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"
15 #include <linux/ceph/osd_client.h>
18 * Ceph address space ops.
20 * There are a few funny things going on here.
22 * The page->private field is used to reference a struct
23 * ceph_snap_context for _every_ dirty page. This indicates which
24 * snapshot the page was logically dirtied in, and thus which snap
25 * context needs to be associated with the osd write during writeback.
27 * Similarly, struct ceph_inode_info maintains a set of counters to
28 * count dirty pages on the inode. In the absence of snapshots,
29 * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count.
31 * When a snapshot is taken (that is, when the client receives
32 * notification that a snapshot was taken), each inode with caps and
33 * with dirty pages (dirty pages implies there is a cap) gets a new
34 * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending
35 * order, new snaps go to the tail). The i_wrbuffer_ref_head count is
36 * moved to capsnap->dirty. (Unless a sync write is currently in
37 * progress. In that case, the capsnap is said to be "pending", new
38 * writes cannot start, and the capsnap isn't "finalized" until the
39 * write completes (or fails) and a final size/mtime for the inode for
40 * that snap can be settled upon.) i_wrbuffer_ref_head is reset to 0.
42 * On writeback, we must submit writes to the osd IN SNAP ORDER. So,
43 * we look for the first capsnap in i_cap_snaps and write out pages in
44 * that snap context _only_. Then we move on to the next capsnap,
45 * eventually reaching the "live" or "head" context (i.e., pages that
46 * are not yet snapped) and are writing the most recently dirtied
49 * Invalidate and so forth must take care to ensure the dirty page
50 * accounting is preserved.
53 #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10))
54 #define CONGESTION_OFF_THRESH(congestion_kb) \
55 (CONGESTION_ON_THRESH(congestion_kb) - \
56 (CONGESTION_ON_THRESH(congestion_kb) >> 2))
58 static inline struct ceph_snap_context
*page_snap_context(struct page
*page
)
60 if (PagePrivate(page
))
61 return (void *)page
->private;
66 * Dirty a page. Optimistically adjust accounting, on the assumption
67 * that we won't race with invalidate. If we do, readjust.
69 static int ceph_set_page_dirty(struct page
*page
)
71 struct address_space
*mapping
= page
->mapping
;
73 struct ceph_inode_info
*ci
;
74 struct ceph_snap_context
*snapc
;
77 if (unlikely(!mapping
))
78 return !TestSetPageDirty(page
);
80 if (PageDirty(page
)) {
81 dout("%p set_page_dirty %p idx %lu -- already dirty\n",
82 mapping
->host
, page
, page
->index
);
83 BUG_ON(!PagePrivate(page
));
87 inode
= mapping
->host
;
88 ci
= ceph_inode(inode
);
91 * Note that we're grabbing a snapc ref here without holding
94 snapc
= ceph_get_snap_context(ci
->i_snap_realm
->cached_context
);
97 spin_lock(&ci
->i_ceph_lock
);
98 if (ci
->i_head_snapc
== NULL
)
99 ci
->i_head_snapc
= ceph_get_snap_context(snapc
);
100 ++ci
->i_wrbuffer_ref_head
;
101 if (ci
->i_wrbuffer_ref
== 0)
103 ++ci
->i_wrbuffer_ref
;
104 dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d "
105 "snapc %p seq %lld (%d snaps)\n",
106 mapping
->host
, page
, page
->index
,
107 ci
->i_wrbuffer_ref
-1, ci
->i_wrbuffer_ref_head
-1,
108 ci
->i_wrbuffer_ref
, ci
->i_wrbuffer_ref_head
,
109 snapc
, snapc
->seq
, snapc
->num_snaps
);
110 spin_unlock(&ci
->i_ceph_lock
);
113 * Reference snap context in page->private. Also set
114 * PagePrivate so that we get invalidatepage callback.
