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 = file_inode(filp);
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 int rc = req->r_result;
240 int bytes = le32_to_cpu(msg->hdr.data_len);
243 dout("finish_read %p req %p rc %d bytes %d\n", inode, req, rc, bytes);
245 /* unlock all pages, zeroing any data we didn't read */
246 for (i = 0; i < req->r_data.num_pages; i++, bytes -= PAGE_CACHE_SIZE) {
247 struct page *page = req->r_data.pages[i];
249 if (bytes < (int)PAGE_CACHE_SIZE) {
250 /* zero (remainder of) page */
251 int s = bytes < 0 ? 0 : bytes;
252 zero_user_segment(page, s, PAGE_CACHE_SIZE);
254 dout("finish_read %p uptodate %p idx %lu\n", inode, page,
256 flush_dcache_page(page);
257 SetPageUptodate(page);
259 page_cache_release(page);
261 kfree(req->r_data.pages);
264 static void ceph_unlock_page_vector(struct page **pages, int num_pages)
268 for (i = 0; i < num_pages; i++)
269 unlock_page(pages[i]);
273 * start an async read(ahead) operation. return nr_pages we submitted
274 * a read for on success, or negative error code.
276 static int start_read(struct inode *inode, struct list_head *page_list, int max)
278 struct ceph_osd_client *osdc =
279 &ceph_inode_to_client(inode)->client->osdc;
280 struct ceph_inode_info *ci = ceph_inode(inode);
281 struct page *page = list_entry(page_list->prev, struct page, lru);
282 struct ceph_osd_request *req;
291 off = (u64) page_offset(page);
294 next_index = page->index;
295 list_for_each_entry_reverse(page, page_list, lru) {
296 if (page->index != next_index)
300 if (max && nr_pages == max)
303 len = nr_pages << PAGE_CACHE_SHIFT;
304 dout("start_read %p nr_pages %d is %lld~%lld\n", inode, nr_pages,
307 req = ceph_osdc_new_request(osdc, &ci->i_layout, ceph_vino(inode),
309 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
311 ci->i_truncate_seq, ci->i_truncate_size,
316 /* build page vector */
317 nr_pages = calc_pages_for(0, len);
318 pages = kmalloc(sizeof(*pages) * nr_pages, GFP_NOFS);
322 for (i = 0; i < nr_pages; ++i) {
323 page = list_entry(page_list->prev, struct page, lru);
324 BUG_ON(PageLocked(page));
325 list_del(&page->lru);
327 dout("start_read %p adding %p idx %lu\n", inode, page,
329 if (add_to_page_cache_lru(page, &inode->i_data, page->index,
331 page_cache_release(page);
332 dout("start_read %p add_to_page_cache failed %p\n",
339 req->r_data.pages = pages;
340 req->r_data.num_pages = nr_pages;
341 req->r_data.alignment = 0;
342 req->r_callback = finish_read;
343 req->r_inode = inode;
345 dout("start_read %p starting %p %lld~%lld\n", inode, req, off, len);
346 ret = ceph_osdc_start_request(osdc, req, false);
349 ceph_osdc_put_request(req);
353 ceph_unlock_page_vector(pages, nr_pages);
354 ceph_release_page_vector(pages, nr_pages);
356 ceph_osdc_put_request(req);
362 * Read multiple pages. Leave pages we don't read + unlock in page_list;
363 * the caller (VM) cleans them up.
365 static int ceph_readpages(struct file *file, struct address_space *mapping,
366 struct list_head *page_list, unsigned nr_pages)
368 struct inode *inode = file_inode(file);
369 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
373 if (fsc->mount_options->rsize >= PAGE_CACHE_SIZE)
374 max = (fsc->mount_options->rsize + PAGE_CACHE_SIZE - 1)
377 dout("readpages %p file %p nr_pages %d max %d\n", inode,
380 while (!list_empty(page_list)) {
381 rc = start_read(inode, page_list, max);
387 dout("readpages %p file %p ret %d\n", inode, file, rc);
392 * Get ref for the oldest snapc for an inode with dirty data... that is, the
393 * only snap context we are allowed to write back.
