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

TOMOYO Linux Cross Reference
Linux/fs/buffer.c

Version: ~ [ linux-5.8-rc4 ] ~ [ linux-5.7.7 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.50 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.131 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.187 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.229 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.229 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ 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-2.6.32.71 ] ~ [ 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/syscalls.h>
 23 #include <linux/fs.h>
 24 #include <linux/mm.h>
 25 #include <linux/percpu.h>
 26 #include <linux/slab.h>
 27 #include <linux/capability.h>
 28 #include <linux/blkdev.h>
 29 #include <linux/file.h>
 30 #include <linux/quotaops.h>
 31 #include <linux/highmem.h>
 32 #include <linux/export.h>
 33 #include <linux/writeback.h>
 34 #include <linux/hash.h>
 35 #include <linux/suspend.h>
 36 #include <linux/buffer_head.h>
 37 #include <linux/task_io_accounting_ops.h>
 38 #include <linux/bio.h>
 39 #include <linux/notifier.h>
 40 #include <linux/cpu.h>
 41 #include <linux/bitops.h>
 42 #include <linux/mpage.h>
 43 #include <linux/bit_spinlock.h>
 44 #include <trace/events/block.h>
 45 
 46 static int fsync_buffers_list(spinlock_t *lock, struct list_head *list);
 47 
 48 #define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers)
 49 
 50 void init_buffer(struct buffer_head *bh, bh_end_io_t *handler, void *private)
 51 {
 52         bh->b_end_io = handler;
 53         bh->b_private = private;
 54 }
 55 EXPORT_SYMBOL(init_buffer);
 56 
 57 inline void touch_buffer(struct buffer_head *bh)
 58 {
 59         trace_block_touch_buffer(bh);
 60         mark_page_accessed(bh->b_page);
 61 }
 62 EXPORT_SYMBOL(touch_buffer);
 63 
 64 void __lock_buffer(struct buffer_head *bh)
 65 {
 66         wait_on_bit_lock_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE);
 67 }
 68 EXPORT_SYMBOL(__lock_buffer);
 69 
 70 void unlock_buffer(struct buffer_head *bh)
 71 {
 72         clear_bit_unlock(BH_Lock, &bh->b_state);
 73         smp_mb__after_atomic();
 74         wake_up_bit(&bh->b_state, BH_Lock);
 75 }
 76 EXPORT_SYMBOL(unlock_buffer);
 77 
 78 /*
 79  * Returns if the page has dirty or writeback buffers. If all the buffers
 80  * are unlocked and clean then the PageDirty information is stale. If
 81  * any of the pages are locked, it is assumed they are locked for IO.
 82  */
 83 void buffer_check_dirty_writeback(struct page *page,
 84                                      bool *dirty, bool *writeback)
 85 {
 86         struct buffer_head *head, *bh;
 87         *dirty = false;
 88         *writeback = false;
 89 
 90         BUG_ON(!PageLocked(page));
 91 
 92         if (!page_has_buffers(page))
 93                 return;
 94 
 95         if (PageWriteback(page))
 96                 *writeback = true;
 97 
 98         head = page_buffers(page);
 99         bh = head;
100         do {
101                 if (buffer_locked(bh))
102                         *writeback = true;
103 
104                 if (buffer_dirty(bh))
105                         *dirty = true;
106 
107                 bh = bh->b_this_page;
108         } while (bh != head);
109 }
110 EXPORT_SYMBOL(buffer_check_dirty_writeback);
111 
112 /*
113  * Block until a buffer comes unlocked.  This doesn't stop it
114  * from becoming locked again - you have to lock it yourself
115  * if you want to preserve its state.
116  */
117 void __wait_on_buffer(struct buffer_head * bh)
118 {
119         wait_on_bit_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE);
120 }
121 EXPORT_SYMBOL(__wait_on_buffer);
122 
123 static void
124 __clear_page_buffers(struct page *page)
125 {
126         ClearPagePrivate(page);
127         set_page_private(page, 0);
128         page_cache_release(page);
129 }
130 
131 static void buffer_io_error(struct buffer_head *bh, char *msg)
132 {
133         char b[BDEVNAME_SIZE];
134 
135         if (!test_bit(BH_Quiet, &bh->b_state))
136                 printk_ratelimited(KERN_ERR
137                         "Buffer I/O error on dev %s, logical block %llu%s\n",
138                         bdevname(bh->b_bdev, b),
139                         (unsigned long long)bh->b_blocknr, msg);
140 }
141 
142 /*
143  * End-of-IO handler helper function which does not touch the bh after
144  * unlocking it.
145  * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but
146  * a race there is benign: unlock_buffer() only use the bh's address for
147  * hashing after unlocking the buffer, so it doesn't actually touch the bh
148  * itself.
149  */
150 static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate)
151 {
152         if (uptodate) {
153                 set_buffer_uptodate(bh);
154         } else {
155                 /* This happens, due to failed READA attempts. */
156                 clear_buffer_uptodate(bh);
157         }
158         unlock_buffer(bh);
159 }
160 
161 /*
162  * Default synchronous end-of-IO handler..  Just mark it up-to-date and
163  * unlock the buffer. This is what ll_rw_block uses too.
164  */
165 void end_buffer_read_sync(struct buffer_head *bh, int uptodate)
166 {
167         __end_buffer_read_notouch(bh, uptodate);
168         put_bh(bh);
169 }
170 EXPORT_SYMBOL(end_buffer_read_sync);
171 
172 void end_buffer_write_sync(struct buffer_head *bh, int uptodate)
173 {
174         if (uptodate) {
175                 set_buffer_uptodate(bh);
176         } else {
177                 buffer_io_error(bh, ", lost sync page write");
178                 set_buffer_write_io_error(bh);
179                 clear_buffer_uptodate(bh);
180         }
181         unlock_buffer(bh);
182         put_bh(bh);
183 }
184 EXPORT_SYMBOL(end_buffer_write_sync);
185 
186 /*
187  * Various filesystems appear to want __find_get_block to be non-blocking.
188  * But it's the page lock which protects the buffers.  To get around this,
189  * we get exclusion from try_to_free_buffers with the blockdev mapping's
190  * private_lock.
191  *
192  * Hack idea: for the blockdev mapping, i_bufferlist_lock contention
193  * may be quite high.  This code could TryLock the page, and if that
194  * succeeds, there is no need to take private_lock. (But if
195  * private_lock is contended then so is mapping->tree_lock).
196  */
197 static struct buffer_head *
198 __find_get_block_slow(struct block_device *bdev, sector_t block)
199 {
200         struct inode *bd_inode = bdev->bd_inode;
201         struct address_space *bd_mapping = bd_inode->i_mapping;
202         struct buffer_head *ret = NULL;
203         pgoff_t index;
204         struct buffer_head *bh;
205         struct buffer_head *head;
206         struct page *page;
207         int all_mapped = 1;
208 
209         index = block >> (PAGE_CACHE_SHIFT - bd_inode->i_blkbits);
210         page = find_get_page_flags(bd_mapping, index, FGP_ACCESSED);
211         if (!page)
212                 goto out;
213 
214         spin_lock(&bd_mapping->private_lock);
215         if (!page_has_buffers(page))
216                 goto out_unlock;
217         head = page_buffers(page);
218         bh = head;
219         do {
220                 if (!buffer_mapped(bh))
221                         all_mapped = 0;
222                 else if (bh->b_blocknr == block) {
223                         ret = bh;
224                         get_bh(bh);
225                         goto out_unlock;
226                 }
227                 bh = bh->b_this_page;
228         } while (bh != head);
229 
230         /* we might be here because some of the buffers on this page are
231          * not mapped.  This is due to various races between
232          * file io on the block device and getblk.  It gets dealt with
233          * elsewhere, don't buffer_error if we had some unmapped buffers
234          */
235         if (all_mapped) {
236                 char b[BDEVNAME_SIZE];
237 
238                 printk("__find_get_block_slow() failed. "
239                         "block=%llu, b_blocknr=%llu\n",
240                         (unsigned long long)block,
241                         (unsigned long long)bh->b_blocknr);
242                 printk("b_state=0x%08lx, b_size=%zu\n",
243                         bh->b_state, bh->b_size);
244                 printk("device %s blocksize: %d\n", bdevname(bdev, b),
245                         1 << bd_inode->i_blkbits);
246         }
247 out_unlock:
248         spin_unlock(&bd_mapping->private_lock);
249         page_cache_release(page);
250 out:
251         return ret;
252 }
253 
254 /*
255  * Kick the writeback threads then try to free up some ZONE_NORMAL memory.
256  */
257 static void free_more_memory(void)
258 {
259         struct zone *zone;
260         int nid;
261 
262         wakeup_flusher_threads(1024, WB_REASON_FREE_MORE_MEM);
263         yield();
264 
265         for_each_online_node(nid) {
266                 (void)first_zones_zonelist(node_zonelist(nid, GFP_NOFS),
267                                                 gfp_zone(GFP_NOFS), NULL,
268                                                 &zone);
269                 if (zone)
270                         try_to_free_pages(node_zonelist(nid, GFP_NOFS), 0,
271                                                 GFP_NOFS, NULL);
272         }
273 }
274 
275 /*
276  * I/O completion handler for block_read_full_page() - pages
277  * which come unlocked at the end of I/O.
278  */
279 static void end_buffer_async_read(struct buffer_head *bh, int uptodate)
280 {
281         unsigned long flags;
282         struct buffer_head *first;
283         struct buffer_head *tmp;
284         struct page *page;
285         int page_uptodate = 1;
286 
287         BUG_ON(!buffer_async_read(bh));
288 
289         page = bh->b_page;
290         if (uptodate) {
291                 set_buffer_uptodate(bh);
292         } else {
293                 clear_buffer_uptodate(bh);
294                 buffer_io_error(bh, ", async page read");
295                 SetPageError(page);
296         }
297 
298         /*
299          * Be _very_ careful from here on. Bad things can happen if
300          * two buffer heads end IO at almost the same time and both
301          * decide that the page is now completely done.
302          */
303         first = page_buffers(page);
304         local_irq_save(flags);
305         bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
306         clear_buffer_async_read(bh);
307         unlock_buffer(bh);
308         tmp = bh;
309         do {
310                 if (!buffer_uptodate(tmp))
311                         page_uptodate = 0;
312                 if (buffer_async_read(tmp)) {
313                         BUG_ON(!buffer_locked(tmp));
314                         goto still_busy;
315                 }
316                 tmp = tmp->b_this_page;
317         } while (tmp != bh);
318         bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
319         local_irq_restore(flags);
320 
321         /*
322          * If none of the buffers had errors and they are all
323          * uptodate then we can set the page uptodate.
324          */
325         if (page_uptodate && !PageError(page))
326                 SetPageUptodate(page);
327         unlock_page(page);
328         return;
329 
330 still_busy:
331         bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
332         local_irq_restore(flags);
333         return;
334 }
335 
336 /*
337  * Completion handler for block_write_full_page() - pages which are unlocked
338  * during I/O, and which have PageWriteback cleared upon I/O completion.
339  */
340 void end_buffer_async_write(struct buffer_head *bh, int uptodate)
341 {
342         unsigned long flags;
343         struct buffer_head *first;
344         struct buffer_head *tmp;
345         struct page *page;
346 
347         BUG_ON(!buffer_async_write(bh));
348 
349         page = bh->b_page;
350         if (uptodate) {
351                 set_buffer_uptodate(bh);
352         } else {
353                 buffer_io_error(bh, ", lost async page write");
354                 set_bit(AS_EIO, &page->mapping->flags);
355                 set_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         if (buffer_write_io_error(bh))
484                 set_bit(AS_EIO, &bh->b_assoc_map->flags);
485         bh->b_assoc_map = NULL;
486 }
487 
488 int inode_has_buffers(struct inode *inode)
489 {
490         return !list_empty(&inode->i_data.private_list);
491 }
492 
493 /*
494  * osync is designed to support O_SYNC io.  It waits synchronously for
495  * all already-submitted IO to complete, but does not queue any new
496  * writes to the disk.
497  *
498  * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as
499  * you dirty the buffers, and then use osync_inode_buffers to wait for
500  * completion.  Any other dirty buffers which are not yet queued for
501  * write will not be flushed to disk by the osync.
502  */
503 static int osync_buffers_list(spinlock_t *lock, struct list_head *list)
504 {
505         struct buffer_head *bh;
506         struct list_head *p;
507         int err = 0;
508 
509         spin_lock(lock);
510 repeat:
511         list_for_each_prev(p, list) {
512                 bh = BH_ENTRY(p);
513                 if (buffer_locked(bh)) {
514                         get_bh(bh);
515                         spin_unlock(lock);
516                         wait_on_buffer(bh);
517                         if (!buffer_uptodate(bh))
518                                 err = -EIO;
519                         brelse(bh);
520                         spin_lock(lock);
521                         goto repeat;
522                 }
523         }
524         spin_unlock(lock);
525         return err;
526 }
527 
528 static void do_thaw_one(struct super_block *sb, void *unused)
529 {
530         char b[BDEVNAME_SIZE];
531         while (sb->s_bdev && !thaw_bdev(sb->s_bdev, sb))
532                 printk(KERN_WARNING "Emergency Thaw on %s\n",
533                        bdevname(sb->s_bdev, b));
534 }
535 
536 static void do_thaw_all(struct work_struct *work)
537 {
538         iterate_supers(do_thaw_one, NULL);
539         kfree(work);
540         printk(KERN_WARNING "Emergency Thaw complete\n");
541 }
542 
543 /**
544  * emergency_thaw_all -- forcibly thaw every frozen filesystem
545  *
546  * Used for emergency unfreeze of all filesystems via SysRq
547  */
548 void emergency_thaw_all(void)
549 {
550         struct work_struct *work;
551 
552         work = kmalloc(sizeof(*work), GFP_ATOMIC);
553         if (work) {
554                 INIT_WORK(work, do_thaw_all);
555                 schedule_work(work);
556         }
557 }
558 
559 /**
560  * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers
561  * @mapping: the mapping which wants those buffers written
562  *
563  * Starts I/O against the buffers at mapping->private_list, and waits upon
564  * that I/O.
565  *
566  * Basically, this is a convenience function for fsync().
567  * @mapping is a file or directory which needs those buffers to be written for
568  * a successful fsync().
