1 /* handling of writes to regular files and writing back to the server
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
11 #include <linux/backing-dev.h>
12 #include <linux/slab.h>
14 #include <linux/pagemap.h>
15 #include <linux/writeback.h>
16 #include <linux/pagevec.h>
19 static int afs_write_back_from_locked_page(struct afs_writeback
*wb
,
23 * mark a page as having been made dirty and thus needing writeback
25 int afs_set_page_dirty(struct page
*page
)
28 return __set_page_dirty_nobuffers(page
);
32 * unlink a writeback record because its usage has reached zero
33 * - must be called with the wb->vnode->writeback_lock held
35 static void afs_unlink_writeback(struct afs_writeback
*wb
)
37 struct afs_writeback
*front
;
38 struct afs_vnode
*vnode
= wb
->vnode
;
40 list_del_init(&wb
->link
);
41 if (!list_empty(&vnode
->writebacks
)) {
42 /* if an fsync rises to the front of the queue then wake it
44 front
= list_entry(vnode
->writebacks
.next
,
45 struct afs_writeback
, link
);
46 if (front
->state
== AFS_WBACK_SYNCING
) {
47 _debug("wake up sync");
48 front
->state
= AFS_WBACK_COMPLETE
;
49 wake_up(&front
->waitq
);
55 * free a writeback record
57 static void afs_free_writeback(struct afs_writeback
*wb
)
65 * dispose of a reference to a writeback record
67 void afs_put_writeback(struct afs_writeback
*wb
)
69 struct afs_vnode
*vnode
= wb
->vnode
;
71 _enter("{%d}", wb
->usage
);
73 spin_lock(&vnode
->writeback_lock
);
75 afs_unlink_writeback(wb
);
78 spin_unlock(&vnode
->writeback_lock
);
80 afs_free_writeback(wb
);
84 * partly or wholly fill a page that's under preparation for writing
86 static int afs_fill_page(struct afs_vnode
*vnode
, struct key
*key
,
87 loff_t pos
, struct page
*page
)
93 _enter(",,%llu", (unsigned long long)pos
);
95 i_size
= i_size_read(&vnode
->vfs_inode
);
96 if (pos
+ PAGE_CACHE_SIZE
> i_size
)
99 len
= PAGE_CACHE_SIZE
;
101 ret
= afs_vnode_fetch_data(vnode
, key
, pos
, len
, page
);
103 if (ret
== -ENOENT
) {
104 _debug("got NOENT from server"
105 " - marking file deleted and stale");
106 set_bit(AFS_VNODE_DELETED
, &vnode
->flags
);
111 _leave(" = %d", ret
);
116 * prepare to perform part of a write to a page
118 int afs_write_begin(struct file
*file
, struct address_space
*mapping
,
119 loff_t pos
, unsigned len
, unsigned flags
,
120 struct page
**pagep
, void **fsdata
)
122 struct afs_writeback
*candidate
, *wb
;
123 struct afs_vnode
*vnode
= AFS_FS_I(file_inode(file
));
125 struct key
*key
= file
->private_data
;
126 unsigned from
= pos
& (PAGE_CACHE_SIZE
- 1);
127 unsigned to
= from
+ len
;
128 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
131 _enter("{%x:%u},{%lx},%u,%u",
132 vnode
->fid
.vid
, vnode
->fid
.vnode
, index
, from
, to
);
134 candidate
= kzalloc(sizeof(*candidate
), GFP_KERNEL
);
137 candidate
->vnode
= vnode
;
138 candidate
->first
= candidate
->last
= index
;
139 candidate
->offset_first
= from
;
140 candidate
->to_last
= to
;
141 INIT_LIST_HEAD(&candidate
->link
);
142 candidate
->usage
= 1;
143 candidate
->state
= AFS_WBACK_PENDING
;
144 init_waitqueue_head(&candidate
->waitq
);
146 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
152 if (!PageUptodate(page
) && len
!= PAGE_CACHE_SIZE
) {
153 ret
= afs_fill_page(vnode
, key
, index
<< PAGE_CACHE_SHIFT
, page
);
158 _leave(" = %d [prep]", ret
);
161 SetPageUptodate(page
);
164 /* page won't leak in error case: it eventually gets cleaned off LRU */
168 spin_lock(&vnode
->writeback_lock
);
