1 #include "ceph_debug.h"
3 #include <linux/backing-dev.h>
6 #include <linux/pagemap.h>
7 #include <linux/writeback.h> /* generic_writepages */
8 #include <linux/slab.h>
9 #include <linux/pagevec.h>
10 #include <linux/task_io_accounting_ops.h>
13 #include "osd_client.h"
16 * Ceph address space ops.
18 * There are a few funny things going on here.
20 * The page->private field is used to reference a struct
21 * ceph_snap_context for _every_ dirty page. This indicates which
22 * snapshot the page was logically dirtied in, and thus which snap
23 * context needs to be associated with the osd write during writeback.
25 * Similarly, struct ceph_inode_info maintains a set of counters to
26 * count dirty pages on the inode. In the absense of snapshots,
27 * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count.
29 * When a snapshot is taken (that is, when the client receives
30 * notification that a snapshot was taken), each inode with caps and
31 * with dirty pages (dirty pages implies there is a cap) gets a new
32 * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending
33 * order, new snaps go to the tail). The i_wrbuffer_ref_head count is
34 * moved to capsnap->dirty. (Unless a sync write is currently in
35 * progress. In that case, the capsnap is said to be "pending", new
36 * writes cannot start, and the capsnap isn't "finalized" until the
37 * write completes (or fails) and a final size/mtime for the inode for
38 * that snap can be settled upon.) i_wrbuffer_ref_head is reset to 0.
40 * On writeback, we must submit writes to the osd IN SNAP ORDER. So,
41 * we look for the first capsnap in i_cap_snaps and write out pages in
42 * that snap context _only_. Then we move on to the next capsnap,
43 * eventually reaching the "live" or "head" context (i.e., pages that
44 * are not yet snapped) and are writing the most recently dirtied
47 * Invalidate and so forth must take care to ensure the dirty page
48 * accounting is preserved.
51 #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10))
52 #define CONGESTION_OFF_THRESH(congestion_kb) \
53 (CONGESTION_ON_THRESH(congestion_kb) - \
54 (CONGESTION_ON_THRESH(congestion_kb) >> 2))
59 * Dirty a page. Optimistically adjust accounting, on the assumption
60 * that we won't race with invalidate. If we do, readjust.
62 static int ceph_set_page_dirty(struct page
*page
)
64 struct address_space
*mapping
= page
->mapping
;
66 struct ceph_inode_info
*ci
;
68 struct ceph_snap_context
*snapc
;
70 if (unlikely(!mapping
))
71 return !TestSetPageDirty(page
);
73 if (TestSetPageDirty(page
)) {
74 dout("%p set_page_dirty %p idx %lu -- already dirty\n",
75 mapping
->host
, page
, page
->index
);
79 inode
= mapping
->host
;
80 ci
= ceph_inode(inode
);
83 * Note that we're grabbing a snapc ref here without holding
86 snapc
= ceph_get_snap_context(ci
->i_snap_realm
->cached_context
);
89 spin_lock(&inode
->i_lock
);
90 if (ci
->i_head_snapc
== NULL
)
91 ci
->i_head_snapc
= ceph_get_snap_context(snapc
);
92 ++ci
->i_wrbuffer_ref_head
;
93 if (ci
->i_wrbuffer_ref
== 0)
96 dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d "
97 "snapc %p seq %lld (%d snaps)\n",
98 mapping
->host
, page
, page
->index
,
99 ci
->i_wrbuffer_ref
-1, ci
->i_wrbuffer_ref_head
-1,
100 ci
->i_wrbuffer_ref
, ci
->i_wrbuffer_ref_head
,
101 snapc
, snapc
->seq
, snapc
->num_snaps
);
102 spin_unlock(&inode
->i_lock
);
104 /* now adjust page */
105 spin_lock_irq(&mapping
->tree_lock
);
106 if (page
->mapping
) { /* Race with truncate? */
107 WARN_ON_ONCE(!PageUptodate(page
));
108 account_page_dirtied(page
, page
->mapping
);
109 radix_tree_tag_set(&mapping
->page_tree
,
110 page_index(page
), PAGECACHE_TAG_DIRTY
);
113 * Reference snap context in page->private. Also set
114 * PagePrivate so that we get invalidatepage callback.