116 BUG_ON(PagePrivate(page
));
117 page
->private = (unsigned long)snapc
;
118 SetPagePrivate(page
);
120 ret
= __set_page_dirty_nobuffers(page
);
121 WARN_ON(!PageLocked(page
));
122 WARN_ON(!page
->mapping
);
128 * If we are truncating the full page (i.e. offset == 0), adjust the
129 * dirty page counters appropriately. Only called if there is private
132 static void ceph_invalidatepage(struct page
*page
, unsigned int offset
,
136 struct ceph_inode_info
*ci
;
137 struct ceph_snap_context
*snapc
= page_snap_context(page
);
139 inode
= page
->mapping
->host
;
140 ci
= ceph_inode(inode
);
142 if (offset
!= 0 || length
!= PAGE_CACHE_SIZE
) {
143 dout("%p invalidatepage %p idx %lu partial dirty page %u~%u\n",
144 inode
, page
, page
->index
, offset
, length
);
148 ceph_invalidate_fscache_page(inode
, page
);
150 if (!PagePrivate(page
))
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
);
161 ClearPageChecked(page
);
163 dout("%p invalidatepage %p idx %lu full dirty page\n",
164 inode
, page
, page
->index
);
166 ceph_put_wrbuffer_cap_refs(ci
, 1, snapc
);
167 ceph_put_snap_context(snapc
);
169 ClearPagePrivate(page
);
172 static int ceph_releasepage(struct page
*page
, gfp_t g
)
174 struct inode
*inode
= page
->mapping
? page
->mapping
->host
: NULL
;
175 dout("%p releasepage %p idx %lu\n", inode
, page
, page
->index
);
176 WARN_ON(PageDirty(page
));
178 /* Can we release the page from the cache? */
179 if (!ceph_release_fscache_page(page
, g
))
182 return !PagePrivate(page
);
186 * read a single page, without unlocking it.
188 static int readpage_nounlock(struct file
*filp
, struct page
*page
)
190 struct inode
*inode
= file_inode(filp
);
191 struct ceph_inode_info
*ci
= ceph_inode(inode
);
192 struct ceph_osd_client
*osdc
=
193 &ceph_inode_to_client(inode
)->client
->osdc
;
195 u64 len
= PAGE_CACHE_SIZE
;
197 err
= ceph_readpage_from_fscache(inode
, page
);
202 dout("readpage inode %p file %p page %p index %lu\n",
203 inode
, filp
, page
, page
->index
);
204 err
= ceph_osdc_readpages(osdc
, ceph_vino(inode
), &ci
->i_layout
,
205 (u64
) page_offset(page
), &len
,
206 ci
->i_truncate_seq
, ci
->i_truncate_size
,
213 } else if (err
< PAGE_CACHE_SIZE
) {
214 /* zero fill remainder of page */
215 zero_user_segment(page
, err
, PAGE_CACHE_SIZE
);
217 SetPageUptodate(page
);
220 ceph_readpage_to_fscache(inode
, 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_data
*osd_data
;
240 int rc
= req
->r_result
;
241 int bytes
= le32_to_cpu(msg
->hdr
.data_len
);
245 dout("finish_read %p req %p rc %d bytes %d\n", inode
, req
, rc
, bytes
);
247 /* unlock all pages, zeroing any data we didn't read */
248 osd_data
= osd_req_op_extent_osd_data(req
, 0);
249 BUG_ON(osd_data
->type
!= CEPH_OSD_DATA_TYPE_PAGES
);
250 num_pages
= calc_pages_for((u64
)osd_data
->alignment
,
251 (u64
)osd_data
->length
);
252 for (i
= 0; i
< num_pages
; i
++) {
253 struct page
*page
= osd_data
->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
);
264 ceph_readpage_to_fscache(inode
, page
);
266 page_cache_release(page
);
267 bytes
-= PAGE_CACHE_SIZE
;
269 kfree(osd_data
->pages
);
272 static void ceph_unlock_page_vector(struct page
**pages
, int num_pages
)
276 for (i
= 0; i
< num_pages
; i
++)
277 unlock_page(pages
[i
]);
281 * start an async read(ahead) operation. return nr_pages we submitted
282 * a read for on success, or negative error code.
284 static int start_read(struct inode
*inode
, struct list_head
*page_list
, int max
)
286 struct ceph_osd_client
*osdc
=
287 &ceph_inode_to_client(inode
)->client
->osdc
;
288 struct ceph_inode_info
*ci
= ceph_inode(inode
);
289 struct page
*page
= list_entry(page_list
->prev
, struct page
, lru
);
290 struct ceph_vino vino
;
291 struct ceph_osd_request
*req
;
300 off
= (u64
) page_offset(page
);
303 next_index
= page
->index
;
304 list_for_each_entry_reverse(page
, page_list
, lru
) {
305 if (page
->index
!= next_index
)
309 if (max
&& nr_pages
== max
)
312 len
= nr_pages
<< PAGE_CACHE_SHIFT
;
313 dout("start_read %p nr_pages %d is %lld~%lld\n", inode
, nr_pages
,
315 vino
= ceph_vino(inode
);
316 req
= ceph_osdc_new_request(osdc
, &ci
->i_layout
, vino
, off
, &len
,
318 CEPH_OSD_FLAG_READ
, NULL
,
319 ci
->i_truncate_seq
, ci
->i_truncate_size
,
324 /* build page vector */
325 nr_pages
= calc_pages_for(0, len
);
326 pages
= kmalloc(sizeof(*pages
) * nr_pages
, GFP_NOFS
);
330 for (i
= 0; i
< nr_pages
; ++i
) {
331 page
= list_entry(page_list
->prev
, struct page
, lru
);
332 BUG_ON(PageLocked(page
));
333 list_del(&page
->lru
);
335 dout("start_read %p adding %p idx %lu\n", inode
, page
,
337 if (add_to_page_cache_lru(page
, &inode
->i_data
, page
->index
,
339 page_cache_release(page
);
340 dout("start_read %p add_to_page_cache failed %p\n",
347 osd_req_op_extent_osd_data_pages(req
, 0, pages
, len
, 0, false, false);
348 req
->r_callback
= finish_read
;
349 req
->r_inode
= inode
;
351 ceph_osdc_build_request(req
, off
, NULL
, vino
.snap
, NULL
);
353 dout("start_read %p starting %p %lld~%lld\n", inode
, req
, off
, len
);
354 ret
= ceph_osdc_start_request(osdc
, req
, false);
357 ceph_osdc_put_request(req
);
361 ceph_unlock_page_vector(pages
, nr_pages
);
362 ceph_release_page_vector(pages
, nr_pages
);
364 ceph_osdc_put_request(req
);
370 * Read multiple pages. Leave pages we don't read + unlock in page_list;
371 * the caller (VM) cleans them up.
