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 BUG_ON(req->r_data.type != CEPH_OSD_DATA_TYPE_PAGES);
247 for (i = 0; i < req->r_data.num_pages; i++, bytes -= PAGE_CACHE_SIZE) {
248 struct page *page = req->r_data.pages[i];
250 if (bytes < (int)PAGE_CACHE_SIZE) {
251 /* zero (remainder of) page */
252 int s = bytes < 0 ? 0 : bytes;
253 zero_user_segment(page, s, PAGE_CACHE_SIZE);
255 dout("finish_read %p uptodate %p idx %lu\n", inode, page,
257 flush_dcache_page(page);
258 SetPageUptodate(page);
260 page_cache_release(page);
262 kfree(req->r_data.pages);
265 static void ceph_unlock_page_vector(struct page **pages, int num_pages)
269 for (i = 0; i < num_pages; i++)
270 unlock_page(pages[i]);
274 * start an async read(ahead) operation. return nr_pages we submitted
275 * a read for on success, or negative error code.
277 static int start_read(struct inode *inode, struct list_head *page_list, int max)
279 struct ceph_osd_client *osdc =
280 &ceph_inode_to_client(inode)->client->osdc;
281 struct ceph_inode_info *ci = ceph_inode(inode);
282 struct page *page = list_entry(page_list->prev, struct page, lru);
283 struct ceph_osd_request *req;
292 off = (u64) page_offset(page);
295 next_index = page->index;
296 list_for_each_entry_reverse(page, page_list, lru) {
297 if (page->index != next_index)
301 if (max && nr_pages == max)
304 len = nr_pages << PAGE_CACHE_SHIFT;
305 dout("start_read %p nr_pages %d is %lld~%lld\n", inode, nr_pages,
308 req = ceph_osdc_new_request(osdc, &ci->i_layout, ceph_vino(inode),
310 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ,
312 ci->i_truncate_seq, ci->i_truncate_size,
317 /* build page vector */
318 nr_pages = calc_pages_for(0, len);
319 pages = kmalloc(sizeof(*pages) * nr_pages, GFP_NOFS);
323 for (i = 0; i < nr_pages; ++i) {
324 page = list_entry(page_list->prev, struct page, lru);
325 BUG_ON(PageLocked(page));
326 list_del(&page->lru);
328 dout("start_read %p adding %p idx %lu\n", inode, page,
330 if (add_to_page_cache_lru(page, &inode->i_data, page->index,
332 page_cache_release(page);
333 dout("start_read %p add_to_page_cache failed %p\n",
340 req->r_data.type = CEPH_OSD_DATA_TYPE_PAGES;
341 req->r_data.pages = pages;
342 req->r_data.num_pages = nr_pages;
343 req->r_data.alignment = 0;
344 req->r_callback = finish_read;
345 req->r_inode = inode;
347 dout("start_read %p starting %p %lld~%lld\n", inode, req, off, len);
348 ret = ceph_osdc_start_request(osdc, req, false);
351 ceph_osdc_put_request(req);
355 ceph_unlock_page_vector(pages, nr_pages);
356 ceph_release_page_vector(pages, nr_pages);
358 ceph_osdc_put_request(req);
364 * Read multiple pages. Leave pages we don't read + unlock in page_list;
365 * the caller (VM) cleans them up.
367 static int ceph_readpages(struct file *file, struct address_space *mapping,
368 struct list_head *page_list, unsigned nr_pages)
370 struct inode *inode = file_inode(file);
371 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
375 if (fsc->mount_options->rsize >= PAGE_CACHE_SIZE)
376 max = (fsc->mount_options->rsize + PAGE_CACHE_SIZE - 1)
379 dout("readpages %p file %p nr_pages %d max %d\n", inode,
382 while (!list_empty(page_list)) {
383 rc = start_read(inode, page_list, max);
389 dout("readpages %p file %p ret %d\n", inode, file, rc);
394 * Get ref for the oldest snapc for an inode with dirty data... that is, the
395 * only snap context we are allowed to write back.
