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
)
40 _enter(",,%llu", (unsigned long long)pos
);
42 if (pos
>= vnode
->vfs_inode
.i_size
) {
44 ASSERTCMP(p
+ len
, <=, PAGE_SIZE
);
46 memset(data
+ p
, 0, len
);
51 req
= kzalloc(sizeof(struct afs_read
) + sizeof(struct page
*),
56 refcount_set(&req
->usage
, 1);
60 req
->pages
= req
->array
;
64 ret
= afs_fetch_data(vnode
, key
, req
);
68 _debug("got NOENT from server"
69 " - marking file deleted and stale");
70 set_bit(AFS_VNODE_DELETED
, &vnode
->flags
);
80 * prepare to perform part of a write to a page
82 int afs_write_begin(struct file
*file
, struct address_space
*mapping
,
83 loff_t pos
, unsigned len
, unsigned flags
,
84 struct page
**pagep
, void **fsdata
)
86 struct afs_vnode
*vnode
= AFS_FS_I(file_inode(file
));
88 struct key
*key
= afs_file_key(file
);
90 unsigned f
, from
= pos
& (PAGE_SIZE
- 1);
91 unsigned t
, to
= from
+ len
;
92 pgoff_t index
= pos
>> PAGE_SHIFT
;
95 _enter("{%llx:%llu},{%lx},%u,%u",
96 vnode
->fid
.vid
, vnode
->fid
.vnode
, index
, from
, to
);
98 /* We want to store information about how much of a page is altered in
101 BUILD_BUG_ON(PAGE_SIZE
> 32768 && sizeof(page
->private) < 8);
103 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
107 if (!PageUptodate(page
) && len
!= PAGE_SIZE
) {
108 ret
= afs_fill_page(vnode
, key
, pos
& PAGE_MASK
, PAGE_SIZE
, page
);
112 _leave(" = %d [prep]", ret
);
115 SetPageUptodate(page
);
118 /* page won't leak in error case: it eventually gets cleaned off LRU */
122 /* See if this page is already partially written in a way that we can
123 * merge the new write with.
126 if (PagePrivate(page
)) {
127 priv
= page_private(page
);
128 f
= priv
& AFS_PRIV_MAX
;
129 t
= priv
>> AFS_PRIV_SHIFT
;
134 if (PageWriteback(page
)) {
135 trace_afs_page_dirty(vnode
, tracepoint_string("alrdy"),
137 goto flush_conflicting_write
;
139 /* If the file is being filled locally, allow inter-write
140 * spaces to be merged into writes. If it's not, only write
141 * back what the user gives us.
143 if (!test_bit(AFS_VNODE_NEW_CONTENT
, &vnode
->flags
) &&
144 (to
< f
|| from
> t
))
145 goto flush_conflicting_write
;
155 priv
= (unsigned long)t
<< AFS_PRIV_SHIFT
;
157 trace_afs_page_dirty(vnode
, tracepoint_string("begin"),
159 SetPagePrivate(page
);
160 set_page_private(page
, priv
);
164 /* The previous write and this write aren't adjacent or overlapping, so
165 * flush the page out.
167 flush_conflicting_write
:
168 _debug("flush conflict");
169 ret
= write_one_page(page
);
171 _leave(" = %d", ret
);
175 ret
= lock_page_killable(page
);
177 _leave(" = %d", ret
);
184 * finalise part of a write to a page
186 int afs_write_end(struct file
*file
, struct address_space
*mapping
,
187 loff_t pos
, unsigned len
, unsigned copied
,
188 struct page
*page
, void *fsdata
)
190 struct afs_vnode
*vnode
= AFS_FS_I(file_inode(file
));
191 struct key
*key
= afs_file_key(file
);
192 loff_t i_size
, maybe_i_size
;
195 _enter("{%llx:%llu},{%lx}",
196 vnode
->fid
.vid
, vnode
->fid
.vnode
, page
->index
);
198 maybe_i_size
= pos
+ copied
;
200 i_size
= i_size_read(&vnode
->vfs_inode
);
201 if (maybe_i_size
> i_size
) {
202 spin_lock(&vnode
->wb_lock
);
203 i_size
= i_size_read(&vnode
->vfs_inode
);
204 if (maybe_i_size
> i_size
)
205 i_size_write(&vnode
->vfs_inode
, maybe_i_size
);
206 spin_unlock(&vnode
->wb_lock
);
209 if (!PageUptodate(page
)) {
211 /* Try and load any missing data from the server. The
212 * unmarshalling routine will take care of clearing any
213 * bits that are beyond the EOF.
