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 atomic_set(&req
->usage
, 1);
52 ret
= afs_fetch_data(vnode
, key
, req
);
56 _debug("got NOENT from server"
57 " - marking file deleted and stale");
58 set_bit(AFS_VNODE_DELETED
, &vnode
->flags
);
68 * prepare to perform part of a write to a page
70 int afs_write_begin(struct file
*file
, struct address_space
*mapping
,
71 loff_t pos
, unsigned len
, unsigned flags
,
72 struct page
**pagep
, void **fsdata
)
74 struct afs_vnode
*vnode
= AFS_FS_I(file_inode(file
));
76 struct key
*key
= afs_file_key(file
);
78 unsigned f
, from
= pos
& (PAGE_SIZE
- 1);
79 unsigned t
, to
= from
+ len
;
80 pgoff_t index
= pos
>> PAGE_SHIFT
;
83 _enter("{%x:%u},{%lx},%u,%u",
84 vnode
->fid
.vid
, vnode
->fid
.vnode
, index
, from
, to
);
86 /* We want to store information about how much of a page is altered in
89 BUILD_BUG_ON(PAGE_SIZE
> 32768 && sizeof(page
->private) < 8);
91 page
= grab_cache_page_write_begin(mapping
, index
, flags
);
95 if (!PageUptodate(page
) && len
!= PAGE_SIZE
) {
96 ret
= afs_fill_page(vnode
, key
, pos
& PAGE_MASK
, PAGE_SIZE
, page
);
100 _leave(" = %d [prep]", ret
);
103 SetPageUptodate(page
);
106 /* page won't leak in error case: it eventually gets cleaned off LRU */
110 /* See if this page is already partially written in a way that we can
111 * merge the new write with.
114 if (PagePrivate(page
)) {
115 priv
= page_private(page
);
116 f
= priv
& AFS_PRIV_MAX
;
117 t
= priv
>> AFS_PRIV_SHIFT
;
122 if (PageWriteback(page
)) {
123 trace_afs_page_dirty(vnode
, tracepoint_string("alrdy"),
125 goto flush_conflicting_write
;
127 if (to
< f
|| from
> t
)
128 goto flush_conflicting_write
;
138 priv
= (unsigned long)t
<< AFS_PRIV_SHIFT
;
140 trace_afs_page_dirty(vnode
, tracepoint_string("begin"),
142 SetPagePrivate(page
);
143 set_page_private(page
, priv
);
147 /* The previous write and this write aren't adjacent or overlapping, so
148 * flush the page out.
150 flush_conflicting_write
:
151 _debug("flush conflict");
152 ret
= write_one_page(page
);
154 _leave(" = %d", ret
);
158 ret
= lock_page_killable(page
);
160 _leave(" = %d", ret
);
167 * finalise part of a write to a page
169 int afs_write_end(struct file
*file
, struct address_space
*mapping
,
170 loff_t pos
, unsigned len
, unsigned copied
,
171 struct page
*page
, void *fsdata
)
173 struct afs_vnode
*vnode
= AFS_FS_I(file_inode(file
));
174 struct key
*key
= afs_file_key(file
);
175 loff_t i_size
, maybe_i_size
;
178 _enter("{%x:%u},{%lx}",
179 vnode
->fid
.vid
, vnode
->fid
.vnode
, page
->index
);
181 maybe_i_size
= pos
+ copied
;
183 i_size
= i_size_read(&vnode
->vfs_inode
);
184 if (maybe_i_size
> i_size
) {
185 spin_lock(&vnode
->wb_lock
);
186 i_size
= i_size_read(&vnode
->vfs_inode
);
187 if (maybe_i_size
> i_size
)
188 i_size_write(&vnode
->vfs_inode
, maybe_i_size
);
189 spin_unlock(&vnode
->wb_lock
);
192 if (!PageUptodate(page
)) {
194 /* Try and load any missing data from the server. The
195 * unmarshalling routine will take care of clearing any
196 * bits that are beyond the EOF.