395 static struct ceph_snap_context *get_oldest_context(struct inode *inode,
398 struct ceph_inode_info *ci = ceph_inode(inode);
399 struct ceph_snap_context *snapc = NULL;
400 struct ceph_cap_snap *capsnap = NULL;
402 spin_lock(&ci->i_ceph_lock);
403 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
404 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap,
405 capsnap->context, capsnap->dirty_pages);
406 if (capsnap->dirty_pages) {
407 snapc = ceph_get_snap_context(capsnap->context);
409 *snap_size = capsnap->size;
413 if (!snapc && ci->i_wrbuffer_ref_head) {
414 snapc = ceph_get_snap_context(ci->i_head_snapc);
415 dout(" head snapc %p has %d dirty pages\n",
416 snapc, ci->i_wrbuffer_ref_head);
418 spin_unlock(&ci->i_ceph_lock);
423 * Write a single page, but leave the page locked.
425 * If we get a write error, set the page error bit, but still adjust the
426 * dirty page accounting (i.e., page is no longer dirty).
428 static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
431 struct ceph_inode_info *ci;
432 struct ceph_fs_client *fsc;
433 struct ceph_osd_client *osdc;
434 loff_t page_off = page_offset(page);
435 int len = PAGE_CACHE_SIZE;
438 struct ceph_snap_context *snapc, *oldest;
442 dout("writepage %p idx %lu\n", page, page->index);
444 if (!page->mapping || !page->mapping->host) {
445 dout("writepage %p - no mapping\n", page);
448 inode = page->mapping->host;
449 ci = ceph_inode(inode);
450 fsc = ceph_inode_to_client(inode);
451 osdc = &fsc->client->osdc;
453 /* verify this is a writeable snap context */
454 snapc = page_snap_context(page);
456 dout("writepage %p page %p not dirty?\n", inode, page);
459 oldest = get_oldest_context(inode, &snap_size);
460 if (snapc->seq > oldest->seq) {
461 dout("writepage %p page %p snapc %p not writeable - noop\n",
463 /* we should only noop if called by kswapd */
464 WARN_ON((current->flags & PF_MEMALLOC) == 0);
465 ceph_put_snap_context(oldest);
468 ceph_put_snap_context(oldest);
470 /* is this a partial page at end of file? */
474 i_size = i_size_read(inode);
475 if (i_size < page_off + len)
476 len = i_size - page_off;
478 dout("writepage %p page %p index %lu on %llu~%u snapc %p\n",
479 inode, page, page->index, page_off, len, snapc);
481 writeback_stat = atomic_long_inc_return(&fsc->writeback_count);
483 CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb))
484 set_bdi_congested(&fsc->backing_dev_info, BLK_RW_ASYNC);
486 set_page_writeback(page);
487 err = ceph_osdc_writepages(osdc, ceph_vino(inode),
488 &ci->i_layout, snapc,
490 ci->i_truncate_seq, ci->i_truncate_size,
491 &inode->i_mtime, &page, 1);
493 dout("writepage setting page/mapping error %d %p\n", err, page);
495 mapping_set_error(&inode->i_data, err);
497 wbc->pages_skipped++;
499 dout("writepage cleaned page %p\n", page);
500 err = 0; /* vfs expects us to return 0 */
503 ClearPagePrivate(page);
504 end_page_writeback(page);
505 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
506 ceph_put_snap_context(snapc); /* page's reference */
511 static int ceph_writepage(struct page *page, struct writeback_control *wbc)
514 struct inode *inode = page->mapping->host;
517 err = writepage_nounlock(page, wbc);
525 * lame release_pages helper. release_pages() isn't exported to
528 static void ceph_release_pages(struct page **pages, int num)
533 pagevec_init(&pvec, 0);
534 for (i = 0; i < num; i++) {
535 if (pagevec_add(&pvec, pages[i]) == 0)
536 pagevec_release(&pvec);
538 pagevec_release(&pvec);
543 * async writeback completion handler.
545 * If we get an error, set the mapping error bit, but not the individual
548 static void writepages_finish(struct ceph_osd_request *req,
549 struct ceph_msg *msg)
551 struct inode *inode = req->r_inode;
552 struct ceph_inode_info *ci = ceph_inode(inode);
556 struct ceph_snap_context *snapc = req->r_snapc;
557 struct address_space *mapping = inode->i_mapping;
558 int rc = req->r_result;
559 u64 bytes = le64_to_cpu(req->r_request_ops[0].extent.length);
560 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
562 unsigned issued = ceph_caps_issued(ci);
566 * Assume we wrote the pages we originally sent. The
567 * osd might reply with fewer pages if our writeback
568 * raced with a truncation and was adjusted at the osd,
569 * so don't believe the reply.