215 ret
= afs_fill_page(vnode
, key
, pos
+ copied
,
220 SetPageUptodate(page
);
223 set_page_dirty(page
);
235 * kill all the pages in the given range
237 static void afs_kill_pages(struct address_space
*mapping
,
238 pgoff_t first
, pgoff_t last
)
240 struct afs_vnode
*vnode
= AFS_FS_I(mapping
->host
);
242 unsigned count
, loop
;
244 _enter("{%llx:%llu},%lx-%lx",
245 vnode
->fid
.vid
, vnode
->fid
.vnode
, first
, last
);
250 _debug("kill %lx-%lx", first
, last
);
252 count
= last
- first
+ 1;
253 if (count
> PAGEVEC_SIZE
)
254 count
= PAGEVEC_SIZE
;
255 pv
.nr
= find_get_pages_contig(mapping
, first
, count
, pv
.pages
);
256 ASSERTCMP(pv
.nr
, ==, count
);
258 for (loop
= 0; loop
< count
; loop
++) {
259 struct page
*page
= pv
.pages
[loop
];
260 ClearPageUptodate(page
);
262 end_page_writeback(page
);
263 if (page
->index
>= first
)
264 first
= page
->index
+ 1;
266 generic_error_remove_page(mapping
, page
);
269 __pagevec_release(&pv
);
270 } while (first
<= last
);
276 * Redirty all the pages in a given range.
278 static void afs_redirty_pages(struct writeback_control
*wbc
,
279 struct address_space
*mapping
,
280 pgoff_t first
, pgoff_t last
)
282 struct afs_vnode
*vnode
= AFS_FS_I(mapping
->host
);
284 unsigned count
, loop
;
286 _enter("{%llx:%llu},%lx-%lx",
287 vnode
->fid
.vid
, vnode
->fid
.vnode
, first
, last
);
292 _debug("redirty %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(mapping
, first
, count
, pv
.pages
);
298 ASSERTCMP(pv
.nr
, ==, count
);
300 for (loop
= 0; loop
< count
; loop
++) {
301 struct page
*page
= pv
.pages
[loop
];
303 redirty_page_for_writepage(wbc
, page
);
304 end_page_writeback(page
);
305 if (page
->index
>= first
)
306 first
= page
->index
+ 1;
309 __pagevec_release(&pv
);
310 } while (first
<= last
);
318 static int afs_store_data(struct address_space
*mapping
,
319 pgoff_t first
, pgoff_t last
,
320 unsigned offset
, unsigned to
)
322 struct afs_vnode
*vnode
= AFS_FS_I(mapping
->host
);
323 struct afs_fs_cursor fc
;
324 struct afs_wb_key
*wbk
= NULL
;
326 int ret
= -ENOKEY
, ret2
;
328 _enter("%s{%llx:%llu.%u},%lx,%lx,%x,%x",
333 first
, last
, offset
, to
);
335 spin_lock(&vnode
->wb_lock
);
336 p
= vnode
->wb_keys
.next
;
338 /* Iterate through the list looking for a valid key to use. */
340 while (p
!= &vnode
->wb_keys
) {
341 wbk
= list_entry(p
, struct afs_wb_key
, vnode_link
);
342 _debug("wbk %u", key_serial(wbk
->key
));
343 ret2
= key_validate(wbk
->key
);
351 spin_unlock(&vnode
->wb_lock
);
353 _leave(" = %d [no keys]", ret
);
357 refcount_inc(&wbk
->usage
);
358 spin_unlock(&vnode
->wb_lock
);
360 _debug("USE WB KEY %u", key_serial(wbk
->key
));
363 if (afs_begin_vnode_operation(&fc
, vnode
, wbk
->key
)) {
364 while (afs_select_fileserver(&fc
)) {
365 fc
.cb_break
= afs_calc_vnode_cb_break(vnode
);
366 afs_fs_store_data(&fc
, mapping
, first
, last
, offset
, to
);
369 afs_check_for_remote_deletion(&fc
, fc
.vnode
);
370 afs_vnode_commit_status(&fc
, vnode
, fc
.cb_break
);
371 ret
= afs_end_vnode_operation(&fc
);
376 afs_stat_v(vnode
, n_stores
);
377 atomic_long_add((last
* PAGE_SIZE
+ to
) -
378 (first
* PAGE_SIZE
+ offset
),
379 &afs_v2net(vnode
)->n_store_bytes
);
388 spin_lock(&vnode
->wb_lock
);
389 p
= wbk
->vnode_link
.next
;
395 _leave(" = %d", ret
);
400 * Synchronously write back the locked page and any subsequent non-locked dirty