198 ret
= afs_fill_page(vnode
, key
, pos
+ copied
,
203 SetPageUptodate(page
);
206 set_page_dirty(page
);
218 * kill all the pages in the given range
220 static void afs_kill_pages(struct address_space
*mapping
,
221 pgoff_t first
, pgoff_t last
)
223 struct afs_vnode
*vnode
= AFS_FS_I(mapping
->host
);
225 unsigned count
, loop
;
227 _enter("{%x:%u},%lx-%lx",
228 vnode
->fid
.vid
, vnode
->fid
.vnode
, first
, last
);
233 _debug("kill %lx-%lx", first
, last
);
235 count
= last
- first
+ 1;
236 if (count
> PAGEVEC_SIZE
)
237 count
= PAGEVEC_SIZE
;
238 pv
.nr
= find_get_pages_contig(mapping
, first
, count
, pv
.pages
);
239 ASSERTCMP(pv
.nr
, ==, count
);
241 for (loop
= 0; loop
< count
; loop
++) {
242 struct page
*page
= pv
.pages
[loop
];
243 ClearPageUptodate(page
);
245 end_page_writeback(page
);
246 if (page
->index
>= first
)
247 first
= page
->index
+ 1;
249 generic_error_remove_page(mapping
, page
);
252 __pagevec_release(&pv
);
253 } while (first
<= last
);
259 * Redirty all the pages in a given range.
261 static void afs_redirty_pages(struct writeback_control
*wbc
,
262 struct address_space
*mapping
,
263 pgoff_t first
, pgoff_t last
)
265 struct afs_vnode
*vnode
= AFS_FS_I(mapping
->host
);
267 unsigned count
, loop
;
269 _enter("{%x:%u},%lx-%lx",
270 vnode
->fid
.vid
, vnode
->fid
.vnode
, first
, last
);
275 _debug("redirty %lx-%lx", first
, last
);
277 count
= last
- first
+ 1;
278 if (count
> PAGEVEC_SIZE
)
279 count
= PAGEVEC_SIZE
;
280 pv
.nr
= find_get_pages_contig(mapping
, first
, count
, pv
.pages
);
281 ASSERTCMP(pv
.nr
, ==, count
);
283 for (loop
= 0; loop
< count
; loop
++) {
284 struct page
*page
= pv
.pages
[loop
];
286 redirty_page_for_writepage(wbc
, page
);
287 end_page_writeback(page
);
288 if (page
->index
>= first
)
289 first
= page
->index
+ 1;
292 __pagevec_release(&pv
);
293 } while (first
<= last
);
301 static int afs_store_data(struct address_space
*mapping
,
302 pgoff_t first
, pgoff_t last
,
303 unsigned offset
, unsigned to
)
305 struct afs_vnode
*vnode
= AFS_FS_I(mapping
->host
);
306 struct afs_fs_cursor fc
;
307 struct afs_wb_key
*wbk
= NULL
;
309 int ret
= -ENOKEY
, ret2
;
311 _enter("%s{%x:%u.%u},%lx,%lx,%x,%x",
316 first
, last
, offset
, to
);
318 spin_lock(&vnode
->wb_lock
);
319 p
= vnode
->wb_keys
.next
;
321 /* Iterate through the list looking for a valid key to use. */
323 while (p
!= &vnode
->wb_keys
) {
324 wbk
= list_entry(p
, struct afs_wb_key
, vnode_link
);
325 _debug("wbk %u", key_serial(wbk
->key
));
326 ret2
= key_validate(wbk
->key
);
334 spin_unlock(&vnode
->wb_lock
);
336 _leave(" = %d [no keys]", ret
);
340 refcount_inc(&wbk
->usage
);
341 spin_unlock(&vnode
->wb_lock
);
343 _debug("USE WB KEY %u", key_serial(wbk
->key
));
346 if (afs_begin_vnode_operation(&fc
, vnode
, wbk
->key
)) {
347 while (afs_select_fileserver(&fc
)) {
348 fc
.cb_break
= vnode
->cb_break
+ vnode
->cb_s_break
;
349 afs_fs_store_data(&fc
, mapping
, first
, last
, offset
, to
);
352 afs_check_for_remote_deletion(&fc
, fc
.vnode
);
353 afs_vnode_commit_status(&fc
, vnode
, fc
.cb_break
);
354 ret
= afs_end_vnode_operation(&fc
);
365 spin_lock(&vnode
->wb_lock
);
366 p
= wbk
->vnode_link
.next
;
372 _leave(" = %d", ret
);
377 * Synchronously write back the locked page and any subsequent non-locked dirty
