE - blocksize)
2198 : : return 0;
2199 : :
2200 : : bh = head;
2201 : : block_start = 0;
2202 : : do {
2203 : 0 : block_end = block_start + blocksize;
2204 : 0 : if (block_end > from && block_start < to) {
2205 : 0 : if (!buffer_uptodate(bh)) {
2206 : : ret = 0;
2207 : : break;
2208 : : }
2209 : 0 : if (block_end >= to)
2210 : : break;
2211 : : }
2212 : : block_start = block_end;
2213 : 0 : bh = bh->b_this_page;
2214 : 0 : } while (bh != head);
2215 : :
2216 : 0 : return ret;
2217 : : }
2218 : : EXPORT_SYMBOL(block_is_partially_uptodate);
2219 : :
2220 : : /*
2221 : : * Generic "read page" function for block devices that have the normal
2222 : : * get_block functionality. This is most of the block device filesystems.
2223 : : * Reads the page asynchronously --- the unlock_buffer() and
2224 : : * set/clear_buffer_uptodate() functions propagate buffer state into the
2225 : : * page struct once IO has completed.
2226 : : */
2227 : 3 : int block_read_full_page(struct page *page, get_block_t *get_block)
2228 : : {
2229 : 3 : struct inode *inode = page->mapping->host;
2230 : : sector_t iblock, lblock;
2231 : : struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
2232 : : unsigned int blocksize, bbits;
2233 : : int nr, i;
2234 : : int fully_mapped = 1;
2235 : :
2236 : 3 : head = create_page_buffers(page, inode, 0);
2237 : 3 : blocksize = head->b_size;
2238 : 3 : bbits = block_size_bits(blocksize);
2239 : :
2240 : 3 : iblock = (sector_t)page->index << (PAGE_SHIFT - bbits);
2241 : 3 : lblock = (i_size_read(inode)+blocksize-1) >> bbits;
2242 : : bh = head;
2243 : : nr = 0;
2244 : : i = 0;
2245 : :
2246 : : do {
2247 : 3 : if (buffer_uptodate(bh))
2248 : 3 : continue;
2249 : :
2250 : 3 : if (!buffer_mapped(bh)) {
2251 : : int err = 0;
2252 : :
2253 : : fully_mapped = 0;
2254 : 3 : if (iblock < lblock) {
2255 : 3 : WARN_ON(bh->b_size != blocksize);
2256 : 3 : err = get_block(inode, iblock, bh, 0);
2257 : 3 : if (err)
2258 : : SetPageError(page);
2259 : : }
2260 : 3 : if (!buffer_mapped(bh)) {
2261 : 0 : zero_user(page, i * blocksize, blocksize);
2262 : 0 : if (!err)
2263 : : set_buffer_uptodate(bh);
2264 : 0 : continue;
2265 : : }
2266 : : /*
2267 : : * get_block() might have updated the buffer
2268 : : * synchronously
2269 : : */
2270 : 3 : if (buffer_uptodate(bh))
2271 : 0 : continue;
2272 : : }
2273 : 3 : arr[nr++] = bh;
2274 : 3 : } while (i++, iblock++, (bh = bh->b_this_page) != head);
2275 : :
2276 : 3 : if (fully_mapped)
2277 : : SetPageMappedToDisk(page);
2278 : :
2279 : 3 : if (!nr) {
2280 : : /*
2281 : : * All buffers are uptodate - we can set the page uptodate
2282 : : * as well. But not if get_block() returned an error.
2283 : : */
2284 : 3 : if (!PageError(page))
2285 : : SetPageUptodate(page);
2286 : 3 : unlock_page(page);
2287 : 3 : return 0;
2288 : : }
2289 : :
2290 : : /* Stage two: lock the buffers */
2291 : 3 : for (i = 0; i < nr; i++) {
2292 : 3 : bh = arr[i];
2293 : 3 : lock_buffer(bh);
2294 : 3 : mark_buffer_async_read(bh);
2295 : : }
2296 : :
2297 : : /*
2298 : : * Stage 3: start the IO. Check for uptodateness
2299 : : * inside the buffer lock in case another process reading
2300 : : * the underlying blockdev brought it uptodate (the sct fix).
2301 : : */
2302 : 3 : for (i = 0; i < nr; i++) {
2303 : 3 : bh = arr[i];
2304 : 3 : if (buffer_uptodate(bh))
2305 : 0 : end_buffer_async_read(bh, 1);
2306 : : else
2307 : : submit_bh(REQ_OP_READ, 0, bh);
2308 : : }
2309 : : return 0;
2310 : : }
2311 : : EXPORT_SYMBOL(block_read_full_page);
2312 : :
2313 : : /* utility function for filesystems that need to do work on expanding
2314 : : * truncates. Uses filesystem pagecache writes to allow the filesystem to
2315 : : * deal with the hole.