373 static int ceph_readpages(struct file
*file
, struct address_space
*mapping
,
374 struct list_head
*page_list
, unsigned nr_pages
)
376 struct inode
*inode
= file_inode(file
);
377 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
381 rc
= ceph_readpages_from_fscache(mapping
->host
, mapping
, page_list
,
387 if (fsc
->mount_options
->rsize
>= PAGE_CACHE_SIZE
)
388 max
= (fsc
->mount_options
->rsize
+ PAGE_CACHE_SIZE
- 1)
391 dout("readpages %p file %p nr_pages %d max %d\n", inode
,
394 while (!list_empty(page_list
)) {
395 rc
= start_read(inode
, page_list
, max
);
401 dout("readpages %p file %p ret %d\n", inode
, file
, rc
);
406 * Get ref for the oldest snapc for an inode with dirty data... that is, the
407 * only snap context we are allowed to write back.
409 static struct ceph_snap_context
*get_oldest_context(struct inode
*inode
,
412 struct ceph_inode_info
*ci
= ceph_inode(inode
);
413 struct ceph_snap_context
*snapc
= NULL
;
414 struct ceph_cap_snap
*capsnap
= NULL
;
416 spin_lock(&ci
->i_ceph_lock
);
417 list_for_each_entry(capsnap
, &ci
->i_cap_snaps
, ci_item
) {
418 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap
,
419 capsnap
->context
, capsnap
->dirty_pages
);
420 if (capsnap
->dirty_pages
) {
421 snapc
= ceph_get_snap_context(capsnap
->context
);
423 *snap_size
= capsnap
->size
;
427 if (!snapc
&& ci
->i_wrbuffer_ref_head
) {
428 snapc
= ceph_get_snap_context(ci
->i_head_snapc
);
429 dout(" head snapc %p has %d dirty pages\n",
430 snapc
, ci
->i_wrbuffer_ref_head
);
432 spin_unlock(&ci
->i_ceph_lock
);
437 * Write a single page, but leave the page locked.
439 * If we get a write error, set the page error bit, but still adjust the
440 * dirty page accounting (i.e., page is no longer dirty).
