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.
12 #include <linux/backing-dev.h>
13 #include <linux/slab.h>
15 #include <linux/pagemap.h>
16 #include <linux/writeback.h>
17 #include <linux/pagevec.h>
21 * mark a page as having been made dirty and thus needing writeback
23 int afs_set_page_dirty(struct page
*page
)
26 return __set_page_dirty_nobuffers(page
);
30 * partly or wholly fill a page that's under preparation for writing
32 static int afs_fill_page(struct afs_vnode
*vnode
, struct key
*key
,
33 loff_t pos
, unsigned int len
, struct page
*page
)
38 _enter(",,%llu", (unsigned long long)pos
);
40 req
= kzalloc(sizeof(struct afs_read
) + sizeof(struct page
*),
45 refcount_set(&req
->usage
, 1);
49 req
->pages
= req
->array
;
53 ret
= afs_fetch_data(vnode
, key
, req
);
57 _debug("got NOENT from server"
58 " - marking file deleted and stale");
59 set_bit(AFS_VNODE_DELETED
, &vnode
->flags
);
69 * prepare to perform part of a write to a page
71 int afs_write_begin(struct file
*file
, struct address_space
*mapping
,
72 loff_t pos
, unsigned len
, unsigned flags
,
73 struct page
**pagep
, void **fsdata
)
75 struct afs_vnode
*vnode
= AFS_FS_I(file_inode(file
));
77 struct key
*key
= afs_file_key(file
);
79 unsigned f
, from
= pos
& (PAGE_SIZE
- 1);
80 unsigned t
, to
= from
+ len
;
81 pgoff_t index
= pos
>> PAGE_SHIFT
;
84 _enter("{%x:%u},{%lx},%u,%u",
85 vnode
->fid
.vid
, vnode
->fid
.vnode
, index
, from
, to
);
87 /* We want to store information about how much of a page is altered in
90 BUILD_BUG_ON(PAGE_SIZE
> 32768 && sizeof(page
->private) < 8);
92 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
96 if (!PageUptodate(page
) && len
!= PAGE_SIZE
) {
97 ret
= afs_fill_page(vnode
, key
, pos
& PAGE_MASK
, PAGE_SIZE
, page
);
101 _leave(" = %d [prep]", ret
);
104 SetPageUptodate(page
);
107 /* page won't leak in error case: it eventually gets cleaned off LRU */
111 /* See if this page is already partially written in a way that we can
112 * merge the new write with.
115 if (PagePrivate(page
)) {
116 priv
= page_private(page
);
117 f
= priv
& AFS_PRIV_MAX
;
118 t
= priv
>> AFS_PRIV_SHIFT
;
123 if (PageWriteback(page
)) {
124 trace_afs_page_dirty(vnode
, tracepoint_string("alrdy"),
126 goto flush_conflicting_write
;
128 /* If the file is being filled locally, allow inter-write
129 * spaces to be merged into writes. If it's not, only write
130 * back what the user gives us.
132 if (!test_bit(AFS_VNODE_NEW_CONTENT
, &vnode
->flags
) &&
133 (to
< f
|| from
> t
))
134 goto flush_conflicting_write
;
144 priv
= (unsigned long)t
<< AFS_PRIV_SHIFT
;
146 trace_afs_page_dirty(vnode
, tracepoint_string("begin"),
148 SetPagePrivate(page
);
149 set_page_private(page
, priv
);
153 /* The previous write and this write aren't adjacent or overlapping, so
154 * flush the page out.
156 flush_conflicting_write
:
157 _debug("flush conflict");
158 ret
= write_one_page(page
);
160 _leave(" = %d", ret
);
164 ret
= lock_page_killable(page
);
166 _leave(" = %d", ret
);
173 * finalise part of a write to a page
175 int afs_write_end(struct file
*file
, struct address_space
*mapping
,
176 loff_t pos
, unsigned len
, unsigned copied
,
177 struct page
*page
, void *fsdata
)
179 struct afs_vnode
*vnode
= AFS_FS_I(file_inode(file
));
180 struct key
*key
= afs_file_key(file
);
181 loff_t i_size
, maybe_i_size
;
184 _enter("{%x:%u},{%lx}",
185 vnode
->fid
.vid
, vnode
->fid
.vnode
, page
->index
);
187 maybe_i_size
= pos
+ copied
;
189 i_size
= i_size_read(&vnode
->vfs_inode
);
190 if (maybe_i_size
> i_size
) {
191 spin_lock(&vnode
->wb_lock
);
192 i_size
= i_size_read(&vnode
->vfs_inode
);
193 if (maybe_i_size
> i_size
)
194 i_size_write(&vnode
->vfs_inode
, maybe_i_size
);
195 spin_unlock(&vnode
->wb_lock
);
198 if (!PageUptodate(page
)) {
200 /* Try and load any missing data from the server. The
201 * unmarshalling routine will take care of clearing any
202 * bits that are beyond the EOF.