2316 : : */
2317 : 0 : int generic_cont_expand_simple(struct inode *inode, loff_t size)
2318 : : {
2319 : 0 : struct address_space *mapping = inode->i_mapping;
2320 : : struct page *page;
2321 : : void *fsdata;
2322 : : int err;
2323 : :
2324 : 0 : err = inode_newsize_ok(inode, size);
2325 : 0 : if (err)
2326 : : goto out;
2327 : :
2328 : 0 : err = pagecache_write_begin(NULL, mapping, size, 0,
2329 : : AOP_FLAG_CONT_EXPAND, &page, &fsdata);
2330 : 0 : if (err)
2331 : : goto out;
2332 : :
2333 : 0 : err = pagecache_write_end(NULL, mapping, size, 0, 0, page, fsdata);
2334 : 0 : BUG_ON(err > 0);
2335 : :
2336 : : out:
2337 : 0 : return err;
2338 : : }
2339 : : EXPORT_SYMBOL(generic_cont_expand_simple);
2340 : :
2341 : 0 : static int cont_expand_zero(struct file *file, struct address_space *mapping,
2342 : : loff_t pos, loff_t *bytes)
2343 : : {
2344 : 0 : struct inode *inode = mapping->host;
2345 : : unsigned int blocksize = i_blocksize(inode);
2346 : : struct page *page;
2347 : : void *fsdata;
2348 : : pgoff_t index, curidx;
2349 : : loff_t curpos;
2350 : : unsigned zerofrom, offset, len;
2351 : : int err = 0;
2352 : :
2353 : 0 : index = pos >> PAGE_SHIFT;
2354 : 0 : offset = pos & ~PAGE_MASK;
2355 : :
2356 : 0 : while (index > (curidx = (curpos = *bytes)>>PAGE_SHIFT)) {
2357 : 0 : zerofrom = curpos & ~PAGE_MASK;
2358 : 0 : if (zerofrom & (blocksize-1)) {
2359 : 0 : *bytes |= (blocksize-1);
2360 : 0 : (*bytes)++;
2361 : : }
2362 : 0 : len = PAGE_SIZE - zerofrom;
2363 : :
2364 : 0 : err = pagecache_write_begin(file, mapping, curpos, len, 0,
2365 : : &page, &fsdata);
2366 : 0 : if (err)
2367 : : goto out;
2368 : 0 : zero_user(page, zerofrom, len);
2369 : 0 : err = pagecache_write_end(file, mapping, curpos, len, len,
2370 : : page, fsdata);
2371 : 0 : if (err < 0)
2372 : : goto out;
2373 : 0 : BUG_ON(err != len);
2374 : : err = 0;
2375 : :
2376 : 0 : balance_dirty_pages_ratelimited(mapping);
2377 : :
2378 : 0 : if (fatal_signal_pending(current)) {
2379 : : err = -EINTR;
2380 : : goto out;
2381 : : }
2382 : : }
2383 : :
2384 : : /* page covers the boundary, find the boundary offset */
2385 : 0 : if (index == curidx) {
2386 : 0 : zerofrom = curpos & ~PAGE_MASK;
2387 : : /* if we will expand the thing last block will be filled */
2388 : 0 : if (offset <= zerofrom) {
2389 : : goto out;
2390 : : }
2391 : 0 : if (zerofrom & (blocksize-1)) {
2392 : 0 : *bytes |= (blocksize-1);
2393 : 0 : (*bytes)++;
2394 : : }
2395 : 0 : len = offset - zerofrom;
2396 : :
2397 : 0 : err = pagecache_write_begin(file, mapping, curpos, len, 0,
2398 : : &page, &fsdata);
2399 : 0 : if (err)
2400 : : goto out;
2401 : 0 : zero_user(page, zerofrom, len);
2402 : 0 : err = pagecache_write_end(file, mapping, curpos, len, len,
2403 : : page, fsdata);
2404 : 0 : if (err < 0)
2405 : : goto out;
2406 : 0 : BUG_ON(err != len);
2407 : : err = 0;
2408 : : }
2409 : : out:
2410 : 0 : return err;
2411 : : }
2412 : :
2413 : : /*
2414 : : * For moronic filesystems that do not allow holes in file.
2415 : : * We may have to extend the file.
2416 : : */
2417 : 0 : int cont_write_begin(struct file *file, struct address_space *mapping,
2418 : : loff_t pos, unsigned len, unsigned flags,
2419 : : struct page **pagep, void **fsdata,
2420 : : get_block_t *get_block, loff_t *bytes)
2421 : : {
2422 : 0 : struct inode *inode = mapping->host;
2423 : : unsigned int blocksize = i_blocksize(inode);
2424 : : unsigned int zerofrom;
2425 : : int err;
2426 : :
2427 : 0 : err = cont_expand_zero(file, mapping, pos, bytes);
2428 : 0 : if (err)
2429 : : return err;
2430 : :
2431 : 0 : zerofrom = *bytes & ~PAGE_MASK;
2432 : 0 : if (pos+len > *bytes && zerofrom & (blocksize-1)) {
2433 : 0 : *bytes |= (blocksize-1);
2434 : 0 : (*bytes)++;
2435 : : }
2436 : :
2437 : 0 : return block_write_begin(mapping, pos, len, flags, pagep, get_block);
2438 : : }
2439 : : EXPORT_SYMBOL(cont_write_begin);
2440 : :
2441 : 0 : int block_commit_write(struct page *page, unsigned from, unsigned to)
2442 : : {
2443 : 3 : struct inode *inode = page->mapping->host;
2444 : 3 : __block_commit_write(inode,page,from,to);
2445 : 0 : return 0;
2446 : : }
2447 : : EXPORT_SYMBOL(block_commit_write);
2448 : :
2449 : : /*
2450 : : * block_page_mkwrite() is not allowed to change the file size as it gets
2451 : : * called from a page fault handler when a page is first dirtied. Hence we must
2452 : : * be careful to check for EOF conditions here. We set the page up correctly
2453 : : * for a written page which means we get ENOSPC checking when writing into
2454 : : * holes and correct delalloc and unwritten extent mapping on filesystems that
2455 : : * support these features.
2456 : : *
2457 : : * We are not allowed to take the i_mutex here so we have to play games to
2458 : : * protect against truncate races as the page could now be beyond EOF. Because
2459 : : * truncate writes the inode size before removing pages, once we have the
2460 : : * page lock we can determine safely if the page is beyond EOF. If it is not
2461 : : * beyond EOF, then the page is guaranteed safe against truncation until we
2462 : : * unlock the page.
2463 : : *
2464 : : * Direct callers of this function should protect against filesystem freezing
2465 : : * using sb_start_pagefault() - sb_end_pagefault() functions.
2466 : : */
2467 : 3 : int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
2468 : : get_block_t get_block)
2469 : : {
2470 : 3 : struct page *page = vmf->page;
2471 : 3 : struct inode *inode = file_inode(vma->vm_file);
2472 : : unsigned long end;
2473 : : loff_t size;
2474 : : int ret;
2475 : :
2476 : 3 : lock_page(page);
2477 : : size = i_size_read(inode);
2478 : 3 : if ((page->mapping != inode->i_mapping) ||
2479 : : (page_offset(page) > size)) {
2480 : : /* We overload EFAULT to mean page got truncated */
2481 : : ret = -EFAULT;
2482 : : goto out_unlock;
2483 : : }
2484 : :
2485 : : /* page is wholly or partially inside EOF */
2486 : 3 : if (((page->index + 1) << PAGE_SHIFT) > size)
2487 : 0 : end = size & ~PAGE_MASK;
2488 : : else
2489 : : end = PAGE_SIZE;
2490 : :
2491 : : ret = __block_write_begin(page, 0, end, get_block);
2492 : 3 : if (!ret)
2493 : : ret = block_commit_write(page, 0, end);
2494 : :
2495 : 3 : if (unlikely(ret < 0))
2496 : : goto out_unlock;
2497 : 3 : set_page_dirty(page);
2498 : 3 : wait_for_stable_page(page);
2499 : 3 : return 0;
2500 : : out_unlock:
2501 : 0 : unlock_page(page);
2502 : 0 : return ret;
2503 : : }
2504 : : EXPORT_SYMBOL(block_page_mkwrite);
2505 : :
2506 : : /*
2507 : : * nobh_write_begin()'s prereads are special: the buffer_heads are freed
2508 : : * immediately, while under the page lock. So it needs a special end_io
2509 : : * handler which does not touch the bh after unlocking it.