403 static int afs_write_back_from_locked_page(struct address_space
*mapping
,
404 struct writeback_control
*wbc
,
405 struct page
*primary_page
,
408 struct afs_vnode
*vnode
= AFS_FS_I(mapping
->host
);
409 struct page
*pages
[8], *page
;
410 unsigned long count
, priv
;
411 unsigned n
, offset
, to
, f
, t
;
412 pgoff_t start
, first
, last
;
415 _enter(",%lx", primary_page
->index
);
418 if (test_set_page_writeback(primary_page
))
421 /* Find all consecutive lockable dirty pages that have contiguous
422 * written regions, stopping when we find a page that is not
423 * immediately lockable, is not dirty or is missing, or we reach the
426 start
= primary_page
->index
;
427 priv
= page_private(primary_page
);
428 offset
= priv
& AFS_PRIV_MAX
;
429 to
= priv
>> AFS_PRIV_SHIFT
;
430 trace_afs_page_dirty(vnode
, tracepoint_string("store"),
431 primary_page
->index
, priv
);
433 WARN_ON(offset
== to
);
435 trace_afs_page_dirty(vnode
, tracepoint_string("WARN"),
436 primary_page
->index
, priv
);
438 if (start
>= final_page
||
439 (to
< PAGE_SIZE
&& !test_bit(AFS_VNODE_NEW_CONTENT
, &vnode
->flags
)))
444 _debug("more %lx [%lx]", start
, count
);
445 n
= final_page
- start
+ 1;
446 if (n
> ARRAY_SIZE(pages
))
447 n
= ARRAY_SIZE(pages
);
448 n
= find_get_pages_contig(mapping
, start
, ARRAY_SIZE(pages
), pages
);
449 _debug("fgpc %u", n
);
452 if (pages
[0]->index
!= start
) {
454 put_page(pages
[--n
]);
459 for (loop
= 0; loop
< n
; loop
++) {
461 if (to
!= PAGE_SIZE
&&
462 !test_bit(AFS_VNODE_NEW_CONTENT
, &vnode
->flags
))
464 if (page
->index
> final_page
)
466 if (!trylock_page(page
))
468 if (!PageDirty(page
) || PageWriteback(page
)) {
473 priv
= page_private(page
);
474 f
= priv
& AFS_PRIV_MAX
;
475 t
= priv
>> AFS_PRIV_SHIFT
;
477 !test_bit(AFS_VNODE_NEW_CONTENT
, &vnode
->flags
)) {
483 trace_afs_page_dirty(vnode
, tracepoint_string("store+"),
486 if (!clear_page_dirty_for_io(page
))
488 if (test_set_page_writeback(page
))
495 for (; loop
< n
; loop
++)
496 put_page(pages
[loop
]);
501 } while (start
<= final_page
&& count
< 65536);
504 /* We now have a contiguous set of dirty pages, each with writeback
505 * set; the first page is still locked at this point, but all the rest
506 * have been unlocked.
508 unlock_page(primary_page
);
510 first
= primary_page
->index
;
511 last
= first
+ count
- 1;
513 _debug("write back %lx[%u..] to %lx[..%u]", first
, offset
, last
, to
);
515 ret
= afs_store_data(mapping
, first
, last
, offset
, to
);
522 pr_notice("kAFS: Unexpected error from FS.StoreData %d\n", ret
);
530 afs_redirty_pages(wbc
, mapping
, first
, last
);
531 mapping_set_error(mapping
, ret
);
536 afs_redirty_pages(wbc
, mapping
, first
, last
);
537 mapping_set_error(mapping
, -ENOSPC
);
547 trace_afs_file_error(vnode
, ret
, afs_file_error_writeback_fail
);
548 afs_kill_pages(mapping
, first
, last
);
549 mapping_set_error(mapping
, ret
);
553 _leave(" = %d", ret
);
558 * write a page back to the server
559 * - the caller locked the page for us
561 int afs_writepage(struct page
*page
, struct writeback_control
*wbc
)
565 _enter("{%lx},", page
->index
);
567 ret
= afs_write_back_from_locked_page(page
->mapping
, wbc
, page
,
568 wbc
->range_end
>> PAGE_SHIFT
);
570 _leave(" = %d", ret
);
574 wbc
->nr_to_write
-= ret
;
581 * write a region of pages back to the server
583 static int afs_writepages_region(struct address_space
*mapping
,
