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_num_pages; i++, bytes -= PAGE_CACHE_SIZE) {
247 struct page *page = req->r_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);
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 = len >> PAGE_CACHE_SHIFT;
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_pages = pages;
340 req->r_num_pages = nr_pages;
341 req->r_callback = finish_read;
342 req->r_inode = inode;
344 dout("start_read %p starting %p %lld~%lld\n", inode, req, off, len);
345 ret = ceph_osdc_start_request(osdc, req, false);
348 ceph_osdc_put_request(req);
352 ceph_unlock_page_vector(pages, nr_pages);
353 ceph_release_page_vector(pages, nr_pages);
355 ceph_osdc_put_request(req);
361 * Read multiple pages. Leave pages we don't read + unlock in page_list;
362 * the caller (VM) cleans them up.
364 static int ceph_readpages(struct file *file, struct address_space *mapping,
365 struct list_head *page_list, unsigned nr_pages)
367 struct inode *inode = file_inode(file);
368 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
372 if (fsc->mount_options->rsize >= PAGE_CACHE_SIZE)
373 max = (fsc->mount_options->rsize + PAGE_CACHE_SIZE - 1)
376 dout("readpages %p file %p nr_pages %d max %d\n", inode, file, nr_pages,
378 while (!list_empty(page_list)) {
379 rc = start_read(inode, page_list, max);
385 dout("readpages %p file %p ret %d\n", inode, file, rc);
390 * Get ref for the oldest snapc for an inode with dirty data... that is, the
391 * only snap context we are allowed to write back.
393 static struct ceph_snap_context *get_oldest_context(struct inode *inode,
396 struct ceph_inode_info *ci = ceph_inode(inode);
397 struct ceph_snap_context *snapc = NULL;
398 struct ceph_cap_snap *capsnap = NULL;
400 spin_lock(&ci->i_ceph_lock);
401 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
402 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap,
403 capsnap->context, capsnap->dirty_pages);
404 if (capsnap->dirty_pages) {
405 snapc = ceph_get_snap_context(capsnap->context);
407 *snap_size = capsnap->size;
411 if (!snapc && ci->i_wrbuffer_ref_head) {
412 snapc = ceph_get_snap_context(ci->i_head_snapc);
413 dout(" head snapc %p has %d dirty pages\n",
414 snapc, ci->i_wrbuffer_ref_head);
416 spin_unlock(&ci->i_ceph_lock);
421 * Write a single page, but leave the page locked.
423 * If we get a write error, set the page error bit, but still adjust the
424 * dirty page accounting (i.e., page is no longer dirty).
426 static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
429 struct ceph_inode_info *ci;
430 struct ceph_fs_client *fsc;
431 struct ceph_osd_client *osdc;
432 loff_t page_off = page_offset(page);
433 int len = PAGE_CACHE_SIZE;
436 struct ceph_snap_context *snapc, *oldest;
440 dout("writepage %p idx %lu\n", page, page->index);
442 if (!page->mapping || !page->mapping->host) {
443 dout("writepage %p - no mapping\n", page);
446 inode = page->mapping->host;
447 ci = ceph_inode(inode);
448 fsc = ceph_inode_to_client(inode);
449 osdc = &fsc->client->osdc;
451 /* verify this is a writeable snap context */
452 snapc = page_snap_context(page);
454 dout("writepage %p page %p not dirty?\n", inode, page);
457 oldest = get_oldest_context(inode, &snap_size);
458 if (snapc->seq > oldest->seq) {
459 dout("writepage %p page %p snapc %p not writeable - noop\n",
461 /* we should only noop if called by kswapd */
462 WARN_ON((current->flags & PF_MEMALLOC) == 0);
463 ceph_put_snap_context(oldest);
466 ceph_put_snap_context(oldest);
468 /* is this a partial page at end of file? */
472 i_size = i_size_read(inode);
473 if (i_size < page_off + len)