2510 : : */
2511 : 0 : static void end_buffer_read_nobh(struct buffer_head *bh, int uptodate)
2512 : : {
2513 : 0 : __end_buffer_read_notouch(bh, uptodate);
2514 : 0 : }
2515 : :
2516 : : /*
2517 : : * Attach the singly-linked list of buffers created by nobh_write_begin, to
2518 : : * the page (converting it to circular linked list and taking care of page
2519 : : * dirty races).
2520 : : */
2521 : 0 : static void attach_nobh_buffers(struct page *page, struct buffer_head *head)
2522 : : {
2523 : : struct buffer_head *bh;
2524 : :
2525 : 0 : BUG_ON(!PageLocked(page));
2526 : :
2527 : 0 : spin_lock(&page->mapping->private_lock);
2528 : : bh = head;
2529 : : do {
2530 : 0 : if (PageDirty(page))
2531 : : set_buffer_dirty(bh);
2532 : 0 : if (!bh->b_this_page)
2533 : 0 : bh->b_this_page = head;
2534 : 0 : bh = bh->b_this_page;
2535 : 0 : } while (bh != head);
2536 : : attach_page_buffers(page, head);
2537 : 0 : spin_unlock(&page->mapping->private_lock);
2538 : 0 : }
2539 : :
2540 : : /*
2541 : : * On entry, the page is fully not uptodate.
2542 : : * On exit the page is fully uptodate in the areas outside (from,to)
2543 : : * The filesystem needs to handle block truncation upon failure.
2544 : : */
2545 : 0 : int nobh_write_begin(struct address_space *mapping,
2546 : : loff_t pos, unsigned len, unsigned flags,
2547 : : struct page **pagep, void **fsdata,
2548 : : get_block_t *get_block)
2549 : : {
2550 : 0 : struct inode *inode = mapping->host;
2551 : 0 : const unsigned blkbits = inode->i_blkbits;
2552 : 0 : const unsigned blocksize = 1 << blkbits;
2553 : : struct buffer_head *head, *bh;
2554 : : struct page *page;
2555 : : pgoff_t index;
2556 : : unsigned from, to;
2557 : : unsigned block_in_page;
2558 : : unsigned block_start, block_end;
2559 : : sector_t block_in_file;
2560 : : int nr_reads = 0;
2561 : : int ret = 0;
2562 : : int is_mapped_to_disk = 1;
2563 : :
2564 : 0 : index = pos >> PAGE_SHIFT;
2565 : 0 : from = pos & (PAGE_SIZE - 1);
2566 : 0 : to = from + len;
2567 : :
2568 : 0 : page = grab_cache_page_write_begin(mapping, index, flags);
2569 : 0 : if (!page)
2570 : : return -ENOMEM;
2571 : 0 : *pagep = page;
2572 : 0 : *fsdata = NULL;
2573 : :
2574 : 0 : if (page_has_buffers(page)) {
2575 : : ret = __block_write_begin(page, pos, len, get_block);
2576 : 0 : if (unlikely(ret))
2577 : : goto out_release;
2578 : : return ret;
2579 : : }
2580 : :
2581 : 0 : if (PageMappedToDisk(page))
2582 : : return 0;
2583 : :
2584 : : /*
2585 : : * Allocate buffers so that we can keep track of state, and potentially
2586 : : * attach them to the page if an error occurs. In the common case of
2587 : : * no error, they will just be freed again without ever being attached
2588 : : * to the page (which is all OK, because we're under the page lock).
2589 : : *
2590 : : * Be careful: the buffer linked list is a NULL terminated one, rather
2591 : : * than the circular one we're used to.
2592 : : */
2593 : 0 : head = alloc_page_buffers(page, blocksize, false);
2594 : 0 : if (!head) {
2595 : : ret = -ENOMEM;
2596 : : goto out_release;
2597 : : }
2598 : :
2599 : 0 : block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
2600 : :
2601 : : /*
2602 : : * We loop across all blocks in the page, whether or not they are
2603 : : * part of the affected region. This is so we can discover if the
2604 : : * page is fully mapped-to-disk.
2605 : : */
2606 : 0 : for (block_start = 0, block_in_page = 0, bh = head;
2607 : : block_start < PAGE_SIZE;
2608 : 0 : block_in_page++, block_start += blocksize, bh = bh->b_this_page) {
2609 : : int create;
2610 : :
2611 : 0 : block_end = block_start + blocksize;
2612 : 0 : bh->b_state = 0;
2613 : : create = 1;
2614 : 0 : if (block_start >= to)
2615 : : create = 0;
2616 : 0 : ret = get_block(inode, block_in_file + block_in_page,
2617 : : bh, create);
2618 : 0 : if (ret)
2619 : : goto failed;
2620 : 0 : if (!buffer_mapped(bh))
2621 : : is_mapped_to_disk = 0;
2622 : 0 : if (buffer_new(bh))
2623 : 0 : clean_bdev_bh_alias(bh);
2624 : 0 : if (PageUptodate(page)) {
2625 : : set_buffer_uptodate(bh);
2626 : 0 : continue;
2627 : : }
2628 : 0 : if (buffer_new(bh) || !buffer_mapped(bh)) {
2629 : 0 : zero_user_segments(page, block_start, from,
2630 : : to, block_end);
2631 : 0 : continue;
2632 : : }
2633 : 0 : if (buffer_uptodate(bh))
2634 : 0 : continue; /* reiserfs does this */
2635 : 0 : if (block_start < from || block_end > to) {
2636 : 0 : lock_buffer(bh);
2637 : 0 : bh->b_end_io = end_buffer_read_nobh;
2638 : : submit_bh(REQ_OP_READ, 0, bh);
2639 : 0 : nr_reads++;
2640 : : }
2641 : : }
2642 : :
2643 : 0 : if (nr_reads) {
2644 : : /*
2645 : : * The page is locked, so these buffers are protected from
2646 : : * any VM or truncate activity. Hence we don't need to care
2647 : : * for the buffer_head refcounts.
2648 : : */
2649 : 0 : for (bh = head; bh; bh = bh->b_this_page) {
2650 : 0 : wait_on_buffer(bh);
2651 : 0 : if (!buffer_uptodate(bh))
2652 : : ret = -EIO;
2653 : : }
2654 : 0 : if (ret)
2655 : : goto failed;
2656 : : }
2657 : :
2658 : 0 : if (is_mapped_to_disk)
2659 : : SetPageMappedToDisk(page);
2660 : :
2661 : 0 : *fsdata = head; /* to be released by nobh_write_end */
2662 : :
2663 : 0 : return 0;
2664 : :
2665 : : failed:
2666 : 0 : BUG_ON(!ret);
2667 : : /*
2668 : : * Error recovery is a bit difficult. We need to zero out blocks that
2669 : : * were newly allocated, and dirty them to ensure they get written out.