170 /* see if this page is already pending a writeback under a suitable key
171 * - if so we can just join onto that one */
172 wb
= (struct afs_writeback
*) page_private(page
);
174 if (wb
->key
== key
&& wb
->state
== AFS_WBACK_PENDING
)
175 goto subsume_in_current_wb
;
176 goto flush_conflicting_wb
;
180 /* see if we can find an already pending writeback that we can
181 * append this page to */
182 list_for_each_entry(wb
, &vnode
->writebacks
, link
) {
183 if (wb
->last
== index
- 1 && wb
->key
== key
&&
184 wb
->state
== AFS_WBACK_PENDING
)
185 goto append_to_previous_wb
;
189 list_add_tail(&candidate
->link
, &vnode
->writebacks
);
190 candidate
->key
= key_get(key
);
191 spin_unlock(&vnode
->writeback_lock
);
192 SetPagePrivate(page
);
193 set_page_private(page
, (unsigned long) candidate
);
194 _leave(" = 0 [new]");
197 subsume_in_current_wb
:
199 ASSERTRANGE(wb
->first
, <=, index
, <=, wb
->last
);
200 if (index
== wb
->first
&& from
< wb
->offset_first
)
201 wb
->offset_first
= from
;
202 if (index
== wb
->last
&& to
> wb
->to_last
)
204 spin_unlock(&vnode
->writeback_lock
);
206 _leave(" = 0 [sub]");
209 append_to_previous_wb
:
210 _debug("append into %lx-%lx", wb
->first
, wb
->last
);
214 spin_unlock(&vnode
->writeback_lock
);
215 SetPagePrivate(page
);
216 set_page_private(page
, (unsigned long) wb
);
218 _leave(" = 0 [app]");
221 /* the page is currently bound to another context, so if it's dirty we
222 * need to flush it before we can use the new context */
223 flush_conflicting_wb
:
224 _debug("flush conflict");
225 if (wb
->state
== AFS_WBACK_PENDING
)
226 wb
->state
= AFS_WBACK_CONFLICTING
;
227 spin_unlock(&vnode
->writeback_lock
);
228 if (PageDirty(page
)) {
229 ret
= afs_write_back_from_locked_page(wb
, page
);
231 afs_put_writeback(candidate
);
232 _leave(" = %d", ret
);
237 /* the page holds a ref on the writeback record */
238 afs_put_writeback(wb
);
239 set_page_private(page
, 0);
240 ClearPagePrivate(page
);
245 * finalise part of a write to a page
247 int afs_write_end(struct file
*file
, struct address_space
*mapping
,
248 loff_t pos
, unsigned len
, unsigned copied
,
249 struct page
*page
, void *fsdata
)
251 struct afs_vnode
*vnode
= AFS_FS_I(file_inode(file
));
252 loff_t i_size
, maybe_i_size
;
254 _enter("{%x:%u},{%lx}",
255 vnode
->fid
.vid
, vnode
->fid
.vnode
, page
->index
);
257 maybe_i_size
= pos
+ copied
;
259 i_size
= i_size_read(&vnode
->vfs_inode
);
260 if (maybe_i_size
> i_size
) {
261 spin_lock(&vnode
->writeback_lock
);
262 i_size
= i_size_read(&vnode
->vfs_inode
);
263 if (maybe_i_size
> i_size
)
264 i_size_write(&vnode
->vfs_inode
, maybe_i_size
);
265 spin_unlock(&vnode
->writeback_lock
);
268 set_page_dirty(page
);
272 page_cache_release(page
);
278 * kill all the pages in the given range
280 static void afs_kill_pages(struct afs_vnode
*vnode
, bool error
,
281 pgoff_t first
, pgoff_t last
)
284 unsigned count
, loop
;
286 _enter("{%x:%u},%lx-%lx",
287 vnode
->fid
.vid
, vnode
->fid
.vnode
, first
, last
);
289 pagevec_init(&pv
, 0);
292 _debug("kill %lx-%lx", first
, last
);
294 count
= last
- first
+ 1;
295 if (count
> PAGEVEC_SIZE
)
296 count
= PAGEVEC_SIZE
;
297 pv
.nr
= find_get_pages_contig(vnode
->vfs_inode
.i_mapping
,
298 first
, count
, pv
.pages
);
299 ASSERTCMP(pv
.nr
, ==, count
);
301 for (loop
= 0; loop
< count
; loop
++) {
302 struct page
*page
= pv
.pages
[loop
];
303 ClearPageUptodate(page
);
306 if (PageWriteback(page
))
307 end_page_writeback(page
);
308 if (page
->index
>= first
)
309 first
= page
->index