571 wrote = req->r_data.num_pages;
574 mapping_set_error(mapping, rc);
576 dout("writepages_finish %p rc %d bytes %llu wrote %d (pages)\n",
577 inode, rc, bytes, wrote);
579 /* clean all pages */
580 for (i = 0; i < req->r_data.num_pages; i++) {
581 page = req->r_data.pages[i];
583 WARN_ON(!PageUptodate(page));
586 atomic_long_dec_return(&fsc->writeback_count);
588 CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb))
589 clear_bdi_congested(&fsc->backing_dev_info,
592 ceph_put_snap_context(page_snap_context(page));
594 ClearPagePrivate(page);
595 dout("unlocking %d %p\n", i, page);
596 end_page_writeback(page);
599 * We lost the cache cap, need to truncate the page before
600 * it is unlocked, otherwise we'd truncate it later in the
601 * page truncation thread, possibly losing some data that
604 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0)
605 generic_error_remove_page(inode->i_mapping, page);
609 dout("%p wrote+cleaned %d pages\n", inode, wrote);
610 ceph_put_wrbuffer_cap_refs(ci, req->r_data.num_pages, snapc);
612 ceph_release_pages(req->r_data.pages, req->r_data.num_pages);
613 if (req->r_data.pages_from_pool)
614 mempool_free(req->r_data.pages,
615 ceph_sb_to_client(inode->i_sb)->wb_pagevec_pool);
617 kfree(req->r_data.pages);
618 ceph_osdc_put_request(req);
622 * allocate a page vec, either directly, or if necessary, via a the
623 * mempool. we avoid the mempool if we can because req->r_data.num_pages
624 * may be less than the maximum write size.
626 static void alloc_page_vec(struct ceph_fs_client *fsc,
627 struct ceph_osd_request *req)
629 req->r_data.pages = kmalloc(sizeof(struct page *) * req->r_data.num_pages,
631 if (!req->r_data.pages) {
632 req->r_data.pages = mempool_alloc(fsc->wb_pagevec_pool, GFP_NOFS);
633 req->r_data.pages_from_pool = 1;
634 WARN_ON(!req->r_data.pages);
639 * initiate async writeback
641 static int ceph_writepages_start(struct address_space *mapping,
642 struct writeback_control *wbc)
644 struct inode *inode = mapping->host;
645 struct ceph_inode_info *ci = ceph_inode(inode);
646 struct ceph_fs_client *fsc;
647 pgoff_t index, start, end;
650 pgoff_t max_pages = 0, max_pages_ever = 0;
651 struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc;
655 unsigned wsize = 1 << inode->i_blkbits;
656 struct ceph_osd_request *req = NULL;
661 * Include a 'sync' in the OSD request if this is a data
662 * integrity write (e.g., O_SYNC write or fsync()), or if our
663 * cap is being revoked.
665 do_sync = wbc->sync_mode == WB_SYNC_ALL;
666 if (ceph_caps_revoking(ci, CEPH_CAP_FILE_BUFFER))
668 dout("writepages_start %p dosync=%d (mode=%s)\n",
670 wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
671 (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
673 fsc = ceph_inode_to_client(inode);
674 if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) {
675 pr_warning("writepage_start %p on forced umount\n", inode);
676 return -EIO; /* we're in a forced umount, don't write! */
678 if (fsc->mount_options->wsize && fsc->mount_options->wsize < wsize)
679 wsize = fsc->mount_options->wsize;
680 if (wsize < PAGE_CACHE_SIZE)
681 wsize = PAGE_CACHE_SIZE;
682 max_pages_ever = wsize >> PAGE_CACHE_SHIFT;
684 pagevec_init(&pvec, 0);
686 /* where to start/end? */
687 if (wbc->range_cyclic) {
688 start = mapping->writeback_index; /* Start from prev offset */
690 dout(" cyclic, start at %lu\n", start);
692 start = wbc->range_start >> PAGE_CACHE_SHIFT;
693 end = wbc->range_end >> PAGE_CACHE_SHIFT;
694 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
697 dout(" not cyclic, %lu to %lu\n", start, end);
702 /* find oldest snap context with dirty data */
703 ceph_put_snap_context(snapc);
704 snapc = get_oldest_context(inode, &snap_size);
706 /* hmm, why does writepages get called when there
708 dout(" no snap context with dirty data?\n");
711 dout(" oldest snapc is %p seq %lld (%d snaps)\n",
712 snapc, snapc->seq, snapc->num_snaps);
713 if (last_snapc && snapc != last_snapc) {
714 /* if we switched to a newer snapc, restart our scan at the
715 * start of the original file range. */
716 dout(" snapc differs from last pass, restarting at %lu\n",
722 while (!done && index <= end) {