2670 : : * Buffers need to be attached to the page at this point, otherwise
2671 : : * the handling of potential IO errors during writeout would be hard
2672 : : * (could try doing synchronous writeout, but what if that fails too?)
2673 : : */
2674 : 0 : attach_nobh_buffers(page, head);
2675 : 0 : page_zero_new_buffers(page, from, to);
2676 : :
2677 : : out_release:
2678 : 0 : unlock_page(page);
2679 : 0 : put_page(page);
2680 : 0 : *pagep = NULL;
2681 : :
2682 : 0 : return ret;
2683 : : }
2684 : : EXPORT_SYMBOL(nobh_write_begin);
2685 : :
2686 : 0 : int nobh_write_end(struct file *file, struct address_space *mapping,
2687 : : loff_t pos, unsigned len, unsigned copied,
2688 : : struct page *page, void *fsdata)
2689 : : {
2690 : 0 : struct inode *inode = page->mapping->host;
2691 : : struct buffer_head *head = fsdata;
2692 : : struct buffer_head *bh;
2693 : 0 : BUG_ON(fsdata != NULL && page_has_buffers(page));
2694 : :
2695 : 0 : if (unlikely(copied < len) && head)
2696 : 0 : attach_nobh_buffers(page, head);
2697 : 0 : if (page_has_buffers(page))
2698 : 0 : return generic_write_end(file, mapping, pos, len,
2699 : : copied, page, fsdata);
2700 : :
2701 : : SetPageUptodate(page);
2702 : 0 : set_page_dirty(page);
2703 : 0 : if (pos+copied > inode->i_size) {
2704 : : i_size_write(inode, pos+copied);
2705 : : mark_inode_dirty(inode);
2706 : : }
2707 : :
2708 : 0 : unlock_page(page);
2709 : 0 : put_page(page);
2710 : :
2711 : 0 : while (head) {
2712 : : bh = head;
2713 : 0 : head = head->b_this_page;
2714 : 0 : free_buffer_head(bh);
2715 : : }
2716 : :
2717 : 0 : return copied;
2718 : : }
2719 : : EXPORT_SYMBOL(nobh_write_end);
2720 : :
2721 : : /*
2722 : : * nobh_writepage() - based on block_full_write_page() except
2723 : : * that it tries to operate without attaching bufferheads to
2724 : : * the page.
2725 : : */
2726 : 0 : int nobh_writepage(struct page *page, get_block_t *get_block,
2727 : : struct writeback_control *wbc)
2728 : : {
2729 : 0 : struct inode * const inode = page->mapping->host;
2730 : : loff_t i_size = i_size_read(inode);
2731 : 0 : const pgoff_t end_index = i_size >> PAGE_SHIFT;
2732 : : unsigned offset;
2733 : : int ret;
2734 : :
2735 : : /* Is the page fully inside i_size? */
2736 : 0 : if (page->index < end_index)
2737 : : goto out;
2738 : :
2739 : : /* Is the page fully outside i_size? (truncate in progress) */
2740 : 0 : offset = i_size & (PAGE_SIZE-1);
2741 : 0 : if (page->index >= end_index+1 || !offset) {
2742 : : /*
2743 : : * The page may have dirty, unmapped buffers. For example,
2744 : : * they may have been added in ext3_writepage(). Make them
2745 : : * freeable here, so the page does not leak.
2746 : : */
2747 : : #if 0
2748 : : /* Not really sure about this - do we need this ? */
2749 : : if (page->mapping->a_ops->invalidatepage)
2750 : : page->mapping->a_ops->invalidatepage(page, offset);
2751 : : #endif
2752 : 0 : unlock_page(page);
2753 : 0 : return 0; /* don't care */
2754 : : }
2755 : :
2756 : : /*
2757 : : * The page straddles i_size. It must be zeroed out on each and every
2758 : : * writepage invocation because it may be mmapped. "A file is mapped
2759 : : * in multiples of the page size. For a file that is not a multiple of
2760 : : * the page size, the remaining memory is zeroed when mapped, and
2761 : : * writes to that region are not written out to the file."
2762 : : */
2763 : : zero_user_segment(page, offset, PAGE_SIZE);
2764 : : out:
2765 : 0 : ret = mpage_writepage(page, get_block, wbc);
2766 : 0 : if (ret == -EAGAIN)
2767 : 0 : ret = __block_write_full_page(inode, page, get_block, wbc,
2768 : : end_buffer_async_write);
2769 : 0 : return ret;
2770 : : }
2771 : : EXPORT_SYMBOL(nobh_writepage);
2772 : :
2773 : 0 : int nobh_truncate_page(struct address_space *mapping,
2774 : : loff_t from, get_block_t *get_block)
2775 : : {
2776 : 0 : pgoff_t index = from >> PAGE_SHIFT;
2777 : 0 : unsigned offset = from & (PAGE_SIZE-1);
2778 : : unsigned blocksize;
2779 : : sector_t iblock;
2780 : : unsigned length, pos;
2781 : 0 : struct inode *inode = mapping->host;
2782 : : struct page *page;
2783 : : struct buffer_head map_bh;
2784 : : int err;
2785 : :
2786 : : blocksize = i_blocksize(inode);
2787 : 0 : length = offset & (blocksize - 1);
2788 : :
2789 : : /* Block boundary? Nothing to do */
2790 : 0 : if (!length)
2791 : : return 0;
2792 : :
2793 : 0 : length = blocksize - length;
2794 : 0 : iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits);
2795 : :
2796 : : page = grab_cache_page(mapping, index);
2797 : : err = -ENOMEM;
2798 : 0 : if (!page)
2799 : : goto out;
2800 : :
2801 : 0 : if (page_has_buffers(page)) {
2802 : : has_buffers:
2803 : 0 : unlock_page(page);
2804 : 0 : put_page(page);
2805 : 0 : return block_truncate_page(mapping, from, get_block);
2806 : : }
2807 : :
2808 : : /* Find the buffer that contains "offset" */
2809 : : pos = blocksize;
2810 : 0 : while (offset >= pos) {
2811 : 0 : iblock++;
2812 : 0 : pos += blocksize;
2813 : : }
2814 : :
2815 : 0 : map_bh.b_size = blocksize;
2816 : 0 : map_bh.b_state = 0;
2817 : 0 : err = get_block(inode, iblock, &map_bh, 0);
2818 : 0 : if (err)
2819 : : goto unlock;
2820 : : /* unmapped? It's a hole - nothing to do */
2821 : 0 : if (!buffer_mapped(&map_bh))
2822 : : goto unlock;
2823 : :
2824 : : /* Ok, it's mapped. Make sure it's up-to-date */
2825 : 0 : if (!PageUptodate(page)) {
2826 : 0 : err = mapping->a_ops->readpage(NULL, page);
2827 : 0 : if (err) {
2828 : 0 : put_page(page);
