1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
4 #include <linux/backing-dev.h>
7 #include <linux/pagemap.h>
8 #include <linux/writeback.h> /* generic_writepages */
9 #include <linux/slab.h>
10 #include <linux/pagevec.h>
11 #include <linux/task_io_accounting_ops.h>
12 #include <linux/signal.h>
15 #include "mds_client.h"
17 #include <linux/ceph/osd_client.h>
20 * Ceph address space ops.
22 * There are a few funny things going on here.
24 * The page->private field is used to reference a struct
25 * ceph_snap_context for _every_ dirty page. This indicates which
26 * snapshot the page was logically dirtied in, and thus which snap
27 * context needs to be associated with the osd write during writeback.
29 * Similarly, struct ceph_inode_info maintains a set of counters to
30 * count dirty pages on the inode. In the absence of snapshots,
31 * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count.
33 * When a snapshot is taken (that is, when the client receives
34 * notification that a snapshot was taken), each inode with caps and
35 * with dirty pages (dirty pages implies there is a cap) gets a new
36 * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending
37 * order, new snaps go to the tail). The i_wrbuffer_ref_head count is
38 * moved to capsnap->dirty. (Unless a sync write is currently in
39 * progress. In that case, the capsnap is said to be "pending", new
40 * writes cannot start, and the capsnap isn't "finalized" until the
41 * write completes (or fails) and a final size/mtime for the inode for
42 * that snap can be settled upon.) i_wrbuffer_ref_head is reset to 0.
44 * On writeback, we must submit writes to the osd IN SNAP ORDER. So,
45 * we look for the first capsnap in i_cap_snaps and write out pages in
46 * that snap context _only_. Then we move on to the next capsnap,
47 * eventually reaching the "live" or "head" context (i.e., pages that
48 * are not yet snapped) and are writing the most recently dirtied
51 * Invalidate and so forth must take care to ensure the dirty page
52 * accounting is preserved.
55 #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10))
56 #define CONGESTION_OFF_THRESH(congestion_kb) \
57 (CONGESTION_ON_THRESH(congestion_kb) - \
58 (CONGESTION_ON_THRESH(congestion_kb) >> 2))
60 static inline struct ceph_snap_context *page_snap_context(struct page *page)
62 if (PagePrivate(page))
63 return (void *)page->private;
68 * Dirty a page. Optimistically adjust accounting, on the assumption
69 * that we won't race with invalidate. If we do, readjust.
71 static int ceph_set_page_dirty(struct page *page)
73 struct address_space *mapping = page->mapping;
75 struct ceph_inode_info *ci;
76 struct ceph_snap_context *snapc;
79 if (unlikely(!mapping))
80 return !TestSetPageDirty(page);
82 if (PageDirty(page)) {
83 dout("%p set_page_dirty %p idx %lu -- already dirty\n",
84 mapping->host, page, page->index);
85 BUG_ON(!PagePrivate(page));
89 inode = mapping->host;
90 ci = ceph_inode(inode);
93 spin_lock(&ci->i_ceph_lock);
94 BUG_ON(ci->i_wr_ref == 0); // caller should hold Fw reference
95 if (__ceph_have_pending_cap_snap(ci)) {
96 struct ceph_cap_snap *capsnap =
97 list_last_entry(&ci->i_cap_snaps,
100 snapc = ceph_get_snap_context(capsnap->context);
101 capsnap->dirty_pages++;
103 BUG_ON(!ci->i_head_snapc);
104 snapc = ceph_get_snap_context(ci->i_head_snapc);
105 ++ci->i_wrbuffer_ref_head;
107 if (ci->i_wrbuffer_ref == 0)
109 ++ci->i_wrbuffer_ref;
110 dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d "
111 "snapc %p seq %lld (%d snaps)\n",
112 mapping->host, page, page->index,
113 ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1,
114 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
115 snapc, snapc->seq, snapc->num_snaps);
116 spin_unlock(&ci->i_ceph_lock);
119 * Reference snap context in page->private. Also set
120 * PagePrivate so that we get invalidatepage callback.
122 BUG_ON(PagePrivate(page));
123 page->private = (unsigned long)snapc;
124 SetPagePrivate(page);
126 ret = __set_page_dirty_nobuffers(page);
127 WARN_ON(!PageLocked(page));
128 WARN_ON(!page->mapping);
134 * If we are truncating the full page (i.e. offset == 0), adjust the
135 * dirty page counters appropriately. Only called if there is private
138 static void ceph_invalidatepage(struct page *page, unsigned int offset,
142 struct ceph_inode_info *ci;
143 struct ceph_snap_context *snapc = page_snap_context(page);
145 inode = page->mapping->host;
146 ci = ceph_inode(inode);
148 if (offset != 0 || length != PAGE_SIZE) {
149 dout("%p invalidatepage %p idx %lu partial dirty page %u~%u\n",
150 inode, page, page->index, offset, length);
154 ceph_invalidate_fscache_page(inode, page);
156 WARN_ON(!PageLocked(page));
157 if (!PagePrivate(page))
160 ClearPageChecked(page);
162 dout("%p invalidatepage %p idx %lu full dirty page\n",
163 inode, page, page->index);
165 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
166 ceph_put_snap_context(snapc);
168 ClearPagePrivate(page);
171 static int ceph_releasepage(struct page *page, gfp_t g)
173 dout("%p releasepage %p idx %lu (%sdirty)\n", page->mapping->host,
174 page, page->index, PageDirty(page) ? "" : "not ");
176 /* Can we release the page from the cache? */
177 if (!ceph_release_fscache_page(page, g))
180 return !PagePrivate(page);
184 * read a single page, without unlocking it.
