1 #include <linux/ceph/ceph_debug.h>
3 #include <linux/backing-dev.h>
6 #include <linux/pagemap.h>
7 #include <linux/writeback.h> /* generic_writepages */
8 #include <linux/slab.h>
9 #include <linux/pagevec.h>
10 #include <linux/task_io_accounting_ops.h>
13 #include "mds_client.h"
15 #include <linux/ceph/osd_client.h>
18 * Ceph address space ops.
20 * There are a few funny things going on here.
22 * The page->private field is used to reference a struct
23 * ceph_snap_context for _every_ dirty page. This indicates which
24 * snapshot the page was logically dirtied in, and thus which snap
25 * context needs to be associated with the osd write during writeback.
27 * Similarly, struct ceph_inode_info maintains a set of counters to
28 * count dirty pages on the inode. In the absence of snapshots,
29 * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count.
31 * When a snapshot is taken (that is, when the client receives
32 * notification that a snapshot was taken), each inode with caps and
33 * with dirty pages (dirty pages implies there is a cap) gets a new
34 * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending
35 * order, new snaps go to the tail). The i_wrbuffer_ref_head count is
36 * moved to capsnap->dirty. (Unless a sync write is currently in
37 * progress. In that case, the capsnap is said to be "pending", new
38 * writes cannot start, and the capsnap isn't "finalized" until the
39 * write completes (or fails) and a final size/mtime for the inode for
40 * that snap can be settled upon.) i_wrbuffer_ref_head is reset to 0.
42 * On writeback, we must submit writes to the osd IN SNAP ORDER. So,
43 * we look for the first capsnap in i_cap_snaps and write out pages in
44 * that snap context _only_. Then we move on to the next capsnap,
45 * eventually reaching the "live" or "head" context (i.e., pages that
46 * are not yet snapped) and are writing the most recently dirtied
49 * Invalidate and so forth must take care to ensure the dirty page
50 * accounting is preserved.
53 #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10))
54 #define CONGESTION_OFF_THRESH(congestion_kb) \
55 (CONGESTION_ON_THRESH(congestion_kb) - \
56 (CONGESTION_ON_THRESH(congestion_kb) >> 2))
58 static inline struct ceph_snap_context *page_snap_context(struct page *page)
60 if (PagePrivate(page))
61 return (void *)page->private;
66 * Dirty a page. Optimistically adjust accounting, on the assumption
67 * that we won't race with invalidate. If we do, readjust.
69 static int ceph_set_page_dirty(struct page *page)
71 struct address_space *mapping = page->mapping;
73 struct ceph_inode_info *ci;
74 struct ceph_snap_context *snapc;
77 if (unlikely(!mapping))
78 return !TestSetPageDirty(page);
80 if (PageDirty(page)) {
81 dout("%p set_page_dirty %p idx %lu -- already dirty\n",
82 mapping->host, page, page->index);
83 BUG_ON(!PagePrivate(page));
87 inode = mapping->host;
88 ci = ceph_inode(inode);
91 * Note that we're grabbing a snapc ref here without holding
94 snapc = ceph_get_snap_context(ci->i_snap_realm->cached_context);
97 spin_lock(&ci->i_ceph_lock);
98 if (ci->i_head_snapc == NULL)
99 ci->i_head_snapc = ceph_get_snap_context(snapc);
100 ++ci->i_wrbuffer_ref_head;
101 if (ci->i_wrbuffer_ref == 0)
103 ++ci->i_wrbuffer_ref;
104 dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d "
105 "snapc %p seq %lld (%d snaps)\n",
106 mapping->host, page, page->index,
107 ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1,
108 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
109 snapc, snapc->seq, snapc->num_snaps);
110 spin_unlock(&ci->i_ceph_lock);
113 * Reference snap context in page->private. Also set
114 * PagePrivate so that we get invalidatepage callback.
116 BUG_ON(PagePrivate(page));
117 page->private = (unsigned long)snapc;
118 SetPagePrivate(page);
120 ret = __set_page_dirty_nobuffers(page);
121 WARN_ON(!PageLocked(page));
122 WARN_ON(!page->mapping);
128 * If we are truncating the full page (i.e. offset == 0), adjust the
129 * dirty page counters appropriately. Only called if there is private
132 static void ceph_invalidatepage(struct page *page, unsigned int offset,
136 struct ceph_inode_info *ci;
137 struct ceph_snap_context *snapc = page_snap_context(page);
139 inode = page->mapping->host;
140 ci = ceph_inode(inode);
142 if (offset != 0 || length != PAGE_CACHE_SIZE) {
143 dout("%p invalidatepage %p idx %lu partial dirty page %u~%u\n",
144 inode, page, page->index, offset, length);
148 ceph_invalidate_fscache_page(inode, page);
150 if (!PagePrivate(page))
154 * We can get non-dirty pages here due to races between
155 * set_page_dirty and truncate_complete_page; just spit out a
156 * warning, in case we end up with accounting problems later.