116 page
->private = (unsigned long)snapc
;
117 SetPagePrivate(page
);
119 dout("ANON set_page_dirty %p (raced truncate?)\n", page
);
123 spin_unlock_irq(&mapping
->tree_lock
);
126 /* whoops, we failed to dirty the page */
127 ceph_put_wrbuffer_cap_refs(ci
, 1, snapc
);
129 __mark_inode_dirty(mapping
->host
, I_DIRTY_PAGES
);
131 BUG_ON(!PageDirty(page
));
136 * If we are truncating the full page (i.e. offset == 0), adjust the
137 * dirty page counters appropriately. Only called if there is private
140 static void ceph_invalidatepage(struct page
*page
, unsigned long offset
)
143 struct ceph_inode_info
*ci
;
144 struct ceph_snap_context
*snapc
= (void *)page
->private;
146 BUG_ON(!PageLocked(page
));
147 BUG_ON(!page
->private);
148 BUG_ON(!PagePrivate(page
));
149 BUG_ON(!page
->mapping
);
151 inode
= page
->mapping
->host
;
154 * We can get non-dirty pages here due to races between
155 * set_page_dirty and truncate_complete_page; just spit out a
156 * warning, in case we end up with accounting problems later.
158 if (!PageDirty(page
))
159 pr_err("%p invalidatepage %p page not dirty\n", inode
, page
);
162 ClearPageChecked(page
);
164 ci
= ceph_inode(inode
);
166 dout("%p invalidatepage %p idx %lu full dirty page %lu\n",
167 inode
, page
, page
->index
, offset
);
168 ceph_put_wrbuffer_cap_refs(ci
, 1, snapc
);
169 ceph_put_snap_context(snapc
);
171 ClearPagePrivate(page
);
173 dout("%p invalidatepage %p idx %lu partial dirty page\n",
174 inode
, page
, page
->index
);
178 /* just a sanity check */
179 static int ceph_releasepage(struct page
*page
, gfp_t g
)
181 struct inode
*inode
= page
->mapping
? page
->mapping
->host
: NULL
;
182 dout("%p releasepage %p idx %lu\n", inode
, page
, page
->index
);
183 WARN_ON(PageDirty(page
));
184 WARN_ON(page
->private);
185 WARN_ON(PagePrivate(page
));
190 * read a single page, without unlocking it.
192 static int readpage_nounlock(struct file
*filp
, struct page
*page
)
194 struct inode
*inode
= filp
->f_dentry
->d_inode
;
195 struct ceph_inode_info
*ci
= ceph_inode(inode
);
196 struct ceph_osd_client
*osdc
= &ceph_inode_to_client(inode
)->osdc
;
198 u64 len
= PAGE_CACHE_SIZE
;
200 dout("readpage inode %p file %p page %p index %lu\n",
201 inode
, filp
, page
, page
->index
);
202 err
= ceph_osdc_readpages(osdc
, ceph_vino(inode
), &ci
->i_layout
,
203 page
->index
<< PAGE_CACHE_SHIFT
, &len
,
204 ci
->i_truncate_seq
, ci
->i_truncate_size
,
211 } else if (err
< PAGE_CACHE_SIZE
) {
212 /* zero fill remainder of page */
213 zero_user_segment(page
, err
, PAGE_CACHE_SIZE
);
215 SetPageUptodate(page
);
218 return err
< 0 ? err
: 0;
221 static int ceph_readpage(struct file
*filp
, struct page
*page
)
223 int r
= readpage_nounlock(filp
, page
);
229 * Build a vector of contiguous pages from the provided page list.
231 static struct page
**page_vector_from_list(struct list_head
*page_list
,
236 int next_index
, contig_pages
= 0;
238 /* build page vector */
239 pages
= kmalloc(sizeof(*pages
) * *nr_pages
, GFP_NOFS
);
241 return ERR_PTR(-ENOMEM
);
243 BUG_ON(list_empty(page_list
));
244 next_index
= list_entry(page_list
->prev
, struct page
, lru
)->index
;
245 list_for_each_entry_reverse(page
, page_list
, lru
) {
246 if (page
->index
== next_index
) {
247 dout("readpages page %d %p\n", contig_pages
, page
);
248 pages
[contig_pages
] = page
;
255 *nr_pages
= contig_pages
;
260 * Read multiple pages. Leave pages we don't read + unlock in page_list;
261 * the caller (VM) cleans them up.