186 static int ceph_do_readpage(struct file *filp, struct page *page)
188 struct inode *inode = file_inode(filp);
189 struct ceph_inode_info *ci = ceph_inode(inode);
190 struct ceph_osd_client *osdc =
191 &ceph_inode_to_client(inode)->client->osdc;
193 u64 off = page_offset(page);
196 if (off >= i_size_read(inode)) {
197 zero_user_segment(page, 0, PAGE_SIZE);
198 SetPageUptodate(page);
202 if (ci->i_inline_version != CEPH_INLINE_NONE) {
204 * Uptodate inline data should have been added
205 * into page cache while getting Fcr caps.
209 zero_user_segment(page, 0, PAGE_SIZE);
210 SetPageUptodate(page);
214 err = ceph_readpage_from_fscache(inode, page);
218 dout("readpage inode %p file %p page %p index %lu\n",
219 inode, filp, page, page->index);
220 err = ceph_osdc_readpages(osdc, ceph_vino(inode), &ci->i_layout,
222 ci->i_truncate_seq, ci->i_truncate_size,
228 ceph_fscache_readpage_cancel(inode, page);
232 /* zero fill remainder of page */
233 zero_user_segment(page, err, PAGE_SIZE);
235 flush_dcache_page(page);
237 SetPageUptodate(page);
238 ceph_readpage_to_fscache(inode, page);
241 return err < 0 ? err : 0;
244 static int ceph_readpage(struct file *filp, struct page *page)
246 int r = ceph_do_readpage(filp, page);
247 if (r != -EINPROGRESS)
255 * Finish an async read(ahead) op.
257 static void finish_read(struct ceph_osd_request *req)
259 struct inode *inode = req->r_inode;
260 struct ceph_osd_data *osd_data;
261 int rc = req->r_result <= 0 ? req->r_result : 0;
262 int bytes = req->r_result >= 0 ? req->r_result : 0;
266 dout("finish_read %p req %p rc %d bytes %d\n", inode, req, rc, bytes);
268 /* unlock all pages, zeroing any data we didn't read */
269 osd_data = osd_req_op_extent_osd_data(req, 0);
270 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
271 num_pages = calc_pages_for((u64)osd_data->alignment,
272 (u64)osd_data->length);
273 for (i = 0; i < num_pages; i++) {
274 struct page *page = osd_data->pages[i];
276 if (rc < 0 && rc != -ENOENT) {
277 ceph_fscache_readpage_cancel(inode, page);
280 if (bytes < (int)PAGE_SIZE) {
281 /* zero (remainder of) page */
282 int s = bytes < 0 ? 0 : bytes;
283 zero_user_segment(page, s, PAGE_SIZE);
285 dout("finish_read %p uptodate %p idx %lu\n", inode, page,
287 flush_dcache_page(page);
288 SetPageUptodate(page);
289 ceph_readpage_to_fscache(inode, page);
295 kfree(osd_data->pages);
299 * start an async read(ahead) operation. return nr_pages we submitted
300 * a read for on success, or negative error code.
302 static int start_read(struct inode *inode, struct list_head *page_list, int max)
304 struct ceph_osd_client *osdc =
305 &ceph_inode_to_client(inode)->client->osdc;
306 struct ceph_inode_info *ci = ceph_inode(inode);
307 struct page *page = list_entry(page_list->prev, struct page, lru);
308 struct ceph_vino vino;
309 struct ceph_osd_request *req;
319 if (!current->journal_info) {
320 /* caller of readpages does not hold buffer and read caps
321 * (fadvise, madvise and readahead cases) */
322 int want = CEPH_CAP_FILE_CACHE;
323 ret = ceph_try_get_caps(ci, CEPH_CAP_FILE_RD, want, &got);
325 dout("start_read %p, error getting cap\n", inode);
326 } else if (!(got & want)) {
327 dout("start_read %p, no cache cap\n", inode);
332 ceph_put_cap_refs(ci, got);
333 while (!list_empty(page_list)) {
334 page = list_entry(page_list->prev,
336 list_del(&page->lru);
343 off = (u64) page_offset(page);
346 next_index = page->index;
347 list_for_each_entry_reverse(page, page_list, lru) {
348 if (page->index != next_index)
352 if (max && nr_pages == max)
355 len = nr_pages << PAGE_SHIFT;
356 dout("start_read %p nr_pages %d is %lld~%lld\n", inode, nr_pages,
358 vino = ceph_vino(inode);
359 req = ceph_osdc_new_request(osdc, &ci->i_layout, vino, off, &len,
360 0, 1, CEPH_OSD_OP_READ,
361 CEPH_OSD_FLAG_READ, NULL,
362 ci->i_truncate_seq, ci->i_truncate_size,
369 /* build page vector */
370 nr_pages = calc_pages_for(0, len);
371 pages = kmalloc(sizeof(*pages) * nr_pages, GFP_KERNEL);
376 for (i = 0; i < nr_pages; ++i) {
377 page = list_entry(page_list->prev, struct page, lru);
378 BUG_ON(PageLocked(page));
379 list_del(&page->lru);
381 dout("start_read %p adding %p idx %lu\n", inode, page,
383 if (add_to_page_cache_lru(page, &inode->i_data, page->index,
385 ceph_fscache_uncache_page(inode, page);
387 dout("start_read %p add_to_page_cache failed %p\n",
391 len = nr_pages << PAGE_SHIFT;
392 osd_req_op_extent_update(req, 0, len);
399 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false);
400 req->r_callback = finish_read;
401 req->r_inode = inode;
403 dout("start_read %p starting %p %lld~%lld\n", inode, req, off, len);
404 ret = ceph_osdc_start_request(osdc, req, false);
407 ceph_osdc_put_request(req);
409 /* After adding locked pages to page cache, the inode holds cache cap.