397 static struct ceph_snap_context *get_oldest_context(struct inode *inode,
400 struct ceph_inode_info *ci = ceph_inode(inode);
401 struct ceph_snap_context *snapc = NULL;
402 struct ceph_cap_snap *capsnap = NULL;
404 spin_lock(&ci->i_ceph_lock);
405 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
406 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap,
407 capsnap->context, capsnap->dirty_pages);
408 if (capsnap->dirty_pages) {
409 snapc = ceph_get_snap_context(capsnap->context);
411 *snap_size = capsnap->size;
415 if (!snapc && ci->i_wrbuffer_ref_head) {
416 snapc = ceph_get_snap_context(ci->i_head_snapc);
417 dout(" head snapc %p has %d dirty pages\n",
418 snapc, ci->i_wrbuffer_ref_head);
420 spin_unlock(&ci->i_ceph_lock);
425 * Write a single page, but leave the page locked.
427 * If we get a write error, set the page error bit, but still adjust the
428 * dirty page accounting (i.e., page is no longer dirty).
430 static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
433 struct ceph_inode_info *ci;
434 struct ceph_fs_client *fsc;
435 struct ceph_osd_client *osdc;
436 loff_t page_off = page_offset(page);
437 int len = PAGE_CACHE_SIZE;
440 struct ceph_snap_context *snapc, *oldest;
444 dout("writepage %p idx %lu\n", page, page->index);
446 if (!page->mapping || !page->mapping->host) {
447 dout("writepage %p - no mapping\n", page);
450 inode = page->mapping->host;
451 ci = ceph_inode(inode);
452 fsc = ceph_inode_to_client(inode);
453 osdc = &fsc->client->osdc;
455 /* verify this is a writeable snap context */
456 snapc = page_snap_context(page);
458 dout("writepage %p page %p not dirty?\n", inode, page);
461 oldest = get_oldest_context(inode, &snap_size);
462 if (snapc->seq > oldest->seq) {
463 dout("writepage %p page %p snapc %p not writeable - noop\n",
465 /* we should only noop if called by kswapd */
466 WARN_ON((current->flags & PF_MEMALLOC) == 0);
467 ceph_put_snap_context(oldest);
470 ceph_put_snap_context(oldest);
472 /* is this a partial page at end of file? */
476 i_size = i_size_read(inode);
477 if (i_size < page_off + len)
478 len = i_size - page_off;
480 dout("writepage %p page %p index %lu on %llu~%u snapc %p\n",
481 inode, page, page->index, page_off, len, snapc);
483 writeback_stat = atomic_long_inc_return(&fsc->writeback_count);
485 CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb))
486 set_bdi_congested(&fsc->backing_dev_info, BLK_RW_ASYNC);
488 set_page_writeback(page);
489 err = ceph_osdc_writepages(osdc, ceph_vino(inode),
490 &ci->i_layout, snapc,
492 ci->i_truncate_seq, ci->i_truncate_size,
493 &inode->i_mtime, &page, 1);
495 dout("writepage setting page/mapping error %d %p\n", err, page);
497 mapping_set_error(&inode->i_data, err);
499 wbc->pages_skipped++;
501 dout("writepage cleaned page %p\n", page);
502 err = 0; /* vfs expects us to return 0 */
505 ClearPagePrivate(page);
506 end_page_writeback(page);
507 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
508 ceph_put_snap_context(snapc); /* page's reference */
513 static int ceph_writepage(struct page *page, struct writeback_control *wbc)
516 struct inode *inode = page->mapping->host;
519 err = writepage_nounlock(page, wbc);
527 * lame release_pages helper. release_pages() isn't exported to
530 static void ceph_release_pages(struct page **pages, int num)
535 pagevec_init(&pvec, 0);
536 for (i = 0; i < num; i++) {
537 if (pagevec_add(&pvec, pages[i]) == 0)
538 pagevec_release(&pvec);
540 pagevec_release(&pvec);
545 * async writeback completion handler.
547 * If we get an error, set the mapping error bit, but not the individual
550 static void writepages_finish(struct ceph_osd_request *req,
551 struct ceph_msg *msg)
553 struct inode *inode = req->r_inode;
554 struct ceph_inode_info *ci = ceph_inode(inode);
558 struct ceph_snap_context *snapc = req->r_snapc;
559 struct address_space *mapping = inode->i_mapping;
560 int rc = req->r_result;
561 u64 bytes = le64_to_cpu(req->r_request_ops[0].extent.length);
562 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
564 unsigned issued = ceph_caps_issued(ci);
566 BUG_ON(req->r_data.type != CEPH_OSD_DATA_TYPE_PAGES);
569 * Assume we wrote the pages we originally sent. The
570 * osd might reply with fewer pages if our writeback
571 * raced with a truncation and was adjusted at the osd,
572 * so don't believe the reply.
