1 #include "ceph_debug.h"
3 #include <linux/backing-dev.h>
6 #include <linux/pagemap.h>
7 #include <linux/writeback.h> /* generic_writepages */
8 #include <linux/pagevec.h>
9 #include <linux/task_io_accounting_ops.h>
12 #include "osd_client.h"
15 * Ceph address space ops.
17 * There are a few funny things going on here.
19 * The page->private field is used to reference a struct
20 * ceph_snap_context for _every_ dirty page. This indicates which
21 * snapshot the page was logically dirtied in, and thus which snap
22 * context needs to be associated with the osd write during writeback.
24 * Similarly, struct ceph_inode_info maintains a set of counters to
25 * count dirty pages on the inode. In the absense of snapshots,
26 * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count.
28 * When a snapshot is taken (that is, when the client receives
29 * notification that a snapshot was taken), each inode with caps and
30 * with dirty pages (dirty pages implies there is a cap) gets a new
31 * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending
32 * order, new snaps go to the tail). The i_wrbuffer_ref_head count is
33 * moved to capsnap->dirty. (Unless a sync write is currently in
34 * progress. In that case, the capsnap is said to be "pending", new
35 * writes cannot start, and the capsnap isn't "finalized" until the
36 * write completes (or fails) and a final size/mtime for the inode for
37 * that snap can be settled upon.) i_wrbuffer_ref_head is reset to 0.
39 * On writeback, we must submit writes to the osd IN SNAP ORDER. So,
40 * we look for the first capsnap in i_cap_snaps and write out pages in
41 * that snap context _only_. Then we move on to the next capsnap,
42 * eventually reaching the "live" or "head" context (i.e., pages that
43 * are not yet snapped) and are writing the most recently dirtied
46 * Invalidate and so forth must take care to ensure the dirty page
47 * accounting is preserved.
50 #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10))
51 #define CONGESTION_OFF_THRESH(congestion_kb) \
52 (CONGESTION_ON_THRESH(congestion_kb) - \
53 (CONGESTION_ON_THRESH(congestion_kb) >> 2))
58 * Dirty a page. Optimistically adjust accounting, on the assumption
59 * that we won't race with invalidate. If we do, readjust.
61 static int ceph_set_page_dirty(struct page *page)
63 struct address_space *mapping = page->mapping;
65 struct ceph_inode_info *ci;
67 struct ceph_snap_context *snapc;
69 if (unlikely(!mapping))
70 return !TestSetPageDirty(page);
72 if (TestSetPageDirty(page)) {
73 dout("%p set_page_dirty %p idx %lu -- already dirty\n",
74 mapping->host, page, page->index);
78 inode = mapping->host;
79 ci = ceph_inode(inode);
82 * Note that we're grabbing a snapc ref here without holding
85 snapc = ceph_get_snap_context(ci->i_snap_realm->cached_context);
88 spin_lock(&inode->i_lock);
89 if (ci->i_wrbuffer_ref_head == 0)
90 ci->i_head_snapc = ceph_get_snap_context(snapc);
91 ++ci->i_wrbuffer_ref_head;
92 if (ci->i_wrbuffer_ref == 0)
95 dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d "
96 "snapc %p seq %lld (%d snaps)\n",
97 mapping->host, page, page->index,
98 ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1,
99 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
100 snapc, snapc->seq, snapc->num_snaps);
101 spin_unlock(&inode->i_lock);
103 /* now adjust page */
104 spin_lock_irq(&mapping->tree_lock);
105 if (page->mapping) { /* Race with truncate? */
106 WARN_ON_ONCE(!PageUptodate(page));
108 if (mapping_cap_account_dirty(mapping)) {
109 __inc_zone_page_state(page, NR_FILE_DIRTY);
110 __inc_bdi_stat(mapping->backing_dev_info,
112 task_io_account_write(PAGE_CACHE_SIZE);
114 radix_tree_tag_set(&mapping->page_tree,
115 page_index(page), PAGECACHE_TAG_DIRTY);
118 * Reference snap context in page->private. Also set
119 * PagePrivate so that we get invalidatepage callback.