158 if (!PageDirty(page))
159 pr_err("%p invalidatepage %p page not dirty\n", inode, page);
161 ClearPageChecked(page);
163 dout("%p invalidatepage %p idx %lu full dirty page\n",
164 inode, page, page->index);
166 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
167 ceph_put_snap_context(snapc);
169 ClearPagePrivate(page);
172 static int ceph_releasepage(struct page *page, gfp_t g)
174 struct inode *inode = page->mapping ? page->mapping->host : NULL;
175 dout("%p releasepage %p idx %lu\n", inode, page, page->index);
176 WARN_ON(PageDirty(page));
178 /* Can we release the page from the cache? */
179 if (!ceph_release_fscache_page(page, g))
182 return !PagePrivate(page);
186 * read a single page, without unlocking it.
188 static int readpage_nounlock(struct file *filp, struct page *page)
190 struct inode *inode = file_inode(filp);
191 struct ceph_inode_info *ci = ceph_inode(inode);
192 struct ceph_osd_client *osdc =
193 &ceph_inode_to_client(inode)->client->osdc;
195 u64 len = PAGE_CACHE_SIZE;
197 err = ceph_readpage_from_fscache(inode, page);
202 dout("readpage inode %p file %p page %p index %lu\n",
203 inode, filp, page, page->index);
204 err = ceph_osdc_readpages(osdc, ceph_vino(inode), &ci->i_layout,
205 (u64) page_offset(page), &len,
206 ci->i_truncate_seq, ci->i_truncate_size,
212 ceph_fscache_readpage_cancel(inode, page);
215 if (err < PAGE_CACHE_SIZE)
216 /* zero fill remainder of page */
217 zero_user_segment(page, err, PAGE_CACHE_SIZE);
219 flush_dcache_page(page);
221 SetPageUptodate(page);
222 ceph_readpage_to_fscache(inode, page);
225 return err < 0 ? err : 0;
228 static int ceph_readpage(struct file *filp, struct page *page)
230 int r = readpage_nounlock(filp, page);
236 * Finish an async read(ahead) op.
238 static void finish_read(struct ceph_osd_request *req, struct ceph_msg *msg)
240 struct inode *inode = req->r_inode;
241 struct ceph_osd_data *osd_data;
242 int rc = req->r_result;
243 int bytes = le32_to_cpu(msg->hdr.data_len);
247 dout("finish_read %p req %p rc %d bytes %d\n", inode, req, rc, bytes);
249 /* unlock all pages, zeroing any data we didn't read */
250 osd_data = osd_req_op_extent_osd_data(req, 0);
251 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
252 num_pages = calc_pages_for((u64)osd_data->alignment,
253 (u64)osd_data->length);
254 for (i = 0; i < num_pages; i++) {
255 struct page *page = osd_data->pages[i];
259 if (bytes < (int)PAGE_CACHE_SIZE) {
260 /* zero (remainder of) page */
261 int s = bytes < 0 ? 0 : bytes;
262 zero_user_segment(page, s, PAGE_CACHE_SIZE);
264 dout("finish_read %p uptodate %p idx %lu\n", inode, page,
266 flush_dcache_page(page);
267 SetPageUptodate(page);
268 ceph_readpage_to_fscache(inode, page);
271 page_cache_release(page);
272 bytes -= PAGE_CACHE_SIZE;
274 kfree(osd_data->pages);
277 static void ceph_unlock_page_vector(struct page **pages, int num_pages)
281 for (i = 0; i < num_pages; i++)
282 unlock_page(pages[i]);
286 * start an async read(ahead) operation. return nr_pages we submitted
287 * a read for on success, or negative error code.