584 struct writeback_control
*wbc
,
585 pgoff_t index
, pgoff_t end
, pgoff_t
*_next
)
590 _enter(",,%lx,%lx,", index
, end
);
593 n
= find_get_pages_range_tag(mapping
, &index
, end
,
594 PAGECACHE_TAG_DIRTY
, 1, &page
);
598 _debug("wback %lx", page
->index
);
601 * at this point we hold neither the i_pages lock nor the
602 * page lock: the page may be truncated or invalidated
603 * (changing page->mapping to NULL), or even swizzled
604 * back from swapper_space to tmpfs file mapping
606 ret
= lock_page_killable(page
);
609 _leave(" = %d", ret
);
613 if (page
->mapping
!= mapping
|| !PageDirty(page
)) {
619 if (PageWriteback(page
)) {
621 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
622 wait_on_page_writeback(page
);
627 if (!clear_page_dirty_for_io(page
))
629 ret
= afs_write_back_from_locked_page(mapping
, wbc
, page
, end
);
632 _leave(" = %d", ret
);
636 wbc
->nr_to_write
-= ret
;
639 } while (index
< end
&& wbc
->nr_to_write
> 0);
642 _leave(" = 0 [%lx]", *_next
);
647 * write some of the pending data back to the server
649 int afs_writepages(struct address_space
*mapping
,
650 struct writeback_control
*wbc
)
652 pgoff_t start
, end
, next
;
657 if (wbc
->range_cyclic
) {
658 start
= mapping
->writeback_index
;
660 ret
= afs_writepages_region(mapping
, wbc
, start
, end
, &next
);
661 if (start
> 0 && wbc
->nr_to_write
> 0 && ret
== 0)
662 ret
= afs_writepages_region(mapping
, wbc
, 0, start
,
664 mapping
->writeback_index
= next
;
665 } else if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
) {
666 end
= (pgoff_t
)(LLONG_MAX
>> PAGE_SHIFT
);
667 ret
= afs_writepages_region(mapping
, wbc
, 0, end
, &next
);
668 if (wbc
->nr_to_write
> 0)
669 mapping
->writeback_index
= next
;
671 start
= wbc
->range_start
>> PAGE_SHIFT
;
672 end
= wbc
->range_end
>> PAGE_SHIFT
;
673 ret
= afs_writepages_region(mapping
, wbc
, start
, end
, &next
);
676 _leave(" = %d", ret
);
681 * completion of write to server
683 void afs_pages_written_back(struct afs_vnode
*vnode
, struct afs_call
*call
)
687 unsigned count
, loop
;
688 pgoff_t first
= call
->first
, last
= call
->last
;
690 _enter("{%llx:%llu},{%lx-%lx}",
691 vnode
->fid
.vid
, vnode
->fid
.vnode
, first
, last
);
696 _debug("done %lx-%lx", first
, last
);
698 count
= last
- first
+ 1;
699 if (count
> PAGEVEC_SIZE
)
700 count
= PAGEVEC_SIZE
;
701 pv
.nr
= find_get_pages_contig(vnode
->vfs_inode
.i_mapping
,
702 first
, count
, pv
.pages
);
703 ASSERTCMP(pv
.nr
, ==, count
);
705 for (loop
= 0; loop
< count
; loop
++) {
706 priv
= page_private(pv
.pages
[loop
]);
707 trace_afs_page_dirty(vnode
, tracepoint_string("clear"),
708 pv
.pages
[loop
]->index
, priv
);
709 set_page_private(pv
.pages
[loop
], 0);
710 end_page_writeback(pv
.pages
[loop
]);
713 __pagevec_release(&pv
);
714 } while (first
<= last
);
716 afs_prune_wb_keys(vnode
);
721 * write to an AFS file
723 ssize_t
afs_file_write(struct kiocb
*iocb
, struct iov_iter
*from
)
725 struct afs_vnode
*vnode
= AFS_FS_I(file_inode(iocb
->ki_filp
));
727 size_t count
= iov_iter_count(from
);
729 _enter("{%llx:%llu},{%zu},",
730 vnode
->fid
.vid
, vnode
->fid
.vnode
, count
);
732 if (IS_SWAPFILE(&vnode
->vfs_inode
)) {
734 "AFS: Attempt to write to active swap file!\n");
741 result
= generic_file_write_iter(iocb
, from
);
743 _leave(" = %zd", result
);
748 * flush any dirty pages for this process, and check for write errors.