204 ret
= afs_fill_page(vnode
, key
, pos
+ copied
,
209 SetPageUptodate(page
);
212 set_page_dirty(page
);
224 * kill all the pages in the given range
226 static void afs_kill_pages(struct address_space
*mapping
,
227 pgoff_t first
, pgoff_t last
)
229 struct afs_vnode
*vnode
= AFS_FS_I(mapping
->host
);
231 unsigned count
, loop
;
233 _enter("{%x:%u},%lx-%lx",
234 vnode
->fid
.vid
, vnode
->fid
.vnode
, first
, last
);
239 _debug("kill %lx-%lx", first
, last
);
241 count
= last
- first
+ 1;
242 if (count
> PAGEVEC_SIZE
)
243 count
= PAGEVEC_SIZE
;
244 pv
.nr
= find_get_pages_contig(mapping
, first
, count
, pv
.pages
);
245 ASSERTCMP(pv
.nr
, ==, count
);
247 for (loop
= 0; loop
< count
; loop
++) {
248 struct page
*page
= pv
.pages
[loop
];
249 ClearPageUptodate(page
);
251 end_page_writeback(page
);
252 if (page
->index
>= first
)
253 first
= page
->index
+ 1;
255 generic_error_remove_page(mapping
, page
);
259 __pagevec_release(&pv
);
260 } while (first
<= last
);
266 * Redirty all the pages in a given range.
268 static void afs_redirty_pages(struct writeback_control
*wbc
,
269 struct address_space
*mapping
,
270 pgoff_t first
, pgoff_t last
)
272 struct afs_vnode
*vnode
= AFS_FS_I(mapping
->host
);
274 unsigned count
, loop
;
276 _enter("{%x:%u},%lx-%lx",
277 vnode
->fid
.vid
, vnode
->fid
.vnode
, first
, last
);
282 _debug("redirty %lx-%lx", first
, last
);
284 count
= last
- first
+ 1;
285 if (count
> PAGEVEC_SIZE
)
286 count
= PAGEVEC_SIZE
;
287 pv
.nr
= find_get_pages_contig(mapping
, first
, count
, pv
.pages
);
288 ASSERTCMP(pv
.nr
, ==, count
);
290 for (loop
= 0; loop
< count
; loop
++) {
291 struct page
*page
= pv
.pages
[loop
];
293 redirty_page_for_writepage(wbc
, page
);
294 end_page_writeback(page
);
295 if (page
->index
>= first
)
296 first
= page
->index
+ 1;
299 __pagevec_release(&pv
);
300 } while (first
<= last
);
308 static int afs_store_data(struct address_space
*mapping
,
309 pgoff_t first
, pgoff_t last
,
310 unsigned offset
, unsigned to
)
312 struct afs_vnode
*vnode
= AFS_FS_I(mapping
->host
);
313 struct afs_fs_cursor fc
;
314 struct afs_wb_key
*wbk
= NULL
;
316 int ret
= -ENOKEY
, ret2
;
318 _enter("%s{%x:%u.%u},%lx,%lx,%x,%x",
323 first
, last
, offset
, to
);
325 spin_lock(&vnode
->wb_lock
);
326 p
= vnode
->wb_keys
.next
;
328 /* Iterate through the list looking for a valid key to use. */
330 while (p
!= &vnode
->wb_keys
) {
331 wbk
= list_entry(p
, struct afs_wb_key
, vnode_link
);
332 _debug("wbk %u", key_serial(wbk
->key
));
333 ret2
= key_validate(wbk
->key
);
341 spin_unlock(&vnode
->wb_lock
);
343 _leave(" = %d [no keys]", ret
);
347 refcount_inc(&wbk
->usage
);
348 spin_unlock(&vnode
->wb_lock
);
350 _debug("USE WB KEY %u", key_serial(wbk
->key
));
353 if (afs_begin_vnode_operation(&fc
, vnode
, wbk
->key
)) {
354 while (afs_select_fileserver(&fc
)) {