569  */
570 int sync_mapping_buffers(struct address_space *mapping)
571 {
572         struct address_space *buffer_mapping = mapping->private_data;
573 
574         if (buffer_mapping == NULL || list_empty(&mapping->private_list))
575                 return 0;
576 
577         return fsync_buffers_list(&buffer_mapping->private_lock,
578                                         &mapping->private_list);
579 }
580 EXPORT_SYMBOL(sync_mapping_buffers);
581 
582 /*
583  * Called when we've recently written block `bblock', and it is known that
584  * `bblock' was for a buffer_boundary() buffer.  This means that the block at
585  * `bblock + 1' is probably a dirty indirect block.  Hunt it down and, if it's
586  * dirty, schedule it for IO.  So that indirects merge nicely with their data.
587  */
588 void write_boundary_block(struct block_device *bdev,
589                         sector_t bblock, unsigned blocksize)
590 {
591         struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize);
592         if (bh) {
593                 if (buffer_dirty(bh))
594                         ll_rw_block(WRITE, 1, &bh);
595                 put_bh(bh);
596         }
597 }
598 
599 void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode)
600 {
601         struct address_space *mapping = inode->i_mapping;
602         struct address_space *buffer_mapping = bh->b_page->mapping;
603 
604         mark_buffer_dirty(bh);
605         if (!mapping->private_data) {
606                 mapping->private_data = buffer_mapping;
607         } else {
608                 BUG_ON(mapping->private_data != buffer_mapping);
609         }
610         if (!bh->b_assoc_map) {
611                 spin_lock(&buffer_mapping->private_lock);
612                 list_move_tail(&bh->b_assoc_buffers,
613                                 &mapping->private_list);
614                 bh->b_assoc_map = mapping;
615                 spin_unlock(&buffer_mapping->private_lock);
616         }
617 }
618 EXPORT_SYMBOL(mark_buffer_dirty_inode);
619 
620 /*
621  * Mark the page dirty, and set it dirty in the radix tree, and mark the inode
622  * dirty.
623  *
624  * If warn is true, then emit a warning if the page is not uptodate and has
625  * not been truncated.
626  */
627 static void __set_page_dirty(struct page *page,
628                 struct address_space *mapping, int warn)
629 {
630         unsigned long flags;
631 
632         spin_lock_irqsave(&mapping->tree_lock, flags);
633         if (page->mapping) {    /* Race with truncate? */
634                 WARN_ON_ONCE(warn && !PageUptodate(page));
635                 account_page_dirtied(page, mapping);
636                 radix_tree_tag_set(&mapping->page_tree,
637                                 page_index(page), PAGECACHE_TAG_DIRTY);
638         }
639         spin_unlock_irqrestore(&mapping->tree_lock, flags);
640         __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
641 }
642 
643 /*
644  * Add a page to the dirty page list.
645  *
646  * It is a sad fact of life that this function is called from several places
647  * deeply under spinlocking.  It may not sleep.
648  *
649  * If the page has buffers, the uptodate buffers are set dirty, to preserve
650  * dirty-state coherency between the page and the buffers.  It the page does
651  * not have buffers then when they are later attached they will all be set
652  * dirty.
653  *
654  * The buffers are dirtied before the page is dirtied.  There's a small race
655  * window in which a writepage caller may see the page cleanness but not the
656  * buffer dirtiness.  That's fine.  If this code were to set the page dirty
657  * before the buffers, a concurrent writepage caller could clear the page dirty
658  * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean
659  * page on the dirty page list.
660  *
661  * We use private_lock to lock against try_to_free_buffers while using the
662  * page's buffer list.  Also use this to protect against clean buffers being
663  * added to the page after it was set dirty.
664  *
665  * FIXME: may need to call ->reservepage here as well.  That's rather up to the
666  * address_space though.
667  */
668 int __set_page_dirty_buffers(struct page *page)
669 {
670         int newly_dirty;
671         struct address_space *mapping = page_mapping(page);
672 
673         if (unlikely(!mapping))
674                 return !TestSetPageDirty(page);
675 
676         spin_lock(&mapping->private_lock);
677         if (page_has_buffers(page)) {
678                 struct buffer_head *head = page_buffers(page);
679                 struct buffer_head *bh = head;
680 
681                 do {
682                         set_buffer_dirty(bh);
683                         bh = bh->b_this_page;
684                 } while (bh != head);
685         }
686         newly_dirty = !TestSetPageDirty(page);
687         spin_unlock(&mapping->private_lock);
688 
689         if (newly_dirty)
690                 __set_page_dirty(page, mapping, 1);
691         return newly_dirty;
692 }
693 EXPORT_SYMBOL(__set_page_dirty_buffers);
694 
695 /*
696  * Write out and wait upon a list of buffers.
697  *
698  * We have conflicting pressures: we want to make sure that all
699  * initially dirty buffers get waited on, but that any subsequently
700  * dirtied buffers don't.  After all, we don't want fsync to last
701  * forever if somebody is actively writing to the file.
702  *
703  * Do this in two main stages: first we copy dirty buffers to a
704  * temporary inode list, queueing the writes as we go.  Then we clean
705  * up, waiting for those writes to complete.
706  * 
707  * During this second stage, any subsequent updates to the file may end
708  * up refiling the buffer on the original inode's dirty list again, so
709  * there is a chance we will end up with a buffer queued for write but
710  * not yet completed on that list.  So, as a final cleanup we go through
711  * the osync code to catch these locked, dirty buffers without requeuing
712  * any newly dirty buffers for write.
713  */
714 static int fsync_buffers_list(spinlock_t *lock, struct list_head *list)
715 {
716         struct buffer_head *bh;
717         struct list_head tmp;
718         struct address_space *mapping;
719         int err = 0, err2;
720         struct blk_plug plug;
721 
722         INIT_LIST_HEAD(&tmp);
723         blk_start_plug(&plug);
724 
725         spin_lock(lock);
726         while (!list_empty(list)) {
727                 bh = BH_ENTRY(list->next);
728                 mapping = bh->b_assoc_map;
729                 __remove_assoc_queue(bh);
730                 /* Avoid race with mark_buffer_dirty_inode() which does
731                  * a lockless check and we rely on seeing the dirty bit */
732                 smp_mb();
733                 if (buffer_dirty(bh) || buffer_locked(bh)) {
734                         list_add(&bh->b_assoc_buffers, &tmp);
735                         bh->b_assoc_map = mapping;
736                         if (buffer_dirty(bh)) {
737                                 get_bh(bh);
738                                 spin_unlock(lock);
739                                 /*
740                                  * Ensure any pending I/O completes so that
741                                  * write_dirty_buffer() actually writes the
742                                  * current contents - it is a noop if I/O is
743                                  * still in flight on potentially older
744                                  * contents.
745                                  */
746                                 write_dirty_buffer(bh, WRITE_SYNC);
747 
748                                 /*
749                                  * Kick off IO for the previous mapping. Note
750                                  * that we will not run the very last mapping,
751                                  * wait_on_buffer() will do that for us
752                                  * through sync_buffer().
753                                  */
754                                 brelse(bh);
755                                 spin_lock(lock);
756                         }
757                 }
758         }
759 
760         spin_unlock(lock);
761         blk_finish_plug(&plug);
762         spin_lock(lock);
763 
764         while (!list_empty(&tmp)) {
765                 bh = BH_ENTRY(tmp.prev);
766                 get_bh(bh);
767                 mapping = bh->b_assoc_map;
768                 __remove_assoc_queue(bh);
769                 /* Avoid race with mark_buffer_dirty_inode() which does
770                  * a lockless check and we rely on seeing the dirty bit */
771                 smp_mb();
772                 if (buffer_dirty(bh)) {
773                         list_add(&bh->b_assoc_buffers,
774                                  &mapping->private_list);
775                         bh->b_assoc_map = mapping;
776                 }
777                 spin_unlock(lock);
778                 wait_on_buffer(bh);
779                 if (!buffer_uptodate(bh))
780                         err = -EIO;
781                 brelse(bh);
782                 spin_lock(lock);
783         }
784         
785         spin_unlock(lock);
786         err2 = osync_buffers_list(lock, list);
787         if (err)
788                 return err;
789         else
790                 return err2;
791 }
792 
793 /*
794  * Invalidate any and all dirty buffers on a given inode.  We are
795  * probably unmounting the fs, but that doesn't mean we have already
796  * done a sync().  Just drop the buffers from the inode list.
797  *
798  * NOTE: we take the inode's blockdev's mapping's private_lock.  Which
799  * assumes that all the buffers are against the blockdev.  Not true
800  * for reiserfs.
801  */
802 void invalidate_inode_buffers(struct inode *inode)
803 {
804         if (inode_has_buffers(inode)) {
805                 struct address_space *mapping = &inode->i_data;
806                 struct list_head *list = &mapping->private_list;
807                 struct address_space *buffer_mapping = mapping->private_data;
808 
809                 spin_lock(&buffer_mapping->private_lock);
810                 while (!list_empty(list))
811                         __remove_assoc_queue(BH_ENTRY(list->next));
812                 spin_unlock(&buffer_mapping->private_lock);
813         }
814 }
815 EXPORT_SYMBOL(invalidate_inode_buffers);
816 
817 /*
818  * Remove any clean buffers from the inode's buffer list.  This is called
819  * when we're trying to free the inode itself.  Those buffers can pin it.
820  *
821  * Returns true if all buffers were removed.
822  */
823 int remove_inode_buffers(struct inode *inode)
824 {
825         int ret = 1;
826 
827         if (inode_has_buffers(inode)) {
828                 struct address_space *mapping = &inode->i_data;
829                 struct list_head *list = &mapping->private_list;
830                 struct address_space *buffer_mapping = mapping->private_data;
831 
832                 spin_lock(&buffer_mapping->private_lock);
833                 while (!list_empty(list)) {
834                         struct buffer_head *bh = BH_ENTRY(list->next);
835                         if (buffer_dirty(bh)) {
836                                 ret = 0;
837                                 break;
838                         }
839                         __remove_assoc_queue(bh);
840                 }
841                 spin_unlock(&buffer_mapping->private_lock);
842         }
843         return ret;
844 }
845 
846 /*
847  * Create the appropriate buffers when given a page for data area and
848  * the size of each buffer.. Use the bh->b_this_page linked list to
849  * follow the buffers created.  Return NULL if unable to create more
850  * buffers.
851  *
852  * The retry flag is used to differentiate async IO (paging, swapping)
853  * which may not fail from ordinary buffer allocations.
854  */
855 struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size,
856                 int retry)
857 {
858         struct buffer_head *bh, *head;
859         long offset;
860 
861 try_again:
862         head = NULL;
863         offset = PAGE_SIZE;
864         while ((offset -= size) >= 0) {
865                 bh = alloc_buffer_head(GFP_NOFS);
866                 if (!bh)
867                         goto no_grow;
868 
869                 bh->b_this_page = head;
870                 bh->b_blocknr = -1;
871                 head = bh;
872 
873                 bh->b_size = size;
874 
875                 /* Link the buffer to its page */
876                 set_bh_page(bh, page, offset);
877         }
878         return head;
879 /*
880  * In case anything failed, we just free everything we got.
881  */
882 no_grow:
883         if (head) {
884                 do {
885                         bh = head;
886                         head = head->b_this_page;
887                         free_buffer_head(bh);
888                 } while (head);
889         }
890 
891         /*
892          * Return failure for non-async IO requests.  Async IO requests
893          * are not allowed to fail, so we have to wait until buffer heads
894          * become available.  But we don't want tasks sleeping with 
895          * partially complete buffers, so all were released above.
896          */
897         if (!retry)
898                 return NULL;
899 
900         /* We're _really_ low on memory. Now we just
901          * wait for old buffer heads to become free due to
902          * finishing IO.  Since this is an async request and
903          * the reserve list is empty, we're sure there are 
904          * async buffer heads in use.
905          */
906         free_more_memory();
907         goto try_again;
908 }
909 EXPORT_SYMBOL_GPL(alloc_page_buffers);
910 
911 static inline void
912 link_dev_buffers(struct page *page, struct buffer_head *head)
913 {
914         struct buffer_head *bh, *tail;
915 
916         bh = head;
917         do {
918                 tail = bh;
919                 bh = bh->b_this_page;
920         } while (bh);
921         tail->b_this_page = head;
922         attach_page_buffers(page, head);
923 }
924 
925 static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size)
926 {
927         sector_t retval = ~((sector_t)0);
928         loff_t sz = i_size_read(bdev->bd_inode);
929 
930         if (sz) {
931                 unsigned int sizebits = blksize_bits(size);
932                 retval = (sz >> sizebits);
933         }
934         return retval;
935 }
936 
937 /*
938  * Initialise the state of a blockdev page's buffers.
939  */ 
940 static sector_t
941 init_page_buffers(struct page *page, struct block_device *bdev,
942                         sector_t block, int size)
943 {
944         struct buffer_head *head = page_buffers(page);
945         struct buffer_head *bh = head;
946         int uptodate = PageUptodate(page);
947         sector_t end_block = blkdev_max_block(I_BDEV(bdev->bd_inode), size);
948 
949         do {
950                 if (!buffer_mapped(bh)) {
951                         init_buffer(bh, NULL, NULL);
952                         bh->b_bdev = bdev;
953                         bh->b_blocknr = block;
954                         if (uptodate)
955                                 set_buffer_uptodate(bh);
956                         if (block < end_block)
957                                 set_buffer_mapped(bh);
958                 }
959                 block++;
960                 bh = bh->b_this_page;
961         } while (bh != head);
962 
963         /*
964          * Caller needs to validate requested block against end of device.
965          */
966         return end_block;
967 }
968 
969 /*
970  * Create the page-cache page that contains the requested block.
971  *
972  * This is used purely for blockdev mappings.
973  */
974 static int
975 grow_dev_page(struct block_device *bdev, sector_t block,
976               pgoff_t index, int size, int sizebits, gfp_t gfp)
977 {
978         struct inode *inode = bdev->bd_inode;
979         struct page *page;
980         struct buffer_head *bh;
981         sector_t end_block;
982         int ret = 0;            /* Will call free_more_memory() */
983         gfp_t gfp_mask;
984 
985         gfp_mask = (mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS) | gfp;
986 
987         /*
988          * XXX: __getblk_slow() can not really deal with failure and
989          * will endlessly loop on improvised global reclaim.  Prefer
990          * looping in the allocator rather than here, at least that
991          * code knows what it's doing.