263 static int ceph_readpages(struct file
*file
, struct address_space
*mapping
,
264 struct list_head
*page_list
, unsigned nr_pages
)
266 struct inode
*inode
= file
->f_dentry
->d_inode
;
267 struct ceph_inode_info
*ci
= ceph_inode(inode
);
268 struct ceph_osd_client
*osdc
= &ceph_inode_to_client(inode
)->osdc
;
274 dout("readpages %p file %p nr_pages %d\n",
275 inode
, file
, nr_pages
);
277 pages
= page_vector_from_list(page_list
, &nr_pages
);
279 return PTR_ERR(pages
);
281 /* guess read extent */
282 offset
= pages
[0]->index
<< PAGE_CACHE_SHIFT
;
283 len
= nr_pages
<< PAGE_CACHE_SHIFT
;
284 rc
= ceph_osdc_readpages(osdc
, ceph_vino(inode
), &ci
->i_layout
,
286 ci
->i_truncate_seq
, ci
->i_truncate_size
,
293 for (; !list_empty(page_list
) && len
> 0;
294 rc
-= PAGE_CACHE_SIZE
, len
-= PAGE_CACHE_SIZE
) {
296 list_entry(page_list
->prev
, struct page
, lru
);
298 list_del(&page
->lru
);
300 if (rc
< (int)PAGE_CACHE_SIZE
) {
301 /* zero (remainder of) page */
302 int s
= rc
< 0 ? 0 : rc
;
303 zero_user_segment(page
, s
, PAGE_CACHE_SIZE
);
306 if (add_to_page_cache_lru(page
, mapping
, page
->index
,
308 page_cache_release(page
);
309 dout("readpages %p add_to_page_cache failed %p\n",
313 dout("readpages %p adding %p idx %lu\n", inode
, page
,
315 flush_dcache_page(page
);
316 SetPageUptodate(page
);
318 page_cache_release(page
);
328 * Get ref for the oldest snapc for an inode with dirty data... that is, the
329 * only snap context we are allowed to write back.
331 static struct ceph_snap_context
*get_oldest_context(struct inode
*inode
,
334 struct ceph_inode_info
*ci
= ceph_inode(inode
);
335 struct ceph_snap_context
*snapc
= NULL
;
336 struct ceph_cap_snap
*capsnap
= NULL
;
338 spin_lock(&inode
->i_lock
);
339 list_for_each_entry(capsnap
, &ci
->i_cap_snaps
, ci_item
) {
340 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap
,
341 capsnap
->context
, capsnap
->dirty_pages
);
342 if (capsnap
->dirty_pages
) {
343 snapc
= ceph_get_snap_context(capsnap
->context
);
345 *snap_size
= capsnap
->size
;
349 if (!snapc
&& ci
->i_wrbuffer_ref_head
) {
350 snapc
= ceph_get_snap_context(ci
->i_head_snapc
);
351 dout(" head snapc %p has %d dirty pages\n",
352 snapc
, ci
->i_wrbuffer_ref_head
);
354 spin_unlock(&inode
->i_lock
);
359 * Write a single page, but leave the page locked.
361 * If we get a write error, set the page error bit, but still adjust the
362 * dirty page accounting (i.e., page is no longer dirty).
364 static int writepage_nounlock(struct page
*page
, struct writeback_control
*wbc
)
367 struct ceph_inode_info
*ci
;
368 struct ceph_client
*client
;
369 struct ceph_osd_client
*osdc
;
370 loff_t page_off
= page
->index
<< PAGE_CACHE_SHIFT
;
371 int len
= PAGE_CACHE_SIZE
;
374 struct ceph_snap_context
*snapc
, *oldest
;
378 dout("writepage %p idx %lu\n", page
, page
->index
);
380 if (!page
->mapping
|| !page
->mapping
->host
) {
381 dout("writepage %p - no mapping\n", page
);
384 inode
= page
->mapping
->host
;
385 ci
= ceph_inode(inode
);
386 client
= ceph_inode_to_client(inode
);
387 osdc
= &client
->osdc
;
389 /* verify this is a writeable snap context */
390 snapc
= (void *)page
->private;
392 dout("writepage %p page %p not dirty?\n", inode
, page
);
395 oldest
= get_oldest_context(inode
, &snap_size
);
396 if (snapc
->seq
> oldest
->seq
) {
397 dout("writepage %p page %p snapc %p not writeable - noop\n",
398 inode
, page
, (void *)page
->private);
399 /* we should only noop if called by kswapd */
400 WARN_ON((current
->flags
& PF_MEMALLOC
) == 0);
401 ceph_put_snap_context(oldest
);
404 ceph_put_snap_context(oldest
);
406 /* is this a partial page at end of file? */
410 i_size
= i_size_read(inode
);
411 if (i_size
< page_off
+ len
)
412 len
= i_size
- page_off
;
414 dout("writepage %p page %p index %lu on %llu~%u\n",
415 inode
, page