574 wrote = req->r_data.num_pages;
577 mapping_set_error(mapping, rc);
579 dout("writepages_finish %p rc %d bytes %llu wrote %d (pages)\n",
580 inode, rc, bytes, wrote);
582 /* clean all pages */
583 for (i = 0; i < req->r_data.num_pages; i++) {
584 page = req->r_data.pages[i];
586 WARN_ON(!PageUptodate(page));
589 atomic_long_dec_return(&fsc->writeback_count);
591 CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb))
592 clear_bdi_congested(&fsc->backing_dev_info,
595 ceph_put_snap_context(page_snap_context(page));
597 ClearPagePrivate(page);
598 dout("unlocking %d %p\n", i, page);
599 end_page_writeback(page);
602 * We lost the cache cap, need to truncate the page before
603 * it is unlocked, otherwise we'd truncate it later in the
604 * page truncation thread, possibly losing some data that
607 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0)
608 generic_error_remove_page(inode->i_mapping, page);
612 dout("%p wrote+cleaned %d pages\n", inode, wrote);
613 ceph_put_wrbuffer_cap_refs(ci, req->r_data.num_pages, snapc);
615 ceph_release_pages(req->r_data.pages, req->r_data.num_pages);
616 if (req->r_data.pages_from_pool)
617 mempool_free(req->r_data.pages,
618 ceph_sb_to_client(inode->i_sb)->wb_pagevec_pool);
620 kfree(req->r_data.pages);
621 ceph_osdc_put_request(req);
625 * allocate a page vec, either directly, or if necessary, via a the
626 * mempool. we avoid the mempool if we can because req->r_data.num_pages
627 * may be less than the maximum write size.
629 static void alloc_page_vec(struct ceph_fs_client *fsc,
630 struct ceph_osd_request *req)
632 req->r_data.pages = kmalloc(sizeof(struct page *) * req->r_data.num_pages,
634 if (!req->r_data.pages) {
635 req->r_data.pages = mempool_alloc(fsc->wb_pagevec_pool, GFP_NOFS);
636 req->r_data.pages_from_pool = 1;
637 WARN_ON(!req->r_data.pages);
642 * initiate async writeback
644 static int ceph_writepages_start(struct address_space *mapping,
645 struct writeback_control *wbc)
647 struct inode *inode = mapping->host;
648 struct ceph_inode_info *ci = ceph_inode(inode);
649 struct ceph_fs_client *fsc;
650 pgoff_t index, start, end;
653 pgoff_t max_pages = 0, max_pages_ever = 0;
654 struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc;
658 unsigned wsize = 1 << inode->i_blkbits;
659 struct ceph_osd_request *req = NULL;
664 * Include a 'sync' in the OSD request if this is a data
665 * integrity write (e.g., O_SYNC write or fsync()), or if our
666 * cap is being revoked.
668 do_sync = wbc->sync_mode == WB_SYNC_ALL;
669 if (ceph_caps_revoking(ci, CEPH_CAP_FILE_BUFFER))
671 dout("writepages_start %p dosync=%d (mode=%s)\n",
673 wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
674 (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
676 fsc = ceph_inode_to_client(inode);
677 if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) {
678 pr_warning("writepage_start %p on forced umount\n", inode);
679 return -EIO; /* we're in a forced umount, don't write! */
681 if (fsc->mount_options->wsize && fsc->mount_options->wsize < wsize)
682 wsize = fsc->mount_options->wsize;
683 if (wsize < PAGE_CACHE_SIZE)
684 wsize = PAGE_CACHE_SIZE;
685 max_pages_ever = wsize >> PAGE_CACHE_SHIFT;
687 pagevec_init(&pvec, 0);
689 /* where to start/end? */
690 if (wbc->range_cyclic) {
691 start = mapping->writeback_index; /* Start from prev offset */
693 dout(" cyclic, start at %lu\n", start);
695 start = wbc->range_start >> PAGE_CACHE_SHIFT;
696 end = wbc->range_end >> PAGE_CACHE_SHIFT;
697 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
700 dout(" not cyclic, %lu to %lu\n", start, end);
705 /* find oldest snap context with dirty data */
706 ceph_put_snap_context(snapc);
707 snapc = get_oldest_context(inode, &snap_size);
709 /* hmm, why does writepages get called when there
711 dout(" no snap context with dirty data?\n");
714 dout(" oldest snapc is %p seq %lld (%d snaps)\n",
715 snapc, snapc->seq, snapc->num_snaps);
716 if (last_snapc && snapc != last_snapc) {
717 /* if we switched to a newer snapc, restart our scan at the
718 * start of the original file range. */
719 dout(" snapc differs from last pass, restarting at %lu\n",
725 while (!done && index <= end) {
729 int pvec_pages, locked_pages;