289 static int start_read(struct inode *inode, struct list_head *page_list, int max)
291 struct ceph_osd_client *osdc =
292 &ceph_inode_to_client(inode)->client->osdc;
293 struct ceph_inode_info *ci = ceph_inode(inode);
294 struct page *page = list_entry(page_list->prev, struct page, lru);
295 struct ceph_vino vino;
296 struct ceph_osd_request *req;
305 off = (u64) page_offset(page);
308 next_index = page->index;
309 list_for_each_entry_reverse(page, page_list, lru) {
310 if (page->index != next_index)
314 if (max && nr_pages == max)
317 len = nr_pages << PAGE_CACHE_SHIFT;
318 dout("start_read %p nr_pages %d is %lld~%lld\n", inode, nr_pages,
320 vino = ceph_vino(inode);
321 req = ceph_osdc_new_request(osdc, &ci->i_layout, vino, off, &len,
322 0, 1, CEPH_OSD_OP_READ,
323 CEPH_OSD_FLAG_READ, NULL,
324 ci->i_truncate_seq, ci->i_truncate_size,
329 /* build page vector */
330 nr_pages = calc_pages_for(0, len);
331 pages = kmalloc(sizeof(*pages) * nr_pages, GFP_NOFS);
335 for (i = 0; i < nr_pages; ++i) {
336 page = list_entry(page_list->prev, struct page, lru);
337 BUG_ON(PageLocked(page));
338 list_del(&page->lru);
340 dout("start_read %p adding %p idx %lu\n", inode, page,
342 if (add_to_page_cache_lru(page, &inode->i_data, page->index,
344 ceph_fscache_uncache_page(inode, page);
345 page_cache_release(page);
346 dout("start_read %p add_to_page_cache failed %p\n",
353 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false);
354 req->r_callback = finish_read;
355 req->r_inode = inode;
357 ceph_osdc_build_request(req, off, NULL, vino.snap, NULL);
359 dout("start_read %p starting %p %lld~%lld\n", inode, req, off, len);
360 ret = ceph_osdc_start_request(osdc, req, false);
363 ceph_osdc_put_request(req);
367 ceph_unlock_page_vector(pages, nr_pages);
368 ceph_release_page_vector(pages, nr_pages);
370 ceph_osdc_put_request(req);
376 * Read multiple pages. Leave pages we don't read + unlock in page_list;
377 * the caller (VM) cleans them up.
379 static int ceph_readpages(struct file *file, struct address_space *mapping,
380 struct list_head *page_list, unsigned nr_pages)
382 struct inode *inode = file_inode(file);
383 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
387 rc = ceph_readpages_from_fscache(mapping->host, mapping, page_list,
393 if (fsc->mount_options->rsize >= PAGE_CACHE_SIZE)
394 max = (fsc->mount_options->rsize + PAGE_CACHE_SIZE - 1)
397 dout("readpages %p file %p nr_pages %d max %d\n", inode,
400 while (!list_empty(page_list)) {
401 rc = start_read(inode, page_list, max);
407 ceph_fscache_readpages_cancel(inode, page_list);
409 dout("readpages %p file %p ret %d\n", inode, file, rc);
414 * Get ref for the oldest snapc for an inode with dirty data... that is, the
415 * only snap context we are allowed to write back.
417 static struct ceph_snap_context *get_oldest_context(struct inode *inode,
420 struct ceph_inode_info *ci = ceph_inode(inode);
421 struct ceph_snap_context *snapc = NULL;
422 struct ceph_cap_snap *capsnap = NULL;
424 spin_lock(&ci->i_ceph_lock);
425 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
426 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap,
427 capsnap->context, capsnap->dirty_pages);
428 if (capsnap->dirty_pages) {
429 snapc = ceph_get_snap_context(capsnap->context);
431 *snap_size = capsnap->size;
435 if (!snapc && ci->i_wrbuffer_ref_head) {
436 snapc = ceph_get_snap_context(ci->i_head_snapc);
437 dout(" head snapc %p has %d dirty pages\n",
438 snapc, ci->i_wrbuffer_ref_head);
440 spin_unlock(&ci->i_ceph_lock);
445 * Write a single page, but leave the page locked.
447 * If we get a write error, set the page error bit, but still adjust the
448 * dirty page accounting (i.e., page is no longer dirty).