992          */
993         gfp_mask |= __GFP_NOFAIL;
994 
995         page = find_or_create_page(inode->i_mapping, index, gfp_mask);
996         if (!page)
997                 return ret;
998 
999         BUG_ON(!PageLocked(page));
1000 
1001         if (page_has_buffers(page)) {
1002                 bh = page_buffers(page);
1003                 if (bh->b_size == size) {
1004                         end_block = init_page_buffers(page, bdev,
1005                                                 (sector_t)index << sizebits,
1006                                                 size);
1007                         goto done;
1008                 }
1009                 if (!try_to_free_buffers(page))
1010                         goto failed;
1011         }
1012 
1013         /*
1014          * Allocate some buffers for this page
1015          */
1016         bh = alloc_page_buffers(page, size, 0);
1017         if (!bh)
1018                 goto failed;
1019 
1020         /*
1021          * Link the page to the buffers and initialise them.  Take the
1022          * lock to be atomic wrt __find_get_block(), which does not
1023          * run under the page lock.
1024          */
1025         spin_lock(&inode->i_mapping->private_lock);
1026         link_dev_buffers(page, bh);
1027         end_block = init_page_buffers(page, bdev, (sector_t)index << sizebits,
1028                         size);
1029         spin_unlock(&inode->i_mapping->private_lock);
1030 done:
1031         ret = (block < end_block) ? 1 : -ENXIO;
1032 failed:
1033         unlock_page(page);
1034         page_cache_release(page);
1035         return ret;
1036 }
1037 
1038 /*
1039  * Create buffers for the specified block device block's page.  If
1040  * that page was dirty, the buffers are set dirty also.
1041  */
1042 static int
1043 grow_buffers(struct block_device *bdev, sector_t block, int size, gfp_t gfp)
1044 {
1045         pgoff_t index;
1046         int sizebits;
1047 
1048         sizebits = -1;
1049         do {
1050                 sizebits++;
1051         } while ((size << sizebits) < PAGE_SIZE);
1052 
1053         index = block >> sizebits;
1054 
1055         /*
1056          * Check for a block which wants to lie outside our maximum possible
1057          * pagecache index.  (this comparison is done using sector_t types).
1058          */
1059         if (unlikely(index != block >> sizebits)) {
1060                 char b[BDEVNAME_SIZE];
1061 
1062                 printk(KERN_ERR "%s: requested out-of-range block %llu for "
1063                         "device %s\n",
1064                         __func__, (unsigned long long)block,
1065                         bdevname(bdev, b));
1066                 return -EIO;
1067         }
1068 
1069         /* Create a page with the proper size buffers.. */
1070         return grow_dev_page(bdev, block, index, size, sizebits, gfp);
1071 }
1072 
1073 struct buffer_head *
1074 __getblk_slow(struct block_device *bdev, sector_t block,
1075              unsigned size, gfp_t gfp)
1076 {
1077         /* Size must be multiple of hard sectorsize */
1078         if (unlikely(size & (bdev_logical_block_size(bdev)-1) ||
1079                         (size < 512 || size > PAGE_SIZE))) {
1080                 printk(KERN_ERR "getblk(): invalid block size %d requested\n",
1081                                         size);
1082                 printk(KERN_ERR "logical block size: %d\n",
1083                                         bdev_logical_block_size(bdev));
1084 
1085                 dump_stack();
1086                 return NULL;
1087         }
1088 
1089         for (;;) {
1090                 struct buffer_head *bh;
1091                 int ret;
1092 
1093                 bh = __find_get_block(bdev, block, size);
1094                 if (bh)
1095                         return bh;
1096 
1097                 ret = grow_buffers(bdev, block, size, gfp);
1098                 if (ret < 0)
1099                         return NULL;
1100                 if (ret == 0)
1101                         free_more_memory();
1102         }
1103 }
1104 EXPORT_SYMBOL(__getblk_slow);
1105 
1106 /*
1107  * The relationship between dirty buffers and dirty pages:
1108  *
1109  * Whenever a page has any dirty buffers, the page's dirty bit is set, and
1110  * the page is tagged dirty in its radix tree.
1111  *
1112  * At all times, the dirtiness of the buffers represents the dirtiness of
1113  * subsections of the page.  If the page has buffers, the page dirty bit is
1114  * merely a hint about the true dirty state.
1115  *
1116  * When a page is set dirty in its entirety, all its buffers are marked dirty
1117  * (if the page has buffers).
1118  *
1119  * When a buffer is marked dirty, its page is dirtied, but the page's other
1120  * buffers are not.
1121  *
1122  * Also.  When blockdev buffers are explicitly read with bread(), they
1123  * individually become uptodate.  But their backing page remains not
1124  * uptodate - even if all of its buffers are uptodate.  A subsequent
1125  * block_read_full_page() against that page will discover all the uptodate
1126  * buffers, will set the page uptodate and will perform no I/O.
1127  */
1128 
1129 /**
1130  * mark_buffer_dirty - mark a buffer_head as needing writeout
1131  * @bh: the buffer_head to mark dirty
1132  *
1133  * mark_buffer_dirty() will set the dirty bit against the buffer, then set its
1134  * backing page dirty, then tag the page as dirty in its address_space's radix
1135  * tree and then attach the address_space's inode to its superblock's dirty
1136  * inode list.
1137  *
1138  * mark_buffer_dirty() is atomic.  It takes bh->b_page->mapping->private_lock,
1139  * mapping->tree_lock and mapping->host->i_lock.
1140  */
1141 void mark_buffer_dirty(struct buffer_head *bh)
1142 {
1143         WARN_ON_ONCE(!buffer_uptodate(bh));
1144 
1145         trace_block_dirty_buffer(bh);
1146 
1147         /*
1148          * Very *carefully* optimize the it-is-already-dirty case.
1149          *
1150          * Don't let the final "is it dirty" escape to before we
1151          * perhaps modified the buffer.
1152          */
1153         if (buffer_dirty(bh)) {
1154                 smp_mb();
1155                 if (buffer_dirty(bh))
1156                         return;
1157         }
1158 
1159         if (!test_set_buffer_dirty(bh)) {
1160                 struct page *page = bh->b_page;
1161                 if (!TestSetPageDirty(page)) {
1162                         struct address_space *mapping = page_mapping(page);
1163                         if (mapping)
1164                                 __set_page_dirty(page, mapping, 0);
1165                 }
1166         }
1167 }
1168 EXPORT_SYMBOL(mark_buffer_dirty);
1169 
1170 /*
1171  * Decrement a buffer_head's reference count.  If all buffers against a page
1172  * have zero reference count, are clean and unlocked, and if the page is clean
1173  * and unlocked then try_to_free_buffers() may strip the buffers from the page
1174  * in preparation for freeing it (sometimes, rarely, buffers are removed from
1175  * a page but it ends up not being freed, and buffers may later be reattached).
1176  */
1177 void __brelse(struct buffer_head * buf)
1178 {
1179         if (atomic_read(&buf->b_count)) {
1180                 put_bh(buf);
1181                 return;
1182         }
1183         WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n");
1184 }
1185 EXPORT_SYMBOL(__brelse);
1186 
1187 /*
1188  * bforget() is like brelse(), except it discards any
1189  * potentially dirty data.
1190  */
1191 void __bforget(struct buffer_head *bh)
1192 {
1193         clear_buffer_dirty(bh);
1194         if (bh->b_assoc_map) {
1195                 struct address_space *buffer_mapping = bh->b_page->mapping;
1196 
1197                 spin_lock(&buffer_mapping->private_lock);
1198                 list_del_init(&bh->b_assoc_buffers);
1199                 bh->b_assoc_map = NULL;
1200                 spin_unlock(&buffer_mapping->private_lock);
1201         }
1202         __brelse(bh);
1203 }
1204 EXPORT_SYMBOL(__bforget);
1205 
1206 static struct buffer_head *__bread_slow(struct buffer_head *bh)
1207 {
1208         lock_buffer(bh);
1209         if (buffer_uptodate(bh)) {
1210                 unlock_buffer(bh);
1211                 return bh;
1212         } else {
1213                 get_bh(bh);
1214                 bh->b_end_io = end_buffer_read_sync;
1215                 submit_bh(READ, bh);
1216                 wait_on_buffer(bh);
1217                 if (buffer_uptodate(bh))
1218                         return bh;
1219         }
1220         brelse(bh);
1221         return NULL;
1222 }
1223 
1224 /*
1225  * Per-cpu buffer LRU implementation.  To reduce the cost of __find_get_block().
1226  * The bhs[] array is sorted - newest buffer is at bhs[0].  Buffers have their
1227  * refcount elevated by one when they're in an LRU.  A buffer can only appear
1228  * once in a particular CPU's LRU.  A single buffer can be present in multiple
1229  * CPU's LRUs at the same time.
1230  *
1231  * This is a transparent caching front-end to sb_bread(), sb_getblk() and
1232  * sb_find_get_block().
1233  *
1234  * The LRUs themselves only need locking against invalidate_bh_lrus.  We use
1235  * a local interrupt disable for that.
1236  */
1237 
1238 #define BH_LRU_SIZE     16
1239 
1240 struct bh_lru {
1241         struct buffer_head *bhs[BH_LRU_SIZE];
1242 };
1243 
1244 static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }};
1245 
1246 #ifdef CONFIG_SMP
1247 #define bh_lru_lock()   local_irq_disable()
1248 #define bh_lru_unlock() local_irq_enable()
1249 #else
1250 #define bh_lru_lock()   preempt_disable()
1251 #define bh_lru_unlock() preempt_enable()
1252 #endif
1253 
1254 static inline void check_irqs_on(void)
1255 {
1256 #ifdef irqs_disabled
1257         BUG_ON(irqs_disabled());
1258 #endif
1259 }
1260 
1261 /*
1262  * The LRU management algorithm is dopey-but-simple.  Sorry.
1263  */
1264 static void bh_lru_install(struct buffer_head *bh)
1265 {
1266         struct buffer_head *evictee = NULL;
1267 
1268         check_irqs_on();
1269         bh_lru_lock();
1270         if (__this_cpu_read(bh_lrus.bhs[0]) != bh) {
1271                 struct buffer_head *bhs[BH_LRU_SIZE];
1272                 int in;
1273                 int out = 0;
1274 
1275                 get_bh(bh);
1276                 bhs[out++] = bh;
1277                 for (in = 0; in < BH_LRU_SIZE; in++) {
1278                         struct buffer_head *bh2 =
1279                                 __this_cpu_read(bh_lrus.bhs[in]);
1280 
1281                         if (bh2 == bh) {
1282                                 __brelse(bh2);
1283                         } else {
1284                                 if (out >= BH_LRU_SIZE) {
1285                                         BUG_ON(evictee != NULL);
1286                                         evictee = bh2;
1287                                 } else {
1288                                         bhs[out++] = bh2;
1289                                 }
1290                         }
1291                 }
1292                 while (out < BH_LRU_SIZE)
1293                         bhs[out++] = NULL;
1294                 memcpy(this_cpu_ptr(&bh_lrus.bhs), bhs, sizeof(bhs));
1295         }
1296         bh_lru_unlock();
1297 
1298         if (evictee)
1299                 __brelse(evictee);
1300 }
1301 
1302 /*
1303  * Look up the bh in this cpu's LRU.  If it's there, move it to the head.
1304  */
1305 static struct buffer_head *
1306 lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size)
1307 {
1308         struct buffer_head *ret = NULL;
1309         unsigned int i;
1310 
1311         check_irqs_on();
1312         bh_lru_lock();
1313         for (i = 0; i < BH_LRU_SIZE; i++) {
1314                 struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]);
1315 
1316                 if (bh && bh->b_blocknr == block && bh->b_bdev == bdev &&
1317                     bh->b_size == size) {
1318                         if (i) {
1319                                 while (i) {
1320                                         __this_cpu_write(bh_lrus.bhs[i],
1321                                                 __this_cpu_read(bh_lrus.bhs[i - 1]));
1322                                         i--;
1323                                 }
1324                                 __this_cpu_write(bh_lrus.bhs[0], bh);
1325                         }
1326                         get_bh(bh);
1327                         ret = bh;
1328                         break;
1329                 }
1330         }
1331         bh_lru_unlock();
1332         return ret;
1333 }
1334 
1335 /*
1336  * Perform a pagecache lookup for the matching buffer.  If it's there, refresh
1337  * it in the LRU and mark it as accessed.  If it is not present then return
1338  * NULL
1339  */
1340 struct buffer_head *
1341 __find_get_block(struct block_device *bdev, sector_t block, unsigned size)
1342 {
1343         struct buffer_head *bh = lookup_bh_lru(bdev, block, size);
1344 
1345         if (bh == NULL) {
1346                 /* __find_get_block_slow will mark the page accessed */
1347                 bh = __find_get_block_slow(bdev, block);
1348                 if (bh)
1349                         bh_lru_install(bh);
1350         } else
1351                 touch_buffer(bh);
1352 
1353         return bh;
1354 }
1355 EXPORT_SYMBOL(__find_get_block);
1356 
1357 /*
1358  * __getblk_gfp() will locate (and, if necessary, create) the buffer_head
1359  * which corresponds to the passed block_device, block and size. The
1360  * returned buffer has its reference count incremented.
1361  *
1362  * __getblk_gfp() will lock up the machine if grow_dev_page's
1363  * try_to_free_buffers() attempt is failing.  FIXME, perhaps?
1364  */
1365 struct buffer_head *
1366 __getblk_gfp(struct block_device *bdev, sector_t block,
1367              unsigned size, gfp_t gfp)
1368 {
1369         struct buffer_head *bh = __find_get_block(bdev, block, size);
1370 
1371         might_sleep();
1372         if (bh == NULL)
1373                 bh = __getblk_slow(bdev, block, size, gfp);
1374         return bh;
1375 }
1376 EXPORT_SYMBOL(__getblk_gfp);
1377 
1378 /*
1379  * Do async read-ahead on a buffer..
1380  */
1381 void __breadahead(struct block_device *bdev, sector_t block, unsigned size)
1382 {
1383         struct buffer_head *bh = __getblk(bdev, block, size);
1384         if (likely(bh)) {
1385                 ll_rw_block(READA, 1, &bh);
1386                 brelse(bh);
1387         }
1388 }
1389 EXPORT_SYMBOL(__breadahead);
1390 
1391 /**
1392  *  __bread_gfp() - reads a specified block and returns the bh
1393  *  @bdev: the block_device to read from
1394  *  @block: number of block
1395  *  @size: size (in bytes) to read
1396  *  @gfp: page allocation flag
1397  *
1398  *  Reads a specified block, and returns buffer head that contains it.