121 page->private = (unsigned long)snapc;
122 SetPagePrivate(page);
124 dout("ANON set_page_dirty %p (raced truncate?)\n", page);
128 spin_unlock_irq(&mapping->tree_lock);
131 /* whoops, we failed to dirty the page */
132 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
134 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
136 BUG_ON(!PageDirty(page));
141 * If we are truncating the full page (i.e. offset == 0), adjust the
142 * dirty page counters appropriately. Only called if there is private
145 static void ceph_invalidatepage(struct page *page, unsigned long offset)
147 struct inode *inode = page->mapping->host;
148 struct ceph_inode_info *ci;
149 struct ceph_snap_context *snapc = (void *)page->private;
151 BUG_ON(!PageLocked(page));
152 BUG_ON(!page->private);
153 BUG_ON(!PagePrivate(page));
154 BUG_ON(!page->mapping);
157 * We can get non-dirty pages here due to races between
158 * set_page_dirty and truncate_complete_page; just spit out a
159 * warning, in case we end up with accounting problems later.
161 if (!PageDirty(page))
162 pr_err("%p invalidatepage %p page not dirty\n", inode, page);
165 ClearPageChecked(page);
167 ci = ceph_inode(inode);
169 dout("%p invalidatepage %p idx %lu full dirty page %lu\n",
170 inode, page, page->index, offset);
171 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
172 ceph_put_snap_context(snapc);
174 ClearPagePrivate(page);
176 dout("%p invalidatepage %p idx %lu partial dirty page\n",
177 inode, page, page->index);
181 /* just a sanity check */
182 static int ceph_releasepage(struct page *page, gfp_t g)
184 struct inode *inode = page->mapping ? page->mapping->host : NULL;
185 dout("%p releasepage %p idx %lu\n", inode, page, page->index);
186 WARN_ON(PageDirty(page));
187 WARN_ON(page->private);
188 WARN_ON(PagePrivate(page));
193 * read a single page, without unlocking it.
195 static int readpage_nounlock(struct file *filp, struct page *page)
197 struct inode *inode = filp->f_dentry->d_inode;
198 struct ceph_inode_info *ci = ceph_inode(inode);
199 struct ceph_osd_client *osdc = &ceph_inode_to_client(inode)->osdc;
201 u64 len = PAGE_CACHE_SIZE;
203 dout("readpage inode %p file %p page %p index %lu\n",
204 inode, filp, page, page->index);
205 err = ceph_osdc_readpages(osdc, ceph_vino(inode), &ci->i_layout,
206 page->index << PAGE_CACHE_SHIFT, &len,
207 ci->i_truncate_seq, ci->i_truncate_size,
214 } else if (err < PAGE_CACHE_SIZE) {
215 /* zero fill remainder of page */
216 zero_user_segment(page, err, PAGE_CACHE_SIZE);
218 SetPageUptodate(page);
221 return err < 0 ? err : 0;
224 static int ceph_readpage(struct file *filp, struct page *page)
226 int r = readpage_nounlock(filp, page);
232 * Build a vector of contiguous pages from the provided page list.
234 static struct page **page_vector_from_list(struct list_head *page_list,
239 int next_index, contig_pages = 0;
241 /* build page vector */
242 pages = kmalloc(sizeof(*pages) * *nr_pages, GFP_NOFS);
244 return ERR_PTR(-ENOMEM);
246 BUG_ON(list_empty(page_list));
247 next_index = list_entry(page_list->prev, struct page, lru)->index;
248 list_for_each_entry_reverse(page, page_list, lru) {
249 if (page->index == next_index) {
250 dout("readpages page %d %p\n", contig_pages, page);
251 pages[contig_pages] = page;
258 *nr_pages = contig_pages;
263 * Read multiple pages. Leave pages we don't read + unlock in page_list;
264 * the caller (VM) cleans them up.
