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 : : #include <linux/fscrypt.h>
51 : :
52 : : #include "internal.h"
53 : :
54 : : static int fsync_buffers_list(spinlock_t *lock, struct list_head *list);
55 : : static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh,
56 : : enum rw_hint hint, struct writeback_control *wbc);
57 : :
58 : : #define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers)
59 : :
60 : 264124 : inline void touch_buffer(struct buffer_head *bh)
61 : : {
62 : 0 : trace_block_touch_buffer(bh);
63 : 264124 : mark_page_accessed(bh->b_page);
64 : 264124 : }
65 : : EXPORT_SYMBOL(touch_buffer);
66 : :
67 : 68 : void __lock_buffer(struct buffer_head *bh)
68 : : {
69 : 68 : wait_on_bit_lock_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE);
70 : 66 : }
71 : : EXPORT_SYMBOL(__lock_buffer);
72 : :
73 : 37925 : void unlock_buffer(struct buffer_head *bh)
74 : : {
75 : 18799 : clear_bit_unlock(BH_Lock, &bh->b_state);
76 : 37925 : smp_mb__after_atomic();
77 : 37799 : wake_up_bit(&bh->b_state, BH_Lock);
78 : 240 : }
79 : : EXPORT_SYMBOL(unlock_buffer);
80 : :
81 : : /*
82 : : * Returns if the page has dirty or writeback buffers. If all the buffers
83 : : * are unlocked and clean then the PageDirty information is stale. If
84 : : * any of the pages are locked, it is assumed they are locked for IO.
85 : : */
86 : 0 : void buffer_check_dirty_writeback(struct page *page,
87 : : bool *dirty, bool *writeback)
88 : : {
89 : 0 : struct buffer_head *head, *bh;
90 : 0 : *dirty = false;
91 : 0 : *writeback = false;
92 : :
93 [ # # # # ]: 0 : BUG_ON(!PageLocked(page));
94 : :
95 [ # # ]: 0 : if (!page_has_buffers(page))
96 : : return;
97 : :
98 [ # # # # ]: 0 : if (PageWriteback(page))
99 : 0 : *writeback = true;
100 : :
101 [ # # ]: 0 : head = page_buffers(page);
102 : 0 : bh = head;
103 : 0 : do {
104 [ # # ]: 0 : if (buffer_locked(bh))
105 : 0 : *writeback = true;
106 : :
107 [ # # ]: 0 : if (buffer_dirty(bh))
108 : 0 : *dirty = true;
109 : :
110 : 0 : bh = bh->b_this_page;
111 [ # # ]: 0 : } while (bh != head);
112 : : }
113 : : EXPORT_SYMBOL(buffer_check_dirty_writeback);
114 : :
115 : : /*
116 : : * Block until a buffer comes unlocked. This doesn't stop it
117 : : * from becoming locked again - you have to lock it yourself
118 : : * if you want to preserve its state.
119 : : */
120 : 3795 : void __wait_on_buffer(struct buffer_head * bh)
121 : : {
122 : 3795 : wait_on_bit_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE);
123 : 219 : }
124 : : EXPORT_SYMBOL(__wait_on_buffer);
125 : :
126 : : static void
127 : 120 : __clear_page_buffers(struct page *page)
128 : : {
129 : 120 : ClearPagePrivate(page);
130 : 120 : set_page_private(page, 0);
131 : 120 : put_page(page);
132 : : }
133 : :
134 : 0 : static void buffer_io_error(struct buffer_head *bh, char *msg)
135 : : {
136 [ # # ]: 0 : if (!test_bit(BH_Quiet, &bh->b_state))
137 [ # # ]: 0 : printk_ratelimited(KERN_ERR
138 : : "Buffer I/O error on dev %pg, logical block %llu%s\n",
139 : : bh->b_bdev, (unsigned long long)bh->b_blocknr, msg);
140 : 0 : }
141 : :
142 : : /*
143 : : * End-of-IO handler helper function which does not touch the bh after
144 : : * unlocking it.
145 : : * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but
146 : : * a race there is benign: unlock_buffer() only use the bh's address for
147 : : * hashing after unlocking the buffer, so it doesn't actually touch the bh
148 : : * itself.
149 : : */
150 : 18129 : static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate)
151 : : {
152 [ + - ]: 18129 : if (uptodate) {
153 : 18129 : set_buffer_uptodate(bh);
154 : : } else {
155 : : /* This happens, due to failed read-ahead attempts. */
156 : 0 : clear_buffer_uptodate(bh);
157 : : }
158 : 18129 : unlock_buffer(bh);
159 : 18129 : }
160 : :
161 : : /*
162 : : * Default synchronous end-of-IO handler.. Just mark it up-to-date and
163 : : * unlock the buffer. This is what ll_rw_block uses too.
164 : : */
165 : 18129 : void end_buffer_read_sync(struct buffer_head *bh, int uptodate)
166 : : {
167 : 18129 : __end_buffer_read_notouch(bh, uptodate);
168 : 18129 : put_bh(bh);
169 : 18129 : }
170 : : EXPORT_SYMBOL(end_buffer_read_sync);
171 : :
172 : 91 : void end_buffer_write_sync(struct buffer_head *bh, int uptodate)
173 : : {
174 [ + - ]: 91 : if (uptodate) {
175 : 91 : set_buffer_uptodate(bh);
176 : : } else {
177 : 0 : buffer_io_error(bh, ", lost sync page write");
178 : 0 : mark_buffer_write_io_error(bh);
179 : 0 : clear_buffer_uptodate(bh);
180 : : }
181 : 91 : unlock_buffer(bh);
182 : 91 : put_bh(bh);
183 : 91 : }
184 : : EXPORT_SYMBOL(end_buffer_write_sync);
185 : :
186 : : /*
187 : : * Various filesystems appear to want __find_get_block to be non-blocking.
188 : : * But it's the page lock which protects the buffers. To get around this,
189 : : * we get exclusion from try_to_free_buffers with the blockdev mapping's
190 : : * private_lock.
191 : : *
192 : : * Hack idea: for the blockdev mapping, private_lock contention
193 : : * may be quite high. This code could TryLock the page, and if that
194 : : * succeeds, there is no need to take private_lock.
195 : : */
196 : : static struct buffer_head *
197 : 100864 : __find_get_block_slow(struct block_device *bdev, sector_t block)
198 : : {
199 : 100864 : struct inode *bd_inode = bdev->bd_inode;
200 : 100864 : struct address_space *bd_mapping = bd_inode->i_mapping;
201 : 100864 : struct buffer_head *ret = NULL;
202 : 100864 : pgoff_t index;
203 : 100864 : struct buffer_head *bh;
204 : 100864 : struct buffer_head *head;
205 : 100864 : struct page *page;
206 : 100864 : int all_mapped = 1;
207 : 100864 : static DEFINE_RATELIMIT_STATE(last_warned, HZ, 1);
208 : :
209 : 100864 : index = block >> (PAGE_SHIFT - bd_inode->i_blkbits);
210 : 100864 : page = find_get_page_flags(bd_mapping, index, FGP_ACCESSED);
211 [ + + ]: 100864 : if (!page)
212 : 54550 : goto out;
213 : :
214 : 46314 : spin_lock(&bd_mapping->private_lock);
215 [ - + ]: 46314 : if (!page_has_buffers(page))
216 : 0 : goto out_unlock;
217 [ - + ]: 46314 : head = page_buffers(page);
218 : 46314 : bh = head;
219 : 46434 : do {
220 [ + - ]: 46434 : if (!buffer_mapped(bh))
221 : : all_mapped = 0;
222 [ + + ]: 46434 : else if (bh->b_blocknr == block) {
223 : 46314 : ret = bh;
224 : 46314 : get_bh(bh);
225 : 46314 : goto out_unlock;
226 : : }
227 : 120 : bh = bh->b_this_page;
228 [ + - ]: 120 : } while (bh != head);
229 : :
230 : : /* we might be here because some of the buffers on this page are
231 : : * not mapped. This is due to various races between
232 : : * file io on the block device and getblk. It gets dealt with
233 : : * elsewhere, don't buffer_error if we had some unmapped buffers
234 : : */
235 [ # # ]: 0 : ratelimit_set_flags(&last_warned, RATELIMIT_MSG_ON_RELEASE);
236 [ # # # # ]: 0 : if (all_mapped && __ratelimit(&last_warned)) {
237 : 0 : printk("__find_get_block_slow() failed. block=%llu, "
238 : : "b_blocknr=%llu, b_state=0x%08lx, b_size=%zu, "
239 : : "device %pg blocksize: %d\n",
240 : : (unsigned long long)block,
241 : 0 : (unsigned long long)bh->b_blocknr,
242 : : bh->b_state, bh->b_size, bdev,
243 : 0 : 1 << bd_inode->i_blkbits);
244 : : }
245 : 0 : out_unlock:
246 : 46314 : spin_unlock(&bd_mapping->private_lock);
247 : 46314 : put_page(page);
248 : 100864 : out:
249 : 100864 : return ret;
250 : : }
251 : :
252 : 60 : static void end_buffer_async_read(struct buffer_head *bh, int uptodate)
253 : : {
254 : 60 : unsigned long flags;
255 : 60 : struct buffer_head *first;
256 : 60 : struct buffer_head *tmp;
257 : 60 : struct page *page;
258 : 60 : int page_uptodate = 1;
259 : :
260 [ - + ]: 60 : BUG_ON(!buffer_async_read(bh));
261 : :
262 : 60 : page = bh->b_page;
263 [ + - ]: 60 : if (uptodate) {
264 : 60 : set_buffer_uptodate(bh);
265 : : } else {
266 : 0 : clear_buffer_uptodate(bh);
267 : 0 : buffer_io_error(bh, ", async page read");
268 [ # # ]: 0 : SetPageError(page);
269 : : }
270 : :
271 : : /*
272 : : * Be _very_ careful from here on. Bad things can happen if
273 : : * two buffer heads end IO at almost the same time and both
274 : : * decide that the page is now completely done.
275 : : */
276 [ - + ]: 60 : first = page_buffers(page);
277 : 60 : local_irq_save(flags);
278 : 60 : bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
279 : 60 : clear_buffer_async_read(bh);
280 : 60 : unlock_buffer(bh);
281 : 60 : tmp = bh;
282 : 60 : do {
283 [ - + ]: 60 : if (!buffer_uptodate(tmp))
284 : 0 : page_uptodate = 0;
285 [ - + ]: 60 : if (buffer_async_read(tmp)) {
286 [ # # ]: 0 : BUG_ON(!buffer_locked(tmp));
287 : 0 : goto still_busy;
288 : : }
289 : 60 : tmp = tmp->b_this_page;
290 [ - + ]: 60 : } while (tmp != bh);
291 : 60 : bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
292 : 60 : local_irq_restore(flags);
293 : :
294 : : /*
295 : : * If none of the buffers had errors and they are all
296 : : * uptodate then we can set the page uptodate.
297 : : */
298 [ + - + - ]: 120 : if (page_uptodate && !PageError(page))
299 : 60 : SetPageUptodate(page);
300 : 60 : unlock_page(page);
301 : 60 : return;
302 : :
303 : : still_busy:
304 : 0 : bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
305 : 0 : local_irq_restore(flags);
306 : : return;
307 : : }
308 : :
309 : : struct decrypt_bh_ctx {
310 : : struct work_struct work;
311 : : struct buffer_head *bh;
312 : : };
313 : :
314 : : static void decrypt_bh(struct work_struct *work)
315 : : {
316 : : struct decrypt_bh_ctx *ctx =
317 : : container_of(work, struct decrypt_bh_ctx, work);
318 : : struct buffer_head *bh = ctx->bh;
319 : : int err;
320 : :
321 : : err = fscrypt_decrypt_pagecache_blocks(bh->b_page, bh->b_size,
322 : : bh_offset(bh));
323 : : end_buffer_async_read(bh, err == 0);
324 : : kfree(ctx);
325 : : }
326 : :
327 : : /*
328 : : * I/O completion handler for block_read_full_page() - pages
329 : : * which come unlocked at the end of I/O.
330 : : */
331 : 60 : static void end_buffer_async_read_io(struct buffer_head *bh, int uptodate)
332 : : {
333 : : /* Decrypt if needed */
334 : 60 : if (uptodate && IS_ENABLED(CONFIG_FS_ENCRYPTION) &&
335 : : IS_ENCRYPTED(bh->b_page->mapping->host) &&
336 : : S_ISREG(bh->b_page->mapping->host->i_mode)) {
337 : : struct decrypt_bh_ctx *ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
338 : :
339 : : if (ctx) {
340 : : INIT_WORK(&ctx->work, decrypt_bh);
341 : : ctx->bh = bh;
342 : : fscrypt_enqueue_decrypt_work(&ctx->work);
343 : : return;
344 : : }
345 : : uptodate = 0;
346 : : }
347 : 60 : end_buffer_async_read(bh, uptodate);
348 : : }
349 : :
350 : : /*
351 : : * Completion handler for block_write_full_page() - pages which are unlocked
352 : : * during I/O, and which have PageWriteback cleared upon I/O completion.
353 : : */
354 : 0 : void end_buffer_async_write(struct buffer_head *bh, int uptodate)
355 : : {
356 : 0 : unsigned long flags;
357 : 0 : struct buffer_head *first;
358 : 0 : struct buffer_head *tmp;
359 : 0 : struct page *page;
360 : :
361 [ # # ]: 0 : BUG_ON(!buffer_async_write(bh));
362 : :
363 : 0 : page = bh->b_page;
364 [ # # ]: 0 : if (uptodate) {
365 : 0 : set_buffer_uptodate(bh);
366 : : } else {
367 : 0 : buffer_io_error(bh, ", lost async page write");
368 : 0 : mark_buffer_write_io_error(bh);
369 : 0 : clear_buffer_uptodate(bh);
370 [ # # ]: 0 : SetPageError(page);
371 : : }
372 : :
373 [ # # ]: 0 : first = page_buffers(page);
374 : 0 : local_irq_save(flags);
375 : 0 : bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
376 : :
377 : 0 : clear_buffer_async_write(bh);
378 : 0 : unlock_buffer(bh);
379 : 0 : tmp = bh->b_this_page;
380 [ # # ]: 0 : while (tmp != bh) {
381 [ # # ]: 0 : if (buffer_async_write(tmp)) {
382 [ # # ]: 0 : BUG_ON(!buffer_locked(tmp));
383 : 0 : goto still_busy;
384 : : }
385 : 0 : tmp = tmp->b_this_page;
386 : : }
387 : 0 : bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
388 : 0 : local_irq_restore(flags);
389 : 0 : end_page_writeback(page);
390 : 0 : return;
391 : :
392 : : still_busy:
393 : 0 : bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
394 : 0 : local_irq_restore(flags);
395 : : return;
396 : : }
397 : : EXPORT_SYMBOL(end_buffer_async_write);
398 : :
399 : : /*
400 : : * If a page's buffers are under async readin (end_buffer_async_read
401 : : * completion) then there is a possibility that another thread of
402 : : * control could lock one of the buffers after it has completed
403 : : * but while some of the other buffers have not completed. This
404 : : * locked buffer would confuse end_buffer_async_read() into not unlocking
405 : : * the page. So the absence of BH_Async_Read tells end_buffer_async_read()
406 : : * that this buffer is not under async I/O.
407 : : *
408 : : * The page comes unlocked when it has no locked buffer_async buffers
409 : : * left.
410 : : *
411 : : * PageLocked prevents anyone starting new async I/O reads any of
412 : : * the buffers.
413 : : *
414 : : * PageWriteback is used to prevent simultaneous writeout of the same
415 : : * page.
416 : : *
417 : : * PageLocked prevents anyone from starting writeback of a page which is
418 : : * under read I/O (PageWriteback is only ever set against a locked page).
419 : : */
420 : 60 : static void mark_buffer_async_read(struct buffer_head *bh)
421 : : {
422 : 60 : bh->b_end_io = end_buffer_async_read_io;
423 : 60 : set_buffer_async_read(bh);
424 : 60 : }
425 : :
426 : 0 : static void mark_buffer_async_write_endio(struct buffer_head *bh,
427 : : bh_end_io_t *handler)
428 : : {
429 : 0 : bh->b_end_io = handler;
430 : 0 : set_buffer_async_write(bh);
431 : 0 : }
432 : :
433 : 0 : void mark_buffer_async_write(struct buffer_head *bh)
434 : : {
435 : 0 : mark_buffer_async_write_endio(bh, end_buffer_async_write);
436 : 0 : }
437 : : EXPORT_SYMBOL(mark_buffer_async_write);
438 : :
439 : :
440 : : /*
441 : : * fs/buffer.c contains helper functions for buffer-backed address space's
442 : : * fsync functions. A common requirement for buffer-based filesystems is
443 : : * that certain data from the backing blockdev needs to be written out for
444 : : * a successful fsync(). For example, ext2 indirect blocks need to be
445 : : * written back and waited upon before fsync() returns.
446 : : *
447 : : * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(),
448 : : * inode_has_buffers() and invalidate_inode_buffers() are provided for the
449 : : * management of a list of dependent buffers at ->i_mapping->private_list.
450 : : *
451 : : * Locking is a little subtle: try_to_free_buffers() will remove buffers
452 : : * from their controlling inode's queue when they are being freed. But
453 : : * try_to_free_buffers() will be operating against the *blockdev* mapping
454 : : * at the time, not against the S_ISREG file which depends on those buffers.
455 : : * So the locking for private_list is via the private_lock in the address_space
456 : : * which backs the buffers. Which is different from the address_space
457 : : * against which the buffers are listed. So for a particular address_space,
458 : : * mapping->private_lock does *not* protect mapping->private_list! In fact,
459 : : * mapping->private_list will always be protected by the backing blockdev's
460 : : * ->private_lock.
461 : : *
462 : : * Which introduces a requirement: all buffers on an address_space's
463 : : * ->private_list must be from the same address_space: the blockdev's.
464 : : *
465 : : * address_spaces which do not place buffers at ->private_list via these
466 : : * utility functions are free to use private_lock and private_list for
467 : : * whatever they want. The only requirement is that list_empty(private_list)
468 : : * be true at clear_inode() time.
469 : : *
470 : : * FIXME: clear_inode should not call invalidate_inode_buffers(). The
471 : : * filesystems should do that. invalidate_inode_buffers() should just go
472 : : * BUG_ON(!list_empty).
473 : : *
474 : : * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should
475 : : * take an address_space, not an inode. And it should be called
476 : : * mark_buffer_dirty_fsync() to clearly define why those buffers are being
477 : : * queued up.
478 : : *
479 : : * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the
480 : : * list if it is already on a list. Because if the buffer is on a list,
481 : : * it *must* already be on the right one. If not, the filesystem is being
482 : : * silly. This will save a ton of locking. But first we have to ensure
483 : : * that buffers are taken *off* the old inode's list when they are freed
484 : : * (presumably in truncate). That requires careful auditing of all
485 : : * filesystems (do it inside bforget()). It could also be done by bringing
486 : : * b_inode back.