474 len = i_size - page_off;
476 dout("writepage %p page %p index %lu on %llu~%u snapc %p\n",
477 inode, page, page->index, page_off, len, snapc);
479 writeback_stat = atomic_long_inc_return(&fsc->writeback_count);
481 CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb))
482 set_bdi_congested(&fsc->backing_dev_info, BLK_RW_ASYNC);
484 set_page_writeback(page);
485 err = ceph_osdc_writepages(osdc, ceph_vino(inode),
486 &ci->i_layout, snapc,
488 ci->i_truncate_seq, ci->i_truncate_size,
489 &inode->i_mtime, &page, 1);
491 dout("writepage setting page/mapping error %d %p\n", err, page);
493 mapping_set_error(&inode->i_data, err);
495 wbc->pages_skipped++;
497 dout("writepage cleaned page %p\n", page);
498 err = 0; /* vfs expects us to return 0 */
501 ClearPagePrivate(page);
502 end_page_writeback(page);
503 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
504 ceph_put_snap_context(snapc); /* page's reference */
509 static int ceph_writepage(struct page *page, struct writeback_control *wbc)
512 struct inode *inode = page->mapping->host;
515 err = writepage_nounlock(page, wbc);
523 * lame release_pages helper. release_pages() isn't exported to
526 static void ceph_release_pages(struct page **pages, int num)
531 pagevec_init(&pvec, 0);
532 for (i = 0; i < num; i++) {
533 if (pagevec_add(&pvec, pages[i]) == 0)
534 pagevec_release(&pvec);
536 pagevec_release(&pvec);
541 * async writeback completion handler.
543 * If we get an error, set the mapping error bit, but not the individual
546 static void writepages_finish(struct ceph_osd_request *req,
547 struct ceph_msg *msg)
549 struct inode *inode = req->r_inode;
550 struct ceph_inode_info *ci = ceph_inode(inode);
554 struct ceph_snap_context *snapc = req->r_snapc;
555 struct address_space *mapping = inode->i_mapping;
556 int rc = req->r_result;
557 u64 bytes = le64_to_cpu(req->r_request_ops[0].extent.length);
558 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
560 unsigned issued = ceph_caps_issued(ci);
564 * Assume we wrote the pages we originally sent. The
565 * osd might reply with fewer pages if our writeback
566 * raced with a truncation and was adjusted at the osd,
567 * so don't believe the reply.
569 wrote = req->r_num_pages;
572 mapping_set_error(mapping, rc);
574 dout("writepages_finish %p rc %d bytes %llu wrote %d (pages)\n",
575 inode, rc, bytes, wrote);
577 /* clean all pages */
578 for (i = 0; i < req->r_num_pages; i++) {
579 page = req->r_pages[i];
581 WARN_ON(!PageUptodate(page));
584 atomic_long_dec_return(&fsc->writeback_count);
586 CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb))
587 clear_bdi_congested(&fsc->backing_dev_info,
590 ceph_put_snap_context(page_snap_context(page));
592 ClearPagePrivate(page);
593 dout("unlocking %d %p\n", i, page);
594 end_page_writeback(page);
597 * We lost the cache cap, need to truncate the page before
598 * it is unlocked, otherwise we'd truncate it later in the
599 * page truncation thread, possibly losing some data that
602 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0)
603 generic_error_remove_page(inode->i_mapping, page);
607 dout("%p wrote+cleaned %d pages\n", inode, wrote);
608 ceph_put_wrbuffer_cap_refs(ci, req->r_num_pages, snapc);
610 ceph_release_pages(req->r_pages, req->r_num_pages);
611 if (req->r_pages_from_pool)
612 mempool_free(req->r_pages,
613 ceph_sb_to_client(inode->i_sb)->wb_pagevec_pool);
616 ceph_osdc_put_request(req);
620 * allocate a page vec, either directly, or if necessary, via a the
621 * mempool. we avoid the mempool if we can because req->r_num_pages
622 * may be less than the maximum write size.
