~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/buffer.c

Version: ~ [ linux-4.14 ] ~ [ linux-4.13.12 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.61 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.97 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.46 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.80 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.50 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.95 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.27.62 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  *  linux/fs/buffer.c
  3  *
  4  *  Copyright (C) 1991, 1992, 2002  Linus Torvalds
  5  */
  6 
  7 /*
  8  * Start bdflush() with kernel_thread not syscall - Paul Gortmaker, 12/95
  9  *
 10  * Removed a lot of unnecessary code and simplified things now that
 11  * the buffer cache isn't our primary cache - Andrew Tridgell 12/96
 12  *
 13  * Speed up hash, lru, and free list operations.  Use gfp() for allocating
 14  * hash table, use SLAB cache for buffer heads. SMP threading.  -DaveM
 15  *
 16  * Added 32k buffer block sizes - these are required older ARM systems. - RMK
 17  *
 18  * async buffer flushing, 1999 Andrea Arcangeli <andrea@suse.de>
 19  */
 20 
 21 #include <linux/kernel.h>
 22 #include <linux/sched/signal.h>
 23 #include <linux/syscalls.h>
 24 #include <linux/fs.h>
 25 #include <linux/iomap.h>
 26 #include <linux/mm.h>
 27 #include <linux/percpu.h>
 28 #include <linux/slab.h>
 29 #include <linux/capability.h>
 30 #include <linux/blkdev.h>
 31 #include <linux/file.h>
 32 #include <linux/quotaops.h>
 33 #include <linux/highmem.h>
 34 #include <linux/export.h>
 35 #include <linux/backing-dev.h>
 36 #include <linux/writeback.h>
 37 #include <linux/hash.h>
 38 #include <linux/suspend.h>
 39 #include <linux/buffer_head.h>
 40 #include <linux/task_io_accounting_ops.h>
 41 #include <linux/bio.h>
 42 #include <linux/notifier.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 <trace/events/block.h>
 49 
 50 static int fsync_buffers_list(spinlock_t *lock, struct list_head *list);
 51 static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh,
 52                          enum rw_hint hint, struct writeback_control *wbc);
 53 
 54 #define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers)
 55 
 56 void init_buffer(struct buffer_head *bh, bh_end_io_t *handler, void *private)
 57 {
 58         bh->b_end_io = handler;
 59         bh->b_private = private;
 60 }
 61 EXPORT_SYMBOL(init_buffer);
 62 
 63 inline void touch_buffer(struct buffer_head *bh)
 64 {
 65         trace_block_touch_buffer(bh);
 66         mark_page_accessed(bh->b_page);
 67 }
 68 EXPORT_SYMBOL(touch_buffer);
 69 
 70 void __lock_buffer(struct buffer_head *bh)
 71 {
 72         wait_on_bit_lock_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE);
 73 }
 74 EXPORT_SYMBOL(__lock_buffer);
 75 
 76 void unlock_buffer(struct buffer_head *bh)
 77 {
 78         clear_bit_unlock(BH_Lock, &bh->b_state);
 79         smp_mb__after_atomic();
 80         wake_up_bit(&bh->b_state, BH_Lock);
 81 }
 82 EXPORT_SYMBOL(unlock_buffer);
 83 
 84 /*
 85  * Returns if the page has dirty or writeback buffers. If all the buffers
 86  * are unlocked and clean then the PageDirty information is stale. If
 87  * any of the pages are locked, it is assumed they are locked for IO.
 88  */
 89 void buffer_check_dirty_writeback(struct page *page,
 90                                      bool *dirty, bool *writeback)
 91 {
 92         struct buffer_head *head, *bh;
 93         *dirty = false;
 94         *writeback = false;
 95 
 96         BUG_ON(!PageLocked(page));
 97 
 98         if (!page_has_buffers(page))
 99                 return;
100 
101         if (PageWriteback(page))
102                 *writeback = true;
103 
104         head = page_buffers(page);
105         bh = head;
106         do {
107                 if (buffer_locked(bh))
108                         *writeback = true;
109 
110                 if (buffer_dirty(bh))
111                         *dirty = true;
112 
113                 bh = bh->b_this_page;
114         } while (bh != head);
115 }
116 EXPORT_SYMBOL(buffer_check_dirty_writeback);
117 
118 /*
119  * Block until a buffer comes unlocked.  This doesn't stop it
120  * from becoming locked again - you have to lock it yourself
121  * if you want to preserve its state.
122  */
123 void __wait_on_buffer(struct buffer_head * bh)
124 {
125         wait_on_bit_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE);
126 }
127 EXPORT_SYMBOL(__wait_on_buffer);
128 
129 static void
130 __clear_page_buffers(struct page *page)
131 {
132         ClearPagePrivate(page);
133         set_page_private(page, 0);
134         put_page(page);
135 }
136 
137 static void buffer_io_error(struct buffer_head *bh, char *msg)
138 {
139         if (!test_bit(BH_Quiet, &bh->b_state))
140                 printk_ratelimited(KERN_ERR
141                         "Buffer I/O error on dev %pg, logical block %llu%s\n",
142                         bh->b_bdev, (unsigned long long)bh->b_blocknr, msg);
143 }
144 
145 /*
146  * End-of-IO handler helper function which does not touch the bh after
147  * unlocking it.
148  * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but
149  * a race there is benign: unlock_buffer() only use the bh's address for
150  * hashing after unlocking the buffer, so it doesn't actually touch the bh
151  * itself.
152  */
153 static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate)
154 {
155         if (uptodate) {
156                 set_buffer_uptodate(bh);
157         } else {
158                 /* This happens, due to failed read-ahead attempts. */
159                 clear_buffer_uptodate(bh);
160         }
161         unlock_buffer(bh);
162 }
163 
164 /*
165  * Default synchronous end-of-IO handler..  Just mark it up-to-date and
166  * unlock the buffer. This is what ll_rw_block uses too.
167  */
168 void end_buffer_read_sync(struct buffer_head *bh, int uptodate)
169 {
170         __end_buffer_read_notouch(bh, uptodate);
171         put_bh(bh);
172 }
173 EXPORT_SYMBOL(end_buffer_read_sync);
174 
175 void end_buffer_write_sync(struct buffer_head *bh, int uptodate)
176 {
177         if (uptodate) {
178                 set_buffer_uptodate(bh);
179         } else {
180                 buffer_io_error(bh, ", lost sync page write");
181                 mark_buffer_write_io_error(bh);
182                 clear_buffer_uptodate(bh);
183         }
184         unlock_buffer(bh);
185         put_bh(bh);
186 }
187 EXPORT_SYMBOL(end_buffer_write_sync);
188 
189 /*
190  * Various filesystems appear to want __find_get_block to be non-blocking.
191  * But it's the page lock which protects the buffers.  To get around this,
192  * we get exclusion from try_to_free_buffers with the blockdev mapping's
193  * private_lock.
194  *
195  * Hack idea: for the blockdev mapping, i_bufferlist_lock contention
196  * may be quite high.  This code could TryLock the page, and if that
197  * succeeds, there is no need to take private_lock. (But if
198  * private_lock is contended then so is mapping->tree_lock).
199  */
200 static struct buffer_head *
201 __find_get_block_slow(struct block_device *bdev, sector_t block)
202 {
203         struct inode *bd_inode = bdev->bd_inode;
204         struct address_space *bd_mapping = bd_inode->i_mapping;
205         struct buffer_head *ret = NULL;
206         pgoff_t index;
207         struct buffer_head *bh;
208         struct buffer_head *head;
209         struct page *page;
210         int all_mapped = 1;
211 
212         index = block >> (PAGE_SHIFT - bd_inode->i_blkbits);
213         page = find_get_page_flags(bd_mapping, index, FGP_ACCESSED);
214         if (!page)
215                 goto out;
216 
217         spin_lock(&bd_mapping->private_lock);
218         if (!page_has_buffers(page))
219                 goto out_unlock;
220         head = page_buffers(page);
221         bh = head;
222         do {
223                 if (!buffer_mapped(bh))
224                         all_mapped = 0;
225                 else if (bh->b_blocknr == block) {
226                         ret = bh;
227                         get_bh(bh);
228                         goto out_unlock;
229                 }
230                 bh = bh->b_this_page;
231         } while (bh != head);
232 
233         /* we might be here because some of the buffers on this page are
234          * not mapped.  This is due to various races between
235          * file io on the block device and getblk.  It gets dealt with
236          * elsewhere, don't buffer_error if we had some unmapped buffers
237          */
238         if (all_mapped) {
239                 printk("__find_get_block_slow() failed. "
240                         "block=%llu, b_blocknr=%llu\n",
241                         (unsigned long long)block,
242                         (unsigned long long)bh->b_blocknr);
243                 printk("b_state=0x%08lx, b_size=%zu\n",
244                         bh->b_state, bh->b_size);
245                 printk("device %pg blocksize: %d\n", bdev,
246                         1 << bd_inode->i_blkbits);
247         }
248 out_unlock:
249         spin_unlock(&bd_mapping->private_lock);
250         put_page(page);
251 out:
252         return ret;
253 }
254 
255 /*
256  * Kick the writeback threads then try to free up some ZONE_NORMAL memory.
257  */
258 static void free_more_memory(void)
259 {
260         struct zoneref *z;
261         int nid;
262 
263         wakeup_flusher_threads(1024, WB_REASON_FREE_MORE_MEM);
264         yield();
265 
266         for_each_online_node(nid) {
267 
268                 z = first_zones_zonelist(node_zonelist(nid, GFP_NOFS),
269                                                 gfp_zone(GFP_NOFS), NULL);
270                 if (z->zone)
271                         try_to_free_pages(node_zonelist(nid, GFP_NOFS), 0,
272                                                 GFP_NOFS, NULL);
273         }
274 }
275 
276 /*
277  * I/O completion handler for block_read_full_page() - pages
278  * which come unlocked at the end of I/O.
279  */
280 static void end_buffer_async_read(struct buffer_head *bh, int uptodate)
281 {
282         unsigned long flags;
283         struct buffer_head *first;
284         struct buffer_head *tmp;
285         struct page *page;
286         int page_uptodate = 1;
287 
288         BUG_ON(!buffer_async_read(bh));
289 
290         page = bh->b_page;
291         if (uptodate) {
292                 set_buffer_uptodate(bh);
293         } else {
294                 clear_buffer_uptodate(bh);
295                 buffer_io_error(bh, ", async page read");
296                 SetPageError(page);
297         }
298 
299         /*
300          * Be _very_ careful from here on. Bad things can happen if
301          * two buffer heads end IO at almost the same time and both
302          * decide that the page is now completely done.
303          */
304         first = page_buffers(page);
305         local_irq_save(flags);
306         bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
307         clear_buffer_async_read(bh);
308         unlock_buffer(bh);
309         tmp = bh;
310         do {
311                 if (!buffer_uptodate(tmp))
312                         page_uptodate = 0;
313                 if (buffer_async_read(tmp)) {
314                         BUG_ON(!buffer_locked(tmp));
315                         goto still_busy;
316                 }
317                 tmp = tmp->b_this_page;
318         } while (tmp != bh);
319         bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
320         local_irq_restore(flags);
321 
322         /*
323          * If none of the buffers had errors and they are all
324          * uptodate then we can set the page uptodate.
325          */
326         if (page_uptodate && !PageError(page))
327                 SetPageUptodate(page);
328         unlock_page(page);
329         return;
330 
331 still_busy:
332         bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
333         local_irq_restore(flags);
334         return;
335 }
336 
337 /*
338  * Completion handler for block_write_full_page() - pages which are unlocked
339  * during I/O, and which have PageWriteback cleared upon I/O completion.
340  */
341 void end_buffer_async_write(struct buffer_head *bh, int uptodate)
342 {
343         unsigned long flags;
344         struct buffer_head *first;
345         struct buffer_head *tmp;
346         struct page *page;
347 
348         BUG_ON(!buffer_async_write(bh));
349 
350         page = bh->b_page;
351         if (uptodate) {
352                 set_buffer_uptodate(bh);
353         } else {
354                 buffer_io_error(bh, ", lost async page write");
355                 mark_buffer_write_io_error(bh);
356                 clear_buffer_uptodate(bh);
357                 SetPageError(page);
358         }
359 
360         first = page_buffers(page);
361         local_irq_save(flags);
362         bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
363 
364         clear_buffer_async_write(bh);
365         unlock_buffer(bh);
366         tmp = bh->b_this_page;
367         while (tmp != bh) {
368                 if (buffer_async_write(tmp)) {
369                         BUG_ON(!buffer_locked(tmp));
370                         goto still_busy;
371                 }
372                 tmp = tmp->b_this_page;
373         }
374         bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
375         local_irq_restore(flags);
376         end_page_writeback(page);
377         return;
378 
379 still_busy:
380         bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
381         local_irq_restore(flags);
382         return;
383 }
384 EXPORT_SYMBOL(end_buffer_async_write);
385 
386 /*
387  * If a page's buffers are under async readin (end_buffer_async_read
388  * completion) then there is a possibility that another thread of
389  * control could lock one of the buffers after it has completed
390  * but while some of the other buffers have not completed.  This
391  * locked buffer would confuse end_buffer_async_read() into not unlocking
392  * the page.  So the absence of BH_Async_Read tells end_buffer_async_read()
393  * that this buffer is not under async I/O.
394  *
395  * The page comes unlocked when it has no locked buffer_async buffers
396  * left.
397  *
398  * PageLocked prevents anyone starting new async I/O reads any of
399  * the buffers.
400  *
401  * PageWriteback is used to prevent simultaneous writeout of the same
402  * page.
403  *
404  * PageLocked prevents anyone from starting writeback of a page which is
405  * under read I/O (PageWriteback is only ever set against a locked page).
406  */
407 static void mark_buffer_async_read(struct buffer_head *bh)
408 {
409         bh->b_end_io = end_buffer_async_read;
410         set_buffer_async_read(bh);
411 }
412 
413 static void mark_buffer_async_write_endio(struct buffer_head *bh,
414                                           bh_end_io_t *handler)
415 {
416         bh->b_end_io = handler;
417         set_buffer_async_write(bh);
418 }
419 
420 void mark_buffer_async_write(struct buffer_head *bh)
421 {
422         mark_buffer_async_write_endio(bh, end_buffer_async_write);
423 }
424 EXPORT_SYMBOL(mark_buffer_async_write);
425 
426 
427 /*
428  * fs/buffer.c contains helper functions for buffer-backed address space's
429  * fsync functions.  A common requirement for buffer-based filesystems is
430  * that certain data from the backing blockdev needs to be written out for
431  * a successful fsync().  For example, ext2 indirect blocks need to be
432  * written back and waited upon before fsync() returns.
433  *
434  * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(),
435  * inode_has_buffers() and invalidate_inode_buffers() are provided for the
436  * management of a list of dependent buffers at ->i_mapping->private_list.
437  *
438  * Locking is a little subtle: try_to_free_buffers() will remove buffers
439  * from their controlling inode's queue when they are being freed.  But
440  * try_to_free_buffers() will be operating against the *blockdev* mapping
441  * at the time, not against the S_ISREG file which depends on those buffers.
442  * So the locking for private_list is via the private_lock in the address_space
443  * which backs the buffers.  Which is different from the address_space 
444  * against which the buffers are listed.  So for a particular address_space,
445  * mapping->private_lock does *not* protect mapping->private_list!  In fact,
446  * mapping->private_list will always be protected by the backing blockdev's
447  * ->private_lock.
448  *
449  * Which introduces a requirement: all buffers on an address_space's
450  * ->private_list must be from the same address_space: the blockdev's.
451  *
452  * address_spaces which do not place buffers at ->private_list via these
453  * utility functions are free to use private_lock and private_list for
454  * whatever they want.  The only requirement is that list_empty(private_list)
455  * be true at clear_inode() time.
456  *
457  * FIXME: clear_inode should not call invalidate_inode_buffers().  The
458  * filesystems should do that.  invalidate_inode_buffers() should just go
459  * BUG_ON(!list_empty).