487 : : */
488 : :
489 : : /*
490 : : * The buffer's backing address_space's private_lock must be held
491 : : */
492 : 0 : static void __remove_assoc_queue(struct buffer_head *bh)
493 : : {
494 : 0 : list_del_init(&bh->b_assoc_buffers);
495 [ # # # # : 0 : WARN_ON(!bh->b_assoc_map);
# # # # #
# ]
496 : 0 : bh->b_assoc_map = NULL;
497 : 0 : }
498 : :
499 : 54657 : int inode_has_buffers(struct inode *inode)
500 : : {
501 : 54657 : return !list_empty(&inode->i_data.private_list);
502 : : }
503 : :
504 : : /*
505 : : * osync is designed to support O_SYNC io. It waits synchronously for
506 : : * all already-submitted IO to complete, but does not queue any new
507 : : * writes to the disk.
508 : : *
509 : : * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as
510 : : * you dirty the buffers, and then use osync_inode_buffers to wait for
511 : : * completion. Any other dirty buffers which are not yet queued for
512 : : * write will not be flushed to disk by the osync.
513 : : */
514 : 0 : static int osync_buffers_list(spinlock_t *lock, struct list_head *list)
515 : : {
516 : 0 : struct buffer_head *bh;
517 : 0 : struct list_head *p;
518 : 0 : int err = 0;
519 : :
520 : 0 : spin_lock(lock);
521 : 0 : repeat:
522 [ # # ]: 0 : list_for_each_prev(p, list) {
523 : 0 : bh = BH_ENTRY(p);
524 [ # # ]: 0 : if (buffer_locked(bh)) {
525 : 0 : get_bh(bh);
526 : 0 : spin_unlock(lock);
527 : 0 : wait_on_buffer(bh);
528 [ # # ]: 0 : if (!buffer_uptodate(bh))
529 : 0 : err = -EIO;
530 [ # # ]: 0 : brelse(bh);
531 : 0 : spin_lock(lock);
532 : 0 : goto repeat;
533 : : }
534 : : }
535 : 0 : spin_unlock(lock);
536 : 0 : return err;
537 : : }
538 : :
539 : 0 : void emergency_thaw_bdev(struct super_block *sb)
540 : : {
541 [ # # # # ]: 0 : while (sb->s_bdev && !thaw_bdev(sb->s_bdev, sb))
542 : 0 : printk(KERN_WARNING "Emergency Thaw on %pg\n", sb->s_bdev);
543 : 0 : }
544 : :
545 : : /**
546 : : * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers
547 : : * @mapping: the mapping which wants those buffers written
548 : : *
549 : : * Starts I/O against the buffers at mapping->private_list, and waits upon
550 : : * that I/O.
551 : : *
552 : : * Basically, this is a convenience function for fsync().
553 : : * @mapping is a file or directory which needs those buffers to be written for
554 : : * a successful fsync().
555 : : */
556 : 0 : int sync_mapping_buffers(struct address_space *mapping)
557 : : {
558 : 0 : struct address_space *buffer_mapping = mapping->private_data;
559 : :
560 [ # # # # ]: 0 : if (buffer_mapping == NULL || list_empty(&mapping->private_list))
561 : : return 0;
562 : :
563 : 0 : return fsync_buffers_list(&buffer_mapping->private_lock,
564 : : &mapping->private_list);
565 : : }
566 : : EXPORT_SYMBOL(sync_mapping_buffers);
567 : :
568 : : /*
569 : : * Called when we've recently written block `bblock', and it is known that
570 : : * `bblock' was for a buffer_boundary() buffer. This means that the block at
571 : : * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's
572 : : * dirty, schedule it for IO. So that indirects merge nicely with their data.
573 : : */
574 : 0 : void write_boundary_block(struct block_device *bdev,
575 : : sector_t bblock, unsigned blocksize)
576 : : {
577 : 0 : struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize);
578 [ # # ]: 0 : if (bh) {
579 [ # # ]: 0 : if (buffer_dirty(bh))
580 : 0 : ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
581 : 0 : put_bh(bh);
582 : : }
583 : 0 : }
584 : :
585 : 0 : void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode)
586 : : {
587 : 0 : struct address_space *mapping = inode->i_mapping;
588 : 0 : struct address_space *buffer_mapping = bh->b_page->mapping;
589 : :
590 : 0 : mark_buffer_dirty(bh);
591 [ # # ]: 0 : if (!mapping->private_data) {
592 : 0 : mapping->private_data = buffer_mapping;
593 : : } else {
594 [ # # ]: 0 : BUG_ON(mapping->private_data != buffer_mapping);
595 : : }
596 [ # # ]: 0 : if (!bh->b_assoc_map) {
597 : 0 : spin_lock(&buffer_mapping->private_lock);
598 : 0 : list_move_tail(&bh->b_assoc_buffers,
599 : : &mapping->private_list);
600 : 0 : bh->b_assoc_map = mapping;
601 : 0 : spin_unlock(&buffer_mapping->private_lock);
602 : : }
603 : 0 : }
604 : : EXPORT_SYMBOL(mark_buffer_dirty_inode);
605 : :
606 : : /*
607 : : * Mark the page dirty, and set it dirty in the page cache, and mark the inode
608 : : * dirty.
609 : : *
610 : : * If warn is true, then emit a warning if the page is not uptodate and has
611 : : * not been truncated.
612 : : *
613 : : * The caller must hold lock_page_memcg().
614 : : */
615 : 19323 : void __set_page_dirty(struct page *page, struct address_space *mapping,
616 : : int warn)
617 : : {
618 : 19323 : unsigned long flags;
619 : :
620 : 19323 : xa_lock_irqsave(&mapping->i_pages, flags);
621 [ + - ]: 19323 : if (page->mapping) { /* Race with truncate? */
622 [ - + - - : 19323 : WARN_ON_ONCE(warn && !PageUptodate(page));
- + ]
623 : 19323 : account_page_dirtied(page, mapping);
624 : 19323 : __xa_set_mark(&mapping->i_pages, page_index(page),
625 : : PAGECACHE_TAG_DIRTY);
626 : : }
627 : 19323 : xa_unlock_irqrestore(&mapping->i_pages, flags);
628 : 19323 : }
629 : : EXPORT_SYMBOL_GPL(__set_page_dirty);
630 : :
631 : : /*
632 : : * Add a page to the dirty page list.
633 : : *
634 : : * It is a sad fact of life that this function is called from several places
635 : : * deeply under spinlocking. It may not sleep.
636 : : *
637 : : * If the page has buffers, the uptodate buffers are set dirty, to preserve
638 : : * dirty-state coherency between the page and the buffers. It the page does
639 : : * not have buffers then when they are later attached they will all be set
640 : : * dirty.
641 : : *
642 : : * The buffers are dirtied before the page is dirtied. There's a small race
643 : : * window in which a writepage caller may see the page cleanness but not the
644 : : * buffer dirtiness. That's fine. If this code were to set the page dirty
645 : : * before the buffers, a concurrent writepage caller could clear the page dirty
646 : : * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean
647 : : * page on the dirty page list.
648 : : *
649 : : * We use private_lock to lock against try_to_free_buffers while using the
650 : : * page's buffer list. Also use this to protect against clean buffers being
651 : : * added to the page after it was set dirty.
652 : : *
653 : : * FIXME: may need to call ->reservepage here as well. That's rather up to the
654 : : * address_space though.
655 : : */
656 : 0 : int __set_page_dirty_buffers(struct page *page)
657 : : {
658 : 0 : int newly_dirty;
659 : 0 : struct address_space *mapping = page_mapping(page);
660 : :
661 [ # # ]: 0 : if (unlikely(!mapping))
662 [ # # ]: 0 : return !TestSetPageDirty(page);
663 : :
664 : 0 : spin_lock(&mapping->private_lock);
665 [ # # ]: 0 : if (page_has_buffers(page)) {
666 [ # # ]: 0 : struct buffer_head *head = page_buffers(page);
667 : 0 : struct buffer_head *bh = head;
668 : :
669 : 0 : do {
670 : 0 : set_buffer_dirty(bh);
671 : 0 : bh = bh->b_this_page;
672 [ # # ]: 0 : } while (bh != head);
673 : : }
674 : : /*
675 : : * Lock out page->mem_cgroup migration to keep PageDirty
676 : : * synchronized with per-memcg dirty page counters.
677 : : */
678 [ # # ]: 0 : lock_page_memcg(page);
679 [ # # ]: 0 : newly_dirty = !TestSetPageDirty(page);
680 : 0 : spin_unlock(&mapping->private_lock);
681 : :
682 [ # # ]: 0 : if (newly_dirty)
683 : 0 : __set_page_dirty(page, mapping, 1);
684 : :
685 [ # # ]: 0 : unlock_page_memcg(page);
686 : :
687 [ # # ]: 0 : if (newly_dirty)
688 : 0 : __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
689 : :
690 : : return newly_dirty;
691 : : }
692 : : EXPORT_SYMBOL(__set_page_dirty_buffers);
693 : :
694 : : /*
695 : : * Write out and wait upon a list of buffers.
696 : : *
697 : : * We have conflicting pressures: we want to make sure that all
698 : : * initially dirty buffers get waited on, but that any subsequently
699 : : * dirtied buffers don't. After all, we don't want fsync to last
700 : : * forever if somebody is actively writing to the file.
701 : : *
702 : : * Do this in two main stages: first we copy dirty buffers to a
703 : : * temporary inode list, queueing the writes as we go. Then we clean
704 : : * up, waiting for those writes to complete.
705 : : *
706 : : * During this second stage, any subsequent updates to the file may end
707 : : * up refiling the buffer on the original inode's dirty list again, so
708 : : * there is a chance we will end up with a buffer queued for write but
709 : : * not yet completed on that list. So, as a final cleanup we go through
710 : : * the osync code to catch these locked, dirty buffers without requeuing
711 : : * any newly dirty buffers for write.
712 : : */
713 : 0 : static int fsync_buffers_list(spinlock_t *lock, struct list_head *list)
714 : : {
715 : 0 : struct buffer_head *bh;
716 : 0 : struct list_head tmp;
717 : 0 : struct address_space *mapping;
718 : 0 : int err = 0, err2;
719 : 0 : struct blk_plug plug;
720 : :
721 : 0 : INIT_LIST_HEAD(&tmp);
722 : 0 : blk_start_plug(&plug);
723 : :
724 : 0 : spin_lock(lock);
725 [ # # ]: 0 : while (!list_empty(list)) {
726 : 0 : bh = BH_ENTRY(list->next);
727 : 0 : mapping = bh->b_assoc_map;
728 [ # # ]: 0 : __remove_assoc_queue(bh);
729 : : /* Avoid race with mark_buffer_dirty_inode() which does
730 : : * a lockless check and we rely on seeing the dirty bit */
731 : 0 : smp_mb();
732 [ # # # # ]: 0 : if (buffer_dirty(bh) || buffer_locked(bh)) {
733 : 0 : list_add(&bh->b_assoc_buffers, &tmp);
734 : 0 : bh->b_assoc_map = mapping;
735 [ # # ]: 0 : if (buffer_dirty(bh)) {
736 : 0 : get_bh(bh);
737 : 0 : spin_unlock(lock);
738 : : /*
739 : : * Ensure any pending I/O completes so that
740 : : * write_dirty_buffer() actually writes the
741 : : * current contents - it is a noop if I/O is
742 : : * still in flight on potentially older
743 : : * contents.
744 : : */
745 : 0 : write_dirty_buffer(bh, REQ_SYNC);
746 : :
747 : : /*
748 : : * Kick off IO for the previous mapping. Note
749 : : * that we will not run the very last mapping,
750 : : * wait_on_buffer() will do that for us
751 : : * through sync_buffer().
752 : : */
753 [ # # ]: 0 : brelse(bh);
754 : 0 : spin_lock(lock);
755 : : }
756 : : }
757 : : }
758 : :
759 : 0 : spin_unlock(lock);
760 : 0 : blk_finish_plug(&plug);
761 : 0 : spin_lock(lock);
762 : :
763 [ # # ]: 0 : while (!list_empty(&tmp)) {
764 : 0 : bh = BH_ENTRY(tmp.prev);
765 : 0 : get_bh(bh);
766 : 0 : mapping = bh->b_assoc_map;
767 [ # # ]: 0 : __remove_assoc_queue(bh);
768 : : /* Avoid race with mark_buffer_dirty_inode() which does
769 : : * a lockless check and we rely on seeing the dirty bit */
770 : 0 : smp_mb();
771 [ # # ]: 0 : if (buffer_dirty(bh)) {
772 : 0 : list_add(&bh->b_assoc_buffers,
773 : : &mapping->private_list);
774 : 0 : bh->b_assoc_map = mapping;
775 : : }
776 : 0 : spin_unlock(lock);
777 : 0 : wait_on_buffer(bh);
778 [ # # ]: 0 : if (!buffer_uptodate(bh))
779 : 0 : err = -EIO;
780 [ # # ]: 0 : brelse(bh);
781 : 0 : spin_lock(lock);
782 : : }
783 : :
784 : 0 : spin_unlock(lock);
785 : 0 : err2 = osync_buffers_list(lock, list);
786 [ # # ]: 0 : if (err)
787 : : return err;
788 : : else
789 : 0 : return err2;
790 : : }
791 : :
792 : : /*
793 : : * Invalidate any and all dirty buffers on a given inode. We are
794 : : * probably unmounting the fs, but that doesn't mean we have already
795 : : * done a sync(). Just drop the buffers from the inode list.
796 : : *
797 : : * NOTE: we take the inode's blockdev's mapping's private_lock. Which
798 : : * assumes that all the buffers are against the blockdev. Not true
799 : : * for reiserfs.
800 : : */
801 : 360 : void invalidate_inode_buffers(struct inode *inode)
802 : : {
803 [ - + ]: 360 : if (inode_has_buffers(inode)) {
804 : 0 : struct address_space *mapping = &inode->i_data;
805 : 0 : struct list_head *list = &mapping->private_list;
806 : 0 : struct address_space *buffer_mapping = mapping->private_data;
807 : :
808 : 0 : spin_lock(&buffer_mapping->private_lock);
809 [ # # ]: 0 : while (!list_empty(list))
810 [ # # ]: 0 : __remove_assoc_queue(BH_ENTRY(list->next));
811 : 0 : spin_unlock(&buffer_mapping->private_lock);
812 : : }
813 : 360 : }
814 : : EXPORT_SYMBOL(invalidate_inode_buffers);
815 : :
816 : : /*
817 : : * Remove any clean buffers from the inode's buffer list. This is called
818 : : * when we're trying to free the inode itself. Those buffers can pin it.
819 : : *
820 : : * Returns true if all buffers were removed.
821 : : */
822 : 0 : int remove_inode_buffers(struct inode *inode)
823 : : {
824 : 0 : int ret = 1;
825 : :
826 [ # # ]: 0 : if (inode_has_buffers(inode)) {
827 : 0 : struct address_space *mapping = &inode->i_data;
828 : 0 : struct list_head *list = &mapping->private_list;
829 : 0 : struct address_space *buffer_mapping = mapping->private_data;
830 : :
831 : 0 : spin_lock(&buffer_mapping->private_lock);
832 [ # # ]: 0 : while (!list_empty(list)) {
833 : 0 : struct buffer_head *bh = BH_ENTRY(list->next);
834 [ # # ]: 0 : if (buffer_dirty(bh)) {
835 : : ret = 0;
836 : : break;
837 : : }
838 [ # # ]: 0 : __remove_assoc_queue(bh);
839 : : }
840 : 0 : spin_unlock(&buffer_mapping->private_lock);
841 : : }
842 : 0 : return ret;
843 : : }
844 : :
845 : : /*
846 : : * Create the appropriate buffers when given a page for data area and
847 : : * the size of each buffer.. Use the bh->b_this_page linked list to
848 : : * follow the buffers created. Return NULL if unable to create more
849 : : * buffers.
850 : : *
851 : : * The retry flag is used to differentiate async IO (paging, swapping)
852 : : * which may not fail from ordinary buffer allocations.
853 : : */
854 : 45071 : struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size,
855 : : bool retry)
856 : : {
857 : 45071 : struct buffer_head *bh, *head;
858 : 45071 : gfp_t gfp = GFP_NOFS | __GFP_ACCOUNT;
859 : 45071 : long offset;
860 : 45071 : struct mem_cgroup *memcg;
861 : :
862 [ + - ]: 45071 : if (retry)
863 : 45071 : gfp |= __GFP_NOFAIL;
864 : :
865 : 45071 : memcg = get_mem_cgroup_from_page(page);
866 : 45071 : memalloc_use_memcg(memcg);
867 : :
868 : 45071 : head = NULL;
869 : 45071 : offset = PAGE_SIZE;
870 [ + + ]: 90502 : while ((offset -= size) >= 0) {
871 : 45431 : bh = alloc_buffer_head(gfp);
872 [ - + ]: 45431 : if (!bh)
873 : 0 : goto no_grow;
874 : :
875 : 45431 : bh->b_this_page = head;
876 : 45431 : bh->b_blocknr = -1;
877 : 45431 : head = bh;
878 : :
879 : 45431 : bh->b_size = size;
880 : :
881 : : /* Link the buffer to its page */
882 : 45431 : set_bh_page(bh, page, offset);
883 : : }
884 : 45071 : out:
885 : 45071 : memalloc_unuse_memcg();
886 : 45071 : mem_cgroup_put(memcg);
887 : 45071 : return head;
888 : : /*
889 : : * In case anything failed, we just free everything we got.
890 : : */
891 : : no_grow:
892 [ # # ]: 0 : if (head) {
893 : 0 : do {
894 : 0 : bh = head;
895 : 0 : head = head->b_this_page;
896 : 0 : free_buffer_head(bh);
897 [ # # ]: 0 : } while (head);
898 : : }
899 : :
900 : 0 : goto out;
901 : : }
902 : : EXPORT_SYMBOL_GPL(alloc_page_buffers);
903 : :
904 : : static inline void
905 : : link_dev_buffers(struct page *page, struct buffer_head *head)
906 : : {
907 : : struct buffer_head *bh, *tail;
908 : :
909 : : bh = head;
910 : 27635 : do {
911 : 27635 : tail = bh;
912 : 27635 : bh = bh->b_this_page;
913 [ + + ]: 27635 : } while (bh);
914 : 27275 : tail->b_this_page = head;
915 : 27275 : attach_page_buffers(page, head);
916 : : }
917 : :
918 : 27275 : static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size)
919 : : {
920 : 27275 : sector_t retval = ~((sector_t)0);
921 : 27275 : loff_t sz = i_size_read(bdev->bd_inode);
922 : :
923 [ + - ]: 27275 : if (sz) {
924 : : unsigned int sizebits = blksize_bits(size);
925 : 27275 : retval = (sz >> sizebits);
926 : : }
927 : 27275 : return retval;
928 : : }
929 : :
930 : : /*
931 : : * Initialise the state of a blockdev page's buffers.