624 static void alloc_page_vec(struct ceph_fs_client *fsc,
625 struct ceph_osd_request *req)
627 req->r_pages = kmalloc(sizeof(struct page *) * req->r_num_pages,
630 req->r_pages = mempool_alloc(fsc->wb_pagevec_pool, GFP_NOFS);
631 req->r_pages_from_pool = 1;
632 WARN_ON(!req->r_pages);
637 * initiate async writeback
639 static int ceph_writepages_start(struct address_space *mapping,
640 struct writeback_control *wbc)
642 struct inode *inode = mapping->host;
643 struct ceph_inode_info *ci = ceph_inode(inode);
644 struct ceph_fs_client *fsc;
645 pgoff_t index, start, end;
648 pgoff_t max_pages = 0, max_pages_ever = 0;
649 struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc;
653 unsigned wsize = 1 << inode->i_blkbits;
654 struct ceph_osd_request *req = NULL;
659 * Include a 'sync' in the OSD request if this is a data
660 * integrity write (e.g., O_SYNC write or fsync()), or if our
661 * cap is being revoked.
663 do_sync = wbc->sync_mode == WB_SYNC_ALL;
664 if (ceph_caps_revoking(ci, CEPH_CAP_FILE_BUFFER))
666 dout("writepages_start %p dosync=%d (mode=%s)\n",
668 wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
669 (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
671 fsc = ceph_inode_to_client(inode);
672 if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) {
673 pr_warning("writepage_start %p on forced umount\n", inode);
674 return -EIO; /* we're in a forced umount, don't write! */
676 if (fsc->mount_options->wsize && fsc->mount_options->wsize < wsize)
677 wsize = fsc->mount_options->wsize;
678 if (wsize < PAGE_CACHE_SIZE)
679 wsize = PAGE_CACHE_SIZE;
680 max_pages_ever = wsize >> PAGE_CACHE_SHIFT;
682 pagevec_init(&pvec, 0);
684 /* where to start/end? */
685 if (wbc->range_cyclic) {
686 start = mapping->writeback_index; /* Start from prev offset */
688 dout(" cyclic, start at %lu\n", start);
690 start = wbc->range_start >> PAGE_CACHE_SHIFT;
691 end = wbc->range_end >> PAGE_CACHE_SHIFT;
692 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
695 dout(" not cyclic, %lu to %lu\n", start, end);
700 /* find oldest snap context with dirty data */
701 ceph_put_snap_context(snapc);
702 snapc = get_oldest_context(inode, &snap_size);
704 /* hmm, why does writepages get called when there
706 dout(" no snap context with dirty data?\n");
709 dout(" oldest snapc is %p seq %lld (%d snaps)\n",
710 snapc, snapc->seq, snapc->num_snaps);
711 if (last_snapc && snapc != last_snapc) {
712 /* if we switched to a newer snapc, restart our scan at the
713 * start of the original file range. */
714 dout(" snapc differs from last pass, restarting at %lu\n",
720 while (!done && index <= end) {
724 int pvec_pages, locked_pages;
732 max_pages = max_pages_ever;
736 want = min(end - index,
737 min((pgoff_t)PAGEVEC_SIZE,
738 max_pages - (pgoff_t)locked_pages) - 1)
740 pvec_pages = pagevec_lookup_tag(&pvec, mapping, &index,
743 dout("pagevec_lookup_tag got %d\n", pvec_pages);
744 if (!pvec_pages && !locked_pages)
746 for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) {
747 page = pvec.pages[i];
748 dout("? %p idx %lu\n", page, page->index);
749 if (locked_pages == 0)
750 lock_page(page); /* first page */
751 else if (!trylock_page(page))
754 /* only dirty pages, or our accounting breaks */
755 if (unlikely(!PageDirty(page)) ||
756 unlikely(page->mapping != mapping)) {
757 dout("!dirty or !mapping %p\n", page);
761 if (!wbc->range_cyclic && page->index > end) {
762 dout("end of range %p\n", page);
767 if (next && (page->index != next)) {
768 dout("not consecutive %p\n", page);
772 if (wbc->sync_mode != WB_SYNC_NONE) {
773 dout("waiting on writeback %p\n", page);
774 wait_on_page_writeback(page);
776 if ((snap_size && page_offset(page) > snap_size) ||
778 page_offset(page) > i_size_read(inode))) {
779 dout("%p page eof %llu\n", page, snap_size ?