2829 : 0 : goto out;
2830 : : }
2831 : 0 : lock_page(page);
2832 : 0 : if (!PageUptodate(page)) {
2833 : : err = -EIO;
2834 : : goto unlock;
2835 : : }
2836 : 0 : if (page_has_buffers(page))
2837 : : goto has_buffers;
2838 : : }
2839 : : zero_user(page, offset, length);
2840 : 0 : set_page_dirty(page);
2841 : : err = 0;
2842 : :
2843 : : unlock:
2844 : 0 : unlock_page(page);
2845 : 0 : put_page(page);
2846 : : out:
2847 : 0 : return err;
2848 : : }
2849 : : EXPORT_SYMBOL(nobh_truncate_page);
2850 : :
2851 : 0 : int block_truncate_page(struct address_space *mapping,
2852 : : loff_t from, get_block_t *get_block)
2853 : : {
2854 : 0 : pgoff_t index = from >> PAGE_SHIFT;
2855 : 0 : unsigned offset = from & (PAGE_SIZE-1);
2856 : : unsigned blocksize;
2857 : : sector_t iblock;
2858 : : unsigned length, pos;
2859 : 0 : struct inode *inode = mapping->host;
2860 : : struct page *page;
2861 : : struct buffer_head *bh;
2862 : : int err;
2863 : :
2864 : : blocksize = i_blocksize(inode);
2865 : 0 : length = offset & (blocksize - 1);
2866 : :
2867 : : /* Block boundary? Nothing to do */
2868 : 0 : if (!length)
2869 : : return 0;
2870 : :
2871 : 0 : length = blocksize - length;
2872 : 0 : iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits);
2873 : :
2874 : : page = grab_cache_page(mapping, index);
2875 : : err = -ENOMEM;
2876 : 0 : if (!page)
2877 : : goto out;
2878 : :
2879 : 0 : if (!page_has_buffers(page))
2880 : 0 : create_empty_buffers(page, blocksize, 0);
2881 : :
2882 : : /* Find the buffer that contains "offset" */
2883 : 0 : bh = page_buffers(page);
2884 : : pos = blocksize;
2885 : 0 : while (offset >= pos) {
2886 : 0 : bh = bh->b_this_page;
2887 : 0 : iblock++;
2888 : 0 : pos += blocksize;
2889 : : }
2890 : :
2891 : : err = 0;
2892 : 0 : if (!buffer_mapped(bh)) {
2893 : 0 : WARN_ON(bh->b_size != blocksize);
2894 : 0 : err = get_block(inode, iblock, bh, 0);
2895 : 0 : if (err)
2896 : : goto unlock;
2897 : : /* unmapped? It's a hole - nothing to do */
2898 : 0 : if (!buffer_mapped(bh))
2899 : : goto unlock;
2900 : : }
2901 : :
2902 : : /* Ok, it's mapped. Make sure it's up-to-date */
2903 : 0 : if (PageUptodate(page))
2904 : 0 : set_buffer_uptodate(bh);
2905 : :
2906 : 0 : if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) {
2907 : : err = -EIO;
2908 : 0 : ll_rw_block(REQ_OP_READ, 0, 1, &bh);
2909 : 0 : wait_on_buffer(bh);
2910 : : /* Uhhuh. Read error. Complain and punt. */
2911 : 0 : if (!buffer_uptodate(bh))
2912 : : goto unlock;
2913 : : }
2914 : :
2915 : : zero_user(page, offset, length);
2916 : 0 : mark_buffer_dirty(bh);
2917 : : err = 0;
2918 : :
2919 : : unlock:
2920 : 0 : unlock_page(page);
2921 : 0 : put_page(page);
2922 : : out:
2923 : 0 : return err;
2924 : : }
2925 : : EXPORT_SYMBOL(block_truncate_page);
2926 : :
2927 : : /*
2928 : : * The generic ->writepage function for buffer-backed address_spaces
2929 : : */
2930 : 3 : int block_write_full_page(struct page *page, get_block_t *get_block,
2931 : : struct writeback_control *wbc)
2932 : : {
2933 : 3 : struct inode * const inode = page->mapping->host;
2934 : : loff_t i_size = i_size_read(inode);
2935 : 3 : const pgoff_t end_index = i_size >> PAGE_SHIFT;
2936 : : unsigned offset;
2937 : :
2938 : : /* Is the page fully inside i_size? */
2939 : 3 : if (page->index < end_index)
2940 : 3 : return __block_write_full_page(inode, page, get_block, wbc,
2941 : : end_buffer_async_write);
2942 : :
2943 : : /* Is the page fully outside i_size? (truncate in progress) */
2944 : 0 : offset = i_size & (PAGE_SIZE-1);
2945 : 0 : if (page->index >= end_index+1 || !offset) {
2946 : : /*
2947 : : * The page may have dirty, unmapped buffers. For example,
2948 : : * they may have been added in ext3_writepage(). Make them
2949 : : * freeable here, so the page does not leak.
2950 : : */
2951 : 0 : do_invalidatepage(page, 0, PAGE_SIZE);
2952 : 0 : unlock_page(page);
2953 : 0 : return 0; /* don't care */
2954 : : }
2955 : :
2956 : : /*
2957 : : * The page straddles i_size. It must be zeroed out on each and every
2958 : : * writepage invocation because it may be mmapped. "A file is mapped
2959 : : * in multiples of the page size. For a file that is not a multiple of
2960 : : * the page size, the remaining memory is zeroed when mapped, and
2961 : : * writes to that region are not written out to the file."
2962 : : */
2963 : : zero_user_segment(page, offset, PAGE_SIZE);
2964 : 0 : return __block_write_full_page(inode, page, get_block, wbc,
2965 : : end_buffer_async_write);
2966 : : }
2967 : : EXPORT_SYMBOL(block_write_full_page);
2968 : :
2969 : 3 : sector_t generic_block_bmap(struct address_space *mapping, sector_t block,
2970 : : get_block_t *get_block)
2971 : : {
2972 : 3 : struct inode *inode = mapping->host;
2973 : 3 : struct buffer_head tmp = {
2974 : : .b_size = i_blocksize(inode),
2975 : : };
2976 : :
2977 : 3 : get_block(inode, block, &tmp, 0);
2978 : 3 : return tmp.b_blocknr;
2979 : : }
2980 : : EXPORT_SYMBOL(generic_block_bmap);
2981 : :
2982 : 3 : static void end_bio_bh_io_sync(struct bio *bio)
2983 : : {
2984 : 3 : struct buffer_head *bh = bio->bi_private;
2985 : :
2986 : 3 : if (unlikely(bio_flagged(bio, BIO_QUIET)))
2987 : 0 : set_bit(BH_Quiet, &bh->b_state);
2988 : :
2989 : 3 : bh->b_end_io(bh, !bio->bi_status);
2990 : 3 : bio_put(bio);
2991 : 3 : }
2992 : :
2993 : : /*
2994 : : * This allows us to do IO even on the odd last sectors
2995 : : * of a device, even if the block size is some multiple
2996 : : * of the physical sector size.