932 : : */
933 : : static sector_t
934 : 27275 : init_page_buffers(struct page *page, struct block_device *bdev,
935 : : sector_t block, int size)
936 : : {
937 [ - + ]: 27275 : struct buffer_head *head = page_buffers(page);
938 : 27275 : struct buffer_head *bh = head;
939 : 27275 : int uptodate = PageUptodate(page);
940 [ + - ]: 27275 : sector_t end_block = blkdev_max_block(I_BDEV(bdev->bd_inode), size);
941 : :
942 : 27635 : do {
943 [ + - ]: 27635 : if (!buffer_mapped(bh)) {
944 : 27635 : bh->b_end_io = NULL;
945 : 27635 : bh->b_private = NULL;
946 : 27635 : bh->b_bdev = bdev;
947 : 27635 : bh->b_blocknr = block;
948 [ - + ]: 27635 : if (uptodate)
949 : 0 : set_buffer_uptodate(bh);
950 [ + - ]: 27635 : if (block < end_block)
951 : 27635 : set_buffer_mapped(bh);
952 : : }
953 : 27635 : block++;
954 : 27635 : bh = bh->b_this_page;
955 [ + + ]: 27635 : } while (bh != head);
956 : :
957 : : /*
958 : : * Caller needs to validate requested block against end of device.
959 : : */
960 : 27275 : return end_block;
961 : : }
962 : :
963 : : /*
964 : : * Create the page-cache page that contains the requested block.
965 : : *
966 : : * This is used purely for blockdev mappings.
967 : : */
968 : : static int
969 : 27275 : grow_dev_page(struct block_device *bdev, sector_t block,
970 : : pgoff_t index, int size, int sizebits, gfp_t gfp)
971 : : {
972 : 27275 : struct inode *inode = bdev->bd_inode;
973 : 27275 : struct page *page;
974 : 27275 : struct buffer_head *bh;
975 : 27275 : sector_t end_block;
976 : 27275 : int ret = 0; /* Will call free_more_memory() */
977 : 27275 : gfp_t gfp_mask;
978 : :
979 : 27275 : gfp_mask = mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS) | gfp;
980 : :
981 : : /*
982 : : * XXX: __getblk_slow() can not really deal with failure and
983 : : * will endlessly loop on improvised global reclaim. Prefer
984 : : * looping in the allocator rather than here, at least that
985 : : * code knows what it's doing.
986 : : */
987 : 27275 : gfp_mask |= __GFP_NOFAIL;
988 : :
989 : 27275 : page = find_or_create_page(inode->i_mapping, index, gfp_mask);
990 : :
991 [ - + - + ]: 54550 : BUG_ON(!PageLocked(page));
992 : :
993 [ - + ]: 27275 : if (page_has_buffers(page)) {
994 [ # # ]: 0 : bh = page_buffers(page);
995 [ # # ]: 0 : if (bh->b_size == size) {
996 : 0 : end_block = init_page_buffers(page, bdev,
997 : : (sector_t)index << sizebits,
998 : : size);
999 : 0 : goto done;
1000 : : }
1001 [ # # ]: 0 : if (!try_to_free_buffers(page))
1002 : 0 : goto failed;
1003 : : }
1004 : :
1005 : : /*
1006 : : * Allocate some buffers for this page
1007 : : */
1008 : 27275 : bh = alloc_page_buffers(page, size, true);
1009 : :
1010 : : /*
1011 : : * Link the page to the buffers and initialise them. Take the
1012 : : * lock to be atomic wrt __find_get_block(), which does not
1013 : : * run under the page lock.
1014 : : */
1015 : 27275 : spin_lock(&inode->i_mapping->private_lock);
1016 : : link_dev_buffers(page, bh);
1017 : 27275 : end_block = init_page_buffers(page, bdev, (sector_t)index << sizebits,
1018 : : size);
1019 : 27275 : spin_unlock(&inode->i_mapping->private_lock);
1020 : 27275 : done:
1021 [ - + ]: 27275 : ret = (block < end_block) ? 1 : -ENXIO;
1022 : 27275 : failed:
1023 : 27275 : unlock_page(page);
1024 : 27275 : put_page(page);
1025 : 27275 : return ret;
1026 : : }
1027 : :
1028 : : /*
1029 : : * Create buffers for the specified block device block's page. If
1030 : : * that page was dirty, the buffers are set dirty also.
1031 : : */
1032 : : static int
1033 : 27275 : grow_buffers(struct block_device *bdev, sector_t block, int size, gfp_t gfp)
1034 : : {
1035 : 27275 : pgoff_t index;
1036 : 27275 : int sizebits;
1037 : :
1038 : 27275 : sizebits = -1;
1039 : 27515 : do {
1040 : 27515 : sizebits++;
1041 [ + + ]: 27515 : } while ((size << sizebits) < PAGE_SIZE);
1042 : :
1043 : 27275 : index = block >> sizebits;
1044 : :
1045 : : /*
1046 : : * Check for a block which wants to lie outside our maximum possible
1047 : : * pagecache index. (this comparison is done using sector_t types).
1048 : : */
1049 : 27275 : if (unlikely(index != block >> sizebits)) {
1050 : : printk(KERN_ERR "%s: requested out-of-range block %llu for "
1051 : : "device %pg\n",
1052 : : __func__, (unsigned long long)block,
1053 : : bdev);
1054 : : return -EIO;
1055 : : }
1056 : :
1057 : : /* Create a page with the proper size buffers.. */
1058 : 27275 : return grow_dev_page(bdev, block, index, size, sizebits, gfp);
1059 : : }
1060 : :
1061 : : static struct buffer_head *
1062 : 27275 : __getblk_slow(struct block_device *bdev, sector_t block,
1063 : : unsigned size, gfp_t gfp)
1064 : : {
1065 : : /* Size must be multiple of hard sectorsize */
1066 [ + - + - : 54550 : if (unlikely(size & (bdev_logical_block_size(bdev)-1) ||
- + ]
1067 : : (size < 512 || size > PAGE_SIZE))) {
1068 : 0 : printk(KERN_ERR "getblk(): invalid block size %d requested\n",
1069 : : size);
1070 [ # # ]: 0 : printk(KERN_ERR "logical block size: %d\n",
1071 : : bdev_logical_block_size(bdev));
1072 : :
1073 : 0 : dump_stack();
1074 : 0 : return NULL;
1075 : : }
1076 : :
1077 : 54550 : for (;;) {
1078 : 54550 : struct buffer_head *bh;
1079 : 54550 : int ret;
1080 : :
1081 : 54550 : bh = __find_get_block(bdev, block, size);
1082 [ + + ]: 54550 : if (bh)
1083 : 27275 : return bh;
1084 : :
1085 : 27275 : ret = grow_buffers(bdev, block, size, gfp);
1086 [ + - ]: 27275 : if (ret < 0)
1087 : : return NULL;
1088 : : }
1089 : : }
1090 : :
1091 : : /*
1092 : : * The relationship between dirty buffers and dirty pages:
1093 : : *
1094 : : * Whenever a page has any dirty buffers, the page's dirty bit is set, and
1095 : : * the page is tagged dirty in the page cache.
1096 : : *
1097 : : * At all times, the dirtiness of the buffers represents the dirtiness of
1098 : : * subsections of the page. If the page has buffers, the page dirty bit is
1099 : : * merely a hint about the true dirty state.
1100 : : *
1101 : : * When a page is set dirty in its entirety, all its buffers are marked dirty
1102 : : * (if the page has buffers).
1103 : : *
1104 : : * When a buffer is marked dirty, its page is dirtied, but the page's other
1105 : : * buffers are not.
1106 : : *
1107 : : * Also. When blockdev buffers are explicitly read with bread(), they
1108 : : * individually become uptodate. But their backing page remains not
1109 : : * uptodate - even if all of its buffers are uptodate. A subsequent
1110 : : * block_read_full_page() against that page will discover all the uptodate
1111 : : * buffers, will set the page uptodate and will perform no I/O.
1112 : : */
1113 : :
1114 : : /**
1115 : : * mark_buffer_dirty - mark a buffer_head as needing writeout
1116 : : * @bh: the buffer_head to mark dirty
1117 : : *
1118 : : * mark_buffer_dirty() will set the dirty bit against the buffer, then set
1119 : : * its backing page dirty, then tag the page as dirty in the page cache
1120 : : * and then attach the address_space's inode to its superblock's dirty
1121 : : * inode list.
1122 : : *
1123 : : * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock,
1124 : : * i_pages lock and mapping->host->i_lock.
1125 : : */
1126 : 23916 : void mark_buffer_dirty(struct buffer_head *bh)
1127 : : {
1128 [ - + ]: 23916 : WARN_ON_ONCE(!buffer_uptodate(bh));
1129 : :
1130 : 23916 : trace_block_dirty_buffer(bh);
1131 : :
1132 : : /*
1133 : : * Very *carefully* optimize the it-is-already-dirty case.
1134 : : *
1135 : : * Don't let the final "is it dirty" escape to before we
1136 : : * perhaps modified the buffer.
1137 : : */
1138 [ + + ]: 23916 : if (buffer_dirty(bh)) {
1139 : 3166 : smp_mb();
1140 [ - + ]: 3166 : if (buffer_dirty(bh))
1141 : : return;
1142 : : }
1143 : :
1144 [ + - ]: 20750 : if (!test_set_buffer_dirty(bh)) {
1145 : 20750 : struct page *page = bh->b_page;
1146 : 20750 : struct address_space *mapping = NULL;
1147 : :
1148 [ - + ]: 20750 : lock_page_memcg(page);
1149 [ - + + + ]: 41500 : if (!TestSetPageDirty(page)) {
1150 : 19323 : mapping = page_mapping(page);
1151 [ + - ]: 19323 : if (mapping)
1152 : 19323 : __set_page_dirty(page, mapping, 0);
1153 : : }
1154 [ + - ]: 19323 : unlock_page_memcg(page);
1155 [ + - ]: 19323 : if (mapping)
1156 : 19323 : __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1157 : : }
1158 : : }
1159 : : EXPORT_SYMBOL(mark_buffer_dirty);
1160 : :
1161 : 0 : void mark_buffer_write_io_error(struct buffer_head *bh)
1162 : : {
1163 : 0 : set_buffer_write_io_error(bh);
1164 : : /* FIXME: do we need to set this in both places? */
1165 [ # # # # ]: 0 : if (bh->b_page && bh->b_page->mapping)
1166 : 0 : mapping_set_error(bh->b_page->mapping, -EIO);
1167 [ # # ]: 0 : if (bh->b_assoc_map)
1168 : 0 : mapping_set_error(bh->b_assoc_map, -EIO);
1169 : 0 : }
1170 : : EXPORT_SYMBOL(mark_buffer_write_io_error);
1171 : :
1172 : : /*
1173 : : * Decrement a buffer_head's reference count. If all buffers against a page
1174 : : * have zero reference count, are clean and unlocked, and if the page is clean
1175 : : * and unlocked then try_to_free_buffers() may strip the buffers from the page
1176 : : * in preparation for freeing it (sometimes, rarely, buffers are removed from
1177 : : * a page but it ends up not being freed, and buffers may later be reattached).
1178 : : */
1179 : 364768 : void __brelse(struct buffer_head * buf)
1180 : : {
1181 [ + - ]: 364768 : if (atomic_read(&buf->b_count)) {
1182 : 364768 : put_bh(buf);
1183 : 364768 : return;
1184 : : }
1185 : 0 : WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n");
1186 : : }
1187 : : EXPORT_SYMBOL(__brelse);
1188 : :
1189 : : /*
1190 : : * bforget() is like brelse(), except it discards any
1191 : : * potentially dirty data.
1192 : : */
1193 : 0 : void __bforget(struct buffer_head *bh)
1194 : : {
1195 : 0 : clear_buffer_dirty(bh);
1196 [ # # ]: 0 : if (bh->b_assoc_map) {
1197 : 0 : struct address_space *buffer_mapping = bh->b_page->mapping;
1198 : :
1199 : 0 : spin_lock(&buffer_mapping->private_lock);
1200 : 0 : list_del_init(&bh->b_assoc_buffers);
1201 : 0 : bh->b_assoc_map = NULL;
1202 : 0 : spin_unlock(&buffer_mapping->private_lock);
1203 : : }
1204 : 0 : __brelse(bh);
1205 : 0 : }
1206 : : EXPORT_SYMBOL(__bforget);
1207 : :
1208 : 246 : static struct buffer_head *__bread_slow(struct buffer_head *bh)
1209 : : {
1210 : 246 : lock_buffer(bh);
1211 [ + + ]: 246 : if (buffer_uptodate(bh)) {
1212 : 66 : unlock_buffer(bh);
1213 : 66 : return bh;
1214 : : } else {
1215 : 180 : get_bh(bh);
1216 : 180 : bh->b_end_io = end_buffer_read_sync;
1217 : 180 : submit_bh(REQ_OP_READ, 0, bh);
1218 : 180 : wait_on_buffer(bh);
1219 [ - + ]: 180 : if (buffer_uptodate(bh))
1220 : : return bh;
1221 : : }
1222 [ # # ]: 0 : brelse(bh);
1223 : : return NULL;
1224 : : }
1225 : :
1226 : : /*
1227 : : * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block().
1228 : : * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their
1229 : : * refcount elevated by one when they're in an LRU. A buffer can only appear
1230 : : * once in a particular CPU's LRU. A single buffer can be present in multiple
1231 : : * CPU's LRUs at the same time.
1232 : : *
1233 : : * This is a transparent caching front-end to sb_bread(), sb_getblk() and
1234 : : * sb_find_get_block().
1235 : : *
1236 : : * The LRUs themselves only need locking against invalidate_bh_lrus. We use
1237 : : * a local interrupt disable for that.
1238 : : */
1239 : :
1240 : : #define BH_LRU_SIZE 16
1241 : :
1242 : : struct bh_lru {
1243 : : struct buffer_head *bhs[BH_LRU_SIZE];
1244 : : };
1245 : :
1246 : : static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }};
1247 : :
1248 : : #ifdef CONFIG_SMP
1249 : : #define bh_lru_lock() local_irq_disable()
1250 : : #define bh_lru_unlock() local_irq_enable()
1251 : : #else
1252 : : #define bh_lru_lock() preempt_disable()
1253 : : #define bh_lru_unlock() preempt_enable()
1254 : : #endif
1255 : :
1256 : 411302 : static inline void check_irqs_on(void)
1257 : : {
1258 : : #ifdef irqs_disabled
1259 [ - + - + ]: 411302 : BUG_ON(irqs_disabled());
1260 : : #endif
1261 : : }
1262 : :
1263 : : /*
1264 : : * Install a buffer_head into this cpu's LRU. If not already in the LRU, it is
1265 : : * inserted at the front, and the buffer_head at the back if any is evicted.
1266 : : * Or, if already in the LRU it is moved to the front.
1267 : : */
1268 : 46314 : static void bh_lru_install(struct buffer_head *bh)
1269 : : {
1270 : 46314 : struct buffer_head *evictee = bh;
1271 : 46314 : struct bh_lru *b;
1272 : 46314 : int i;
1273 : :
1274 : 46314 : check_irqs_on();
1275 : 46314 : bh_lru_lock();
1276 : :
1277 : 46314 : b = this_cpu_ptr(&bh_lrus);
1278 [ + + ]: 787338 : for (i = 0; i < BH_LRU_SIZE; i++) {
1279 : 741024 : swap(evictee, b->bhs[i]);
1280 [ - + ]: 741024 : if (evictee == bh) {
1281 : 0 : bh_lru_unlock();
1282 : 0 : return;
1283 : : }
1284 : : }
1285 : :
1286 : 46314 : get_bh(bh);
1287 : 46314 : bh_lru_unlock();
1288 [ + + ]: 46314 : brelse(evictee);
1289 : : }
1290 : :
1291 : : /*
1292 : : * Look up the bh in this cpu's LRU. If it's there, move it to the head.
1293 : : */
1294 : : static struct buffer_head *
1295 : 364988 : lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size)
1296 : : {
1297 : 364988 : struct buffer_head *ret = NULL;
1298 : 364988 : unsigned int i;
1299 : :
1300 : 364988 : check_irqs_on();
1301 : 364988 : bh_lru_lock();
1302 [ + + ]: 2887060 : for (i = 0; i < BH_LRU_SIZE; i++) {
1303 [ + + ]: 2421208 : struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]);
1304 : :
1305 [ + + + + : 2421208 : if (bh && bh->b_blocknr == block && bh->b_bdev == bdev &&
+ - ]
1306 [ + - ]: 264124 : bh->b_size == size) {
1307 [ + + ]: 264124 : if (i) {
1308 [ + + ]: 722980 : while (i) {
1309 : 543260 : __this_cpu_write(bh_lrus.bhs[i],
1310 : : __this_cpu_read(bh_lrus.bhs[i - 1]));
1311 : 543260 : i--;
1312 : : }
1313 : 264124 : __this_cpu_write(bh_lrus.bhs[0], bh);
1314 : : }
1315 : 264124 : get_bh(bh);
1316 : 264124 : ret = bh;
1317 : 264124 : break;
1318 : : }
1319 : : }
1320 : 364988 : bh_lru_unlock();
1321 : 364988 : return ret;
1322 : : }
1323 : :
1324 : : /*
1325 : : * Perform a pagecache lookup for the matching buffer. If it's there, refresh
1326 : : * it in the LRU and mark it as accessed. If it is not present then return
1327 : : * NULL
1328 : : */
1329 : : struct buffer_head *
1330 : 364988 : __find_get_block(struct block_device *bdev, sector_t block, unsigned size)
1331 : : {
1332 : 364988 : struct buffer_head *bh = lookup_bh_lru(bdev, block, size);
1333 : :
1334 [ + + ]: 364988 : if (bh == NULL) {
1335 : : /* __find_get_block_slow will mark the page accessed */
1336 : 100864 : bh = __find_get_block_slow(bdev, block);
1337 [ + + ]: 100864 : if (bh)
1338 : 46314 : bh_lru_install(bh);
1339 : : } else
1340 : 264124 : touch_buffer(bh);
1341 : :
1342 : 364988 : return bh;
1343 : : }
1344 : : EXPORT_SYMBOL(__find_get_block);
1345 : :
1346 : : /*
1347 : : * __getblk_gfp() will locate (and, if necessary, create) the buffer_head
1348 : : * which corresponds to the passed block_device, block and size. The
1349 : : * returned buffer has its reference count incremented.
1350 : : *
1351 : : * __getblk_gfp() will lock up the machine if grow_dev_page's
1352 : : * try_to_free_buffers() attempt is failing. FIXME, perhaps?
1353 : : */
1354 : : struct buffer_head *
1355 : 299084 : __getblk_gfp(struct block_device *bdev, sector_t block,
1356 : : unsigned size, gfp_t gfp)
1357 : : {
1358 : 299084 : struct buffer_head *bh = __find_get_block(bdev, block, size);
1359 : :
1360 : 299084 : might_sleep();
1361 [ + + ]: 299084 : if (bh == NULL)
1362 : 27275 : bh = __getblk_slow(bdev, block, size, gfp);
1363 : 299084 : return bh;
1364 : : }
1365 : : EXPORT_SYMBOL(__getblk_gfp);
1366 : :
1367 : : /*
1368 : : * Do async read-ahead on a buffer..