410 * So we can drop our cap refs. */
412 ceph_put_cap_refs(ci, got);
417 for (i = 0; i < nr_pages; ++i) {
418 ceph_fscache_readpage_cancel(inode, pages[i]);
419 unlock_page(pages[i]);
421 ceph_put_page_vector(pages, nr_pages, false);
423 ceph_osdc_put_request(req);
426 ceph_put_cap_refs(ci, got);
432 * Read multiple pages. Leave pages we don't read + unlock in page_list;
433 * the caller (VM) cleans them up.
435 static int ceph_readpages(struct file *file, struct address_space *mapping,
436 struct list_head *page_list, unsigned nr_pages)
438 struct inode *inode = file_inode(file);
439 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
443 if (ceph_inode(inode)->i_inline_version != CEPH_INLINE_NONE)
446 rc = ceph_readpages_from_fscache(mapping->host, mapping, page_list,
452 max = fsc->mount_options->rsize >> PAGE_SHIFT;
453 dout("readpages %p file %p nr_pages %d max %d\n",
454 inode, file, nr_pages, max);
455 while (!list_empty(page_list)) {
456 rc = start_read(inode, page_list, max);
461 ceph_fscache_readpages_cancel(inode, page_list);
463 dout("readpages %p file %p ret %d\n", inode, file, rc);
467 struct ceph_writeback_ctl
477 * Get ref for the oldest snapc for an inode with dirty data... that is, the
478 * only snap context we are allowed to write back.
480 static struct ceph_snap_context *
481 get_oldest_context(struct inode *inode, struct ceph_writeback_ctl *ctl,
482 struct ceph_snap_context *page_snapc)
484 struct ceph_inode_info *ci = ceph_inode(inode);
485 struct ceph_snap_context *snapc = NULL;
486 struct ceph_cap_snap *capsnap = NULL;
488 spin_lock(&ci->i_ceph_lock);
489 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
490 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap,
491 capsnap->context, capsnap->dirty_pages);
492 if (!capsnap->dirty_pages)
495 /* get i_size, truncate_{seq,size} for page_snapc? */
496 if (snapc && capsnap->context != page_snapc)
500 if (capsnap->writing) {
501 ctl->i_size = i_size_read(inode);
502 ctl->size_stable = false;
504 ctl->i_size = capsnap->size;
505 ctl->size_stable = true;
507 ctl->truncate_size = capsnap->truncate_size;
508 ctl->truncate_seq = capsnap->truncate_seq;
509 ctl->head_snapc = false;
515 snapc = ceph_get_snap_context(capsnap->context);
517 page_snapc == snapc ||
518 page_snapc->seq > snapc->seq)
521 if (!snapc && ci->i_wrbuffer_ref_head) {
522 snapc = ceph_get_snap_context(ci->i_head_snapc);
523 dout(" head snapc %p has %d dirty pages\n",
524 snapc, ci->i_wrbuffer_ref_head);
526 ctl->i_size = i_size_read(inode);
527 ctl->truncate_size = ci->i_truncate_size;
528 ctl->truncate_seq = ci->i_truncate_seq;
529 ctl->size_stable = false;
530 ctl->head_snapc = true;
533 spin_unlock(&ci->i_ceph_lock);
537 static u64 get_writepages_data_length(struct inode *inode,
538 struct page *page, u64 start)
540 struct ceph_inode_info *ci = ceph_inode(inode);
541 struct ceph_snap_context *snapc = page_snap_context(page);
542 struct ceph_cap_snap *capsnap = NULL;
543 u64 end = i_size_read(inode);
545 if (snapc != ci->i_head_snapc) {
547 spin_lock(&ci->i_ceph_lock);
548 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
549 if (capsnap->context == snapc) {
550 if (!capsnap->writing)
556 spin_unlock(&ci->i_ceph_lock);
559 if (end > page_offset(page) + PAGE_SIZE)
560 end = page_offset(page) + PAGE_SIZE;
561 return end > start ? end - start : 0;
565 * Write a single page, but leave the page locked.
567 * If we get a write error, set the page error bit, but still adjust the
568 * dirty page accounting (i.e., page is no longer dirty).