450 static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
453 struct ceph_inode_info *ci;
454 struct ceph_fs_client *fsc;
455 struct ceph_osd_client *osdc;
456 struct ceph_snap_context *snapc, *oldest;
457 loff_t page_off = page_offset(page);
459 u64 truncate_size, snap_size = 0;
461 int err = 0, len = PAGE_CACHE_SIZE;
463 dout("writepage %p idx %lu\n", page, page->index);
465 if (!page->mapping || !page->mapping->host) {
466 dout("writepage %p - no mapping\n", page);
469 inode = page->mapping->host;
470 ci = ceph_inode(inode);
471 fsc = ceph_inode_to_client(inode);
472 osdc = &fsc->client->osdc;
474 /* verify this is a writeable snap context */
475 snapc = page_snap_context(page);
477 dout("writepage %p page %p not dirty?\n", inode, page);
480 oldest = get_oldest_context(inode, &snap_size);
481 if (snapc->seq > oldest->seq) {
482 dout("writepage %p page %p snapc %p not writeable - noop\n",
484 /* we should only noop if called by kswapd */
485 WARN_ON((current->flags & PF_MEMALLOC) == 0);
486 ceph_put_snap_context(oldest);
489 ceph_put_snap_context(oldest);
491 spin_lock(&ci->i_ceph_lock);
492 truncate_seq = ci->i_truncate_seq;
493 truncate_size = ci->i_truncate_size;
495 snap_size = i_size_read(inode);
496 spin_unlock(&ci->i_ceph_lock);
498 /* is this a partial page at end of file? */
499 if (page_off >= snap_size) {
500 dout("%p page eof %llu\n", page, snap_size);
503 if (snap_size < page_off + len)
504 len = snap_size - page_off;
506 dout("writepage %p page %p index %lu on %llu~%u snapc %p\n",
507 inode, page, page->index, page_off, len, snapc);
509 writeback_stat = atomic_long_inc_return(&fsc->writeback_count);
511 CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb))
512 set_bdi_congested(&fsc->backing_dev_info, BLK_RW_ASYNC);
514 ceph_readpage_to_fscache(inode, page);
516 set_page_writeback(page);
517 err = ceph_osdc_writepages(osdc, ceph_vino(inode),
518 &ci->i_layout, snapc,
520 truncate_seq, truncate_size,
521 &inode->i_mtime, &page, 1);
523 dout("writepage setting page/mapping error %d %p\n", err, page);
525 mapping_set_error(&inode->i_data, err);
527 wbc->pages_skipped++;
529 dout("writepage cleaned page %p\n", page);
530 err = 0; /* vfs expects us to return 0 */
533 ClearPagePrivate(page);
534 end_page_writeback(page);
535 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
536 ceph_put_snap_context(snapc); /* page's reference */
541 static int ceph_writepage(struct page *page, struct writeback_control *wbc)
544 struct inode *inode = page->mapping->host;
547 err = writepage_nounlock(page, wbc);
555 * lame release_pages helper. release_pages() isn't exported to
558 static void ceph_release_pages(struct page **pages, int num)
563 pagevec_init(&pvec, 0);
564 for (i = 0; i < num; i++) {
565 if (pagevec_add(&pvec, pages[i]) == 0)
566 pagevec_release(&pvec);
568 pagevec_release(&pvec);
572 * async writeback completion handler.
574 * If we get an error, set the mapping error bit, but not the individual
577 static void writepages_finish(struct ceph_osd_request *req,
578 struct ceph_msg *msg)
580 struct inode *inode = req->r_inode;
581 struct ceph_inode_info *ci = ceph_inode(inode);
582 struct ceph_osd_data *osd_data;
587 struct ceph_snap_context *snapc = req->r_snapc;
588 struct address_space *mapping = inode->i_mapping;
589 int rc = req->r_result;
590 u64 bytes = req->r_ops[0].extent.length;
591 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
593 unsigned issued = ceph_caps_issued(ci);
595 osd_data = osd_req_op_extent_osd_data(req, 0);
596 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
597 num_pages = calc_pages_for((u64)osd_data->alignment,
598 (u64)osd_data->length);
601 * Assume we wrote the pages we originally sent. The
602 * osd might reply with fewer pages if our writeback
603 * raced with a truncation and was adjusted at the osd,
604 * so don't believe the reply.