1369 : : */
1370 : 14889 : void __breadahead(struct block_device *bdev, sector_t block, unsigned size)
1371 : : {
1372 : 14889 : struct buffer_head *bh = __getblk(bdev, block, size);
1373 [ + - ]: 14889 : if (likely(bh)) {
1374 : 14889 : ll_rw_block(REQ_OP_READ, REQ_RAHEAD, 1, &bh);
1375 [ + - ]: 14889 : brelse(bh);
1376 : : }
1377 : 14889 : }
1378 : : EXPORT_SYMBOL(__breadahead);
1379 : :
1380 : : /**
1381 : : * __bread_gfp() - reads a specified block and returns the bh
1382 : : * @bdev: the block_device to read from
1383 : : * @block: number of block
1384 : : * @size: size (in bytes) to read
1385 : : * @gfp: page allocation flag
1386 : : *
1387 : : * Reads a specified block, and returns buffer head that contains it.
1388 : : * The page cache can be allocated from non-movable area
1389 : : * not to prevent page migration if you set gfp to zero.
1390 : : * It returns NULL if the block was unreadable.
1391 : : */
1392 : : struct buffer_head *
1393 : 270 : __bread_gfp(struct block_device *bdev, sector_t block,
1394 : : unsigned size, gfp_t gfp)
1395 : : {
1396 : 270 : struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp);
1397 : :
1398 [ + - + + ]: 540 : if (likely(bh) && !buffer_uptodate(bh))
1399 : 246 : bh = __bread_slow(bh);
1400 : 270 : return bh;
1401 : : }
1402 : : EXPORT_SYMBOL(__bread_gfp);
1403 : :
1404 : : /*
1405 : : * invalidate_bh_lrus() is called rarely - but not only at unmount.
1406 : : * This doesn't race because it runs in each cpu either in irq
1407 : : * or with preempt disabled.
1408 : : */
1409 : 180 : static void invalidate_bh_lru(void *arg)
1410 : : {
1411 : 180 : struct bh_lru *b = &get_cpu_var(bh_lrus);
1412 : 180 : int i;
1413 : :
1414 [ + + ]: 3060 : for (i = 0; i < BH_LRU_SIZE; i++) {
1415 [ + + ]: 2880 : brelse(b->bhs[i]);
1416 : 2880 : b->bhs[i] = NULL;
1417 : : }
1418 : 180 : put_cpu_var(bh_lrus);
1419 : 180 : }
1420 : :
1421 : 240 : static bool has_bh_in_lru(int cpu, void *dummy)
1422 : : {
1423 : 240 : struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu);
1424 : 240 : int i;
1425 : :
1426 [ + + ]: 1200 : for (i = 0; i < BH_LRU_SIZE; i++) {
1427 [ + + ]: 1140 : if (b->bhs[i])
1428 : : return true;
1429 : : }
1430 : :
1431 : : return false;
1432 : : }
1433 : :
1434 : 240 : void invalidate_bh_lrus(void)
1435 : : {
1436 : 240 : on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1);
1437 : 240 : }
1438 : : EXPORT_SYMBOL_GPL(invalidate_bh_lrus);
1439 : :
1440 : 48386 : void set_bh_page(struct buffer_head *bh,
1441 : : struct page *page, unsigned long offset)
1442 : : {
1443 : 48386 : bh->b_page = page;
1444 [ - + - + ]: 48386 : BUG_ON(offset >= PAGE_SIZE);
1445 : 45431 : if (PageHighMem(page))
1446 : : /*
1447 : : * This catches illegal uses and preserves the offset:
1448 : : */
1449 : : bh->b_data = (char *)(0 + offset);
1450 : : else
1451 : 45431 : bh->b_data = page_address(page) + offset;
1452 : 45431 : }
1453 : : EXPORT_SYMBOL(set_bh_page);
1454 : :
1455 : : /*
1456 : : * Called when truncating a buffer on a page completely.
1457 : : */
1458 : :
1459 : : /* Bits that are cleared during an invalidate */
1460 : : #define BUFFER_FLAGS_DISCARD \
1461 : : (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \
1462 : : 1 << BH_Delay | 1 << BH_Unwritten)
1463 : :
1464 : 540 : static void discard_buffer(struct buffer_head * bh)
1465 : : {
1466 : 540 : unsigned long b_state, b_state_old;
1467 : :
1468 : 540 : lock_buffer(bh);
1469 : 540 : clear_buffer_dirty(bh);
1470 : 540 : bh->b_bdev = NULL;
1471 : 540 : b_state = bh->b_state;
1472 : 540 : for (;;) {
1473 : 540 : b_state_old = cmpxchg(&bh->b_state, b_state,
1474 : : (b_state & ~BUFFER_FLAGS_DISCARD));
1475 [ - + ]: 540 : if (b_state_old == b_state)
1476 : : break;
1477 : : b_state = b_state_old;
1478 : : }
1479 : 540 : unlock_buffer(bh);
1480 : 540 : }
1481 : :
1482 : : /**
1483 : : * block_invalidatepage - invalidate part or all of a buffer-backed page
1484 : : *
1485 : : * @page: the page which is affected
1486 : : * @offset: start of the range to invalidate
1487 : : * @length: length of the range to invalidate
1488 : : *
1489 : : * block_invalidatepage() is called when all or part of the page has become
1490 : : * invalidated by a truncate operation.
1491 : : *
1492 : : * block_invalidatepage() does not have to release all buffers, but it must
1493 : : * ensure that no dirty buffer is left outside @offset and that no I/O
1494 : : * is underway against any of the blocks which are outside the truncation
1495 : : * point. Because the caller is about to free (and possibly reuse) those
1496 : : * blocks on-disk.
1497 : : */
1498 : 180 : void block_invalidatepage(struct page *page, unsigned int offset,
1499 : : unsigned int length)
1500 : : {
1501 : 180 : struct buffer_head *head, *bh, *next;
1502 : 180 : unsigned int curr_off = 0;
1503 : 180 : unsigned int stop = length + offset;
1504 : :
1505 [ - + - + ]: 360 : BUG_ON(!PageLocked(page));
1506 [ - + ]: 180 : if (!page_has_buffers(page))
1507 : 0 : goto out;
1508 : :
1509 : : /*
1510 : : * Check for overflow
1511 : : */
1512 [ - + ]: 180 : BUG_ON(stop > PAGE_SIZE || stop < length);
1513 : :
1514 [ - + ]: 180 : head = page_buffers(page);
1515 : 180 : bh = head;
1516 : 540 : do {
1517 : 540 : unsigned int next_off = curr_off + bh->b_size;
1518 : 540 : next = bh->b_this_page;
1519 : :
1520 : : /*
1521 : : * Are we still fully in range ?
1522 : : */
1523 [ - + ]: 540 : if (next_off > stop)
1524 : 0 : goto out;
1525 : :
1526 : : /*
1527 : : * is this block fully invalidated?
1528 : : */
1529 [ + - ]: 540 : if (offset <= curr_off)
1530 : 540 : discard_buffer(bh);
1531 : 540 : curr_off = next_off;
1532 : 540 : bh = next;
1533 [ + + ]: 540 : } while (bh != head);
1534 : :
1535 : : /*
1536 : : * We release buffers only if the entire page is being invalidated.
1537 : : * The get_block cached value has been unconditionally invalidated,
1538 : : * so real IO is not possible anymore.
1539 : : */
1540 [ - + ]: 180 : if (length == PAGE_SIZE)
1541 : 180 : try_to_release_page(page, 0);
1542 : 0 : out:
1543 : 180 : return;
1544 : : }
1545 : : EXPORT_SYMBOL(block_invalidatepage);
1546 : :
1547 : :
1548 : : /*
1549 : : * We attach and possibly dirty the buffers atomically wrt
1550 : : * __set_page_dirty_buffers() via private_lock. try_to_free_buffers
1551 : : * is already excluded via the page lock.
1552 : : */
1553 : 17796 : void create_empty_buffers(struct page *page,
1554 : : unsigned long blocksize, unsigned long b_state)
1555 : : {
1556 : 17796 : struct buffer_head *bh, *head, *tail;
1557 : :
1558 : 17796 : head = alloc_page_buffers(page, blocksize, true);
1559 : 17796 : bh = head;
1560 : 17796 : do {
1561 : 17796 : bh->b_state |= b_state;
1562 : 17796 : tail = bh;
1563 : 17796 : bh = bh->b_this_page;
1564 [ - + ]: 17796 : } while (bh);
1565 : 17796 : tail->b_this_page = head;
1566 : :
1567 : 17796 : spin_lock(&page->mapping->private_lock);
1568 [ + + - + ]: 35562 : if (PageUptodate(page) || PageDirty(page)) {
1569 : : bh = head;
1570 : 30 : do {
1571 [ - + - + ]: 60 : if (PageDirty(page))
1572 : 0 : set_buffer_dirty(bh);
1573 [ + - ]: 30 : if (PageUptodate(page))
1574 : 30 : set_buffer_uptodate(bh);
1575 : 30 : bh = bh->b_this_page;
1576 [ - + ]: 30 : } while (bh != head);
1577 : : }
1578 : 17796 : attach_page_buffers(page, head);
1579 : 17796 : spin_unlock(&page->mapping->private_lock);
1580 : 17796 : }
1581 : : EXPORT_SYMBOL(create_empty_buffers);
1582 : :
1583 : : /**
1584 : : * clean_bdev_aliases: clean a range of buffers in block device
1585 : : * @bdev: Block device to clean buffers in
1586 : : * @block: Start of a range of blocks to clean
1587 : : * @len: Number of blocks to clean
1588 : : *
1589 : : * We are taking a range of blocks for data and we don't want writeback of any
1590 : : * buffer-cache aliases starting from return from this function and until the
1591 : : * moment when something will explicitly mark the buffer dirty (hopefully that
1592 : : * will not happen until we will free that block ;-) We don't even need to mark
1593 : : * it not-uptodate - nobody can expect anything from a newly allocated buffer
1594 : : * anyway. We used to use unmap_buffer() for such invalidation, but that was
1595 : : * wrong. We definitely don't want to mark the alias unmapped, for example - it
1596 : : * would confuse anyone who might pick it with bread() afterwards...
1597 : : *
1598 : : * Also.. Note that bforget() doesn't lock the buffer. So there can be
1599 : : * writeout I/O going on against recently-freed buffers. We don't wait on that
1600 : : * I/O in bforget() - it's more efficient to wait on the I/O only if we really
1601 : : * need to. That happens here.
1602 : : */
1603 : 17736 : void clean_bdev_aliases(struct block_device *bdev, sector_t block, sector_t len)
1604 : : {
1605 : 17736 : struct inode *bd_inode = bdev->bd_inode;
1606 : 17736 : struct address_space *bd_mapping = bd_inode->i_mapping;
1607 : 17736 : struct pagevec pvec;
1608 : 17736 : pgoff_t index = block >> (PAGE_SHIFT - bd_inode->i_blkbits);
1609 : 17736 : pgoff_t end;
1610 : 17736 : int i, count;
1611 : 17736 : struct buffer_head *bh;
1612 : 17736 : struct buffer_head *head;
1613 : :
1614 : 17736 : end = (block + len - 1) >> (PAGE_SHIFT - bd_inode->i_blkbits);
1615 : 17736 : pagevec_init(&pvec);
1616 [ - + ]: 17736 : while (pagevec_lookup_range(&pvec, bd_mapping, &index, end)) {
1617 : 0 : count = pagevec_count(&pvec);
1618 [ # # ]: 0 : for (i = 0; i < count; i++) {
1619 : 0 : struct page *page = pvec.pages[i];
1620 : :
1621 [ # # ]: 0 : if (!page_has_buffers(page))
1622 : 0 : continue;
1623 : : /*
1624 : : * We use page lock instead of bd_mapping->private_lock
1625 : : * to pin buffers here since we can afford to sleep and
1626 : : * it scales better than a global spinlock lock.
1627 : : */
1628 : 0 : lock_page(page);
1629 : : /* Recheck when the page is locked which pins bhs */
1630 [ # # ]: 0 : if (!page_has_buffers(page))
1631 : 0 : goto unlock_page;
1632 [ # # ]: 0 : head = page_buffers(page);
1633 : 0 : bh = head;
1634 : 0 : do {
1635 [ # # # # ]: 0 : if (!buffer_mapped(bh) || (bh->b_blocknr < block))
1636 : 0 : goto next;
1637 [ # # ]: 0 : if (bh->b_blocknr >= block + len)
1638 : : break;
1639 : 0 : clear_buffer_dirty(bh);
1640 : 0 : wait_on_buffer(bh);
1641 : 0 : clear_buffer_req(bh);
1642 : 0 : next:
1643 : 0 : bh = bh->b_this_page;
1644 [ # # ]: 0 : } while (bh != head);
1645 : 0 : unlock_page:
1646 : 0 : unlock_page(page);
1647 : : }
1648 [ # # ]: 0 : pagevec_release(&pvec);
1649 : 0 : cond_resched();
1650 : : /* End of range already reached? */
1651 [ # # # # ]: 0 : if (index > end || !index)
1652 : : break;
1653 : : }
1654 : 17736 : }
1655 : : EXPORT_SYMBOL(clean_bdev_aliases);
1656 : :
1657 : : /*
1658 : : * Size is a power-of-two in the range 512..PAGE_SIZE,
1659 : : * and the case we care about most is PAGE_SIZE.
1660 : : *
1661 : : * So this *could* possibly be written with those
1662 : : * constraints in mind (relevant mostly if some
1663 : : * architecture has a slow bit-scan instruction)
1664 : : */
1665 : 21022 : static inline int block_size_bits(unsigned int blocksize)
1666 : : {
1667 [ - + - - : 21022 : return ilog2(blocksize);
- - - - -
- - - - -
- - - - -
- - - - -
- - - - -
- - - - -
- - - - -
- - - - -
- - - - -
- - - - -
- - - - -
- - - -
- ]
1668 : : }
1669 : :
1670 : 21022 : static struct buffer_head *create_page_buffers(struct page *page, struct inode *inode, unsigned int b_state)
1671 : : {
1672 [ - + - + ]: 42044 : BUG_ON(!PageLocked(page));
1673 : :
1674 [ + + ]: 21022 : if (!page_has_buffers(page))
1675 : 17796 : create_empty_buffers(page, 1 << READ_ONCE(inode->i_blkbits),
1676 : : b_state);
1677 [ - + ]: 21022 : return page_buffers(page);
1678 : : }
1679 : :
1680 : : /*
1681 : : * NOTE! All mapped/uptodate combinations are valid:
1682 : : *
1683 : : * Mapped Uptodate Meaning
1684 : : *
1685 : : * No No "unknown" - must do get_block()
1686 : : * No Yes "hole" - zero-filled
1687 : : * Yes No "allocated" - allocated on disk, not read in
1688 : : * Yes Yes "valid" - allocated and up-to-date in memory.
1689 : : *
1690 : : * "Dirty" is valid only with the last case (mapped+uptodate).
1691 : : */
1692 : :
1693 : : /*
1694 : : * While block_write_full_page is writing back the dirty buffers under
1695 : : * the page lock, whoever dirtied the buffers may decide to clean them
1696 : : * again at any time. We handle that by only looking at the buffer
1697 : : * state inside lock_buffer().
1698 : : *
1699 : : * If block_write_full_page() is called for regular writeback
1700 : : * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1701 : : * locked buffer. This only can happen if someone has written the buffer
1702 : : * directly, with submit_bh(). At the address_space level PageWriteback
1703 : : * prevents this contention from occurring.
1704 : : *
1705 : : * If block_write_full_page() is called with wbc->sync_mode ==
1706 : : * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this
1707 : : * causes the writes to be flagged as synchronous writes.
1708 : : */
1709 : 0 : int __block_write_full_page(struct inode *inode, struct page *page,
1710 : : get_block_t *get_block, struct writeback_control *wbc,
1711 : : bh_end_io_t *handler)
1712 : : {
1713 : 0 : int err;
1714 : 0 : sector_t block;
1715 : 0 : sector_t last_block;
1716 : 0 : struct buffer_head *bh, *head;
1717 : 0 : unsigned int blocksize, bbits;
1718 : 0 : int nr_underway = 0;
1719 [ # # ]: 0 : int write_flags = wbc_to_write_flags(wbc);
1720 : :
1721 : 0 : head = create_page_buffers(page, inode,
1722 : : (1 << BH_Dirty)|(1 << BH_Uptodate));
1723 : :
1724 : : /*
1725 : : * Be very careful. We have no exclusion from __set_page_dirty_buffers
1726 : : * here, and the (potentially unmapped) buffers may become dirty at
1727 : : * any time. If a buffer becomes dirty here after we've inspected it
1728 : : * then we just miss that fact, and the page stays dirty.
1729 : : *
1730 : : * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
1731 : : * handle that here by just cleaning them.
1732 : : */
1733 : :
1734 : 0 : bh = head;
1735 : 0 : blocksize = bh->b_size;
1736 : 0 : bbits = block_size_bits(blocksize);
1737 : :
1738 : 0 : block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1739 : 0 : last_block = (i_size_read(inode) - 1) >> bbits;
1740 : :
1741 : : /*
1742 : : * Get all the dirty buffers mapped to disk addresses and
1743 : : * handle any aliases from the underlying blockdev's mapping.
1744 : : */
1745 : 0 : do {
1746 [ # # ]: 0 : if (block > last_block) {
1747 : : /*
1748 : : * mapped buffers outside i_size will occur, because
1749 : : * this page can be outside i_size when there is a
1750 : : * truncate in progress.
1751 : : */
1752 : : /*
1753 : : * The buffer was zeroed by block_write_full_page()
1754 : : */
1755 : 0 : clear_buffer_dirty(bh);
1756 : 0 : set_buffer_uptodate(bh);
1757 [ # # # # : 0 : } else if ((!buffer_mapped(bh) || buffer_delay(bh)) &&
# # ]
1758 : : buffer_dirty(bh)) {
1759 [ # # ]: 0 : WARN_ON(bh->b_size != blocksize);
1760 : 0 : err = get_block(inode, block, bh, 1);
1761 [ # # ]: 0 : if (err)
1762 : 0 : goto recover;
1763 : 0 : clear_buffer_delay(bh);
1764 [ # # ]: 0 : if (buffer_new(bh)) {
1765 : : /* blockdev mappings never come here */
1766 : 0 : clear_buffer_new(bh);
1767 : 0 : clean_bdev_bh_alias(bh);
1768 : : }
1769 : : }
1770 : 0 : bh = bh->b_this_page;
1771 : 0 : block++;
1772 [ # # ]: 0 : } while (bh != head);
1773 : :
1774 : 0 : do {
1775 [ # # ]: 0 : if (!buffer_mapped(bh))
1776 : 0 : continue;
1777 : : /*
1778 : : * If it's a fully non-blocking write attempt and we cannot
1779 : : * lock the buffer then redirty the page. Note that this can
1780 : : * potentially cause a busy-wait loop from writeback threads
1781 : : * and kswapd activity, but those code paths have their own
1782 : : * higher-level throttling.