442 static int writepage_nounlock(struct page
*page
, struct writeback_control
*wbc
)
445 struct ceph_inode_info
*ci
;
446 struct ceph_fs_client
*fsc
;
447 struct ceph_osd_client
*osdc
;
448 struct ceph_snap_context
*snapc
, *oldest
;
449 loff_t page_off
= page_offset(page
);
451 u64 truncate_size
, snap_size
= 0;
453 int err
= 0, len
= PAGE_CACHE_SIZE
;
455 dout("writepage %p idx %lu\n", page
, page
->index
);
457 if (!page
->mapping
|| !page
->mapping
->host
) {
458 dout("writepage %p - no mapping\n", page
);
461 inode
= page
->mapping
->host
;
462 ci
= ceph_inode(inode
);
463 fsc
= ceph_inode_to_client(inode
);
464 osdc
= &fsc
->client
->osdc
;
466 /* verify this is a writeable snap context */
467 snapc
= page_snap_context(page
);
469 dout("writepage %p page %p not dirty?\n", inode
, page
);
472 oldest
= get_oldest_context(inode
, &snap_size
);
473 if (snapc
->seq
> oldest
->seq
) {
474 dout("writepage %p page %p snapc %p not writeable - noop\n",
476 /* we should only noop if called by kswapd */
477 WARN_ON((current
->flags
& PF_MEMALLOC
) == 0);
478 ceph_put_snap_context(oldest
);
481 ceph_put_snap_context(oldest
);
483 spin_lock(&ci
->i_ceph_lock
);
484 truncate_seq
= ci
->i_truncate_seq
;
485 truncate_size
= ci
->i_truncate_size
;
487 snap_size
= i_size_read(inode
);
488 spin_unlock(&ci
->i_ceph_lock
);
490 /* is this a partial page at end of file? */
491 if (page_off
>= snap_size
) {
492 dout("%p page eof %llu\n", page
, snap_size
);
495 if (snap_size
< page_off
+ len
)
496 len
= snap_size
- page_off
;
498 dout("writepage %p page %p index %lu on %llu~%u snapc %p\n",
499 inode
, page
, page
->index
, page_off
, len
, snapc
);
501 writeback_stat
= atomic_long_inc_return(&fsc
->writeback_count
);
503 CONGESTION_ON_THRESH(fsc
->mount_options
->congestion_kb
))
504 set_bdi_congested(&fsc
->backing_dev_info
, BLK_RW_ASYNC
);
506 ceph_readpage_to_fscache(inode
, page
);
508 set_page_writeback(page
);
509 err
= ceph_osdc_writepages(osdc
, ceph_vino(inode
),
510 &ci
->i_layout
, snapc
,
512 truncate_seq
, truncate_size
,
513 &inode
->i_mtime
, &page
, 1);
515 dout("writepage setting page/mapping error %d %p\n", err
, page
);
517 mapping_set_error(&inode
->i_data
, err
);
519 wbc
->pages_skipped
++;
521 dout("writepage cleaned page %p\n", page
);
522 err
= 0; /* vfs expects us to return 0 */
525 ClearPagePrivate(page
);
526 end_page_writeback(page
);
527 ceph_put_wrbuffer_cap_refs(ci
, 1, snapc
);
528 ceph_put_snap_context(snapc
); /* page's reference */
533 static int ceph_writepage(struct page
*page
, struct writeback_control
*wbc
)
536 struct inode
*inode
= page
->mapping
->host
;
539 err
= writepage_nounlock(page
, wbc
);
547 * lame release_pages helper. release_pages() isn't exported to
550 static void ceph_release_pages(struct page
**pages
, int num
)
555 pagevec_init(&pvec
, 0);
556 for (i
= 0; i
< num
; i
++) {
557 if (pagevec_add(&pvec
, pages
[i
]) == 0)
558 pagevec_release(&pvec
);
560 pagevec_release(&pvec
);
564 * async writeback completion handler.
566 * If we get an error, set the mapping error bit, but not the individual
569 static void writepages_finish(struct ceph_osd_request
*req
,
570 struct ceph_msg
*msg
)
572 struct inode
*inode
= req
->r_inode
;
573 struct ceph_inode_info
*ci
= ceph_inode(inode
);
574 struct ceph_osd_data
*osd_data
;
579 struct ceph_snap_context
*snapc
= req
->r_snapc
;
580 struct address_space
*mapping
= inode
->i_mapping
;
581 int rc
= req
->r_result
;
582 u64 bytes
= req
->r_ops
[0].extent
.length
;
583 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
585 unsigned issued
= ceph_caps_issued(ci
);
587 osd_data
= osd_req_op_extent_osd_data(req
, 0);
588 BUG_ON(osd_data
->type
!= CEPH_OSD_DATA_TYPE_PAGES
);
589 num_pages
= calc_pages_for((u64
)osd_data
->alignment
,
590 (u64
)osd_data
->length
);
593 * Assume we wrote the pages we originally sent. The
594 * osd might reply with fewer pages if our writeback
595 * raced with a truncation and was adjusted at the osd,
596 * so don't believe the reply.