609 mapping_set_error(mapping, rc);
611 dout("writepages_finish %p rc %d bytes %llu wrote %d (pages)\n",
612 inode, rc, bytes, wrote);
614 /* clean all pages */
615 for (i = 0; i < num_pages; i++) {
616 page = osd_data->pages[i];
618 WARN_ON(!PageUptodate(page));
621 atomic_long_dec_return(&fsc->writeback_count);
623 CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb))
624 clear_bdi_congested(&fsc->backing_dev_info,
627 ceph_put_snap_context(page_snap_context(page));
629 ClearPagePrivate(page);
630 dout("unlocking %d %p\n", i, page);
631 end_page_writeback(page);
634 * We lost the cache cap, need to truncate the page before
635 * it is unlocked, otherwise we'd truncate it later in the
636 * page truncation thread, possibly losing some data that
639 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0)
640 generic_error_remove_page(inode->i_mapping, page);
644 dout("%p wrote+cleaned %d pages\n", inode, wrote);
645 ceph_put_wrbuffer_cap_refs(ci, num_pages, snapc);
647 ceph_release_pages(osd_data->pages, num_pages);
648 if (osd_data->pages_from_pool)
649 mempool_free(osd_data->pages,
650 ceph_sb_to_client(inode->i_sb)->wb_pagevec_pool);
652 kfree(osd_data->pages);
653 ceph_osdc_put_request(req);
657 * initiate async writeback
659 static int ceph_writepages_start(struct address_space *mapping,
660 struct writeback_control *wbc)
662 struct inode *inode = mapping->host;
663 struct ceph_inode_info *ci = ceph_inode(inode);
664 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
665 struct ceph_vino vino = ceph_vino(inode);
666 pgoff_t index, start, end;
669 pgoff_t max_pages = 0, max_pages_ever = 0;
670 struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc;
674 unsigned wsize = 1 << inode->i_blkbits;
675 struct ceph_osd_request *req = NULL;
677 u64 truncate_size, snap_size;
681 * Include a 'sync' in the OSD request if this is a data
682 * integrity write (e.g., O_SYNC write or fsync()), or if our
683 * cap is being revoked.
685 if ((wbc->sync_mode == WB_SYNC_ALL) ||
686 ceph_caps_revoking(ci, CEPH_CAP_FILE_BUFFER))
688 dout("writepages_start %p dosync=%d (mode=%s)\n",
690 wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
691 (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
693 if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) {
694 pr_warn("writepage_start %p on forced umount\n", inode);
695 return -EIO; /* we're in a forced umount, don't write! */
697 if (fsc->mount_options->wsize && fsc->mount_options->wsize < wsize)
698 wsize = fsc->mount_options->wsize;
699 if (wsize < PAGE_CACHE_SIZE)
700 wsize = PAGE_CACHE_SIZE;
701 max_pages_ever = wsize >> PAGE_CACHE_SHIFT;
703 pagevec_init(&pvec, 0);
705 /* where to start/end? */
706 if (wbc->range_cyclic) {
707 start = mapping->writeback_index; /* Start from prev offset */
709 dout(" cyclic, start at %lu\n", start);
711 start = wbc->range_start >> PAGE_CACHE_SHIFT;
712 end = wbc->range_end >> PAGE_CACHE_SHIFT;
713 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
716 dout(" not cyclic, %lu to %lu\n", start, end);
721 /* find oldest snap context with dirty data */
722 ceph_put_snap_context(snapc);
724 snapc = get_oldest_context(inode, &snap_size);
726 /* hmm, why does writepages get called when there
728 dout(" no snap context with dirty data?\n");
732 snap_size = i_size_read(inode);
733 dout(" oldest snapc is %p seq %lld (%d snaps)\n",
734 snapc, snapc->seq, snapc->num_snaps);
736 spin_lock(&ci->i_ceph_lock);
737 truncate_seq = ci->i_truncate_seq;
738 truncate_size = ci->i_truncate_size;
740 snap_size = i_size_read(inode);
741 spin_unlock(&ci->i_ceph_lock);
743 if (last_snapc && snapc != last_snapc) {
744 /* if we switched to a newer snapc, restart our scan at the
745 * start of the original file range. */
746 dout(" snapc differs from last pass, restarting at %lu\n",
752 while (!done && index <= end) {
756 int pvec_pages, locked_pages;
757 struct page **pages = NULL;
758 mempool_t *pool = NULL; /* Becomes non-null if mempool used */
766 max_pages = max_pages_ever;
770 want = min(end - index,
771 min((pgoff_t)PAGEVEC_SIZE,
772 max_pages - (pgoff_t)locked_pages) - 1)
774 pvec_pages = pagevec_lookup_tag(&pvec, mapping, &index,
777 dout("pagevec_lookup_tag got %d\n", pvec_pages);
778 if (!pvec_pages && !locked_pages)
780 for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) {
781 page = pvec.pages[i];
782 dout("? %p idx %lu\n", page, page->index);