460  *
461  * FIXME: mark_buffer_dirty_inode() is a data-plane operation.  It should
462  * take an address_space, not an inode.  And it should be called
463  * mark_buffer_dirty_fsync() to clearly define why those buffers are being
464  * queued up.
465  *
466  * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the
467  * list if it is already on a list.  Because if the buffer is on a list,
468  * it *must* already be on the right one.  If not, the filesystem is being
469  * silly.  This will save a ton of locking.  But first we have to ensure
470  * that buffers are taken *off* the old inode's list when they are freed
471  * (presumably in truncate).  That requires careful auditing of all
472  * filesystems (do it inside bforget()).  It could also be done by bringing
473  * b_inode back.
474  */
475 
476 /*
477  * The buffer's backing address_space's private_lock must be held
478  */
479 static void __remove_assoc_queue(struct buffer_head *bh)
480 {
481         list_del_init(&bh->b_assoc_buffers);
482         WARN_ON(!bh->b_assoc_map);
483         bh->b_assoc_map = NULL;
484 }
485 
486 int inode_has_buffers(struct inode *inode)
487 {
488         return !list_empty(&inode->i_data.private_list);
489 }
490 
491 /*
492  * osync is designed to support O_SYNC io.  It waits synchronously for
493  * all already-submitted IO to complete, but does not queue any new
494  * writes to the disk.
495  *
496  * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as
497  * you dirty the buffers, and then use osync_inode_buffers to wait for
498  * completion.  Any other dirty buffers which are not yet queued for
499  * write will not be flushed to disk by the osync.
500  */
501 static int osync_buffers_list(spinlock_t *lock, struct list_head *list)
502 {
503         struct buffer_head *bh;
504         struct list_head *p;
505         int err = 0;
506 
507         spin_lock(lock);
508 repeat:
509         list_for_each_prev(p, list) {
510                 bh = BH_ENTRY(p);
511                 if (buffer_locked(bh)) {
512                         get_bh(bh);
513                         spin_unlock(lock);
514                         wait_on_buffer(bh);
515                         if (!buffer_uptodate(bh))
516                                 err = -EIO;
517                         brelse(bh);
518                         spin_lock(lock);
519                         goto repeat;
520                 }
521         }
522         spin_unlock(lock);
523         return err;
524 }
525 
526 static void do_thaw_one(struct super_block *sb, void *unused)
527 {
528         while (sb->s_bdev && !thaw_bdev(sb->s_bdev, sb))
529                 printk(KERN_WARNING "Emergency Thaw on %pg\n", sb->s_bdev);
530 }
531 
532 static void do_thaw_all(struct work_struct *work)
533 {
534         iterate_supers(do_thaw_one, NULL);
535         kfree(work);
536         printk(KERN_WARNING "Emergency Thaw complete\n");
537 }
538 
539 /**
540  * emergency_thaw_all -- forcibly thaw every frozen filesystem
541  *
542  * Used for emergency unfreeze of all filesystems via SysRq
543  */
544 void emergency_thaw_all(void)
545 {
546         struct work_struct *work;
547 
548         work = kmalloc(sizeof(*work), GFP_ATOMIC);
549         if (work) {
550                 INIT_WORK(work, do_thaw_all);
551                 schedule_work(work);
552         }
553 }
554 
555 /**
556  * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers
557  * @mapping: the mapping which wants those buffers written
558  *
559  * Starts I/O against the buffers at mapping->private_list, and waits upon
560  * that I/O.
561  *
562  * Basically, this is a convenience function for fsync().
563  * @mapping is a file or directory which needs those buffers to be written for
564  * a successful fsync().
565  */
566 int sync_mapping_buffers(struct address_space *mapping)
567 {
568         struct address_space *buffer_mapping = mapping->private_data;
569 
570         if (buffer_mapping == NULL || list_empty(&mapping->private_list))
571                 return 0;
572 
573         return fsync_buffers_list(&buffer_mapping->private_lock,
574                                         &mapping->private_list);
575 }
576 EXPORT_SYMBOL(sync_mapping_buffers);
577 
578 /*
579  * Called when we've recently written block `bblock', and it is known that
580  * `bblock' was for a buffer_boundary() buffer.  This means that the block at
581  * `bblock + 1' is probably a dirty indirect block.  Hunt it down and, if it's
582  * dirty, schedule it for IO.  So that indirects merge nicely with their data.
583  */
584 void write_boundary_block(struct block_device *bdev,
585                         sector_t bblock, unsigned blocksize)
586 {
587         struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize);
588         if (bh) {
589                 if (buffer_dirty(bh))
590                         ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
591                 put_bh(bh);
592         }
593 }
594 
595 void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode)
596 {
597         struct address_space *mapping = inode->i_mapping;
598         struct address_space *buffer_mapping = bh->b_page->mapping;
599 
600         mark_buffer_dirty(bh);
601         if (!mapping->private_data) {
602                 mapping->private_data = buffer_mapping;
603         } else {
604                 BUG_ON(mapping->private_data != buffer_mapping);
605         }
606         if (!bh->b_assoc_map) {
607                 spin_lock(&buffer_mapping->private_lock);
608                 list_move_tail(&bh->b_assoc_buffers,
609                                 &mapping->private_list);
610                 bh->b_assoc_map = mapping;
611                 spin_unlock(&buffer_mapping->private_lock);
612         }
613 }
614 EXPORT_SYMBOL(mark_buffer_dirty_inode);
615 
616 /*
617  * Mark the page dirty, and set it dirty in the radix tree, and mark the inode
618  * dirty.
619  *
620  * If warn is true, then emit a warning if the page is not uptodate and has
621  * not been truncated.
622  *
623  * The caller must hold lock_page_memcg().
624  */
625 static void __set_page_dirty(struct page *page, struct address_space *mapping,
626                              int warn)
627 {
628         unsigned long flags;
629 
630         spin_lock_irqsave(&mapping->tree_lock, flags);
631         if (page->mapping) {    /* Race with truncate? */
632                 WARN_ON_ONCE(warn && !PageUptodate(page));
633                 account_page_dirtied(page, mapping);
634                 radix_tree_tag_set(&mapping->page_tree,
635                                 page_index(page), PAGECACHE_TAG_DIRTY);
636         }
637         spin_unlock_irqrestore(&mapping->tree_lock, flags);
638 }
639 
640 /*
641  * Add a page to the dirty page list.
642  *
643  * It is a sad fact of life that this function is called from several places
644  * deeply under spinlocking.  It may not sleep.
645  *
646  * If the page has buffers, the uptodate buffers are set dirty, to preserve
647  * dirty-state coherency between the page and the buffers.  It the page does
648  * not have buffers then when they are later attached they will all be set
649  * dirty.
650  *
651  * The buffers are dirtied before the page is dirtied.  There's a small race
652  * window in which a writepage caller may see the page cleanness but not the
653  * buffer dirtiness.  That's fine.  If this code were to set the page dirty
654  * before the buffers, a concurrent writepage caller could clear the page dirty
655  * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean
656  * page on the dirty page list.
657  *
658  * We use private_lock to lock against try_to_free_buffers while using the
659  * page's buffer list.  Also use this to protect against clean buffers being
660  * added to the page after it was set dirty.
661  *
662  * FIXME: may need to call ->reservepage here as well.  That's rather up to the
663  * address_space though.
664  */
665 int __set_page_dirty_buffers(struct page *page)
666 {
667         int newly_dirty;
668         struct address_space *mapping = page_mapping(page);
669 
670         if (unlikely(!mapping))
671                 return !TestSetPageDirty(page);
672 
673         spin_lock(&mapping->private_lock);
674         if (page_has_buffers(page)) {
675                 struct buffer_head *head = page_buffers(page);
676                 struct buffer_head *bh = head;
677 
678                 do {
679                         set_buffer_dirty(bh);
680                         bh = bh->b_this_page;
681                 } while (bh != head);
682         }
683         /*
684          * Lock out page->mem_cgroup migration to keep PageDirty
685          * synchronized with per-memcg dirty page counters.
686          */
687         lock_page_memcg(page);
688         newly_dirty = !TestSetPageDirty(page);
689         spin_unlock(&mapping->private_lock);
690 
691         if (newly_dirty)
692                 __set_page_dirty(page, mapping, 1);
693 
694         unlock_page_memcg(page);
695 
696         if (newly_dirty)
697                 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
698 
699         return newly_dirty;
700 }
701 EXPORT_SYMBOL(__set_page_dirty_buffers);
702 
703 /*
704  * Write out and wait upon a list of buffers.
705  *
706  * We have conflicting pressures: we want to make sure that all
707  * initially dirty buffers get waited on, but that any subsequently
708  * dirtied buffers don't.  After all, we don't want fsync to last
709  * forever if somebody is actively writing to the file.
710  *
711  * Do this in two main stages: first we copy dirty buffers to a
712  * temporary inode list, queueing the writes as we go.  Then we clean
713  * up, waiting for those writes to complete.
714  * 
715  * During this second stage, any subsequent updates to the file may end
716  * up refiling the buffer on the original inode's dirty list again, so
717  * there is a chance we will end up with a buffer queued for write but
718  * not yet completed on that list.  So, as a final cleanup we go through
719  * the osync code to catch these locked, dirty buffers without requeuing
720  * any newly dirty buffers for write.
721  */
722 static int fsync_buffers_list(spinlock_t *lock, struct list_head *list)
723 {
724         struct buffer_head *bh;
725         struct list_head tmp;
726         struct address_space *mapping;
727         int err = 0, err2;
728         struct blk_plug plug;
729 
730         INIT_LIST_HEAD(&tmp);
731         blk_start_plug(&plug);
732 
733         spin_lock(lock);
734         while (!list_empty(list)) {
735                 bh = BH_ENTRY(list->next);
736                 mapping = bh->b_assoc_map;
737                 __remove_assoc_queue(bh);
738                 /* Avoid race with mark_buffer_dirty_inode() which does
739                  * a lockless check and we rely on seeing the dirty bit */
740                 smp_mb();
741                 if (buffer_dirty(bh) || buffer_locked(bh)) {
742                         list_add(&bh->b_assoc_buffers, &tmp);
743                         bh->b_assoc_map = mapping;
744                         if (buffer_dirty(bh)) {
745                                 get_bh(bh);
746                                 spin_unlock(lock);
747                                 /*
748                                  * Ensure any pending I/O completes so that
749                                  * write_dirty_buffer() actually writes the
750                                  * current contents - it is a noop if I/O is
751                                  * still in flight on potentially older
752                                  * contents.
753                                  */
754                                 write_dirty_buffer(bh, REQ_SYNC);
755 
756                                 /*
757                                  * Kick off IO for the previous mapping. Note
758                                  * that we will not run the very last mapping,
759                                  * wait_on_buffer() will do that for us
760                                  * through sync_buffer().
761                                  */
762                                 brelse(bh);
763                                 spin_lock(lock);
764                         }
765                 }
766         }
767 
768         spin_unlock(lock);
769         blk_finish_plug(&plug);
770         spin_lock(lock);
771 
772         while (!list_empty(&tmp)) {
773                 bh = BH_ENTRY(tmp.prev);
774                 get_bh(bh);
775                 mapping = bh->b_assoc_map;
776                 __remove_assoc_queue(bh);
777                 /* Avoid race with mark_buffer_dirty_inode() which does
778                  * a lockless check and we rely on seeing the dirty bit */
779                 smp_mb();
780                 if (buffer_dirty(bh)) {
781                         list_add(&bh->b_assoc_buffers,
782                                  &mapping->private_list);
783                         bh->b_assoc_map = mapping;
784                 }
785                 spin_unlock(lock);
786                 wait_on_buffer(bh);
787                 if (!buffer_uptodate(bh))
788                         err = -EIO;
789                 brelse(bh);
790                 spin_lock(lock);
791         }
792         
793         spin_unlock(lock);
794         err2 = osync_buffers_list(lock, list);
795         if (err)
796                 return err;
797         else
798                 return err2;
799 }
800 
801 /*
802  * Invalidate any and all dirty buffers on a given inode.  We are
803  * probably unmounting the fs, but that doesn't mean we have already
804  * done a sync().  Just drop the buffers from the inode list.
805  *
806  * NOTE: we take the inode's blockdev's mapping's private_lock.  Which
807  * assumes that all the buffers are against the blockdev.  Not true
808  * for reiserfs.
809  */
810 void invalidate_inode_buffers(struct inode *inode)
811 {
812         if (inode_has_buffers(inode)) {
813                 struct address_space *mapping = &inode->i_data;
814                 struct list_head *list = &mapping->private_list;
815                 struct address_space *buffer_mapping = mapping->private_data;
816 
817                 spin_lock(&buffer_mapping->private_lock);
818                 while (!list_empty(list))
819                         __remove_assoc_queue(BH_ENTRY(list->next));
820                 spin_unlock(&buffer_mapping->private_lock);
821         }
822 }
823 EXPORT_SYMBOL(invalidate_inode_buffers);
824 
825 /*
826  * Remove any clean buffers from the inode's buffer list.  This is called
827  * when we're trying to free the inode itself.  Those buffers can pin it.
828  *
829  * Returns true if all buffers were removed.
830  */
831 int remove_inode_buffers(struct inode *inode)
832 {
833         int ret = 1;
834 
835         if (inode_has_buffers(inode)) {
836                 struct address_space *mapping = &inode->i_data;
837                 struct list_head *list = &mapping->private_list;
838                 struct address_space *buffer_mapping = mapping->private_data;
839 
840                 spin_lock(&buffer_mapping->private_lock);
841                 while (!list_empty(list)) {
842                         struct buffer_head *bh = BH_ENTRY(list->next);
843                         if (buffer_dirty(bh)) {
844                                 ret = 0;
845                                 break;
846                         }
847                         __remove_assoc_queue(bh);
848                 }
849                 spin_unlock(&buffer_mapping->private_lock);
850         }
851         return ret;
852 }
853 
854 /*
855  * Create the appropriate buffers when given a page for data area and
856  * the size of each buffer.. Use the bh->b_this_page linked list to
857  * follow the buffers created.  Return NULL if unable to create more
858  * buffers.
859  *
860  * The retry flag is used to differentiate async IO (paging, swapping)
861  * which may not fail from ordinary buffer allocations.
862  */
863 struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size,
864                 int retry)
865 {
866         struct buffer_head *bh, *head;
867         long offset;
868 
869 try_again:
870         head = NULL;
871         offset = PAGE_SIZE;
872         while ((offset -= size) >= 0) {
873                 bh = alloc_buffer_head(GFP_NOFS);
874                 if (!bh)
875                         goto no_grow;
876 
877                 bh->b_this_page = head;
878                 bh->b_blocknr = -1;
879                 head = bh;
880 
881                 bh->b_size = size;
882 
883                 /* Link the buffer to its page */
884                 set_bh_page(bh, page, offset);
885         }
886         return head;
887 /*
888  * In case anything failed, we just free everything we got.
889  */
890 no_grow:
891         if (head) {
892                 do {
893                         bh = head;
894                         head = head->b_this_page;
895                         free_buffer_head(bh);
896                 } while (head);
897         }
898 
899         /*
900          * Return failure for non-async IO requests.  Async IO requests
901          * are not allowed to fail, so we have to wait until buffer heads
902          * become available.  But we don't want tasks sleeping with 
903          * partially complete buffers, so all were released above.
904          */
905         if (!retry)
906                 return NULL;
907 
908         /* We're _really_ low on memory. Now we just
909          * wait for old buffer heads to become free due to
910          * finishing IO.  Since this is an async request and
911          * the reserve list is empty, we're sure there are 
912          * async buffer heads in use.