, page
->index
, page_off
, len
);
417 writeback_stat
= atomic_long_inc_return(&client
->writeback_count
);
419 CONGESTION_ON_THRESH(client
->mount_args
->congestion_kb
))
420 set_bdi_congested(&client
->backing_dev_info
, BLK_RW_ASYNC
);
422 set_page_writeback(page
);
423 err
= ceph_osdc_writepages(osdc
, ceph_vino(inode
),
424 &ci
->i_layout
, snapc
,
426 ci
->i_truncate_seq
, ci
->i_truncate_size
,
428 &page
, 1, 0, 0, true);
430 dout("writepage setting page/mapping error %d %p\n", err
, page
);
432 mapping_set_error(&inode
->i_data
, err
);
434 wbc
->pages_skipped
++;
436 dout("writepage cleaned page %p\n", page
);
437 err
= 0; /* vfs expects us to return 0 */
440 ClearPagePrivate(page
);
441 end_page_writeback(page
);
442 ceph_put_wrbuffer_cap_refs(ci
, 1, snapc
);
443 ceph_put_snap_context(snapc
); /* page's reference */
448 static int ceph_writepage(struct page
*page
, struct writeback_control
*wbc
)
451 struct inode
*inode
= page
->mapping
->host
;
454 err
= writepage_nounlock(page
, wbc
);
462 * lame release_pages helper. release_pages() isn't exported to
465 static void ceph_release_pages(struct page
**pages
, int num
)
470 pagevec_init(&pvec
, 0);
471 for (i
= 0; i
< num
; i
++) {
472 if (pagevec_add(&pvec
, pages
[i
]) == 0)
473 pagevec_release(&pvec
);
475 pagevec_release(&pvec
);
480 * async writeback completion handler.
482 * If we get an error, set the mapping error bit, but not the individual
485 static void writepages_finish(struct ceph_osd_request
*req
,
486 struct ceph_msg
*msg
)
488 struct inode
*inode
= req
->r_inode
;
489 struct ceph_osd_reply_head
*replyhead
;
490 struct ceph_osd_op
*op
;
491 struct ceph_inode_info
*ci
= ceph_inode(inode
);
495 struct ceph_snap_context
*snapc
= req
->r_snapc
;
496 struct address_space
*mapping
= inode
->i_mapping
;
499 struct ceph_client
*client
= ceph_inode_to_client(inode
);
501 unsigned issued
= ceph_caps_issued(ci
);
504 replyhead
= msg
->front
.iov_base
;
505 WARN_ON(le32_to_cpu(replyhead
->num_ops
) == 0);
506 op
= (void *)(replyhead
+ 1);
507 rc
= le32_to_cpu(replyhead
->result
);
508 bytes
= le64_to_cpu(op
->extent
.length
);
512 * Assume we wrote the pages we originally sent. The
513 * osd might reply with fewer pages if our writeback
514 * raced with a truncation and was adjusted at the osd,
515 * so don't believe the reply.
517 wrote
= req
->r_num_pages
;
520 mapping_set_error(mapping
, rc
);
522 dout("writepages_finish %p rc %d bytes %llu wrote %d (pages)\n",
523 inode
, rc
, bytes
, wrote
);
525 /* clean all pages */
526 for (i
= 0; i
< req
->r_num_pages
; i
++) {
527 page
= req
->r_pages
[i
];
529 WARN_ON(!PageUptodate(page
));
532 atomic_long_dec_return(&client
->writeback_count
);
534 CONGESTION_OFF_THRESH(client
->mount_args
->congestion_kb
))
535 clear_bdi_congested(&client
->backing_dev_info
,
538 ceph_put_snap_context((void *)page
->private);
540 ClearPagePrivate(page
);
541 dout("unlocking %d %p\n", i
, page
);
542 end_page_writeback(page
);
545 * We lost the cache cap, need to truncate the page before
546 * it is unlocked, otherwise we'd truncate it later in the
547 * page truncation thread, possibly losing some data that
550 if ((issued
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)) == 0)
551 generic_error_remove_page(inode
->i_mapping
, page
);
555 dout("%p wrote+cleaned %d pages\n", inode
, wrote
);
556 ceph_put_wrbuffer_cap_refs(ci
, req
->r_num_pages
, snapc
);
558 ceph_release_pages(req
->r_pages
, req
->r_num_pages
);
559 if (req
->r_pages_from_pool
)
560 mempool_free(req
->r_pages
,
561 ceph_sb_to_client(inode
->i_sb
)->wb_pagevec_pool
);
564 ceph_osdc_put_request(req
);
568 * allocate a page vec, either directly, or if necessary, via a the
569 * mempool. we avoid the mempool if we can because req->r_num_pages
570 * may be less than the maximum write size.