601 mapping_set_error(mapping
, rc
);
603 dout("writepages_finish %p rc %d bytes %llu wrote %d (pages)\n",
604 inode
, rc
, bytes
, wrote
);
606 /* clean all pages */
607 for (i
= 0; i
< num_pages
; i
++) {
608 page
= osd_data
->pages
[i
];
610 WARN_ON(!PageUptodate(page
));
613 atomic_long_dec_return(&fsc
->writeback_count
);
615 CONGESTION_OFF_THRESH(fsc
->mount_options
->congestion_kb
))
616 clear_bdi_congested(&fsc
->backing_dev_info
,
619 ceph_put_snap_context(page_snap_context(page
));
621 ClearPagePrivate(page
);
622 dout("unlocking %d %p\n", i
, page
);
623 end_page_writeback(page
);
626 * We lost the cache cap, need to truncate the page before
627 * it is unlocked, otherwise we'd truncate it later in the
628 * page truncation thread, possibly losing some data that
631 if ((issued
& (CEPH_CAP_FILE_CACHE
|CEPH_CAP_FILE_LAZYIO
)) == 0)
632 generic_error_remove_page(inode
->i_mapping
, page
);
636 dout("%p wrote+cleaned %d pages\n", inode
, wrote
);
637 ceph_put_wrbuffer_cap_refs(ci
, num_pages
, snapc
);
639 ceph_release_pages(osd_data
->pages
, num_pages
);
640 if (osd_data
->pages_from_pool
)
641 mempool_free(osd_data
->pages
,
642 ceph_sb_to_client(inode
->i_sb
)->wb_pagevec_pool
);
644 kfree(osd_data
->pages
);
645 ceph_osdc_put_request(req
);
649 * initiate async writeback
651 static int ceph_writepages_start(struct address_space
*mapping
,
652 struct writeback_control
*wbc
)
654 struct inode
*inode
= mapping
->host
;
655 struct ceph_inode_info
*ci
= ceph_inode(inode
);
656 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
657 struct ceph_vino vino
= ceph_vino(inode
);
658 pgoff_t index
, start
, end
;
661 pgoff_t max_pages
= 0, max_pages_ever
= 0;
662 struct ceph_snap_context
*snapc
= NULL
, *last_snapc
= NULL
, *pgsnapc
;
666 unsigned wsize
= 1 << inode
->i_blkbits
;
667 struct ceph_osd_request
*req
= NULL
;
669 u64 truncate_size
, snap_size
;
673 * Include a 'sync' in the OSD request if this is a data
674 * integrity write (e.g., O_SYNC write or fsync()), or if our
675 * cap is being revoked.
677 if ((wbc
->sync_mode
== WB_SYNC_ALL
) ||
678 ceph_caps_revoking(ci
, CEPH_CAP_FILE_BUFFER
))
680 dout("writepages_start %p dosync=%d (mode=%s)\n",
682 wbc
->sync_mode
== WB_SYNC_NONE
? "NONE" :
683 (wbc
->sync_mode
== WB_SYNC_ALL
? "ALL" : "HOLD"));
685 if (fsc
->mount_state
== CEPH_MOUNT_SHUTDOWN
) {
686 pr_warning("writepage_start %p on forced umount\n", inode
);
687 return -EIO
; /* we're in a forced umount, don't write! */
689 if (fsc
->mount_options
->wsize
&& fsc
->mount_options
->wsize
< wsize
)
690 wsize
= fsc
->mount_options
->wsize
;
691 if (wsize
< PAGE_CACHE_SIZE
)
692 wsize
= PAGE_CACHE_SIZE
;
693 max_pages_ever
= wsize
>> PAGE_CACHE_SHIFT
;
695 pagevec_init(&pvec
, 0);
697 /* where to start/end? */
698 if (wbc
->range_cyclic
) {
699 start
= mapping
->writeback_index
; /* Start from prev offset */
701 dout(" cyclic, start at %lu\n", start
);
703 start
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
704 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
705 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
708 dout(" not cyclic, %lu to %lu\n", start
, end
);
713 /* find oldest snap context with dirty data */
714 ceph_put_snap_context(snapc
);
716 snapc
= get_oldest_context(inode
, &snap_size
);
718 /* hmm, why does writepages get called when there
720 dout(" no snap context with dirty data?\n");
724 snap_size
= i_size_read(inode
);
725 dout(" oldest snapc is %p seq %lld (%d snaps)\n",
726 snapc
, snapc
->seq
, snapc
->num_snaps
);
728 spin_lock(&ci
->i_ceph_lock
);
729 truncate_seq
= ci
->i_truncate_seq
;
730 truncate_size
= ci
->i_truncate_size
;
732 snap_size
= i_size_read(inode
);
733 spin_unlock(&ci
->i_ceph_lock
);
735 if (last_snapc
&& snapc
!= last_snapc
) {
736 /* if we switched to a newer snapc, restart our scan at the
737 * start of the original file range. */
738 dout(" snapc differs from last pass, restarting at %lu\n",
744 while (!done
&& index
<= end
) {
745 int num_ops
= do_sync
? 2 : 1;
749 int pvec_pages
, locked_pages
;
750 struct page
**pages
= NULL
;
751 mempool_t
*pool
= NULL
; /* Becomes non-null if mempool used */
759 max_pages
= max_pages_ever
;
763 want
= min(end
- index
,
764 min((pgoff_t
)PAGEVEC_SIZE
,
765 max_pages
- (pgoff_t
)locked_pages