913          */
914         free_more_memory();
915         goto try_again;
916 }
917 EXPORT_SYMBOL_GPL(alloc_page_buffers);
918 
919 static inline void
920 link_dev_buffers(struct page *page, struct buffer_head *head)
921 {
922         struct buffer_head *bh, *tail;
923 
924         bh = head;
925         do {
926                 tail = bh;
927                 bh = bh->b_this_page;
928         } while (bh);
929         tail->b_this_page = head;
930         attach_page_buffers(page, head);
931 }
932 
933 static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size)
934 {
935         sector_t retval = ~((sector_t)0);
936         loff_t sz = i_size_read(bdev->bd_inode);
937 
938         if (sz) {
939                 unsigned int sizebits = blksize_bits(size);
940                 retval = (sz >> sizebits);
941         }
942         return retval;
943 }
944 
945 /*
946  * Initialise the state of a blockdev page's buffers.
947  */ 
948 static sector_t
949 init_page_buffers(struct page *page, struct block_device *bdev,
950                         sector_t block, int size)
951 {
952         struct buffer_head *head = page_buffers(page);
953         struct buffer_head *bh = head;
954         int uptodate = PageUptodate(page);
955         sector_t end_block = blkdev_max_block(I_BDEV(bdev->bd_inode), size);
956 
957         do {
958                 if (!buffer_mapped(bh)) {
959                         init_buffer(bh, NULL, NULL);
960                         bh->b_bdev = bdev;
961                         bh->b_blocknr = block;
962                         if (uptodate)
963                                 set_buffer_uptodate(bh);
964                         if (block < end_block)
965                                 set_buffer_mapped(bh);
966                 }
967                 block++;
968                 bh = bh->b_this_page;
969         } while (bh != head);
970 
971         /*
972          * Caller needs to validate requested block against end of device.
973          */
974         return end_block;
975 }
976 
977 /*
978  * Create the page-cache page that contains the requested block.
979  *
980  * This is used purely for blockdev mappings.
981  */
982 static int
983 grow_dev_page(struct block_device *bdev, sector_t block,
984               pgoff_t index, int size, int sizebits, gfp_t gfp)
985 {
986         struct inode *inode = bdev->bd_inode;
987         struct page *page;
988         struct buffer_head *bh;
989         sector_t end_block;
990         int ret = 0;            /* Will call free_more_memory() */
991         gfp_t gfp_mask;
992 
993         gfp_mask = mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS) | gfp;
994 
995         /*
996          * XXX: __getblk_slow() can not really deal with failure and
997          * will endlessly loop on improvised global reclaim.  Prefer
998          * looping in the allocator rather than here, at least that
999          * code knows what it's doing.
1000          */
1001         gfp_mask |= __GFP_NOFAIL;
1002 
1003         page = find_or_create_page(inode->i_mapping, index, gfp_mask);
1004         if (!page)
1005                 return ret;
1006 
1007         BUG_ON(!PageLocked(page));
1008 
1009         if (page_has_buffers(page)) {
1010                 bh = page_buffers(page);
1011                 if (bh->b_size == size) {
1012                         end_block = init_page_buffers(page, bdev,
1013                                                 (sector_t)index << sizebits,
1014                                                 size);
1015                         goto done;
1016                 }
1017                 if (!try_to_free_buffers(page))
1018                         goto failed;
1019         }
1020 
1021         /*
1022          * Allocate some buffers for this page
1023          */
1024         bh = alloc_page_buffers(page, size, 0);
1025         if (!bh)
1026                 goto failed;
1027 
1028         /*
1029          * Link the page to the buffers and initialise them.  Take the
1030          * lock to be atomic wrt __find_get_block(), which does not
1031          * run under the page lock.
1032          */
1033         spin_lock(&inode->i_mapping->private_lock);
1034         link_dev_buffers(page, bh);
1035         end_block = init_page_buffers(page, bdev, (sector_t)index << sizebits,
1036                         size);
1037         spin_unlock(&inode->i_mapping->private_lock);
1038 done:
1039         ret = (block < end_block) ? 1 : -ENXIO;
1040 failed:
1041         unlock_page(page);
1042         put_page(page);
1043         return ret;
1044 }
1045 
1046 /*
1047  * Create buffers for the specified block device block's page.  If
1048  * that page was dirty, the buffers are set dirty also.
1049  */
1050 static int
1051 grow_buffers(struct block_device *bdev, sector_t block, int size, gfp_t gfp)
1052 {
1053         pgoff_t index;
1054         int sizebits;
1055 
1056         sizebits = -1;
1057         do {
1058                 sizebits++;
1059         } while ((size << sizebits) < PAGE_SIZE);
1060 
1061         index = block >> sizebits;
1062 
1063         /*
1064          * Check for a block which wants to lie outside our maximum possible
1065          * pagecache index.  (this comparison is done using sector_t types).
1066          */
1067         if (unlikely(index != block >> sizebits)) {
1068                 printk(KERN_ERR "%s: requested out-of-range block %llu for "
1069                         "device %pg\n",
1070                         __func__, (unsigned long long)block,
1071                         bdev);
1072                 return -EIO;
1073         }
1074 
1075         /* Create a page with the proper size buffers.. */
1076         return grow_dev_page(bdev, block, index, size, sizebits, gfp);
1077 }
1078 
1079 static struct buffer_head *
1080 __getblk_slow(struct block_device *bdev, sector_t block,
1081              unsigned size, gfp_t gfp)
1082 {
1083         /* Size must be multiple of hard sectorsize */
1084         if (unlikely(size & (bdev_logical_block_size(bdev)-1) ||
1085                         (size < 512 || size > PAGE_SIZE))) {
1086                 printk(KERN_ERR "getblk(): invalid block size %d requested\n",
1087                                         size);
1088                 printk(KERN_ERR "logical block size: %d\n",
1089                                         bdev_logical_block_size(bdev));
1090 
1091                 dump_stack();
1092                 return NULL;
1093         }
1094 
1095         for (;;) {
1096                 struct buffer_head *bh;
1097                 int ret;
1098 
1099                 bh = __find_get_block(bdev, block, size);
1100                 if (bh)
1101                         return bh;
1102 
1103                 ret = grow_buffers(bdev, block, size, gfp);
1104                 if (ret < 0)
1105                         return NULL;
1106                 if (ret == 0)
1107                         free_more_memory();
1108         }
1109 }
1110 
1111 /*
1112  * The relationship between dirty buffers and dirty pages:
1113  *
1114  * Whenever a page has any dirty buffers, the page's dirty bit is set, and
1115  * the page is tagged dirty in its radix tree.
1116  *
1117  * At all times, the dirtiness of the buffers represents the dirtiness of
1118  * subsections of the page.  If the page has buffers, the page dirty bit is
1119  * merely a hint about the true dirty state.
1120  *
1121  * When a page is set dirty in its entirety, all its buffers are marked dirty
1122  * (if the page has buffers).
1123  *
1124  * When a buffer is marked dirty, its page is dirtied, but the page's other
1125  * buffers are not.
1126  *
1127  * Also.  When blockdev buffers are explicitly read with bread(), they
1128  * individually become uptodate.  But their backing page remains not
1129  * uptodate - even if all of its buffers are uptodate.  A subsequent
1130  * block_read_full_page() against that page will discover all the uptodate
1131  * buffers, will set the page uptodate and will perform no I/O.
1132  */
1133 
1134 /**
1135  * mark_buffer_dirty - mark a buffer_head as needing writeout
1136  * @bh: the buffer_head to mark dirty
1137  *
1138  * mark_buffer_dirty() will set the dirty bit against the buffer, then set its
1139  * backing page dirty, then tag the page as dirty in its address_space's radix
1140  * tree and then attach the address_space's inode to its superblock's dirty
1141  * inode list.
1142  *
1143  * mark_buffer_dirty() is atomic.  It takes bh->b_page->mapping->private_lock,
1144  * mapping->tree_lock and mapping->host->i_lock.
1145  */
1146 void mark_buffer_dirty(struct buffer_head *bh)
1147 {
1148         WARN_ON_ONCE(!buffer_uptodate(bh));
1149 
1150         trace_block_dirty_buffer(bh);
1151 
1152         /*
1153          * Very *carefully* optimize the it-is-already-dirty case.
1154          *
1155          * Don't let the final "is it dirty" escape to before we
1156          * perhaps modified the buffer.
1157          */
1158         if (buffer_dirty(bh)) {
1159                 smp_mb();
1160                 if (buffer_dirty(bh))
1161                         return;
1162         }
1163 
1164         if (!test_set_buffer_dirty(bh)) {
1165                 struct page *page = bh->b_page;
1166                 struct address_space *mapping = NULL;
1167 
1168                 lock_page_memcg(page);
1169                 if (!TestSetPageDirty(page)) {
1170                         mapping = page_mapping(page);
1171                         if (mapping)
1172                                 __set_page_dirty(page, mapping, 0);
1173                 }
1174                 unlock_page_memcg(page);
1175                 if (mapping)
1176                         __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
1177         }
1178 }
1179 EXPORT_SYMBOL(mark_buffer_dirty);
1180 
1181 void mark_buffer_write_io_error(struct buffer_head *bh)
1182 {
1183         set_buffer_write_io_error(bh);
1184         /* FIXME: do we need to set this in both places? */
1185         if (bh->b_page && bh->b_page->mapping)
1186                 mapping_set_error(bh->b_page->mapping, -EIO);
1187         if (bh->b_assoc_map)
1188                 mapping_set_error(bh->b_assoc_map, -EIO);
1189 }
1190 EXPORT_SYMBOL(mark_buffer_write_io_error);
1191 
1192 /*
1193  * Decrement a buffer_head's reference count.  If all buffers against a page
1194  * have zero reference count, are clean and unlocked, and if the page is clean
1195  * and unlocked then try_to_free_buffers() may strip the buffers from the page
1196  * in preparation for freeing it (sometimes, rarely, buffers are removed from
1197  * a page but it ends up not being freed, and buffers may later be reattached).
1198  */
1199 void __brelse(struct buffer_head * buf)
1200 {
1201         if (atomic_read(&buf->b_count)) {
1202                 put_bh(buf);
1203                 return;
1204         }
1205         WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n");
1206 }
1207 EXPORT_SYMBOL(__brelse);
1208 
1209 /*
1210  * bforget() is like brelse(), except it discards any
1211  * potentially dirty data.
1212  */
1213 void __bforget(struct buffer_head *bh)
1214 {
1215         clear_buffer_dirty(bh);
1216         if (bh->b_assoc_map) {
1217                 struct address_space *buffer_mapping = bh->b_page->mapping;
1218 
1219                 spin_lock(&buffer_mapping->private_lock);
1220                 list_del_init(&bh->b_assoc_buffers);
1221                 bh->b_assoc_map = NULL;
1222                 spin_unlock(&buffer_mapping->private_lock);
1223         }
1224         __brelse(bh);
1225 }
1226 EXPORT_SYMBOL(__bforget);
1227 
1228 static struct buffer_head *__bread_slow(struct buffer_head *bh)
1229 {
1230         lock_buffer(bh);
1231         if (buffer_uptodate(bh)) {
1232                 unlock_buffer(bh);
1233                 return bh;
1234         } else {
1235                 get_bh(bh);
1236                 bh->b_end_io = end_buffer_read_sync;
1237                 submit_bh(REQ_OP_READ, 0, bh);
1238                 wait_on_buffer(bh);
1239                 if (buffer_uptodate(bh))
1240                         return bh;
1241         }
1242         brelse(bh);
1243         return NULL;
1244 }
1245 
1246 /*
1247  * Per-cpu buffer LRU implementation.  To reduce the cost of __find_get_block().
1248  * The bhs[] array is sorted - newest buffer is at bhs[0].  Buffers have their
1249  * refcount elevated by one when they're in an LRU.  A buffer can only appear
1250  * once in a particular CPU's LRU.  A single buffer can be present in multiple
1251  * CPU's LRUs at the same time.
1252  *
1253  * This is a transparent caching front-end to sb_bread(), sb_getblk() and
1254  * sb_find_get_block().
1255  *
1256  * The LRUs themselves only need locking against invalidate_bh_lrus.  We use
1257  * a local interrupt disable for that.
1258  */
1259 
1260 #define BH_LRU_SIZE     16
1261 
1262 struct bh_lru {
1263         struct buffer_head *bhs[BH_LRU_SIZE];
1264 };
1265 
1266 static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }};
1267 
1268 #ifdef CONFIG_SMP
1269 #define bh_lru_lock()   local_irq_disable()
1270 #define bh_lru_unlock() local_irq_enable()
1271 #else
1272 #define bh_lru_lock()   preempt_disable()
1273 #define bh_lru_unlock() preempt_enable()
1274 #endif
1275 
1276 static inline void check_irqs_on(void)
1277 {
1278 #ifdef irqs_disabled
1279         BUG_ON(irqs_disabled());
1280 #endif
1281 }
1282 
1283 /*
1284  * Install a buffer_head into this cpu's LRU.  If not already in the LRU, it is
1285  * inserted at the front, and the buffer_head at the back if any is evicted.
1286  * Or, if already in the LRU it is moved to the front.
1287  */
1288 static void bh_lru_install(struct buffer_head *bh)
1289 {
1290         struct buffer_head *evictee = bh;
1291         struct bh_lru *b;
1292         int i;
1293 
1294         check_irqs_on();
1295         bh_lru_lock();
1296 
1297         b = this_cpu_ptr(&bh_lrus);
1298         for (i = 0; i < BH_LRU_SIZE; i++) {
1299                 swap(evictee, b->bhs[i]);
1300                 if (evictee == bh) {
1301                         bh_lru_unlock();
1302                         return;
1303                 }
1304         }
1305 
1306         get_bh(bh);
1307         bh_lru_unlock();
1308         brelse(evictee);
1309 }
1310 
1311 /*
1312  * Look up the bh in this cpu's LRU.  If it's there, move it to the head.
1313  */
1314 static struct buffer_head *
1315 lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size)
1316 {
1317         struct buffer_head *ret = NULL;
1318         unsigned int i;
1319 
1320         check_irqs_on();
1321         bh_lru_lock();
1322         for (i = 0; i < BH_LRU_SIZE; i++) {
1323                 struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]);
1324 
1325                 if (bh && bh->b_blocknr == block && bh->b_bdev == bdev &&
1326                     bh->b_size == size) {
1327                         if (i) {
1328                                 while (i) {
1329                                         __this_cpu_write(bh_lrus.bhs[i],
1330                                                 __this_cpu_read(bh_lrus.bhs[i - 1]));
1331                                         i--;
1332                                 }
1333                                 __this_cpu_write(bh_lrus.bhs[0], bh);
1334                         }
1335                         get_bh(bh);
1336                         ret = bh;
1337                         break;
1338                 }
1339         }
1340         bh_lru_unlock();
1341         return ret;
1342 }
1343 
1344 /*
1345  * Perform a pagecache lookup for the matching buffer.  If it's there, refresh
1346  * it in the LRU and mark it as accessed.  If it is not present then return
1347  * NULL
1348  */
1349 struct buffer_head *
1350 __find_get_block(struct block_device *bdev, sector_t block, unsigned size)
1351 {
1352         struct buffer_head *bh = lookup_bh_lru(bdev, block, size);
1353 
1354         if (bh == NULL) {
1355                 /* __find_get_block_slow will mark the page accessed */
1356                 bh = __find_get_block_slow(bdev, block);
1357                 if (bh)
1358                         bh_lru_install(bh);
1359         } else
1360                 touch_buffer(bh);
1361 
1362         return bh;
1363 }
1364 EXPORT_SYMBOL(__find_get_block);
1365 
1366 /*
1367  * __getblk_gfp() will locate (and, if necessary, create) the buffer_head
1368  * which corresponds to the passed block_device, block and size. The
1369  * returned buffer has its reference count incremented.
1370  *
1371  * __getblk_gfp() will lock up the machine if grow_dev_page's
1372  * try_to_free_buffers() attempt is failing.  FIXME, perhaps?
1373  */
1374 struct buffer_head *
1375 __getblk_gfp(struct block_device *bdev, sector_t block,
1376              unsigned size, gfp_t gfp)
1377 {
1378         struct buffer_head *bh = __find_get_block(bdev, block, size);
1379 
1380         might_sleep();
1381         if (bh == NULL)
1382                 bh = __getblk_slow(bdev, block, size, gfp);
1383         return bh;
1384 }
1385 EXPORT_SYMBOL(__getblk_gfp);
1386 
1387 /*
1388  * Do async read-ahead on a buffer..