572 static void alloc_page_vec(struct ceph_client
*client
,
573 struct ceph_osd_request
*req
)
575 req
->r_pages
= kmalloc(sizeof(struct page
*) * req
->r_num_pages
,
578 req
->r_pages
= mempool_alloc(client
->wb_pagevec_pool
, GFP_NOFS
);
579 req
->r_pages_from_pool
= 1;
580 WARN_ON(!req
->r_pages
);
585 * initiate async writeback
587 static int ceph_writepages_start(struct address_space
*mapping
,
588 struct writeback_control
*wbc
)
590 struct inode
*inode
= mapping
->host
;
591 struct backing_dev_info
*bdi
= mapping
->backing_dev_info
;
592 struct ceph_inode_info
*ci
= ceph_inode(inode
);
593 struct ceph_client
*client
;
594 pgoff_t index
, start
, end
;
597 pgoff_t max_pages
= 0, max_pages_ever
= 0;
598 struct ceph_snap_context
*snapc
= NULL
, *last_snapc
= NULL
, *pgsnapc
;
602 unsigned wsize
= 1 << inode
->i_blkbits
;
603 struct ceph_osd_request
*req
= NULL
;
608 * Include a 'sync' in the OSD request if this is a data
609 * integrity write (e.g., O_SYNC write or fsync()), or if our
610 * cap is being revoked.
612 do_sync
= wbc
->sync_mode
== WB_SYNC_ALL
;
613 if (ceph_caps_revoking(ci
, CEPH_CAP_FILE_BUFFER
))
615 dout("writepages_start %p dosync=%d (mode=%s)\n",
617 wbc
->sync_mode
== WB_SYNC_NONE
? "NONE" :
618 (wbc
->sync_mode
== WB_SYNC_ALL
? "ALL" : "HOLD"));
620 client
= ceph_inode_to_client(inode
);
621 if (client
->mount_state
== CEPH_MOUNT_SHUTDOWN
) {
622 pr_warning("writepage_start %p on forced umount\n", inode
);
623 return -EIO
; /* we're in a forced umount, don't write! */
625 if (client
->mount_args
->wsize
&& client
->mount_args
->wsize
< wsize
)
626 wsize
= client
->mount_args
->wsize
;
627 if (wsize
< PAGE_CACHE_SIZE
)
628 wsize
= PAGE_CACHE_SIZE
;
629 max_pages_ever
= wsize
>> PAGE_CACHE_SHIFT
;
631 pagevec_init(&pvec
, 0);
634 if (wbc
->nonblocking
&& bdi_write_congested(bdi
)) {
635 dout(" writepages congested\n");
636 wbc
->encountered_congestion
= 1;
640 /* where to start/end? */
641 if (wbc
->range_cyclic
) {
642 start
= mapping
->writeback_index
; /* Start from prev offset */
644 dout(" cyclic, start at %lu\n", start
);
646 start
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
647 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
648 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
651 dout(" not cyclic, %lu to %lu\n", start
, end
);
656 /* find oldest snap context with dirty data */
657 ceph_put_snap_context(snapc
);
658 snapc
= get_oldest_context(inode
, &snap_size
);
660 /* hmm, why does writepages get called when there
662 dout(" no snap context with dirty data?\n");
665 dout(" oldest snapc is %p seq %lld (%d snaps)\n",
666 snapc
, snapc
->seq
, snapc
->num_snaps
);
667 if (last_snapc
&& snapc
!= last_snapc
) {
668 /* if we switched to a newer snapc, restart our scan at the
669 * start of the original file range. */
670 dout(" snapc differs from last pass, restarting at %lu\n",
676 while (!done
&& index
<= end
) {
680 int pvec_pages
, locked_pages
;
684 struct ceph_osd_request_head
*reqhead
;
685 struct ceph_osd_op
*op
;
690 max_pages
= max_pages_ever
;
694 want
= min(end
- index
,
695 min((pgoff_t
)PAGEVEC_SIZE
,
696 max_pages
- (pgoff_t
)locked_pages
) - 1)
698 pvec_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
701 dout("pagevec_lookup_tag got %d\n", pvec_pages
);
702 if (!pvec_pages
&& !locked_pages
)
704 for (i
= 0; i
< pvec_pages
&& locked_pages
< max_pages
; i
++) {
705 page
= pvec
.pages
[i
];
706 dout("? %p idx %lu\n", page
, page
->index
);
707 if (locked_pages
== 0)
708 lock_page(page
); /* first page */
709 else if (!trylock_page(page
))
712 /* only dirty pages, or our accounting breaks */
713 if (unlikely(!PageDirty(page
)) ||
714 unlikely(page
->mapping
!= mapping
)) {
715 dout("!dirty or !mapping %p\n", page
);
719 if (!wbc
->range_cyclic
&& page
->index
> end
) {
720 dout("end of range %p\n", page
);
725 if (next
&& (page
->index
!= next
)) {
726 dout("not consecutive %p\n", page
);
730 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
731 dout("waiting on writeback %p\n", page
);
732 wait_on_page_writeback(page
);
734 if ((snap_size
&& page_offset(page
) > snap_size
) ||
736 page_offset(page
) > i_size_read(inode
))) {
737 dout("%p page eof %llu\n", page
, snap_size
?