570 static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
573 struct ceph_inode_info *ci;
574 struct ceph_fs_client *fsc;
575 struct ceph_snap_context *snapc, *oldest;
576 loff_t page_off = page_offset(page);
578 int err, len = PAGE_SIZE;
579 struct ceph_writeback_ctl ceph_wbc;
581 dout("writepage %p idx %lu\n", page, page->index);
583 inode = page->mapping->host;
584 ci = ceph_inode(inode);
585 fsc = ceph_inode_to_client(inode);
587 /* verify this is a writeable snap context */
588 snapc = page_snap_context(page);
590 dout("writepage %p page %p not dirty?\n", inode, page);
593 oldest = get_oldest_context(inode, &ceph_wbc, snapc);
594 if (snapc->seq > oldest->seq) {
595 dout("writepage %p page %p snapc %p not writeable - noop\n",
597 /* we should only noop if called by kswapd */
598 WARN_ON(!(current->flags & PF_MEMALLOC));
599 ceph_put_snap_context(oldest);
600 redirty_page_for_writepage(wbc, page);
603 ceph_put_snap_context(oldest);
605 /* is this a partial page at end of file? */
606 if (page_off >= ceph_wbc.i_size) {
607 dout("%p page eof %llu\n", page, ceph_wbc.i_size);
608 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
612 if (ceph_wbc.i_size < page_off + len)
613 len = ceph_wbc.i_size - page_off;
615 dout("writepage %p page %p index %lu on %llu~%u snapc %p seq %lld\n",
616 inode, page, page->index, page_off, len, snapc, snapc->seq);
618 writeback_stat = atomic_long_inc_return(&fsc->writeback_count);
620 CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb))
621 set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC);
623 set_page_writeback(page);
624 err = ceph_osdc_writepages(&fsc->client->osdc, ceph_vino(inode),
625 &ci->i_layout, snapc, page_off, len,
626 ceph_wbc.truncate_seq,
627 ceph_wbc.truncate_size,
628 &inode->i_mtime, &page, 1);
630 struct writeback_control tmp_wbc;
633 if (err == -ERESTARTSYS) {
634 /* killed by SIGKILL */
635 dout("writepage interrupted page %p\n", page);
636 redirty_page_for_writepage(wbc, page);
637 end_page_writeback(page);
640 dout("writepage setting page/mapping error %d %p\n",
643 mapping_set_error(&inode->i_data, err);
644 wbc->pages_skipped++;
646 dout("writepage cleaned page %p\n", page);
647 err = 0; /* vfs expects us to return 0 */
650 ClearPagePrivate(page);
651 end_page_writeback(page);
652 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
653 ceph_put_snap_context(snapc); /* page's reference */
657 static int ceph_writepage(struct page *page, struct writeback_control *wbc)
660 struct inode *inode = page->mapping->host;
663 err = writepage_nounlock(page, wbc);
664 if (err == -ERESTARTSYS) {
665 /* direct memory reclaimer was killed by SIGKILL. return 0
666 * to prevent caller from setting mapping/page error */
675 * lame release_pages helper. release_pages() isn't exported to
678 static void ceph_release_pages(struct page **pages, int num)
683 pagevec_init(&pvec, 0);
684 for (i = 0; i < num; i++) {
685 if (pagevec_add(&pvec, pages[i]) == 0)
686 pagevec_release(&pvec);
688 pagevec_release(&pvec);
692 * async writeback completion handler.
694 * If we get an error, set the mapping error bit, but not the individual
697 static void writepages_finish(struct ceph_osd_request *req)
699 struct inode *inode = req->r_inode;
700 struct ceph_inode_info *ci = ceph_inode(inode);
701 struct ceph_osd_data *osd_data;
703 int num_pages, total_pages = 0;
705 int rc = req->r_result;
706 struct ceph_snap_context *snapc = req->r_snapc;
707 struct address_space *mapping = inode->i_mapping;
708 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
711 dout("writepages_finish %p rc %d\n", inode, rc);
713 mapping_set_error(mapping, rc);
714 ceph_set_error_write(ci);
716 ceph_clear_error_write(ci);
720 * We lost the cache cap, need to truncate the page before
721 * it is unlocked, otherwise we'd truncate it later in the
722 * page truncation thread, possibly losing some data that
725 remove_page = !(ceph_caps_issued(ci) &
726 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO));
728 /* clean all pages */
729 for (i = 0; i < req->r_num_ops; i++) {
730 if (req->r_ops[i].op != CEPH_OSD_OP_WRITE)
733 osd_data = osd_req_op_extent_osd_data(req, i);
734 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
735 num_pages = calc_pages_for((u64)osd_data->alignment,
736 (u64)osd_data->length);
737 total_pages += num_pages;
738 for (j = 0; j < num_pages; j++) {
739 page = osd_data->pages[j];
741 WARN_ON(!PageUptodate(page));
743 if (atomic_long_dec_return(&fsc->writeback_count) <
744 CONGESTION_OFF_THRESH(
745 fsc->mount_options->congestion_kb))
746 clear_bdi_congested(inode_to_bdi(inode),
749 ceph_put_snap_context(page_snap_context(page));
751 ClearPagePrivate(page);
752 dout("unlocking %p\n", page);
753 end_page_writeback(page);
756 generic_error_remove_page(inode->i_mapping,
761 dout("writepages_finish %p wrote %llu bytes cleaned %d pages\n",
762 inode, osd_data->length, rc >= 0 ? num_pages : 0);
764 ceph_release_pages(osd_data->pages, num_pages);