749 * - the return status from this call provides a reliable indication of
750 * whether any write errors occurred for this process.
752 int afs_fsync(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
754 struct inode
*inode
= file_inode(file
);
755 struct afs_vnode
*vnode
= AFS_FS_I(inode
);
757 _enter("{%llx:%llu},{n=%pD},%d",
758 vnode
->fid
.vid
, vnode
->fid
.vnode
, file
,
761 return file_write_and_wait_range(file
, start
, end
);
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 vm_fault_t
afs_page_mkwrite(struct vm_fault
*vmf
)
770 struct file
*file
= vmf
->vma
->vm_file
;
771 struct inode
*inode
= file_inode(file
);
772 struct afs_vnode
*vnode
= AFS_FS_I(inode
);
775 _enter("{{%llx:%llu}},{%lx}",
776 vnode
->fid
.vid
, vnode
->fid
.vnode
, vmf
->page
->index
);
778 sb_start_pagefault(inode
->i_sb
);
780 /* Wait for the page to be written to the cache before we allow it to
781 * be modified. We then assume the entire page will need writing back.
783 #ifdef CONFIG_AFS_FSCACHE
784 fscache_wait_on_page_write(vnode
->cache
, vmf
->page
);
787 if (PageWriteback(vmf
->page
) &&
788 wait_on_page_bit_killable(vmf
->page
, PG_writeback
) < 0)
789 return VM_FAULT_RETRY
;
791 if (lock_page_killable(vmf
->page
) < 0)
792 return VM_FAULT_RETRY
;
794 /* We mustn't change page->private until writeback is complete as that
795 * details the portion of the page we need to write back and we might
796 * need to redirty the page if there's a problem.
798 wait_on_page_writeback(vmf
->page
);
800 priv
= (unsigned long)PAGE_SIZE
<< AFS_PRIV_SHIFT
; /* To */
801 priv
|= 0; /* From */
802 trace_afs_page_dirty(vnode
, tracepoint_string("mkwrite"),
803 vmf
->page
->index
, priv
);
804 SetPagePrivate(vmf
->page
);
805 set_page_private(vmf
->page
, priv
);
807 sb_end_pagefault(inode
->i_sb
);
808 return VM_FAULT_LOCKED
;
812 * Prune the keys cached for writeback. The caller must hold vnode->wb_lock.
814 void afs_prune_wb_keys(struct afs_vnode
*vnode
)
816 LIST_HEAD(graveyard
);
817 struct afs_wb_key
*wbk
, *tmp
;
819 /* Discard unused keys */
820 spin_lock(&vnode
->wb_lock
);
822 if (!mapping_tagged(&vnode
->vfs_inode
.i_data
, PAGECACHE_TAG_WRITEBACK
) &&
823 !mapping_tagged(&vnode
->vfs_inode
.i_data
, PAGECACHE_TAG_DIRTY
)) {
824 list_for_each_entry_safe(wbk
, tmp
, &vnode
->wb_keys
, vnode_link
) {
825 if (refcount_read(&wbk
->usage
) == 1)
826 list_move(&wbk
->vnode_link
, &graveyard
);
830 spin_unlock(&vnode
->wb_lock
);
832 while (!list_empty(&graveyard
)) {
833 wbk
= list_entry(graveyard
.next
, struct afs_wb_key
, vnode_link
);
834 list_del(&wbk
->vnode_link
);
840 * Clean up a page during invalidation.
842 int afs_launder_page(struct page
*page
)
844 struct address_space
*mapping
= page
->mapping
;
845 struct afs_vnode
*vnode
= AFS_FS_I(mapping
->host
);
850 _enter("{%lx}", page
->index
);
852 priv
= page_private(page
);
853 if (clear_page_dirty_for_io(page
)) {
856 if (PagePrivate(page
)) {
857 f
= priv
& AFS_PRIV_MAX
;
858 t
= priv
>> AFS_PRIV_SHIFT
;
861 trace_afs_page_dirty(vnode
, tracepoint_string("launder"),
863 ret
= afs_store_data(mapping
, page
->index
, page
->index
, t
, f
);
866 trace_afs_page_dirty(vnode
, tracepoint_string("laundered"),
868 set_page_private(page
, 0);
869 ClearPagePrivate(page
);
871 #ifdef CONFIG_AFS_FSCACHE
872 if (PageFsCache(page
)) {
873 fscache_wait_on_page_write(vnode
->cache
, page
);
874 fscache_uncache_page(vnode
->cache
, page
);