2997 : : *
2998 : : * We'll just truncate the bio to the size of the device,
2999 : : * and clear the end of the buffer head manually.
3000 : : *
3001 : : * Truly out-of-range accesses will turn into actual IO
3002 : : * errors, this only handles the "we need to be able to
3003 : : * do IO at the final sector" case.
3004 : : */
3005 : 3 : void guard_bio_eod(struct bio *bio)
3006 : : {
3007 : : sector_t maxsector;
3008 : : struct hd_struct *part;
3009 : :
3010 : : rcu_read_lock();
3011 : 3 : part = __disk_get_part(bio->bi_disk, bio->bi_partno);
3012 : 3 : if (part)
3013 : : maxsector = part_nr_sects_read(part);
3014 : : else
3015 : 0 : maxsector = get_capacity(bio->bi_disk);
3016 : : rcu_read_unlock();
3017 : :
3018 : 3 : if (!maxsector)
3019 : : return;
3020 : :
3021 : : /*
3022 : : * If the *whole* IO is past the end of the device,
3023 : : * let it through, and the IO layer will turn it into
3024 : : * an EIO.
3025 : : */
3026 : 3 : if (unlikely(bio->bi_iter.bi_sector >= maxsector))
3027 : : return;
3028 : :
3029 : 3 : maxsector -= bio->bi_iter.bi_sector;
3030 : 3 : if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
3031 : : return;
3032 : :
3033 : 0 : bio_truncate(bio, maxsector << 9);
3034 : : }
3035 : :
3036 : 3 : static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh,
3037 : : enum rw_hint write_hint, struct writeback_control *wbc)
3038 : : {
3039 : : struct bio *bio;
3040 : :
3041 : 3 : BUG_ON(!buffer_locked(bh));
3042 : 3 : BUG_ON(!buffer_mapped(bh));
3043 : 3 : BUG_ON(!bh->b_end_io);
3044 : 3 : BUG_ON(buffer_delay(bh));
3045 : 3 : BUG_ON(buffer_unwritten(bh));
3046 : :
3047 : : /*
3048 : : * Only clear out a write error when rewriting
3049 : : */
3050 : 3 : if (test_set_buffer_req(bh) && (op == REQ_OP_WRITE))
3051 : : clear_buffer_write_io_error(bh);
3052 : :
3053 : : /*
3054 : : * from here on down, it's all bio -- do the initial mapping,
3055 : : * submit_bio -> generic_make_request may further map this bio around
3056 : : */
3057 : : bio = bio_alloc(GFP_NOIO, 1);
3058 : :
3059 : 3 : bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
3060 : 3 : bio_set_dev(bio, bh->b_bdev);
3061 : 3 : bio->bi_write_hint = write_hint;
3062 : :
3063 : 3 : bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
3064 : 3 : BUG_ON(bio->bi_iter.bi_size != bh->b_size);
3065 : :
3066 : 3 : bio->bi_end_io = end_bio_bh_io_sync;
3067 : 3 : bio->bi_private = bh;
3068 : :
3069 : 3 : if (buffer_meta(bh))
3070 : 3 : op_flags |= REQ_META;
3071 : 3 : if (buffer_prio(bh))
3072 : 3 : op_flags |= REQ_PRIO;
3073 : 3 : bio_set_op_attrs(bio, op, op_flags);
3074 : :
3075 : : /* Take care of bh's that straddle the end of the device */
3076 : 3 : guard_bio_eod(bio);
3077 : :
3078 : 3 : if (wbc) {
3079 : : wbc_init_bio(wbc, bio);
3080 : 3 : wbc_account_cgroup_owner(wbc, bh->b_page, bh->b_size);
3081 : : }
3082 : :
3083 : 3 : submit_bio(bio);
3084 : 3 : return 0;
3085 : : }
3086 : :
3087 : 3 : int submit_bh(int op, int op_flags, struct buffer_head *bh)
3088 : : {
3089 : 3 : return submit_bh_wbc(op, op_flags, bh, 0, NULL);
3090 : : }
3091 : : EXPORT_SYMBOL(submit_bh);
3092 : :
3093 : : /**
3094 : : * ll_rw_block: low-level access to block devices (DEPRECATED)
3095 : : * @op: whether to %READ or %WRITE
3096 : : * @op_flags: req_flag_bits
3097 : : * @nr: number of &struct buffer_heads in the array
3098 : : * @bhs: array of pointers to &struct buffer_head
3099 : : *
3100 : : * ll_rw_block() takes an array of pointers to &struct buffer_heads, and
3101 : : * requests an I/O operation on them, either a %REQ_OP_READ or a %REQ_OP_WRITE.
3102 : : * @op_flags contains flags modifying the detailed I/O behavior, most notably
3103 : : * %REQ_RAHEAD.
3104 : : *
3105 : : * This function drops any buffer that it cannot get a lock on (with the
3106 : : * BH_Lock state bit), any buffer that appears to be clean when doing a write
3107 : : * request, and any buffer that appears to be up-to-date when doing read
3108 : : * request. Further it marks as clean buffers that are processed for
3109 : : * writing (the buffer cache won't assume that they are actually clean
3110 : : * until the buffer gets unlocked).
3111 : : *
3112 : : * ll_rw_block sets b_end_io to simple completion handler that marks
3113 : : * the buffer up-to-date (if appropriate), unlocks the buffer and wakes
3114 : : * any waiters.
3115 : : *
3116 : : * All of the buffers must be for the same device, and must also be a
3117 : : * multiple of the current approved size for the device.