266 static int ceph_readpages(struct file *file, struct address_space *mapping,
267 struct list_head *page_list, unsigned nr_pages)
269 struct inode *inode = file->f_dentry->d_inode;
270 struct ceph_inode_info *ci = ceph_inode(inode);
271 struct ceph_osd_client *osdc = &ceph_inode_to_client(inode)->osdc;
278 dout("readpages %p file %p nr_pages %d\n",
279 inode, file, nr_pages);
281 pages = page_vector_from_list(page_list, &nr_pages);
283 return PTR_ERR(pages);
285 /* guess read extent */
286 offset = pages[0]->index << PAGE_CACHE_SHIFT;
287 len = nr_pages << PAGE_CACHE_SHIFT;
288 rc = ceph_osdc_readpages(osdc, ceph_vino(inode), &ci->i_layout,
290 ci->i_truncate_seq, ci->i_truncate_size,
297 /* set uptodate and add to lru in pagevec-sized chunks */
298 pagevec_init(&pvec, 0);
299 for (; !list_empty(page_list) && len > 0;
300 rc -= PAGE_CACHE_SIZE, len -= PAGE_CACHE_SIZE) {
302 list_entry(page_list->prev, struct page, lru);
304 list_del(&page->lru);
306 if (rc < (int)PAGE_CACHE_SIZE) {
307 /* zero (remainder of) page */
308 int s = rc < 0 ? 0 : rc;
309 zero_user_segment(page, s, PAGE_CACHE_SIZE);
312 if (add_to_page_cache(page, mapping, page->index, GFP_NOFS)) {
313 page_cache_release(page);
314 dout("readpages %p add_to_page_cache failed %p\n",
318 dout("readpages %p adding %p idx %lu\n", inode, page,
320 flush_dcache_page(page);
321 SetPageUptodate(page);
323 if (pagevec_add(&pvec, page) == 0)
324 pagevec_lru_add_file(&pvec); /* add to lru */
326 pagevec_lru_add_file(&pvec);
335 * Get ref for the oldest snapc for an inode with dirty data... that is, the
336 * only snap context we are allowed to write back.
338 * Caller holds i_lock.
340 static struct ceph_snap_context *__get_oldest_context(struct inode *inode,
343 struct ceph_inode_info *ci = ceph_inode(inode);
344 struct ceph_snap_context *snapc = NULL;
345 struct ceph_cap_snap *capsnap = NULL;
347 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
348 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap,
349 capsnap->context, capsnap->dirty_pages);
350 if (capsnap->dirty_pages) {
351 snapc = ceph_get_snap_context(capsnap->context);
353 *snap_size = capsnap->size;
357 if (!snapc && ci->i_snap_realm) {
358 snapc = ceph_get_snap_context(ci->i_snap_realm->cached_context);
359 dout(" head snapc %p has %d dirty pages\n",
360 snapc, ci->i_wrbuffer_ref_head);
365 static struct ceph_snap_context *get_oldest_context(struct inode *inode,
368 struct ceph_snap_context *snapc = NULL;
370 spin_lock(&inode->i_lock);
371 snapc = __get_oldest_context(inode, snap_size);
372 spin_unlock(&inode->i_lock);
377 * Write a single page, but leave the page locked.
379 * If we get a write error, set the page error bit, but still adjust the
380 * dirty page accounting (i.e., page is no longer dirty).