355 fc
.cb_break
= afs_calc_vnode_cb_break(vnode
);
356 afs_fs_store_data(&fc
, mapping
, first
, last
, offset
, to
);
359 afs_check_for_remote_deletion(&fc
, fc
.vnode
);
360 afs_vnode_commit_status(&fc
, vnode
, fc
.cb_break
);
361 ret
= afs_end_vnode_operation(&fc
);
366 afs_stat_v(vnode
, n_stores
);
367 atomic_long_add((last
* PAGE_SIZE
+ to
) -
368 (first
* PAGE_SIZE
+ offset
),
369 &afs_v2net(vnode
)->n_store_bytes
);
378 spin_lock(&vnode
->wb_lock
);
379 p
= wbk
->vnode_link
.next
;
385 _leave(" = %d", ret
);
390 * Synchronously write back the locked page and any subsequent non-locked dirty
393 static int afs_write_back_from_locked_page(struct address_space
*mapping
,
394 struct writeback_control
*wbc
,
395 struct page
*primary_page
,
398 struct afs_vnode
*vnode
= AFS_FS_I(mapping
->host
);
399 struct page
*pages
[8], *page
;
400 unsigned long count
, priv
;
401 unsigned n
, offset
, to
, f
, t
;
402 pgoff_t start
, first
, last
;
405 _enter(",%lx", primary_page
->index
);
408 if (test_set_page_writeback(primary_page
))
411 /* Find all consecutive lockable dirty pages that have contiguous
412 * written regions, stopping when we find a page that is not
413 * immediately lockable, is not dirty or is missing, or we reach the
416 start
= primary_page
->index
;
417 priv
= page_private(primary_page
);
418 offset
= priv
& AFS_PRIV_MAX
;
419 to
= priv
>> AFS_PRIV_SHIFT
;
420 trace_afs_page_dirty(vnode
, tracepoint_string("store"),
421 primary_page
->index
, priv
);
423 WARN_ON(offset
== to
);
425 trace_afs_page_dirty(vnode
, tracepoint_string("WARN"),
426 primary_page
->index
, priv
);
428 if (start
>= final_page
||
429 (to
< PAGE_SIZE
&& !test_bit(AFS_VNODE_NEW_CONTENT
, &vnode
->flags
)))
434 _debug("more %lx [%lx]", start
, count
);
435 n
= final_page
- start
+ 1;
436 if (n
> ARRAY_SIZE(pages
))
437 n
= ARRAY_SIZE(pages
);
438 n
= find_get_pages_contig(mapping
, start
, ARRAY_SIZE(pages
), pages
);
439 _debug("fgpc %u", n
);
442 if (pages
[0]->index
!= start
) {
444 put_page(pages
[--n
]);
449 for (loop
= 0; loop
< n
; loop
++) {
451 if (to
!= PAGE_SIZE
&&
452 !test_bit(AFS_VNODE_NEW_CONTENT
, &vnode
->flags
))
454 if (page
->index
> final_page
)
456 if (!trylock_page(page
))
458 if (!PageDirty(page
) || PageWriteback(page
)) {
463 priv
= page_private(page
);
464 f
= priv
& AFS_PRIV_MAX
;
465 t
= priv
>> AFS_PRIV_SHIFT
;
467 !test_bit(AFS_VNODE_NEW_CONTENT
, &vnode
->flags
)) {
473 trace_afs_page_dirty(vnode
, tracepoint_string("store+"),
476 if (!clear_page_dirty_for_io(page
))
478 if (test_set_page_writeback(page
))
485 for (; loop
< n
; loop
++)
486 put_page(pages
[loop
]);
491 } while (start
<= final_page
&& count
< 65536);
494 /* We now have a contiguous set of dirty pages, each with writeback
495 * set; the first page is still locked at this point, but all the rest
496 * have been unlocked.