380 static int afs_write_back_from_locked_page(struct address_space
*mapping
,
381 struct writeback_control
*wbc
,
382 struct page
*primary_page
,
385 struct afs_vnode
*vnode
= AFS_FS_I(mapping
->host
);
386 struct page
*pages
[8], *page
;
387 unsigned long count
, priv
;
388 unsigned n
, offset
, to
, f
, t
;
389 pgoff_t start
, first
, last
;
392 _enter(",%lx", primary_page
->index
);
395 if (test_set_page_writeback(primary_page
))
398 /* Find all consecutive lockable dirty pages that have contiguous
399 * written regions, stopping when we find a page that is not
400 * immediately lockable, is not dirty or is missing, or we reach the
403 start
= primary_page
->index
;
404 priv
= page_private(primary_page
);
405 offset
= priv
& AFS_PRIV_MAX
;
406 to
= priv
>> AFS_PRIV_SHIFT
;
407 trace_afs_page_dirty(vnode
, tracepoint_string("store"),
408 primary_page
->index
, priv
);
410 WARN_ON(offset
== to
);
412 trace_afs_page_dirty(vnode
, tracepoint_string("WARN"),
413 primary_page
->index
, priv
);
415 if (start
>= final_page
|| to
< PAGE_SIZE
)
420 _debug("more %lx [%lx]", start
, count
);
421 n
= final_page
- start
+ 1;
422 if (n
> ARRAY_SIZE(pages
))
423 n
= ARRAY_SIZE(pages
);
424 n
= find_get_pages_contig(mapping
, start
, ARRAY_SIZE(pages
), pages
);
425 _debug("fgpc %u", n
);
428 if (pages
[0]->index
!= start
) {
430 put_page(pages
[--n
]);
435 for (loop
= 0; loop
< n
; loop
++) {
439 if (page
->index
> final_page
)
441 if (!trylock_page(page
))
443 if (!PageDirty(page
) || PageWriteback(page
)) {
448 priv
= page_private(page
);
449 f
= priv
& AFS_PRIV_MAX
;
450 t
= priv
>> AFS_PRIV_SHIFT
;
457 trace_afs_page_dirty(vnode
, tracepoint_string("store+"),
460 if (!clear_page_dirty_for_io(page
))
462 if (test_set_page_writeback(page
))
469 for (; loop
< n
; loop
++)
470 put_page(pages
[loop
]);
475 } while (start
<= final_page
&& count
< 65536);
478 /* We now have a contiguous set of dirty pages, each with writeback
479 * set; the first page is still locked at this point, but all the rest
480 * have been unlocked.
482 unlock_page(primary_page
);
484 first
= primary_page
->index
;
485 last
= first
+ count
- 1;
487 _debug("write back %lx[%u..] to %lx[..%u]", first
, offset
, last
, to
);
489 ret
= afs_store_data(mapping
, first
, last
, offset
, to
);
496 pr_notice("kAFS: Unexpected error from FS.StoreData %d\n", ret
);
504 afs_redirty_pages(wbc
, mapping
, first
, last
);
505 mapping_set_error(mapping
, ret
);
510 afs_redirty_pages(wbc
, mapping
, first
, last
);
511 mapping_set_error(mapping
, -ENOSPC
);
521 afs_kill_pages(mapping
, first
, last
);
522 mapping_set_error(mapping
, ret
);
526 _leave(" = %d", ret
);
531 * write a page back to the server
532 * - the caller locked the page for us
534 int afs_writepage(struct page
*page
, struct writeback_control
*wbc
)
538 _enter("{%lx},", page
->index
);
540 ret
= afs_write_back_from_locked_page(page
->mapping
, wbc
, page
,
541 wbc
->range_end
>> PAGE_SHIFT
);
543 _leave(" = %d", ret
);
547 wbc
->nr_to_write
-= ret
;
554 * write a region of pages back to the server
556 static int afs_writepages_region(struct address_space
*mapping
,
557 struct writeback_control
*wbc
,
558 pgoff_t index
, pgoff_t end
, pgoff_t
*_next
)
563 _enter(",,%lx,%lx,", index
, end
);
566 n
= find_get_pages_range_tag(mapping
, &index
, end
,
567 PAGECACHE_TAG_DIRTY