382 static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
385 struct ceph_inode_info *ci;
386 struct ceph_client *client;
387 struct ceph_osd_client *osdc;
388 loff_t page_off = page->index << PAGE_CACHE_SHIFT;
389 int len = PAGE_CACHE_SIZE;
392 struct ceph_snap_context *snapc;
396 dout("writepage %p idx %lu\n", page, page->index);
398 if (!page->mapping || !page->mapping->host) {
399 dout("writepage %p - no mapping\n", page);
402 inode = page->mapping->host;
403 ci = ceph_inode(inode);
404 client = ceph_inode_to_client(inode);
405 osdc = &client->osdc;
407 /* verify this is a writeable snap context */
408 snapc = (void *)page->private;
410 dout("writepage %p page %p not dirty?\n", inode, page);
413 if (snapc != get_oldest_context(inode, &snap_size)) {
414 dout("writepage %p page %p snapc %p not writeable - noop\n",
415 inode, page, (void *)page->private);
416 /* we should only noop if called by kswapd */
417 WARN_ON((current->flags & PF_MEMALLOC) == 0);
421 /* is this a partial page at end of file? */
425 i_size = i_size_read(inode);
426 if (i_size < page_off + len)
427 len = i_size - page_off;
429 dout("writepage %p page %p index %lu on %llu~%u\n",
430 inode, page, page->index, page_off, len);
432 writeback_stat = atomic_long_inc_return(&client->writeback_count);
434 CONGESTION_ON_THRESH(client->mount_args->congestion_kb))
435 set_bdi_congested(&client->backing_dev_info, BLK_RW_ASYNC);
437 set_page_writeback(page);
438 err = ceph_osdc_writepages(osdc, ceph_vino(inode),
439 &ci->i_layout, snapc,
441 ci->i_truncate_seq, ci->i_truncate_size,
443 &page, 1, 0, 0, true);
445 dout("writepage setting page/mapping error %d %p\n", err, page);
447 mapping_set_error(&inode->i_data, err);
449 wbc->pages_skipped++;
451 dout("writepage cleaned page %p\n", page);
452 err = 0; /* vfs expects us to return 0 */
455 ClearPagePrivate(page);
456 end_page_writeback(page);
457 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
458 ceph_put_snap_context(snapc);
463 static int ceph_writepage(struct page *page, struct writeback_control *wbc)
466 struct inode *inode = page->mapping->host;
469 err = writepage_nounlock(page, wbc);
477 * lame release_pages helper. release_pages() isn't exported to
480 static void ceph_release_pages(struct page **pages, int num)
485 pagevec_init(&pvec, 0);
486 for (i = 0; i < num; i++) {
487 if (pagevec_add(&pvec, pages[i]) == 0)
488 pagevec_release(&pvec);
490 pagevec_release(&pvec);
495 * async writeback completion handler.
497 * If we get an error, set the mapping error bit, but not the individual
500 static void writepages_finish(struct ceph_osd_request *req,
501 struct ceph_msg *msg)
503 struct inode *inode = req->r_inode;
504 struct ceph_osd_reply_head *replyhead;
505 struct ceph_osd_op *op;
506 struct ceph_inode_info *ci = ceph_inode(inode);
510 struct ceph_snap_context *snapc = req->r_snapc;
511 struct address_space *mapping = inode->i_mapping;
512 struct writeback_control *wbc = req->r_wbc;
515 struct ceph_client *client = ceph_inode_to_client(inode);
519 replyhead = msg->front.iov_base;
520 WARN_ON(le32_to_cpu(replyhead->num_ops) == 0);
521 op = (void *)(replyhead + 1);
522 rc = le32_to_cpu(replyhead->result);
523 bytes = le64_to_cpu(op->extent.length);
527 * Assume we wrote the pages we originally sent. The
528 * osd might reply with fewer pages if our writeback
529 * raced with a truncation and was adjusted at the osd,
530 * so don't believe the reply.
532 wrote = req->r_num_pages;
535 mapping_set_error(mapping, rc);
537 dout("writepages_finish %p rc %d bytes %llu wrote %d (pages)\n",
538 inode, rc, bytes, wrote);
540 /* clean all pages */
541 for (i = 0; i < req->r_num_pages; i++) {
542 page = req->r_pages[i];
544 WARN_ON(!PageUptodate(page));
547 atomic_long_dec_return(&client->writeback_count);
549 CONGESTION_OFF_THRESH(client->mount_args->congestion_kb))
550 clear_bdi_congested(&client->backing_dev_info,
554 dout("inode %p skipping page %p\n", inode, page);
555 wbc->pages_skipped++;
558 ClearPagePrivate(page);
559 ceph_put_snap_context(snapc);
560 dout("unlocking %d %p\n", i, page);
561 end_page_writeback(page);
564 dout("%p wrote+cleaned %d pages\n", inode, wrote);
565 ceph_put_wrbuffer_cap_refs(ci, req->r_num_pages, snapc);
567 ceph_release_pages(req->r_pages, req->r_num_pages);
568 if (req->r_pages_from_pool)
569 mempool_free(req->r_pages,
570 ceph_client(inode->i_sb)->wb_pagevec_pool);
573 ceph_osdc_put_request(req);
577 * allocate a page vec, either directly, or if necessary, via a the
578 * mempool. we avoid the mempool if we can because req->r_num_pages
579 * may be less than the maximum write size.