498 unlock_page(primary_page
);
500 first
= primary_page
->index
;
501 last
= first
+ count
- 1;
503 _debug("write back %lx[%u..] to %lx[..%u]", first
, offset
, last
, to
);
505 ret
= afs_store_data(mapping
, first
, last
, offset
, to
);
512 pr_notice("kAFS: Unexpected error from FS.StoreData %d\n", ret
);
520 afs_redirty_pages(wbc
, mapping
, first
, last
);
521 mapping_set_error(mapping
, ret
);
526 afs_redirty_pages(wbc
, mapping
, first
, last
);
527 mapping_set_error(mapping
, -ENOSPC
);
537 afs_kill_pages(mapping
, first
, last
);
538 mapping_set_error(mapping
, ret
);
542 _leave(" = %d", ret
);
547 * write a page back to the server
548 * - the caller locked the page for us
550 int afs_writepage(struct page
*page
, struct writeback_control
*wbc
)
554 _enter("{%lx},", page
->index
);
556 ret
= afs_write_back_from_locked_page(page
->mapping
, wbc
, page
,
557 wbc
->range_end
>> PAGE_SHIFT
);
559 _leave(" = %d", ret
);
563 wbc
->nr_to_write
-= ret
;
570 * write a region of pages back to the server
572 static int afs_writepages_region(struct address_space
*mapping
,
573 struct writeback_control
*wbc
,
574 pgoff_t index
, pgoff_t end
, pgoff_t
*_next
)
579 _enter(",,%lx,%lx,", index
, end
);
582 n
= find_get_pages_range_tag(mapping
, &index
, end
,
583 PAGECACHE_TAG_DIRTY
, 1, &page
);
587 _debug("wback %lx", page
->index
);
590 * at this point we hold neither the i_pages lock nor the
591 * page lock: the page may be truncated or invalidated
592 * (changing page->mapping to NULL), or even swizzled
593 * back from swapper_space to tmpfs file mapping
595 ret
= lock_page_killable(page
);
598 _leave(" = %d", ret
);
602 if (page
->mapping
!= mapping
|| !PageDirty(page
)) {
608 if (PageWriteback(page
)) {
610 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
611 wait_on_page_writeback(page
);
616 if (!clear_page_dirty_for_io(page
))
618 ret
= afs_write_back_from_locked_page(mapping
, wbc
, page
, end
);
621 _leave(" = %d", ret
);
625 wbc
->nr_to_write
-= ret
;
628 } while (index
< end
&& wbc
->nr_to_write
> 0);
631 _leave(" = 0 [%lx]", *_next
);
636 * write some of the pending data back to the server
638 int afs_writepages(struct address_space
*mapping
,
639 struct writeback_control
*wbc
)
641 pgoff_t start
, end
, next
;
646 if (wbc
->range_cyclic
) {
647 start
= mapping
->writeback_index
;
649 ret
= afs_writepages_region(mapping
, wbc
, start
, end
, &next
);
650 if (start
> 0 && wbc
->nr_to_write
> 0 && ret
== 0)
651 ret
= afs_writepages_region(mapping
, wbc
, 0, start
,
653 mapping
->writeback_index
= next
;
654 } else if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
) {
655 end
= (pgoff_t
)(LLONG_MAX
>> PAGE_SHIFT
);
656 ret
= afs_writepages_region(mapping
, wbc
, 0, end
, &next
);
657 if (wbc
->nr_to_write
> 0)
658 mapping
->writeback_index
= next
;
660 start
= wbc
->range_start
>> PAGE_SHIFT
;
661 end
= wbc
->range_end
>> PAGE_SHIFT
;
662 ret
= afs_writepages_region(mapping
, wbc
, start
, end
, &next
);
665 _leave(" = %d", ret
);
670 * completion of write to server
672 void afs_pages_written_back(struct afs_vnode
*vnode
, struct afs_call
*call
)
676 unsigned count
, loop
;
677 pgoff_t first
= call
->first
, last
= call
->last
;
679 _enter("{%x:%u},{%lx-%lx}",
680 vnode
->fid
.vid
, vnode
->fid
.vnode
, first
, last
);
685 _debug("done %lx-%lx", first
, last
);
687 count
= last
- first
+ 1;
688 if (count
> PAGEVEC_SIZE
)
689 count
= PAGEVEC_SIZE
;
690 pv
.nr
= find_get_pages_contig(vnode
->vfs_inode
.i_mapping
,
691 first
, count
, pv
.pages
);
692 ASSERTCMP(pv
.nr
, ==, count
);
694 for (loop
= 0; loop
< count
; loop
++) {
695 priv
= page_private(pv
.pages
[loop
]);
696 trace_afs_page_dirty(vnode
, tracepoint_string("clear"),
697 pv
.pages
[loop
]->index
, priv
);
698 set_page_private(pv
.pages
[loop
], 0);
699 end_page_writeback(pv
.pages
[loop
]);
702 __pagevec_release(&pv
);
703 } while (first
<= last
);
705 afs_prune_wb_keys(vnode
);
710 * write to an AFS file
712 ssize_t
afs_file_write(struct kiocb
*iocb
, struct iov_iter
*from
)
714 struct afs_vnode
*vnode
= AFS_FS_I(file_inode(iocb
->ki_filp
));
716 size_t count
= iov_iter_count(from
);
718 _enter("{%x.%u},{%zu},",
719 vnode
->fid
.vid
, vnode
->fid
.vnode
, count
);
721 if (IS_SWAPFILE(&vnode
->vfs_inode
)) {
723 "AFS: Attempt to write to active swap file!\n");
730 result
= generic_file_write_iter(iocb
, from
);
732 _leave(" = %zd", result
);
737 * flush any dirty pages for this process, and check for write errors.