, 1, &page
);
571 _debug("wback %lx", page
->index
);
573 /* at this point we hold neither mapping->tree_lock nor lock on
574 * the page itself: the page may be truncated or invalidated
575 * (changing page->mapping to NULL), or even swizzled back from
576 * swapper_space to tmpfs file mapping
578 ret
= lock_page_killable(page
);
581 _leave(" = %d", ret
);
585 if (page
->mapping
!= mapping
|| !PageDirty(page
)) {
591 if (PageWriteback(page
)) {
593 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
594 wait_on_page_writeback(page
);
599 if (!clear_page_dirty_for_io(page
))
601 ret
= afs_write_back_from_locked_page(mapping
, wbc
, page
, end
);
604 _leave(" = %d", ret
);
608 wbc
->nr_to_write
-= ret
;
611 } while (index
< end
&& wbc
->nr_to_write
> 0);
614 _leave(" = 0 [%lx]", *_next
);
619 * write some of the pending data back to the server
621 int afs_writepages(struct address_space
*mapping
,
622 struct writeback_control
*wbc
)
624 pgoff_t start
, end
, next
;
629 if (wbc
->range_cyclic
) {
630 start
= mapping
->writeback_index
;
632 ret
= afs_writepages_region(mapping
, wbc
, start
, end
, &next
);
633 if (start
> 0 && wbc
->nr_to_write
> 0 && ret
== 0)
634 ret
= afs_writepages_region(mapping
, wbc
, 0, start
,
636 mapping
->writeback_index
= next
;
637 } else if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
) {
638 end
= (pgoff_t
)(LLONG_MAX
>> PAGE_SHIFT
);
639 ret
= afs_writepages_region(mapping
, wbc
, 0, end
, &next
);
640 if (wbc
->nr_to_write
> 0)
641 mapping
->writeback_index
= next
;
643 start
= wbc
->range_start
>> PAGE_SHIFT
;
644 end
= wbc
->range_end
>> PAGE_SHIFT
;
645 ret
= afs_writepages_region(mapping
, wbc
, start
, end
, &next
);
648 _leave(" = %d", ret
);
653 * completion of write to server
655 void afs_pages_written_back(struct afs_vnode
*vnode
, struct afs_call
*call
)
659 unsigned count
, loop
;
660 pgoff_t first
= call
->first
, last
= call
->last
;
662 _enter("{%x:%u},{%lx-%lx}",
663 vnode
->fid
.vid
, vnode
->fid
.vnode
, first
, last
);
668 _debug("done %lx-%lx", first
, last
);
670 count
= last
- first
+ 1;
671 if (count
> PAGEVEC_SIZE
)
672 count
= PAGEVEC_SIZE
;
673 pv
.nr
= find_get_pages_contig(vnode
->vfs_inode
.i_mapping
,
674 first
, count
, pv
.pages
);
675 ASSERTCMP(pv
.nr
, ==, count
);
677 for (loop
= 0; loop
< count
; loop
++) {
678 priv
= page_private(pv
.pages
[loop
]);
679 trace_afs_page_dirty(vnode
, tracepoint_string("clear"),
680 pv
.pages
[loop
]->index
, priv
);
681 set_page_private(pv
.pages
[loop
], 0);
682 end_page_writeback(pv
.pages
[loop
]);
685 __pagevec_release(&pv
);
686 } while (first
<= last
);
688 afs_prune_wb_keys(vnode
);
693 * write to an AFS file
695 ssize_t
afs_file_write(struct kiocb
*iocb
, struct iov_iter
*from
)
697 struct afs_vnode
*vnode
= AFS_FS_I(file_inode(iocb
->ki_filp
));
699 size_t count
= iov_iter_count(from
);
701 _enter("{%x.%u},{%zu},",
702 vnode
->fid
.vid
, vnode
->fid
.vnode
, count
);
704 if (IS_SWAPFILE(&vnode
->vfs_inode
)) {
706 "AFS: Attempt to write to active swap file!\n");
713 result
= generic_file_write_iter(iocb
, from
);
715 _leave(" = %zd", result
);
720 * flush any dirty pages for this process, and check for write errors.
721 * - the return status from this call provides a reliable indication of
722 * whether any write errors occurred for this process.