) - 1)
767 pvec_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
770 dout("pagevec_lookup_tag got %d\n", pvec_pages
);
771 if (!pvec_pages
&& !locked_pages
)
773 for (i
= 0; i
< pvec_pages
&& locked_pages
< max_pages
; i
++) {
774 page
= pvec
.pages
[i
];
775 dout("? %p idx %lu\n", page
, page
->index
);
776 if (locked_pages
== 0)
777 lock_page(page
); /* first page */
778 else if (!trylock_page(page
))
781 /* only dirty pages, or our accounting breaks */
782 if (unlikely(!PageDirty(page
)) ||
783 unlikely(page
->mapping
!= mapping
)) {
784 dout("!dirty or !mapping %p\n", page
);
788 if (!wbc
->range_cyclic
&& page
->index
> end
) {
789 dout("end of range %p\n", page
);
794 if (next
&& (page
->index
!= next
)) {
795 dout("not consecutive %p\n", page
);
799 if (wbc
->sync_mode
!= WB_SYNC_NONE
) {
800 dout("waiting on writeback %p\n", page
);
801 wait_on_page_writeback(page
);
803 if (page_offset(page
) >= snap_size
) {
804 dout("%p page eof %llu\n", page
, snap_size
);
809 if (PageWriteback(page
)) {
810 dout("%p under writeback\n", page
);
815 /* only if matching snap context */
816 pgsnapc
= page_snap_context(page
);
817 if (pgsnapc
->seq
> snapc
->seq
) {
818 dout("page snapc %p %lld > oldest %p %lld\n",
819 pgsnapc
, pgsnapc
->seq
, snapc
, snapc
->seq
);
822 continue; /* keep looking for snap */
826 if (!clear_page_dirty_for_io(page
)) {
827 dout("%p !clear_page_dirty_for_io\n", page
);
833 * We have something to write. If this is
834 * the first locked page this time through,
835 * allocate an osd request and a page array
838 if (locked_pages
== 0) {
840 /* prepare async write request */
841 offset
= (u64
)page_offset(page
);
843 req
= ceph_osdc_new_request(&fsc
->client
->osdc
,
845 offset
, &len
, num_ops
,
847 CEPH_OSD_FLAG_WRITE
|
848 CEPH_OSD_FLAG_ONDISK
,
850 truncate_size
, true);
857 req
->r_callback
= writepages_finish
;
858 req
->r_inode
= inode
;
860 max_pages
= calc_pages_for(0, (u64
)len
);
861 pages
= kmalloc(max_pages
* sizeof (*pages
),
864 pool
= fsc
->wb_pagevec_pool
;
865 pages
= mempool_alloc(pool
, GFP_NOFS
);
870 /* note position of first page in pvec */
873 dout("%p will write page %p idx %lu\n",
874 inode
, page
, page
->index
);
877 atomic_long_inc_return(&fsc
->writeback_count
);
878 if (writeback_stat
> CONGESTION_ON_THRESH(
879 fsc
->mount_options
->congestion_kb
)) {
880 set_bdi_congested(&fsc
->backing_dev_info
,
884 set_page_writeback(page
);
885 pages
[locked_pages
] = page
;
887 next
= page
->index
+ 1;
890 /* did we get anything? */
892 goto release_pvec_pages
;
895 BUG_ON(!locked_pages
|| first
< 0);
897 if (pvec_pages
&& i
== pvec_pages
&&
898 locked_pages
< max_pages
) {
899 dout("reached end pvec, trying for more\n");
900 pagevec_reinit(&pvec
);
904 /* shift unused pages over in the pvec... we
905 * will need to release them below. */
906 for (j
= i
; j
< pvec_pages
; j
++) {
907 dout(" pvec leftover page %p\n",
909 pvec
.pages
[j
-i
+first
] = pvec
.pages
[j
];
914 /* Format the osd request message and submit the write */
916 offset
= page_offset(pages
[0]);
917 len
= min(snap_size
- offset
,
918 (u64
)locked_pages
<< PAGE_CACHE_SHIFT
);
919 dout("writepages got %d pages at %llu~%llu\n",
920 locked_pages
, offset
, len
);
922 osd_req_op_extent_osd_data_pages(req
, 0, pages
, len
, 0,
925 pages
= NULL
; /* request message now owns the pages array */
928 /* Update the write op length in case we changed it */
930 osd_req_op_extent_update(req
, 0, len
);
932 vino
= ceph_vino(inode
);
933 ceph_osdc_build_request(req
, offset
, snapc
, vino
.snap
,
936 rc
= ceph_osdc_start_request(&fsc
->client
->osdc
, req
, true);
942 wbc
->nr_to_write
-= locked_pages
;
943 if (wbc
->nr_to_write
<= 0)
947 dout("pagevec_release on %d pages (%p)\n", (int)pvec
.nr
,
948 pvec
.nr
? pvec
.pages
[0] : NULL
);
949 pagevec_release(&pvec
);
951 if (locked_pages
&& !done
)
955 if (should_loop
&& !done
) {
956 /* more to do; loop back to beginning of file */
957 dout("writepages looping back to beginning of file\n");
963 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
964 mapping
->writeback_index
= index
;
968 ceph_osdc_put_request(req
);
969 ceph_put_snap_context(snapc
);
970 dout("writepages done, rc = %d\n", rc
);
977 * See if a given @snapc is either writeable, or already written.