1389  */
1390 void __breadahead(struct block_device *bdev, sector_t block, unsigned size)
1391 {
1392         struct buffer_head *bh = __getblk(bdev, block, size);
1393         if (likely(bh)) {
1394                 ll_rw_block(REQ_OP_READ, REQ_RAHEAD, 1, &bh);
1395                 brelse(bh);
1396         }
1397 }
1398 EXPORT_SYMBOL(__breadahead);
1399 
1400 /**
1401  *  __bread_gfp() - reads a specified block and returns the bh
1402  *  @bdev: the block_device to read from
1403  *  @block: number of block
1404  *  @size: size (in bytes) to read
1405  *  @gfp: page allocation flag
1406  *
1407  *  Reads a specified block, and returns buffer head that contains it.
1408  *  The page cache can be allocated from non-movable area
1409  *  not to prevent page migration if you set gfp to zero.
1410  *  It returns NULL if the block was unreadable.
1411  */
1412 struct buffer_head *
1413 __bread_gfp(struct block_device *bdev, sector_t block,
1414                    unsigned size, gfp_t gfp)
1415 {
1416         struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp);
1417 
1418         if (likely(bh) && !buffer_uptodate(bh))
1419                 bh = __bread_slow(bh);
1420         return bh;
1421 }
1422 EXPORT_SYMBOL(__bread_gfp);
1423 
1424 /*
1425  * invalidate_bh_lrus() is called rarely - but not only at unmount.
1426  * This doesn't race because it runs in each cpu either in irq
1427  * or with preempt disabled.
1428  */
1429 static void invalidate_bh_lru(void *arg)
1430 {
1431         struct bh_lru *b = &get_cpu_var(bh_lrus);
1432         int i;
1433 
1434         for (i = 0; i < BH_LRU_SIZE; i++) {
1435                 brelse(b->bhs[i]);
1436                 b->bhs[i] = NULL;
1437         }
1438         put_cpu_var(bh_lrus);
1439 }
1440 
1441 static bool has_bh_in_lru(int cpu, void *dummy)
1442 {
1443         struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu);
1444         int i;
1445         
1446         for (i = 0; i < BH_LRU_SIZE; i++) {
1447                 if (b->bhs[i])
1448                         return 1;
1449         }
1450 
1451         return 0;
1452 }
1453 
1454 void invalidate_bh_lrus(void)
1455 {
1456         on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1, GFP_KERNEL);
1457 }
1458 EXPORT_SYMBOL_GPL(invalidate_bh_lrus);
1459 
1460 void set_bh_page(struct buffer_head *bh,
1461                 struct page *page, unsigned long offset)
1462 {
1463         bh->b_page = page;
1464         BUG_ON(offset >= PAGE_SIZE);
1465         if (PageHighMem(page))
1466                 /*
1467                  * This catches illegal uses and preserves the offset:
1468                  */
1469                 bh->b_data = (char *)(0 + offset);
1470         else
1471                 bh->b_data = page_address(page) + offset;
1472 }
1473 EXPORT_SYMBOL(set_bh_page);
1474 
1475 /*
1476  * Called when truncating a buffer on a page completely.
1477  */
1478 
1479 /* Bits that are cleared during an invalidate */
1480 #define BUFFER_FLAGS_DISCARD \
1481         (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \
1482          1 << BH_Delay | 1 << BH_Unwritten)
1483 
1484 static void discard_buffer(struct buffer_head * bh)
1485 {
1486         unsigned long b_state, b_state_old;
1487 
1488         lock_buffer(bh);
1489         clear_buffer_dirty(bh);
1490         bh->b_bdev = NULL;
1491         b_state = bh->b_state;
1492         for (;;) {
1493                 b_state_old = cmpxchg(&bh->b_state, b_state,
1494                                       (b_state & ~BUFFER_FLAGS_DISCARD));
1495                 if (b_state_old == b_state)
1496                         break;
1497                 b_state = b_state_old;
1498         }
1499         unlock_buffer(bh);
1500 }
1501 
1502 /**
1503  * block_invalidatepage - invalidate part or all of a buffer-backed page
1504  *
1505  * @page: the page which is affected
1506  * @offset: start of the range to invalidate
1507  * @length: length of the range to invalidate
1508  *
1509  * block_invalidatepage() is called when all or part of the page has become
1510  * invalidated by a truncate operation.
1511  *
1512  * block_invalidatepage() does not have to release all buffers, but it must
1513  * ensure that no dirty buffer is left outside @offset and that no I/O
1514  * is underway against any of the blocks which are outside the truncation
1515  * point.  Because the caller is about to free (and possibly reuse) those
1516  * blocks on-disk.
1517  */
1518 void block_invalidatepage(struct page *page, unsigned int offset,
1519                           unsigned int length)
1520 {
1521         struct buffer_head *head, *bh, *next;
1522         unsigned int curr_off = 0;
1523         unsigned int stop = length + offset;
1524 
1525         BUG_ON(!PageLocked(page));
1526         if (!page_has_buffers(page))
1527                 goto out;
1528 
1529         /*
1530          * Check for overflow
1531          */
1532         BUG_ON(stop > PAGE_SIZE || stop < length);
1533 
1534         head = page_buffers(page);
1535         bh = head;
1536         do {
1537                 unsigned int next_off = curr_off + bh->b_size;
1538                 next = bh->b_this_page;
1539 
1540                 /*
1541                  * Are we still fully in range ?
1542                  */
1543                 if (next_off > stop)
1544                         goto out;
1545 
1546                 /*
1547                  * is this block fully invalidated?
1548                  */
1549                 if (offset <= curr_off)
1550                         discard_buffer(bh);
1551                 curr_off = next_off;
1552                 bh = next;
1553         } while (bh != head);
1554 
1555         /*
1556          * We release buffers only if the entire page is being invalidated.
1557          * The get_block cached value has been unconditionally invalidated,
1558          * so real IO is not possible anymore.
1559          */
1560         if (offset == 0)
1561                 try_to_release_page(page, 0);
1562 out:
1563         return;
1564 }
1565 EXPORT_SYMBOL(block_invalidatepage);
1566 
1567 
1568 /*
1569  * We attach and possibly dirty the buffers atomically wrt
1570  * __set_page_dirty_buffers() via private_lock.  try_to_free_buffers
1571  * is already excluded via the page lock.
1572  */
1573 void create_empty_buffers(struct page *page,
1574                         unsigned long blocksize, unsigned long b_state)
1575 {
1576         struct buffer_head *bh, *head, *tail;
1577 
1578         head = alloc_page_buffers(page, blocksize, 1);
1579         bh = head;
1580         do {
1581                 bh->b_state |= b_state;
1582                 tail = bh;
1583                 bh = bh->b_this_page;
1584         } while (bh);
1585         tail->b_this_page = head;
1586 
1587         spin_lock(&page->mapping->private_lock);
1588         if (PageUptodate(page) || PageDirty(page)) {
1589                 bh = head;
1590                 do {
1591                         if (PageDirty(page))
1592                                 set_buffer_dirty(bh);
1593                         if (PageUptodate(page))
1594                                 set_buffer_uptodate(bh);
1595                         bh = bh->b_this_page;
1596                 } while (bh != head);
1597         }
1598         attach_page_buffers(page, head);
1599         spin_unlock(&page->mapping->private_lock);
1600 }
1601 EXPORT_SYMBOL(create_empty_buffers);
1602 
1603 /**
1604  * clean_bdev_aliases: clean a range of buffers in block device
1605  * @bdev: Block device to clean buffers in
1606  * @block: Start of a range of blocks to clean
1607  * @len: Number of blocks to clean
1608  *
1609  * We are taking a range of blocks for data and we don't want writeback of any
1610  * buffer-cache aliases starting from return from this function and until the
1611  * moment when something will explicitly mark the buffer dirty (hopefully that
1612  * will not happen until we will free that block ;-) We don't even need to mark
1613  * it not-uptodate - nobody can expect anything from a newly allocated buffer
1614  * anyway. We used to use unmap_buffer() for such invalidation, but that was
1615  * wrong. We definitely don't want to mark the alias unmapped, for example - it
1616  * would confuse anyone who might pick it with bread() afterwards...
1617  *
1618  * Also..  Note that bforget() doesn't lock the buffer.  So there can be
1619  * writeout I/O going on against recently-freed buffers.  We don't wait on that
1620  * I/O in bforget() - it's more efficient to wait on the I/O only if we really
1621  * need to.  That happens here.
1622  */
1623 void clean_bdev_aliases(struct block_device *bdev, sector_t block, sector_t len)
1624 {
1625         struct inode *bd_inode = bdev->bd_inode;
1626         struct address_space *bd_mapping = bd_inode->i_mapping;
1627         struct pagevec pvec;
1628         pgoff_t index = block >> (PAGE_SHIFT - bd_inode->i_blkbits);
1629         pgoff_t end;
1630         int i, count;
1631         struct buffer_head *bh;
1632         struct buffer_head *head;
1633 
1634         end = (block + len - 1) >> (PAGE_SHIFT - bd_inode->i_blkbits);
1635         pagevec_init(&pvec, 0);
1636         while (pagevec_lookup_range(&pvec, bd_mapping, &index, end)) {
1637                 count = pagevec_count(&pvec);
1638                 for (i = 0; i < count; i++) {
1639                         struct page *page = pvec.pages[i];
1640 
1641                         if (!page_has_buffers(page))
1642                                 continue;
1643                         /*
1644                          * We use page lock instead of bd_mapping->private_lock
1645                          * to pin buffers here since we can afford to sleep and
1646                          * it scales better than a global spinlock lock.
1647                          */
1648                         lock_page(page);
1649                         /* Recheck when the page is locked which pins bhs */
1650                         if (!page_has_buffers(page))
1651                                 goto unlock_page;
1652                         head = page_buffers(page);
1653                         bh = head;
1654                         do {
1655                                 if (!buffer_mapped(bh) || (bh->b_blocknr < block))
1656                                         goto next;
1657                                 if (bh->b_blocknr >= block + len)
1658                                         break;
1659                                 clear_buffer_dirty(bh);
1660                                 wait_on_buffer(bh);
1661                                 clear_buffer_req(bh);
1662 next:
1663                                 bh = bh->b_this_page;
1664                         } while (bh != head);
1665 unlock_page:
1666                         unlock_page(page);
1667                 }
1668                 pagevec_release(&pvec);
1669                 cond_resched();
1670                 /* End of range already reached? */
1671                 if (index > end || !index)
1672                         break;
1673         }
1674 }
1675 EXPORT_SYMBOL(clean_bdev_aliases);
1676 
1677 /*
1678  * Size is a power-of-two in the range 512..PAGE_SIZE,
1679  * and the case we care about most is PAGE_SIZE.
1680  *
1681  * So this *could* possibly be written with those
1682  * constraints in mind (relevant mostly if some
1683  * architecture has a slow bit-scan instruction)
1684  */
1685 static inline int block_size_bits(unsigned int blocksize)
1686 {
1687         return ilog2(blocksize);
1688 }
1689 
1690 static struct buffer_head *create_page_buffers(struct page *page, struct inode *inode, unsigned int b_state)
1691 {
1692         BUG_ON(!PageLocked(page));
1693 
1694         if (!page_has_buffers(page))
1695                 create_empty_buffers(page, 1 << ACCESS_ONCE(inode->i_blkbits), b_state);
1696         return page_buffers(page);
1697 }
1698 
1699 /*
1700  * NOTE! All mapped/uptodate combinations are valid:
1701  *
1702  *      Mapped  Uptodate        Meaning
1703  *
1704  *      No      No              "unknown" - must do get_block()
1705  *      No      Yes             "hole" - zero-filled
1706  *      Yes     No              "allocated" - allocated on disk, not read in
1707  *      Yes     Yes             "valid" - allocated and up-to-date in memory.
1708  *
1709  * "Dirty" is valid only with the last case (mapped+uptodate).
1710  */
1711 
1712 /*
1713  * While block_write_full_page is writing back the dirty buffers under
1714  * the page lock, whoever dirtied the buffers may decide to clean them
1715  * again at any time.  We handle that by only looking at the buffer
1716  * state inside lock_buffer().
1717  *
1718  * If block_write_full_page() is called for regular writeback
1719  * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1720  * locked buffer.   This only can happen if someone has written the buffer
1721  * directly, with submit_bh().  At the address_space level PageWriteback
1722  * prevents this contention from occurring.
1723  *
1724  * If block_write_full_page() is called with wbc->sync_mode ==
1725  * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this
1726  * causes the writes to be flagged as synchronous writes.
1727  */
1728 int __block_write_full_page(struct inode *inode, struct page *page,
1729                         get_block_t *get_block, struct writeback_control *wbc,
1730                         bh_end_io_t *handler)
1731 {
1732         int err;
1733         sector_t block;
1734         sector_t last_block;
1735         struct buffer_head *bh, *head;
1736         unsigned int blocksize, bbits;
1737         int nr_underway = 0;
1738         int write_flags = wbc_to_write_flags(wbc);
1739 
1740         head = create_page_buffers(page, inode,
1741                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
1742 
1743         /*
1744          * Be very careful.  We have no exclusion from __set_page_dirty_buffers
1745          * here, and the (potentially unmapped) buffers may become dirty at
1746          * any time.  If a buffer becomes dirty here after we've inspected it
1747          * then we just miss that fact, and the page stays dirty.
1748          *
1749          * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
1750          * handle that here by just cleaning them.
1751          */
1752 
1753         bh = head;
1754         blocksize = bh->b_size;
1755         bbits = block_size_bits(blocksize);
1756 
1757         block = (sector_t)page->index << (PAGE_SHIFT - bbits);
1758         last_block = (i_size_read(inode) - 1) >> bbits;
1759 
1760         /*
1761          * Get all the dirty buffers mapped to disk addresses and
1762          * handle any aliases from the underlying blockdev's mapping.
1763          */
1764         do {
1765                 if (block > last_block) {
1766                         /*
1767                          * mapped buffers outside i_size will occur, because
1768                          * this page can be outside i_size when there is a
1769                          * truncate in progress.
1770                          */
1771                         /*
1772                          * The buffer was zeroed by block_write_full_page()
1773                          */
1774                         clear_buffer_dirty(bh);
1775                         set_buffer_uptodate(bh);
1776                 } else if ((!buffer_mapped(bh) || buffer_delay(bh)) &&
1777                            buffer_dirty(bh)) {
1778                         WARN_ON(bh->b_size != blocksize);
1779                         err = get_block(inode, block, bh, 1);
1780                         if (err)
1781                                 goto recover;
1782                         clear_buffer_delay(bh);
1783                         if (buffer_new(bh)) {
1784                                 /* blockdev mappings never come here */
1785                                 clear_buffer_new(bh);
1786                                 clean_bdev_bh_alias(bh);
1787                         }
1788                 }
1789                 bh = bh->b_this_page;
1790                 block++;
1791         } while (bh != head);
1792 
1793         do {
1794                 if (!buffer_mapped(bh))
1795                         continue;
1796                 /*
1797                  * If it's a fully non-blocking write attempt and we cannot
1798                  * lock the buffer then redirty the page.  Note that this can
1799                  * potentially cause a busy-wait loop from writeback threads
1800                  * and kswapd activity, but those code paths have their own
1801                  * higher-level throttling.
1802                  */
1803                 if (wbc->sync_mode != WB_SYNC_NONE) {
1804                         lock_buffer(bh);
1805                 } else if (!trylock_buffer(bh)) {
1806                         redirty_page_for_writepage(wbc, page);
1807                         continue;
1808                 }
1809                 if (test_clear_buffer_dirty(bh)) {
1810                         mark_buffer_async_write_endio(bh, handler);
1811                 } else {
1812                         unlock_buffer(bh);
1813                 }
1814         } while ((bh = bh->b_this_page) != head);
1815 
1816         /*
1817          * The page and its buffers are protected by PageWriteback(), so we can
1818          * drop the bh refcounts early.