783 if (locked_pages == 0)
784 lock_page(page); /* first page */
785 else if (!trylock_page(page))
788 /* only dirty pages, or our accounting breaks */
789 if (unlikely(!PageDirty(page)) ||
790 unlikely(page->mapping != mapping)) {
791 dout("!dirty or !mapping %p\n", page);
795 if (!wbc->range_cyclic && page->index > end) {
796 dout("end of range %p\n", page);
801 if (next && (page->index != next)) {
802 dout("not consecutive %p\n", page);
806 if (wbc->sync_mode != WB_SYNC_NONE) {
807 dout("waiting on writeback %p\n", page);
808 wait_on_page_writeback(page);
810 if (page_offset(page) >= snap_size) {
811 dout("%p page eof %llu\n", page, snap_size);
816 if (PageWriteback(page)) {
817 dout("%p under writeback\n", page);
822 /* only if matching snap context */
823 pgsnapc = page_snap_context(page);
824 if (pgsnapc->seq > snapc->seq) {
825 dout("page snapc %p %lld > oldest %p %lld\n",
826 pgsnapc, pgsnapc->seq, snapc, snapc->seq);
829 continue; /* keep looking for snap */
833 if (!clear_page_dirty_for_io(page)) {
834 dout("%p !clear_page_dirty_for_io\n", page);
840 * We have something to write. If this is
841 * the first locked page this time through,
842 * allocate an osd request and a page array
845 if (locked_pages == 0) {
847 /* prepare async write request */
848 offset = (u64)page_offset(page);
850 req = ceph_osdc_new_request(&fsc->client->osdc,
855 CEPH_OSD_FLAG_WRITE |
856 CEPH_OSD_FLAG_ONDISK,
858 truncate_size, true);
866 osd_req_op_init(req, 1, CEPH_OSD_OP_STARTSYNC);
868 req->r_callback = writepages_finish;
869 req->r_inode = inode;
871 max_pages = calc_pages_for(0, (u64)len);
872 pages = kmalloc(max_pages * sizeof (*pages),
875 pool = fsc->wb_pagevec_pool;
876 pages = mempool_alloc(pool, GFP_NOFS);
881 /* note position of first page in pvec */
884 dout("%p will write page %p idx %lu\n",
885 inode, page, page->index);
888 atomic_long_inc_return(&fsc->writeback_count);
889 if (writeback_stat > CONGESTION_ON_THRESH(
890 fsc->mount_options->congestion_kb)) {
891 set_bdi_congested(&fsc->backing_dev_info,
895 set_page_writeback(page);
896 pages[locked_pages] = page;
898 next = page->index + 1;
901 /* did we get anything? */
903 goto release_pvec_pages;
906 BUG_ON(!locked_pages || first < 0);
908 if (pvec_pages && i == pvec_pages &&
909 locked_pages < max_pages) {
910 dout("reached end pvec, trying for more\n");
911 pagevec_reinit(&pvec);
915 /* shift unused pages over in the pvec... we
916 * will need to release them below. */
917 for (j = i; j < pvec_pages; j++) {
918 dout(" pvec leftover page %p\n",
920 pvec.pages[j-i+first] = pvec.pages[j];
925 /* Format the osd request message and submit the write */
927 offset = page_offset(pages[0]);
928 len = min(snap_size - offset,
929 (u64)locked_pages << PAGE_CACHE_SHIFT);
930 dout("writepages got %d pages at %llu~%llu\n",
931 locked_pages, offset, len);
933 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0,
936 pages = NULL; /* request message now owns the pages array */
939 /* Update the write op length in case we changed it */
941 osd_req_op_extent_update(req, 0, len);
943 vino = ceph_vino(inode);
944 ceph_osdc_build_request(req, offset, snapc, vino.snap,
947 rc = ceph_osdc_start_request(&fsc->client->osdc, req, true);
953 wbc->nr_to_write -= locked_pages;
954 if (wbc->nr_to_write <= 0)
958 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr,
959 pvec.nr ? pvec.pages[0] : NULL);
960 pagevec_release(&pvec);
962 if (locked_pages && !done)
966 if (should_loop && !done) {
967 /* more to do; loop back to beginning of file */
968 dout("writepages looping back to beginning of file\n");
974 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
975 mapping->writeback_index = index;
979 ceph_osdc_put_request(req);
980 ceph_put_snap_context(snapc);
981 dout("writepages done, rc = %d\n", rc);
988 * See if a given @snapc is either writeable, or already written.
990 static int context_is_writeable_or_written(struct inode *inode,
991 struct ceph_snap_context *snapc)
993 struct ceph_snap_context *oldest = get_oldest_context(inode, NULL);
994 int ret = !oldest || snapc->seq <= oldest->seq;
996 ceph_put_snap_context(oldest);
1001 * We are only allowed to write into/dirty the page if the page is
1002 * clean, or already dirty within the same snap context.
1004 * called with page locked.
1005 * return success with page locked,
1006 * or any failure (incl -EAGAIN) with page unlocked.