581 static void alloc_page_vec(struct ceph_client *client,
582 struct ceph_osd_request *req)
584 req->r_pages = kmalloc(sizeof(struct page *) * req->r_num_pages,
587 req->r_pages = mempool_alloc(client->wb_pagevec_pool, GFP_NOFS);
588 req->r_pages_from_pool = 1;
589 WARN_ON(!req->r_pages);
594 * initiate async writeback
596 static int ceph_writepages_start(struct address_space *mapping,
597 struct writeback_control *wbc)
599 struct inode *inode = mapping->host;
600 struct backing_dev_info *bdi = mapping->backing_dev_info;
601 struct ceph_inode_info *ci = ceph_inode(inode);
602 struct ceph_client *client;
603 pgoff_t index, start, end;
606 pgoff_t max_pages = 0, max_pages_ever = 0;
607 struct ceph_snap_context *snapc = NULL, *last_snapc = NULL;
611 unsigned wsize = 1 << inode->i_blkbits;
612 struct ceph_osd_request *req = NULL;
617 * Include a 'sync' in the OSD request if this is a data
618 * integrity write (e.g., O_SYNC write or fsync()), or if our
619 * cap is being revoked.
621 do_sync = wbc->sync_mode == WB_SYNC_ALL;
622 if (ceph_caps_revoking(ci, CEPH_CAP_FILE_BUFFER))
624 dout("writepages_start %p dosync=%d (mode=%s)\n",
626 wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
627 (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
629 client = ceph_inode_to_client(inode);
630 if (client->mount_state == CEPH_MOUNT_SHUTDOWN) {
631 pr_warning("writepage_start %p on forced umount\n", inode);
632 return -EIO; /* we're in a forced umount, don't write! */
634 if (client->mount_args->wsize && client->mount_args->wsize < wsize)
635 wsize = client->mount_args->wsize;
636 if (wsize < PAGE_CACHE_SIZE)
637 wsize = PAGE_CACHE_SIZE;
638 max_pages_ever = wsize >> PAGE_CACHE_SHIFT;
640 pagevec_init(&pvec, 0);
643 if (wbc->nonblocking && bdi_write_congested(bdi)) {
644 dout(" writepages congested\n");
645 wbc->encountered_congestion = 1;
649 /* where to start/end? */
650 if (wbc->range_cyclic) {
651 start = mapping->writeback_index; /* Start from prev offset */
653 dout(" cyclic, start at %lu\n", start);
655 start = wbc->range_start >> PAGE_CACHE_SHIFT;
656 end = wbc->range_end >> PAGE_CACHE_SHIFT;
657 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
660 dout(" not cyclic, %lu to %lu\n", start, end);
665 /* find oldest snap context with dirty data */
666 ceph_put_snap_context(snapc);
667 snapc = get_oldest_context(inode, &snap_size);
669 /* hmm, why does writepages get called when there
671 dout(" no snap context with dirty data?\n");
674 dout(" oldest snapc is %p seq %lld (%d snaps)\n",
675 snapc, snapc->seq, snapc->num_snaps);
676 if (last_snapc && snapc != last_snapc) {
677 /* if we switched to a newer snapc, restart our scan at the
678 * start of the original file range. */
679 dout(" snapc differs from last pass, restarting at %lu\n",
685 while (!done && index <= end) {
689 int pvec_pages, locked_pages;
693 struct ceph_osd_request_head *reqhead;
694 struct ceph_osd_op *op;
699 max_pages = max_pages_ever;
703 want = min(end - index,
704 min((pgoff_t)PAGEVEC_SIZE,
705 max_pages - (pgoff_t)locked_pages) - 1)
707 pvec_pages = pagevec_lookup_tag(&pvec, mapping, &index,
710 dout("pagevec_lookup_tag got %d\n", pvec_pages);
711 if (!pvec_pages && !locked_pages)
713 for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) {
714 page = pvec.pages[i];
715 dout("? %p idx %lu\n", page, page->index);
716 if (locked_pages == 0)
717 lock_page(page); /* first page */
718 else if (!trylock_page(page))
721 /* only dirty pages, or our accounting breaks */
722 if (unlikely(!PageDirty(page)) ||
723 unlikely(page->mapping != mapping)) {
724 dout("!dirty or !mapping %p\n", page);
728 if (!wbc->range_cyclic && page->index > end) {
729 dout("end of range %p\n", page);
734 if (next && (page->index != next)) {
735 dout("not consecutive %p\n", page);
739 if (wbc->sync_mode != WB_SYNC_NONE) {
740 dout("waiting on writeback %p\n", page);
741 wait_on_page_writeback(page);
743 if ((snap_size && page_offset(page) > snap_size) ||
745 page_offset(page) > i_size_read(inode))) {
746 dout("%p page eof %llu\n", page, snap_size ?