737 max_pages = max_pages_ever;
741 want = min(end - index,
742 min((pgoff_t)PAGEVEC_SIZE,
743 max_pages - (pgoff_t)locked_pages) - 1)
745 pvec_pages = pagevec_lookup_tag(&pvec, mapping, &index,
748 dout("pagevec_lookup_tag got %d\n", pvec_pages);
749 if (!pvec_pages && !locked_pages)
751 for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) {
752 page = pvec.pages[i];
753 dout("? %p idx %lu\n", page, page->index);
754 if (locked_pages == 0)
755 lock_page(page); /* first page */
756 else if (!trylock_page(page))
759 /* only dirty pages, or our accounting breaks */
760 if (unlikely(!PageDirty(page)) ||
761 unlikely(page->mapping != mapping)) {
762 dout("!dirty or !mapping %p\n", page);
766 if (!wbc->range_cyclic && page->index > end) {
767 dout("end of range %p\n", page);
772 if (next && (page->index != next)) {
773 dout("not consecutive %p\n", page);
777 if (wbc->sync_mode != WB_SYNC_NONE) {
778 dout("waiting on writeback %p\n", page);
779 wait_on_page_writeback(page);
781 if ((snap_size && page_offset(page) > snap_size) ||
783 page_offset(page) > i_size_read(inode))) {
784 dout("%p page eof %llu\n", page, snap_size ?
785 snap_size : i_size_read(inode));
790 if (PageWriteback(page)) {
791 dout("%p under writeback\n", page);
796 /* only if matching snap context */
797 pgsnapc = page_snap_context(page);
798 if (pgsnapc->seq > snapc->seq) {
799 dout("page snapc %p %lld > oldest %p %lld\n",
800 pgsnapc, pgsnapc->seq, snapc, snapc->seq);
803 continue; /* keep looking for snap */
807 if (!clear_page_dirty_for_io(page)) {
808 dout("%p !clear_page_dirty_for_io\n", page);
814 if (locked_pages == 0) {
815 /* prepare async write request */
816 offset = (u64) page_offset(page);
818 req = ceph_osdc_new_request(&fsc->client->osdc,
823 CEPH_OSD_FLAG_WRITE |
824 CEPH_OSD_FLAG_ONDISK,
828 &inode->i_mtime, true);
836 req->r_data.type = CEPH_OSD_DATA_TYPE_PAGES;
837 req->r_data.num_pages = calc_pages_for(0, len);
838 req->r_data.alignment = 0;
839 max_pages = req->r_data.num_pages;
841 alloc_page_vec(fsc, req);
842 req->r_callback = writepages_finish;
843 req->r_inode = inode;
846 /* note position of first page in pvec */
849 dout("%p will write page %p idx %lu\n",
850 inode, page, page->index);
853 atomic_long_inc_return(&fsc->writeback_count);
854 if (writeback_stat > CONGESTION_ON_THRESH(
855 fsc->mount_options->congestion_kb)) {
856 set_bdi_congested(&fsc->backing_dev_info,
860 set_page_writeback(page);
861 req->r_data.pages[locked_pages] = page;
863 next = page->index + 1;
866 /* did we get anything? */
868 goto release_pvec_pages;
871 BUG_ON(!locked_pages || first < 0);
873 if (pvec_pages && i == pvec_pages &&
874 locked_pages < max_pages) {
875 dout("reached end pvec, trying for more\n");
876 pagevec_reinit(&pvec);
880 /* shift unused pages over in the pvec... we
881 * will need to release them below. */
882 for (j = i; j < pvec_pages; j++) {
883 dout(" pvec leftover page %p\n",
885 pvec.pages[j-i+first] = pvec.pages[j];
890 /* submit the write */
891 offset = req->r_data.pages[0]->index << PAGE_CACHE_SHIFT;
892 len = min((snap_size ? snap_size : i_size_read(inode)) - offset,
893 (u64)locked_pages << PAGE_CACHE_SHIFT);
894 dout("writepages got %d pages at %llu~%llu\n",
895 locked_pages, offset, len);
897 /* revise final length, page count */
898 req->r_data.num_pages = locked_pages;
899 req->r_request_ops[0].extent.length = cpu_to_le64(len);
900 req->r_request_ops[0].payload_len = cpu_to_le32(len);
901 req->r_request->hdr.data_len = cpu_to_le32(len);
903 rc = ceph_osdc_start_request(&fsc->client->osdc, req, true);
909 wbc->nr_to_write -= locked_pages;
910 if (wbc->nr_to_write <= 0)
914 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr,
915 pvec.nr ? pvec.pages[0] : NULL);
916 pagevec_release(&pvec);
918 if (locked_pages && !done)
922 if (should_loop && !done) {
923 /* more to do; loop back to beginning of file */
924 dout("writepages looping back to beginning of file\n");
930 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
931 mapping->writeback_index = index;
935 ceph_osdc_put_request(req);
936 ceph_put_snap_context(snapc);
937 dout("writepages done, rc = %d\n", rc);
944 * See if a given @snapc is either writeable, or already written.