+ 1;
312 __pagevec_release(&pv
);
313 } while (first
< last
);
319 * synchronously write back the locked page and any subsequent non-locked dirty
320 * pages also covered by the same writeback record
322 static int afs_write_back_from_locked_page(struct afs_writeback
*wb
,
323 struct page
*primary_page
)
325 struct page
*pages
[8], *page
;
327 unsigned n
, offset
, to
;
328 pgoff_t start
, first
, last
;
331 _enter(",%lx", primary_page
->index
);
334 if (!clear_page_dirty_for_io(primary_page
))
336 if (test_set_page_writeback(primary_page
))
339 /* find all consecutive lockable dirty pages, stopping when we find a
340 * page that is not immediately lockable, is not dirty or is missing,
341 * or we reach the end of the range */
342 start
= primary_page
->index
;
343 if (start
>= wb
->last
)
347 _debug("more %lx [%lx]", start
, count
);
348 n
= wb
->last
- start
+ 1;
349 if (n
> ARRAY_SIZE(pages
))
350 n
= ARRAY_SIZE(pages
);
351 n
= find_get_pages_contig(wb
->vnode
->vfs_inode
.i_mapping
,
353 _debug("fgpc %u", n
);
356 if (pages
[0]->index
!= start
) {
358 put_page(pages
[--n
]);
363 for (loop
= 0; loop
< n
; loop
++) {
365 if (page
->index
> wb
->last
)
367 if (!trylock_page(page
))
369 if (!PageDirty(page
) ||
370 page_private(page
) != (unsigned long) wb
) {
374 if (!clear_page_dirty_for_io(page
))
376 if (test_set_page_writeback(page
))
383 for (; loop
< n
; loop
++)
384 put_page(pages
[loop
]);
389 } while (start
<= wb
->last
&& count
< 65536);
392 /* we now have a contiguous set of dirty pages, each with writeback set
393 * and the dirty mark cleared; the first page is locked and must remain
394 * so, all the rest are unlocked */
395 first
= primary_page
->index
;
396 last
= first
+ count
- 1;
398 offset
= (first
== wb
->first
) ? wb
->offset_first
: 0;
399 to
= (last
== wb
->last
) ? wb
->to_last
: PAGE_SIZE
;
401 _debug("write back %lx[%u..] to %lx[..%u]", first
, offset
, last
, to
);
403 ret
= afs_vnode_store_data(wb
, first
, last
, offset
, to
);
409 &wb
->vnode
->vfs_inode
.i_mapping
->flags
);
418 afs_kill_pages(wb
->vnode
, true, first
, last
);
419 set_bit(AS_EIO
, &wb
->vnode
->vfs_inode
.i_mapping
->flags
);
427 afs_kill_pages(wb
->vnode
, false, first
, last
);
436 _leave(" = %d", ret
);
441 * write a page back to the server
442 * - the caller locked the page for us
444 int afs_writepage(struct page
*page
, struct writeback_control
*wbc
)
446 struct afs_writeback
*wb
;
449 _enter("{%lx},", page
->index
);
451 wb
= (struct afs_writeback
*) page_private(page
);
454 ret
= afs_write_back_from_locked_page(wb
, page
);
457 _leave(" = %d", ret
);
461 wbc
->nr_to_write
-= ret
;
468 * write a region of pages back to the server
470 static int afs_writepages_region(struct address_space
*mapping
,
471 struct writeback_control
*wbc
,
472 pgoff_t index
, pgoff_t end
, pgoff_t
*_next
)
474 struct afs_writeback
*wb
;
478 _enter(",,%lx,%lx,", index
, end
);
481 n
= find_get_pages_tag(mapping
, &index
, PAGECACHE_TAG_DIRTY
,
486 _debug("wback %lx", page
->index
);
488 if (page
->index
> end
) {
490 page_cache_release(page
);
491 _leave(" = 0 [%lx]", *_next
);
495 /* at this point we hold neither mapping->tree_lock nor lock on
496 * the page itself: the page may be truncated or invalidated
497 * (changing page->mapping to NULL), or even swizzled back from
498 * swapper_space to tmpfs file mapping
502 if (page
->mapping
!= mapping
) {
504 page_cache_release(page
);
508 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
509 wait_on_page_writeback(page