979 static int context_is_writeable_or_written(struct inode
*inode
,
980 struct ceph_snap_context
*snapc
)
982 struct ceph_snap_context
*oldest
= get_oldest_context(inode
, NULL
);
983 int ret
= !oldest
|| snapc
->seq
<= oldest
->seq
;
985 ceph_put_snap_context(oldest
);
990 * We are only allowed to write into/dirty the page if the page is
991 * clean, or already dirty within the same snap context.
993 * called with page locked.
994 * return success with page locked,
995 * or any failure (incl -EAGAIN) with page unlocked.
997 static int ceph_update_writeable_page(struct file
*file
,
998 loff_t pos
, unsigned len
,
1001 struct inode
*inode
= file_inode(file
);
1002 struct ceph_inode_info
*ci
= ceph_inode(inode
);
1003 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
1004 loff_t page_off
= pos
& PAGE_CACHE_MASK
;
1005 int pos_in_page
= pos
& ~PAGE_CACHE_MASK
;
1006 int end_in_page
= pos_in_page
+ len
;
1009 struct ceph_snap_context
*snapc
, *oldest
;
1012 /* writepages currently holds page lock, but if we change that later, */
1013 wait_on_page_writeback(page
);
1015 /* check snap context */
1016 BUG_ON(!ci
->i_snap_realm
);
1017 down_read(&mdsc
->snap_rwsem
);
1018 BUG_ON(!ci
->i_snap_realm
->cached_context
);
1019 snapc
= page_snap_context(page
);
1020 if (snapc
&& snapc
!= ci
->i_head_snapc
) {
1022 * this page is already dirty in another (older) snap
1023 * context! is it writeable now?
1025 oldest
= get_oldest_context(inode
, NULL
);
1026 up_read(&mdsc
->snap_rwsem
);
1028 if (snapc
->seq
> oldest
->seq
) {
1029 ceph_put_snap_context(oldest
);
1030 dout(" page %p snapc %p not current or oldest\n",
1033 * queue for writeback, and wait for snapc to
1034 * be writeable or written
1036 snapc
= ceph_get_snap_context(snapc
);
1038 ceph_queue_writeback(inode
);
1039 r
= wait_event_interruptible(ci
->i_cap_wq
,
1040 context_is_writeable_or_written(inode
, snapc
));
1041 ceph_put_snap_context(snapc
);
1042 if (r
== -ERESTARTSYS
)
1046 ceph_put_snap_context(oldest
);
1048 /* yay, writeable, do it now (without dropping page lock) */
1049 dout(" page %p snapc %p not current, but oldest\n",
1051 if (!clear_page_dirty_for_io(page
))
1053 r
= writepage_nounlock(page
, NULL
);
1059 if (PageUptodate(page
)) {
1060 dout(" page %p already uptodate\n", page
);
1065 if (pos_in_page
== 0 && len
== PAGE_CACHE_SIZE
)
1068 /* past end of file? */
1069 i_size
= inode
->i_size
; /* caller holds i_mutex */
1071 if (i_size
+ len
> inode
->i_sb
->s_maxbytes
) {
1072 /* file is too big */
1077 if (page_off
>= i_size
||
1078 (pos_in_page
== 0 && (pos
+len
) >= i_size
&&
1079 end_in_page
- pos_in_page
!= PAGE_CACHE_SIZE
)) {
1080 dout(" zeroing %p 0 - %d and %d - %d\n",
1081 page
, pos_in_page
, end_in_page
, (int)PAGE_CACHE_SIZE
);
1082 zero_user_segments(page
,
1084 end_in_page
, PAGE_CACHE_SIZE
);
1088 /* we need to read it. */
1089 up_read(&mdsc
->snap_rwsem
);
1090 r
= readpage_nounlock(file
, page
);
1096 up_read(&mdsc
->snap_rwsem
);
1103 * We are only allowed to write into/dirty the page if the page is
1104 * clean, or already dirty within the same snap context.