767 ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc);
769 osd_data = osd_req_op_extent_osd_data(req, 0);
770 if (osd_data->pages_from_pool)
771 mempool_free(osd_data->pages,
772 ceph_sb_to_client(inode->i_sb)->wb_pagevec_pool);
774 kfree(osd_data->pages);
775 ceph_osdc_put_request(req);
779 * initiate async writeback
781 static int ceph_writepages_start(struct address_space *mapping,
782 struct writeback_control *wbc)
784 struct inode *inode = mapping->host;
785 struct ceph_inode_info *ci = ceph_inode(inode);
786 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
787 struct ceph_vino vino = ceph_vino(inode);
788 pgoff_t index, start_index, end = -1;
789 struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc;
792 unsigned int wsize = i_blocksize(inode);
793 struct ceph_osd_request *req = NULL;
794 struct ceph_writeback_ctl ceph_wbc;
795 bool should_loop, range_whole = false;
796 bool stop, done = false;
798 dout("writepages_start %p (mode=%s)\n", inode,
799 wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
800 (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
802 if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
803 if (ci->i_wrbuffer_ref > 0) {
805 "writepage_start %p %lld forced umount\n",
806 inode, ceph_ino(inode));
808 mapping_set_error(mapping, -EIO);
809 return -EIO; /* we're in a forced umount, don't write! */
811 if (fsc->mount_options->wsize < wsize)
812 wsize = fsc->mount_options->wsize;
814 pagevec_init(&pvec, 0);
816 start_index = wbc->range_cyclic ? mapping->writeback_index : 0;
820 /* find oldest snap context with dirty data */
821 snapc = get_oldest_context(inode, &ceph_wbc, NULL);
823 /* hmm, why does writepages get called when there
825 dout(" no snap context with dirty data?\n");
828 dout(" oldest snapc is %p seq %lld (%d snaps)\n",
829 snapc, snapc->seq, snapc->num_snaps);
832 if (ceph_wbc.head_snapc && snapc != last_snapc) {
833 /* where to start/end? */
834 if (wbc->range_cyclic) {
839 dout(" cyclic, start at %lu\n", index);
841 index = wbc->range_start >> PAGE_SHIFT;
842 end = wbc->range_end >> PAGE_SHIFT;
843 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
845 dout(" not cyclic, %lu to %lu\n", index, end);
847 } else if (!ceph_wbc.head_snapc) {
848 /* Do not respect wbc->range_{start,end}. Dirty pages
849 * in that range can be associated with newer snapc.
850 * They are not writeable until we write all dirty pages
851 * associated with 'snapc' get written */
852 if (index > 0 || wbc->sync_mode != WB_SYNC_NONE)
854 dout(" non-head snapc, range whole\n");
857 ceph_put_snap_context(last_snapc);
861 while (!stop && index <= end) {
862 int num_ops = 0, op_idx;
863 unsigned i, pvec_pages, max_pages, locked_pages = 0;
864 struct page **pages = NULL, **data_pages;
865 mempool_t *pool = NULL; /* Becomes non-null if mempool used */
867 pgoff_t strip_unit_end = 0;
868 u64 offset = 0, len = 0;
870 max_pages = wsize >> PAGE_SHIFT;
873 pvec_pages = min_t(unsigned, PAGEVEC_SIZE,
874 max_pages - locked_pages);
875 if (end - index < (u64)(pvec_pages - 1))
876 pvec_pages = (unsigned)(end - index) + 1;
878 pvec_pages = pagevec_lookup_tag(&pvec, mapping, &index,
881 dout("pagevec_lookup_tag got %d\n", pvec_pages);
882 if (!pvec_pages && !locked_pages)
884 for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) {
885 page = pvec.pages[i];
886 dout("? %p idx %lu\n", page, page->index);
887 if (locked_pages == 0)
888 lock_page(page); /* first page */
889 else if (!trylock_page(page))
892 /* only dirty pages, or our accounting breaks */
893 if (unlikely(!PageDirty(page)) ||
894 unlikely(page->mapping != mapping)) {
895 dout("!dirty or !mapping %p\n", page);
899 if (page->index > end) {
900 dout("end of range %p\n", page);
901 /* can't be range_cyclic (1st pass) because
902 * end == -1 in that case. */
904 if (ceph_wbc.head_snapc)
909 if (strip_unit_end && (page->index > strip_unit_end)) {
910 dout("end of strip unit %p\n", page);
914 if (page_offset(page) >= ceph_wbc.i_size) {
915 dout("%p page eof %llu\n",
916 page, ceph_wbc.i_size);
917 /* not done if range_cyclic */
922 if (PageWriteback(page)) {
923 if (wbc->sync_mode == WB_SYNC_NONE) {
924 dout("%p under writeback\n", page);
928 dout("waiting on writeback %p\n", page);
929 wait_on_page_writeback(page);
932 /* only if matching snap context */
933 pgsnapc = page_snap_context(page);
934 if (pgsnapc != snapc) {
935 dout("page snapc %p %lld != oldest %p %lld\n",
936 pgsnapc, pgsnapc->seq, snapc, snapc->seq);
941 if (!clear_page_dirty_for_io(page)) {
942 dout("%p !clear_page_dirty_for_io\n", page);
948 * We have something to write. If this is
949 * the first locked page this time through,
950 * calculate max possinle write size and
951 * allocate a page array
953 if (locked_pages == 0) {
957 /* prepare async write request */
958 offset = (u64)page_offset(page);
961 rc = ceph_calc_file_object_mapping(&ci->i_layout,
971 strip_unit_end = page->index +
972 ((len - 1) >> PAGE_SHIFT);
975 max_pages = calc_pages_for(0, (u64)len);