738 snap_size
: i_size_read(inode
));
743 if (PageWriteback(page
)) {
744 dout("%p under writeback\n", page
);
749 /* only if matching snap context */
750 pgsnapc
= (void *)page
->private;
751 if (pgsnapc
->seq
> snapc
->seq
) {
752 dout("page snapc %p %lld > oldest %p %lld\n",
753 pgsnapc
, pgsnapc
->seq
, snapc
, snapc
->seq
);
756 continue; /* keep looking for snap */
760 if (!clear_page_dirty_for_io(page
)) {
761 dout("%p !clear_page_dirty_for_io\n", page
);
767 if (locked_pages
== 0) {
768 /* prepare async write request */
769 offset
= page
->index
<< PAGE_CACHE_SHIFT
;
771 req
= ceph_osdc_new_request(&client
->osdc
,
776 CEPH_OSD_FLAG_WRITE
|
777 CEPH_OSD_FLAG_ONDISK
,
781 &inode
->i_mtime
, true, 1);
782 max_pages
= req
->r_num_pages
;
784 alloc_page_vec(client
, req
);
785 req
->r_callback
= writepages_finish
;
786 req
->r_inode
= inode
;
789 /* note position of first page in pvec */
792 dout("%p will write page %p idx %lu\n",
793 inode
, page
, page
->index
);
796 atomic_long_inc_return(&client
->writeback_count
);
797 if (writeback_stat
> CONGESTION_ON_THRESH(
798 client
->mount_args
->congestion_kb
)) {
799 set_bdi_congested(&client
->backing_dev_info
,
803 set_page_writeback(page
);
804 req
->r_pages
[locked_pages
] = page
;
806 next
= page
->index
+ 1;
809 /* did we get anything? */
811 goto release_pvec_pages
;
814 BUG_ON(!locked_pages
|| first
< 0);
816 if (pvec_pages
&& i
== pvec_pages
&&
817 locked_pages
< max_pages
) {
818 dout("reached end pvec, trying for more\n");
819 pagevec_reinit(&pvec
);
823 /* shift unused pages over in the pvec... we
824 * will need to release them below. */
825 for (j
= i
; j
< pvec_pages
; j
++) {
826 dout(" pvec leftover page %p\n",
828 pvec
.pages
[j
-i
+first
] = pvec
.pages
[j
];
833 /* submit the write */
834 offset
= req
->r_pages
[0]->index
<< PAGE_CACHE_SHIFT
;
835 len
= min((snap_size
? snap_size
: i_size_read(inode
)) - offset
,
836 (u64
)locked_pages
<< PAGE_CACHE_SHIFT
);
837 dout("writepages got %d pages at %llu~%llu\n",
838 locked_pages
, offset
, len
);
840 /* revise final length, page count */
841 req
->r_num_pages
= locked_pages
;
842 reqhead
= req
->r_request
->front
.iov_base
;
843 op
= (void *)(reqhead
+ 1);
844 op
->extent
.length
= cpu_to_le64(len
);
845 op
->payload_len
= cpu_to_le32(len
);
846 req
->r_request
->hdr
.data_len
= cpu_to_le32(len
);
848 ceph_osdc_start_request(&client
->osdc
, req
, true);
853 wbc
->nr_to_write
-= locked_pages
;
854 if (wbc
->nr_to_write
<= 0)
858 dout("pagevec_release on %d pages (%p)\n", (int)pvec
.nr
,
859 pvec
.nr
? pvec
.pages
[0] : NULL
);
860 pagevec_release(&pvec
);
862 if (locked_pages
&& !done
)
866 if (should_loop
&& !done
) {
867 /* more to do; loop back to beginning of file */
868 dout("writepages looping back to beginning of file\n");
874 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
875 mapping
->writeback_index
= index
;
879 ceph_osdc_put_request(req
);
881 rc
= 0; /* vfs expects us to return 0 */
882 ceph_put_snap_context(snapc
);
883 dout("writepages done, rc = %d\n", rc
);
891 * See if a given @snapc is either writeable, or already written.