780 snap_size : i_size_read(inode));
785 if (PageWriteback(page)) {
786 dout("%p under writeback\n", page);
791 /* only if matching snap context */
792 pgsnapc = page_snap_context(page);
793 if (pgsnapc->seq > snapc->seq) {
794 dout("page snapc %p %lld > oldest %p %lld\n",
795 pgsnapc, pgsnapc->seq, snapc, snapc->seq);
798 continue; /* keep looking for snap */
802 if (!clear_page_dirty_for_io(page)) {
803 dout("%p !clear_page_dirty_for_io\n", page);
809 if (locked_pages == 0) {
810 /* prepare async write request */
811 offset = (u64) page_offset(page);
813 req = ceph_osdc_new_request(&fsc->client->osdc,
818 CEPH_OSD_FLAG_WRITE |
819 CEPH_OSD_FLAG_ONDISK,
823 &inode->i_mtime, true, 0);
831 max_pages = req->r_num_pages;
833 alloc_page_vec(fsc, req);
834 req->r_callback = writepages_finish;
835 req->r_inode = inode;
838 /* note position of first page in pvec */
841 dout("%p will write page %p idx %lu\n",
842 inode, page, page->index);
845 atomic_long_inc_return(&fsc->writeback_count);
846 if (writeback_stat > CONGESTION_ON_THRESH(
847 fsc->mount_options->congestion_kb)) {
848 set_bdi_congested(&fsc->backing_dev_info,
852 set_page_writeback(page);
853 req->r_pages[locked_pages] = page;
855 next = page->index + 1;
858 /* did we get anything? */
860 goto release_pvec_pages;
863 BUG_ON(!locked_pages || first < 0);
865 if (pvec_pages && i == pvec_pages &&
866 locked_pages < max_pages) {
867 dout("reached end pvec, trying for more\n");
868 pagevec_reinit(&pvec);
872 /* shift unused pages over in the pvec... we
873 * will need to release them below. */
874 for (j = i; j < pvec_pages; j++) {
875 dout(" pvec leftover page %p\n",
877 pvec.pages[j-i+first] = pvec.pages[j];
882 /* submit the write */
883 offset = req->r_pages[0]->index << PAGE_CACHE_SHIFT;
884 len = min((snap_size ? snap_size : i_size_read(inode)) - offset,
885 (u64)locked_pages << PAGE_CACHE_SHIFT);
886 dout("writepages got %d pages at %llu~%llu\n",
887 locked_pages, offset, len);
889 /* revise final length, page count */
890 req->r_num_pages = locked_pages;
891 req->r_request_ops[0].extent.length = cpu_to_le64(len);
892 req->r_request_ops[0].payload_len = cpu_to_le32(len);
893 req->r_request->hdr.data_len = cpu_to_le32(len);
895 rc = ceph_osdc_start_request(&fsc->client->osdc, req, true);
901 wbc->nr_to_write -= locked_pages;
902 if (wbc->nr_to_write <= 0)
906 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr,
907 pvec.nr ? pvec.pages[0] : NULL);
908 pagevec_release(&pvec);
910 if (locked_pages && !done)
914 if (should_loop && !done) {
915 /* more to do; loop back to beginning of file */
916 dout("writepages looping back to beginning of file\n");
922 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
923 mapping->writeback_index = index;
927 ceph_osdc_put_request(req);
928 ceph_put_snap_context(snapc);
929 dout("writepages done, rc = %d\n", rc);
936 * See if a given @snapc is either writeable, or already written.
938 static int context_is_writeable_or_written(struct inode *inode,
939 struct ceph_snap_context *snapc)
941 struct ceph_snap_context *oldest = get_oldest_context(inode, NULL);
942 int ret = !oldest || snapc->seq <= oldest->seq;
944 ceph_put_snap_context(oldest);
949 * We are only allowed to write into/dirty the page if the page is
950 * clean, or already dirty within the same snap context.
952 * called with page locked.
953 * return success with page locked,
954 * or any failure (incl -EAGAIN) with page unlocked.