1783 : : */
1784 [ # # ]: 0 : if (wbc->sync_mode != WB_SYNC_NONE) {
1785 : 0 : lock_buffer(bh);
1786 [ # # ]: 0 : } else if (!trylock_buffer(bh)) {
1787 : 0 : redirty_page_for_writepage(wbc, page);
1788 : 0 : continue;
1789 : : }
1790 [ # # ]: 0 : if (test_clear_buffer_dirty(bh)) {
1791 : 0 : mark_buffer_async_write_endio(bh, handler);
1792 : : } else {
1793 : 0 : unlock_buffer(bh);
1794 : : }
1795 [ # # ]: 0 : } while ((bh = bh->b_this_page) != head);
1796 : :
1797 : : /*
1798 : : * The page and its buffers are protected by PageWriteback(), so we can
1799 : : * drop the bh refcounts early.
1800 : : */
1801 [ # # # # ]: 0 : BUG_ON(PageWriteback(page));
1802 : 0 : set_page_writeback(page);
1803 : :
1804 : 0 : do {
1805 : 0 : struct buffer_head *next = bh->b_this_page;
1806 [ # # ]: 0 : if (buffer_async_write(bh)) {
1807 : 0 : submit_bh_wbc(REQ_OP_WRITE, write_flags, bh,
1808 : 0 : inode->i_write_hint, wbc);
1809 : 0 : nr_underway++;
1810 : : }
1811 : 0 : bh = next;
1812 [ # # ]: 0 : } while (bh != head);
1813 : 0 : unlock_page(page);
1814 : :
1815 : 0 : err = 0;
1816 : 0 : done:
1817 [ # # ]: 0 : if (nr_underway == 0) {
1818 : : /*
1819 : : * The page was marked dirty, but the buffers were
1820 : : * clean. Someone wrote them back by hand with
1821 : : * ll_rw_block/submit_bh. A rare case.
1822 : : */
1823 : 0 : end_page_writeback(page);
1824 : :
1825 : : /*
1826 : : * The page and buffer_heads can be released at any time from
1827 : : * here on.
1828 : : */
1829 : : }
1830 : 0 : return err;
1831 : :
1832 : : recover:
1833 : : /*
1834 : : * ENOSPC, or some other error. We may already have added some
1835 : : * blocks to the file, so we need to write these out to avoid
1836 : : * exposing stale data.
1837 : : * The page is currently locked and not marked for writeback
1838 : : */
1839 : 0 : bh = head;
1840 : : /* Recovery: lock and submit the mapped buffers */
1841 : 0 : do {
1842 [ # # # # : 0 : if (buffer_mapped(bh) && buffer_dirty(bh) &&
# # ]
1843 : : !buffer_delay(bh)) {
1844 : 0 : lock_buffer(bh);
1845 : 0 : mark_buffer_async_write_endio(bh, handler);
1846 : : } else {
1847 : : /*
1848 : : * The buffer may have been set dirty during
1849 : : * attachment to a dirty page.
1850 : : */
1851 : 0 : clear_buffer_dirty(bh);
1852 : : }
1853 [ # # ]: 0 : } while ((bh = bh->b_this_page) != head);
1854 [ # # ]: 0 : SetPageError(page);
1855 [ # # # # ]: 0 : BUG_ON(PageWriteback(page));
1856 : 0 : mapping_set_error(page->mapping, err);
1857 : 0 : set_page_writeback(page);
1858 : 0 : do {
1859 : 0 : struct buffer_head *next = bh->b_this_page;
1860 [ # # ]: 0 : if (buffer_async_write(bh)) {
1861 : 0 : clear_buffer_dirty(bh);
1862 : 0 : submit_bh_wbc(REQ_OP_WRITE, write_flags, bh,
1863 : 0 : inode->i_write_hint, wbc);
1864 : 0 : nr_underway++;
1865 : : }
1866 : 0 : bh = next;
1867 [ # # ]: 0 : } while (bh != head);
1868 : 0 : unlock_page(page);
1869 : 0 : goto done;
1870 : : }
1871 : : EXPORT_SYMBOL(__block_write_full_page);
1872 : :
1873 : : /*
1874 : : * If a page has any new buffers, zero them out here, and mark them uptodate
1875 : : * and dirty so they'll be written out (in order to prevent uninitialised
1876 : : * block data from leaking). And clear the new bit.
1877 : : */
1878 : 0 : void page_zero_new_buffers(struct page *page, unsigned from, unsigned to)
1879 : : {
1880 : 0 : unsigned int block_start, block_end;
1881 : 0 : struct buffer_head *head, *bh;
1882 : :
1883 [ # # # # ]: 0 : BUG_ON(!PageLocked(page));
1884 [ # # ]: 0 : if (!page_has_buffers(page))
1885 : : return;
1886 : :
1887 [ # # ]: 0 : bh = head = page_buffers(page);
1888 : 0 : block_start = 0;
1889 : 0 : do {
1890 : 0 : block_end = block_start + bh->b_size;
1891 : :
1892 [ # # ]: 0 : if (buffer_new(bh)) {
1893 [ # # ]: 0 : if (block_end > from && block_start < to) {
1894 [ # # ]: 0 : if (!PageUptodate(page)) {
1895 : 0 : unsigned start, size;
1896 : :
1897 : 0 : start = max(from, block_start);
1898 : 0 : size = min(to, block_end) - start;
1899 : :
1900 : 0 : zero_user(page, start, size);
1901 : 0 : set_buffer_uptodate(bh);
1902 : : }
1903 : :
1904 : 0 : clear_buffer_new(bh);
1905 : 0 : mark_buffer_dirty(bh);
1906 : : }
1907 : : }
1908 : :
1909 : 0 : block_start = block_end;
1910 : 0 : bh = bh->b_this_page;
1911 [ # # ]: 0 : } while (bh != head);
1912 : : }
1913 : : EXPORT_SYMBOL(page_zero_new_buffers);
1914 : :
1915 : : static void
1916 : 0 : iomap_to_bh(struct inode *inode, sector_t block, struct buffer_head *bh,
1917 : : struct iomap *iomap)
1918 : : {
1919 : 0 : loff_t offset = block << inode->i_blkbits;
1920 : :
1921 : 0 : bh->b_bdev = iomap->bdev;
1922 : :
1923 : : /*
1924 : : * Block points to offset in file we need to map, iomap contains
1925 : : * the offset at which the map starts. If the map ends before the
1926 : : * current block, then do not map the buffer and let the caller
1927 : : * handle it.
1928 : : */
1929 [ # # ]: 0 : BUG_ON(offset >= iomap->offset + iomap->length);
1930 : :
1931 [ # # # # : 0 : switch (iomap->type) {
# ]
1932 : : case IOMAP_HOLE:
1933 : : /*
1934 : : * If the buffer is not up to date or beyond the current EOF,
1935 : : * we need to mark it as new to ensure sub-block zeroing is
1936 : : * executed if necessary.
1937 : : */
1938 [ # # # # ]: 0 : if (!buffer_uptodate(bh) ||
1939 [ # # ]: 0 : (offset >= i_size_read(inode)))
1940 : 0 : set_buffer_new(bh);
1941 : : break;
1942 : : case IOMAP_DELALLOC:
1943 [ # # # # ]: 0 : if (!buffer_uptodate(bh) ||
1944 [ # # ]: 0 : (offset >= i_size_read(inode)))
1945 : 0 : set_buffer_new(bh);
1946 : 0 : set_buffer_uptodate(bh);
1947 : 0 : set_buffer_mapped(bh);
1948 : 0 : set_buffer_delay(bh);
1949 : : break;
1950 : : case IOMAP_UNWRITTEN:
1951 : : /*
1952 : : * For unwritten regions, we always need to ensure that regions
1953 : : * in the block we are not writing to are zeroed. Mark the
1954 : : * buffer as new to ensure this.
1955 : : */
1956 : 0 : set_buffer_new(bh);
1957 : 0 : set_buffer_unwritten(bh);
1958 : : /* FALLTHRU */
1959 : 0 : case IOMAP_MAPPED:
1960 [ # # # # ]: 0 : if ((iomap->flags & IOMAP_F_NEW) ||
1961 [ # # ]: 0 : offset >= i_size_read(inode))
1962 : 0 : set_buffer_new(bh);
1963 : 0 : bh->b_blocknr = (iomap->addr + offset - iomap->offset) >>
1964 : 0 : inode->i_blkbits;
1965 : 0 : set_buffer_mapped(bh);
1966 : : break;
1967 : : }
1968 : 0 : }
1969 : :
1970 : 20872 : int __block_write_begin_int(struct page *page, loff_t pos, unsigned len,
1971 : : get_block_t *get_block, struct iomap *iomap)
1972 : : {
1973 : 20872 : unsigned from = pos & (PAGE_SIZE - 1);
1974 : 20872 : unsigned to = from + len;
1975 : 20872 : struct inode *inode = page->mapping->host;
1976 : 20872 : unsigned block_start, block_end;
1977 : 20872 : sector_t block;
1978 : 20872 : int err = 0;
1979 : 20872 : unsigned blocksize, bbits;
1980 : 20872 : struct buffer_head *bh, *head, *wait[2], **wait_bh=wait;
1981 : :
1982 [ - + - + ]: 41744 : BUG_ON(!PageLocked(page));
1983 : 20872 : BUG_ON(from > PAGE_SIZE);
1984 [ - + ]: 20872 : BUG_ON(to > PAGE_SIZE);
1985 [ - + ]: 20872 : BUG_ON(from > to);
1986 : :
1987 : 20872 : head = create_page_buffers(page, inode, 0);
1988 : 20872 : blocksize = head->b_size;
1989 : 20872 : bbits = block_size_bits(blocksize);
1990 : :
1991 : 20872 : block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1992 : :
1993 [ + + ]: 41744 : for(bh = head, block_start = 0; bh != head || !block_start;
1994 : 20872 : block++, block_start=block_end, bh = bh->b_this_page) {
1995 : 20872 : block_end = block_start + blocksize;
1996 [ - + ]: 20872 : if (block_end <= from || block_start >= to) {
1997 [ # # ]: 0 : if (PageUptodate(page)) {
1998 [ # # ]: 0 : if (!buffer_uptodate(bh))
1999 : 0 : set_buffer_uptodate(bh);
2000 : : }
2001 : 0 : continue;
2002 : : }
2003 [ - + ]: 20872 : if (buffer_new(bh))
2004 : 0 : clear_buffer_new(bh);
2005 [ + + ]: 20872 : if (!buffer_mapped(bh)) {
2006 [ - + ]: 17736 : WARN_ON(bh->b_size != blocksize);
2007 [ + - ]: 17736 : if (get_block) {
2008 : 17736 : err = get_block(inode, block, bh, 1);
2009 [ - + ]: 17736 : if (err)
2010 : : break;
2011 : : } else {
2012 : 0 : iomap_to_bh(inode, block, bh, iomap);
2013 : : }
2014 : :
2015 [ + - ]: 17736 : if (buffer_new(bh)) {
2016 : 17736 : clean_bdev_bh_alias(bh);
2017 [ + + ]: 17736 : if (PageUptodate(page)) {
2018 : 30 : clear_buffer_new(bh);
2019 : 30 : set_buffer_uptodate(bh);
2020 : 30 : mark_buffer_dirty(bh);
2021 : 30 : continue;
2022 : : }
2023 [ + + ]: 17706 : if (block_end > to || block_start < from)
2024 : 10941 : zero_user_segments(page,
2025 : : to, block_end,
2026 : : block_start, from);
2027 : 17706 : continue;
2028 : : }
2029 : : }
2030 [ + - ]: 3136 : if (PageUptodate(page)) {
2031 [ - + ]: 3136 : if (!buffer_uptodate(bh))
2032 : 0 : set_buffer_uptodate(bh);
2033 : 3136 : continue;
2034 : : }
2035 [ # # # # : 0 : if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
# # ]
2036 : 0 : !buffer_unwritten(bh) &&
2037 [ # # ]: 0 : (block_start < from || block_end > to)) {
2038 : 0 : ll_rw_block(REQ_OP_READ, 0, 1, &bh);
2039 : 0 : *wait_bh++=bh;
2040 : : }
2041 : : }
2042 : : /*
2043 : : * If we issued read requests - let them complete.
2044 : : */
2045 [ - + ]: 20872 : while(wait_bh > wait) {
2046 : 0 : wait_on_buffer(*--wait_bh);
2047 [ # # ]: 0 : if (!buffer_uptodate(*wait_bh))
2048 : 0 : err = -EIO;
2049 : : }
2050 [ - + ]: 20872 : if (unlikely(err))
2051 : 0 : page_zero_new_buffers(page, from, to);
2052 : 20872 : return err;
2053 : : }
2054 : :
2055 : 20872 : int __block_write_begin(struct page *page, loff_t pos, unsigned len,
2056 : : get_block_t *get_block)
2057 : : {
2058 : 20872 : return __block_write_begin_int(page, pos, len, get_block, NULL);
2059 : : }
2060 : : EXPORT_SYMBOL(__block_write_begin);
2061 : :
2062 : : static int __block_commit_write(struct inode *inode, struct page *page,
2063 : : unsigned from, unsigned to)
2064 : : {
2065 : : unsigned block_start, block_end;
2066 : : int partial = 0;
2067 : : unsigned blocksize;
2068 : : struct buffer_head *bh, *head;
2069 : :
2070 : : bh = head = page_buffers(page);
2071 : : blocksize = bh->b_size;
2072 : :
2073 : : block_start = 0;
2074 : : do {
2075 : : block_end = block_start + blocksize;
2076 : : if (block_end <= from || block_start >= to) {
2077 : : if (!buffer_uptodate(bh))
2078 : : partial = 1;
2079 : : } else {
2080 : : set_buffer_uptodate(bh);
2081 : : mark_buffer_dirty(bh);
2082 : : }
2083 : : clear_buffer_new(bh);
2084 : :
2085 : : block_start = block_end;
2086 : : bh = bh->b_this_page;
2087 : : } while (bh != head);
2088 : :
2089 : : /*
2090 : : * If this is a partial write which happened to make all buffers
2091 : : * uptodate then we can optimize away a bogus readpage() for
2092 : : * the next read(). Here we 'discover' whether the page went
2093 : : * uptodate as a result of this (potentially partial) write.
2094 : : */
2095 : : if (!partial)
2096 : : SetPageUptodate(page);
2097 : : return 0;
2098 : : }
2099 : :
2100 : : /*
2101 : : * block_write_begin takes care of the basic task of block allocation and
2102 : : * bringing partial write blocks uptodate first.
2103 : : *
2104 : : * The filesystem needs to handle block truncation upon failure.
2105 : : */
2106 : 0 : int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,
2107 : : unsigned flags, struct page **pagep, get_block_t *get_block)
2108 : : {
2109 : 0 : pgoff_t index = pos >> PAGE_SHIFT;
2110 : 0 : struct page *page;
2111 : 0 : int status;
2112 : :
2113 : 0 : page = grab_cache_page_write_begin(mapping, index, flags);
2114 [ # # ]: 0 : if (!page)
2115 : : return -ENOMEM;
2116 : :
2117 : 0 : status = __block_write_begin(page, pos, len, get_block);
2118 [ # # ]: 0 : if (unlikely(status)) {
2119 : 0 : unlock_page(page);
2120 : 0 : put_page(page);
2121 : 0 : page = NULL;
2122 : : }
2123 : :
2124 : 0 : *pagep = page;
2125 : 0 : return status;
2126 : : }
2127 : : EXPORT_SYMBOL(block_write_begin);
2128 : :
2129 : 20872 : int block_write_end(struct file *file, struct address_space *mapping,
2130 : : loff_t pos, unsigned len, unsigned copied,
2131 : : struct page *page, void *fsdata)
2132 : : {
2133 : 20872 : struct inode *inode = mapping->host;
2134 : 20872 : unsigned start;
2135 : :
2136 : 20872 : start = pos & (PAGE_SIZE - 1);
2137 : :
2138 [ - + ]: 20872 : if (unlikely(copied < len)) {
2139 : : /*
2140 : : * The buffers that were written will now be uptodate, so we
2141 : : * don't have to worry about a readpage reading them and
2142 : : * overwriting a partial write. However if we have encountered
2143 : : * a short write and only partially written into a buffer, it
2144 : : * will not be marked uptodate, so a readpage might come in and
2145 : : * destroy our partial write.
2146 : : *
2147 : : * Do the simplest thing, and just treat any short write to a
2148 : : * non uptodate page as a zero-length write, and force the
2149 : : * caller to redo the whole thing.
2150 : : */
2151 [ # # ]: 0 : if (!PageUptodate(page))
2152 : 0 : copied = 0;
2153 : :
2154 : 0 : page_zero_new_buffers(page, start+copied, start+len);
2155 : : }
2156 : 20872 : flush_dcache_page(page);
2157 : :
2158 : : /* This could be a short (even 0-length) commit */
2159 : 20872 : __block_commit_write(inode, page, start, start+copied);
2160 : :
2161 : 20872 : return copied;
2162 : : }
2163 : : EXPORT_SYMBOL(block_write_end);
2164 : :
2165 : 20872 : int generic_write_end(struct file *file, struct address_space *mapping,
2166 : : loff_t pos, unsigned len, unsigned copied,
2167 : : struct page *page, void *fsdata)
2168 : : {
2169 : 20872 : struct inode *inode = mapping->host;
2170 : 20872 : loff_t old_size = inode->i_size;
2171 : 20872 : bool i_size_changed = false;
2172 : :
2173 : 20872 : copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
2174 : :
2175 : : /*
2176 : : * No need to use i_size_read() here, the i_size cannot change under us
2177 : : * because we hold i_rwsem.
2178 : : *
2179 : : * But it's important to update i_size while still holding page lock:
2180 : : * page writeout could otherwise come in and zero beyond i_size.
2181 : : */
2182 [ + - ]: 20872 : if (pos + copied > inode->i_size) {
2183 : 20872 : i_size_write(inode, pos + copied);
2184 : 20872 : i_size_changed = true;
2185 : : }
2186 : :
2187 : 20872 : unlock_page(page);
2188 : 20872 : put_page(page);
2189 : :
2190 [ - + ]: 20872 : if (old_size < pos)
2191 : 0 : pagecache_isize_extended(inode, old_size, pos);
2192 : : /*
2193 : : * Don't mark the inode dirty under page lock. First, it unnecessarily
2194 : : * makes the holding time of page lock longer. Second, it forces lock
2195 : : * ordering of page lock and transaction start for journaling
2196 : : * filesystems.
2197 : : */
2198 [ + - ]: 20872 : if (i_size_changed)
2199 : 20872 : mark_inode_dirty(inode);
2200 : 20872 : return copied;
2201 : : }
2202 : : EXPORT_SYMBOL(generic_write_end);
2203 : :
2204 : : /*
2205 : : * block_is_partially_uptodate checks whether buffers within a page are
2206 : : * uptodate or not.