893 static int context_is_writeable_or_written(struct inode
*inode
,
894 struct ceph_snap_context
*snapc
)
896 struct ceph_snap_context
*oldest
= get_oldest_context(inode
, NULL
);
897 int ret
= !oldest
|| snapc
->seq
<= oldest
->seq
;
899 ceph_put_snap_context(oldest
);
904 * We are only allowed to write into/dirty the page if the page is
905 * clean, or already dirty within the same snap context.
907 * called with page locked.
908 * return success with page locked,
909 * or any failure (incl -EAGAIN) with page unlocked.
911 static int ceph_update_writeable_page(struct file
*file
,
912 loff_t pos
, unsigned len
,
915 struct inode
*inode
= file
->f_dentry
->d_inode
;
916 struct ceph_inode_info
*ci
= ceph_inode(inode
);
917 struct ceph_mds_client
*mdsc
= &ceph_inode_to_client(inode
)->mdsc
;
918 loff_t page_off
= pos
& PAGE_CACHE_MASK
;
919 int pos_in_page
= pos
& ~PAGE_CACHE_MASK
;
920 int end_in_page
= pos_in_page
+ len
;
923 struct ceph_snap_context
*snapc
, *oldest
;
926 /* writepages currently holds page lock, but if we change that later, */
927 wait_on_page_writeback(page
);
929 /* check snap context */
930 BUG_ON(!ci
->i_snap_realm
);
931 down_read(&mdsc
->snap_rwsem
);
932 BUG_ON(!ci
->i_snap_realm
->cached_context
);
933 snapc
= (void *)page
->private;
934 if (snapc
&& snapc
!= ci
->i_head_snapc
) {
936 * this page is already dirty in another (older) snap
937 * context! is it writeable now?
939 oldest
= get_oldest_context(inode
, NULL
);
940 up_read(&mdsc
->snap_rwsem
);
942 if (snapc
->seq
> oldest
->seq
) {
943 ceph_put_snap_context(oldest
);
944 dout(" page %p snapc %p not current or oldest\n",
947 * queue for writeback, and wait for snapc to
948 * be writeable or written
950 snapc
= ceph_get_snap_context(snapc
);
952 ceph_queue_writeback(inode
);
953 r
= wait_event_interruptible(ci
->i_cap_wq
,
954 context_is_writeable_or_written(inode
, snapc
));
955 ceph_put_snap_context(snapc
);
956 if (r
== -ERESTARTSYS
)
960 ceph_put_snap_context(oldest
);
962 /* yay, writeable, do it now (without dropping page lock) */
963 dout(" page %p snapc %p not current, but oldest\n",
965 if (!clear_page_dirty_for_io(page
))
967 r
= writepage_nounlock(page
, NULL
);
973 if (PageUptodate(page
)) {
974 dout(" page %p already uptodate\n", page
);
979 if (pos_in_page
== 0 && len
== PAGE_CACHE_SIZE
)
982 /* past end of file? */
983 i_size
= inode
->i_size
; /* caller holds i_mutex */
985 if (i_size
+ len
> inode
->i_sb
->s_maxbytes
) {
986 /* file is too big */
991 if (page_off
>= i_size
||
992 (pos_in_page
== 0 && (pos
+len
) >= i_size
&&
993 end_in_page
- pos_in_page
!= PAGE_CACHE_SIZE
)) {
994 dout(" zeroing %p 0 - %d and %d - %d\n",
995 page
, pos_in_page
, end_in_page
, (int)PAGE_CACHE_SIZE
);
996 zero_user_segments(page
,
998 end_in_page
, PAGE_CACHE_SIZE
);
1002 /* we need to read it. */
1003 up_read(&mdsc
->snap_rwsem
);
1004 r
= readpage_nounlock(file
, page
);
1010 up_read(&mdsc
->snap_rwsem
);
1017 * We are only allowed to write into/dirty the page if the page is
1018 * clean, or already dirty within the same snap context.