956 static int ceph_update_writeable_page(struct file *file,
957 loff_t pos, unsigned len,
960 struct inode *inode = file_inode(file);
961 struct ceph_inode_info *ci = ceph_inode(inode);
962 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
963 loff_t page_off = pos & PAGE_CACHE_MASK;
964 int pos_in_page = pos & ~PAGE_CACHE_MASK;
965 int end_in_page = pos_in_page + len;
968 struct ceph_snap_context *snapc, *oldest;
971 /* writepages currently holds page lock, but if we change that later, */
972 wait_on_page_writeback(page);
974 /* check snap context */
975 BUG_ON(!ci->i_snap_realm);
976 down_read(&mdsc->snap_rwsem);
977 BUG_ON(!ci->i_snap_realm->cached_context);
978 snapc = page_snap_context(page);
979 if (snapc && snapc != ci->i_head_snapc) {
981 * this page is already dirty in another (older) snap
982 * context! is it writeable now?
984 oldest = get_oldest_context(inode, NULL);
985 up_read(&mdsc->snap_rwsem);
987 if (snapc->seq > oldest->seq) {
988 ceph_put_snap_context(oldest);
989 dout(" page %p snapc %p not current or oldest\n",
992 * queue for writeback, and wait for snapc to
993 * be writeable or written
995 snapc = ceph_get_snap_context(snapc);
997 ceph_queue_writeback(inode);
998 r = wait_event_interruptible(ci->i_cap_wq,
999 context_is_writeable_or_written(inode, snapc));
1000 ceph_put_snap_context(snapc);
1001 if (r == -ERESTARTSYS)
1005 ceph_put_snap_context(oldest);
1007 /* yay, writeable, do it now (without dropping page lock) */
1008 dout(" page %p snapc %p not current, but oldest\n",
1010 if (!clear_page_dirty_for_io(page))
1012 r = writepage_nounlock(page, NULL);
1018 if (PageUptodate(page)) {
1019 dout(" page %p already uptodate\n", page);
1024 if (pos_in_page == 0 && len == PAGE_CACHE_SIZE)
1027 /* past end of file? */
1028 i_size = inode->i_size; /* caller holds i_mutex */
1030 if (i_size + len > inode->i_sb->s_maxbytes) {
1031 /* file is too big */
1036 if (page_off >= i_size ||
1037 (pos_in_page == 0 && (pos+len) >= i_size &&
1038 end_in_page - pos_in_page != PAGE_CACHE_SIZE)) {
1039 dout(" zeroing %p 0 - %d and %d - %d\n",
1040 page, pos_in_page, end_in_page, (int)PAGE_CACHE_SIZE);
1041 zero_user_segments(page,
1043 end_in_page, PAGE_CACHE_SIZE);
1047 /* we need to read it. */
1048 up_read(&mdsc->snap_rwsem);
1049 r = readpage_nounlock(file, page);
1055 up_read(&mdsc->snap_rwsem);
1062 * We are only allowed to write into/dirty the page if the page is
1063 * clean, or already dirty within the same snap context.
1065 static int ceph_write_begin(struct file *file, struct address_space *mapping,
1066 loff_t pos, unsigned len, unsigned flags,
1067 struct page **pagep, void **fsdata)
1069 struct inode *inode = file_inode(file);
1070 struct ceph_inode_info *ci = ceph_inode(inode);
1071 struct ceph_file_info *fi = file->private_data;
1073 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1074 int r, want, got = 0;
1076 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1077 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1079 want = CEPH_CAP_FILE_BUFFER;
1081 dout("write_begin %p %llx.%llx %llu~%u getting caps. i_size %llu\n",
1082 inode, ceph_vinop(inode), pos, len, inode->i_size);
1083 r = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, pos+len);
1086 dout("write_begin %p %llx.%llx %llu~%u got cap refs on %s\n",
1087 inode, ceph_vinop(inode), pos, len, ceph_cap_string(got));
1088 if (!(got & (CEPH_CAP_FILE_BUFFER|CEPH_CAP_FILE_LAZYIO))) {
1089 ceph_put_cap_refs(ci, got);
1095 page = grab_cache_page_write_begin(mapping, index, 0);
1101 dout("write_begin file %p inode %p page %p %d~%d\n", file,
1102 inode, page, (int)pos, (int)len);
1104 r = ceph_update_writeable_page(file, pos, len, page);
1106 page_cache_release(page);
1107 } while (r == -EAGAIN);