2207 : : *
2208 : : * Returns true if all buffers which correspond to a file portion
2209 : : * we want to read are uptodate.
2210 : : */
2211 : 0 : int block_is_partially_uptodate(struct page *page, unsigned long from,
2212 : : unsigned long count)
2213 : : {
2214 : 0 : unsigned block_start, block_end, blocksize;
2215 : 0 : unsigned to;
2216 : 0 : struct buffer_head *bh, *head;
2217 : 0 : int ret = 1;
2218 : :
2219 [ # # ]: 0 : if (!page_has_buffers(page))
2220 : : return 0;
2221 : :
2222 [ # # ]: 0 : head = page_buffers(page);
2223 : 0 : blocksize = head->b_size;
2224 : 0 : to = min_t(unsigned, PAGE_SIZE - from, count);
2225 : 0 : to = from + to;
2226 [ # # # # ]: 0 : if (from < blocksize && to > PAGE_SIZE - blocksize)
2227 : : return 0;
2228 : :
2229 : : bh = head;
2230 : : block_start = 0;
2231 : 0 : do {
2232 : 0 : block_end = block_start + blocksize;
2233 [ # # # # ]: 0 : if (block_end > from && block_start < to) {
2234 [ # # ]: 0 : if (!buffer_uptodate(bh)) {
2235 : : ret = 0;
2236 : : break;
2237 : : }
2238 [ # # ]: 0 : if (block_end >= to)
2239 : : break;
2240 : : }
2241 : 0 : block_start = block_end;
2242 : 0 : bh = bh->b_this_page;
2243 [ # # ]: 0 : } while (bh != head);
2244 : :
2245 : : return ret;
2246 : : }
2247 : : EXPORT_SYMBOL(block_is_partially_uptodate);
2248 : :
2249 : : /*
2250 : : * Generic "read page" function for block devices that have the normal
2251 : : * get_block functionality. This is most of the block device filesystems.
2252 : : * Reads the page asynchronously --- the unlock_buffer() and
2253 : : * set/clear_buffer_uptodate() functions propagate buffer state into the
2254 : : * page struct once IO has completed.
2255 : : */
2256 : 150 : int block_read_full_page(struct page *page, get_block_t *get_block)
2257 : : {
2258 : 150 : struct inode *inode = page->mapping->host;
2259 : 150 : sector_t iblock, lblock;
2260 : 150 : struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
2261 : 150 : unsigned int blocksize, bbits;
2262 : 150 : int nr, i;
2263 : 150 : int fully_mapped = 1;
2264 : :
2265 : 150 : head = create_page_buffers(page, inode, 0);
2266 : 150 : blocksize = head->b_size;
2267 : 150 : bbits = block_size_bits(blocksize);
2268 : :
2269 : 150 : iblock = (sector_t)page->index << (PAGE_SHIFT - bbits);
2270 : 150 : lblock = (i_size_read(inode)+blocksize-1) >> bbits;
2271 : 150 : bh = head;
2272 : 150 : nr = 0;
2273 : 150 : i = 0;
2274 : :
2275 : 150 : do {
2276 [ + + ]: 150 : if (buffer_uptodate(bh))
2277 : 90 : continue;
2278 : :
2279 [ + - ]: 60 : if (!buffer_mapped(bh)) {
2280 : 60 : int err = 0;
2281 : :
2282 : 60 : fully_mapped = 0;
2283 [ + - ]: 60 : if (iblock < lblock) {
2284 [ - + ]: 60 : WARN_ON(bh->b_size != blocksize);
2285 : 60 : err = get_block(inode, iblock, bh, 0);
2286 [ - + ]: 60 : if (err)
2287 [ # # ]: 0 : SetPageError(page);
2288 : : }
2289 [ - + ]: 60 : if (!buffer_mapped(bh)) {
2290 : 0 : zero_user(page, i * blocksize, blocksize);
2291 [ # # ]: 0 : if (!err)
2292 : 0 : set_buffer_uptodate(bh);
2293 : 0 : continue;
2294 : : }
2295 : : /*
2296 : : * get_block() might have updated the buffer
2297 : : * synchronously
2298 : : */
2299 [ - + ]: 60 : if (buffer_uptodate(bh))
2300 : 0 : continue;
2301 : : }
2302 : 60 : arr[nr++] = bh;
2303 [ - + ]: 150 : } while (i++, iblock++, (bh = bh->b_this_page) != head);
2304 : :
2305 [ + + ]: 150 : if (fully_mapped)
2306 [ - + ]: 90 : SetPageMappedToDisk(page);
2307 : :
2308 [ + + ]: 150 : if (!nr) {
2309 : : /*
2310 : : * All buffers are uptodate - we can set the page uptodate
2311 : : * as well. But not if get_block() returned an error.
2312 : : */
2313 [ - + + - ]: 180 : if (!PageError(page))
2314 : 90 : SetPageUptodate(page);
2315 : 90 : unlock_page(page);
2316 : 90 : return 0;
2317 : : }
2318 : :
2319 : : /* Stage two: lock the buffers */
2320 [ + + ]: 120 : for (i = 0; i < nr; i++) {
2321 : 60 : bh = arr[i];
2322 : 60 : lock_buffer(bh);
2323 : 60 : mark_buffer_async_read(bh);
2324 : : }
2325 : :
2326 : : /*
2327 : : * Stage 3: start the IO. Check for uptodateness
2328 : : * inside the buffer lock in case another process reading
2329 : : * the underlying blockdev brought it uptodate (the sct fix).
2330 : : */
2331 [ + + ]: 120 : for (i = 0; i < nr; i++) {
2332 : 60 : bh = arr[i];
2333 [ - + ]: 60 : if (buffer_uptodate(bh))
2334 : 0 : end_buffer_async_read(bh, 1);
2335 : : else
2336 : 60 : submit_bh(REQ_OP_READ, 0, bh);
2337 : : }
2338 : : return 0;
2339 : : }
2340 : : EXPORT_SYMBOL(block_read_full_page);
2341 : :
2342 : : /* utility function for filesystems that need to do work on expanding
2343 : : * truncates. Uses filesystem pagecache writes to allow the filesystem to
2344 : : * deal with the hole.
2345 : : */
2346 : 0 : int generic_cont_expand_simple(struct inode *inode, loff_t size)
2347 : : {
2348 : 0 : struct address_space *mapping = inode->i_mapping;
2349 : 0 : struct page *page;
2350 : 0 : void *fsdata;
2351 : 0 : int err;
2352 : :
2353 : 0 : err = inode_newsize_ok(inode, size);
2354 [ # # ]: 0 : if (err)
2355 : 0 : goto out;
2356 : :
2357 : 0 : err = pagecache_write_begin(NULL, mapping, size, 0,
2358 : : AOP_FLAG_CONT_EXPAND, &page, &fsdata);
2359 [ # # ]: 0 : if (err)
2360 : 0 : goto out;
2361 : :
2362 : 0 : err = pagecache_write_end(NULL, mapping, size, 0, 0, page, fsdata);
2363 [ # # ]: 0 : BUG_ON(err > 0);
2364 : :
2365 : 0 : out:
2366 : 0 : return err;
2367 : : }
2368 : : EXPORT_SYMBOL(generic_cont_expand_simple);
2369 : :
2370 : 0 : static int cont_expand_zero(struct file *file, struct address_space *mapping,
2371 : : loff_t pos, loff_t *bytes)
2372 : : {
2373 : 0 : struct inode *inode = mapping->host;
2374 : 0 : unsigned int blocksize = i_blocksize(inode);
2375 : 0 : struct page *page;
2376 : 0 : void *fsdata;
2377 : 0 : pgoff_t index, curidx;
2378 : 0 : loff_t curpos;
2379 : 0 : unsigned zerofrom, offset, len;
2380 : 0 : int err = 0;
2381 : :
2382 : 0 : index = pos >> PAGE_SHIFT;
2383 : 0 : offset = pos & ~PAGE_MASK;
2384 : :
2385 [ # # ]: 0 : while (index > (curidx = (curpos = *bytes)>>PAGE_SHIFT)) {
2386 : 0 : zerofrom = curpos & ~PAGE_MASK;
2387 [ # # ]: 0 : if (zerofrom & (blocksize-1)) {
2388 : 0 : *bytes |= (blocksize-1);
2389 : 0 : (*bytes)++;
2390 : : }
2391 : 0 : len = PAGE_SIZE - zerofrom;
2392 : :
2393 : 0 : err = pagecache_write_begin(file, mapping, curpos, len, 0,
2394 : : &page, &fsdata);
2395 [ # # ]: 0 : if (err)
2396 : 0 : goto out;
2397 : 0 : zero_user(page, zerofrom, len);
2398 : 0 : err = pagecache_write_end(file, mapping, curpos, len, len,
2399 : : page, fsdata);
2400 [ # # ]: 0 : if (err < 0)
2401 : 0 : goto out;
2402 [ # # ]: 0 : BUG_ON(err != len);
2403 : 0 : err = 0;
2404 : :
2405 : 0 : balance_dirty_pages_ratelimited(mapping);
2406 : :
2407 [ # # ]: 0 : if (fatal_signal_pending(current)) {
2408 : 0 : err = -EINTR;
2409 : 0 : goto out;
2410 : : }
2411 : : }
2412 : :
2413 : : /* page covers the boundary, find the boundary offset */
2414 [ # # ]: 0 : if (index == curidx) {
2415 : 0 : zerofrom = curpos & ~PAGE_MASK;
2416 : : /* if we will expand the thing last block will be filled */
2417 [ # # ]: 0 : if (offset <= zerofrom) {
2418 : 0 : goto out;
2419 : : }
2420 [ # # ]: 0 : if (zerofrom & (blocksize-1)) {
2421 : 0 : *bytes |= (blocksize-1);
2422 : 0 : (*bytes)++;
2423 : : }
2424 : 0 : len = offset - zerofrom;
2425 : :
2426 : 0 : err = pagecache_write_begin(file, mapping, curpos, len, 0,
2427 : : &page, &fsdata);
2428 [ # # ]: 0 : if (err)
2429 : 0 : goto out;
2430 : 0 : zero_user(page, zerofrom, len);
2431 : 0 : err = pagecache_write_end(file, mapping, curpos, len, len,
2432 : : page, fsdata);
2433 [ # # ]: 0 : if (err < 0)
2434 : 0 : goto out;
2435 [ # # ]: 0 : BUG_ON(err != len);
2436 : : err = 0;
2437 : : }
2438 : 0 : out:
2439 : 0 : return err;
2440 : : }
2441 : :
2442 : : /*
2443 : : * For moronic filesystems that do not allow holes in file.
2444 : : * We may have to extend the file.
2445 : : */
2446 : 0 : int cont_write_begin(struct file *file, struct address_space *mapping,
2447 : : loff_t pos, unsigned len, unsigned flags,
2448 : : struct page **pagep, void **fsdata,
2449 : : get_block_t *get_block, loff_t *bytes)
2450 : : {
2451 : 0 : struct inode *inode = mapping->host;
2452 : 0 : unsigned int blocksize = i_blocksize(inode);
2453 : 0 : unsigned int zerofrom;
2454 : 0 : int err;
2455 : :
2456 : 0 : err = cont_expand_zero(file, mapping, pos, bytes);
2457 [ # # ]: 0 : if (err)
2458 : : return err;
2459 : :
2460 : 0 : zerofrom = *bytes & ~PAGE_MASK;
2461 [ # # # # ]: 0 : if (pos+len > *bytes && zerofrom & (blocksize-1)) {
2462 : 0 : *bytes |= (blocksize-1);
2463 : 0 : (*bytes)++;
2464 : : }
2465 : :
2466 : 0 : return block_write_begin(mapping, pos, len, flags, pagep, get_block);
2467 : : }
2468 : : EXPORT_SYMBOL(cont_write_begin);
2469 : :
2470 : 0 : int block_commit_write(struct page *page, unsigned from, unsigned to)
2471 : : {
2472 : 0 : struct inode *inode = page->mapping->host;
2473 : 0 : __block_commit_write(inode,page,from,to);
2474 : 0 : return 0;
2475 : : }
2476 : : EXPORT_SYMBOL(block_commit_write);
2477 : :
2478 : : /*
2479 : : * block_page_mkwrite() is not allowed to change the file size as it gets
2480 : : * called from a page fault handler when a page is first dirtied. Hence we must
2481 : : * be careful to check for EOF conditions here. We set the page up correctly
2482 : : * for a written page which means we get ENOSPC checking when writing into
2483 : : * holes and correct delalloc and unwritten extent mapping on filesystems that
2484 : : * support these features.
2485 : : *
2486 : : * We are not allowed to take the i_mutex here so we have to play games to
2487 : : * protect against truncate races as the page could now be beyond EOF. Because
2488 : : * truncate writes the inode size before removing pages, once we have the
2489 : : * page lock we can determine safely if the page is beyond EOF. If it is not
2490 : : * beyond EOF, then the page is guaranteed safe against truncation until we
2491 : : * unlock the page.
2492 : : *
2493 : : * Direct callers of this function should protect against filesystem freezing
2494 : : * using sb_start_pagefault() - sb_end_pagefault() functions.
2495 : : */
2496 : 0 : int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
2497 : : get_block_t get_block)
2498 : : {
2499 : 0 : struct page *page = vmf->page;
2500 : 0 : struct inode *inode = file_inode(vma->vm_file);
2501 : 0 : unsigned long end;
2502 : 0 : loff_t size;
2503 : 0 : int ret;
2504 : :
2505 : 0 : lock_page(page);
2506 [ # # ]: 0 : size = i_size_read(inode);
2507 [ # # # # ]: 0 : if ((page->mapping != inode->i_mapping) ||
2508 [ # # ]: 0 : (page_offset(page) > size)) {
2509 : : /* We overload EFAULT to mean page got truncated */
2510 : 0 : ret = -EFAULT;
2511 : 0 : goto out_unlock;
2512 : : }
2513 : :
2514 : : /* page is wholly or partially inside EOF */
2515 [ # # ]: 0 : if (((page->index + 1) << PAGE_SHIFT) > size)
2516 : 0 : end = size & ~PAGE_MASK;
2517 : : else
2518 : : end = PAGE_SIZE;
2519 : :
2520 : 0 : ret = __block_write_begin(page, 0, end, get_block);
2521 [ # # ]: 0 : if (!ret)
2522 : 0 : ret = block_commit_write(page, 0, end);
2523 : :
2524 [ # # ]: 0 : if (unlikely(ret < 0))
2525 : 0 : goto out_unlock;
2526 : 0 : set_page_dirty(page);
2527 : 0 : wait_for_stable_page(page);
2528 : 0 : return 0;
2529 : 0 : out_unlock:
2530 : 0 : unlock_page(page);
2531 : 0 : return ret;
2532 : : }
2533 : : EXPORT_SYMBOL(block_page_mkwrite);
2534 : :
2535 : : /*
2536 : : * nobh_write_begin()'s prereads are special: the buffer_heads are freed
2537 : : * immediately, while under the page lock. So it needs a special end_io
2538 : : * handler which does not touch the bh after unlocking it.
2539 : : */
2540 : 0 : static void end_buffer_read_nobh(struct buffer_head *bh, int uptodate)
2541 : : {
2542 : 0 : __end_buffer_read_notouch(bh, uptodate);
2543 : 0 : }
2544 : :
2545 : : /*
2546 : : * Attach the singly-linked list of buffers created by nobh_write_begin, to
2547 : : * the page (converting it to circular linked list and taking care of page
2548 : : * dirty races).
2549 : : */
2550 : 0 : static void attach_nobh_buffers(struct page *page, struct buffer_head *head)
2551 : : {
2552 : 0 : struct buffer_head *bh;
2553 : :
2554 [ # # # # ]: 0 : BUG_ON(!PageLocked(page));
2555 : :
2556 : 0 : spin_lock(&page->mapping->private_lock);
2557 : 0 : bh = head;
2558 : 0 : do {
2559 [ # # # # ]: 0 : if (PageDirty(page))
2560 : 0 : set_buffer_dirty(bh);
2561 [ # # ]: 0 : if (!bh->b_this_page)
2562 : 0 : bh->b_this_page = head;
2563 : 0 : bh = bh->b_this_page;
2564 [ # # ]: 0 : } while (bh != head);
2565 : 0 : attach_page_buffers(page, head);
2566 : 0 : spin_unlock(&page->mapping->private_lock);
2567 : 0 : }
2568 : :
2569 : : /*
2570 : : * On entry, the page is fully not uptodate.
2571 : : * On exit the page is fully uptodate in the areas outside (from,to)
2572 : : * The filesystem needs to handle block truncation upon failure.
2573 : : */
2574 : 0 : int nobh_write_begin(struct address_space *mapping,
2575 : : loff_t pos, unsigned len, unsigned flags,
2576 : : struct page **pagep, void **fsdata,
2577 : : get_block_t *get_block)
2578 : : {
2579 : 0 : struct inode *inode = mapping->host;
2580 : 0 : const unsigned blkbits = inode->i_blkbits;
2581 : 0 : const unsigned blocksize = 1 << blkbits;
2582 : 0 : struct buffer_head *head, *bh;
2583 : 0 : struct page *page;
2584 : 0 : pgoff_t index;
2585 : 0 : unsigned from, to;
2586 : 0 : unsigned block_in_page;
2587 : 0 : unsigned block_start, block_end;
2588 : 0 : sector_t block_in_file;
2589 : 0 : int nr_reads = 0;
2590 : 0 : int ret = 0;
2591 : 0 : int is_mapped_to_disk = 1;
2592 : :
2593 : 0 : index = pos >> PAGE_SHIFT;
2594 : 0 : from = pos & (PAGE_SIZE - 1);
2595 : 0 : to = from + len;
2596 : :
2597 : 0 : page = grab_cache_page_write_begin(mapping, index, flags);
2598 [ # # ]: 0 : if (!page)
2599 : : return -ENOMEM;
2600 : 0 : *pagep = page;
2601 : 0 : *fsdata = NULL;
2602 : :
2603 [ # # ]: 0 : if (page_has_buffers(page)) {
2604 : 0 : ret = __block_write_begin(page, pos, len, get_block);
2605 [ # # ]: 0 : if (unlikely(ret))
2606 : 0 : goto out_release;
2607 : : return ret;
2608 : : }
2609 : :
2610 [ # # # # ]: 0 : if (PageMappedToDisk(page))
2611 : : return 0;
2612 : :
2613 : : /*
2614 : : * Allocate buffers so that we can keep track of state, and potentially
2615 : : * attach them to the page if an error occurs. In the common case of
2616 : : * no error, they will just be freed again without ever being attached
2617 : : * to the page (which is all OK, because we're under the page lock).
2618 : : *
2619 : : * Be careful: the buffer linked list is a NULL terminated one, rather
2620 : : * than the circular one we're used to.