1399  *  The page cache can be allocated from non-movable area
1400  *  not to prevent page migration if you set gfp to zero.
1401  *  It returns NULL if the block was unreadable.
1402  */
1403 struct buffer_head *
1404 __bread_gfp(struct block_device *bdev, sector_t block,
1405                    unsigned size, gfp_t gfp)
1406 {
1407         struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp);
1408 
1409         if (likely(bh) && !buffer_uptodate(bh))
1410                 bh = __bread_slow(bh);
1411         return bh;
1412 }
1413 EXPORT_SYMBOL(__bread_gfp);
1414 
1415 /*
1416  * invalidate_bh_lrus() is called rarely - but not only at unmount.
1417  * This doesn't race because it runs in each cpu either in irq
1418  * or with preempt disabled.
1419  */
1420 static void invalidate_bh_lru(void *arg)
1421 {
1422         struct bh_lru *b = &get_cpu_var(bh_lrus);
1423         int i;
1424 
1425         for (i = 0; i < BH_LRU_SIZE; i++) {
1426                 brelse(b->bhs[i]);
1427                 b->bhs[i] = NULL;
1428         }
1429         put_cpu_var(bh_lrus);
1430 }
1431 
1432 static bool has_bh_in_lru(int cpu, void *dummy)
1433 {
1434         struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu);
1435         int i;
1436         
1437         for (i = 0; i < BH_LRU_SIZE; i++) {
1438                 if (b->bhs[i])
1439                         return 1;
1440         }
1441 
1442         return 0;
1443 }
1444 
1445 void invalidate_bh_lrus(void)
1446 {
1447         on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1, GFP_KERNEL);
1448 }
1449 EXPORT_SYMBOL_GPL(invalidate_bh_lrus);
1450 
1451 void set_bh_page(struct buffer_head *bh,
1452                 struct page *page, unsigned long offset)
1453 {
1454         bh->b_page = page;
1455         BUG_ON(offset >= PAGE_SIZE);
1456         if (PageHighMem(page))
1457                 /*
1458                  * This catches illegal uses and preserves the offset:
1459                  */
1460                 bh->b_data = (char *)(0 + offset);
1461         else
1462                 bh->b_data = page_address(page) + offset;
1463 }
1464 EXPORT_SYMBOL(set_bh_page);
1465 
1466 /*
1467  * Called when truncating a buffer on a page completely.
1468  */
1469 
1470 /* Bits that are cleared during an invalidate */
1471 #define BUFFER_FLAGS_DISCARD \
1472         (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \
1473          1 << BH_Delay | 1 << BH_Unwritten)
1474 
1475 static void discard_buffer(struct buffer_head * bh)
1476 {
1477         unsigned long b_state, b_state_old;
1478 
1479         lock_buffer(bh);
1480         clear_buffer_dirty(bh);
1481         bh->b_bdev = NULL;
1482         b_state = bh->b_state;
1483         for (;;) {
1484                 b_state_old = cmpxchg(&bh->b_state, b_state,
1485                                       (b_state & ~BUFFER_FLAGS_DISCARD));
1486                 if (b_state_old == b_state)
1487                         break;
1488                 b_state = b_state_old;
1489         }
1490         unlock_buffer(bh);
1491 }
1492 
1493 /**
1494  * block_invalidatepage - invalidate part or all of a buffer-backed page
1495  *
1496  * @page: the page which is affected
1497  * @offset: start of the range to invalidate
1498  * @length: length of the range to invalidate
1499  *
1500  * block_invalidatepage() is called when all or part of the page has become
1501  * invalidated by a truncate operation.
1502  *
1503  * block_invalidatepage() does not have to release all buffers, but it must
1504  * ensure that no dirty buffer is left outside @offset and that no I/O
1505  * is underway against any of the blocks which are outside the truncation
1506  * point.  Because the caller is about to free (and possibly reuse) those
1507  * blocks on-disk.
1508  */
1509 void block_invalidatepage(struct page *page, unsigned int offset,
1510                           unsigned int length)
1511 {
1512         struct buffer_head *head, *bh, *next;
1513         unsigned int curr_off = 0;
1514         unsigned int stop = length + offset;
1515 
1516         BUG_ON(!PageLocked(page));
1517         if (!page_has_buffers(page))
1518                 goto out;
1519 
1520         /*
1521          * Check for overflow
1522          */
1523         BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
1524 
1525         head = page_buffers(page);
1526         bh = head;
1527         do {
1528                 unsigned int next_off = curr_off + bh->b_size;
1529                 next = bh->b_this_page;
1530 
1531                 /*
1532                  * Are we still fully in range ?
1533                  */
1534                 if (next_off > stop)
1535                         goto out;
1536 
1537                 /*
1538                  * is this block fully invalidated?
1539                  */
1540                 if (offset <= curr_off)
1541                         discard_buffer(bh);
1542                 curr_off = next_off;
1543                 bh = next;
1544         } while (bh != head);
1545 
1546         /*
1547          * We release buffers only if the entire page is being invalidated.
1548          * The get_block cached value has been unconditionally invalidated,
1549          * so real IO is not possible anymore.
1550          */
1551         if (offset == 0)
1552                 try_to_release_page(page, 0);
1553 out:
1554         return;
1555 }
1556 EXPORT_SYMBOL(block_invalidatepage);
1557 
1558 
1559 /*
1560  * We attach and possibly dirty the buffers atomically wrt
1561  * __set_page_dirty_buffers() via private_lock.  try_to_free_buffers
1562  * is already excluded via the page lock.
1563  */
1564 void create_empty_buffers(struct page *page,
1565                         unsigned long blocksize, unsigned long b_state)
1566 {
1567         struct buffer_head *bh, *head, *tail;
1568 
1569         head = alloc_page_buffers(page, blocksize, 1);
1570         bh = head;
1571         do {
1572                 bh->b_state |= b_state;
1573                 tail = bh;
1574                 bh = bh->b_this_page;
1575         } while (bh);
1576         tail->b_this_page = head;
1577 
1578         spin_lock(&page->mapping->private_lock);
1579         if (PageUptodate(page) || PageDirty(page)) {
1580                 bh = head;
1581                 do {
1582                         if (PageDirty(page))
1583                                 set_buffer_dirty(bh);
1584                         if (PageUptodate(page))
1585                                 set_buffer_uptodate(bh);
1586                         bh = bh->b_this_page;
1587                 } while (bh != head);
1588         }
1589         attach_page_buffers(page, head);
1590         spin_unlock(&page->mapping->private_lock);
1591 }
1592 EXPORT_SYMBOL(create_empty_buffers);
1593 
1594 /*
1595  * We are taking a block for data and we don't want any output from any
1596  * buffer-cache aliases starting from return from that function and
1597  * until the moment when something will explicitly mark the buffer
1598  * dirty (hopefully that will not happen until we will free that block ;-)
1599  * We don't even need to mark it not-uptodate - nobody can expect
1600  * anything from a newly allocated buffer anyway. We used to used
1601  * unmap_buffer() for such invalidation, but that was wrong. We definitely
1602  * don't want to mark the alias unmapped, for example - it would confuse
1603  * anyone who might pick it with bread() afterwards...
1604  *
1605  * Also..  Note that bforget() doesn't lock the buffer.  So there can
1606  * be writeout I/O going on against recently-freed buffers.  We don't
1607  * wait on that I/O in bforget() - it's more efficient to wait on the I/O
1608  * only if we really need to.  That happens here.
1609  */
1610 void unmap_underlying_metadata(struct block_device *bdev, sector_t block)
1611 {
1612         struct buffer_head *old_bh;
1613 
1614         might_sleep();
1615 
1616         old_bh = __find_get_block_slow(bdev, block);
1617         if (old_bh) {
1618                 clear_buffer_dirty(old_bh);
1619                 wait_on_buffer(old_bh);
1620                 clear_buffer_req(old_bh);
1621                 __brelse(old_bh);
1622         }
1623 }
1624 EXPORT_SYMBOL(unmap_underlying_metadata);
1625 
1626 /*
1627  * Size is a power-of-two in the range 512..PAGE_SIZE,
1628  * and the case we care about most is PAGE_SIZE.
1629  *
1630  * So this *could* possibly be written with those
1631  * constraints in mind (relevant mostly if some
1632  * architecture has a slow bit-scan instruction)
1633  */
1634 static inline int block_size_bits(unsigned int blocksize)
1635 {
1636         return ilog2(blocksize);
1637 }
1638 
1639 static struct buffer_head *create_page_buffers(struct page *page, struct inode *inode, unsigned int b_state)
1640 {
1641         BUG_ON(!PageLocked(page));
1642 
1643         if (!page_has_buffers(page))
1644                 create_empty_buffers(page, 1 << ACCESS_ONCE(inode->i_blkbits), b_state);
1645         return page_buffers(page);
1646 }
1647 
1648 /*
1649  * NOTE! All mapped/uptodate combinations are valid:
1650  *
1651  *      Mapped  Uptodate        Meaning
1652  *
1653  *      No      No              "unknown" - must do get_block()
1654  *      No      Yes             "hole" - zero-filled
1655  *      Yes     No              "allocated" - allocated on disk, not read in
1656  *      Yes     Yes             "valid" - allocated and up-to-date in memory.
1657  *
1658  * "Dirty" is valid only with the last case (mapped+uptodate).
1659  */
1660 
1661 /*
1662  * While block_write_full_page is writing back the dirty buffers under
1663  * the page lock, whoever dirtied the buffers may decide to clean them
1664  * again at any time.  We handle that by only looking at the buffer
1665  * state inside lock_buffer().
1666  *
1667  * If block_write_full_page() is called for regular writeback
1668  * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1669  * locked buffer.   This only can happen if someone has written the buffer
1670  * directly, with submit_bh().  At the address_space level PageWriteback
1671  * prevents this contention from occurring.
1672  *
1673  * If block_write_full_page() is called with wbc->sync_mode ==
1674  * WB_SYNC_ALL, the writes are posted using WRITE_SYNC; this
1675  * causes the writes to be flagged as synchronous writes.
1676  */
1677 static int __block_write_full_page(struct inode *inode, struct page *page,
1678                         get_block_t *get_block, struct writeback_control *wbc,
1679                         bh_end_io_t *handler)
1680 {
1681         int err;
1682         sector_t block;
1683         sector_t last_block;
1684         struct buffer_head *bh, *head;
1685         unsigned int blocksize, bbits;
1686         int nr_underway = 0;
1687         int write_op = (wbc->sync_mode == WB_SYNC_ALL ?
1688                         WRITE_SYNC : WRITE);
1689 
1690         head = create_page_buffers(page, inode,
1691                                         (1 << BH_Dirty)|(1 << BH_Uptodate));
1692 
1693         /*
1694          * Be very careful.  We have no exclusion from __set_page_dirty_buffers
1695          * here, and the (potentially unmapped) buffers may become dirty at
1696          * any time.  If a buffer becomes dirty here after we've inspected it
1697          * then we just miss that fact, and the page stays dirty.
1698          *
1699          * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
1700          * handle that here by just cleaning them.
1701          */
1702 
1703         bh = head;
1704         blocksize = bh->b_size;
1705         bbits = block_size_bits(blocksize);
1706 
1707         block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits);
1708         last_block = (i_size_read(inode) - 1) >> bbits;
1709 
1710         /*
1711          * Get all the dirty buffers mapped to disk addresses and
1712          * handle any aliases from the underlying blockdev's mapping.
1713          */
1714         do {
1715                 if (block > last_block) {
1716                         /*
1717                          * mapped buffers outside i_size will occur, because
1718                          * this page can be outside i_size when there is a
1719                          * truncate in progress.
1720                          */
1721                         /*
1722                          * The buffer was zeroed by block_write_full_page()
1723                          */
1724                         clear_buffer_dirty(bh);
1725                         set_buffer_uptodate(bh);
1726                 } else if ((!buffer_mapped(bh) || buffer_delay(bh)) &&
1727                            buffer_dirty(bh)) {
1728                         WARN_ON(bh->b_size != blocksize);
1729                         err = get_block(inode, block, bh, 1);
1730                         if (err)
1731                                 goto recover;
1732                         clear_buffer_delay(bh);
1733                         if (buffer_new(bh)) {
1734                                 /* blockdev mappings never come here */
1735                                 clear_buffer_new(bh);
1736                                 unmap_underlying_metadata(bh->b_bdev,
1737                                                         bh->b_blocknr);
1738                         }
1739                 }
1740                 bh = bh->b_this_page;
1741                 block++;
1742         } while (bh != head);
1743 
1744         do {
1745                 if (!buffer_mapped(bh))
1746                         continue;
1747                 /*
1748                  * If it's a fully non-blocking write attempt and we cannot
1749                  * lock the buffer then redirty the page.  Note that this can
1750                  * potentially cause a busy-wait loop from writeback threads
1751                  * and kswapd activity, but those code paths have their own
1752                  * higher-level throttling.
1753                  */
1754                 if (wbc->sync_mode != WB_SYNC_NONE) {
1755                         lock_buffer(bh);
1756                 } else if (!trylock_buffer(bh)) {
1757                         redirty_page_for_writepage(wbc, page);
1758                         continue;
1759                 }
1760                 if (test_clear_buffer_dirty(bh)) {
1761                         mark_buffer_async_write_endio(bh, handler);
1762                 } else {
1763                         unlock_buffer(bh);
1764                 }
1765         } while ((bh = bh->b_this_page) != head);
1766 
1767         /*
1768          * The page and its buffers are protected by PageWriteback(), so we can
1769          * drop the bh refcounts early.
1770          */
1771         BUG_ON(PageWriteback(page));
1772         set_page_writeback(page);
1773 
1774         do {
1775                 struct buffer_head *next = bh->b_this_page;
1776                 if (buffer_async_write(bh)) {
1777                         submit_bh(write_op, bh);
1778                         nr_underway++;
1779                 }
1780                 bh = next;
1781         } while (bh != head);
1782         unlock_page(page);
1783 
1784         err = 0;
1785 done:
1786         if (nr_underway == 0) {
1787                 /*
1788                  * The page was marked dirty, but the buffers were
1789                  * clean.  Someone wrote them back by hand with
1790                  * ll_rw_block/submit_bh.  A rare case.
1791                  */
1792                 end_page_writeback(page);
1793 
1794                 /*
1795                  * The page and buffer_heads can be released at any time from
1796                  * here on.
1797                  */
1798         }
1799         return err;
1800 
1801 recover:
1802         /*
1803          * ENOSPC, or some other error.  We may already have added some
1804          * blocks to the file, so we need to write these out to avoid
1805          * exposing stale data.