1819          */
1820         BUG_ON(PageWriteback(page));
1821         set_page_writeback(page);
1822 
1823         do {
1824                 struct buffer_head *next = bh->b_this_page;
1825                 if (buffer_async_write(bh)) {
1826                         submit_bh_wbc(REQ_OP_WRITE, write_flags, bh,
1827                                         inode->i_write_hint, wbc);
1828                         nr_underway++;
1829                 }
1830                 bh = next;
1831         } while (bh != head);
1832         unlock_page(page);
1833 
1834         err = 0;
1835 done:
1836         if (nr_underway == 0) {
1837                 /*
1838                  * The page was marked dirty, but the buffers were
1839                  * clean.  Someone wrote them back by hand with
1840                  * ll_rw_block/submit_bh.  A rare case.
1841                  */
1842                 end_page_writeback(page);
1843 
1844                 /*
1845                  * The page and buffer_heads can be released at any time from
1846                  * here on.
1847                  */
1848         }
1849         return err;
1850 
1851 recover:
1852         /*
1853          * ENOSPC, or some other error.  We may already have added some
1854          * blocks to the file, so we need to write these out to avoid
1855          * exposing stale data.
1856          * The page is currently locked and not marked for writeback
1857          */
1858         bh = head;
1859         /* Recovery: lock and submit the mapped buffers */
1860         do {
1861                 if (buffer_mapped(bh) && buffer_dirty(bh) &&
1862                     !buffer_delay(bh)) {
1863                         lock_buffer(bh);
1864                         mark_buffer_async_write_endio(bh, handler);
1865                 } else {
1866                         /*
1867                          * The buffer may have been set dirty during
1868                          * attachment to a dirty page.
1869                          */
1870                         clear_buffer_dirty(bh);
1871                 }
1872         } while ((bh = bh->b_this_page) != head);
1873         SetPageError(page);
1874         BUG_ON(PageWriteback(page));
1875         mapping_set_error(page->mapping, err);
1876         set_page_writeback(page);
1877         do {
1878                 struct buffer_head *next = bh->b_this_page;
1879                 if (buffer_async_write(bh)) {
1880                         clear_buffer_dirty(bh);
1881                         submit_bh_wbc(REQ_OP_WRITE, write_flags, bh,
1882                                         inode->i_write_hint, wbc);
1883                         nr_underway++;
1884                 }
1885                 bh = next;
1886         } while (bh != head);
1887         unlock_page(page);
1888         goto done;
1889 }
1890 EXPORT_SYMBOL(__block_write_full_page);
1891 
1892 /*
1893  * If a page has any new buffers, zero them out here, and mark them uptodate
1894  * and dirty so they'll be written out (in order to prevent uninitialised
1895  * block data from leaking). And clear the new bit.
1896  */
1897 void page_zero_new_buffers(struct page *page, unsigned from, unsigned to)
1898 {
1899         unsigned int block_start, block_end;
1900         struct buffer_head *head, *bh;
1901 
1902         BUG_ON(!PageLocked(page));
1903         if (!page_has_buffers(page))
1904                 return;
1905 
1906         bh = head = page_buffers(page);
1907         block_start = 0;
1908         do {
1909                 block_end = block_start + bh->b_size;
1910 
1911                 if (buffer_new(bh)) {
1912                         if (block_end > from && block_start < to) {
1913                                 if (!PageUptodate(page)) {
1914                                         unsigned start, size;
1915 
1916                                         start = max(from, block_start);
1917                                         size = min(to, block_end) - start;
1918 
1919                                         zero_user(page, start, size);
1920                                         set_buffer_uptodate(bh);
1921                                 }
1922 
1923                                 clear_buffer_new(bh);
1924                                 mark_buffer_dirty(bh);
1925                         }
1926                 }
1927 
1928                 block_start = block_end;
1929                 bh = bh->b_this_page;
1930         } while (bh != head);
1931 }
1932 EXPORT_SYMBOL(page_zero_new_buffers);
1933 
1934 static void
1935 iomap_to_bh(struct inode *inode, sector_t block, struct buffer_head *bh,
1936                 struct iomap *iomap)
1937 {
1938         loff_t offset = block << inode->i_blkbits;
1939 
1940         bh->b_bdev = iomap->bdev;
1941 
1942         /*
1943          * Block points to offset in file we need to map, iomap contains
1944          * the offset at which the map starts. If the map ends before the
1945          * current block, then do not map the buffer and let the caller
1946          * handle it.
1947          */
1948         BUG_ON(offset >= iomap->offset + iomap->length);
1949 
1950         switch (iomap->type) {
1951         case IOMAP_HOLE:
1952                 /*
1953                  * If the buffer is not up to date or beyond the current EOF,
1954                  * we need to mark it as new to ensure sub-block zeroing is
1955                  * executed if necessary.
1956                  */
1957                 if (!buffer_uptodate(bh) ||
1958                     (offset >= i_size_read(inode)))
1959                         set_buffer_new(bh);
1960                 break;
1961         case IOMAP_DELALLOC:
1962                 if (!buffer_uptodate(bh) ||
1963                     (offset >= i_size_read(inode)))
1964                         set_buffer_new(bh);
1965                 set_buffer_uptodate(bh);
1966                 set_buffer_mapped(bh);
1967                 set_buffer_delay(bh);
1968                 break;
1969         case IOMAP_UNWRITTEN:
1970                 /*
1971                  * For unwritten regions, we always need to ensure that
1972                  * sub-block writes cause the regions in the block we are not
1973                  * writing to are zeroed. Set the buffer as new to ensure this.
1974                  */
1975                 set_buffer_new(bh);
1976                 set_buffer_unwritten(bh);
1977                 /* FALLTHRU */
1978         case IOMAP_MAPPED:
1979                 if (offset >= i_size_read(inode))
1980                         set_buffer_new(bh);
1981                 bh->b_blocknr = (iomap->blkno >> (inode->i_blkbits - 9)) +
1982                                 ((offset - iomap->offset) >> inode->i_blkbits);
1983                 set_buffer_mapped(bh);
1984                 break;
1985         }
1986 }
1987 
1988 int __block_write_begin_int(struct page *page, loff_t pos, unsigned len,
1989                 get_block_t *get_block, struct iomap *iomap)
1990 {
1991         unsigned from = pos & (PAGE_SIZE - 1);
1992         unsigned to = from + len;
1993         struct inode *inode = page->mapping->host;
1994         unsigned block_start, block_end;
1995         sector_t block;
1996         int err = 0;
1997         unsigned blocksize, bbits;
1998         struct buffer_head *bh, *head, *wait[2], **wait_bh=wait;
1999 
2000         BUG_ON(!PageLocked(page));
2001         BUG_ON(from > PAGE_SIZE);
2002         BUG_ON(to > PAGE_SIZE);
2003         BUG_ON(from > to);
2004 
2005         head = create_page_buffers(page, inode, 0);
2006         blocksize = head->b_size;
2007         bbits = block_size_bits(blocksize);
2008 
2009         block = (sector_t)page->index << (PAGE_SHIFT - bbits);
2010 
2011         for(bh = head, block_start = 0; bh != head || !block_start;
2012             block++, block_start=block_end, bh = bh->b_this_page) {
2013                 block_end = block_start + blocksize;
2014                 if (block_end <= from || block_start >= to) {
2015                         if (PageUptodate(page)) {
2016                                 if (!buffer_uptodate(bh))
2017                                         set_buffer_uptodate(bh);
2018                         }
2019                         continue;
2020                 }
2021                 if (buffer_new(bh))
2022                         clear_buffer_new(bh);
2023                 if (!buffer_mapped(bh)) {
2024                         WARN_ON(bh->b_size != blocksize);
2025                         if (get_block) {
2026                                 err = get_block(inode, block, bh, 1);
2027                                 if (err)
2028                                         break;
2029                         } else {
2030                                 iomap_to_bh(inode, block, bh, iomap);
2031                         }
2032 
2033                         if (buffer_new(bh)) {
2034                                 clean_bdev_bh_alias(bh);
2035                                 if (PageUptodate(page)) {
2036                                         clear_buffer_new(bh);
2037                                         set_buffer_uptodate(bh);
2038                                         mark_buffer_dirty(bh);
2039                                         continue;
2040                                 }
2041                                 if (block_end > to || block_start < from)
2042                                         zero_user_segments(page,
2043                                                 to, block_end,
2044                                                 block_start, from);
2045                                 continue;
2046                         }
2047                 }
2048                 if (PageUptodate(page)) {
2049                         if (!buffer_uptodate(bh))
2050                                 set_buffer_uptodate(bh);
2051                         continue; 
2052                 }
2053                 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
2054                     !buffer_unwritten(bh) &&
2055                      (block_start < from || block_end > to)) {
2056                         ll_rw_block(REQ_OP_READ, 0, 1, &bh);
2057                         *wait_bh++=bh;
2058                 }
2059         }
2060         /*
2061          * If we issued read requests - let them complete.
2062          */
2063         while(wait_bh > wait) {
2064                 wait_on_buffer(*--wait_bh);
2065                 if (!buffer_uptodate(*wait_bh))
2066                         err = -EIO;
2067         }
2068         if (unlikely(err))
2069                 page_zero_new_buffers(page, from, to);
2070         return err;
2071 }
2072 
2073 int __block_write_begin(struct page *page, loff_t pos, unsigned len,
2074                 get_block_t *get_block)
2075 {
2076         return __block_write_begin_int(page, pos, len, get_block, NULL);
2077 }
2078 EXPORT_SYMBOL(__block_write_begin);
2079 
2080 static int __block_commit_write(struct inode *inode, struct page *page,
2081                 unsigned from, unsigned to)
2082 {
2083         unsigned block_start, block_end;
2084         int partial = 0;
2085         unsigned blocksize;
2086         struct buffer_head *bh, *head;
2087 
2088         bh = head = page_buffers(page);
2089         blocksize = bh->b_size;
2090 
2091         block_start = 0;
2092         do {
2093                 block_end = block_start + blocksize;
2094                 if (block_end <= from || block_start >= to) {
2095                         if (!buffer_uptodate(bh))
2096                                 partial = 1;
2097                 } else {
2098                         set_buffer_uptodate(bh);
2099                         mark_buffer_dirty(bh);
2100                 }
2101                 clear_buffer_new(bh);
2102 
2103                 block_start = block_end;
2104                 bh = bh->b_this_page;
2105         } while (bh != head);
2106 
2107         /*
2108          * If this is a partial write which happened to make all buffers
2109          * uptodate then we can optimize away a bogus readpage() for
2110          * the next read(). Here we 'discover' whether the page went
2111          * uptodate as a result of this (potentially partial) write.
2112          */
2113         if (!partial)
2114                 SetPageUptodate(page);
2115         return 0;
2116 }
2117 
2118 /*
2119  * block_write_begin takes care of the basic task of block allocation and
2120  * bringing partial write blocks uptodate first.
2121  *
2122  * The filesystem needs to handle block truncation upon failure.
2123  */
2124 int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,
2125                 unsigned flags, struct page **pagep, get_block_t *get_block)
2126 {
2127         pgoff_t index = pos >> PAGE_SHIFT;
2128         struct page *page;
2129         int status;
2130 
2131         page = grab_cache_page_write_begin(mapping, index, flags);
2132         if (!page)
2133                 return -ENOMEM;
2134 
2135         status = __block_write_begin(page, pos, len, get_block);
2136         if (unlikely(status)) {
2137                 unlock_page(page);
2138                 put_page(page);
2139                 page = NULL;
2140         }
2141 
2142         *pagep = page;
2143         return status;
2144 }
2145 EXPORT_SYMBOL(block_write_begin);
2146 
2147 int block_write_end(struct file *file, struct address_space *mapping,
2148                         loff_t pos, unsigned len, unsigned copied,
2149                         struct page *page, void *fsdata)
2150 {
2151         struct inode *inode = mapping->host;
2152         unsigned start;
2153 
2154         start = pos & (PAGE_SIZE - 1);
2155 
2156         if (unlikely(copied < len)) {
2157                 /*
2158                  * The buffers that were written will now be uptodate, so we
2159                  * don't have to worry about a readpage reading them and
2160                  * overwriting a partial write. However if we have encountered
2161                  * a short write and only partially written into a buffer, it
2162                  * will not be marked uptodate, so a readpage might come in and
2163                  * destroy our partial write.
2164                  *
2165                  * Do the simplest thing, and just treat any short write to a
2166                  * non uptodate page as a zero-length write, and force the
2167                  * caller to redo the whole thing.
2168                  */
2169                 if (!PageUptodate(page))
2170                         copied = 0;
2171 
2172                 page_zero_new_buffers(page, start+copied, start+len);
2173         }
2174         flush_dcache_page(page);
2175 
2176         /* This could be a short (even 0-length) commit */
2177         __block_commit_write(inode, page, start, start+copied);
2178 
2179         return copied;
2180 }
2181 EXPORT_SYMBOL(block_write_end);
2182 
2183 int generic_write_end(struct file *file, struct address_space *mapping,
2184                         loff_t pos, unsigned len, unsigned copied,
2185                         struct page *page, void *fsdata)
2186 {
2187         struct inode *inode = mapping->host;
2188         loff_t old_size = inode->i_size;
2189         int i_size_changed = 0;
2190 
2191         copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
2192 
2193         /*
2194          * No need to use i_size_read() here, the i_size
2195          * cannot change under us because we hold i_mutex.
2196          *
2197          * But it's important to update i_size while still holding page lock:
2198          * page writeout could otherwise come in and zero beyond i_size.
2199          */
2200         if (pos+copied > inode->i_size) {
2201                 i_size_write(inode, pos+copied);
2202                 i_size_changed = 1;
2203         }
2204 
2205         unlock_page(page);
2206         put_page(page);
2207 
2208         if (old_size < pos)
2209                 pagecache_isize_extended(inode, old_size, pos);
2210         /*
2211          * Don't mark the inode dirty under page lock. First, it unnecessarily
2212          * makes the holding time of page lock longer. Second, it forces lock
2213          * ordering of page lock and transaction start for journaling
2214          * filesystems.
2215          */
2216         if (i_size_changed)
2217                 mark_inode_dirty(inode);
2218 
2219         return copied;
2220 }
2221 EXPORT_SYMBOL(generic_write_end);
2222 
2223 /*
2224  * block_is_partially_uptodate checks whether buffers within a page are
2225  * uptodate or not.
2226  *
2227  * Returns true if all buffers which correspond to a file portion
2228  * we want to read are uptodate.
2229  */
2230 int block_is_partially_uptodate(struct page *page, unsigned long from,
2231                                         unsigned long count)
2232 {
2233         unsigned block_start, block_end, blocksize;
2234         unsigned to;
2235         struct buffer_head *bh, *head;
2236         int ret = 1;
2237 
2238         if (!page_has_buffers(page))
2239                 return 0;
2240 
2241         head = page_buffers(page);
2242         blocksize = head->b_size;
2243         to = min_t(unsigned, PAGE_SIZE - from, count);
2244         to = from + to;
2245         if (from < blocksize && to > PAGE_SIZE - blocksize)
2246                 return 0;
2247 
2248         bh = head;
2249         block_start = 0;
2250         do {
2251                 block_end = block_start + blocksize;
2252                 if (block_end > from && block_start < to) {
2253                         if (!buffer_uptodate(bh)) {
2254                                 ret = 0;
2255                                 break;
2256                         }
2257                         if (block_end >= to)
2258                                 break;
2259                 }
2260                 block_start = block_end;
2261                 bh = bh->b_this_page;
2262         } while (bh != head);
2263 
2264         return ret;
2265 }
2266 EXPORT_SYMBOL(block_is_partially_uptodate);
2267 
2268 /*
2269  * Generic "read page" function for block devices that have the normal
2270  * get_block functionality. This is most of the block device filesystems.
2271  * Reads the page asynchronously --- the unlock_buffer() and
2272  * set/clear_buffer_uptodate() functions propagate buffer state into the
2273  * page struct once IO has completed.