1110 ceph_put_cap_refs(ci, got);
1113 *(int *)fsdata = got;
1119 * we don't do anything in here that simple_write_end doesn't do
1120 * except adjust dirty page accounting and drop read lock on
1123 static int ceph_write_end(struct file *file, struct address_space *mapping,
1124 loff_t pos, unsigned len, unsigned copied,
1125 struct page *page, void *fsdata)
1127 struct inode *inode = file_inode(file);
1128 struct ceph_inode_info *ci = ceph_inode(inode);
1129 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1130 struct ceph_mds_client *mdsc = fsc->mdsc;
1131 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1133 int got = (unsigned long)fsdata;
1135 dout("write_end file %p inode %p page %p %d~%d (%d)\n", file,
1136 inode, page, (int)pos, (int)copied, (int)len);
1138 /* zero the stale part of the page if we did a short copy */
1140 zero_user_segment(page, from+copied, len);
1142 /* did file size increase? */
1143 /* (no need for i_size_read(); we caller holds i_mutex */
1144 if (pos+copied > inode->i_size)
1145 check_cap = ceph_inode_set_size(inode, pos+copied);
1147 if (!PageUptodate(page))
1148 SetPageUptodate(page);
1150 set_page_dirty(page);
1153 up_read(&mdsc->snap_rwsem);
1154 page_cache_release(page);
1158 spin_lock(&ci->i_ceph_lock);
1159 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR);
1160 spin_unlock(&ci->i_ceph_lock);
1162 __mark_inode_dirty(inode, dirty);
1165 dout("write_end %p %llx.%llx %llu~%u dropping cap refs on %s\n",
1166 inode, ceph_vinop(inode), pos, len, ceph_cap_string(got));
1167 ceph_put_cap_refs(ci, got);
1170 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL);
1176 * we set .direct_IO to indicate direct io is supported, but since we
1177 * intercept O_DIRECT reads and writes early, this function should
1180 static ssize_t ceph_direct_io(int rw, struct kiocb *iocb,
1181 const struct iovec *iov,
1182 loff_t pos, unsigned long nr_segs)
1188 const struct address_space_operations ceph_aops = {
1189 .readpage = ceph_readpage,
1190 .readpages = ceph_readpages,
1191 .writepage = ceph_writepage,
1192 .writepages = ceph_writepages_start,
1193 .write_begin = ceph_write_begin,
1194 .write_end = ceph_write_end,
1195 .set_page_dirty = ceph_set_page_dirty,
1196 .invalidatepage = ceph_invalidatepage,
1197 .releasepage = ceph_releasepage,
1198 .direct_IO = ceph_direct_io,
1207 * Reuse write_begin here for simplicity.
1209 static int ceph_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1211 struct inode *inode = file_inode(vma->vm_file);
1212 struct page *page = vmf->page;
1213 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
1214 loff_t off = page_offset(page);
1218 /* Update time before taking page lock */
1219 file_update_time(vma->vm_file);
1221 size = i_size_read(inode);
1222 if (off + PAGE_CACHE_SIZE <= size)
1223 len = PAGE_CACHE_SIZE;
1225 len = size & ~PAGE_CACHE_MASK;
1227 dout("page_mkwrite %p %llu~%llu page %p idx %lu\n", inode,
1228 off, len, page, page->index);
1232 ret = VM_FAULT_NOPAGE;
1234 (page->mapping != inode->i_mapping))
1237 ret = ceph_update_writeable_page(vma->vm_file, off, len, page);
1239 /* success. we'll keep the page locked. */
1240 set_page_dirty(page);
1241 up_read(&mdsc->snap_rwsem);
1242 ret = VM_FAULT_LOCKED;
1247 ret = VM_FAULT_SIGBUS;
1250 dout("page_mkwrite %p %llu~%llu = %d\n", inode, off, len, ret);
1251 if (ret != VM_FAULT_LOCKED)
1256 static struct vm_operations_struct ceph_vmops = {
1257 .fault = filemap_fault,
1258 .page_mkwrite = ceph_page_mkwrite,
1259 .remap_pages = generic_file_remap_pages,
1262 int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1264 struct address_space *mapping = file->f_mapping;
1266 if (!mapping->a_ops->readpage)
1268 file_accessed(file);
1269 vma->vm_ops = &ceph_vmops;