1806          * The page is currently locked and not marked for writeback
1807          */
1808         bh = head;
1809         /* Recovery: lock and submit the mapped buffers */
1810         do {
1811                 if (buffer_mapped(bh) && buffer_dirty(bh) &&
1812                     !buffer_delay(bh)) {
1813                         lock_buffer(bh);
1814                         mark_buffer_async_write_endio(bh, handler);
1815                 } else {
1816                         /*
1817                          * The buffer may have been set dirty during
1818                          * attachment to a dirty page.
1819                          */
1820                         clear_buffer_dirty(bh);
1821                 }
1822         } while ((bh = bh->b_this_page) != head);
1823         SetPageError(page);
1824         BUG_ON(PageWriteback(page));
1825         mapping_set_error(page->mapping, err);
1826         set_page_writeback(page);
1827         do {
1828                 struct buffer_head *next = bh->b_this_page;
1829                 if (buffer_async_write(bh)) {
1830                         clear_buffer_dirty(bh);
1831                         submit_bh(write_op, bh);
1832                         nr_underway++;
1833                 }
1834                 bh = next;
1835         } while (bh != head);
1836         unlock_page(page);
1837         goto done;
1838 }
1839 
1840 /*
1841  * If a page has any new buffers, zero them out here, and mark them uptodate
1842  * and dirty so they'll be written out (in order to prevent uninitialised
1843  * block data from leaking). And clear the new bit.
1844  */
1845 void page_zero_new_buffers(struct page *page, unsigned from, unsigned to)
1846 {
1847         unsigned int block_start, block_end;
1848         struct buffer_head *head, *bh;
1849 
1850         BUG_ON(!PageLocked(page));
1851         if (!page_has_buffers(page))
1852                 return;
1853 
1854         bh = head = page_buffers(page);
1855         block_start = 0;
1856         do {
1857                 block_end = block_start + bh->b_size;
1858 
1859                 if (buffer_new(bh)) {
1860                         if (block_end > from && block_start < to) {
1861                                 if (!PageUptodate(page)) {
1862                                         unsigned start, size;
1863 
1864                                         start = max(from, block_start);
1865                                         size = min(to, block_end) - start;
1866 
1867                                         zero_user(page, start, size);
1868                                         set_buffer_uptodate(bh);
1869                                 }
1870 
1871                                 clear_buffer_new(bh);
1872                                 mark_buffer_dirty(bh);
1873                         }
1874                 }
1875 
1876                 block_start = block_end;
1877                 bh = bh->b_this_page;
1878         } while (bh != head);
1879 }
1880 EXPORT_SYMBOL(page_zero_new_buffers);
1881 
1882 int __block_write_begin(struct page *page, loff_t pos, unsigned len,
1883                 get_block_t *get_block)
1884 {
1885         unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1886         unsigned to = from + len;
1887         struct inode *inode = page->mapping->host;
1888         unsigned block_start, block_end;
1889         sector_t block;
1890         int err = 0;
1891         unsigned blocksize, bbits;
1892         struct buffer_head *bh, *head, *wait[2], **wait_bh=wait;
1893 
1894         BUG_ON(!PageLocked(page));
1895         BUG_ON(from > PAGE_CACHE_SIZE);
1896         BUG_ON(to > PAGE_CACHE_SIZE);
1897         BUG_ON(from > to);
1898 
1899         head = create_page_buffers(page, inode, 0);
1900         blocksize = head->b_size;
1901         bbits = block_size_bits(blocksize);
1902 
1903         block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits);
1904 
1905         for(bh = head, block_start = 0; bh != head || !block_start;
1906             block++, block_start=block_end, bh = bh->b_this_page) {
1907                 block_end = block_start + blocksize;
1908                 if (block_end <= from || block_start >= to) {
1909                         if (PageUptodate(page)) {
1910                                 if (!buffer_uptodate(bh))
1911                                         set_buffer_uptodate(bh);
1912                         }
1913                         continue;
1914                 }
1915                 if (buffer_new(bh))
1916                         clear_buffer_new(bh);
1917                 if (!buffer_mapped(bh)) {
1918                         WARN_ON(bh->b_size != blocksize);
1919                         err = get_block(inode, block, bh, 1);
1920                         if (err)
1921                                 break;
1922                         if (buffer_new(bh)) {
1923                                 unmap_underlying_metadata(bh->b_bdev,
1924                                                         bh->b_blocknr);
1925                                 if (PageUptodate(page)) {
1926                                         clear_buffer_new(bh);
1927                                         set_buffer_uptodate(bh);
1928                                         mark_buffer_dirty(bh);
1929                                         continue;
1930                                 }
1931                                 if (block_end > to || block_start < from)
1932                                         zero_user_segments(page,
1933                                                 to, block_end,
1934                                                 block_start, from);
1935                                 continue;
1936                         }
1937                 }
1938                 if (PageUptodate(page)) {
1939                         if (!buffer_uptodate(bh))
1940                                 set_buffer_uptodate(bh);
1941                         continue; 
1942                 }
1943                 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
1944                     !buffer_unwritten(bh) &&
1945                      (block_start < from || block_end > to)) {
1946                         ll_rw_block(READ, 1, &bh);
1947                         *wait_bh++=bh;
1948                 }
1949         }
1950         /*
1951          * If we issued read requests - let them complete.
1952          */
1953         while(wait_bh > wait) {
1954                 wait_on_buffer(*--wait_bh);
1955                 if (!buffer_uptodate(*wait_bh))
1956                         err = -EIO;
1957         }
1958         if (unlikely(err))
1959                 page_zero_new_buffers(page, from, to);
1960         return err;
1961 }
1962 EXPORT_SYMBOL(__block_write_begin);
1963 
1964 static int __block_commit_write(struct inode *inode, struct page *page,
1965                 unsigned from, unsigned to)
1966 {
1967         unsigned block_start, block_end;
1968         int partial = 0;
1969         unsigned blocksize;
1970         struct buffer_head *bh, *head;
1971 
1972         bh = head = page_buffers(page);
1973         blocksize = bh->b_size;
1974 
1975         block_start = 0;
1976         do {
1977                 block_end = block_start + blocksize;
1978                 if (block_end <= from || block_start >= to) {
1979                         if (!buffer_uptodate(bh))
1980                                 partial = 1;
1981                 } else {
1982                         set_buffer_uptodate(bh);
1983                         mark_buffer_dirty(bh);
1984                 }
1985                 clear_buffer_new(bh);
1986 
1987                 block_start = block_end;
1988                 bh = bh->b_this_page;
1989         } while (bh != head);
1990 
1991         /*
1992          * If this is a partial write which happened to make all buffers
1993          * uptodate then we can optimize away a bogus readpage() for
1994          * the next read(). Here we 'discover' whether the page went
1995          * uptodate as a result of this (potentially partial) write.
1996          */
1997         if (!partial)
1998                 SetPageUptodate(page);
1999         return 0;
2000 }
2001 
2002 /*
2003  * block_write_begin takes care of the basic task of block allocation and
2004  * bringing partial write blocks uptodate first.
2005  *
2006  * The filesystem needs to handle block truncation upon failure.
2007  */
2008 int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,
2009                 unsigned flags, struct page **pagep, get_block_t *get_block)
2010 {
2011         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
2012         struct page *page;
2013         int status;
2014 
2015         page = grab_cache_page_write_begin(mapping, index, flags);
2016         if (!page)
2017                 return -ENOMEM;
2018 
2019         status = __block_write_begin(page, pos, len, get_block);
2020         if (unlikely(status)) {
2021                 unlock_page(page);
2022                 page_cache_release(page);
2023                 page = NULL;
2024         }
2025 
2026         *pagep = page;
2027         return status;
2028 }
2029 EXPORT_SYMBOL(block_write_begin);
2030 
2031 int block_write_end(struct file *file, struct address_space *mapping,
2032                         loff_t pos, unsigned len, unsigned copied,
2033                         struct page *page, void *fsdata)
2034 {
2035         struct inode *inode = mapping->host;
2036         unsigned start;
2037 
2038         start = pos & (PAGE_CACHE_SIZE - 1);
2039 
2040         if (unlikely(copied < len)) {
2041                 /*
2042                  * The buffers that were written will now be uptodate, so we
2043                  * don't have to worry about a readpage reading them and
2044                  * overwriting a partial write. However if we have encountered
2045                  * a short write and only partially written into a buffer, it
2046                  * will not be marked uptodate, so a readpage might come in and
2047                  * destroy our partial write.
2048                  *
2049                  * Do the simplest thing, and just treat any short write to a
2050                  * non uptodate page as a zero-length write, and force the
2051                  * caller to redo the whole thing.
2052                  */
2053                 if (!PageUptodate(page))
2054                         copied = 0;
2055 
2056                 page_zero_new_buffers(page, start+copied, start+len);
2057         }
2058         flush_dcache_page(page);
2059 
2060         /* This could be a short (even 0-length) commit */
2061         __block_commit_write(inode, page, start, start+copied);
2062 
2063         return copied;
2064 }
2065 EXPORT_SYMBOL(block_write_end);
2066 
2067 int generic_write_end(struct file *file, struct address_space *mapping,
2068                         loff_t pos, unsigned len, unsigned copied,
2069                         struct page *page, void *fsdata)
2070 {
2071         struct inode *inode = mapping->host;
2072         loff_t old_size = inode->i_size;
2073         int i_size_changed = 0;
2074 
2075         copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
2076 
2077         /*
2078          * No need to use i_size_read() here, the i_size
2079          * cannot change under us because we hold i_mutex.
2080          *
2081          * But it's important to update i_size while still holding page lock:
2082          * page writeout could otherwise come in and zero beyond i_size.
2083          */
2084         if (pos+copied > inode->i_size) {
2085                 i_size_write(inode, pos+copied);
2086                 i_size_changed = 1;
2087         }
2088 
2089         unlock_page(page);
2090         page_cache_release(page);
2091 
2092         if (old_size < pos)
2093                 pagecache_isize_extended(inode, old_size, pos);
2094         /*
2095          * Don't mark the inode dirty under page lock. First, it unnecessarily
2096          * makes the holding time of page lock longer. Second, it forces lock
2097          * ordering of page lock and transaction start for journaling
2098          * filesystems.
2099          */
2100         if (i_size_changed)
2101                 mark_inode_dirty(inode);
2102 
2103         return copied;
2104 }
2105 EXPORT_SYMBOL(generic_write_end);
2106 
2107 /*
2108  * block_is_partially_uptodate checks whether buffers within a page are
2109  * uptodate or not.
2110  *
2111  * Returns true if all buffers which correspond to a file portion
2112  * we want to read are uptodate.
2113  */
2114 int block_is_partially_uptodate(struct page *page, unsigned long from,
2115                                         unsigned long count)
2116 {
2117         unsigned block_start, block_end, blocksize;
2118         unsigned to;
2119         struct buffer_head *bh, *head;
2120         int ret = 1;
2121 
2122         if (!page_has_buffers(page))
2123                 return 0;
2124 
2125         head = page_buffers(page);
2126         blocksize = head->b_size;
2127         to = min_t(unsigned, PAGE_CACHE_SIZE - from, count);
2128         to = from + to;
2129         if (from < blocksize && to > PAGE_CACHE_SIZE - blocksize)
2130                 return 0;
2131 
2132         bh = head;
2133         block_start = 0;
2134         do {
2135                 block_end = block_start + blocksize;
2136                 if (block_end > from && block_start < to) {
2137                         if (!buffer_uptodate(bh)) {
2138                                 ret = 0;
2139                                 break;
2140                         }
2141                         if (block_end >= to)
2142                                 break;
2143                 }
2144                 block_start = block_end;
2145                 bh = bh->b_this_page;
2146         } while (bh != head);
2147 
2148         return ret;
2149 }
2150 EXPORT_SYMBOL(block_is_partially_uptodate);
2151 
2152 /*
2153  * Generic "read page" function for block devices that have the normal
2154  * get_block functionality. This is most of the block device filesystems.
2155  * Reads the page asynchronously --- the unlock_buffer() and
2156  * set/clear_buffer_uptodate() functions propagate buffer state into the
2157  * page struct once IO has completed.
2158  */
2159 int block_read_full_page(struct page *page, get_block_t *get_block)
2160 {
2161         struct inode *inode = page->mapping->host;
2162         sector_t iblock, lblock;
2163         struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
2164         unsigned int blocksize, bbits;
2165         int nr, i;
2166         int fully_mapped = 1;
2167 
2168         head = create_page_buffers(page, inode, 0);
2169         blocksize = head->b_size;
2170         bbits = block_size_bits(blocksize);
2171 
2172         iblock = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits);
2173         lblock = (i_size_read(inode)+blocksize-1) >> bbits;
2174         bh = head;
2175         nr = 0;
2176         i = 0;
2177 
2178         do {
2179                 if (buffer_uptodate(bh))
2180                         continue;
2181 
2182                 if (!buffer_mapped(bh)) {
2183                         int err = 0;
2184 
2185                         fully_mapped = 0;
2186                         if (iblock < lblock) {
2187                                 WARN_ON(bh->b_size != blocksize);
2188                                 err = get_block(inode, iblock, bh, 0);
2189                                 if (err)
2190                                         SetPageError(page);
2191                         }
2192                         if (!buffer_mapped(bh)) {
2193                                 zero_user(page, i * blocksize, blocksize);
2194                                 if (!err)
2195                                         set_buffer_uptodate(bh);
2196                                 continue;
2197                         }
2198                         /*
2199                          * get_block() might have updated the buffer
2200                          * synchronously
2201                          */
2202                         if (buffer_uptodate(bh))
2203                                 continue;
2204                 }
2205                 arr[nr++] = bh;
2206         } while (i++, iblock++, (bh = bh->b_this_page) != head);
2207 
2208         if (fully_mapped)
2209                 SetPageMappedToDisk(page);
2210 
2211         if (!nr) {
2212                 /*
2213                  * All buffers are uptodate - we can set the page uptodate
2214                  * as well. But not if get_block() returned an error.
2215                  */
2216                 if (!PageError(page))
2217                         SetPageUptodate(page);
2218                 unlock_page(page);
2219                 return 0;
2220         }
2221 
2222         /* Stage two: lock the buffers */
2223         for (i = 0; i < nr; i++) {
2224                 bh = arr[i];
2225                 lock_buffer(bh);
2226                 mark_buffer_async_read(bh);
2227         }
2228 
2229         /*
2230          * Stage 3: start the IO.  Check for uptodateness
2231          * inside the buffer lock in case another process reading
2232          * the underlying blockdev brought it uptodate (the sct fix).