738 * - the return status from this call provides a reliable indication of
739 * whether any write errors occurred for this process.
741 int afs_fsync(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
743 struct inode
*inode
= file_inode(file
);
744 struct afs_vnode
*vnode
= AFS_FS_I(inode
);
746 _enter("{%x:%u},{n=%pD},%d",
747 vnode
->fid
.vid
, vnode
->fid
.vnode
, file
,
750 return file_write_and_wait_range(file
, start
, end
);
754 * notification that a previously read-only page is about to become writable
755 * - if it returns an error, the caller will deliver a bus error signal
757 vm_fault_t
afs_page_mkwrite(struct vm_fault
*vmf
)
759 struct file
*file
= vmf
->vma
->vm_file
;
760 struct inode
*inode
= file_inode(file
);
761 struct afs_vnode
*vnode
= AFS_FS_I(inode
);
764 _enter("{{%x:%u}},{%lx}",
765 vnode
->fid
.vid
, vnode
->fid
.vnode
, vmf
->page
->index
);
767 sb_start_pagefault(inode
->i_sb
);
769 /* Wait for the page to be written to the cache before we allow it to
770 * be modified. We then assume the entire page will need writing back.
772 #ifdef CONFIG_AFS_FSCACHE
773 fscache_wait_on_page_write(vnode
->cache
, vmf
->page
);
776 if (PageWriteback(vmf
->page
) &&
777 wait_on_page_bit_killable(vmf
->page
, PG_writeback
) < 0)
778 return VM_FAULT_RETRY
;
780 if (lock_page_killable(vmf
->page
) < 0)
781 return VM_FAULT_RETRY
;
783 /* We mustn't change page->private until writeback is complete as that
784 * details the portion of the page we need to write back and we might
785 * need to redirty the page if there's a problem.
787 wait_on_page_writeback(vmf
->page
);
789 priv
= (unsigned long)PAGE_SIZE
<< AFS_PRIV_SHIFT
; /* To */
790 priv
|= 0; /* From */
791 trace_afs_page_dirty(vnode
, tracepoint_string("mkwrite"),
792 vmf
->page
->index
, priv
);
793 SetPagePrivate(vmf
->page
);
794 set_page_private(vmf
->page
, priv
);
796 sb_end_pagefault(inode
->i_sb
);
797 return VM_FAULT_LOCKED
;
801 * Prune the keys cached for writeback. The caller must hold vnode->wb_lock.
803 void afs_prune_wb_keys(struct afs_vnode
*vnode
)
805 LIST_HEAD(graveyard
);
806 struct afs_wb_key
*wbk
, *tmp
;
808 /* Discard unused keys */
809 spin_lock(&vnode
->wb_lock
);
811 if (!mapping_tagged(&vnode
->vfs_inode
.i_data
, PAGECACHE_TAG_WRITEBACK
) &&
812 !mapping_tagged(&vnode
->vfs_inode
.i_data
, PAGECACHE_TAG_DIRTY
)) {
813 list_for_each_entry_safe(wbk
, tmp
, &vnode
->wb_keys
, vnode_link
) {
814 if (refcount_read(&wbk
->usage
) == 1)
815 list_move(&wbk
->vnode_link
, &graveyard
);
819 spin_unlock(&vnode
->wb_lock
);
821 while (!list_empty(&graveyard
)) {
822 wbk
= list_entry(graveyard
.next
, struct afs_wb_key
, vnode_link
);
823 list_del(&wbk
->vnode_link
);
829 * Clean up a page during invalidation.
831 int afs_launder_page(struct page
*page
)
833 struct address_space
*mapping
= page
->mapping
;
834 struct afs_vnode
*vnode
= AFS_FS_I(mapping
->host
);
839 _enter("{%lx}", page
->index
);
841 priv
= page_private(page
);
842 if (clear_page_dirty_for_io(page
)) {
845 if (PagePrivate(page
)) {
846 f
= priv
& AFS_PRIV_MAX
;
847 t
= priv
>> AFS_PRIV_SHIFT
;
850 trace_afs_page_dirty(vnode
, tracepoint_string("launder"),
852 ret
= afs_store_data(mapping
, page
->index
, page
->index
, t
, f
);
855 trace_afs_page_dirty(vnode
, tracepoint_string("laundered"),
857 set_page_private(page
, 0);
858 ClearPagePrivate(page
);
860 #ifdef CONFIG_AFS_FSCACHE
861 if (PageFsCache(page
)) {
862 fscache_wait_on_page_write(vnode
->cache
, page
);
863 fscache_uncache_page(vnode
->cache
, page
);