3118 : : */
3119 : 3 : void ll_rw_block(int op, int op_flags, int nr, struct buffer_head *bhs[])
3120 : : {
3121 : : int i;
3122 : :
3123 : 3 : for (i = 0; i < nr; i++) {
3124 : 3 : struct buffer_head *bh = bhs[i];
3125 : :
3126 : 3 : if (!trylock_buffer(bh))
3127 : 3 : continue;
3128 : 3 : if (op == WRITE) {
3129 : 0 : if (test_clear_buffer_dirty(bh)) {
3130 : 0 : bh->b_end_io = end_buffer_write_sync;
3131 : : get_bh(bh);
3132 : : submit_bh(op, op_flags, bh);
3133 : 0 : continue;
3134 : : }
3135 : : } else {
3136 : 3 : if (!buffer_uptodate(bh)) {
3137 : 3 : bh->b_end_io = end_buffer_read_sync;
3138 : : get_bh(bh);
3139 : : submit_bh(op, op_flags, bh);
3140 : 3 : continue;
3141 : : }
3142 : : }
3143 : 3 : unlock_buffer(bh);
3144 : : }
3145 : 3 : }
3146 : : EXPORT_SYMBOL(ll_rw_block);
3147 : :
3148 : 3 : void write_dirty_buffer(struct buffer_head *bh, int op_flags)
3149 : : {
3150 : 3 : lock_buffer(bh);
3151 : 3 : if (!test_clear_buffer_dirty(bh)) {
3152 : 0 : unlock_buffer(bh);
3153 : 3 : return;
3154 : : }
3155 : 3 : bh->b_end_io = end_buffer_write_sync;
3156 : : get_bh(bh);
3157 : : submit_bh(REQ_OP_WRITE, op_flags, bh);
3158 : : }
3159 : : EXPORT_SYMBOL(write_dirty_buffer);
3160 : :
3161 : : /*
3162 : : * For a data-integrity writeout, we need to wait upon any in-progress I/O
3163 : : * and then start new I/O and then wait upon it. The caller must have a ref on
3164 : : * the buffer_head.
3165 : : */
3166 : 3 : int __sync_dirty_buffer(struct buffer_head *bh, int op_flags)
3167 : : {
3168 : : int ret = 0;
3169 : :
3170 : 3 : WARN_ON(atomic_read(&bh->b_count) < 1);
3171 : 3 : lock_buffer(bh);
3172 : 3 : if (test_clear_buffer_dirty(bh)) {
3173 : : get_bh(bh);
3174 : 3 : bh->b_end_io = end_buffer_write_sync;
3175 : : ret = submit_bh(REQ_OP_WRITE, op_flags, bh);
3176 : 3 : wait_on_buffer(bh);
3177 : 3 : if (!ret && !buffer_uptodate(bh))
3178 : : ret = -EIO;
3179 : : } else {
3180 : 0 : unlock_buffer(bh);
3181 : : }
3182 : 3 : return ret;
3183 : : }
3184 : : EXPORT_SYMBOL(__sync_dirty_buffer);
3185 : :
3186 : 3 : int sync_dirty_buffer(struct buffer_head *bh)
3187 : : {
3188 : 3 : return __sync_dirty_buffer(bh, REQ_SYNC);
3189 : : }
3190 : : EXPORT_SYMBOL(sync_dirty_buffer);
3191 : :
3192 : : /*
3193 : : * try_to_free_buffers() checks if all the buffers on this particular page
3194 : : * are unused, and releases them if so.
3195 : : *
3196 : : * Exclusion against try_to_free_buffers may be obtained by either
3197 : : * locking the page or by holding its mapping's private_lock.
3198 : : *
3199 : : * If the page is dirty but all the buffers are clean then we need to
3200 : : * be sure to mark the page clean as well. This is because the page
3201 : : * may be against a block device, and a later reattachment of buffers
3202 : : * to a dirty page will set *all* buffers dirty. Which would corrupt
3203 : : * filesystem data on the same device.
3204 : : *
3205 : : * The same applies to regular filesystem pages: if all the buffers are
3206 : : * clean then we set the page clean and proceed. To do that, we require
3207 : : * total exclusion from __set_page_dirty_buffers(). That is obtained with
3208 : : * private_lock.
3209 : : *
3210 : : * try_to_free_buffers() is non-blocking.
3211 : : */
3212 : : static inline int buffer_busy(struct buffer_head *bh)
3213 : : {
3214 : 3 : return atomic_read(&bh->b_count) |
3215 : 3 : (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock)));
3216 : : }
3217 : :
3218 : : static int
3219 : 3 : drop_buffers(struct page *page, struct buffer_head **buffers_to_free)
3220 : : {
3221 : 3 : struct buffer_head *head = page_buffers(page);
3222 : : struct buffer_head *bh;
3223 : :
3224 : : bh = head;
3225 : : do {
3226 : 3 : if (buffer_busy(bh))
3227 : : goto failed;
3228 : 3 : bh = bh->b_this_page;
3229 : 3 : } while (bh != head);
3230 : :
3231 : : do {
3232 : 3 : struct buffer_head *next = bh->b_this_page;
3233 : :
3234 : 3 : if (bh->b_assoc_map)
3235 : 0 : __remove_assoc_queue(bh);
3236 : : bh = next;
3237 : 3 : } while (bh != head);
3238 : 3 : *buffers_to_free = head;
3239 : 3 : __clear_page_buffers(page);
3240 : 3 : return 1;
3241 : : failed:
3242 : : return 0;
3243 : : }
3244 : :
3245 : 3 : int try_to_free_buffers(struct page *page)
3246 : : {
3247 : 3 : struct address_space * const mapping = page->mapping;
3248 : 3 : struct buffer_head *buffers_to_free = NULL;
3249 : : int ret = 0;
3250 : :
3251 : 3 : BUG_ON(!PageLocked(page));
3252 : 3 : if (PageWriteback(page))
3253 : : return 0;
3254 : :
3255 : 3 : if (mapping == NULL) { /* can this still happen? */
3256 : 0 : ret = drop_buffers(page, &buffers_to_free);
3257 : 0 : goto out;
3258 : : }
3259 : :
3260 : : spin_lock(&mapping->private_lock);
3261 : 3 : ret = drop_buffers(page, &buffers_to_free);
3262 : :
3263 : : /*
3264 : : * If the filesystem writes its buffers by hand (eg ext3)
3265 : : * then we can have clean buffers against a dirty page. We
3266 : : * clean the page here; otherwise the VM will never notice
3267 : : * that the filesystem did any IO at all.
3268 : : *
3269 : : * Also, during truncate, discard_buffer will have marked all
3270 : : * the page's buffers clean. We discover that here and clean
3271 : : * the page also.
3272 : : *
3273 : : * private_lock must be held over this entire operation in order
3274 : : * to synchronise against __set_page_dirty_buffers and prevent the
3275 : : * dirty bit from being lost.
3276 : : */
3277 : 3 : if (ret)
3278 : 3 : cancel_dirty_page(page);
3279 : : spin_unlock(&mapping->private_lock);
3280 : : out:
3281 : 3 : if (buffers_to_free) {
3282 : : struct buffer_head *bh = buffers_to_free;
3283 : :
3284 : : do {
3285 : 3 : struct buffer_head *next = bh->b_this_page;
3286 : 3 : free_buffer_head(bh);
3287 : : bh = next;
3288 : 3 : } while (bh != buffers_to_free);
3289 : : }
3290 : 3 : return ret;
3291 : : }
3292 : : EXPORT_SYMBOL(try_to_free_buffers);
3293 : :
3294 : : /*
3295 : : * There are no bdflush tunables left. But distributions are
3296 : : * still running obsolete flush daemons, so we terminate them here.