747 snap_size : i_size_read(inode));
752 if (PageWriteback(page)) {
753 dout("%p under writeback\n", page);
758 /* only if matching snap context */
759 if (snapc != (void *)page->private) {
760 dout("page snapc %p != oldest %p\n",
761 (void *)page->private, snapc);
764 continue; /* keep looking for snap */
768 if (!clear_page_dirty_for_io(page)) {
769 dout("%p !clear_page_dirty_for_io\n", page);
775 if (locked_pages == 0) {
776 /* prepare async write request */
777 offset = page->index << PAGE_CACHE_SHIFT;
779 req = ceph_osdc_new_request(&client->osdc,
784 CEPH_OSD_FLAG_WRITE |
785 CEPH_OSD_FLAG_ONDISK,
789 &inode->i_mtime, true, 1);
790 max_pages = req->r_num_pages;
792 alloc_page_vec(client, req);
793 req->r_callback = writepages_finish;
794 req->r_inode = inode;
798 /* note position of first page in pvec */
801 dout("%p will write page %p idx %lu\n",
802 inode, page, page->index);
804 writeback_stat = atomic_long_inc_return(&client->writeback_count);
805 if (writeback_stat > CONGESTION_ON_THRESH(client->mount_args->congestion_kb)) {
806 set_bdi_congested(&client->backing_dev_info, BLK_RW_ASYNC);
809 set_page_writeback(page);
810 req->r_pages[locked_pages] = page;
812 next = page->index + 1;
815 /* did we get anything? */
817 goto release_pvec_pages;
820 BUG_ON(!locked_pages || first < 0);
822 if (pvec_pages && i == pvec_pages &&
823 locked_pages < max_pages) {
824 dout("reached end pvec, trying for more\n");
825 pagevec_reinit(&pvec);
829 /* shift unused pages over in the pvec... we
830 * will need to release them below. */
831 for (j = i; j < pvec_pages; j++) {
832 dout(" pvec leftover page %p\n",
834 pvec.pages[j-i+first] = pvec.pages[j];
839 /* submit the write */
840 offset = req->r_pages[0]->index << PAGE_CACHE_SHIFT;
841 len = min((snap_size ? snap_size : i_size_read(inode)) - offset,
842 (u64)locked_pages << PAGE_CACHE_SHIFT);
843 dout("writepages got %d pages at %llu~%llu\n",
844 locked_pages, offset, len);
846 /* revise final length, page count */
847 req->r_num_pages = locked_pages;
848 reqhead = req->r_request->front.iov_base;
849 op = (void *)(reqhead + 1);
850 op->extent.length = cpu_to_le64(len);
851 op->payload_len = cpu_to_le32(len);
852 req->r_request->hdr.data_len = cpu_to_le32(len);
854 ceph_osdc_start_request(&client->osdc, req, true);
859 wbc->nr_to_write -= locked_pages;
860 if (wbc->nr_to_write <= 0)
864 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr,
865 pvec.nr ? pvec.pages[0] : NULL);
866 pagevec_release(&pvec);
868 if (locked_pages && !done)
872 if (should_loop && !done) {
873 /* more to do; loop back to beginning of file */
874 dout("writepages looping back to beginning of file\n");
880 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
881 mapping->writeback_index = index;
885 ceph_osdc_put_request(req);
887 rc = 0; /* vfs expects us to return 0 */
888 ceph_put_snap_context(snapc);
889 dout("writepages done, rc = %d\n", rc);
897 * See if a given @snapc is either writeable, or already written.