2233          */
2234         for (i = 0; i < nr; i++) {
2235                 bh = arr[i];
2236                 if (buffer_uptodate(bh))
2237                         end_buffer_async_read(bh, 1);
2238                 else
2239                         submit_bh(READ, bh);
2240         }
2241         return 0;
2242 }
2243 EXPORT_SYMBOL(block_read_full_page);
2244 
2245 /* utility function for filesystems that need to do work on expanding
2246  * truncates.  Uses filesystem pagecache writes to allow the filesystem to
2247  * deal with the hole.  
2248  */
2249 int generic_cont_expand_simple(struct inode *inode, loff_t size)
2250 {
2251         struct address_space *mapping = inode->i_mapping;
2252         struct page *page;
2253         void *fsdata;
2254         int err;
2255 
2256         err = inode_newsize_ok(inode, size);
2257         if (err)
2258                 goto out;
2259 
2260         err = pagecache_write_begin(NULL, mapping, size, 0,
2261                                 AOP_FLAG_UNINTERRUPTIBLE|AOP_FLAG_CONT_EXPAND,
2262                                 &page, &fsdata);
2263         if (err)
2264                 goto out;
2265 
2266         err = pagecache_write_end(NULL, mapping, size, 0, 0, page, fsdata);
2267         BUG_ON(err > 0);
2268 
2269 out:
2270         return err;
2271 }
2272 EXPORT_SYMBOL(generic_cont_expand_simple);
2273 
2274 static int cont_expand_zero(struct file *file, struct address_space *mapping,
2275                             loff_t pos, loff_t *bytes)
2276 {
2277         struct inode *inode = mapping->host;
2278         unsigned blocksize = 1 << inode->i_blkbits;
2279         struct page *page;
2280         void *fsdata;
2281         pgoff_t index, curidx;
2282         loff_t curpos;
2283         unsigned zerofrom, offset, len;
2284         int err = 0;
2285 
2286         index = pos >> PAGE_CACHE_SHIFT;
2287         offset = pos & ~PAGE_CACHE_MASK;
2288 
2289         while (index > (curidx = (curpos = *bytes)>>PAGE_CACHE_SHIFT)) {
2290                 zerofrom = curpos & ~PAGE_CACHE_MASK;
2291                 if (zerofrom & (blocksize-1)) {
2292                         *bytes |= (blocksize-1);
2293                         (*bytes)++;
2294                 }
2295                 len = PAGE_CACHE_SIZE - zerofrom;
2296 
2297                 err = pagecache_write_begin(file, mapping, curpos, len,
2298                                                 AOP_FLAG_UNINTERRUPTIBLE,
2299                                                 &page, &fsdata);
2300                 if (err)
2301                         goto out;
2302                 zero_user(page, zerofrom, len);
2303                 err = pagecache_write_end(file, mapping, curpos, len, len,
2304                                                 page, fsdata);
2305                 if (err < 0)
2306                         goto out;
2307                 BUG_ON(err != len);
2308                 err = 0;
2309 
2310                 balance_dirty_pages_ratelimited(mapping);
2311 
2312                 if (unlikely(fatal_signal_pending(current))) {
2313                         err = -EINTR;
2314                         goto out;
2315                 }
2316         }
2317 
2318         /* page covers the boundary, find the boundary offset */
2319         if (index == curidx) {
2320                 zerofrom = curpos & ~PAGE_CACHE_MASK;
2321                 /* if we will expand the thing last block will be filled */
2322                 if (offset <= zerofrom) {
2323                         goto out;
2324                 }
2325                 if (zerofrom & (blocksize-1)) {
2326                         *bytes |= (blocksize-1);
2327                         (*bytes)++;
2328                 }
2329                 len = offset - zerofrom;
2330 
2331                 err = pagecache_write_begin(file, mapping, curpos, len,
2332                                                 AOP_FLAG_UNINTERRUPTIBLE,
2333                                                 &page, &fsdata);
2334                 if (err)
2335                         goto out;
2336                 zero_user(page, zerofrom, len);
2337                 err = pagecache_write_end(file, mapping, curpos, len, len,
2338                                                 page, fsdata);
2339                 if (err < 0)
2340                         goto out;
2341                 BUG_ON(err != len);
2342                 err = 0;
2343         }
2344 out:
2345         return err;
2346 }
2347 
2348 /*
2349  * For moronic filesystems that do not allow holes in file.
2350  * We may have to extend the file.
2351  */
2352 int cont_write_begin(struct file *file, struct address_space *mapping,
2353                         loff_t pos, unsigned len, unsigned flags,
2354                         struct page **pagep, void **fsdata,
2355                         get_block_t *get_block, loff_t *bytes)
2356 {
2357         struct inode *inode = mapping->host;
2358         unsigned blocksize = 1 << inode->i_blkbits;
2359         unsigned zerofrom;
2360         int err;
2361 
2362         err = cont_expand_zero(file, mapping, pos, bytes);
2363         if (err)
2364                 return err;
2365 
2366         zerofrom = *bytes & ~PAGE_CACHE_MASK;
2367         if (pos+len > *bytes && zerofrom & (blocksize-1)) {
2368                 *bytes |= (blocksize-1);
2369                 (*bytes)++;
2370         }
2371 
2372         return block_write_begin(mapping, pos, len, flags, pagep, get_block);
2373 }
2374 EXPORT_SYMBOL(cont_write_begin);
2375 
2376 int block_commit_write(struct page *page, unsigned from, unsigned to)
2377 {
2378         struct inode *inode = page->mapping->host;
2379         __block_commit_write(inode,page,from,to);
2380         return 0;
2381 }
2382 EXPORT_SYMBOL(block_commit_write);
2383 
2384 /*
2385  * block_page_mkwrite() is not allowed to change the file size as it gets
2386  * called from a page fault handler when a page is first dirtied. Hence we must
2387  * be careful to check for EOF conditions here. We set the page up correctly
2388  * for a written page which means we get ENOSPC checking when writing into
2389  * holes and correct delalloc and unwritten extent mapping on filesystems that
2390  * support these features.
2391  *
2392  * We are not allowed to take the i_mutex here so we have to play games to
2393  * protect against truncate races as the page could now be beyond EOF.  Because
2394  * truncate writes the inode size before removing pages, once we have the
2395  * page lock we can determine safely if the page is beyond EOF. If it is not
2396  * beyond EOF, then the page is guaranteed safe against truncation until we
2397  * unlock the page.
2398  *
2399  * Direct callers of this function should protect against filesystem freezing
2400  * using sb_start_write() - sb_end_write() functions.
2401  */
2402 int __block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
2403                          get_block_t get_block)
2404 {
2405         struct page *page = vmf->page;
2406         struct inode *inode = file_inode(vma->vm_file);
2407         unsigned long end;
2408         loff_t size;
2409         int ret;
2410 
2411         lock_page(page);
2412         size = i_size_read(inode);
2413         if ((page->mapping != inode->i_mapping) ||
2414             (page_offset(page) > size)) {
2415                 /* We overload EFAULT to mean page got truncated */
2416                 ret = -EFAULT;
2417                 goto out_unlock;
2418         }
2419 
2420         /* page is wholly or partially inside EOF */
2421         if (((page->index + 1) << PAGE_CACHE_SHIFT) > size)
2422                 end = size & ~PAGE_CACHE_MASK;
2423         else
2424                 end = PAGE_CACHE_SIZE;
2425 
2426         ret = __block_write_begin(page, 0, end, get_block);
2427         if (!ret)
2428                 ret = block_commit_write(page, 0, end);
2429 
2430         if (unlikely(ret < 0))
2431                 goto out_unlock;
2432         set_page_dirty(page);
2433         wait_for_stable_page(page);
2434         return 0;
2435 out_unlock:
2436         unlock_page(page);
2437         return ret;
2438 }
2439 EXPORT_SYMBOL(__block_page_mkwrite);
2440 
2441 int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
2442                    get_block_t get_block)
2443 {
2444         int ret;
2445         struct super_block *sb = file_inode(vma->vm_file)->i_sb;
2446 
2447         sb_start_pagefault(sb);
2448 
2449         /*
2450          * Update file times before taking page lock. We may end up failing the
2451          * fault so this update may be superfluous but who really cares...
2452          */
2453         file_update_time(vma->vm_file);
2454 
2455         ret = __block_page_mkwrite(vma, vmf, get_block);
2456         sb_end_pagefault(sb);
2457         return block_page_mkwrite_return(ret);
2458 }
2459 EXPORT_SYMBOL(block_page_mkwrite);
2460 
2461 /*
2462  * nobh_write_begin()'s prereads are special: the buffer_heads are freed
2463  * immediately, while under the page lock.  So it needs a special end_io
2464  * handler which does not touch the bh after unlocking it.
2465  */
2466 static void end_buffer_read_nobh(struct buffer_head *bh, int uptodate)
2467 {
2468         __end_buffer_read_notouch(bh, uptodate);
2469 }
2470 
2471 /*
2472  * Attach the singly-linked list of buffers created by nobh_write_begin, to
2473  * the page (converting it to circular linked list and taking care of page
2474  * dirty races).
2475  */
2476 static void attach_nobh_buffers(struct page *page, struct buffer_head *head)
2477 {
2478         struct buffer_head *bh;
2479 
2480         BUG_ON(!PageLocked(page));
2481 
2482         spin_lock(&page->mapping->private_lock);
2483         bh = head;
2484         do {
2485                 if (PageDirty(page))
2486                         set_buffer_dirty(bh);
2487                 if (!bh->b_this_page)
2488                         bh->b_this_page = head;
2489                 bh = bh->b_this_page;
2490         } while (bh != head);
2491         attach_page_buffers(page, head);
2492         spin_unlock(&page->mapping->private_lock);
2493 }
2494 
2495 /*
2496  * On entry, the page is fully not uptodate.
2497  * On exit the page is fully uptodate in the areas outside (from,to)
2498  * The filesystem needs to handle block truncation upon failure.
2499  */
2500 int nobh_write_begin(struct address_space *mapping,
2501                         loff_t pos, unsigned len, unsigned flags,
2502                         struct page **pagep, void **fsdata,
2503                         get_block_t *get_block)
2504 {
2505         struct inode *inode = mapping->host;
2506         const unsigned blkbits = inode->i_blkbits;
2507         const unsigned blocksize = 1 << blkbits;
2508         struct buffer_head *head, *bh;
2509         struct page *page;
2510         pgoff_t index;
2511         unsigned from, to;
2512         unsigned block_in_page;
2513         unsigned block_start, block_end;
2514         sector_t block_in_file;
2515         int nr_reads = 0;
2516         int ret = 0;
2517         int is_mapped_to_disk = 1;
2518 
2519         index = pos >> PAGE_CACHE_SHIFT;
2520         from = pos & (PAGE_CACHE_SIZE - 1);
2521         to = from + len;
2522 
2523         page = grab_cache_page_write_begin(mapping, index, flags);
2524         if (!page)
2525                 return -ENOMEM;
2526         *pagep = page;
2527         *fsdata = NULL;
2528 
2529         if (page_has_buffers(page)) {
2530                 ret = __block_write_begin(page, pos, len, get_block);
2531                 if (unlikely(ret))
2532                         goto out_release;
2533                 return ret;
2534         }
2535 
2536         if (PageMappedToDisk(page))
2537                 return 0;
2538 
2539         /*
2540          * Allocate buffers so that we can keep track of state, and potentially
2541          * attach them to the page if an error occurs. In the common case of
2542          * no error, they will just be freed again without ever being attached
2543          * to the page (which is all OK, because we're under the page lock).
2544          *
2545          * Be careful: the buffer linked list is a NULL terminated one, rather
2546          * than the circular one we're used to.
2547          */
2548         head = alloc_page_buffers(page, blocksize, 0);
2549         if (!head) {
2550                 ret = -ENOMEM;
2551                 goto out_release;
2552         }
2553 
2554         block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
2555 
2556         /*
2557          * We loop across all blocks in the page, whether or not they are
2558          * part of the affected region.  This is so we can discover if the
2559          * page is fully mapped-to-disk.
2560          */
2561         for (block_start = 0, block_in_page = 0, bh = head;
2562                   block_start < PAGE_CACHE_SIZE;
2563                   block_in_page++, block_start += blocksize, bh = bh->b_this_page) {
2564                 int create;
2565 
2566                 block_end = block_start + blocksize;
2567                 bh->b_state = 0;
2568                 create = 1;
2569                 if (block_start >= to)
2570                         create = 0;
2571                 ret = get_block(inode, block_in_file + block_in_page,
2572                                         bh, create);
2573                 if (ret)
2574                         goto failed;
2575                 if (!buffer_mapped(bh))
2576                         is_mapped_to_disk = 0;
2577                 if (buffer_new(bh))
2578                         unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
2579                 if (PageUptodate(page)) {
2580                         set_buffer_uptodate(bh);
2581                         continue;
2582                 }
2583                 if (buffer_new(bh) || !buffer_mapped(bh)) {
2584                         zero_user_segments(page, block_start, from,
2585                                                         to, block_end);
2586                         continue;
2587                 }
2588                 if (buffer_uptodate(bh))
2589                         continue;       /* reiserfs does this */
2590                 if (block_start < from || block_end > to) {
2591                         lock_buffer(bh);
2592                         bh->b_end_io = end_buffer_read_nobh;
2593                         submit_bh(READ, bh);
2594                         nr_reads++;
2595                 }
2596         }
2597 
2598         if (nr_reads) {
2599                 /*
2600                  * The page is locked, so these buffers are protected from
2601                  * any VM or truncate activity.  Hence we don't need to care
2602                  * for the buffer_head refcounts.
2603                  */
2604                 for (bh = head; bh; bh = bh->b_this_page) {
2605                         wait_on_buffer(bh);
2606                         if (!buffer_uptodate(bh))
2607                                 ret = -EIO;
2608                 }
2609                 if (ret)
2610                         goto failed;
2611         }
2612 
2613         if (is_mapped_to_disk)
2614                 SetPageMappedToDisk(page);
2615 
2616         *fsdata = head; /* to be released by nobh_write_end */
2617 
2618         return 0;
2619 
2620 failed:
2621         BUG_ON(!ret);
2622         /*
2623          * Error recovery is a bit difficult. We need to zero out blocks that
2624          * were newly allocated, and dirty them to ensure they get written out.