976 pages = kmalloc(max_pages * sizeof (*pages),
979 pool = fsc->wb_pagevec_pool;
980 pages = mempool_alloc(pool, GFP_NOFS);
985 } else if (page->index !=
986 (offset + len) >> PAGE_SHIFT) {
987 if (num_ops >= (pool ? CEPH_OSD_SLAB_OPS :
989 redirty_page_for_writepage(wbc, page);
995 offset = (u64)page_offset(page);
999 /* note position of first page in pvec */
1000 dout("%p will write page %p idx %lu\n",
1001 inode, page, page->index);
1003 if (atomic_long_inc_return(&fsc->writeback_count) >
1004 CONGESTION_ON_THRESH(
1005 fsc->mount_options->congestion_kb)) {
1006 set_bdi_congested(inode_to_bdi(inode),
1011 pages[locked_pages++] = page;
1012 pvec.pages[i] = NULL;
1017 /* did we get anything? */
1019 goto release_pvec_pages;
1022 /* shift unused page to beginning of pvec */
1023 for (j = 0; j < pvec_pages; j++) {
1027 pvec.pages[n] = pvec.pages[j];
1032 if (pvec_pages && i == pvec_pages &&
1033 locked_pages < max_pages) {
1034 dout("reached end pvec, trying for more\n");
1035 pagevec_release(&pvec);
1036 goto get_more_pages;
1041 offset = page_offset(pages[0]);
1044 req = ceph_osdc_new_request(&fsc->client->osdc,
1045 &ci->i_layout, vino,
1046 offset, &len, 0, num_ops,
1047 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
1048 snapc, ceph_wbc.truncate_seq,
1049 ceph_wbc.truncate_size, false);
1051 req = ceph_osdc_new_request(&fsc->client->osdc,
1052 &ci->i_layout, vino,
1057 CEPH_OSD_FLAG_WRITE,
1058 snapc, ceph_wbc.truncate_seq,
1059 ceph_wbc.truncate_size, true);
1060 BUG_ON(IS_ERR(req));
1062 BUG_ON(len < page_offset(pages[locked_pages - 1]) +
1063 PAGE_SIZE - offset);
1065 req->r_callback = writepages_finish;
1066 req->r_inode = inode;
1068 /* Format the osd request message and submit the write */
1072 for (i = 0; i < locked_pages; i++) {
1073 u64 cur_offset = page_offset(pages[i]);
1074 if (offset + len != cur_offset) {
1075 if (op_idx + 1 == req->r_num_ops)
1077 osd_req_op_extent_dup_last(req, op_idx,
1078 cur_offset - offset);
1079 dout("writepages got pages at %llu~%llu\n",
1081 osd_req_op_extent_osd_data_pages(req, op_idx,
1084 osd_req_op_extent_update(req, op_idx, len);
1087 offset = cur_offset;
1088 data_pages = pages + i;
1092 set_page_writeback(pages[i]);
1096 if (ceph_wbc.size_stable) {
1097 len = min(len, ceph_wbc.i_size - offset);
1098 } else if (i == locked_pages) {
1099 /* writepages_finish() clears writeback pages
1100 * according to the data length, so make sure
1101 * data length covers all locked pages */
1102 u64 min_len = len + 1 - PAGE_SIZE;
1103 len = get_writepages_data_length(inode, pages[i - 1],
1105 len = max(len, min_len);
1107 dout("writepages got pages at %llu~%llu\n", offset, len);
1109 osd_req_op_extent_osd_data_pages(req, op_idx, data_pages, len,
1111 osd_req_op_extent_update(req, op_idx, len);
1113 BUG_ON(op_idx + 1 != req->r_num_ops);
1116 if (i < locked_pages) {
1117 BUG_ON(num_ops <= req->r_num_ops);
1118 num_ops -= req->r_num_ops;
1121 /* allocate new pages array for next request */
1123 pages = kmalloc(locked_pages * sizeof (*pages),
1126 pool = fsc->wb_pagevec_pool;
1127 pages = mempool_alloc(pool, GFP_NOFS);
1130 memcpy(pages, data_pages + i,
1131 locked_pages * sizeof(*pages));
1132 memset(data_pages + i, 0,
1133 locked_pages * sizeof(*pages));
1135 BUG_ON(num_ops != req->r_num_ops);
1136 index = pages[i - 1]->index + 1;
1137 /* request message now owns the pages array */
1141 req->r_mtime = inode->i_mtime;
1142 rc = ceph_osdc_start_request(&fsc->client->osdc, req, true);
1146 wbc->nr_to_write -= i;
1151 * We stop writing back only if we are not doing
1152 * integrity sync. In case of integrity sync we have to
1153 * keep going until we have written all the pages
1154 * we tagged for writeback prior to entering this loop.
1156 if (wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE)
1160 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr,
1161 pvec.nr ? pvec.pages[0] : NULL);
1162 pagevec_release(&pvec);
1165 if (should_loop && !done) {
1166 /* more to do; loop back to beginning of file */
1167 dout("writepages looping back to beginning of file\n");
1168 end = start_index - 1; /* OK even when start_index == 0 */
1170 /* to write dirty pages associated with next snapc,
1171 * we need to wait until current writes complete */
1172 if (wbc->sync_mode != WB_SYNC_NONE &&
1173 start_index == 0 && /* all dirty pages were checked */
1174 !ceph_wbc.head_snapc) {
1178 while ((index <= end) &&
1179 (nr = pagevec_lookup_tag(&pvec, mapping, &index,
1180 PAGECACHE_TAG_WRITEBACK,
1182 for (i = 0; i < nr; i++) {
1183 page = pvec.pages[i];
1184 if (page_snap_context(page) != snapc)
1186 wait_on_page_writeback(page);
1188 pagevec_release(&pvec);
1198 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1199 mapping->writeback_index = index;
1202 ceph_osdc_put_request(req);
1203 ceph_put_snap_context(last_snapc);
1204 dout("writepages dend - startone, rc = %d\n", rc);
1211 * See if a given @snapc is either writeable, or already written.