1020 static int ceph_write_begin(struct file
*file
, struct address_space
*mapping
,
1021 loff_t pos
, unsigned len
, unsigned flags
,
1022 struct page
**pagep
, void **fsdata
)
1024 struct inode
*inode
= file
->f_dentry
->d_inode
;
1026 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
1031 page
= grab_cache_page_write_begin(mapping
, index
, 0);
1036 dout("write_begin file %p inode %p page %p %d~%d\n", file
,
1037 inode
, page
, (int)pos
, (int)len
);
1039 r
= ceph_update_writeable_page(file
, pos
, len
, page
);
1040 } while (r
== -EAGAIN
);
1046 * we don't do anything in here that simple_write_end doesn't do
1047 * except adjust dirty page accounting and drop read lock on
1050 static int ceph_write_end(struct file
*file
, struct address_space
*mapping
,
1051 loff_t pos
, unsigned len
, unsigned copied
,
1052 struct page
*page
, void *fsdata
)
1054 struct inode
*inode
= file
->f_dentry
->d_inode
;
1055 struct ceph_client
*client
= ceph_inode_to_client(inode
);
1056 struct ceph_mds_client
*mdsc
= &client
->mdsc
;
1057 unsigned from
= pos
& (PAGE_CACHE_SIZE
- 1);
1060 dout("write_end file %p inode %p page %p %d~%d (%d)\n", file
,
1061 inode
, page
, (int)pos
, (int)copied
, (int)len
);
1063 /* zero the stale part of the page if we did a short copy */
1065 zero_user_segment(page
, from
+copied
, len
);
1067 /* did file size increase? */
1068 /* (no need for i_size_read(); we caller holds i_mutex */
1069 if (pos
+copied
> inode
->i_size
)
1070 check_cap
= ceph_inode_set_size(inode
, pos
+copied
);
1072 if (!PageUptodate(page
))
1073 SetPageUptodate(page
);
1075 set_page_dirty(page
);
1078 up_read(&mdsc
->snap_rwsem
);
1079 page_cache_release(page
);
1082 ceph_check_caps(ceph_inode(inode
), CHECK_CAPS_AUTHONLY
, NULL
);
1088 * we set .direct_IO to indicate direct io is supported, but since we
1089 * intercept O_DIRECT reads and writes early, this function should
1092 static ssize_t
ceph_direct_io(int rw
, struct kiocb
*iocb
,
1093 const struct iovec
*iov
,
1094 loff_t pos
, unsigned long nr_segs
)
1100 const struct address_space_operations ceph_aops
= {
1101 .readpage
= ceph_readpage
,
1102 .readpages
= ceph_readpages
,
1103 .writepage
= ceph_writepage
,
1104 .writepages
= ceph_writepages_start
,
1105 .write_begin
= ceph_write_begin
,
1106 .write_end
= ceph_write_end
,
1107 .set_page_dirty
= ceph_set_page_dirty
,
1108 .invalidatepage
= ceph_invalidatepage
,
1109 .releasepage
= ceph_releasepage
,
1110 .direct_IO
= ceph_direct_io
,
1119 * Reuse write_begin here for simplicity.
1121 static int ceph_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1123 struct inode
*inode
= vma
->vm_file
->f_dentry
->d_inode
;
1124 struct page
*page
= vmf
->page
;
1125 struct ceph_mds_client
*mdsc
= &ceph_inode_to_client(inode
)->mdsc
;
1126 loff_t off
= page
->index
<< PAGE_CACHE_SHIFT
;
1130 size
= i_size_read(inode
);
1131 if (off
+ PAGE_CACHE_SIZE
<= size
)
1132 len
= PAGE_CACHE_SIZE
;
1134 len
= size
& ~PAGE_CACHE_MASK
;
1136 dout("page_mkwrite %p %llu~%llu page %p idx %lu\n", inode
,
1137 off
, len
, page
, page
->index
);
1141 ret
= VM_FAULT_NOPAGE
;
1143 (page
->mapping
!= inode
->i_mapping
))
1146 ret
= ceph_update_writeable_page(vma
->vm_file
, off
, len
, page
);
1148 /* success. we'll keep the page locked. */
1149 set_page_dirty(page
);
1150 up_read(&mdsc
->snap_rwsem
);
1151 ret
= VM_FAULT_LOCKED
;
1156 ret
= VM_FAULT_SIGBUS
;
1159 dout("page_mkwrite %p %llu~%llu = %d\n", inode
, off
, len
, ret
);
1160 if (ret
!= VM_FAULT_LOCKED
)
1165 static struct vm_operations_struct ceph_vmops
= {
1166 .fault
= filemap_fault
,
1167 .page_mkwrite
= ceph_page_mkwrite
,
1170 int ceph_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1172 struct address_space
*mapping
= file
->f_mapping
;
1174 if (!mapping
->a_ops
->readpage
)
1176 file_accessed(file
);
1177 vma
->vm_ops
= &ceph_vmops
;
1178 vma
->vm_flags
|= VM_CAN_NONLINEAR
;