946 static int context_is_writeable_or_written(struct inode *inode,
947 struct ceph_snap_context *snapc)
949 struct ceph_snap_context *oldest = get_oldest_context(inode, NULL);
950 int ret = !oldest || snapc->seq <= oldest->seq;
952 ceph_put_snap_context(oldest);
957 * We are only allowed to write into/dirty the page if the page is
958 * clean, or already dirty within the same snap context.
960 * called with page locked.
961 * return success with page locked,
962 * or any failure (incl -EAGAIN) with page unlocked.
964 static int ceph_update_writeable_page(struct file *file,
965 loff_t pos, unsigned len,
968 struct inode *inode = file_inode(file);
969 struct ceph_inode_info *ci = ceph_inode(inode);
970 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
971 loff_t page_off = pos & PAGE_CACHE_MASK;
972 int pos_in_page = pos & ~PAGE_CACHE_MASK;
973 int end_in_page = pos_in_page + len;
976 struct ceph_snap_context *snapc, *oldest;
979 /* writepages currently holds page lock, but if we change that later, */
980 wait_on_page_writeback(page);
982 /* check snap context */
983 BUG_ON(!ci->i_snap_realm);
984 down_read(&mdsc->snap_rwsem);
985 BUG_ON(!ci->i_snap_realm->cached_context);
986 snapc = page_snap_context(page);
987 if (snapc && snapc != ci->i_head_snapc) {
989 * this page is already dirty in another (older) snap
990 * context! is it writeable now?
992 oldest = get_oldest_context(inode, NULL);
993 up_read(&mdsc->snap_rwsem);
995 if (snapc->seq > oldest->seq) {
996 ceph_put_snap_context(oldest);
997 dout(" page %p snapc %p not current or oldest\n",
1000 * queue for writeback, and wait for snapc to
1001 * be writeable or written
1003 snapc = ceph_get_snap_context(snapc);
1005 ceph_queue_writeback(inode);
1006 r = wait_event_interruptible(ci->i_cap_wq,
1007 context_is_writeable_or_written(inode, snapc));
1008 ceph_put_snap_context(snapc);
1009 if (r == -ERESTARTSYS)
1013 ceph_put_snap_context(oldest);
1015 /* yay, writeable, do it now (without dropping page lock) */
1016 dout(" page %p snapc %p not current, but oldest\n",
1018 if (!clear_page_dirty_for_io(page))
1020 r = writepage_nounlock(page, NULL);
1026 if (PageUptodate(page)) {
1027 dout(" page %p already uptodate\n", page);
1032 if (pos_in_page == 0 && len == PAGE_CACHE_SIZE)
1035 /* past end of file? */
1036 i_size = inode->i_size; /* caller holds i_mutex */
1038 if (i_size + len > inode->i_sb->s_maxbytes) {
1039 /* file is too big */
1044 if (page_off >= i_size ||
1045 (pos_in_page == 0 && (pos+len) >= i_size &&
1046 end_in_page - pos_in_page != PAGE_CACHE_SIZE)) {
1047 dout(" zeroing %p 0 - %d and %d - %d\n",
1048 page, pos_in_page, end_in_page, (int)PAGE_CACHE_SIZE);
1049 zero_user_segments(page,
1051 end_in_page, PAGE_CACHE_SIZE);
1055 /* we need to read it. */
1056 up_read(&mdsc->snap_rwsem);
1057 r = readpage_nounlock(file, page);
1063 up_read(&mdsc->snap_rwsem);
1070 * We are only allowed to write into/dirty the page if the page is
1071 * clean, or already dirty within the same snap context.