899 static int context_is_writeable_or_written(struct inode *inode,
900 struct ceph_snap_context *snapc)
902 struct ceph_snap_context *oldest = get_oldest_context(inode, NULL);
903 return !oldest || snapc->seq <= oldest->seq;
907 * We are only allowed to write into/dirty the page if the page is
908 * clean, or already dirty within the same snap context.
910 static int ceph_update_writeable_page(struct file *file,
911 loff_t pos, unsigned len,
914 struct inode *inode = file->f_dentry->d_inode;
915 struct ceph_inode_info *ci = ceph_inode(inode);
916 struct ceph_mds_client *mdsc = &ceph_inode_to_client(inode)->mdsc;
917 loff_t page_off = pos & PAGE_CACHE_MASK;
918 int pos_in_page = pos & ~PAGE_CACHE_MASK;
919 int end_in_page = pos_in_page + len;
921 struct ceph_snap_context *snapc;
925 /* writepages currently holds page lock, but if we change that later, */
926 wait_on_page_writeback(page);
928 /* check snap context */
929 BUG_ON(!ci->i_snap_realm);
930 down_read(&mdsc->snap_rwsem);
931 BUG_ON(!ci->i_snap_realm->cached_context);
933 (void *)page->private != ci->i_snap_realm->cached_context) {
935 * this page is already dirty in another (older) snap
936 * context! is it writeable now?
938 snapc = get_oldest_context(inode, NULL);
939 up_read(&mdsc->snap_rwsem);
941 if (snapc != (void *)page->private) {
942 dout(" page %p snapc %p not current or oldest\n",
943 page, (void *)page->private);
945 * queue for writeback, and wait for snapc to
946 * be writeable or written
948 snapc = ceph_get_snap_context((void *)page->private);
950 ceph_queue_writeback(inode);
951 wait_event_interruptible(ci->i_cap_wq,
952 context_is_writeable_or_written(inode, snapc));
953 ceph_put_snap_context(snapc);
957 /* yay, writeable, do it now (without dropping page lock) */
958 dout(" page %p snapc %p not current, but oldest\n",
960 if (!clear_page_dirty_for_io(page))
962 r = writepage_nounlock(page, NULL);
968 if (PageUptodate(page)) {
969 dout(" page %p already uptodate\n", page);
974 if (pos_in_page == 0 && len == PAGE_CACHE_SIZE)
977 /* past end of file? */
978 i_size = inode->i_size; /* caller holds i_mutex */
980 if (i_size + len > inode->i_sb->s_maxbytes) {
981 /* file is too big */
986 if (page_off >= i_size ||
987 (pos_in_page == 0 && (pos+len) >= i_size &&
988 end_in_page - pos_in_page != PAGE_CACHE_SIZE)) {
989 dout(" zeroing %p 0 - %d and %d - %d\n",
990 page, pos_in_page, end_in_page, (int)PAGE_CACHE_SIZE);
991 zero_user_segments(page,
993 end_in_page, PAGE_CACHE_SIZE);
997 /* we need to read it. */
998 up_read(&mdsc->snap_rwsem);
999 r = readpage_nounlock(file, page);
1005 up_read(&mdsc->snap_rwsem);
1012 * We are only allowed to write into/dirty the page if the page is
1013 * clean, or already dirty within the same snap context.