1008 static int ceph_update_writeable_page(struct file *file,
1009 loff_t pos, unsigned len,
1012 struct inode *inode = file_inode(file);
1013 struct ceph_inode_info *ci = ceph_inode(inode);
1014 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
1015 loff_t page_off = pos & PAGE_CACHE_MASK;
1016 int pos_in_page = pos & ~PAGE_CACHE_MASK;
1017 int end_in_page = pos_in_page + len;
1020 struct ceph_snap_context *snapc, *oldest;
1023 /* writepages currently holds page lock, but if we change that later, */
1024 wait_on_page_writeback(page);
1026 /* check snap context */
1027 BUG_ON(!ci->i_snap_realm);
1028 down_read(&mdsc->snap_rwsem);
1029 BUG_ON(!ci->i_snap_realm->cached_context);
1030 snapc = page_snap_context(page);
1031 if (snapc && snapc != ci->i_head_snapc) {
1033 * this page is already dirty in another (older) snap
1034 * context! is it writeable now?
1036 oldest = get_oldest_context(inode, NULL);
1037 up_read(&mdsc->snap_rwsem);
1039 if (snapc->seq > oldest->seq) {
1040 ceph_put_snap_context(oldest);
1041 dout(" page %p snapc %p not current or oldest\n",
1044 * queue for writeback, and wait for snapc to
1045 * be writeable or written
1047 snapc = ceph_get_snap_context(snapc);
1049 ceph_queue_writeback(inode);
1050 r = wait_event_interruptible(ci->i_cap_wq,
1051 context_is_writeable_or_written(inode, snapc));
1052 ceph_put_snap_context(snapc);
1053 if (r == -ERESTARTSYS)
1057 ceph_put_snap_context(oldest);
1059 /* yay, writeable, do it now (without dropping page lock) */
1060 dout(" page %p snapc %p not current, but oldest\n",
1062 if (!clear_page_dirty_for_io(page))
1064 r = writepage_nounlock(page, NULL);
1070 if (PageUptodate(page)) {
1071 dout(" page %p already uptodate\n", page);
1076 if (pos_in_page == 0 && len == PAGE_CACHE_SIZE)
1079 /* past end of file? */
1080 i_size = inode->i_size; /* caller holds i_mutex */
1082 if (page_off >= i_size ||
1083 (pos_in_page == 0 && (pos+len) >= i_size &&
1084 end_in_page - pos_in_page != PAGE_CACHE_SIZE)) {
1085 dout(" zeroing %p 0 - %d and %d - %d\n",
1086 page, pos_in_page, end_in_page, (int)PAGE_CACHE_SIZE);
1087 zero_user_segments(page,
1089 end_in_page, PAGE_CACHE_SIZE);
1093 /* we need to read it. */
1094 up_read(&mdsc->snap_rwsem);
1095 r = readpage_nounlock(file, page);
1105 * We are only allowed to write into/dirty the page if the page is
1106 * clean, or already dirty within the same snap context.
1108 static int ceph_write_begin(struct file *file, struct address_space *mapping,
1109 loff_t pos, unsigned len, unsigned flags,
1110 struct page **pagep, void **fsdata)
1112 struct inode *inode = file_inode(file);
1114 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1119 page = grab_cache_page_write_begin(mapping, index, 0);
1124 dout("write_begin file %p inode %p page %p %d~%d\n", file,
1125 inode, page, (int)pos, (int)len);
1127 r = ceph_update_writeable_page(file, pos, len, page);
1128 } while (r == -EAGAIN);
1134 * we don't do anything in here that simple_write_end doesn't do
1135 * except adjust dirty page accounting and drop read lock on
1138 static int ceph_write_end(struct file *file, struct address_space *mapping,
1139 loff_t pos, unsigned len, unsigned copied,
1140 struct page *page, void *fsdata)
1142 struct inode *inode = file_inode(file);
1143 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1144 struct ceph_mds_client *mdsc = fsc->mdsc;
1145 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1148 dout("write_end file %p inode %p page %p %d~%d (%d)\n", file,
1149 inode, page, (int)pos, (int)copied, (int)len);
1151 /* zero the stale part of the page if we did a short copy */
1153 zero_user_segment(page, from+copied, len);
1155 /* did file size increase? */
1156 /* (no need for i_size_read(); we caller holds i_mutex */
1157 if (pos+copied > inode->i_size)
1158 check_cap = ceph_inode_set_size(inode, pos+copied);
1160 if (!PageUptodate(page))
1161 SetPageUptodate(page);
1163 set_page_dirty(page);
1166 up_read(&mdsc->snap_rwsem);
1167 page_cache_release(page);
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 struct iov_iter *iter,
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,
1205 static int ceph_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1207 struct inode *inode = file_inode(vma->vm_file);
1208 struct ceph_inode_info *ci = ceph_inode(inode);
1209 struct ceph_file_info *fi = vma->vm_file->private_data;
1210 loff_t off = vmf->pgoff << PAGE_CACHE_SHIFT;
1213 dout("filemap_fault %p %llx.%llx %llu~%zd trying to get caps\n",
1214 inode, ceph_vinop(inode), off, (size_t)PAGE_CACHE_SIZE);
1215 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1216 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1218 want = CEPH_CAP_FILE_CACHE;
1221 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, &got, -1);
1224 if (ret != -ERESTARTSYS) {
1226 return VM_FAULT_SIGBUS;
1229 dout("filemap_fault %p %llu~%zd got cap refs on %s\n",
1230 inode, off, (size_t)PAGE_CACHE_SIZE, ceph_cap_string(got));
1232 ret = filemap_fault(vma, vmf);
1234 dout("filemap_fault %p %llu~%zd dropping cap refs on %s ret %d\n",
1235 inode, off, (size_t)PAGE_CACHE_SIZE, ceph_cap_string(got), ret);
1236 ceph_put_cap_refs(ci, got);
1242 * Reuse write_begin here for simplicity.