726 int pvec_pages, locked_pages;
734 max_pages = max_pages_ever;
738 want = min(end - index,
739 min((pgoff_t)PAGEVEC_SIZE,
740 max_pages - (pgoff_t)locked_pages) - 1)
742 pvec_pages = pagevec_lookup_tag(&pvec, mapping, &index,
745 dout("pagevec_lookup_tag got %d\n", pvec_pages);
746 if (!pvec_pages && !locked_pages)
748 for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) {
749 page = pvec.pages[i];
750 dout("? %p idx %lu\n", page, page->index);
751 if (locked_pages == 0)
752 lock_page(page); /* first page */
753 else if (!trylock_page(page))
756 /* only dirty pages, or our accounting breaks */
757 if (unlikely(!PageDirty(page)) ||
758 unlikely(page->mapping != mapping)) {
759 dout("!dirty or !mapping %p\n", page);
763 if (!wbc->range_cyclic && page->index > end) {
764 dout("end of range %p\n", page);
769 if (next && (page->index != next)) {
770 dout("not consecutive %p\n", page);
774 if (wbc->sync_mode != WB_SYNC_NONE) {
775 dout("waiting on writeback %p\n", page);
776 wait_on_page_writeback(page);
778 if ((snap_size && page_offset(page) > snap_size) ||
780 page_offset(page) > i_size_read(inode))) {
781 dout("%p page eof %llu\n", page, snap_size ?
782 snap_size : i_size_read(inode));
787 if (PageWriteback(page)) {
788 dout("%p under writeback\n", page);
793 /* only if matching snap context */
794 pgsnapc = page_snap_context(page);
795 if (pgsnapc->seq > snapc->seq) {
796 dout("page snapc %p %lld > oldest %p %lld\n",
797 pgsnapc, pgsnapc->seq, snapc, snapc->seq);
800 continue; /* keep looking for snap */
804 if (!clear_page_dirty_for_io(page)) {
805 dout("%p !clear_page_dirty_for_io\n", page);
811 if (locked_pages == 0) {
812 /* prepare async write request */
813 offset = (u64) page_offset(page);
815 req = ceph_osdc_new_request(&fsc->client->osdc,
820 CEPH_OSD_FLAG_WRITE |
821 CEPH_OSD_FLAG_ONDISK,
825 &inode->i_mtime, true);
833 req->r_data.num_pages = calc_pages_for(0, len);
834 req->r_data.alignment = 0;
835 max_pages = req->r_data.num_pages;
837 alloc_page_vec(fsc, req);
838 req->r_callback = writepages_finish;
839 req->r_inode = inode;
842 /* note position of first page in pvec */
845 dout("%p will write page %p idx %lu\n",
846 inode, page, page->index);
849 atomic_long_inc_return(&fsc->writeback_count);
850 if (writeback_stat > CONGESTION_ON_THRESH(
851 fsc->mount_options->congestion_kb)) {
852 set_bdi_congested(&fsc->backing_dev_info,
856 set_page_writeback(page);
857 req->r_data.pages[locked_pages] = page;
859 next = page->index + 1;
862 /* did we get anything? */
864 goto release_pvec_pages;
867 BUG_ON(!locked_pages || first < 0);
869 if (pvec_pages && i == pvec_pages &&
870 locked_pages < max_pages) {
871 dout("reached end pvec, trying for more\n");
872 pagevec_reinit(&pvec);
876 /* shift unused pages over in the pvec... we
877 * will need to release them below. */
878 for (j = i; j < pvec_pages; j++) {
879 dout(" pvec leftover page %p\n",
881 pvec.pages[j-i+first] = pvec.pages[j];
886 /* submit the write */
887 offset = req->r_data.pages[0]->index << PAGE_CACHE_SHIFT;
888 len = min((snap_size ? snap_size : i_size_read(inode)) - offset,
889 (u64)locked_pages << PAGE_CACHE_SHIFT);
890 dout("writepages got %d pages at %llu~%llu\n",
891 locked_pages, offset, len);
893 /* revise final length, page count */
894 req->r_data.num_pages = locked_pages;
895 req->r_request_ops[0].extent.length = cpu_to_le64(len);
896 req->r_request_ops[0].payload_len = cpu_to_le32(len);
897 req->r_request->hdr.data_len = cpu_to_le32(len);
899 rc = ceph_osdc_start_request(&fsc->client->osdc, req, true);
905 wbc->nr_to_write -= locked_pages;
906 if (wbc->nr_to_write <= 0)
910 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr,
911 pvec.nr ? pvec.pages[0] : NULL);
912 pagevec_release(&pvec);
914 if (locked_pages && !done)
918 if (should_loop && !done) {
919 /* more to do; loop back to beginning of file */
920 dout("writepages looping back to beginning of file\n");
926 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
927 mapping->writeback_index = index;
931 ceph_osdc_put_request(req);
932 ceph_put_snap_context(snapc);
933 dout("writepages done, rc = %d\n", rc);
940 * See if a given @snapc is either writeable, or already written.