3297 : : *
3298 : : * Use of bdflush() is deprecated and will be removed in a future kernel.
3299 : : * The `flush-X' kernel threads fully replace bdflush daemons and this call.
3300 : : */
3301 : 0 : SYSCALL_DEFINE2(bdflush, int, func, long, data)
3302 : : {
3303 : : static int msg_count;
3304 : :
3305 : 0 : if (!capable(CAP_SYS_ADMIN))
3306 : : return -EPERM;
3307 : :
3308 : 0 : if (msg_count < 5) {
3309 : 0 : msg_count++;
3310 : 0 : printk(KERN_INFO
3311 : : "warning: process `%s' used the obsolete bdflush"
3312 : 0 : " system call\n", current->comm);
3313 : 0 : printk(KERN_INFO "Fix your initscripts?\n");
3314 : : }
3315 : :
3316 : 0 : if (func == 1)
3317 : 0 : do_exit(0);
3318 : : return 0;
3319 : : }
3320 : :
3321 : : /*
3322 : : * Buffer-head allocation
3323 : : */
3324 : : static struct kmem_cache *bh_cachep __read_mostly;
3325 : :
3326 : : /*
3327 : : * Once the number of bh's in the machine exceeds this level, we start
3328 : : * stripping them in writeback.
3329 : : */
3330 : : static unsigned long max_buffer_heads;
3331 : :
3332 : : int buffer_heads_over_limit;
3333 : :
3334 : : struct bh_accounting {
3335 : : int nr; /* Number of live bh's */
3336 : : int ratelimit; /* Limit cacheline bouncing */
3337 : : };
3338 : :
3339 : : static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0};
3340 : :
3341 : 3 : static void recalc_bh_state(void)
3342 : : {
3343 : : int i;
3344 : : int tot = 0;
3345 : :
3346 : 3 : if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096)
3347 : 3 : return;
3348 : 3 : __this_cpu_write(bh_accounting.ratelimit, 0);
3349 : 3 : for_each_online_cpu(i)
3350 : 3 : tot += per_cpu(bh_accounting, i).nr;
3351 : 3 : buffer_heads_over_limit = (tot > max_buffer_heads);
3352 : : }
3353 : :
3354 : 3 : struct buffer_head *alloc_buffer_head(gfp_t gfp_flags)
3355 : : {
3356 : 3 : struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags);
3357 : 3 : if (ret) {
3358 : 3 : INIT_LIST_HEAD(&ret->b_assoc_buffers);
3359 : 3 : preempt_disable();
3360 : 3 : __this_cpu_inc(bh_accounting.nr);
3361 : 3 : recalc_bh_state();
3362 : 3 : preempt_enable();
3363 : : }
3364 : 3 : return ret;
3365 : : }
3366 : : EXPORT_SYMBOL(alloc_buffer_head);
3367 : :
3368 : 3 : void free_buffer_head(struct buffer_head *bh)
3369 : : {
3370 : 3 : BUG_ON(!list_empty(&bh->b_assoc_buffers));
3371 : 3 : kmem_cache_free(bh_cachep, bh);
3372 : 3 : preempt_disable();
3373 : 3 : __this_cpu_dec(bh_accounting.nr);
3374 : 3 : recalc_bh_state();
3375 : 3 : preempt_enable();
3376 : 3 : }
3377 : : EXPORT_SYMBOL(free_buffer_head);
3378 : :
3379 : 0 : static int buffer_exit_cpu_dead(unsigned int cpu)
3380 : : {
3381 : : int i;
3382 : 0 : struct bh_lru *b = &per_cpu(bh_lrus, cpu);
3383 : :
3384 : 0 : for (i = 0; i < BH_LRU_SIZE; i++) {
3385 : 0 : brelse(b->bhs[i]);
3386 : 0 : b->bhs[i] = NULL;
3387 : : }
3388 : 0 : this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr);
3389 : 0 : per_cpu(bh_accounting, cpu).nr = 0;
3390 : 0 : return 0;
3391 : : }
3392 : :
3393 : : /**
3394 : : * bh_uptodate_or_lock - Test whether the buffer is uptodate
3395 : : * @bh: struct buffer_head
3396 : : *
3397 : : * Return true if the buffer is up-to-date and false,
3398 : : * with the buffer locked, if not.
3399 : : */
3400 : 3 : int bh_uptodate_or_lock(struct buffer_head *bh)
3401 : : {
3402 : 3 : if (!buffer_uptodate(bh)) {
3403 : 3 : lock_buffer(bh);
3404 : 3 : if (!buffer_uptodate(bh))
3405 : : return 0;
3406 : 3 : unlock_buffer(bh);
3407 : : }
3408 : : return 1;
3409 : : }
3410 : : EXPORT_SYMBOL(bh_uptodate_or_lock);
3411 : :
3412 : : /**
3413 : : * bh_submit_read - Submit a locked buffer for reading
3414 : : * @bh: struct buffer_head
3415 : : *
3416 : : * Returns zero on success and -EIO on error.
3417 : : */
3418 : 3 : int bh_submit_read(struct buffer_head *bh)
3419 : : {
3420 : 3 : BUG_ON(!buffer_locked(bh));
3421 : :
3422 : 3 : if (buffer_uptodate(bh)) {
3423 : 0 : unlock_buffer(bh);
3424 : 0 : return 0;
3425 : : }
3426 : :
3427 : : get_bh(bh);
3428 : 3 : bh->b_end_io = end_buffer_read_sync;
3429 : : submit_bh(REQ_OP_READ, 0, bh);
3430 : 3 : wait_on_buffer(bh);
3431 : 3 : if (buffer_uptodate(bh))
3432 : : return 0;
3433 : 0 : return -EIO;
3434 : : }
3435 : : EXPORT_SYMBOL(bh_submit_read);
3436 : :
3437 : 3 : void __init buffer_init(void)
3438 : : {
3439 : : unsigned long nrpages;
3440 : : int ret;
3441 : :
3442 : 3 : bh_cachep = kmem_cache_create("buffer_head",
3443 : : sizeof(struct buffer_head), 0,
3444 : : (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
3445 : : SLAB_MEM_SPREAD),
3446 : : NULL);
3447 : :
3448 : : /*
3449 : : * Limit the bh occupancy to 10% of ZONE_NORMAL
3450 : : */
3451 : 3 : nrpages = (nr_free_buffer_pages() * 10) / 100;
3452 : 3 : max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head));
3453 : : ret = cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD, "fs/buffer:dead",
3454 : : NULL, buffer_exit_cpu_dead);
3455 : 3 : WARN_ON(ret < 0);
3456 : 3 : }
|