2274  */
2275 int block_read_full_page(struct page *page, get_block_t *get_block)
2276 {
2277         struct inode *inode = page->mapping->host;
2278         sector_t iblock, lblock;
2279         struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
2280         unsigned int blocksize, bbits;
2281         int nr, i;
2282         int fully_mapped = 1;
2283 
2284         head = create_page_buffers(page, inode, 0);
2285         blocksize = head->b_size;
2286         bbits = block_size_bits(blocksize);
2287 
2288         iblock = (sector_t)page->index << (PAGE_SHIFT - bbits);
2289         lblock = (i_size_read(inode)+blocksize-1) >> bbits;
2290         bh = head;
2291         nr = 0;
2292         i = 0;
2293 
2294         do {
2295                 if (buffer_uptodate(bh))
2296                         continue;
2297 
2298                 if (!buffer_mapped(bh)) {
2299                         int err = 0;
2300 
2301                         fully_mapped = 0;
2302                         if (iblock < lblock) {
2303                                 WARN_ON(bh->b_size != blocksize);
2304                                 err = get_block(inode, iblock, bh, 0);
2305                                 if (err)
2306                                         SetPageError(page);
2307                         }
2308                         if (!buffer_mapped(bh)) {
2309                                 zero_user(page, i * blocksize, blocksize);
2310                                 if (!err)
2311                                         set_buffer_uptodate(bh);
2312                                 continue;
2313                         }
2314                         /*
2315                          * get_block() might have updated the buffer
2316                          * synchronously
2317                          */
2318                         if (buffer_uptodate(bh))
2319                                 continue;
2320                 }
2321                 arr[nr++] = bh;
2322         } while (i++, iblock++, (bh = bh->b_this_page) != head);
2323 
2324         if (fully_mapped)
2325                 SetPageMappedToDisk(page);
2326 
2327         if (!nr) {
2328                 /*
2329                  * All buffers are uptodate - we can set the page uptodate
2330                  * as well. But not if get_block() returned an error.
2331                  */
2332                 if (!PageError(page))
2333                         SetPageUptodate(page);
2334                 unlock_page(page);
2335                 return 0;
2336         }
2337 
2338         /* Stage two: lock the buffers */
2339         for (i = 0; i < nr; i++) {
2340                 bh = arr[i];
2341                 lock_buffer(bh);
2342                 mark_buffer_async_read(bh);
2343         }
2344 
2345         /*
2346          * Stage 3: start the IO.  Check for uptodateness
2347          * inside the buffer lock in case another process reading
2348          * the underlying blockdev brought it uptodate (the sct fix).
2349          */
2350         for (i = 0; i < nr; i++) {
2351                 bh = arr[i];
2352                 if (buffer_uptodate(bh))
2353                         end_buffer_async_read(bh, 1);
2354                 else
2355                         submit_bh(REQ_OP_READ, 0, bh);
2356         }
2357         return 0;
2358 }
2359 EXPORT_SYMBOL(block_read_full_page);
2360 
2361 /* utility function for filesystems that need to do work on expanding
2362  * truncates.  Uses filesystem pagecache writes to allow the filesystem to
2363  * deal with the hole.  
2364  */
2365 int generic_cont_expand_simple(struct inode *inode, loff_t size)
2366 {
2367         struct address_space *mapping = inode->i_mapping;
2368         struct page *page;
2369         void *fsdata;
2370         int err;
2371 
2372         err = inode_newsize_ok(inode, size);
2373         if (err)
2374                 goto out;
2375 
2376         err = pagecache_write_begin(NULL, mapping, size, 0,
2377                                     AOP_FLAG_CONT_EXPAND, &page, &fsdata);
2378         if (err)
2379                 goto out;
2380 
2381         err = pagecache_write_end(NULL, mapping, size, 0, 0, page, fsdata);
2382         BUG_ON(err > 0);
2383 
2384 out:
2385         return err;
2386 }
2387 EXPORT_SYMBOL(generic_cont_expand_simple);
2388 
2389 static int cont_expand_zero(struct file *file, struct address_space *mapping,
2390                             loff_t pos, loff_t *bytes)
2391 {
2392         struct inode *inode = mapping->host;
2393         unsigned int blocksize = i_blocksize(inode);
2394         struct page *page;
2395         void *fsdata;
2396         pgoff_t index, curidx;
2397         loff_t curpos;
2398         unsigned zerofrom, offset, len;
2399         int err = 0;
2400 
2401         index = pos >> PAGE_SHIFT;
2402         offset = pos & ~PAGE_MASK;
2403 
2404         while (index > (curidx = (curpos = *bytes)>>PAGE_SHIFT)) {
2405                 zerofrom = curpos & ~PAGE_MASK;
2406                 if (zerofrom & (blocksize-1)) {
2407                         *bytes |= (blocksize-1);
2408                         (*bytes)++;
2409                 }
2410                 len = PAGE_SIZE - zerofrom;
2411 
2412                 err = pagecache_write_begin(file, mapping, curpos, len, 0,
2413                                             &page, &fsdata);
2414                 if (err)
2415                         goto out;
2416                 zero_user(page, zerofrom, len);
2417                 err = pagecache_write_end(file, mapping, curpos, len, len,
2418                                                 page, fsdata);
2419                 if (err < 0)
2420                         goto out;
2421                 BUG_ON(err != len);
2422                 err = 0;
2423 
2424                 balance_dirty_pages_ratelimited(mapping);
2425 
2426                 if (unlikely(fatal_signal_pending(current))) {
2427                         err = -EINTR;
2428                         goto out;
2429                 }
2430         }
2431 
2432         /* page covers the boundary, find the boundary offset */
2433         if (index == curidx) {
2434                 zerofrom = curpos & ~PAGE_MASK;
2435                 /* if we will expand the thing last block will be filled */
2436                 if (offset <= zerofrom) {
2437                         goto out;
2438                 }
2439                 if (zerofrom & (blocksize-1)) {
2440                         *bytes |= (blocksize-1);
2441                         (*bytes)++;
2442                 }
2443                 len = offset - zerofrom;
2444 
2445                 err = pagecache_write_begin(file, mapping, curpos, len, 0,
2446                                             &page, &fsdata);
2447                 if (err)
2448                         goto out;
2449                 zero_user(page, zerofrom, len);
2450                 err = pagecache_write_end(file, mapping, curpos, len, len,
2451                                                 page, fsdata);
2452                 if (err < 0)
2453                         goto out;
2454                 BUG_ON(err != len);
2455                 err = 0;
2456         }
2457 out:
2458         return err;
2459 }
2460 
2461 /*
2462  * For moronic filesystems that do not allow holes in file.
2463  * We may have to extend the file.
2464  */
2465 int cont_write_begin(struct file *file, struct address_space *mapping,
2466                         loff_t pos, unsigned len, unsigned flags,
2467                         struct page **pagep, void **fsdata,
2468                         get_block_t *get_block, loff_t *bytes)
2469 {
2470         struct inode *inode = mapping->host;
2471         unsigned int blocksize = i_blocksize(inode);
2472         unsigned int zerofrom;
2473         int err;
2474 
2475         err = cont_expand_zero(file, mapping, pos, bytes);
2476         if (err)
2477                 return err;
2478 
2479         zerofrom = *bytes & ~PAGE_MASK;
2480         if (pos+len > *bytes && zerofrom & (blocksize-1)) {
2481                 *bytes |= (blocksize-1);
2482                 (*bytes)++;
2483         }
2484 
2485         return block_write_begin(mapping, pos, len, flags, pagep, get_block);
2486 }
2487 EXPORT_SYMBOL(cont_write_begin);
2488 
2489 int block_commit_write(struct page *page, unsigned from, unsigned to)
2490 {
2491         struct inode *inode = page->mapping->host;
2492         __block_commit_write(inode,page,from,to);
2493         return 0;
2494 }
2495 EXPORT_SYMBOL(block_commit_write);
2496 
2497 /*
2498  * block_page_mkwrite() is not allowed to change the file size as it gets
2499  * called from a page fault handler when a page is first dirtied. Hence we must
2500  * be careful to check for EOF conditions here. We set the page up correctly
2501  * for a written page which means we get ENOSPC checking when writing into
2502  * holes and correct delalloc and unwritten extent mapping on filesystems that
2503  * support these features.
2504  *
2505  * We are not allowed to take the i_mutex here so we have to play games to
2506  * protect against truncate races as the page could now be beyond EOF.  Because
2507  * truncate writes the inode size before removing pages, once we have the
2508  * page lock we can determine safely if the page is beyond EOF. If it is not
2509  * beyond EOF, then the page is guaranteed safe against truncation until we
2510  * unlock the page.
2511  *
2512  * Direct callers of this function should protect against filesystem freezing
2513  * using sb_start_pagefault() - sb_end_pagefault() functions.
2514  */
2515 int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
2516                          get_block_t get_block)
2517 {
2518         struct page *page = vmf->page;
2519         struct inode *inode = file_inode(vma->vm_file);
2520         unsigned long end;
2521         loff_t size;
2522         int ret;
2523 
2524         lock_page(page);
2525         size = i_size_read(inode);
2526         if ((page->mapping != inode->i_mapping) ||
2527             (page_offset(page) > size)) {
2528                 /* We overload EFAULT to mean page got truncated */
2529                 ret = -EFAULT;
2530                 goto out_unlock;
2531         }
2532 
2533         /* page is wholly or partially inside EOF */
2534         if (((page->index + 1) << PAGE_SHIFT) > size)
2535                 end = size & ~PAGE_MASK;
2536         else
2537                 end = PAGE_SIZE;
2538 
2539         ret = __block_write_begin(page, 0, end, get_block);
2540         if (!ret)
2541                 ret = block_commit_write(page, 0, end);
2542 
2543         if (unlikely(ret < 0))
2544                 goto out_unlock;
2545         set_page_dirty(page);
2546         wait_for_stable_page(page);
2547         return 0;
2548 out_unlock:
2549         unlock_page(page);
2550         return ret;
2551 }
2552 EXPORT_SYMBOL(block_page_mkwrite);
2553 
2554 /*
2555  * nobh_write_begin()'s prereads are special: the buffer_heads are freed
2556  * immediately, while under the page lock.  So it needs a special end_io
2557  * handler which does not touch the bh after unlocking it.
2558  */
2559 static void end_buffer_read_nobh(struct buffer_head *bh, int uptodate)
2560 {
2561         __end_buffer_read_notouch(bh, uptodate);
2562 }
2563 
2564 /*
2565  * Attach the singly-linked list of buffers created by nobh_write_begin, to
2566  * the page (converting it to circular linked list and taking care of page
2567  * dirty races).
2568  */
2569 static void attach_nobh_buffers(struct page *page, struct buffer_head *head)
2570 {
2571         struct buffer_head *bh;
2572 
2573         BUG_ON(!PageLocked(page));
2574 
2575         spin_lock(&page->mapping->private_lock);
2576         bh = head;
2577         do {
2578                 if (PageDirty(page))
2579                         set_buffer_dirty(bh);
2580                 if (!bh->b_this_page)
2581                         bh->b_this_page = head;
2582                 bh = bh->b_this_page;
2583         } while (bh != head);
2584         attach_page_buffers(page, head);
2585         spin_unlock(&page->mapping->private_lock);
2586 }
2587 
2588 /*
2589  * On entry, the page is fully not uptodate.
2590  * On exit the page is fully uptodate in the areas outside (from,to)
2591  * The filesystem needs to handle block truncation upon failure.
2592  */
2593 int nobh_write_begin(struct address_space *mapping,
2594                         loff_t pos, unsigned len, unsigned flags,
2595                         struct page **pagep, void **fsdata,
2596                         get_block_t *get_block)
2597 {
2598         struct inode *inode = mapping->host;
2599         const unsigned blkbits = inode->i_blkbits;
2600         const unsigned blocksize = 1 << blkbits;
2601         struct buffer_head *head, *bh;
2602         struct page *page;
2603         pgoff_t index;
2604         unsigned from, to;
2605         unsigned block_in_page;
2606         unsigned block_start, block_end;
2607         sector_t block_in_file;
2608         int nr_reads = 0;
2609         int ret = 0;
2610         int is_mapped_to_disk = 1;
2611 
2612         index = pos >> PAGE_SHIFT;
2613         from = pos & (PAGE_SIZE - 1);
2614         to = from + len;
2615 
2616         page = grab_cache_page_write_begin(mapping, index, flags);
2617         if (!page)
2618                 return -ENOMEM;
2619         *pagep = page;
2620         *fsdata = NULL;
2621 
2622         if (page_has_buffers(page)) {
2623                 ret = __block_write_begin(page, pos, len, get_block);
2624                 if (unlikely(ret))
2625                         goto out_release;
2626                 return ret;
2627         }
2628 
2629         if (PageMappedToDisk(page))
2630                 return 0;
2631 
2632         /*
2633          * Allocate buffers so that we can keep track of state, and potentially
2634          * attach them to the page if an error occurs. In the common case of
2635          * no error, they will just be freed again without ever being attached
2636          * to the page (which is all OK, because we're under the page lock).
2637          *
2638          * Be careful: the buffer linked list is a NULL terminated one, rather
2639          * than the circular one we're used to.
2640          */
2641         head = alloc_page_buffers(page, blocksize, 0);
2642         if (!head) {
2643                 ret = -ENOMEM;
2644                 goto out_release;
2645         }
2646 
2647         block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
2648 
2649         /*
2650          * We loop across all blocks in the page, whether or not they are
2651          * part of the affected region.  This is so we can discover if the
2652          * page is fully mapped-to-disk.
2653          */
2654         for (block_start = 0, block_in_page = 0, bh = head;
2655                   block_start < PAGE_SIZE;
2656                   block_in_page++, block_start += blocksize, bh = bh->b_this_page) {
2657                 int create;
2658 
2659                 block_end = block_start + blocksize;
2660                 bh->b_state = 0;
2661                 create = 1;
2662                 if (block_start >= to)
2663                         create = 0;
2664                 ret = get_block(inode, block_in_file + block_in_page,
2665                                         bh, create);
2666                 if (ret)
2667                         goto failed;
2668                 if (!buffer_mapped(bh))
2669                         is_mapped_to_disk = 0;
2670                 if (buffer_new(bh))
2671                         clean_bdev_bh_alias(bh);
2672                 if (PageUptodate(page)) {
2673                         set_buffer_uptodate(bh);
2674                         continue;
2675                 }
2676                 if (buffer_new(bh) || !buffer_mapped(bh)) {
2677                         zero_user_segments(page, block_start, from,
2678                                                         to, block_end);
2679                         continue;
2680                 }
2681                 if (buffer_uptodate(bh))
2682                         continue;       /* reiserfs does this */
2683                 if (block_start < from || block_end > to) {
2684                         lock_buffer(bh);
2685                         bh->b_end_io = end_buffer_read_nobh;
2686                         submit_bh(REQ_OP_READ, 0, bh);
2687                         nr_reads++;
2688                 }
2689         }
2690 
2691         if (nr_reads) {
2692                 /*
2693                  * The page is locked, so these buffers are protected from
2694                  * any VM or truncate activity.  Hence we don't need to care
2695                  * for the buffer_head refcounts.
2696                  */
2697                 for (bh = head; bh; bh = bh->b_this_page) {
2698                         wait_on_buffer(bh);
2699                         if (!buffer_uptodate(bh))
2700                                 ret = -EIO;
2701                 }
2702                 if (ret)
2703                         goto failed;
2704         }
2705 
2706         if (is_mapped_to_disk)
2707                 SetPageMappedToDisk(page);
2708 
2709         *fsdata = head; /* to be released by nobh_write_end */
2710 
2711         return 0;
2712 
2713 failed:
2714         BUG_ON(!ret);
2715         /*
2716          * Error recovery is a bit difficult. We need to zero out blocks that
2717          * were newly allocated, and dirty them to ensure they get written out.
2718          * Buffers need to be attached to the page at this point, otherwise
2719          * the handling of potential IO errors during writeout would be hard
2720          * (could try doing synchronous writeout, but what if that fails too?)