1106 static int ceph_write_begin(struct file
*file
, struct address_space
*mapping
,
1107 loff_t pos
, unsigned len
, unsigned flags
,
1108 struct page
**pagep
, void **fsdata
)
1110 struct inode
*inode
= file_inode(file
);
1112 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
1117 page
= grab_cache_page_write_begin(mapping
, index
, 0);
1122 dout("write_begin file %p inode %p page %p %d~%d\n", file
,
1123 inode
, page
, (int)pos
, (int)len
);
1125 r
= ceph_update_writeable_page(file
, pos
, len
, page
);
1126 } while (r
== -EAGAIN
);
1132 * we don't do anything in here that simple_write_end doesn't do
1133 * except adjust dirty page accounting and drop read lock on
1136 static int ceph_write_end(struct file
*file
, struct address_space
*mapping
,
1137 loff_t pos
, unsigned len
, unsigned copied
,
1138 struct page
*page
, void *fsdata
)
1140 struct inode
*inode
= file_inode(file
);
1141 struct ceph_fs_client
*fsc
= ceph_inode_to_client(inode
);
1142 struct ceph_mds_client
*mdsc
= fsc
->mdsc
;
1143 unsigned from
= pos
& (PAGE_CACHE_SIZE
- 1);
1146 dout("write_end file %p inode %p page %p %d~%d (%d)\n", file
,
1147 inode
, page
, (int)pos
, (int)copied
, (int)len
);
1149 /* zero the stale part of the page if we did a short copy */
1151 zero_user_segment(page
, from
+copied
, len
);
1153 /* did file size increase? */
1154 /* (no need for i_size_read(); we caller holds i_mutex */
1155 if (pos
+copied
> inode
->i_size
)
1156 check_cap
= ceph_inode_set_size(inode
, pos
+copied
);
1158 if (!PageUptodate(page
))
1159 SetPageUptodate(page
);
1161 set_page_dirty(page
);
1164 up_read(&mdsc
->snap_rwsem
);
1165 page_cache_release(page
);
1168 ceph_check_caps(ceph_inode(inode
), CHECK_CAPS_AUTHONLY
, NULL
);
1174 * we set .direct_IO to indicate direct io is supported, but since we
1175 * intercept O_DIRECT reads and writes early, this function should
1178 static ssize_t
ceph_direct_io(int rw
, struct kiocb
*iocb
,
1179 const struct iovec
*iov
,
1180 loff_t pos
, unsigned long nr_segs
)
1186 const struct address_space_operations ceph_aops
= {
1187 .readpage
= ceph_readpage
,
1188 .readpages
= ceph_readpages
,
1189 .writepage
= ceph_writepage
,
1190 .writepages
= ceph_writepages_start
,
1191 .write_begin
= ceph_write_begin
,
1192 .write_end
= ceph_write_end
,
1193 .set_page_dirty
= ceph_set_page_dirty
,
1194 .invalidatepage
= ceph_invalidatepage
,
1195 .releasepage
= ceph_releasepage
,
1196 .direct_IO
= ceph_direct_io
,
1205 * Reuse write_begin here for simplicity.
1207 static int ceph_page_mkwrite(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
1209 struct inode
*inode
= file_inode(vma
->vm_file
);
1210 struct page
*page
= vmf
->page
;
1211 struct ceph_mds_client
*mdsc
= ceph_inode_to_client(inode
)->mdsc
;
1212 loff_t off
= page_offset(page
);
1216 /* Update time before taking page lock */
1217 file_update_time(vma
->vm_file
);
1219 size
= i_size_read(inode
);
1220 if (off
+ PAGE_CACHE_SIZE
<= size
)
1221 len
= PAGE_CACHE_SIZE
;
1223 len
= size
& ~PAGE_CACHE_MASK
;
1225 dout("page_mkwrite %p %llu~%llu page %p idx %lu\n", inode
,
1226 off
, len
, page
, page
->index
);
1230 ret
= VM_FAULT_NOPAGE
;
1232 (page
->mapping
!= inode
->i_mapping
))
1235 ret
= ceph_update_writeable_page(vma
->vm_file
, off
, len
, page
);
1237 /* success. we'll keep the page locked. */
1238 set_page_dirty(page
);
1239 up_read(&mdsc
->snap_rwsem
);
1240 ret
= VM_FAULT_LOCKED
;
1245 ret
= VM_FAULT_SIGBUS
;
1248 dout("page_mkwrite %p %llu~%llu = %d\n", inode
, off
, len
, ret
);
1249 if (ret
!= VM_FAULT_LOCKED
)
1254 static struct vm_operations_struct ceph_vmops
= {
1255 .fault
= filemap_fault
,
1256 .page_mkwrite
= ceph_page_mkwrite
,
1257 .remap_pages
= generic_file_remap_pages
,
1260 int ceph_mmap(struct file
*file
, struct vm_area_struct
*vma
)
1262 struct address_space
*mapping
= file
->f_mapping
;
1264 if (!mapping
->a_ops
->readpage
)
1266 file_accessed(file
);
1267 vma
->vm_ops
= &ceph_vmops
;