2625          * Buffers need to be attached to the page at this point, otherwise
2626          * the handling of potential IO errors during writeout would be hard
2627          * (could try doing synchronous writeout, but what if that fails too?)
2628          */
2629         attach_nobh_buffers(page, head);
2630         page_zero_new_buffers(page, from, to);
2631 
2632 out_release:
2633         unlock_page(page);
2634         page_cache_release(page);
2635         *pagep = NULL;
2636 
2637         return ret;
2638 }
2639 EXPORT_SYMBOL(nobh_write_begin);
2640 
2641 int nobh_write_end(struct file *file, struct address_space *mapping,
2642                         loff_t pos, unsigned len, unsigned copied,
2643                         struct page *page, void *fsdata)
2644 {
2645         struct inode *inode = page->mapping->host;
2646         struct buffer_head *head = fsdata;
2647         struct buffer_head *bh;
2648         BUG_ON(fsdata != NULL && page_has_buffers(page));
2649 
2650         if (unlikely(copied < len) && head)
2651                 attach_nobh_buffers(page, head);
2652         if (page_has_buffers(page))
2653                 return generic_write_end(file, mapping, pos, len,
2654                                         copied, page, fsdata);
2655 
2656         SetPageUptodate(page);
2657         set_page_dirty(page);
2658         if (pos+copied > inode->i_size) {
2659                 i_size_write(inode, pos+copied);
2660                 mark_inode_dirty(inode);
2661         }
2662 
2663         unlock_page(page);
2664         page_cache_release(page);
2665 
2666         while (head) {
2667                 bh = head;
2668                 head = head->b_this_page;
2669                 free_buffer_head(bh);
2670         }
2671 
2672         return copied;
2673 }
2674 EXPORT_SYMBOL(nobh_write_end);
2675 
2676 /*
2677  * nobh_writepage() - based on block_full_write_page() except
2678  * that it tries to operate without attaching bufferheads to
2679  * the page.
2680  */
2681 int nobh_writepage(struct page *page, get_block_t *get_block,
2682                         struct writeback_control *wbc)
2683 {
2684         struct inode * const inode = page->mapping->host;
2685         loff_t i_size = i_size_read(inode);
2686         const pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
2687         unsigned offset;
2688         int ret;
2689 
2690         /* Is the page fully inside i_size? */
2691         if (page->index < end_index)
2692                 goto out;
2693 
2694         /* Is the page fully outside i_size? (truncate in progress) */
2695         offset = i_size & (PAGE_CACHE_SIZE-1);
2696         if (page->index >= end_index+1 || !offset) {
2697                 /*
2698                  * The page may have dirty, unmapped buffers.  For example,
2699                  * they may have been added in ext3_writepage().  Make them
2700                  * freeable here, so the page does not leak.
2701                  */
2702 #if 0
2703                 /* Not really sure about this  - do we need this ? */
2704                 if (page->mapping->a_ops->invalidatepage)
2705                         page->mapping->a_ops->invalidatepage(page, offset);
2706 #endif
2707                 unlock_page(page);
2708                 return 0; /* don't care */
2709         }
2710 
2711         /*
2712          * The page straddles i_size.  It must be zeroed out on each and every
2713          * writepage invocation because it may be mmapped.  "A file is mapped
2714          * in multiples of the page size.  For a file that is not a multiple of
2715          * the  page size, the remaining memory is zeroed when mapped, and
2716          * writes to that region are not written out to the file."
2717          */
2718         zero_user_segment(page, offset, PAGE_CACHE_SIZE);
2719 out:
2720         ret = mpage_writepage(page, get_block, wbc);
2721         if (ret == -EAGAIN)
2722                 ret = __block_write_full_page(inode, page, get_block, wbc,
2723                                               end_buffer_async_write);
2724         return ret;
2725 }
2726 EXPORT_SYMBOL(nobh_writepage);
2727 
2728 int nobh_truncate_page(struct address_space *mapping,
2729                         loff_t from, get_block_t *get_block)
2730 {
2731         pgoff_t index = from >> PAGE_CACHE_SHIFT;
2732         unsigned offset = from & (PAGE_CACHE_SIZE-1);
2733         unsigned blocksize;
2734         sector_t iblock;
2735         unsigned length, pos;
2736         struct inode *inode = mapping->host;
2737         struct page *page;
2738         struct buffer_head map_bh;
2739         int err;
2740 
2741         blocksize = 1 << inode->i_blkbits;
2742         length = offset & (blocksize - 1);
2743 
2744         /* Block boundary? Nothing to do */
2745         if (!length)
2746                 return 0;
2747 
2748         length = blocksize - length;
2749         iblock = (sector_t)index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
2750 
2751         page = grab_cache_page(mapping, index);
2752         err = -ENOMEM;
2753         if (!page)
2754                 goto out;
2755 
2756         if (page_has_buffers(page)) {
2757 has_buffers:
2758                 unlock_page(page);
2759                 page_cache_release(page);
2760                 return block_truncate_page(mapping, from, get_block);
2761         }
2762 
2763         /* Find the buffer that contains "offset" */
2764         pos = blocksize;
2765         while (offset >= pos) {
2766                 iblock++;
2767                 pos += blocksize;
2768         }
2769 
2770         map_bh.b_size = blocksize;
2771         map_bh.b_state = 0;
2772         err = get_block(inode, iblock, &map_bh, 0);
2773         if (err)
2774                 goto unlock;
2775         /* unmapped? It's a hole - nothing to do */
2776         if (!buffer_mapped(&map_bh))
2777                 goto unlock;
2778 
2779         /* Ok, it's mapped. Make sure it's up-to-date */
2780         if (!PageUptodate(page)) {
2781                 err = mapping->a_ops->readpage(NULL, page);
2782                 if (err) {
2783                         page_cache_release(page);
2784                         goto out;
2785                 }
2786                 lock_page(page);
2787                 if (!PageUptodate(page)) {
2788                         err = -EIO;
2789                         goto unlock;
2790                 }
2791                 if (page_has_buffers(page))
2792                         goto has_buffers;
2793         }
2794         zero_user(page, offset, length);
2795         set_page_dirty(page);
2796         err = 0;
2797 
2798 unlock:
2799         unlock_page(page);
2800         page_cache_release(page);
2801 out:
2802         return err;
2803 }
2804 EXPORT_SYMBOL(nobh_truncate_page);
2805 
2806 int block_truncate_page(struct address_space *mapping,
2807                         loff_t from, get_block_t *get_block)
2808 {
2809         pgoff_t index = from >> PAGE_CACHE_SHIFT;
2810         unsigned offset = from & (PAGE_CACHE_SIZE-1);
2811         unsigned blocksize;
2812         sector_t iblock;
2813         unsigned length, pos;
2814         struct inode *inode = mapping->host;
2815         struct page *page;
2816         struct buffer_head *bh;
2817         int err;
2818 
2819         blocksize = 1 << inode->i_blkbits;
2820         length = offset & (blocksize - 1);
2821 
2822         /* Block boundary? Nothing to do */
2823         if (!length)
2824                 return 0;
2825 
2826         length = blocksize - length;
2827         iblock = (sector_t)index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
2828         
2829         page = grab_cache_page(mapping, index);
2830         err = -ENOMEM;
2831         if (!page)
2832                 goto out;
2833 
2834         if (!page_has_buffers(page))
2835                 create_empty_buffers(page, blocksize, 0);
2836 
2837         /* Find the buffer that contains "offset" */
2838         bh = page_buffers(page);
2839         pos = blocksize;
2840         while (offset >= pos) {
2841                 bh = bh->b_this_page;
2842                 iblock++;
2843                 pos += blocksize;
2844         }
2845 
2846         err = 0;
2847         if (!buffer_mapped(bh)) {
2848                 WARN_ON(bh->b_size != blocksize);
2849                 err = get_block(inode, iblock, bh, 0);
2850                 if (err)
2851                         goto unlock;
2852                 /* unmapped? It's a hole - nothing to do */
2853                 if (!buffer_mapped(bh))
2854                         goto unlock;
2855         }
2856 
2857         /* Ok, it's mapped. Make sure it's up-to-date */
2858         if (PageUptodate(page))
2859                 set_buffer_uptodate(bh);
2860 
2861         if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) {
2862                 err = -EIO;
2863                 ll_rw_block(READ, 1, &bh);
2864                 wait_on_buffer(bh);
2865                 /* Uhhuh. Read error. Complain and punt. */
2866                 if (!buffer_uptodate(bh))
2867                         goto unlock;
2868         }
2869 
2870         zero_user(page, offset, length);
2871         mark_buffer_dirty(bh);
2872         err = 0;
2873 
2874 unlock:
2875         unlock_page(page);
2876         page_cache_release(page);
2877 out:
2878         return err;
2879 }
2880 EXPORT_SYMBOL(block_truncate_page);
2881 
2882 /*
2883  * The generic ->writepage function for buffer-backed address_spaces
2884  */
2885 int block_write_full_page(struct page *page, get_block_t *get_block,
2886                         struct writeback_control *wbc)
2887 {
2888         struct inode * const inode = page->mapping->host;
2889         loff_t i_size = i_size_read(inode);
2890         const pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
2891         unsigned offset;
2892 
2893         /* Is the page fully inside i_size? */
2894         if (page->index < end_index)
2895                 return __block_write_full_page(inode, page, get_block, wbc,
2896                                                end_buffer_async_write);
2897 
2898         /* Is the page fully outside i_size? (truncate in progress) */
2899         offset = i_size & (PAGE_CACHE_SIZE-1);
2900         if (page->index >= end_index+1 || !offset) {
2901                 /*
2902                  * The page may have dirty, unmapped buffers.  For example,
2903                  * they may have been added in ext3_writepage().  Make them
2904                  * freeable here, so the page does not leak.
2905                  */
2906                 do_invalidatepage(page, 0, PAGE_CACHE_SIZE);
2907                 unlock_page(page);
2908                 return 0; /* don't care */
2909         }
2910 
2911         /*
2912          * The page straddles i_size.  It must be zeroed out on each and every
2913          * writepage invocation because it may be mmapped.  "A file is mapped
2914          * in multiples of the page size.  For a file that is not a multiple of
2915          * the  page size, the remaining memory is zeroed when mapped, and
2916          * writes to that region are not written out to the file."
2917          */
2918         zero_user_segment(page, offset, PAGE_CACHE_SIZE);
2919         return __block_write_full_page(inode, page, get_block, wbc,
2920                                                         end_buffer_async_write);
2921 }
2922 EXPORT_SYMBOL(block_write_full_page);
2923 
2924 sector_t generic_block_bmap(struct address_space *mapping, sector_t block,
2925                             get_block_t *get_block)
2926 {
2927         struct buffer_head tmp;
2928         struct inode *inode = mapping->host;
2929         tmp.b_state = 0;
2930         tmp.b_blocknr = 0;
2931         tmp.b_size = 1 << inode->i_blkbits;
2932         get_block(inode, block, &tmp, 0);
2933         return tmp.b_blocknr;
2934 }
2935 EXPORT_SYMBOL(generic_block_bmap);
2936 
2937 static void end_bio_bh_io_sync(struct bio *bio, int err)
2938 {
2939         struct buffer_head *bh = bio->bi_private;
2940 
2941         if (err == -EOPNOTSUPP) {
2942                 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
2943         }
2944 
2945         if (unlikely (test_bit(BIO_QUIET,&bio->bi_flags)))
2946                 set_bit(BH_Quiet, &bh->b_state);
2947 
2948         bh->b_end_io(bh, test_bit(BIO_UPTODATE, &bio->bi_flags));
2949         bio_put(bio);
2950 }
2951 
2952 /*
2953  * This allows us to do IO even on the odd last sectors
2954  * of a device, even if the block size is some multiple
2955  * of the physical sector size.
2956  *
2957  * We'll just truncate the bio to the size of the device,
2958  * and clear the end of the buffer head manually.
2959  *
2960  * Truly out-of-range accesses will turn into actual IO
2961  * errors, this only handles the "we need to be able to
2962  * do IO at the final sector" case.
2963  */
2964 void guard_bio_eod(int rw, struct bio *bio)
2965 {
2966         sector_t maxsector;
2967         struct bio_vec *bvec = &bio->bi_io_vec[bio->bi_vcnt - 1];
2968         unsigned truncated_bytes;
2969 
2970         maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
2971         if (!maxsector)
2972                 return;
2973 
2974         /*
2975          * If the *whole* IO is past the end of the device,
2976          * let it through, and the IO layer will turn it into
2977          * an EIO.
2978          */
2979         if (unlikely(bio->bi_iter.bi_sector >= maxsector))
2980                 return;
2981 
2982         maxsector -= bio->bi_iter.bi_sector;
2983         if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
2984                 return;
2985 
2986         /* Uhhuh. We've got a bio that straddles the device size! */
2987         truncated_bytes = bio->bi_iter.bi_size - (maxsector << 9);
2988 
2989         /* Truncate the bio.. */
2990         bio->bi_iter.bi_size -= truncated_bytes;
2991         bvec->bv_len -= truncated_bytes;
2992 
2993         /* ..and clear the end of the buffer for reads */
2994         if ((rw & RW_MASK) == READ) {
2995                 zero_user(bvec->bv_page, bvec->bv_offset + bvec->bv_len,
2996                                 truncated_bytes);
2997         }
2998 }
2999 
3000 int _submit_bh(int rw, struct buffer_head *bh, unsigned long bio_flags)
3001 {
3002         struct bio *bio;
3003         int ret = 0;
3004 
3005         BUG_ON(!buffer_locked(bh));
3006         BUG_ON(!buffer_mapped(bh));
3007         BUG_ON(!bh->b_end_io);
3008         BUG_ON(buffer_delay(bh));
3009         BUG_ON(buffer_unwritten(bh));
3010 
3011         /*
3012          * Only clear out a write error when rewriting
3013          */
3014         if (test_set_buffer_req(bh) && (rw & WRITE))
3015                 clear_buffer_write_io_error(bh);
3016 
3017         /*
3018          * from here on down, it's all bio -- do the initial mapping,
3019          * submit_bio -> generic_make_request may further map this bio around
3020          */
3021         bio = bio_alloc(GFP_NOIO, 1);
3022 
3023         bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
3024         bio->bi_bdev = bh->b_bdev;
3025         bio->bi_io_vec[0].bv_page = bh->b_page;
3026         bio->bi_io_vec[0].bv_len = bh->b_size;
3027         bio->bi_io_vec[0].bv_offset = bh_offset(bh);
3028 
3029         bio->bi_vcnt = 1;
3030         bio->bi_iter.bi_size = bh->b_size;
3031 
3032         bio->bi_end_io = end_bio_bh_io_sync;
3033         bio->bi_private = bh;
3034         bio->bi_flags |= bio_flags;
3035 
3036         /* Take care of bh's that straddle the end of the device */
3037         guard_bio_eod(rw, bio);
3038 
3039         if (buffer_meta(bh))
3040                 rw |= REQ_META;
3041         if (buffer_prio(bh))
3042                 rw |= REQ_PRIO;
3043 
3044         bio_get(bio);
3045         submit_bio(rw, bio);
3046 
3047         if (bio_flagged(bio, BIO_EOPNOTSUPP))
3048                 ret = -EOPNOTSUPP;
3049 
3050         bio_put(bio);
3051         return ret;
3052 }
3053 EXPORT_SYMBOL_GPL(_submit_bh);
3054 
3055 int submit_bh(int rw, struct buffer_head *bh)
3056 {
3057         return _submit_bh(rw, bh, 0);
3058 }
3059 EXPORT_SYMBOL(submit_bh);
3060 
3061 /**
3062  * ll_rw_block: low-level access to block devices (DEPRECATED)
3063  * @rw: whether to %READ or %WRITE or maybe %READA (readahead)
3064  * @nr: number of &struct buffer_heads in the array
3065  * @bhs: array of pointers to &struct buffer_head
3066  *
3067  * ll_rw_block() takes an array of pointers to &struct buffer_heads, and
3068  * requests an I/O operation on them, either a %READ or a %WRITE.  The third
3069  * %READA option is described in the documentation for generic_make_request()
3070  * which ll_rw_block() calls.