2721          */
2722         attach_nobh_buffers(page, head);
2723         page_zero_new_buffers(page, from, to);
2724 
2725 out_release:
2726         unlock_page(page);
2727         put_page(page);
2728         *pagep = NULL;
2729 
2730         return ret;
2731 }
2732 EXPORT_SYMBOL(nobh_write_begin);
2733 
2734 int nobh_write_end(struct file *file, struct address_space *mapping,
2735                         loff_t pos, unsigned len, unsigned copied,
2736                         struct page *page, void *fsdata)
2737 {
2738         struct inode *inode = page->mapping->host;
2739         struct buffer_head *head = fsdata;
2740         struct buffer_head *bh;
2741         BUG_ON(fsdata != NULL && page_has_buffers(page));
2742 
2743         if (unlikely(copied < len) && head)
2744                 attach_nobh_buffers(page, head);
2745         if (page_has_buffers(page))
2746                 return generic_write_end(file, mapping, pos, len,
2747                                         copied, page, fsdata);
2748 
2749         SetPageUptodate(page);
2750         set_page_dirty(page);
2751         if (pos+copied > inode->i_size) {
2752                 i_size_write(inode, pos+copied);
2753                 mark_inode_dirty(inode);
2754         }
2755 
2756         unlock_page(page);
2757         put_page(page);
2758 
2759         while (head) {
2760                 bh = head;
2761                 head = head->b_this_page;
2762                 free_buffer_head(bh);
2763         }
2764 
2765         return copied;
2766 }
2767 EXPORT_SYMBOL(nobh_write_end);
2768 
2769 /*
2770  * nobh_writepage() - based on block_full_write_page() except
2771  * that it tries to operate without attaching bufferheads to
2772  * the page.
2773  */
2774 int nobh_writepage(struct page *page, get_block_t *get_block,
2775                         struct writeback_control *wbc)
2776 {
2777         struct inode * const inode = page->mapping->host;
2778         loff_t i_size = i_size_read(inode);
2779         const pgoff_t end_index = i_size >> PAGE_SHIFT;
2780         unsigned offset;
2781         int ret;
2782 
2783         /* Is the page fully inside i_size? */
2784         if (page->index < end_index)
2785                 goto out;
2786 
2787         /* Is the page fully outside i_size? (truncate in progress) */
2788         offset = i_size & (PAGE_SIZE-1);
2789         if (page->index >= end_index+1 || !offset) {
2790                 /*
2791                  * The page may have dirty, unmapped buffers.  For example,
2792                  * they may have been added in ext3_writepage().  Make them
2793                  * freeable here, so the page does not leak.
2794                  */
2795 #if 0
2796                 /* Not really sure about this  - do we need this ? */
2797                 if (page->mapping->a_ops->invalidatepage)
2798                         page->mapping->a_ops->invalidatepage(page, offset);
2799 #endif
2800                 unlock_page(page);
2801                 return 0; /* don't care */
2802         }
2803 
2804         /*
2805          * The page straddles i_size.  It must be zeroed out on each and every
2806          * writepage invocation because it may be mmapped.  "A file is mapped
2807          * in multiples of the page size.  For a file that is not a multiple of
2808          * the  page size, the remaining memory is zeroed when mapped, and
2809          * writes to that region are not written out to the file."
2810          */
2811         zero_user_segment(page, offset, PAGE_SIZE);
2812 out:
2813         ret = mpage_writepage(page, get_block, wbc);
2814         if (ret == -EAGAIN)
2815                 ret = __block_write_full_page(inode, page, get_block, wbc,
2816                                               end_buffer_async_write);
2817         return ret;
2818 }
2819 EXPORT_SYMBOL(nobh_writepage);
2820 
2821 int nobh_truncate_page(struct address_space *mapping,
2822                         loff_t from, get_block_t *get_block)
2823 {
2824         pgoff_t index = from >> PAGE_SHIFT;
2825         unsigned offset = from & (PAGE_SIZE-1);
2826         unsigned blocksize;
2827         sector_t iblock;
2828         unsigned length, pos;
2829         struct inode *inode = mapping->host;
2830         struct page *page;
2831         struct buffer_head map_bh;
2832         int err;
2833 
2834         blocksize = i_blocksize(inode);
2835         length = offset & (blocksize - 1);
2836 
2837         /* Block boundary? Nothing to do */
2838         if (!length)
2839                 return 0;
2840 
2841         length = blocksize - length;
2842         iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits);
2843 
2844         page = grab_cache_page(mapping, index);
2845         err = -ENOMEM;
2846         if (!page)
2847                 goto out;
2848 
2849         if (page_has_buffers(page)) {
2850 has_buffers:
2851                 unlock_page(page);
2852                 put_page(page);
2853                 return block_truncate_page(mapping, from, get_block);
2854         }
2855 
2856         /* Find the buffer that contains "offset" */
2857         pos = blocksize;
2858         while (offset >= pos) {
2859                 iblock++;
2860                 pos += blocksize;
2861         }
2862 
2863         map_bh.b_size = blocksize;
2864         map_bh.b_state = 0;
2865         err = get_block(inode, iblock, &map_bh, 0);
2866         if (err)
2867                 goto unlock;
2868         /* unmapped? It's a hole - nothing to do */
2869         if (!buffer_mapped(&map_bh))
2870                 goto unlock;
2871 
2872         /* Ok, it's mapped. Make sure it's up-to-date */
2873         if (!PageUptodate(page)) {
2874                 err = mapping->a_ops->readpage(NULL, page);
2875                 if (err) {
2876                         put_page(page);
2877                         goto out;
2878                 }
2879                 lock_page(page);
2880                 if (!PageUptodate(page)) {
2881                         err = -EIO;
2882                         goto unlock;
2883                 }
2884                 if (page_has_buffers(page))
2885                         goto has_buffers;
2886         }
2887         zero_user(page, offset, length);
2888         set_page_dirty(page);
2889         err = 0;
2890 
2891 unlock:
2892         unlock_page(page);
2893         put_page(page);
2894 out:
2895         return err;
2896 }
2897 EXPORT_SYMBOL(nobh_truncate_page);
2898 
2899 int block_truncate_page(struct address_space *mapping,
2900                         loff_t from, get_block_t *get_block)
2901 {
2902         pgoff_t index = from >> PAGE_SHIFT;
2903         unsigned offset = from & (PAGE_SIZE-1);
2904         unsigned blocksize;
2905         sector_t iblock;
2906         unsigned length, pos;
2907         struct inode *inode = mapping->host;
2908         struct page *page;
2909         struct buffer_head *bh;
2910         int err;
2911 
2912         blocksize = i_blocksize(inode);
2913         length = offset & (blocksize - 1);
2914 
2915         /* Block boundary? Nothing to do */
2916         if (!length)
2917                 return 0;
2918 
2919         length = blocksize - length;
2920         iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits);
2921         
2922         page = grab_cache_page(mapping, index);
2923         err = -ENOMEM;
2924         if (!page)
2925                 goto out;
2926 
2927         if (!page_has_buffers(page))
2928                 create_empty_buffers(page, blocksize, 0);
2929 
2930         /* Find the buffer that contains "offset" */
2931         bh = page_buffers(page);
2932         pos = blocksize;
2933         while (offset >= pos) {
2934                 bh = bh->b_this_page;
2935                 iblock++;
2936                 pos += blocksize;
2937         }
2938 
2939         err = 0;
2940         if (!buffer_mapped(bh)) {
2941                 WARN_ON(bh->b_size != blocksize);
2942                 err = get_block(inode, iblock, bh, 0);
2943                 if (err)
2944                         goto unlock;
2945                 /* unmapped? It's a hole - nothing to do */
2946                 if (!buffer_mapped(bh))
2947                         goto unlock;
2948         }
2949 
2950         /* Ok, it's mapped. Make sure it's up-to-date */
2951         if (PageUptodate(page))
2952                 set_buffer_uptodate(bh);
2953 
2954         if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) {
2955                 err = -EIO;
2956                 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
2957                 wait_on_buffer(bh);
2958                 /* Uhhuh. Read error. Complain and punt. */
2959                 if (!buffer_uptodate(bh))
2960                         goto unlock;
2961         }
2962 
2963         zero_user(page, offset, length);
2964         mark_buffer_dirty(bh);
2965         err = 0;
2966 
2967 unlock:
2968         unlock_page(page);
2969         put_page(page);
2970 out:
2971         return err;
2972 }
2973 EXPORT_SYMBOL(block_truncate_page);
2974 
2975 /*
2976  * The generic ->writepage function for buffer-backed address_spaces
2977  */
2978 int block_write_full_page(struct page *page, get_block_t *get_block,
2979                         struct writeback_control *wbc)
2980 {
2981         struct inode * const inode = page->mapping->host;
2982         loff_t i_size = i_size_read(inode);
2983         const pgoff_t end_index = i_size >> PAGE_SHIFT;
2984         unsigned offset;
2985 
2986         /* Is the page fully inside i_size? */
2987         if (page->index < end_index)
2988                 return __block_write_full_page(inode, page, get_block, wbc,
2989                                                end_buffer_async_write);
2990 
2991         /* Is the page fully outside i_size? (truncate in progress) */
2992         offset = i_size & (PAGE_SIZE-1);
2993         if (page->index >= end_index+1 || !offset) {
2994                 /*
2995                  * The page may have dirty, unmapped buffers.  For example,
2996                  * they may have been added in ext3_writepage().  Make them
2997                  * freeable here, so the page does not leak.
2998                  */
2999                 do_invalidatepage(page, 0, PAGE_SIZE);
3000                 unlock_page(page);
3001                 return 0; /* don't care */
3002         }
3003 
3004         /*
3005          * The page straddles i_size.  It must be zeroed out on each and every
3006          * writepage invocation because it may be mmapped.  "A file is mapped
3007          * in multiples of the page size.  For a file that is not a multiple of
3008          * the  page size, the remaining memory is zeroed when mapped, and
3009          * writes to that region are not written out to the file."
3010          */
3011         zero_user_segment(page, offset, PAGE_SIZE);
3012         return __block_write_full_page(inode, page, get_block, wbc,
3013                                                         end_buffer_async_write);
3014 }
3015 EXPORT_SYMBOL(block_write_full_page);
3016 
3017 sector_t generic_block_bmap(struct address_space *mapping, sector_t block,
3018                             get_block_t *get_block)
3019 {
3020         struct inode *inode = mapping->host;
3021         struct buffer_head tmp = {
3022                 .b_size = i_blocksize(inode),
3023         };
3024 
3025         get_block(inode, block, &tmp, 0);
3026         return tmp.b_blocknr;
3027 }
3028 EXPORT_SYMBOL(generic_block_bmap);
3029 
3030 static void end_bio_bh_io_sync(struct bio *bio)
3031 {
3032         struct buffer_head *bh = bio->bi_private;
3033 
3034         if (unlikely(bio_flagged(bio, BIO_QUIET)))
3035                 set_bit(BH_Quiet, &bh->b_state);
3036 
3037         bh->b_end_io(bh, !bio->bi_status);
3038         bio_put(bio);
3039 }
3040 
3041 /*
3042  * This allows us to do IO even on the odd last sectors
3043  * of a device, even if the block size is some multiple
3044  * of the physical sector size.
3045  *
3046  * We'll just truncate the bio to the size of the device,
3047  * and clear the end of the buffer head manually.
3048  *
3049  * Truly out-of-range accesses will turn into actual IO
3050  * errors, this only handles the "we need to be able to
3051  * do IO at the final sector" case.
3052  */
3053 void guard_bio_eod(int op, struct bio *bio)
3054 {
3055         sector_t maxsector;
3056         struct bio_vec *bvec = &bio->bi_io_vec[bio->bi_vcnt - 1];
3057         unsigned truncated_bytes;
3058 
3059         maxsector = get_capacity(bio->bi_disk);
3060         if (!maxsector)
3061                 return;
3062 
3063         /*
3064          * If the *whole* IO is past the end of the device,
3065          * let it through, and the IO layer will turn it into
3066          * an EIO.
3067          */
3068         if (unlikely(bio->bi_iter.bi_sector >= maxsector))
3069                 return;
3070 
3071         maxsector -= bio->bi_iter.bi_sector;
3072         if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
3073                 return;
3074 
3075         /* Uhhuh. We've got a bio that straddles the device size! */
3076         truncated_bytes = bio->bi_iter.bi_size - (maxsector << 9);
3077 
3078         /* Truncate the bio.. */
3079         bio->bi_iter.bi_size -= truncated_bytes;
3080         bvec->bv_len -= truncated_bytes;
3081 
3082         /* ..and clear the end of the buffer for reads */
3083         if (op == REQ_OP_READ) {
3084                 zero_user(bvec->bv_page, bvec->bv_offset + bvec->bv_len,
3085                                 truncated_bytes);
3086         }
3087 }
3088 
3089 static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh,
3090                          enum rw_hint write_hint, struct writeback_control *wbc)
3091 {
3092         struct bio *bio;
3093 
3094         BUG_ON(!buffer_locked(bh));
3095         BUG_ON(!buffer_mapped(bh));
3096         BUG_ON(!bh->b_end_io);
3097         BUG_ON(buffer_delay(bh));
3098         BUG_ON(buffer_unwritten(bh));
3099 
3100         /*
3101          * Only clear out a write error when rewriting
3102          */
3103         if (test_set_buffer_req(bh) && (op == REQ_OP_WRITE))
3104                 clear_buffer_write_io_error(bh);
3105 
3106         /*
3107          * from here on down, it's all bio -- do the initial mapping,
3108          * submit_bio -> generic_make_request may further map this bio around
3109          */
3110         bio = bio_alloc(GFP_NOIO, 1);
3111 
3112         if (wbc) {
3113                 wbc_init_bio(wbc, bio);
3114                 wbc_account_io(wbc, bh->b_page, bh->b_size);
3115         }
3116 
3117         bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
3118         bio_set_dev(bio, bh->b_bdev);
3119         bio->bi_write_hint = write_hint;
3120 
3121         bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
3122         BUG_ON(bio->bi_iter.bi_size != bh->b_size);
3123 
3124         bio->bi_end_io = end_bio_bh_io_sync;
3125         bio->bi_private = bh;
3126 
3127         /* Take care of bh's that straddle the end of the device */
3128         guard_bio_eod(op, bio);
3129 
3130         if (buffer_meta(bh))
3131                 op_flags |= REQ_META;
3132         if (buffer_prio(bh))
3133                 op_flags |= REQ_PRIO;
3134         bio_set_op_attrs(bio, op, op_flags);
3135 
3136         submit_bio(bio);
3137         return 0;
3138 }
3139 
3140 int submit_bh(int op, int op_flags, struct buffer_head *bh)
3141 {
3142         return submit_bh_wbc(op, op_flags, bh, 0, NULL);
3143 }
3144 EXPORT_SYMBOL(submit_bh);
3145 
3146 /**
3147  * ll_rw_block: low-level access to block devices (DEPRECATED)
3148  * @op: whether to %READ or %WRITE
3149  * @op_flags: req_flag_bits
3150  * @nr: number of &struct buffer_heads in the array
3151  * @bhs: array of pointers to &struct buffer_head
3152  *
3153  * ll_rw_block() takes an array of pointers to &struct buffer_heads, and
3154  * requests an I/O operation on them, either a %REQ_OP_READ or a %REQ_OP_WRITE.
3155  * @op_flags contains flags modifying the detailed I/O behavior, most notably
3156  * %REQ_RAHEAD.
3157  *
3158  * This function drops any buffer that it cannot get a lock on (with the
3159  * BH_Lock state bit), any buffer that appears to be clean when doing a write
3160  * request, and any buffer that appears to be up-to-date when doing read
3161  * request.  Further it marks as clean buffers that are processed for
3162  * writing (the buffer cache won't assume that they are actually clean
3163  * until the buffer gets unlocked).