942 static int context_is_writeable_or_written(struct inode *inode,
943 struct ceph_snap_context *snapc)
945 struct ceph_snap_context *oldest = get_oldest_context(inode, NULL);
946 int ret = !oldest || snapc->seq <= oldest->seq;
948 ceph_put_snap_context(oldest);
953 * We are only allowed to write into/dirty the page if the page is
954 * clean, or already dirty within the same snap context.
956 * called with page locked.
957 * return success with page locked,
958 * or any failure (incl -EAGAIN) with page unlocked.
960 static int ceph_update_writeable_page(struct file *file,
961 loff_t pos, unsigned len,
964 struct inode *inode = file_inode(file);
965 struct ceph_inode_info *ci = ceph_inode(inode);
966 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
967 loff_t page_off = pos & PAGE_CACHE_MASK;
968 int pos_in_page = pos & ~PAGE_CACHE_MASK;
969 int end_in_page = pos_in_page + len;
972 struct ceph_snap_context *snapc, *oldest;
975 /* writepages currently holds page lock, but if we change that later, */
976 wait_on_page_writeback(page);
978 /* check snap context */
979 BUG_ON(!ci->i_snap_realm);
980 down_read(&mdsc->snap_rwsem);
981 BUG_ON(!ci->i_snap_realm->cached_context);
982 snapc = page_snap_context(page);
983 if (snapc && snapc != ci->i_head_snapc) {
985 * this page is already dirty in another (older) snap
986 * context! is it writeable now?
988 oldest = get_oldest_context(inode, NULL);
989 up_read(&mdsc->snap_rwsem);
991 if (snapc->seq > oldest->seq) {
992 ceph_put_snap_context(oldest);
993 dout(" page %p snapc %p not current or oldest\n",
996 * queue for writeback, and wait for snapc to
997 * be writeable or written
999 snapc = ceph_get_snap_context(snapc);
1001 ceph_queue_writeback(inode);
1002 r = wait_event_interruptible(ci->i_cap_wq,
1003 context_is_writeable_or_written(inode, snapc));
1004 ceph_put_snap_context(snapc);
1005 if (r == -ERESTARTSYS)
1009 ceph_put_snap_context(oldest);
1011 /* yay, writeable, do it now (without dropping page lock) */
1012 dout(" page %p snapc %p not current, but oldest\n",
1014 if (!clear_page_dirty_for_io(page))
1016 r = writepage_nounlock(page, NULL);
1022 if (PageUptodate(page)) {
1023 dout(" page %p already uptodate\n", page);
1028 if (pos_in_page == 0 && len == PAGE_CACHE_SIZE)
1031 /* past end of file? */
1032 i_size = inode->i_size; /* caller holds i_mutex */
1034 if (i_size + len > inode->i_sb->s_maxbytes) {
1035 /* file is too big */
1040 if (page_off >= i_size ||
1041 (pos_in_page == 0 && (pos+len) >= i_size &&
1042 end_in_page - pos_in_page != PAGE_CACHE_SIZE)) {
1043 dout(" zeroing %p 0 - %d and %d - %d\n",
1044 page, pos_in_page, end_in_page, (int)PAGE_CACHE_SIZE);
1045 zero_user_segments(page,
1047 end_in_page, PAGE_CACHE_SIZE);
1051 /* we need to read it. */
1052 up_read(&mdsc->snap_rwsem);
1053 r = readpage_nounlock(file, page);
1059 up_read(&mdsc->snap_rwsem);
1066 * We are only allowed to write into/dirty the page if the page is
1067 * clean, or already dirty within the same snap context.