3071  *
3072  * This function drops any buffer that it cannot get a lock on (with the
3073  * BH_Lock state bit), any buffer that appears to be clean when doing a write
3074  * request, and any buffer that appears to be up-to-date when doing read
3075  * request.  Further it marks as clean buffers that are processed for
3076  * writing (the buffer cache won't assume that they are actually clean
3077  * until the buffer gets unlocked).
3078  *
3079  * ll_rw_block sets b_end_io to simple completion handler that marks
3080  * the buffer up-to-date (if appropriate), unlocks the buffer and wakes
3081  * any waiters. 
3082  *
3083  * All of the buffers must be for the same device, and must also be a
3084  * multiple of the current approved size for the device.
3085  */
3086 void ll_rw_block(int rw, int nr, struct buffer_head *bhs[])
3087 {
3088         int i;
3089 
3090         for (i = 0; i < nr; i++) {
3091                 struct buffer_head *bh = bhs[i];
3092 
3093                 if (!trylock_buffer(bh))
3094                         continue;
3095                 if (rw == WRITE) {
3096                         if (test_clear_buffer_dirty(bh)) {
3097                                 bh->b_end_io = end_buffer_write_sync;
3098                                 get_bh(bh);
3099                                 submit_bh(WRITE, bh);
3100                                 continue;
3101                         }
3102                 } else {
3103                         if (!buffer_uptodate(bh)) {
3104                                 bh->b_end_io = end_buffer_read_sync;
3105                                 get_bh(bh);
3106                                 submit_bh(rw, bh);
3107                                 continue;
3108                         }
3109                 }
3110                 unlock_buffer(bh);
3111         }
3112 }
3113 EXPORT_SYMBOL(ll_rw_block);
3114 
3115 void write_dirty_buffer(struct buffer_head *bh, int rw)
3116 {
3117         lock_buffer(bh);
3118         if (!test_clear_buffer_dirty(bh)) {
3119                 unlock_buffer(bh);
3120                 return;
3121         }
3122         bh->b_end_io = end_buffer_write_sync;
3123         get_bh(bh);
3124         submit_bh(rw, bh);
3125 }
3126 EXPORT_SYMBOL(write_dirty_buffer);
3127 
3128 /*
3129  * For a data-integrity writeout, we need to wait upon any in-progress I/O
3130  * and then start new I/O and then wait upon it.  The caller must have a ref on
3131  * the buffer_head.
3132  */
3133 int __sync_dirty_buffer(struct buffer_head *bh, int rw)
3134 {
3135         int ret = 0;
3136 
3137         WARN_ON(atomic_read(&bh->b_count) < 1);
3138         lock_buffer(bh);
3139         if (test_clear_buffer_dirty(bh)) {
3140                 get_bh(bh);
3141                 bh->b_end_io = end_buffer_write_sync;
3142                 ret = submit_bh(rw, bh);
3143                 wait_on_buffer(bh);
3144                 if (!ret && !buffer_uptodate(bh))
3145                         ret = -EIO;
3146         } else {
3147                 unlock_buffer(bh);
3148         }
3149         return ret;
3150 }
3151 EXPORT_SYMBOL(__sync_dirty_buffer);
3152 
3153 int sync_dirty_buffer(struct buffer_head *bh)
3154 {
3155         return __sync_dirty_buffer(bh, WRITE_SYNC);
3156 }
3157 EXPORT_SYMBOL(sync_dirty_buffer);
3158 
3159 /*
3160  * try_to_free_buffers() checks if all the buffers on this particular page
3161  * are unused, and releases them if so.
3162  *
3163  * Exclusion against try_to_free_buffers may be obtained by either
3164  * locking the page or by holding its mapping's private_lock.
3165  *
3166  * If the page is dirty but all the buffers are clean then we need to
3167  * be sure to mark the page clean as well.  This is because the page
3168  * may be against a block device, and a later reattachment of buffers
3169  * to a dirty page will set *all* buffers dirty.  Which would corrupt
3170  * filesystem data on the same device.
3171  *
3172  * The same applies to regular filesystem pages: if all the buffers are
3173  * clean then we set the page clean and proceed.  To do that, we require
3174  * total exclusion from __set_page_dirty_buffers().  That is obtained with
3175  * private_lock.
3176  *
3177  * try_to_free_buffers() is non-blocking.
3178  */
3179 static inline int buffer_busy(struct buffer_head *bh)
3180 {
3181         return atomic_read(&bh->b_count) |
3182                 (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock)));
3183 }
3184 
3185 static int
3186 drop_buffers(struct page *page, struct buffer_head **buffers_to_free)
3187 {
3188         struct buffer_head *head = page_buffers(page);
3189         struct buffer_head *bh;
3190 
3191         bh = head;
3192         do {
3193                 if (buffer_write_io_error(bh) && page->mapping)
3194                         set_bit(AS_EIO, &page->mapping->flags);
3195                 if (buffer_busy(bh))
3196                         goto failed;
3197                 bh = bh->b_this_page;
3198         } while (bh != head);
3199 
3200         do {
3201                 struct buffer_head *next = bh->b_this_page;
3202 
3203                 if (bh->b_assoc_map)
3204                         __remove_assoc_queue(bh);
3205                 bh = next;
3206         } while (bh != head);
3207         *buffers_to_free = head;
3208         __clear_page_buffers(page);
3209         return 1;
3210 failed:
3211         return 0;
3212 }
3213 
3214 int try_to_free_buffers(struct page *page)
3215 {
3216         struct address_space * const mapping = page->mapping;
3217         struct buffer_head *buffers_to_free = NULL;
3218         int ret = 0;
3219 
3220         BUG_ON(!PageLocked(page));
3221         if (PageWriteback(page))
3222                 return 0;
3223 
3224         if (mapping == NULL) {          /* can this still happen? */
3225                 ret = drop_buffers(page, &buffers_to_free);
3226                 goto out;
3227         }
3228 
3229         spin_lock(&mapping->private_lock);
3230         ret = drop_buffers(page, &buffers_to_free);
3231 
3232         /*
3233          * If the filesystem writes its buffers by hand (eg ext3)
3234          * then we can have clean buffers against a dirty page.  We
3235          * clean the page here; otherwise the VM will never notice
3236          * that the filesystem did any IO at all.
3237          *
3238          * Also, during truncate, discard_buffer will have marked all
3239          * the page's buffers clean.  We discover that here and clean
3240          * the page also.
3241          *
3242          * private_lock must be held over this entire operation in order
3243          * to synchronise against __set_page_dirty_buffers and prevent the
3244          * dirty bit from being lost.
3245          */
3246         if (ret)
3247                 cancel_dirty_page(page, PAGE_CACHE_SIZE);
3248         spin_unlock(&mapping->private_lock);
3249 out:
3250         if (buffers_to_free) {
3251                 struct buffer_head *bh = buffers_to_free;
3252 
3253                 do {
3254                         struct buffer_head *next = bh->b_this_page;
3255                         free_buffer_head(bh);
3256                         bh = next;
3257                 } while (bh != buffers_to_free);
3258         }
3259         return ret;
3260 }
3261 EXPORT_SYMBOL(try_to_free_buffers);
3262 
3263 /*
3264  * There are no bdflush tunables left.  But distributions are
3265  * still running obsolete flush daemons, so we terminate them here.
3266  *
3267  * Use of bdflush() is deprecated and will be removed in a future kernel.
3268  * The `flush-X' kernel threads fully replace bdflush daemons and this call.
3269  */
3270 SYSCALL_DEFINE2(bdflush, int, func, long, data)
3271 {
3272         static int msg_count;
3273 
3274         if (!capable(CAP_SYS_ADMIN))
3275                 return -EPERM;
3276 
3277         if (msg_count < 5) {
3278                 msg_count++;
3279                 printk(KERN_INFO
3280                         "warning: process `%s' used the obsolete bdflush"
3281                         " system call\n", current->comm);
3282                 printk(KERN_INFO "Fix your initscripts?\n");
3283         }
3284 
3285         if (func == 1)
3286                 do_exit(0);
3287         return 0;
3288 }
3289 
3290 /*
3291  * Buffer-head allocation
3292  */
3293 static struct kmem_cache *bh_cachep __read_mostly;
3294 
3295 /*
3296  * Once the number of bh's in the machine exceeds this level, we start
3297  * stripping them in writeback.
3298  */
3299 static unsigned long max_buffer_heads;
3300 
3301 int buffer_heads_over_limit;
3302 
3303 struct bh_accounting {
3304         int nr;                 /* Number of live bh's */
3305         int ratelimit;          /* Limit cacheline bouncing */
3306 };
3307 
3308 static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0};
3309 
3310 static void recalc_bh_state(void)
3311 {
3312         int i;
3313         int tot = 0;
3314 
3315         if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096)
3316                 return;
3317         __this_cpu_write(bh_accounting.ratelimit, 0);
3318         for_each_online_cpu(i)
3319                 tot += per_cpu(bh_accounting, i).nr;
3320         buffer_heads_over_limit = (tot > max_buffer_heads);
3321 }
3322 
3323 struct buffer_head *alloc_buffer_head(gfp_t gfp_flags)
3324 {
3325         struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags);
3326         if (ret) {
3327                 INIT_LIST_HEAD(&ret->b_assoc_buffers);
3328                 preempt_disable();
3329                 __this_cpu_inc(bh_accounting.nr);
3330                 recalc_bh_state();
3331                 preempt_enable();
3332         }
3333         return ret;
3334 }
3335 EXPORT_SYMBOL(alloc_buffer_head);
3336 
3337 void free_buffer_head(struct buffer_head *bh)
3338 {
3339         BUG_ON(!list_empty(&bh->b_assoc_buffers));
3340         kmem_cache_free(bh_cachep, bh);
3341         preempt_disable();
3342         __this_cpu_dec(bh_accounting.nr);
3343         recalc_bh_state();
3344         preempt_enable();
3345 }
3346 EXPORT_SYMBOL(free_buffer_head);
3347 
3348 static void buffer_exit_cpu(int cpu)
3349 {
3350         int i;
3351         struct bh_lru *b = &per_cpu(bh_lrus, cpu);
3352 
3353         for (i = 0; i < BH_LRU_SIZE; i++) {
3354                 brelse(b->bhs[i]);
3355                 b->bhs[i] = NULL;
3356         }
3357         this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr);
3358         per_cpu(bh_accounting, cpu).nr = 0;
3359 }
3360 
3361 static int buffer_cpu_notify(struct notifier_block *self,
3362                               unsigned long action, void *hcpu)
3363 {
3364         if (action == CPU_DEAD || action == CPU_DEAD_FROZEN)
3365                 buffer_exit_cpu((unsigned long)hcpu);
3366         return NOTIFY_OK;
3367 }
3368 
3369 /**
3370  * bh_uptodate_or_lock - Test whether the buffer is uptodate
3371  * @bh: struct buffer_head
3372  *
3373  * Return true if the buffer is up-to-date and false,
3374  * with the buffer locked, if not.
3375  */
3376 int bh_uptodate_or_lock(struct buffer_head *bh)
3377 {
3378         if (!buffer_uptodate(bh)) {
3379                 lock_buffer(bh);
3380                 if (!buffer_uptodate(bh))
3381                         return 0;
3382                 unlock_buffer(bh);
3383         }
3384         return 1;
3385 }
3386 EXPORT_SYMBOL(bh_uptodate_or_lock);
3387 
3388 /**
3389  * bh_submit_read - Submit a locked buffer for reading
3390  * @bh: struct buffer_head
3391  *
3392  * Returns zero on success and -EIO on error.
3393  */
3394 int bh_submit_read(struct buffer_head *bh)
3395 {
3396         BUG_ON(!buffer_locked(bh));
3397 
3398         if (buffer_uptodate(bh)) {
3399                 unlock_buffer(bh);
3400                 return 0;
3401         }
3402 
3403         get_bh(bh);
3404         bh->b_end_io = end_buffer_read_sync;
3405         submit_bh(READ, bh);
3406         wait_on_buffer(bh);
3407         if (buffer_uptodate(bh))
3408                 return 0;
3409         return -EIO;
3410 }
3411 EXPORT_SYMBOL(bh_submit_read);
3412 
3413 void __init buffer_init(void)
3414 {
3415         unsigned long nrpages;
3416 
3417         bh_cachep = kmem_cache_create("buffer_head",
3418                         sizeof(struct buffer_head), 0,
3419                                 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
3420                                 SLAB_MEM_SPREAD),
3421                                 NULL);
3422 
3423         /*
3424          * Limit the bh occupancy to 10% of ZONE_NORMAL
3425          */
3426         nrpages = (nr_free_buffer_pages() * 10) / 100;
3427         max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head));
3428         hotcpu_notifier(buffer_cpu_notify, 0);
3429 }
3430 

~ [ 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