);
511 if (PageWriteback(page
) || !PageDirty(page
)) {
517 wb
= (struct afs_writeback
*) page_private(page
);
520 spin_lock(&wb
->vnode
->writeback_lock
);
521 wb
->state
= AFS_WBACK_WRITING
;
522 spin_unlock(&wb
->vnode
->writeback_lock
);
524 ret
= afs_write_back_from_locked_page(wb
, page
);
526 page_cache_release(page
);
528 _leave(" = %d", ret
);
532 wbc
->nr_to_write
-= ret
;
535 } while (index
< end
&& wbc
->nr_to_write
> 0);
538 _leave(" = 0 [%lx]", *_next
);
543 * write some of the pending data back to the server
545 int afs_writepages(struct address_space
*mapping
,
546 struct writeback_control
*wbc
)
548 pgoff_t start
, end
, next
;
553 if (wbc
->range_cyclic
) {
554 start
= mapping
->writeback_index
;
556 ret
= afs_writepages_region(mapping
, wbc
, start
, end
, &next
);
557 if (start
> 0 && wbc
->nr_to_write
> 0 && ret
== 0)
558 ret
= afs_writepages_region(mapping
, wbc
, 0, start
,
560 mapping
->writeback_index
= next
;
561 } else if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
) {
562 end
= (pgoff_t
)(LLONG_MAX
>> PAGE_CACHE_SHIFT
);
563 ret
= afs_writepages_region(mapping
, wbc
, 0, end
, &next
);
564 if (wbc
->nr_to_write
> 0)
565 mapping
->writeback_index
= next
;
567 start
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
568 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
569 ret
= afs_writepages_region(mapping
, wbc
, start
, end
, &next
);
572 _leave(" = %d", ret
);
577 * completion of write to server
579 void afs_pages_written_back(struct afs_vnode
*vnode
, struct afs_call
*call
)
581 struct afs_writeback
*wb
= call
->wb
;
583 unsigned count
, loop
;
584 pgoff_t first
= call
->first
, last
= call
->last
;
587 _enter("{%x:%u},{%lx-%lx}",
588 vnode
->fid
.vid
, vnode
->fid
.vnode
, first
, last
);
592 pagevec_init(&pv
, 0);
595 _debug("done %lx-%lx", first
, last
);
597 count
= last
- first
+ 1;
598 if (count
> PAGEVEC_SIZE
)
599 count
= PAGEVEC_SIZE
;
600 pv
.nr
= find_get_pages_contig(call
->mapping
, first
, count
,
602 ASSERTCMP(pv
.nr
, ==, count
);
604 spin_lock(&vnode
->writeback_lock
);
605 for (loop
= 0; loop
< count
; loop
++) {
606 struct page
*page
= pv
.pages
[loop
];
607 end_page_writeback(page
);
608 if (page_private(page
) == (unsigned long) wb
) {
609 set_page_private(page
, 0);
610 ClearPagePrivate(page
);
615 if (wb
->usage
== 0) {
616 afs_unlink_writeback(wb
);
619 spin_unlock(&vnode
->writeback_lock
);
622 afs_free_writeback(wb
);
626 __pagevec_release(&pv
);
627 } while (first
<= last
);
633 * write to an AFS file
635 ssize_t
afs_file_write(struct kiocb
*iocb
, struct iov_iter
*from
)
637 struct afs_vnode
*vnode
= AFS_FS_I(file_inode(iocb
->ki_filp
));
639 size_t count
= iov_iter_count(from
);
641 _enter("{%x.%u},{%zu},",
642 vnode
->fid
.vid
, vnode
->fid
.vnode
, count
);
644 if (IS_SWAPFILE(&vnode
->vfs_inode
)) {
646 "AFS: Attempt to write to active swap file!\n");
653 result
= generic_file_write_iter(iocb
, from
);
654 if (IS_ERR_VALUE(result
)) {
655 _leave(" = %zd", result
);
659 _leave(" = %zd", result
);
664 * flush the vnode to the fileserver
666 int afs_writeback_all(struct afs_vnode
*vnode
)
668 struct address_space
*mapping
= vnode
->vfs_inode
.i_mapping
;
669 struct writeback_control wbc
= {
670 .sync_mode
= WB_SYNC_ALL
,
671 .nr_to_write
= LONG_MAX
,
678 ret
= mapping
->a_ops
->writepages(mapping
, &wbc
);
679 __mark_inode_dirty(mapping
->host
, I_DIRTY_PAGES
);
681 _leave(" = %d", ret
);
686 * flush any dirty pages for this process, and check for write errors.