3164  *
3165  * ll_rw_block sets b_end_io to simple completion handler that marks
3166  * the buffer up-to-date (if appropriate), unlocks the buffer and wakes
3167  * any waiters. 
3168  *
3169  * All of the buffers must be for the same device, and must also be a
3170  * multiple of the current approved size for the device.
3171  */
3172 void ll_rw_block(int op, int op_flags,  int nr, struct buffer_head *bhs[])
3173 {
3174         int i;
3175 
3176         for (i = 0; i < nr; i++) {
3177                 struct buffer_head *bh = bhs[i];
3178 
3179                 if (!trylock_buffer(bh))
3180                         continue;
3181                 if (op == WRITE) {
3182                         if (test_clear_buffer_dirty(bh)) {
3183                                 bh->b_end_io = end_buffer_write_sync;
3184                                 get_bh(bh);
3185                                 submit_bh(op, op_flags, bh);
3186                                 continue;
3187                         }
3188                 } else {
3189                         if (!buffer_uptodate(bh)) {
3190                                 bh->b_end_io = end_buffer_read_sync;
3191                                 get_bh(bh);
3192                                 submit_bh(op, op_flags, bh);
3193                                 continue;
3194                         }
3195                 }
3196                 unlock_buffer(bh);
3197         }
3198 }
3199 EXPORT_SYMBOL(ll_rw_block);
3200 
3201 void write_dirty_buffer(struct buffer_head *bh, int op_flags)
3202 {
3203         lock_buffer(bh);
3204         if (!test_clear_buffer_dirty(bh)) {
3205                 unlock_buffer(bh);
3206                 return;
3207         }
3208         bh->b_end_io = end_buffer_write_sync;
3209         get_bh(bh);
3210         submit_bh(REQ_OP_WRITE, op_flags, bh);
3211 }
3212 EXPORT_SYMBOL(write_dirty_buffer);
3213 
3214 /*
3215  * For a data-integrity writeout, we need to wait upon any in-progress I/O
3216  * and then start new I/O and then wait upon it.  The caller must have a ref on
3217  * the buffer_head.
3218  */
3219 int __sync_dirty_buffer(struct buffer_head *bh, int op_flags)
3220 {
3221         int ret = 0;
3222 
3223         WARN_ON(atomic_read(&bh->b_count) < 1);
3224         lock_buffer(bh);
3225         if (test_clear_buffer_dirty(bh)) {
3226                 get_bh(bh);
3227                 bh->b_end_io = end_buffer_write_sync;
3228                 ret = submit_bh(REQ_OP_WRITE, op_flags, bh);
3229                 wait_on_buffer(bh);
3230                 if (!ret && !buffer_uptodate(bh))
3231                         ret = -EIO;
3232         } else {
3233                 unlock_buffer(bh);
3234         }
3235         return ret;
3236 }
3237 EXPORT_SYMBOL(__sync_dirty_buffer);
3238 
3239 int sync_dirty_buffer(struct buffer_head *bh)
3240 {
3241         return __sync_dirty_buffer(bh, REQ_SYNC);
3242 }
3243 EXPORT_SYMBOL(sync_dirty_buffer);
3244 
3245 /*
3246  * try_to_free_buffers() checks if all the buffers on this particular page
3247  * are unused, and releases them if so.
3248  *
3249  * Exclusion against try_to_free_buffers may be obtained by either
3250  * locking the page or by holding its mapping's private_lock.
3251  *
3252  * If the page is dirty but all the buffers are clean then we need to
3253  * be sure to mark the page clean as well.  This is because the page
3254  * may be against a block device, and a later reattachment of buffers
3255  * to a dirty page will set *all* buffers dirty.  Which would corrupt
3256  * filesystem data on the same device.
3257  *
3258  * The same applies to regular filesystem pages: if all the buffers are
3259  * clean then we set the page clean and proceed.  To do that, we require
3260  * total exclusion from __set_page_dirty_buffers().  That is obtained with
3261  * private_lock.
3262  *
3263  * try_to_free_buffers() is non-blocking.
3264  */
3265 static inline int buffer_busy(struct buffer_head *bh)
3266 {
3267         return atomic_read(&bh->b_count) |
3268                 (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock)));
3269 }
3270 
3271 static int
3272 drop_buffers(struct page *page, struct buffer_head **buffers_to_free)
3273 {
3274         struct buffer_head *head = page_buffers(page);
3275         struct buffer_head *bh;
3276 
3277         bh = head;
3278         do {
3279                 if (buffer_busy(bh))
3280                         goto failed;
3281                 bh = bh->b_this_page;
3282         } while (bh != head);
3283 
3284         do {
3285                 struct buffer_head *next = bh->b_this_page;
3286 
3287                 if (bh->b_assoc_map)
3288                         __remove_assoc_queue(bh);
3289                 bh = next;
3290         } while (bh != head);
3291         *buffers_to_free = head;
3292         __clear_page_buffers(page);
3293         return 1;
3294 failed:
3295         return 0;
3296 }
3297 
3298 int try_to_free_buffers(struct page *page)
3299 {
3300         struct address_space * const mapping = page->mapping;
3301         struct buffer_head *buffers_to_free = NULL;
3302         int ret = 0;
3303 
3304         BUG_ON(!PageLocked(page));
3305         if (PageWriteback(page))
3306                 return 0;
3307 
3308         if (mapping == NULL) {          /* can this still happen? */
3309                 ret = drop_buffers(page, &buffers_to_free);
3310                 goto out;
3311         }
3312 
3313         spin_lock(&mapping->private_lock);
3314         ret = drop_buffers(page, &buffers_to_free);
3315 
3316         /*
3317          * If the filesystem writes its buffers by hand (eg ext3)
3318          * then we can have clean buffers against a dirty page.  We
3319          * clean the page here; otherwise the VM will never notice
3320          * that the filesystem did any IO at all.
3321          *
3322          * Also, during truncate, discard_buffer will have marked all
3323          * the page's buffers clean.  We discover that here and clean
3324          * the page also.
3325          *
3326          * private_lock must be held over this entire operation in order
3327          * to synchronise against __set_page_dirty_buffers and prevent the
3328          * dirty bit from being lost.
3329          */
3330         if (ret)
3331                 cancel_dirty_page(page);
3332         spin_unlock(&mapping->private_lock);
3333 out:
3334         if (buffers_to_free) {
3335                 struct buffer_head *bh = buffers_to_free;
3336 
3337                 do {
3338                         struct buffer_head *next = bh->b_this_page;
3339                         free_buffer_head(bh);
3340                         bh = next;
3341                 } while (bh != buffers_to_free);
3342         }
3343         return ret;
3344 }
3345 EXPORT_SYMBOL(try_to_free_buffers);
3346 
3347 /*
3348  * There are no bdflush tunables left.  But distributions are
3349  * still running obsolete flush daemons, so we terminate them here.
3350  *
3351  * Use of bdflush() is deprecated and will be removed in a future kernel.
3352  * The `flush-X' kernel threads fully replace bdflush daemons and this call.
3353  */
3354 SYSCALL_DEFINE2(bdflush, int, func, long, data)
3355 {
3356         static int msg_count;
3357 
3358         if (!capable(CAP_SYS_ADMIN))
3359                 return -EPERM;
3360 
3361         if (msg_count < 5) {
3362                 msg_count++;
3363                 printk(KERN_INFO
3364                         "warning: process `%s' used the obsolete bdflush"
3365                         " system call\n", current->comm);
3366                 printk(KERN_INFO "Fix your initscripts?\n");
3367         }
3368 
3369         if (func == 1)
3370                 do_exit(0);
3371         return 0;
3372 }
3373 
3374 /*
3375  * Buffer-head allocation
3376  */
3377 static struct kmem_cache *bh_cachep __read_mostly;
3378 
3379 /*
3380  * Once the number of bh's in the machine exceeds this level, we start
3381  * stripping them in writeback.
3382  */
3383 static unsigned long max_buffer_heads;
3384 
3385 int buffer_heads_over_limit;
3386 
3387 struct bh_accounting {
3388         int nr;                 /* Number of live bh's */
3389         int ratelimit;          /* Limit cacheline bouncing */
3390 };
3391 
3392 static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0};
3393 
3394 static void recalc_bh_state(void)
3395 {
3396         int i;
3397         int tot = 0;
3398 
3399         if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096)
3400                 return;
3401         __this_cpu_write(bh_accounting.ratelimit, 0);
3402         for_each_online_cpu(i)
3403                 tot += per_cpu(bh_accounting, i).nr;
3404         buffer_heads_over_limit = (tot > max_buffer_heads);
3405 }
3406 
3407 struct buffer_head *alloc_buffer_head(gfp_t gfp_flags)
3408 {
3409         struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags);
3410         if (ret) {
3411                 INIT_LIST_HEAD(&ret->b_assoc_buffers);
3412                 preempt_disable();
3413                 __this_cpu_inc(bh_accounting.nr);
3414                 recalc_bh_state();
3415                 preempt_enable();
3416         }
3417         return ret;
3418 }
3419 EXPORT_SYMBOL(alloc_buffer_head);
3420 
3421 void free_buffer_head(struct buffer_head *bh)
3422 {
3423         BUG_ON(!list_empty(&bh->b_assoc_buffers));
3424         kmem_cache_free(bh_cachep, bh);
3425         preempt_disable();
3426         __this_cpu_dec(bh_accounting.nr);
3427         recalc_bh_state();
3428         preempt_enable();
3429 }
3430 EXPORT_SYMBOL(free_buffer_head);
3431 
3432 static int buffer_exit_cpu_dead(unsigned int cpu)
3433 {
3434         int i;
3435         struct bh_lru *b = &per_cpu(bh_lrus, cpu);
3436 
3437         for (i = 0; i < BH_LRU_SIZE; i++) {
3438                 brelse(b->bhs[i]);
3439                 b->bhs[i] = NULL;
3440         }
3441         this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr);
3442         per_cpu(bh_accounting, cpu).nr = 0;
3443         return 0;
3444 }
3445 
3446 /**
3447  * bh_uptodate_or_lock - Test whether the buffer is uptodate
3448  * @bh: struct buffer_head
3449  *
3450  * Return true if the buffer is up-to-date and false,
3451  * with the buffer locked, if not.
3452  */
3453 int bh_uptodate_or_lock(struct buffer_head *bh)
3454 {
3455         if (!buffer_uptodate(bh)) {
3456                 lock_buffer(bh);
3457                 if (!buffer_uptodate(bh))
3458                         return 0;
3459                 unlock_buffer(bh);
3460         }
3461         return 1;
3462 }
3463 EXPORT_SYMBOL(bh_uptodate_or_lock);
3464 
3465 /**
3466  * bh_submit_read - Submit a locked buffer for reading
3467  * @bh: struct buffer_head
3468  *
3469  * Returns zero on success and -EIO on error.
3470  */
3471 int bh_submit_read(struct buffer_head *bh)
3472 {
3473         BUG_ON(!buffer_locked(bh));
3474 
3475         if (buffer_uptodate(bh)) {
3476                 unlock_buffer(bh);
3477                 return 0;
3478         }
3479 
3480         get_bh(bh);
3481         bh->b_end_io = end_buffer_read_sync;
3482         submit_bh(REQ_OP_READ, 0, bh);
3483         wait_on_buffer(bh);
3484         if (buffer_uptodate(bh))
3485                 return 0;
3486         return -EIO;
3487 }
3488 EXPORT_SYMBOL(bh_submit_read);
3489 
3490 /*
3491  * Seek for SEEK_DATA / SEEK_HOLE within @page, starting at @lastoff.
3492  *
3493  * Returns the offset within the file on success, and -ENOENT otherwise.
3494  */
3495 static loff_t
3496 page_seek_hole_data(struct page *page, loff_t lastoff, int whence)
3497 {
3498         loff_t offset = page_offset(page);
3499         struct buffer_head *bh, *head;
3500         bool seek_data = whence == SEEK_DATA;
3501 
3502         if (lastoff < offset)
3503                 lastoff = offset;
3504 
3505         bh = head = page_buffers(page);
3506         do {
3507                 offset += bh->b_size;
3508                 if (lastoff >= offset)
3509                         continue;
3510 
3511                 /*
3512                  * Unwritten extents that have data in the page cache covering
3513                  * them can be identified by the BH_Unwritten state flag.
3514                  * Pages with multiple buffers might have a mix of holes, data
3515                  * and unwritten extents - any buffer with valid data in it
3516                  * should have BH_Uptodate flag set on it.
3517                  */
3518 
3519                 if ((buffer_unwritten(bh) || buffer_uptodate(bh)) == seek_data)
3520                         return lastoff;
3521 
3522                 lastoff = offset;
3523         } while ((bh = bh->b_this_page) != head);
3524         return -ENOENT;
3525 }
3526 
3527 /*
3528  * Seek for SEEK_DATA / SEEK_HOLE in the page cache.
3529  *
3530  * Within unwritten extents, the page cache determines which parts are holes
3531  * and which are data: unwritten and uptodate buffer heads count as data;
3532  * everything else counts as a hole.
3533  *
3534  * Returns the resulting offset on successs, and -ENOENT otherwise.
3535  */
3536 loff_t
3537 page_cache_seek_hole_data(struct inode *inode, loff_t offset, loff_t length,
3538                           int whence)
3539 {
3540         pgoff_t index = offset >> PAGE_SHIFT;
3541         pgoff_t end = DIV_ROUND_UP(offset + length, PAGE_SIZE);
3542         loff_t lastoff = offset;
3543         struct pagevec pvec;
3544 
3545         if (length <= 0)
3546                 return -ENOENT;
3547 
3548         pagevec_init(&pvec, 0);
3549 
3550         do {
3551                 unsigned nr_pages, i;
3552 
3553                 nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping, &index,
3554                                                 end - 1);
3555                 if (nr_pages == 0)
3556                         break;
3557 
3558                 for (i = 0; i < nr_pages; i++) {
3559                         struct page *page = pvec.pages[i];
3560 
3561                         /*
3562                          * At this point, the page may be truncated or
3563                          * invalidated (changing page->mapping to NULL), or
3564                          * even swizzled back from swapper_space to tmpfs file
3565                          * mapping.  However, page->index will not change
3566                          * because we have a reference on the page.
3567                          *
3568                          * If current page offset is beyond where we've ended,
3569                          * we've found a hole.
3570                          */
3571                         if (whence == SEEK_HOLE &&
3572                             lastoff < page_offset(page))
3573                                 goto check_range;
3574 
3575                         lock_page(page);
3576                         if (likely(page->mapping == inode->i_mapping) &&
3577                             page_has_buffers(page)) {
3578                                 lastoff = page_seek_hole_data(page, lastoff, whence);
3579                                 if (lastoff >= 0) {
3580                                         unlock_page(page);
3581                                         goto check_range;
3582                                 }
3583                         }
3584                         unlock_page(page);
3585                         lastoff = page_offset(page) + PAGE_SIZE;
3586                 }
3587                 pagevec_release(&pvec);
3588         } while (index < end);
3589 
3590         /* When no page at lastoff and we are not done, we found a hole. */
3591         if (whence != SEEK_HOLE)
3592                 goto not_found;
3593 
3594 check_range:
3595         if (lastoff < offset + length)
3596                 goto out;
3597 not_found:
3598         lastoff = -ENOENT;
3599 out:
3600         pagevec_release(&pvec);
3601         return lastoff;
3602 }
3603 
3604 void __init buffer_init(void)
3605 {
3606         unsigned long nrpages;
3607         int ret;
3608 
3609         bh_cachep = kmem_cache_create("buffer_head",
3610                         sizeof(struct buffer_head), 0,
3611                                 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
3612                                 SLAB_MEM_SPREAD),
3613                                 NULL);
3614 
3615         /*
3616          * Limit the bh occupancy to 10% of ZONE_NORMAL
3617          */
3618         nrpages = (nr_free_buffer_pages() * 10) / 100;
3619         max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head));
3620         ret = cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD, "fs/buffer:dead",
3621                                         NULL, buffer_exit_cpu_dead);
3622         WARN_ON(ret < 0);
3623 }
3624 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp