1 #include <linux/ceph/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 "mds_client.h"
14 #include <linux/ceph/osd_client.h>
17 * Ceph address space ops.
19 * There are a few funny things going on here.
21 * The page->private field is used to reference a struct
22 * ceph_snap_context for _every_ dirty page. This indicates which
23 * snapshot the page was logically dirtied in, and thus which snap
24 * context needs to be associated with the osd write during writeback.
26 * Similarly, struct ceph_inode_info maintains a set of counters to
27 * count dirty pages on the inode. In the absence of snapshots,
28 * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count.
30 * When a snapshot is taken (that is, when the client receives
31 * notification that a snapshot was taken), each inode with caps and
32 * with dirty pages (dirty pages implies there is a cap) gets a new
33 * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending
34 * order, new snaps go to the tail). The i_wrbuffer_ref_head count is
35 * moved to capsnap->dirty. (Unless a sync write is currently in
36 * progress. In that case, the capsnap is said to be "pending", new
37 * writes cannot start, and the capsnap isn't "finalized" until the
38 * write completes (or fails) and a final size/mtime for the inode for
39 * that snap can be settled upon.) i_wrbuffer_ref_head is reset to 0.
41 * On writeback, we must submit writes to the osd IN SNAP ORDER. So,
42 * we look for the first capsnap in i_cap_snaps and write out pages in
43 * that snap context _only_. Then we move on to the next capsnap,
44 * eventually reaching the "live" or "head" context (i.e., pages that
45 * are not yet snapped) and are writing the most recently dirtied
48 * Invalidate and so forth must take care to ensure the dirty page
49 * accounting is preserved.
52 #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10))
53 #define CONGESTION_OFF_THRESH(congestion_kb) \
54 (CONGESTION_ON_THRESH(congestion_kb) - \
55 (CONGESTION_ON_THRESH(congestion_kb) >> 2))
57 static inline struct ceph_snap_context
*page_snap_context(struct page
*page
)
59 if (PagePrivate(page
))
60 return (void *)page
->private;
65 * Dirty a page. Optimistically adjust accounting, on the assumption
66 * that we won't race with invalidate. If we do, readjust.
68 static int ceph_set_page_dirty(struct page
*page
)
70 struct address_space
*mapping
= page
->mapping
;
72 struct ceph_inode_info
*ci
;
74 struct ceph_snap_context
*snapc
;
76 if (unlikely(!mapping
))
77 return !TestSetPageDirty(page
);
79 if (TestSetPageDirty(page
)) {
80 dout("%p set_page_dirty %p idx %lu -- already dirty\n",
81 mapping
->host
, page
, page
->index
);
85 inode
= mapping
->host
;
86 ci
= ceph_inode(inode
);
89 * Note that we're grabbing a snapc ref here without holding
92 snapc
= ceph_get_snap_context(ci
->i_snap_realm
->cached_context
);
95 spin_lock(&ci
->i_ceph_lock
);
96 if (ci
->i_head_snapc
== NULL
)
97 ci
->i_head_snapc
= ceph_get_snap_context(snapc
);
98 ++ci
->i_wrbuffer_ref_head
;
99 if (ci
->i_wrbuffer_ref
== 0)
101 ++ci
->i_wrbuffer_ref
;
102 dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d "
103 "snapc %p seq %lld (%d snaps)\n",
104 mapping
->host
, page
, page
->index
,
105 ci
->i_wrbuffer_ref
-1, ci
->i_wrbuffer_ref_head
-1,
106 ci
->i_wrbuffer_ref
, ci
->i_wrbuffer_ref_head
,
107 snapc
, snapc
->seq
, snapc
->num_snaps
);
108 spin_unlock(&ci
->i_ceph_lock
);
110 /* now adjust page */
111 spin_lock_irq(&mapping
->tree_lock
);
112 if (page
->mapping
) { /* Race with truncate? */
113 WARN_ON_ONCE(!PageUptodate(page
));
114 account_page_dirtied(page
, page
->mapping
);
115 radix_tree_tag_set(&mapping
->page_tree
,
116 page_index(page
), PAGECACHE_TAG_DIRTY
);
119 * Reference snap context in page->private. Also set
120 * PagePrivate so that we get invalidatepage callback.
122 page
->private = (unsigned long)snapc
;
123 SetPagePrivate(page
);
125 dout("ANON set_page_dirty %p (raced truncate?)\n", page
);
129 spin_unlock_irq(&mapping
->tree_lock
);
132 /* whoops, we failed to dirty the page */
133 ceph_put_wrbuffer_cap_refs(ci
, 1, snapc
);
135 __mark_inode_dirty(mapping
->host
, I_DIRTY_PAGES
);
137 BUG_ON(!PageDirty(page
));
142 * If we are truncating the full page (i.e. offset == 0), adjust the
143 * dirty page counters appropriately. Only called if there is private
146 static void ceph_invalidatepage(struct page
*page
, unsigned long offset
)
149 struct ceph_inode_info
*ci
;
150 struct ceph_snap_context
*snapc
= page_snap_context(page
);
152 BUG_ON(!PageLocked(page
));
153 BUG_ON(!PagePrivate(page
));
154 BUG_ON(!page
->mapping
);
156 inode
= page
->mapping
->host
;
159 * We can get non-dirty pages here due to races between
160 * set_page_dirty and truncate_complete_page; just spit out a
161 * warning, in case we end up with accounting problems later.
163 if (!PageDirty(page
))
164 pr_err("%p invalidatepage %p page not dirty\n", inode
, page
);
167 ClearPageChecked(page
);
169 ci
= ceph_inode(inode
);
171 dout("%p invalidatepage %p idx %lu full dirty page %lu\n",
172 inode
, page
, page
->index
, offset
);
173 ceph_put_wrbuffer_cap_refs(ci
, 1, snapc
);
174 ceph_put_snap_context(snapc
);
176 ClearPagePrivate(page
);
178 dout("%p invalidatepage %p idx %lu partial dirty page\n",
179 inode
, page
, page
->index
);
183 /* just a sanity check */
184 static int ceph_releasepage(struct page
*page
, gfp_t g
)
186 struct inode
*inode
= page
->mapping
? page
->mapping
->host
: NULL
;
187 dout("%p releasepage %p idx %lu\n", inode
, page
, page
->index
);
188 WARN_ON(PageDirty(page
));
189 WARN_ON(PagePrivate(page
));
194 * read a single page, without unlocking it.
196 static int readpage_nounlock(struct file
*filp
, struct page
*page
)
198 struct inode
*inode
= filp
->f_dentry
->d_inode
;
199 struct ceph_inode_info
*ci
= ceph_inode(inode
);
200 struct ceph_osd_client
*osdc
=
201 &ceph_inode_to_client(inode
)->client
->osdc
;
203 u64 len
= PAGE_CACHE_SIZE
;
205 dout("readpage inode %p file %p page %p index %lu\n",
206 inode
, filp
, page
, page
->index
);
207 err
= ceph_osdc_readpages(osdc
, ceph_vino(inode
), &ci
->i_layout
,
208 (u64
) page_offset(page
), &len
,
209 ci
->i_truncate_seq
, ci
->i_truncate_size
,
216 } else if (err
< PAGE_CACHE_SIZE
) {
217 /* zero fill remainder of page */
218 zero_user_segment(page
, err
, PAGE_CACHE_SIZE
);
220 SetPageUptodate(page
);
223 return err
< 0 ? err
: 0;
226 static int ceph_readpage(struct file
*filp
, struct page
*page
)
228 int r
= readpage_nounlock(filp
, page
);
234 * Finish an async read(ahead) op.
236 static void finish_read(struct ceph_osd_request
*req
, struct ceph_msg
*msg
)
238 struct inode
*inode
= req
->r_inode
;
239 struct ceph_osd_reply_head
*replyhead
;
244 replyhead
= msg
->front
.iov_base
;
245 WARN_ON(le32_to_cpu(replyhead
->num_ops
) == 0);
246 rc
= le32_to_cpu(replyhead
->result
);
247 bytes
= le32_to_cpu(msg
->hdr
.data_len
);
249 dout("finish_read %p req %p rc %d bytes %d\n", inode
, req
, rc
, bytes
);
251 /* unlock all pages, zeroing any data we didn't read */
252 for (i
= 0; i
< req
->r_num_pages
; i
++, bytes
-= PAGE_CACHE_SIZE
) {
253 struct page
*page
= req
->r_pages
[i
];
255 if (bytes
< (int)PAGE_CACHE_SIZE
) {
256 /* zero (remainder of) page */
257 int s
= bytes
< 0 ? 0 : bytes
;
258 zero_user_segment(page
, s
, PAGE_CACHE_SIZE
);
260 dout("finish_read %p uptodate %p idx %lu\n", inode
, page
,
262 flush_dcache_page(page
);
263 SetPageUptodate(page
);
265 page_cache_release(page
);
270 static void ceph_unlock_page_vector(struct page
**pages
, int num_pages
)
274 for (i
= 0; i
< num_pages
; i
++)
275 unlock_page(pages
[i
]);
279 * start an async read(ahead) operation. return nr_pages we submitted
280 * a read for on success, or negative error code.
282 static int start_read(struct inode
*inode
, struct list_head
*page_list
, int max
)
284 struct ceph_osd_client
*osdc
=
285 &ceph_inode_to_client(inode
)->client
->osdc
;
286 struct ceph_inode_info
*ci
= ceph_inode(inode
);
287 struct page
*page
= list_entry(page_list
->prev
, struct page
, lru
);
288 struct ceph_osd_request
*req
;
297 off
= (u64
) page_offset(page
);
300 next_index
= page
->index
;
301 list_for_each_entry_reverse(page
, page_list
, lru
) {
302 if (page
->index
!= next_index
)
306 if (max
&& nr_pages
== max
)
309 len
= nr_pages
<< PAGE_CACHE_SHIFT
;
310 dout("start_read %p nr_pages %d is %lld~%lld\n", inode
, nr_pages
,
313 req
= ceph_osdc_new_request(osdc
, &ci
->i_layout
, ceph_vino(inode
),
315 CEPH_OSD_OP_READ
, CEPH_OSD_FLAG_READ
,
317 ci
->i_truncate_seq
, ci
->i_truncate_size
,
322 /* build page vector */
323 nr_pages
= len
>> PAGE_CACHE_SHIFT
;
324 pages
= kmalloc(sizeof(*pages
) * nr_pages
, GFP_NOFS
);
328 for (i
= 0; i
< nr_pages
; ++i
) {
329 page
= list_entry(page_list
->prev
, struct page
, lru
);
330 BUG_ON(PageLocked(page
));
331 list_del(&page
->lru
);
333 dout("start_read %p adding %p idx %lu\n", inode
, page
,
335 if (add_to_page_cache_lru(page
, &inode
->i_data
, page
->index
,
337 page_cache_release(page
);
338 dout("start_read %p add_to_page_cache failed %p\n",
345 req
->r_pages
= pages
;
346 req
->r_num_pages
= nr_pages
;
347 req
->r_callback
= finish_read
;
348 req
->r_inode
= inode
;
350 dout("start_read %p starting %p %lld~%lld\n", inode
, req
, off
, len
);
351 ret
= ceph_osdc_start_request(osdc
, req
, false);
354 ceph_osdc_put_request(req
);
358 ceph_unlock_page_vector(pages
, nr_pages
);
359 ceph_release_page_vector(pages
, nr_pages
);
361 ceph_osdc_put_request(req
);
367 * Read multiple pages. Leave pages we don't read + unlock in page_list;
368 * the caller (VM) cleans them up.
370 static int ceph_readpages(struct file
*file
, struct address_space
*mapping
,
371 struct list_head
*page_list
, unsigned nr_pages
)
373 struct inode
*inode
= file
->f_dentry
->d_inode
;
374 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
378 if (fsc
->mount_options
->rsize
>= PAGE_CACHE_SIZE
)
379 max
= (fsc
->mount_options
->rsize
+ PAGE_CACHE_SIZE
- 1)
382 dout("readpages %p file %p nr_pages %d max %d\n", inode
, file
, nr_pages
,
384 while (!list_empty(page_list
)) {
385 rc
= start_read(inode
, page_list
, max
);
391 dout("readpages %p file %p ret %d\n", inode
, file
, rc
);
396 * Get ref for the oldest snapc for an inode with dirty data... that is, the
397 * only snap context we are allowed to write back.
399 static struct ceph_snap_context
*get_oldest_context(struct inode
*inode
,
402 struct ceph_inode_info
*ci
= ceph_inode(inode
);
403 struct ceph_snap_context
*snapc
= NULL
;
404 struct ceph_cap_snap
*capsnap
= NULL
;
406 spin_lock(&ci
->i_ceph_lock
);
407 list_for_each_entry(capsnap
, &ci
->i_cap_snaps
, ci_item
) {
408 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap
,
409 capsnap
->context
, capsnap
->dirty_pages
);
410 if (capsnap
->dirty_pages
) {
411 snapc
= ceph_get_snap_context(capsnap
->context
);
413 *snap_size
= capsnap
->size
;
417 if (!snapc
&& ci
->i_wrbuffer_ref_head
) {
418 snapc
= ceph_get_snap_context(ci
->i_head_snapc
);
419 dout(" head snapc %p has %d dirty pages\n",
420 snapc
, ci
->i_wrbuffer_ref_head
);
422 spin_unlock(&ci
->i_ceph_lock
);
427 * Write a single page, but leave the page locked.
429 * If we get a write error, set the page error bit, but still adjust the
430 * dirty page accounting (i.e., page is no longer dirty).
432 static int writepage_nounlock(struct page
*page
, struct writeback_control
*wbc
)
435 struct ceph_inode_info
*ci
;
436 struct ceph_fs_client
*fsc
;
437 struct ceph_osd_client
*osdc
;
438 loff_t page_off
= page_offset(page
);
439 int len
= PAGE_CACHE_SIZE
;
442 struct ceph_snap_context
*snapc
, *oldest
;
446 dout("writepage %p idx %lu\n", page
, page
->index
);
448 if (!page
->mapping
|| !page
->mapping
->host
) {
449 dout("writepage %p - no mapping\n", page
);
452 inode
= page
->mapping
->host
;
453 ci
= ceph_inode(inode
);
454 fsc
= ceph_inode_to_client(inode
);
455 osdc
= &fsc
->client
->osdc
;
457 /* verify this is a writeable snap context */
458 snapc
= page_snap_context(page
);
460 dout("writepage %p page %p not dirty?\n", inode
, page
);
463 oldest
= get_oldest_context(inode
, &snap_size
);
464 if (snapc
->seq
> oldest
->seq
) {
465 dout("writepage %p page %p snapc %p not writeable - noop\n",
467 /* we should only noop if called by kswapd */
468 WARN_ON((current
->flags
& PF_MEMALLOC
) == 0);
469 ceph_put_snap_context(oldest
);
472 ceph_put_snap_context(oldest
);
474 /* is this a partial page at end of file? */
478 i_size
= i_size_read(inode
);
479 if (i_size
< page_off
+ len
)
480 len
= i_size
- page_off
;
482 dout("writepage %p page %p index %lu on %llu~%u snapc %p\n",
483 inode
, page
, page
->index
, page_off
, len
, snapc
);
485 writeback_stat
= atomic_long_inc_return(&fsc
->writeback_count
);
487 CONGESTION_ON_THRESH(fsc
->mount_options
->congestion_kb
))
488 set_bdi_congested(&fsc
->backing_dev_info
, BLK_RW_ASYNC
);
490 set_page_writeback(page
);
491 err
= ceph_osdc_writepages(osdc
, ceph_vino(inode
),
492 &ci
->i_layout
, snapc
,
494 ci
->i_truncate_seq
, ci
->i_truncate_size
,
495 &inode
->i_mtime
, &page
, 1);
497 dout("writepage setting page/mapping error %d %p\n", err
, page
);
499 mapping_set_error(&inode
->i_data
, err
);
501 wbc
->pages_skipped
++;
503 dout("writepage cleaned page %p\n", page
);
504 err
= 0; /* vfs expects us to return 0 */
507 ClearPagePrivate(page
);
508 end_page_writeback(page
);
509 ceph_put_wrbuffer_cap_refs(ci
, 1, snapc
);
510 ceph_put_snap_context(snapc
); /* page's reference */
515 static int ceph_writepage(struct page
*page
, struct writeback_control
*wbc
)
518 struct inode
*inode
= page
->mapping
->host
;
521 err
= writepage_nounlock(page
, wbc
);
529 * lame release_pages helper. release_pages() isn't exported to
532 static void ceph_release_pages(struct page
**pages
, int num
)
537 pagevec_init(&pvec
, 0);
538 for (i
= 0; i
< num
; i
++) {
539 if (pagevec_add(&pvec
, pages
[i
]) == 0)
540 pagevec_release(&pvec
);
542 pagevec_release(&pvec
);
547 * async writeback completion handler.
549 * If we get an error, set the mapping error bit, but not the individual
552 static void writepages_finish(struct ceph_osd_request
*req
,
553 struct ceph_msg
*msg
)
555 struct inode
*inode
= req
->r_inode
;
556 struct ceph_osd_reply_head
*replyhead
;
557 struct ceph_osd_op
*op
;
558 struct ceph_inode_info
*ci
= ceph_inode(inode
);
562 struct ceph_snap_context
*snapc
= req
->r_snapc
;
563 struct address_space
*mapping
= inode
->i_mapping
;
566 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
568 unsigned issued
= ceph_caps_issued(ci
);
571 replyhead
= msg
->front
.iov_base
;
572 WARN_ON(le32_to_cpu(replyhead
->num_ops
) == 0);
573 op
= (void *)(replyhead
+ 1);
574 rc
= le32_to_cpu(replyhead
->result
);
575 bytes
= le64_to_cpu(op
->extent
.length
);
579 * Assume we wrote the pages we originally sent. The
580 * osd might reply with fewer pages if our writeback
581 * raced with a truncation and was adjusted at the osd,
582 * so don't believe the reply.
584 wrote
= req
->r_num_pages
;
587 mapping_set_error(mapping
, rc
);
589 dout("writepages_finish %p rc %d bytes %llu wrote %d (pages)\n",
590 inode
, rc
, bytes
, wrote
);
592 /* clean all pages */
593 for (i
= 0; i
< req
->r_num_pages
; i
++) {
594 page
= req
->r_pages
[i
];
596 WARN_ON(!PageUptodate(page
));
599 atomic_long_dec_return(&fsc
->writeback_count
);
601 CONGESTION_OFF_THRESH(fsc
->mount_options
->congestion_kb
))
602 clear_bdi_congested(&fsc
->backing_dev_info
,
605 ceph_put_snap_context(page_snap_context(page
));
607 ClearPagePrivate(page
);
608 dout("unlocking %d %p\n", i
, page
);
609 end_page_writeback(page
);
612 * We lost the cache cap, need to truncate the page before
613 * it is unlocked, otherwise we'd truncate it later in the
614 * page truncation thread, possibly losing some data that
617 if ((issued
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)) == 0)
618 generic_error_remove_page(inode
->i_mapping
, page
);
622 dout("%p wrote+cleaned %d pages\n", inode
, wrote
);
623 ceph_put_wrbuffer_cap_refs(ci
, req
->r_num_pages
, snapc
);
625 ceph_release_pages(req
->r_pages
, req
->r_num_pages
);
626 if (req
->r_pages_from_pool
)
627 mempool_free(req
->r_pages
,
628 ceph_sb_to_client(inode
->i_sb
)->wb_pagevec_pool
);
631 ceph_osdc_put_request(req
);
635 * allocate a page vec, either directly, or if necessary, via a the
636 * mempool. we avoid the mempool if we can because req->r_num_pages
637 * may be less than the maximum write size.
639 static void alloc_page_vec(struct ceph_fs_client
*fsc
,
640 struct ceph_osd_request
*req
)
642 req
->r_pages
= kmalloc(sizeof(struct page
*) * req
->r_num_pages
,
645 req
->r_pages
= mempool_alloc(fsc
->wb_pagevec_pool
, GFP_NOFS
);
646 req
->r_pages_from_pool
= 1;
647 WARN_ON(!req
->r_pages
);
652 * initiate async writeback
654 static int ceph_writepages_start(struct address_space
*mapping
,
655 struct writeback_control
*wbc
)
657 struct inode
*inode
= mapping
->host
;
658 struct ceph_inode_info
*ci
= ceph_inode(inode
);
659 struct ceph_fs_client
*fsc
;
660 pgoff_t index
, start
, end
;
663 pgoff_t max_pages
= 0, max_pages_ever
= 0;
664 struct ceph_snap_context
*snapc
= NULL
, *last_snapc
= NULL
, *pgsnapc
;
668 unsigned wsize
= 1 << inode
->i_blkbits
;
669 struct ceph_osd_request
*req
= NULL
;
674 * Include a 'sync' in the OSD request if this is a data
675 * integrity write (e.g., O_SYNC write or fsync()), or if our
676 * cap is being revoked.
678 do_sync
= wbc
->sync_mode
== WB_SYNC_ALL
;
679 if (ceph_caps_revoking(ci
, CEPH_CAP_FILE_BUFFER
))
681 dout("writepages_start %p dosync=%d (mode=%s)\n",
683 wbc
->sync_mode
== WB_SYNC_NONE
? "NONE" :
684 (wbc
->sync_mode
== WB_SYNC_ALL
? "ALL" : "HOLD"));
686 fsc
= ceph_inode_to_client(inode
);
687 if (fsc
->mount_state
== CEPH_MOUNT_SHUTDOWN
) {
688 pr_warning("writepage_start %p on forced umount\n", inode
);
689 return -EIO
; /* we're in a forced umount, don't write! */
691 if (fsc
->mount_options
->wsize
&& fsc
->mount_options
->wsize
< wsize
)
692 wsize
= fsc
->mount_options
->wsize
;
693 if (wsize
< PAGE_CACHE_SIZE
)
694 wsize
= PAGE_CACHE_SIZE
;
695 max_pages_ever
= wsize
>> PAGE_CACHE_SHIFT
;
697 pagevec_init(&pvec
, 0);
699 /* where to start/end? */
700 if (wbc
->range_cyclic
) {
701 start
= mapping
->writeback_index
; /* Start from prev offset */
703 dout(" cyclic, start at %lu\n", start
);
705 start
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
706 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
707 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
710 dout(" not cyclic, %lu to %lu\n", start
, end
);
715 /* find oldest snap context with dirty data */
716 ceph_put_snap_context(snapc
);
717 snapc
= get_oldest_context(inode
, &snap_size
);
719 /* hmm, why does writepages get called when there
721 dout(" no snap context with dirty data?\n");
724 dout(" oldest snapc is %p seq %lld (%d snaps)\n",
725 snapc
, snapc
->seq
, snapc
->num_snaps
);
726 if (last_snapc
&& snapc
!= last_snapc
) {
727 /* if we switched to a newer snapc, restart our scan at the
728 * start of the original file range. */
729 dout(" snapc differs from last pass, restarting at %lu\n",
735 while (!done
&& index
<= end
) {
739 int pvec_pages
, locked_pages
;
743 struct ceph_osd_request_head
*reqhead
;
744 struct ceph_osd_op
*op
;
749 max_pages
= max_pages_ever
;
753 want
= min(end
- index
,
754 min((pgoff_t
)PAGEVEC_SIZE
,
755 max_pages
- (pgoff_t
)locked_pages
) - 1)
757 pvec_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
760 dout("pagevec_lookup_tag got %d\n", pvec_pages
);
761 if (!pvec_pages
&& !locked_pages
)
763 for (i
= 0; i
< pvec_pages
&& locked_pages
< max_pages
; i
++) {
764 page
= pvec
.pages
[i
];
765 dout("? %p idx %lu\n", page
, page
->index
);
766 if (locked_pages
== 0)
767 lock_page(page
); /* first page */
768 else if (!trylock_page(page
))
771 /* only dirty pages, or our accounting breaks */
772 if (unlikely(!PageDirty(page
)) ||
773 unlikely(page
->mapping
!= mapping
)) {
774 dout("!dirty or !mapping %p\n", page
);
778 if (!wbc
->range_cyclic
&& page
->index
> end
) {
779 dout("end of range %p\n", page
);
784 if (next
&& (page
->index
!= next
)) {
785 dout("not consecutive %p\n", page
);
789 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
790 dout("waiting on writeback %p\n", page
);
791 wait_on_page_writeback(page
);
793 if ((snap_size
&& page_offset(page
) > snap_size
) ||
795 page_offset(page
) > i_size_read(inode
))) {
796 dout("%p page eof %llu\n", page
, snap_size
?
797 snap_size
: i_size_read(inode
));
802 if (PageWriteback(page
)) {
803 dout("%p under writeback\n", page
);
808 /* only if matching snap context */
809 pgsnapc
= page_snap_context(page
);
810 if (pgsnapc
->seq
> snapc
->seq
) {
811 dout("page snapc %p %lld > oldest %p %lld\n",
812 pgsnapc
, pgsnapc
->seq
, snapc
, snapc
->seq
);
815 continue; /* keep looking for snap */
819 if (!clear_page_dirty_for_io(page
)) {
820 dout("%p !clear_page_dirty_for_io\n", page
);
826 if (locked_pages
== 0) {
827 /* prepare async write request */
828 offset
= (u64
) page_offset(page
);
830 req
= ceph_osdc_new_request(&fsc
->client
->osdc
,
835 CEPH_OSD_FLAG_WRITE
|
836 CEPH_OSD_FLAG_ONDISK
,
840 &inode
->i_mtime
, true, 0);
848 max_pages
= req
->r_num_pages
;
850 alloc_page_vec(fsc
, req
);
851 req
->r_callback
= writepages_finish
;
852 req
->r_inode
= inode
;
855 /* note position of first page in pvec */
858 dout("%p will write page %p idx %lu\n",
859 inode
, page
, page
->index
);
862 atomic_long_inc_return(&fsc
->writeback_count
);
863 if (writeback_stat
> CONGESTION_ON_THRESH(
864 fsc
->mount_options
->congestion_kb
)) {
865 set_bdi_congested(&fsc
->backing_dev_info
,
869 set_page_writeback(page
);
870 req
->r_pages
[locked_pages
] = page
;
872 next
= page
->index
+ 1;
875 /* did we get anything? */
877 goto release_pvec_pages
;
880 BUG_ON(!locked_pages
|| first
< 0);
882 if (pvec_pages
&& i
== pvec_pages
&&
883 locked_pages
< max_pages
) {
884 dout("reached end pvec, trying for more\n");
885 pagevec_reinit(&pvec
);
889 /* shift unused pages over in the pvec... we
890 * will need to release them below. */
891 for (j
= i
; j
< pvec_pages
; j
++) {
892 dout(" pvec leftover page %p\n",
894 pvec
.pages
[j
-i
+first
] = pvec
.pages
[j
];
899 /* submit the write */
900 offset
= req
->r_pages
[0]->index
<< PAGE_CACHE_SHIFT
;
901 len
= min((snap_size
? snap_size
: i_size_read(inode
)) - offset
,
902 (u64
)locked_pages
<< PAGE_CACHE_SHIFT
);
903 dout("writepages got %d pages at %llu~%llu\n",
904 locked_pages
, offset
, len
);
906 /* revise final length, page count */
907 req
->r_num_pages
= locked_pages
;
908 reqhead
= req
->r_request
->front
.iov_base
;
909 op
= (void *)(reqhead
+ 1);
910 op
->extent
.length
= cpu_to_le64(len
);
911 op
->payload_len
= cpu_to_le32(len
);
912 req
->r_request
->hdr
.data_len
= cpu_to_le32(len
);
914 rc
= ceph_osdc_start_request(&fsc
->client
->osdc
, req
, true);
920 wbc
->nr_to_write
-= locked_pages
;
921 if (wbc
->nr_to_write
<= 0)
925 dout("pagevec_release on %d pages (%p)\n", (int)pvec
.nr
,
926 pvec
.nr
? pvec
.pages
[0] : NULL
);
927 pagevec_release(&pvec
);
929 if (locked_pages
&& !done
)
933 if (should_loop
&& !done
) {
934 /* more to do; loop back to beginning of file */
935 dout("writepages looping back to beginning of file\n");
941 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
942 mapping
->writeback_index
= index
;
946 ceph_osdc_put_request(req
);
947 ceph_put_snap_context(snapc
);
948 dout("writepages done, rc = %d\n", rc
);
955 * See if a given @snapc is either writeable, or already written.
957 static int context_is_writeable_or_written(struct inode
*inode
,
958 struct ceph_snap_context
*snapc
)
960 struct ceph_snap_context
*oldest
= get_oldest_context(inode
, NULL
);
961 int ret
= !oldest
|| snapc
->seq
<= oldest
->seq
;
963 ceph_put_snap_context(oldest
);
968 * We are only allowed to write into/dirty the page if the page is
969 * clean, or already dirty within the same snap context.
971 * called with page locked.
972 * return success with page locked,
973 * or any failure (incl -EAGAIN) with page unlocked.
975 static int ceph_update_writeable_page(struct file
*file
,
976 loff_t pos
, unsigned len
,
979 struct inode
*inode
= file
->f_dentry
->d_inode
;
980 struct ceph_inode_info
*ci
= ceph_inode(inode
);
981 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
982 loff_t page_off
= pos
& PAGE_CACHE_MASK
;
983 int pos_in_page
= pos
& ~PAGE_CACHE_MASK
;
984 int end_in_page
= pos_in_page
+ len
;
987 struct ceph_snap_context
*snapc
, *oldest
;
990 /* writepages currently holds page lock, but if we change that later, */
991 wait_on_page_writeback(page
);
993 /* check snap context */
994 BUG_ON(!ci
->i_snap_realm
);
995 down_read(&mdsc
->snap_rwsem
);
996 BUG_ON(!ci
->i_snap_realm
->cached_context
);
997 snapc
= page_snap_context(page
);
998 if (snapc
&& snapc
!= ci
->i_head_snapc
) {
1000 * this page is already dirty in another (older) snap
1001 * context! is it writeable now?
1003 oldest
= get_oldest_context(inode
, NULL
);
1004 up_read(&mdsc
->snap_rwsem
);
1006 if (snapc
->seq
> oldest
->seq
) {
1007 ceph_put_snap_context(oldest
);
1008 dout(" page %p snapc %p not current or oldest\n",
1011 * queue for writeback, and wait for snapc to
1012 * be writeable or written
1014 snapc
= ceph_get_snap_context(snapc
);
1016 ceph_queue_writeback(inode
);
1017 r
= wait_event_interruptible(ci
->i_cap_wq
,
1018 context_is_writeable_or_written(inode
, snapc
));
1019 ceph_put_snap_context(snapc
);
1020 if (r
== -ERESTARTSYS
)
1024 ceph_put_snap_context(oldest
);
1026 /* yay, writeable, do it now (without dropping page lock) */
1027 dout(" page %p snapc %p not current, but oldest\n",
1029 if (!clear_page_dirty_for_io(page
))
1031 r
= writepage_nounlock(page
, NULL
);
1037 if (PageUptodate(page
)) {
1038 dout(" page %p already uptodate\n", page
);
1043 if (pos_in_page
== 0 && len
== PAGE_CACHE_SIZE
)
1046 /* past end of file? */
1047 i_size
= inode
->i_size
; /* caller holds i_mutex */
1049 if (i_size
+ len
> inode
->i_sb
->s_maxbytes
) {
1050 /* file is too big */
1055 if (page_off
>= i_size
||
1056 (pos_in_page
== 0 && (pos
+len
) >= i_size
&&
1057 end_in_page
- pos_in_page
!= PAGE_CACHE_SIZE
)) {
1058 dout(" zeroing %p 0 - %d and %d - %d\n",
1059 page
, pos_in_page
, end_in_page
, (int)PAGE_CACHE_SIZE
);
1060 zero_user_segments(page
,
1062 end_in_page
, PAGE_CACHE_SIZE
);
1066 /* we need to read it. */
1067 up_read(&mdsc
->snap_rwsem
);
1068 r
= readpage_nounlock(file
, page
);
1074 up_read(&mdsc
->snap_rwsem
);
1081 * We are only allowed to write into/dirty the page if the page is
1082 * clean, or already dirty within the same snap context.
1084 static int ceph_write_begin(struct file
*file
, struct address_space
*mapping
,
1085 loff_t pos
, unsigned len
, unsigned flags
,
1086 struct page
**pagep
, void **fsdata
)
1088 struct inode
*inode
= file
->f_dentry
->d_inode
;
1089 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1090 struct ceph_file_info
*fi
= file
->private_data
;
1092 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
1093 int r
, want
, got
= 0;
1095 if (fi
->fmode
& CEPH_FILE_MODE_LAZY
)
1096 want
= CEPH_CAP_FILE_BUFFER
| CEPH_CAP_FILE_LAZYIO
;
1098 want
= CEPH_CAP_FILE_BUFFER
;
1100 dout("write_begin %p %llx.%llx %llu~%u getting caps. i_size %llu\n",
1101 inode
, ceph_vinop(inode
), pos
, len
, inode
->i_size
);
1102 r
= ceph_get_caps(ci
, CEPH_CAP_FILE_WR
, want
, &got
, pos
+len
);
1105 dout("write_begin %p %llx.%llx %llu~%u got cap refs on %s\n",
1106 inode
, ceph_vinop(inode
), pos
, len
, ceph_cap_string(got
));
1107 if (!(got
& (CEPH_CAP_FILE_BUFFER
|CEPH_CAP_FILE_LAZYIO
))) {
1108 ceph_put_cap_refs(ci
, got
);
1114 page
= grab_cache_page_write_begin(mapping
, index
, 0);
1120 dout("write_begin file %p inode %p page %p %d~%d\n", file
,
1121 inode
, page
, (int)pos
, (int)len
);
1123 r
= ceph_update_writeable_page(file
, pos
, len
, page
);
1125 page_cache_release(page
);
1126 } while (r
== -EAGAIN
);
1129 ceph_put_cap_refs(ci
, got
);
1132 *(int *)fsdata
= got
;
1138 * we don't do anything in here that simple_write_end doesn't do
1139 * except adjust dirty page accounting and drop read lock on
1142 static int ceph_write_end(struct file
*file
, struct address_space
*mapping
,
1143 loff_t pos
, unsigned len
, unsigned copied
,
1144 struct page
*page
, void *fsdata
)
1146 struct inode
*inode
= file
->f_dentry
->d_inode
;
1147 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1148 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
1149 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
1150 unsigned from
= pos
& (PAGE_CACHE_SIZE
- 1);
1152 int got
= (unsigned long)fsdata
;
1154 dout("write_end file %p inode %p page %p %d~%d (%d)\n", file
,
1155 inode
, page
, (int)pos
, (int)copied
, (int)len
);
1157 /* zero the stale part of the page if we did a short copy */
1159 zero_user_segment(page
, from
+copied
, len
);
1161 /* did file size increase? */
1162 /* (no need for i_size_read(); we caller holds i_mutex */
1163 if (pos
+copied
> inode
->i_size
)
1164 check_cap
= ceph_inode_set_size(inode
, pos
+copied
);
1166 if (!PageUptodate(page
))
1167 SetPageUptodate(page
);
1169 set_page_dirty(page
);
1172 up_read(&mdsc
->snap_rwsem
);
1173 page_cache_release(page
);
1177 spin_lock(&ci
->i_ceph_lock
);
1178 dirty
= __ceph_mark_dirty_caps(ci
, CEPH_CAP_FILE_WR
);
1179 spin_unlock(&ci
->i_ceph_lock
);
1181 __mark_inode_dirty(inode
, dirty
);
1184 dout("write_end %p %llx.%llx %llu~%u dropping cap refs on %s\n",
1185 inode
, ceph_vinop(inode
), pos
, len
, ceph_cap_string(got
));
1186 ceph_put_cap_refs(ci
, got
);
1189 ceph_check_caps(ceph_inode(inode
), CHECK_CAPS_AUTHONLY
, NULL
);
1195 * we set .direct_IO to indicate direct io is supported, but since we
1196 * intercept O_DIRECT reads and writes early, this function should
1199 static ssize_t
ceph_direct_io(int rw
, struct kiocb
*iocb
,
1200 const struct iovec
*iov
,
1201 loff_t pos
, unsigned long nr_segs
)
1207 const struct address_space_operations ceph_aops
= {
1208 .readpage
= ceph_readpage
,
1209 .readpages
= ceph_readpages
,
1210 .writepage
= ceph_writepage
,
1211 .writepages
= ceph_writepages_start
,
1212 .write_begin
= ceph_write_begin
,
1213 .write_end
= ceph_write_end
,
1214 .set_page_dirty
= ceph_set_page_dirty
,
1215 .invalidatepage
= ceph_invalidatepage
,
1216 .releasepage
= ceph_releasepage
,
1217 .direct_IO
= ceph_direct_io
,
1226 * Reuse write_begin here for simplicity.
1228 static int ceph_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1230 struct inode
*inode
= vma
->vm_file
->f_dentry
->d_inode
;
1231 struct page
*page
= vmf
->page
;
1232 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
1233 loff_t off
= page_offset(page
);
1237 /* Update time before taking page lock */
1238 file_update_time(vma
->vm_file
);
1240 size
= i_size_read(inode
);
1241 if (off
+ PAGE_CACHE_SIZE
<= size
)
1242 len
= PAGE_CACHE_SIZE
;
1244 len
= size
& ~PAGE_CACHE_MASK
;
1246 dout("page_mkwrite %p %llu~%llu page %p idx %lu\n", inode
,
1247 off
, len
, page
, page
->index
);
1251 ret
= VM_FAULT_NOPAGE
;
1253 (page
->mapping
!= inode
->i_mapping
))
1256 ret
= ceph_update_writeable_page(vma
->vm_file
, off
, len
, page
);
1258 /* success. we'll keep the page locked. */
1259 set_page_dirty(page
);
1260 up_read(&mdsc
->snap_rwsem
);
1261 ret
= VM_FAULT_LOCKED
;
1266 ret
= VM_FAULT_SIGBUS
;
1269 dout("page_mkwrite %p %llu~%llu = %d\n", inode
, off
, len
, ret
);
1270 if (ret
!= VM_FAULT_LOCKED
)
1275 static struct vm_operations_struct ceph_vmops
= {
1276 .fault
= filemap_fault
,
1277 .page_mkwrite
= ceph_page_mkwrite
,
1278 .remap_pages
= generic_file_remap_pages
,
1281 int ceph_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1283 struct address_space
*mapping
= file
->f_mapping
;
1285 if (!mapping
->a_ops
->readpage
)
1287 file_accessed(file
);
1288 vma
->vm_ops
= &ceph_vmops
;