1073 static int ceph_write_begin(struct file *file, struct address_space *mapping,
1074 loff_t pos, unsigned len, unsigned flags,
1075 struct page **pagep, void **fsdata)
1077 struct inode *inode = file_inode(file);
1079 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1084 page = grab_cache_page_write_begin(mapping, index, 0);
1089 dout("write_begin file %p inode %p page %p %d~%d\n", file,
1090 inode, page, (int)pos, (int)len);
1092 r = ceph_update_writeable_page(file, pos, len, page);
1093 } while (r == -EAGAIN);
1099 * we don't do anything in here that simple_write_end doesn't do
1100 * except adjust dirty page accounting and drop read lock on
1103 static int ceph_write_end(struct file *file, struct address_space *mapping,
1104 loff_t pos, unsigned len, unsigned copied,
1105 struct page *page, void *fsdata)
1107 struct inode *inode = file_inode(file);
1108 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1109 struct ceph_mds_client *mdsc = fsc->mdsc;
1110 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1113 dout("write_end file %p inode %p page %p %d~%d (%d)\n", file,
1114 inode, page, (int)pos, (int)copied, (int)len);
1116 /* zero the stale part of the page if we did a short copy */
1118 zero_user_segment(page, from+copied, len);
1120 /* did file size increase? */
1121 /* (no need for i_size_read(); we caller holds i_mutex */
1122 if (pos+copied > inode->i_size)
1123 check_cap = ceph_inode_set_size(inode, pos+copied);
1125 if (!PageUptodate(page))
1126 SetPageUptodate(page);
1128 set_page_dirty(page);
1131 up_read(&mdsc->snap_rwsem);
1132 page_cache_release(page);
1135 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL);
1141 * we set .direct_IO to indicate direct io is supported, but since we
1142 * intercept O_DIRECT reads and writes early, this function should
1145 static ssize_t ceph_direct_io(int rw, struct kiocb *iocb,
1146 const struct iovec *iov,
1147 loff_t pos, unsigned long nr_segs)
1153 const struct address_space_operations ceph_aops = {
1154 .readpage = ceph_readpage,
1155 .readpages = ceph_readpages,
1156 .writepage = ceph_writepage,
1157 .writepages = ceph_writepages_start,
1158 .write_begin = ceph_write_begin,
1159 .write_end = ceph_write_end,
1160 .set_page_dirty = ceph_set_page_dirty,
1161 .invalidatepage = ceph_invalidatepage,
1162 .releasepage = ceph_releasepage,
1163 .direct_IO = ceph_direct_io,
1172 * Reuse write_begin here for simplicity.
1174 static int ceph_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1176 struct inode *inode = file_inode(vma->vm_file);
1177 struct page *page = vmf->page;
1178 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
1179 loff_t off = page_offset(page);
1183 /* Update time before taking page lock */
1184 file_update_time(vma->vm_file);
1186 size = i_size_read(inode);
1187 if (off + PAGE_CACHE_SIZE <= size)
1188 len = PAGE_CACHE_SIZE;
1190 len = size & ~PAGE_CACHE_MASK;
1192 dout("page_mkwrite %p %llu~%llu page %p idx %lu\n", inode,
1193 off, len, page, page->index);
1197 ret = VM_FAULT_NOPAGE;
1199 (page->mapping != inode->i_mapping))
1202 ret = ceph_update_writeable_page(vma->vm_file, off, len, page);
1204 /* success. we'll keep the page locked. */
1205 set_page_dirty(page);
1206 up_read(&mdsc->snap_rwsem);
1207 ret = VM_FAULT_LOCKED;
1212 ret = VM_FAULT_SIGBUS;
1215 dout("page_mkwrite %p %llu~%llu = %d\n", inode, off, len, ret);
1216 if (ret != VM_FAULT_LOCKED)
1221 static struct vm_operations_struct ceph_vmops = {
1222 .fault = filemap_fault,
1223 .page_mkwrite = ceph_page_mkwrite,
1224 .remap_pages = generic_file_remap_pages,
1227 int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1229 struct address_space *mapping = file->f_mapping;
1231 if (!mapping->a_ops->readpage)
1233 file_accessed(file);
1234 vma->vm_ops = &ceph_vmops;