2621 : : */
2622 : 0 : head = alloc_page_buffers(page, blocksize, false);
2623 [ # # ]: 0 : if (!head) {
2624 : 0 : ret = -ENOMEM;
2625 : 0 : goto out_release;
2626 : : }
2627 : :
2628 : 0 : block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
2629 : :
2630 : : /*
2631 : : * We loop across all blocks in the page, whether or not they are
2632 : : * part of the affected region. This is so we can discover if the
2633 : : * page is fully mapped-to-disk.
2634 : : */
2635 : 0 : for (block_start = 0, block_in_page = 0, bh = head;
2636 [ # # ]: 0 : block_start < PAGE_SIZE;
2637 : 0 : block_in_page++, block_start += blocksize, bh = bh->b_this_page) {
2638 : 0 : int create;
2639 : :
2640 : 0 : block_end = block_start + blocksize;
2641 : 0 : bh->b_state = 0;
2642 : 0 : create = 1;
2643 [ # # ]: 0 : if (block_start >= to)
2644 : 0 : create = 0;
2645 : 0 : ret = get_block(inode, block_in_file + block_in_page,
2646 : : bh, create);
2647 [ # # ]: 0 : if (ret)
2648 : 0 : goto failed;
2649 [ # # ]: 0 : if (!buffer_mapped(bh))
2650 : 0 : is_mapped_to_disk = 0;
2651 [ # # ]: 0 : if (buffer_new(bh))
2652 : 0 : clean_bdev_bh_alias(bh);
2653 [ # # ]: 0 : if (PageUptodate(page)) {
2654 : 0 : set_buffer_uptodate(bh);
2655 : 0 : continue;
2656 : : }
2657 [ # # # # ]: 0 : if (buffer_new(bh) || !buffer_mapped(bh)) {
2658 : 0 : zero_user_segments(page, block_start, from,
2659 : : to, block_end);
2660 : 0 : continue;
2661 : : }
2662 [ # # ]: 0 : if (buffer_uptodate(bh))
2663 : 0 : continue; /* reiserfs does this */
2664 [ # # ]: 0 : if (block_start < from || block_end > to) {
2665 : 0 : lock_buffer(bh);
2666 : 0 : bh->b_end_io = end_buffer_read_nobh;
2667 : 0 : submit_bh(REQ_OP_READ, 0, bh);
2668 : 0 : nr_reads++;
2669 : : }
2670 : : }
2671 : :
2672 [ # # ]: 0 : if (nr_reads) {
2673 : : /*
2674 : : * The page is locked, so these buffers are protected from
2675 : : * any VM or truncate activity. Hence we don't need to care
2676 : : * for the buffer_head refcounts.
2677 : : */
2678 [ # # ]: 0 : for (bh = head; bh; bh = bh->b_this_page) {
2679 : 0 : wait_on_buffer(bh);
2680 [ # # ]: 0 : if (!buffer_uptodate(bh))
2681 : 0 : ret = -EIO;
2682 : : }
2683 [ # # ]: 0 : if (ret)
2684 : 0 : goto failed;
2685 : : }
2686 : :
2687 [ # # ]: 0 : if (is_mapped_to_disk)
2688 [ # # ]: 0 : SetPageMappedToDisk(page);
2689 : :
2690 : 0 : *fsdata = head; /* to be released by nobh_write_end */
2691 : :
2692 : 0 : return 0;
2693 : :
2694 : 0 : failed:
2695 [ # # ]: 0 : BUG_ON(!ret);
2696 : : /*
2697 : : * Error recovery is a bit difficult. We need to zero out blocks that
2698 : : * were newly allocated, and dirty them to ensure they get written out.
2699 : : * Buffers need to be attached to the page at this point, otherwise
2700 : : * the handling of potential IO errors during writeout would be hard
2701 : : * (could try doing synchronous writeout, but what if that fails too?)
2702 : : */
2703 : 0 : attach_nobh_buffers(page, head);
2704 : 0 : page_zero_new_buffers(page, from, to);
2705 : :
2706 : 0 : out_release:
2707 : 0 : unlock_page(page);
2708 : 0 : put_page(page);
2709 : 0 : *pagep = NULL;
2710 : :
2711 : 0 : return ret;
2712 : : }
2713 : : EXPORT_SYMBOL(nobh_write_begin);
2714 : :
2715 : 0 : int nobh_write_end(struct file *file, struct address_space *mapping,
2716 : : loff_t pos, unsigned len, unsigned copied,
2717 : : struct page *page, void *fsdata)
2718 : : {
2719 : 0 : struct inode *inode = page->mapping->host;
2720 : 0 : struct buffer_head *head = fsdata;
2721 : 0 : struct buffer_head *bh;
2722 [ # # # # ]: 0 : BUG_ON(fsdata != NULL && page_has_buffers(page));
2723 : :
2724 [ # # # # ]: 0 : if (unlikely(copied < len) && head)
2725 : 0 : attach_nobh_buffers(page, head);
2726 [ # # ]: 0 : if (page_has_buffers(page))
2727 : 0 : return generic_write_end(file, mapping, pos, len,
2728 : : copied, page, fsdata);
2729 : :
2730 : 0 : SetPageUptodate(page);
2731 : 0 : set_page_dirty(page);
2732 [ # # ]: 0 : if (pos+copied > inode->i_size) {
2733 : 0 : i_size_write(inode, pos+copied);
2734 : 0 : mark_inode_dirty(inode);
2735 : : }
2736 : :
2737 : 0 : unlock_page(page);
2738 : 0 : put_page(page);
2739 : :
2740 [ # # ]: 0 : while (head) {
2741 : 0 : bh = head;
2742 : 0 : head = head->b_this_page;
2743 : 0 : free_buffer_head(bh);
2744 : : }
2745 : :
2746 : 0 : return copied;
2747 : : }
2748 : : EXPORT_SYMBOL(nobh_write_end);
2749 : :
2750 : : /*
2751 : : * nobh_writepage() - based on block_full_write_page() except
2752 : : * that it tries to operate without attaching bufferheads to
2753 : : * the page.
2754 : : */
2755 : 0 : int nobh_writepage(struct page *page, get_block_t *get_block,
2756 : : struct writeback_control *wbc)
2757 : : {
2758 : 0 : struct inode * const inode = page->mapping->host;
2759 [ # # ]: 0 : loff_t i_size = i_size_read(inode);
2760 : 0 : const pgoff_t end_index = i_size >> PAGE_SHIFT;
2761 : 0 : unsigned offset;
2762 : 0 : int ret;
2763 : :
2764 : : /* Is the page fully inside i_size? */
2765 [ # # ]: 0 : if (page->index < end_index)
2766 : 0 : goto out;
2767 : :
2768 : : /* Is the page fully outside i_size? (truncate in progress) */
2769 : 0 : offset = i_size & (PAGE_SIZE-1);
2770 [ # # # # ]: 0 : if (page->index >= end_index+1 || !offset) {
2771 : : /*
2772 : : * The page may have dirty, unmapped buffers. For example,
2773 : : * they may have been added in ext3_writepage(). Make them
2774 : : * freeable here, so the page does not leak.
2775 : : */
2776 : : #if 0
2777 : : /* Not really sure about this - do we need this ? */
2778 : : if (page->mapping->a_ops->invalidatepage)
2779 : : page->mapping->a_ops->invalidatepage(page, offset);
2780 : : #endif
2781 : 0 : unlock_page(page);
2782 : 0 : return 0; /* don't care */
2783 : : }
2784 : :
2785 : : /*
2786 : : * The page straddles i_size. It must be zeroed out on each and every
2787 : : * writepage invocation because it may be mmapped. "A file is mapped
2788 : : * in multiples of the page size. For a file that is not a multiple of
2789 : : * the page size, the remaining memory is zeroed when mapped, and
2790 : : * writes to that region are not written out to the file."
2791 : : */
2792 : 0 : zero_user_segment(page, offset, PAGE_SIZE);
2793 : 0 : out:
2794 : 0 : ret = mpage_writepage(page, get_block, wbc);
2795 [ # # ]: 0 : if (ret == -EAGAIN)
2796 : 0 : ret = __block_write_full_page(inode, page, get_block, wbc,
2797 : : end_buffer_async_write);
2798 : : return ret;
2799 : : }
2800 : : EXPORT_SYMBOL(nobh_writepage);
2801 : :
2802 : 0 : int nobh_truncate_page(struct address_space *mapping,
2803 : : loff_t from, get_block_t *get_block)
2804 : : {
2805 : 0 : pgoff_t index = from >> PAGE_SHIFT;
2806 : 0 : unsigned offset = from & (PAGE_SIZE-1);
2807 : 0 : unsigned blocksize;
2808 : 0 : sector_t iblock;
2809 : 0 : unsigned length, pos;
2810 : 0 : struct inode *inode = mapping->host;
2811 : 0 : struct page *page;
2812 : 0 : struct buffer_head map_bh;
2813 : 0 : int err;
2814 : :
2815 [ # # ]: 0 : blocksize = i_blocksize(inode);
2816 : 0 : length = offset & (blocksize - 1);
2817 : :
2818 : : /* Block boundary? Nothing to do */
2819 [ # # ]: 0 : if (!length)
2820 : : return 0;
2821 : :
2822 : 0 : length = blocksize - length;
2823 : 0 : iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits);
2824 : :
2825 : 0 : page = grab_cache_page(mapping, index);
2826 : 0 : err = -ENOMEM;
2827 [ # # ]: 0 : if (!page)
2828 : 0 : goto out;
2829 : :
2830 [ # # ]: 0 : if (page_has_buffers(page)) {
2831 : 0 : has_buffers:
2832 : 0 : unlock_page(page);
2833 : 0 : put_page(page);
2834 : 0 : return block_truncate_page(mapping, from, get_block);
2835 : : }
2836 : :
2837 : : /* Find the buffer that contains "offset" */
2838 : : pos = blocksize;
2839 [ # # ]: 0 : while (offset >= pos) {
2840 : 0 : iblock++;
2841 : 0 : pos += blocksize;
2842 : : }
2843 : :
2844 : 0 : map_bh.b_size = blocksize;
2845 : 0 : map_bh.b_state = 0;
2846 : 0 : err = get_block(inode, iblock, &map_bh, 0);
2847 [ # # ]: 0 : if (err)
2848 : 0 : goto unlock;
2849 : : /* unmapped? It's a hole - nothing to do */
2850 [ # # ]: 0 : if (!buffer_mapped(&map_bh))
2851 : 0 : goto unlock;
2852 : :
2853 : : /* Ok, it's mapped. Make sure it's up-to-date */
2854 [ # # ]: 0 : if (!PageUptodate(page)) {
2855 : 0 : err = mapping->a_ops->readpage(NULL, page);
2856 [ # # ]: 0 : if (err) {
2857 : 0 : put_page(page);
2858 : 0 : goto out;
2859 : : }
2860 : 0 : lock_page(page);
2861 [ # # ]: 0 : if (!PageUptodate(page)) {
2862 : 0 : err = -EIO;
2863 : 0 : goto unlock;
2864 : : }
2865 [ # # ]: 0 : if (page_has_buffers(page))
2866 : 0 : goto has_buffers;
2867 : : }
2868 : 0 : zero_user(page, offset, length);
2869 : 0 : set_page_dirty(page);
2870 : 0 : err = 0;
2871 : :
2872 : 0 : unlock:
2873 : 0 : unlock_page(page);
2874 : 0 : put_page(page);
2875 : : out:
2876 : : return err;
2877 : : }
2878 : : EXPORT_SYMBOL(nobh_truncate_page);
2879 : :
2880 : 0 : int block_truncate_page(struct address_space *mapping,
2881 : : loff_t from, get_block_t *get_block)
2882 : : {
2883 : 0 : pgoff_t index = from >> PAGE_SHIFT;
2884 : 0 : unsigned offset = from & (PAGE_SIZE-1);
2885 : 0 : unsigned blocksize;
2886 : 0 : sector_t iblock;
2887 : 0 : unsigned length, pos;
2888 : 0 : struct inode *inode = mapping->host;
2889 : 0 : struct page *page;
2890 : 0 : struct buffer_head *bh;
2891 : 0 : int err;
2892 : :
2893 [ # # ]: 0 : blocksize = i_blocksize(inode);
2894 : 0 : length = offset & (blocksize - 1);
2895 : :
2896 : : /* Block boundary? Nothing to do */
2897 [ # # ]: 0 : if (!length)
2898 : : return 0;
2899 : :
2900 : 0 : length = blocksize - length;
2901 : 0 : iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits);
2902 : :
2903 : 0 : page = grab_cache_page(mapping, index);
2904 : 0 : err = -ENOMEM;
2905 [ # # ]: 0 : if (!page)
2906 : 0 : goto out;
2907 : :
2908 [ # # ]: 0 : if (!page_has_buffers(page))
2909 : 0 : create_empty_buffers(page, blocksize, 0);
2910 : :
2911 : : /* Find the buffer that contains "offset" */
2912 [ # # ]: 0 : bh = page_buffers(page);
2913 : 0 : pos = blocksize;
2914 [ # # ]: 0 : while (offset >= pos) {
2915 : 0 : bh = bh->b_this_page;
2916 : 0 : iblock++;
2917 : 0 : pos += blocksize;
2918 : : }
2919 : :
2920 : 0 : err = 0;
2921 [ # # ]: 0 : if (!buffer_mapped(bh)) {
2922 [ # # ]: 0 : WARN_ON(bh->b_size != blocksize);
2923 : 0 : err = get_block(inode, iblock, bh, 0);
2924 [ # # ]: 0 : if (err)
2925 : 0 : goto unlock;
2926 : : /* unmapped? It's a hole - nothing to do */
2927 [ # # ]: 0 : if (!buffer_mapped(bh))
2928 : 0 : goto unlock;
2929 : : }
2930 : :
2931 : : /* Ok, it's mapped. Make sure it's up-to-date */
2932 [ # # ]: 0 : if (PageUptodate(page))
2933 : 0 : set_buffer_uptodate(bh);
2934 : :
2935 [ # # # # : 0 : if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) {
# # ]
2936 : 0 : err = -EIO;
2937 : 0 : ll_rw_block(REQ_OP_READ, 0, 1, &bh);
2938 : 0 : wait_on_buffer(bh);
2939 : : /* Uhhuh. Read error. Complain and punt. */
2940 [ # # ]: 0 : if (!buffer_uptodate(bh))
2941 : 0 : goto unlock;
2942 : : }
2943 : :
2944 : 0 : zero_user(page, offset, length);
2945 : 0 : mark_buffer_dirty(bh);
2946 : 0 : err = 0;
2947 : :
2948 : 0 : unlock:
2949 : 0 : unlock_page(page);
2950 : 0 : put_page(page);
2951 : : out:
2952 : : return err;
2953 : : }
2954 : : EXPORT_SYMBOL(block_truncate_page);
2955 : :
2956 : : /*
2957 : : * The generic ->writepage function for buffer-backed address_spaces
2958 : : */
2959 : 0 : int block_write_full_page(struct page *page, get_block_t *get_block,
2960 : : struct writeback_control *wbc)
2961 : : {
2962 : 0 : struct inode * const inode = page->mapping->host;
2963 [ # # ]: 0 : loff_t i_size = i_size_read(inode);
2964 : 0 : const pgoff_t end_index = i_size >> PAGE_SHIFT;
2965 : 0 : unsigned offset;
2966 : :
2967 : : /* Is the page fully inside i_size? */
2968 [ # # ]: 0 : if (page->index < end_index)
2969 : 0 : return __block_write_full_page(inode, page, get_block, wbc,
2970 : : end_buffer_async_write);
2971 : :
2972 : : /* Is the page fully outside i_size? (truncate in progress) */
2973 : 0 : offset = i_size & (PAGE_SIZE-1);
2974 [ # # # # ]: 0 : if (page->index >= end_index+1 || !offset) {
2975 : : /*
2976 : : * The page may have dirty, unmapped buffers. For example,
2977 : : * they may have been added in ext3_writepage(). Make them
2978 : : * freeable here, so the page does not leak.
2979 : : */
2980 : 0 : do_invalidatepage(page, 0, PAGE_SIZE);
2981 : 0 : unlock_page(page);
2982 : 0 : return 0; /* don't care */
2983 : : }
2984 : :
2985 : : /*
2986 : : * The page straddles i_size. It must be zeroed out on each and every
2987 : : * writepage invocation because it may be mmapped. "A file is mapped
2988 : : * in multiples of the page size. For a file that is not a multiple of
2989 : : * the page size, the remaining memory is zeroed when mapped, and
2990 : : * writes to that region are not written out to the file."
2991 : : */
2992 : 0 : zero_user_segment(page, offset, PAGE_SIZE);
2993 : 0 : return __block_write_full_page(inode, page, get_block, wbc,
2994 : : end_buffer_async_write);
2995 : : }
2996 : : EXPORT_SYMBOL(block_write_full_page);
2997 : :
2998 : 3455 : sector_t generic_block_bmap(struct address_space *mapping, sector_t block,
2999 : : get_block_t *get_block)
3000 : : {
3001 : 3455 : struct inode *inode = mapping->host;
3002 : 3455 : struct buffer_head tmp = {
3003 : 3455 : .b_size = i_blocksize(inode),
3004 : : };
3005 : :
3006 : 3455 : get_block(inode, block, &tmp, 0);
3007 : 3455 : return tmp.b_blocknr;
3008 : : }
3009 : : EXPORT_SYMBOL(generic_block_bmap);
3010 : :
3011 : 21794 : static void end_bio_bh_io_sync(struct bio *bio)
3012 : : {
3013 : 21794 : struct buffer_head *bh = bio->bi_private;
3014 : :
3015 [ - + ]: 21794 : if (unlikely(bio_flagged(bio, BIO_QUIET)))
3016 : 0 : set_bit(BH_Quiet, &bh->b_state);
3017 : :
3018 : 21794 : bh->b_end_io(bh, !bio->bi_status);
3019 : 21794 : bio_put(bio);
3020 : 21794 : }
3021 : :
3022 : : /*
3023 : : * This allows us to do IO even on the odd last sectors
3024 : : * of a device, even if the block size is some multiple
3025 : : * of the physical sector size.
3026 : : *
3027 : : * We'll just truncate the bio to the size of the device,
3028 : : * and clear the end of the buffer head manually.
3029 : : *
3030 : : * Truly out-of-range accesses will turn into actual IO
3031 : : * errors, this only handles the "we need to be able to
3032 : : * do IO at the final sector" case.
3033 : : */
3034 : 26024 : void guard_bio_eod(struct bio *bio)
3035 : : {
3036 : 26024 : sector_t maxsector;
3037 : 26024 : struct hd_struct *part;
3038 : :
3039 : 26024 : rcu_read_lock();
3040 : 26024 : part = __disk_get_part(bio->bi_disk, bio->bi_partno);
3041 [ + - ]: 26024 : if (part)
3042 : 26024 : maxsector = part_nr_sects_read(part);
3043 : : else
3044 : 0 : maxsector = get_capacity(bio->bi_disk);
3045 : 26024 : rcu_read_unlock();
3046 : :
3047 [ + - ]: 26024 : if (!maxsector)
3048 : : return;
3049 : :
3050 : : /*
3051 : : * If the *whole* IO is past the end of the device,
3052 : : * let it through, and the IO layer will turn it into
3053 : : * an EIO.