1069 static int ceph_write_begin(struct file *file, struct address_space *mapping,
1070 loff_t pos, unsigned len, unsigned flags,
1071 struct page **pagep, void **fsdata)
1073 struct inode *inode = file_inode(file);
1075 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1080 page = grab_cache_page_write_begin(mapping, index, 0);
1085 dout("write_begin file %p inode %p page %p %d~%d\n", file,
1086 inode, page, (int)pos, (int)len);
1088 r = ceph_update_writeable_page(file, pos, len, page);
1089 } while (r == -EAGAIN);
1095 * we don't do anything in here that simple_write_end doesn't do
1096 * except adjust dirty page accounting and drop read lock on
1099 static int ceph_write_end(struct file *file, struct address_space *mapping,
1100 loff_t pos, unsigned len, unsigned copied,
1101 struct page *page, void *fsdata)
1103 struct inode *inode = file_inode(file);
1104 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1105 struct ceph_mds_client *mdsc = fsc->mdsc;
1106 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1109 dout("write_end file %p inode %p page %p %d~%d (%d)\n", file,
1110 inode, page, (int)pos, (int)copied, (int)len);
1112 /* zero the stale part of the page if we did a short copy */
1114 zero_user_segment(page, from+copied, len);
1116 /* did file size increase? */
1117 /* (no need for i_size_read(); we caller holds i_mutex */
1118 if (pos+copied > inode->i_size)
1119 check_cap = ceph_inode_set_size(inode, pos+copied);
1121 if (!PageUptodate(page))
1122 SetPageUptodate(page);
1124 set_page_dirty(page);
1127 up_read(&mdsc->snap_rwsem);
1128 page_cache_release(page);
1131 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL);
1137 * we set .direct_IO to indicate direct io is supported, but since we
1138 * intercept O_DIRECT reads and writes early, this function should
1141 static ssize_t ceph_direct_io(int rw, struct kiocb *iocb,
1142 const struct iovec *iov,
1143 loff_t pos, unsigned long nr_segs)
1149 const struct address_space_operations ceph_aops = {
1150 .readpage = ceph_readpage,
1151 .readpages = ceph_readpages,
1152 .writepage = ceph_writepage,
1153 .writepages = ceph_writepages_start,
1154 .write_begin = ceph_write_begin,
1155 .write_end = ceph_write_end,
1156 .set_page_dirty = ceph_set_page_dirty,
1157 .invalidatepage = ceph_invalidatepage,
1158 .releasepage = ceph_releasepage,
1159 .direct_IO = ceph_direct_io,
1168 * Reuse write_begin here for simplicity.
1170 static int ceph_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1172 struct inode *inode = file_inode(vma->vm_file);
1173 struct page *page = vmf->page;
1174 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
1175 loff_t off = page_offset(page);
1179 /* Update time before taking page lock */
1180 file_update_time(vma->vm_file);
1182 size = i_size_read(inode);
1183 if (off + PAGE_CACHE_SIZE <= size)
1184 len = PAGE_CACHE_SIZE;
1186 len = size & ~PAGE_CACHE_MASK;
1188 dout("page_mkwrite %p %llu~%llu page %p idx %lu\n", inode,
1189 off, len, page, page->index);
1193 ret = VM_FAULT_NOPAGE;
1195 (page->mapping != inode->i_mapping))
1198 ret = ceph_update_writeable_page(vma->vm_file, off, len, page);
1200 /* success. we'll keep the page locked. */
1201 set_page_dirty(page);
1202 up_read(&mdsc->snap_rwsem);
1203 ret = VM_FAULT_LOCKED;
1208 ret = VM_FAULT_SIGBUS;
1211 dout("page_mkwrite %p %llu~%llu = %d\n", inode, off, len, ret);
1212 if (ret != VM_FAULT_LOCKED)
1217 static struct vm_operations_struct ceph_vmops = {
1218 .fault = filemap_fault,
1219 .page_mkwrite = ceph_page_mkwrite,
1220 .remap_pages = generic_file_remap_pages,
1223 int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1225 struct address_space *mapping = file->f_mapping;
1227 if (!mapping->a_ops->readpage)
1229 file_accessed(file);
1230 vma->vm_ops = &ceph_vmops;