1213 static int context_is_writeable_or_written(struct inode *inode,
1214 struct ceph_snap_context *snapc)
1216 struct ceph_snap_context *oldest = get_oldest_context(inode, NULL, NULL);
1217 int ret = !oldest || snapc->seq <= oldest->seq;
1219 ceph_put_snap_context(oldest);
1224 * We are only allowed to write into/dirty the page if the page is
1225 * clean, or already dirty within the same snap context.
1227 * called with page locked.
1228 * return success with page locked,
1229 * or any failure (incl -EAGAIN) with page unlocked.
1231 static int ceph_update_writeable_page(struct file *file,
1232 loff_t pos, unsigned len,
1235 struct inode *inode = file_inode(file);
1236 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1237 struct ceph_inode_info *ci = ceph_inode(inode);
1238 loff_t page_off = pos & PAGE_MASK;
1239 int pos_in_page = pos & ~PAGE_MASK;
1240 int end_in_page = pos_in_page + len;
1243 struct ceph_snap_context *snapc, *oldest;
1245 if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1246 dout(" page %p forced umount\n", page);
1252 /* writepages currently holds page lock, but if we change that later, */
1253 wait_on_page_writeback(page);
1255 snapc = page_snap_context(page);
1256 if (snapc && snapc != ci->i_head_snapc) {
1258 * this page is already dirty in another (older) snap
1259 * context! is it writeable now?
1261 oldest = get_oldest_context(inode, NULL, NULL);
1262 if (snapc->seq > oldest->seq) {
1263 ceph_put_snap_context(oldest);
1264 dout(" page %p snapc %p not current or oldest\n",
1267 * queue for writeback, and wait for snapc to
1268 * be writeable or written
1270 snapc = ceph_get_snap_context(snapc);
1272 ceph_queue_writeback(inode);
1273 r = wait_event_killable(ci->i_cap_wq,
1274 context_is_writeable_or_written(inode, snapc));
1275 ceph_put_snap_context(snapc);
1276 if (r == -ERESTARTSYS)
1280 ceph_put_snap_context(oldest);
1282 /* yay, writeable, do it now (without dropping page lock) */
1283 dout(" page %p snapc %p not current, but oldest\n",
1285 if (!clear_page_dirty_for_io(page))
1287 r = writepage_nounlock(page, NULL);
1293 if (PageUptodate(page)) {
1294 dout(" page %p already uptodate\n", page);
1299 if (pos_in_page == 0 && len == PAGE_SIZE)
1302 /* past end of file? */
1303 i_size = i_size_read(inode);
1305 if (page_off >= i_size ||
1306 (pos_in_page == 0 && (pos+len) >= i_size &&
1307 end_in_page - pos_in_page != PAGE_SIZE)) {
1308 dout(" zeroing %p 0 - %d and %d - %d\n",
1309 page, pos_in_page, end_in_page, (int)PAGE_SIZE);
1310 zero_user_segments(page,
1312 end_in_page, PAGE_SIZE);
1316 /* we need to read it. */
1317 r = ceph_do_readpage(file, page);
1319 if (r == -EINPROGRESS)
1330 * We are only allowed to write into/dirty the page if the page is
1331 * clean, or already dirty within the same snap context.
1333 static int ceph_write_begin(struct file *file, struct address_space *mapping,
1334 loff_t pos, unsigned len, unsigned flags,
1335 struct page **pagep, void **fsdata)
1337 struct inode *inode = file_inode(file);
1339 pgoff_t index = pos >> PAGE_SHIFT;
1344 page = grab_cache_page_write_begin(mapping, index, 0);
1348 dout("write_begin file %p inode %p page %p %d~%d\n", file,
1349 inode, page, (int)pos, (int)len);
1351 r = ceph_update_writeable_page(file, pos, len, page);
1356 } while (r == -EAGAIN);
1362 * we don't do anything in here that simple_write_end doesn't do
1363 * except adjust dirty page accounting
1365 static int ceph_write_end(struct file *file, struct address_space *mapping,
1366 loff_t pos, unsigned len, unsigned copied,
1367 struct page *page, void *fsdata)
1369 struct inode *inode = file_inode(file);
1370 bool check_cap = false;
1372 dout("write_end file %p inode %p page %p %d~%d (%d)\n", file,
1373 inode, page, (int)pos, (int)copied, (int)len);
1375 /* zero the stale part of the page if we did a short copy */
1376 if (!PageUptodate(page)) {
1381 SetPageUptodate(page);
1384 /* did file size increase? */
1385 if (pos+copied > i_size_read(inode))
1386 check_cap = ceph_inode_set_size(inode, pos+copied);
1388 set_page_dirty(page);
1395 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL);
1401 * we set .direct_IO to indicate direct io is supported, but since we
1402 * intercept O_DIRECT reads and writes early, this function should
1405 static ssize_t ceph_direct_io(struct kiocb *iocb, struct iov_iter *iter)
1411 const struct address_space_operations ceph_aops = {
1412 .readpage = ceph_readpage,
1413 .readpages = ceph_readpages,
1414 .writepage = ceph_writepage,
1415 .writepages = ceph_writepages_start,
1416 .write_begin = ceph_write_begin,