724 int afs_fsync(struct file
*file
, loff_t start
, loff_t end
, int datasync
)
726 struct inode
*inode
= file_inode(file
);
727 struct afs_vnode
*vnode
= AFS_FS_I(inode
);
729 _enter("{%x:%u},{n=%pD},%d",
730 vnode
->fid
.vid
, vnode
->fid
.vnode
, file
,
733 return file_write_and_wait_range(file
, start
, end
);
737 * Flush out all outstanding writes on a file opened for writing when it is
740 int afs_flush(struct file
*file
, fl_owner_t id
)
744 if ((file
->f_mode
& FMODE_WRITE
) == 0)
747 return vfs_fsync(file
, 0);
751 * notification that a previously read-only page is about to become writable
752 * - if it returns an error, the caller will deliver a bus error signal
754 int afs_page_mkwrite(struct vm_fault
*vmf
)
756 struct file
*file
= vmf
->vma
->vm_file
;
757 struct inode
*inode
= file_inode(file
);
758 struct afs_vnode
*vnode
= AFS_FS_I(inode
);
761 _enter("{{%x:%u}},{%lx}",
762 vnode
->fid
.vid
, vnode
->fid
.vnode
, vmf
->page
->index
);
764 sb_start_pagefault(inode
->i_sb
);
766 /* Wait for the page to be written to the cache before we allow it to
767 * be modified. We then assume the entire page will need writing back.
769 #ifdef CONFIG_AFS_FSCACHE
770 fscache_wait_on_page_write(vnode
->cache
, vmf
->page
);
773 if (PageWriteback(vmf
->page
) &&
774 wait_on_page_bit_killable(vmf
->page
, PG_writeback
) < 0)
775 return VM_FAULT_RETRY
;
777 if (lock_page_killable(vmf
->page
) < 0)
778 return VM_FAULT_RETRY
;
780 /* We mustn't change page->private until writeback is complete as that
781 * details the portion of the page we need to write back and we might
782 * need to redirty the page if there's a problem.
784 wait_on_page_writeback(vmf
->page
);
786 priv
= (unsigned long)PAGE_SIZE
<< AFS_PRIV_SHIFT
; /* To */
787 priv
|= 0; /* From */
788 trace_afs_page_dirty(vnode
, tracepoint_string("mkwrite"),
789 vmf
->page
->index
, priv
);
790 SetPagePrivate(vmf
->page
);
791 set_page_private(vmf
->page
, priv
);
793 sb_end_pagefault(inode
->i_sb
);
794 return VM_FAULT_LOCKED
;
798 * Prune the keys cached for writeback. The caller must hold vnode->wb_lock.
800 void afs_prune_wb_keys(struct afs_vnode
*vnode
)
802 LIST_HEAD(graveyard
);
803 struct afs_wb_key
*wbk
, *tmp
;
805 /* Discard unused keys */
806 spin_lock(&vnode
->wb_lock
);
808 if (!mapping_tagged(&vnode
->vfs_inode
.i_data
, PAGECACHE_TAG_WRITEBACK
) &&
809 !mapping_tagged(&vnode
->vfs_inode
.i_data
, PAGECACHE_TAG_DIRTY
)) {
810 list_for_each_entry_safe(wbk
, tmp
, &vnode
->wb_keys
, vnode_link
) {
811 if (refcount_read(&wbk
->usage
) == 1)
812 list_move(&wbk
->vnode_link
, &graveyard
);
816 spin_unlock(&vnode
->wb_lock
);
818 while (!list_empty(&graveyard
)) {
819 wbk
= list_entry(graveyard
.next
, struct afs_wb_key
, vnode_link
);
820 list_del(&wbk
->vnode_link
);
826 * Clean up a page during invalidation.
828 int afs_launder_page(struct page
*page
)
830 struct address_space
*mapping
= page
->mapping
;
831 struct afs_vnode
*vnode
= AFS_FS_I(mapping
->host
);
836 _enter("{%lx}", page
->index
);
838 priv
= page_private(page
);
839 if (clear_page_dirty_for_io(page
)) {
842 if (PagePrivate(page
)) {
843 f
= priv
& AFS_PRIV_MAX
;
844 t
= priv
>> AFS_PRIV_SHIFT
;
847 trace_afs_page_dirty(vnode
, tracepoint_string("launder"),
849 ret
= afs_store_data(mapping
, page
->index
, page
->index
, t
, f
);
852 trace_afs_page_dirty(vnode
, tracepoint_string("laundered"),
854 set_page_private(page
, 0);
855 ClearPagePrivate(page
);
857 #ifdef CONFIG_AFS_FSCACHE
858 if (PageFsCache(page
)) {
859 fscache_wait_on_page_write(vnode
->cache
, page
);
860 fscache_uncache_page(vnode
->cache
, page
);