1244 static int ceph_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1246 struct inode *inode = file_inode(vma->vm_file);
1247 struct ceph_inode_info *ci = ceph_inode(inode);
1248 struct ceph_file_info *fi = vma->vm_file->private_data;
1249 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
1250 struct page *page = vmf->page;
1251 loff_t off = page_offset(page);
1252 loff_t size = i_size_read(inode);
1256 if (off + PAGE_CACHE_SIZE <= size)
1257 len = PAGE_CACHE_SIZE;
1259 len = size & ~PAGE_CACHE_MASK;
1261 dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n",
1262 inode, ceph_vinop(inode), off, len, size);
1263 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1264 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1266 want = CEPH_CAP_FILE_BUFFER;
1269 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, &got, off + len);
1272 if (ret != -ERESTARTSYS) {
1274 return VM_FAULT_SIGBUS;
1277 dout("page_mkwrite %p %llu~%zd got cap refs on %s\n",
1278 inode, off, len, ceph_cap_string(got));
1280 /* Update time before taking page lock */
1281 file_update_time(vma->vm_file);
1285 ret = VM_FAULT_NOPAGE;
1287 (page->mapping != inode->i_mapping))
1290 ret = ceph_update_writeable_page(vma->vm_file, off, len, page);
1292 /* success. we'll keep the page locked. */
1293 set_page_dirty(page);
1294 up_read(&mdsc->snap_rwsem);
1295 ret = VM_FAULT_LOCKED;
1300 ret = VM_FAULT_SIGBUS;
1303 if (ret != VM_FAULT_LOCKED) {
1307 spin_lock(&ci->i_ceph_lock);
1308 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR);
1309 spin_unlock(&ci->i_ceph_lock);
1311 __mark_inode_dirty(inode, dirty);
1314 dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %d\n",
1315 inode, off, len, ceph_cap_string(got), ret);
1316 ceph_put_cap_refs(ci, got);
1321 void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
1322 char *data, size_t len)
1324 struct address_space *mapping = inode->i_mapping;
1330 if (i_size_read(inode) == 0)
1332 page = find_or_create_page(mapping, 0,
1333 mapping_gfp_mask(mapping) & ~__GFP_FS);
1336 if (PageUptodate(page)) {
1338 page_cache_release(page);
1343 dout("fill_inline_data %p %llx.%llx len %lu locked_page %p\n",
1344 inode, ceph_vinop(inode), len, locked_page);
1347 void *kaddr = kmap_atomic(page);
1348 memcpy(kaddr, data, len);
1349 kunmap_atomic(kaddr);
1352 if (page != locked_page) {
1353 if (len < PAGE_CACHE_SIZE)
1354 zero_user_segment(page, len, PAGE_CACHE_SIZE);
1356 flush_dcache_page(page);
1358 SetPageUptodate(page);
1360 page_cache_release(page);
1364 static struct vm_operations_struct ceph_vmops = {
1365 .fault = ceph_filemap_fault,
1366 .page_mkwrite = ceph_page_mkwrite,
1367 .remap_pages = generic_file_remap_pages,
1370 int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1372 struct address_space *mapping = file->f_mapping;
1374 if (!mapping->a_ops->readpage)
1376 file_accessed(file);
1377 vma->vm_ops = &ceph_vmops;