687 * - the return status from this call provides a reliable indication of
688 * whether any write errors occurred for this process.
690 int afs_fsync(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
692 struct inode
*inode
= file_inode(file
);
693 struct afs_writeback
*wb
, *xwb
;
694 struct afs_vnode
*vnode
= AFS_FS_I(inode
);
697 _enter("{%x:%u},{n=%pD},%d",
698 vnode
->fid
.vid
, vnode
->fid
.vnode
, file
,
701 ret
= filemap_write_and_wait_range(inode
->i_mapping
, start
, end
);
704 mutex_lock(&inode
->i_mutex
);
706 /* use a writeback record as a marker in the queue - when this reaches
707 * the front of the queue, all the outstanding writes are either
708 * completed or rejected */
709 wb
= kzalloc(sizeof(*wb
), GFP_KERNEL
);
717 wb
->offset_first
= 0;
718 wb
->to_last
= PAGE_SIZE
;
720 wb
->state
= AFS_WBACK_SYNCING
;
721 init_waitqueue_head(&wb
->waitq
);
723 spin_lock(&vnode
->writeback_lock
);
724 list_for_each_entry(xwb
, &vnode
->writebacks
, link
) {
725 if (xwb
->state
== AFS_WBACK_PENDING
)
726 xwb
->state
= AFS_WBACK_CONFLICTING
;
728 list_add_tail(&wb
->link
, &vnode
->writebacks
);
729 spin_unlock(&vnode
->writeback_lock
);
731 /* push all the outstanding writebacks to the server */
732 ret
= afs_writeback_all(vnode
);
734 afs_put_writeback(wb
);
735 _leave(" = %d [wb]", ret
);
739 /* wait for the preceding writes to actually complete */
740 ret
= wait_event_interruptible(wb
->waitq
,
741 wb
->state
== AFS_WBACK_COMPLETE
||
742 vnode
->writebacks
.next
== &wb
->link
);
743 afs_put_writeback(wb
);
744 _leave(" = %d", ret
);
746 mutex_unlock(&inode
->i_mutex
);
751 * Flush out all outstanding writes on a file opened for writing when it is
754 int afs_flush(struct file
*file
, fl_owner_t id
)
758 if ((file
->f_mode
& FMODE_WRITE
) == 0)
761 return vfs_fsync(file
, 0);
765 * notification that a previously read-only page is about to become writable
766 * - if it returns an error, the caller will deliver a bus error signal
768 int afs_page_mkwrite(struct vm_area_struct
*vma
, struct page
*page
)
770 struct afs_vnode
*vnode
= AFS_FS_I(vma
->vm_file
->f_mapping
->host
);
772 _enter("{{%x:%u}},{%lx}",
773 vnode
->fid
.vid
, vnode
->fid
.vnode
, page
->index
);
775 /* wait for the page to be written to the cache before we allow it to
777 #ifdef CONFIG_AFS_FSCACHE
778 fscache_wait_on_page_write(vnode
->cache
, page
);