3054 : : */
3055 [ + - ]: 26024 : if (unlikely(bio->bi_iter.bi_sector >= maxsector))
3056 : : return;
3057 : :
3058 : 26024 : maxsector -= bio->bi_iter.bi_sector;
3059 [ - + ]: 26024 : if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
3060 : : return;
3061 : :
3062 : 0 : bio_truncate(bio, maxsector << 9);
3063 : : }
3064 : :
3065 : : static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh,
3066 : : enum rw_hint write_hint, struct writeback_control *wbc)
3067 : : {
3068 : : struct bio *bio;
3069 : :
3070 : : BUG_ON(!buffer_locked(bh));
3071 : : BUG_ON(!buffer_mapped(bh));
3072 : : BUG_ON(!bh->b_end_io);
3073 : : BUG_ON(buffer_delay(bh));
3074 : : BUG_ON(buffer_unwritten(bh));
3075 : :
3076 : : /*
3077 : : * Only clear out a write error when rewriting
3078 : : */
3079 : : if (test_set_buffer_req(bh) && (op == REQ_OP_WRITE))
3080 : : clear_buffer_write_io_error(bh);
3081 : :
3082 : : /*
3083 : : * from here on down, it's all bio -- do the initial mapping,
3084 : : * submit_bio -> generic_make_request may further map this bio around
3085 : : */
3086 : : bio = bio_alloc(GFP_NOIO, 1);
3087 : :
3088 : : bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
3089 : : bio_set_dev(bio, bh->b_bdev);
3090 : : bio->bi_write_hint = write_hint;
3091 : :
3092 : : bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
3093 : : BUG_ON(bio->bi_iter.bi_size != bh->b_size);
3094 : :
3095 : : bio->bi_end_io = end_bio_bh_io_sync;
3096 : : bio->bi_private = bh;
3097 : :
3098 : : if (buffer_meta(bh))
3099 : : op_flags |= REQ_META;
3100 : : if (buffer_prio(bh))
3101 : : op_flags |= REQ_PRIO;
3102 : : bio_set_op_attrs(bio, op, op_flags);
3103 : :
3104 : : /* Take care of bh's that straddle the end of the device */
3105 : : guard_bio_eod(bio);
3106 : :
3107 : : if (wbc) {
3108 : : wbc_init_bio(wbc, bio);
3109 : : wbc_account_cgroup_owner(wbc, bh->b_page, bh->b_size);
3110 : : }
3111 : :
3112 : : submit_bio(bio);
3113 : : return 0;
3114 : : }
3115 : :
3116 : 21794 : int submit_bh(int op, int op_flags, struct buffer_head *bh)
3117 : : {
3118 : 4294 : return submit_bh_wbc(op, op_flags, bh, 0, NULL);
3119 : : }
3120 : : EXPORT_SYMBOL(submit_bh);
3121 : :
3122 : : /**
3123 : : * ll_rw_block: low-level access to block devices (DEPRECATED)
3124 : : * @op: whether to %READ or %WRITE
3125 : : * @op_flags: req_flag_bits
3126 : : * @nr: number of &struct buffer_heads in the array
3127 : : * @bhs: array of pointers to &struct buffer_head
3128 : : *
3129 : : * ll_rw_block() takes an array of pointers to &struct buffer_heads, and
3130 : : * requests an I/O operation on them, either a %REQ_OP_READ or a %REQ_OP_WRITE.
3131 : : * @op_flags contains flags modifying the detailed I/O behavior, most notably
3132 : : * %REQ_RAHEAD.
3133 : : *
3134 : : * This function drops any buffer that it cannot get a lock on (with the
3135 : : * BH_Lock state bit), any buffer that appears to be clean when doing a write
3136 : : * request, and any buffer that appears to be up-to-date when doing read
3137 : : * request. Further it marks as clean buffers that are processed for
3138 : : * writing (the buffer cache won't assume that they are actually clean
3139 : : * until the buffer gets unlocked).
3140 : : *
3141 : : * ll_rw_block sets b_end_io to simple completion handler that marks
3142 : : * the buffer up-to-date (if appropriate), unlocks the buffer and wakes
3143 : : * any waiters.
3144 : : *
3145 : : * All of the buffers must be for the same device, and must also be a
3146 : : * multiple of the current approved size for the device.
3147 : : */
3148 : 18190 : void ll_rw_block(int op, int op_flags, int nr, struct buffer_head *bhs[])
3149 : : {
3150 : 18190 : int i;
3151 : :
3152 [ + + ]: 36380 : for (i = 0; i < nr; i++) {
3153 : 18190 : struct buffer_head *bh = bhs[i];
3154 : :
3155 [ + + ]: 18190 : if (!trylock_buffer(bh))
3156 : 511 : continue;
3157 [ - + ]: 17679 : if (op == WRITE) {
3158 [ # # ]: 0 : if (test_clear_buffer_dirty(bh)) {
3159 : 0 : bh->b_end_io = end_buffer_write_sync;
3160 : 0 : get_bh(bh);
3161 : 0 : submit_bh(op, op_flags, bh);
3162 : 0 : continue;
3163 : : }
3164 : : } else {
3165 [ + + ]: 17679 : if (!buffer_uptodate(bh)) {
3166 : 17439 : bh->b_end_io = end_buffer_read_sync;
3167 : 17439 : get_bh(bh);
3168 : 17439 : submit_bh(op, op_flags, bh);
3169 : 17439 : continue;
3170 : : }
3171 : : }
3172 : 240 : unlock_buffer(bh);
3173 : : }
3174 : 18190 : }
3175 : : EXPORT_SYMBOL(ll_rw_block);
3176 : :
3177 : 1 : void write_dirty_buffer(struct buffer_head *bh, int op_flags)
3178 : : {
3179 : 1 : lock_buffer(bh);
3180 [ - + ]: 1 : if (!test_clear_buffer_dirty(bh)) {
3181 : 0 : unlock_buffer(bh);
3182 : 0 : return;
3183 : : }
3184 : 1 : bh->b_end_io = end_buffer_write_sync;
3185 : 1 : get_bh(bh);
3186 : 1 : submit_bh(REQ_OP_WRITE, op_flags, bh);
3187 : : }
3188 : : EXPORT_SYMBOL(write_dirty_buffer);
3189 : :
3190 : : /*
3191 : : * For a data-integrity writeout, we need to wait upon any in-progress I/O
3192 : : * and then start new I/O and then wait upon it. The caller must have a ref on
3193 : : * the buffer_head.
3194 : : */
3195 : 60 : int __sync_dirty_buffer(struct buffer_head *bh, int op_flags)
3196 : : {
3197 : 60 : int ret = 0;
3198 : :
3199 [ - + ]: 60 : WARN_ON(atomic_read(&bh->b_count) < 1);
3200 : 60 : lock_buffer(bh);
3201 [ + - ]: 60 : if (test_clear_buffer_dirty(bh)) {
3202 : 60 : get_bh(bh);
3203 : 60 : bh->b_end_io = end_buffer_write_sync;
3204 : 60 : ret = submit_bh(REQ_OP_WRITE, op_flags, bh);
3205 : 60 : wait_on_buffer(bh);
3206 [ + - - + ]: 120 : if (!ret && !buffer_uptodate(bh))
3207 : 0 : ret = -EIO;
3208 : : } else {
3209 : 0 : unlock_buffer(bh);
3210 : : }
3211 : 60 : return ret;
3212 : : }
3213 : : EXPORT_SYMBOL(__sync_dirty_buffer);
3214 : :
3215 : 0 : int sync_dirty_buffer(struct buffer_head *bh)
3216 : : {
3217 : 0 : return __sync_dirty_buffer(bh, REQ_SYNC);
3218 : : }
3219 : : EXPORT_SYMBOL(sync_dirty_buffer);
3220 : :
3221 : : /*
3222 : : * try_to_free_buffers() checks if all the buffers on this particular page
3223 : : * are unused, and releases them if so.
3224 : : *
3225 : : * Exclusion against try_to_free_buffers may be obtained by either
3226 : : * locking the page or by holding its mapping's private_lock.
3227 : : *
3228 : : * If the page is dirty but all the buffers are clean then we need to
3229 : : * be sure to mark the page clean as well. This is because the page
3230 : : * may be against a block device, and a later reattachment of buffers
3231 : : * to a dirty page will set *all* buffers dirty. Which would corrupt
3232 : : * filesystem data on the same device.
3233 : : *
3234 : : * The same applies to regular filesystem pages: if all the buffers are
3235 : : * clean then we set the page clean and proceed. To do that, we require
3236 : : * total exclusion from __set_page_dirty_buffers(). That is obtained with
3237 : : * private_lock.
3238 : : *
3239 : : * try_to_free_buffers() is non-blocking.
3240 : : */
3241 : 420 : static inline int buffer_busy(struct buffer_head *bh)
3242 : : {
3243 : 840 : return atomic_read(&bh->b_count) |
3244 : 420 : (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock)));
3245 : : }
3246 : :
3247 : : static int
3248 : 180 : drop_buffers(struct page *page, struct buffer_head **buffers_to_free)
3249 : : {
3250 [ - + ]: 180 : struct buffer_head *head = page_buffers(page);
3251 : 180 : struct buffer_head *bh;
3252 : :
3253 : 180 : bh = head;
3254 : 420 : do {
3255 [ + + ]: 420 : if (buffer_busy(bh))
3256 : 60 : goto failed;
3257 : 360 : bh = bh->b_this_page;
3258 [ + + ]: 360 : } while (bh != head);
3259 : :
3260 : 300 : do {
3261 : 300 : struct buffer_head *next = bh->b_this_page;
3262 : :
3263 [ - + ]: 300 : if (bh->b_assoc_map)
3264 [ # # ]: 0 : __remove_assoc_queue(bh);
3265 : 300 : bh = next;
3266 [ + + ]: 300 : } while (bh != head);
3267 : 120 : *buffers_to_free = head;
3268 : 120 : __clear_page_buffers(page);
3269 : 120 : return 1;
3270 : : failed:
3271 : 60 : return 0;
3272 : : }
3273 : :
3274 : 180 : int try_to_free_buffers(struct page *page)
3275 : : {
3276 : 180 : struct address_space * const mapping = page->mapping;
3277 : 180 : struct buffer_head *buffers_to_free = NULL;
3278 : 180 : int ret = 0;
3279 : :
3280 [ - + - + ]: 360 : BUG_ON(!PageLocked(page));
3281 [ - + + - ]: 360 : if (PageWriteback(page))
3282 : : return 0;
3283 : :
3284 [ - + ]: 180 : if (mapping == NULL) { /* can this still happen? */
3285 : 0 : ret = drop_buffers(page, &buffers_to_free);
3286 : 0 : goto out;
3287 : : }
3288 : :
3289 : 180 : spin_lock(&mapping->private_lock);
3290 : 180 : ret = drop_buffers(page, &buffers_to_free);
3291 : :
3292 : : /*
3293 : : * If the filesystem writes its buffers by hand (eg ext3)
3294 : : * then we can have clean buffers against a dirty page. We
3295 : : * clean the page here; otherwise the VM will never notice
3296 : : * that the filesystem did any IO at all.
3297 : : *
3298 : : * Also, during truncate, discard_buffer will have marked all
3299 : : * the page's buffers clean. We discover that here and clean
3300 : : * the page also.
3301 : : *
3302 : : * private_lock must be held over this entire operation in order
3303 : : * to synchronise against __set_page_dirty_buffers and prevent the
3304 : : * dirty bit from being lost.
3305 : : */
3306 [ + + ]: 180 : if (ret)
3307 : 120 : cancel_dirty_page(page);
3308 : 180 : spin_unlock(&mapping->private_lock);
3309 : 180 : out:
3310 [ + + ]: 180 : if (buffers_to_free) {
3311 : : struct buffer_head *bh = buffers_to_free;
3312 : :
3313 : 300 : do {
3314 : 300 : struct buffer_head *next = bh->b_this_page;
3315 : 300 : free_buffer_head(bh);
3316 : 300 : bh = next;
3317 [ + + ]: 300 : } while (bh != buffers_to_free);
3318 : : }
3319 : : return ret;
3320 : : }
3321 : : EXPORT_SYMBOL(try_to_free_buffers);
3322 : :
3323 : : /*
3324 : : * There are no bdflush tunables left. But distributions are
3325 : : * still running obsolete flush daemons, so we terminate them here.
3326 : : *
3327 : : * Use of bdflush() is deprecated and will be removed in a future kernel.
3328 : : * The `flush-X' kernel threads fully replace bdflush daemons and this call.
3329 : : */
3330 : 0 : SYSCALL_DEFINE2(bdflush, int, func, long, data)
3331 : : {
3332 : : static int msg_count;
3333 : :
3334 : : if (!capable(CAP_SYS_ADMIN))
3335 : : return -EPERM;
3336 : :
3337 : : if (msg_count < 5) {
3338 : : msg_count++;
3339 : : printk(KERN_INFO
3340 : : "warning: process `%s' used the obsolete bdflush"
3341 : : " system call\n", current->comm);
3342 : : printk(KERN_INFO "Fix your initscripts?\n");
3343 : : }
3344 : :
3345 : : if (func == 1)
3346 : : do_exit(0);
3347 : : return 0;
3348 : : }
3349 : :
3350 : : /*
3351 : : * Buffer-head allocation
3352 : : */
3353 : : static struct kmem_cache *bh_cachep __read_mostly;
3354 : :
3355 : : /*
3356 : : * Once the number of bh's in the machine exceeds this level, we start
3357 : : * stripping them in writeback.
3358 : : */
3359 : : static unsigned long max_buffer_heads;
3360 : :
3361 : : int buffer_heads_over_limit;
3362 : :
3363 : : struct bh_accounting {
3364 : : int nr; /* Number of live bh's */
3365 : : int ratelimit; /* Limit cacheline bouncing */
3366 : : };
3367 : :
3368 : : static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0};
3369 : :
3370 : 51641 : static void recalc_bh_state(void)
3371 : : {
3372 : 51641 : int i;
3373 : 51641 : int tot = 0;
3374 : :
3375 [ - + ]: 51641 : if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096)
3376 : : return;
3377 : 0 : __this_cpu_write(bh_accounting.ratelimit, 0);
3378 [ # # ]: 0 : for_each_online_cpu(i)
3379 : 0 : tot += per_cpu(bh_accounting, i).nr;
3380 : 0 : buffer_heads_over_limit = (tot > max_buffer_heads);
3381 : : }
3382 : :
3383 : 48386 : struct buffer_head *alloc_buffer_head(gfp_t gfp_flags)
3384 : : {
3385 : 48386 : struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags);
3386 [ + - ]: 48386 : if (ret) {
3387 : 48386 : INIT_LIST_HEAD(&ret->b_assoc_buffers);
3388 : 48386 : preempt_disable();
3389 : 48386 : __this_cpu_inc(bh_accounting.nr);
3390 : 48386 : recalc_bh_state();
3391 : 48386 : preempt_enable();
3392 : : }
3393 : 48386 : return ret;
3394 : : }
3395 : : EXPORT_SYMBOL(alloc_buffer_head);
3396 : :
3397 : 3255 : void free_buffer_head(struct buffer_head *bh)
3398 : : {
3399 [ - + ]: 3255 : BUG_ON(!list_empty(&bh->b_assoc_buffers));
3400 : 3255 : kmem_cache_free(bh_cachep, bh);
3401 : 3255 : preempt_disable();
3402 : 3255 : __this_cpu_dec(bh_accounting.nr);
3403 : 3255 : recalc_bh_state();
3404 : 3255 : preempt_enable();
3405 : 3255 : }
3406 : : EXPORT_SYMBOL(free_buffer_head);
3407 : :
3408 : 0 : static int buffer_exit_cpu_dead(unsigned int cpu)
3409 : : {
3410 : 0 : int i;
3411 : 0 : struct bh_lru *b = &per_cpu(bh_lrus, cpu);
3412 : :
3413 [ # # ]: 0 : for (i = 0; i < BH_LRU_SIZE; i++) {
3414 [ # # ]: 0 : brelse(b->bhs[i]);
3415 : 0 : b->bhs[i] = NULL;
3416 : : }
3417 : 0 : this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr);
3418 : 0 : per_cpu(bh_accounting, cpu).nr = 0;
3419 : 0 : return 0;
3420 : : }
3421 : :
3422 : : /**
3423 : : * bh_uptodate_or_lock - Test whether the buffer is uptodate
3424 : : * @bh: struct buffer_head
3425 : : *
3426 : : * Return true if the buffer is up-to-date and false,
3427 : : * with the buffer locked, if not.
3428 : : */
3429 : 0 : int bh_uptodate_or_lock(struct buffer_head *bh)
3430 : : {
3431 [ # # ]: 0 : if (!buffer_uptodate(bh)) {
3432 : 0 : lock_buffer(bh);
3433 [ # # ]: 0 : if (!buffer_uptodate(bh))
3434 : : return 0;
3435 : 0 : unlock_buffer(bh);
3436 : : }
3437 : : return 1;
3438 : : }
3439 : : EXPORT_SYMBOL(bh_uptodate_or_lock);
3440 : :
3441 : : /**
3442 : : * bh_submit_read - Submit a locked buffer for reading
3443 : : * @bh: struct buffer_head
3444 : : *
3445 : : * Returns zero on success and -EIO on error.
3446 : : */
3447 : 0 : int bh_submit_read(struct buffer_head *bh)
3448 : : {
3449 [ # # ]: 0 : BUG_ON(!buffer_locked(bh));
3450 : :
3451 [ # # ]: 0 : if (buffer_uptodate(bh)) {
3452 : 0 : unlock_buffer(bh);
3453 : 0 : return 0;
3454 : : }
3455 : :
3456 : 0 : get_bh(bh);
3457 : 0 : bh->b_end_io = end_buffer_read_sync;
3458 : 0 : submit_bh(REQ_OP_READ, 0, bh);
3459 : 0 : wait_on_buffer(bh);
3460 [ # # ]: 0 : if (buffer_uptodate(bh))
3461 : 0 : return 0;
3462 : : return -EIO;
3463 : : }
3464 : : EXPORT_SYMBOL(bh_submit_read);
3465 : :
3466 : 30 : void __init buffer_init(void)
3467 : : {
3468 : 30 : unsigned long nrpages;
3469 : 30 : int ret;
3470 : :
3471 : 30 : bh_cachep = kmem_cache_create("buffer_head",
3472 : : sizeof(struct buffer_head), 0,
3473 : : (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
3474 : : SLAB_MEM_SPREAD),
3475 : : NULL);
3476 : :
3477 : : /*
3478 : : * Limit the bh occupancy to 10% of ZONE_NORMAL
3479 : : */
3480 : 30 : nrpages = (nr_free_buffer_pages() * 10) / 100;
3481 : 30 : max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head));
3482 : 30 : ret = cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD, "fs/buffer:dead",
3483 : : NULL, buffer_exit_cpu_dead);
3484 [ - + ]: 30 : WARN_ON(ret < 0);
3485 : 30 : }
|