1015 static int ceph_write_begin(struct file *file, struct address_space *mapping,
1016 loff_t pos, unsigned len, unsigned flags,
1017 struct page **pagep, void **fsdata)
1019 struct inode *inode = file->f_dentry->d_inode;
1021 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1026 page = grab_cache_page_write_begin(mapping, index, 0);
1031 dout("write_begin file %p inode %p page %p %d~%d\n", file,
1032 inode, page, (int)pos, (int)len);
1034 r = ceph_update_writeable_page(file, pos, len, page);
1035 } while (r == -EAGAIN);
1041 * we don't do anything in here that simple_write_end doesn't do
1042 * except adjust dirty page accounting and drop read lock on
1045 static int ceph_write_end(struct file *file, struct address_space *mapping,
1046 loff_t pos, unsigned len, unsigned copied,
1047 struct page *page, void *fsdata)
1049 struct inode *inode = file->f_dentry->d_inode;
1050 struct ceph_client *client = ceph_inode_to_client(inode);
1051 struct ceph_mds_client *mdsc = &client->mdsc;
1052 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1055 dout("write_end file %p inode %p page %p %d~%d (%d)\n", file,
1056 inode, page, (int)pos, (int)copied, (int)len);
1058 /* zero the stale part of the page if we did a short copy */
1060 zero_user_segment(page, from+copied, len);
1062 /* did file size increase? */
1063 /* (no need for i_size_read(); we caller holds i_mutex */
1064 if (pos+copied > inode->i_size)
1065 check_cap = ceph_inode_set_size(inode, pos+copied);
1067 if (!PageUptodate(page))
1068 SetPageUptodate(page);
1070 set_page_dirty(page);
1073 up_read(&mdsc->snap_rwsem);
1074 page_cache_release(page);
1077 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL);
1083 * we set .direct_IO to indicate direct io is supported, but since we
1084 * intercept O_DIRECT reads and writes early, this function should
1087 static ssize_t ceph_direct_io(int rw, struct kiocb *iocb,
1088 const struct iovec *iov,
1089 loff_t pos, unsigned long nr_segs)
1095 const struct address_space_operations ceph_aops = {
1096 .readpage = ceph_readpage,
1097 .readpages = ceph_readpages,
1098 .writepage = ceph_writepage,
1099 .writepages = ceph_writepages_start,
1100 .write_begin = ceph_write_begin,
1101 .write_end = ceph_write_end,
1102 .set_page_dirty = ceph_set_page_dirty,
1103 .invalidatepage = ceph_invalidatepage,
1104 .releasepage = ceph_releasepage,
1105 .direct_IO = ceph_direct_io,
1114 * Reuse write_begin here for simplicity.
1116 static int ceph_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1118 struct inode *inode = vma->vm_file->f_dentry->d_inode;
1119 struct page *page = vmf->page;
1120 struct ceph_mds_client *mdsc = &ceph_inode_to_client(inode)->mdsc;
1121 loff_t off = page->index << PAGE_CACHE_SHIFT;
1125 size = i_size_read(inode);
1126 if (off + PAGE_CACHE_SIZE <= size)
1127 len = PAGE_CACHE_SIZE;
1129 len = size & ~PAGE_CACHE_MASK;
1131 dout("page_mkwrite %p %llu~%llu page %p idx %lu\n", inode,
1132 off, len, page, page->index);
1136 ret = VM_FAULT_NOPAGE;
1138 (page->mapping != inode->i_mapping))
1141 ret = ceph_update_writeable_page(vma->vm_file, off, len, page);
1143 /* success. we'll keep the page locked. */
1144 set_page_dirty(page);
1145 up_read(&mdsc->snap_rwsem);
1146 ret = VM_FAULT_LOCKED;
1151 ret = VM_FAULT_SIGBUS;
1154 dout("page_mkwrite %p %llu~%llu = %d\n", inode, off, len, ret);
1155 if (ret != VM_FAULT_LOCKED)
1160 static struct vm_operations_struct ceph_vmops = {
1161 .fault = filemap_fault,
1162 .page_mkwrite = ceph_page_mkwrite,
1165 int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1167 struct address_space *mapping = file->f_mapping;
1169 if (!mapping->a_ops->readpage)
1171 file_accessed(file);
1172 vma->vm_ops = &ceph_vmops;
1173 vma->vm_flags |= VM_CAN_NONLINEAR;