1417 .write_end = ceph_write_end,
1418 .set_page_dirty = ceph_set_page_dirty,
1419 .invalidatepage = ceph_invalidatepage,
1420 .releasepage = ceph_releasepage,
1421 .direct_IO = ceph_direct_io,
1424 static void ceph_block_sigs(sigset_t *oldset)
1427 siginitsetinv(&mask, sigmask(SIGKILL));
1428 sigprocmask(SIG_BLOCK, &mask, oldset);
1431 static void ceph_restore_sigs(sigset_t *oldset)
1433 sigprocmask(SIG_SETMASK, oldset, NULL);
1439 static int ceph_filemap_fault(struct vm_fault *vmf)
1441 struct vm_area_struct *vma = vmf->vma;
1442 struct inode *inode = file_inode(vma->vm_file);
1443 struct ceph_inode_info *ci = ceph_inode(inode);
1444 struct ceph_file_info *fi = vma->vm_file->private_data;
1445 struct page *pinned_page = NULL;
1446 loff_t off = vmf->pgoff << PAGE_SHIFT;
1450 ceph_block_sigs(&oldset);
1452 dout("filemap_fault %p %llx.%llx %llu~%zd trying to get caps\n",
1453 inode, ceph_vinop(inode), off, (size_t)PAGE_SIZE);
1454 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1455 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1457 want = CEPH_CAP_FILE_CACHE;
1460 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page);
1464 dout("filemap_fault %p %llu~%zd got cap refs on %s\n",
1465 inode, off, (size_t)PAGE_SIZE, ceph_cap_string(got));
1467 if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) ||
1468 ci->i_inline_version == CEPH_INLINE_NONE) {
1469 current->journal_info = vma->vm_file;
1470 ret = filemap_fault(vmf);
1471 current->journal_info = NULL;
1475 dout("filemap_fault %p %llu~%zd dropping cap refs on %s ret %d\n",
1476 inode, off, (size_t)PAGE_SIZE, ceph_cap_string(got), ret);
1478 put_page(pinned_page);
1479 ceph_put_cap_refs(ci, got);
1484 /* read inline data */
1485 if (off >= PAGE_SIZE) {
1486 /* does not support inline data > PAGE_SIZE */
1487 ret = VM_FAULT_SIGBUS;
1490 struct address_space *mapping = inode->i_mapping;
1491 struct page *page = find_or_create_page(mapping, 0,
1492 mapping_gfp_constraint(mapping,
1498 ret1 = __ceph_do_getattr(inode, page,
1499 CEPH_STAT_CAP_INLINE_DATA, true);
1500 if (ret1 < 0 || off >= i_size_read(inode)) {
1506 ret = VM_FAULT_SIGBUS;
1509 if (ret1 < PAGE_SIZE)
1510 zero_user_segment(page, ret1, PAGE_SIZE);
1512 flush_dcache_page(page);
1513 SetPageUptodate(page);
1515 ret = VM_FAULT_MAJOR | VM_FAULT_LOCKED;
1517 dout("filemap_fault %p %llu~%zd read inline data ret %d\n",
1518 inode, off, (size_t)PAGE_SIZE, ret);
1521 ceph_restore_sigs(&oldset);
1523 ret = (ret == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS;
1529 * Reuse write_begin here for simplicity.
1531 static int ceph_page_mkwrite(struct vm_fault *vmf)
1533 struct vm_area_struct *vma = vmf->vma;
1534 struct inode *inode = file_inode(vma->vm_file);
1535 struct ceph_inode_info *ci = ceph_inode(inode);
1536 struct ceph_file_info *fi = vma->vm_file->private_data;
1537 struct ceph_cap_flush *prealloc_cf;
1538 struct page *page = vmf->page;
1539 loff_t off = page_offset(page);
1540 loff_t size = i_size_read(inode);
1545 prealloc_cf = ceph_alloc_cap_flush();
1547 return VM_FAULT_OOM;
1549 ceph_block_sigs(&oldset);
1551 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1552 struct page *locked_page = NULL;
1557 ret = ceph_uninline_data(vma->vm_file, locked_page);
1559 unlock_page(locked_page);
1564 if (off + PAGE_SIZE <= size)
1567 len = size & ~PAGE_MASK;
1569 dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n",
1570 inode, ceph_vinop(inode), off, len, size);
1571 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1572 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1574 want = CEPH_CAP_FILE_BUFFER;
1577 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, off + len,
1582 dout("page_mkwrite %p %llu~%zd got cap refs on %s\n",
1583 inode, off, len, ceph_cap_string(got));
1585 /* Update time before taking page lock */
1586 file_update_time(vma->vm_file);
1591 if ((off > size) || (page->mapping != inode->i_mapping)) {
1593 ret = VM_FAULT_NOPAGE;
1597 ret = ceph_update_writeable_page(vma->vm_file, off, len, page);
1599 /* success. we'll keep the page locked. */
1600 set_page_dirty(page);
1601 ret = VM_FAULT_LOCKED;
1603 } while (ret == -EAGAIN);
1605 if (ret == VM_FAULT_LOCKED ||
1606 ci->i_inline_version != CEPH_INLINE_NONE) {
1608 spin_lock(&ci->i_ceph_lock);
1609 ci->i_inline_version = CEPH_INLINE_NONE;
1610 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1612 spin_unlock(&ci->i_ceph_lock);
1614 __mark_inode_dirty(inode, dirty);
1617 dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %d\n",
1618 inode, off, len, ceph_cap_string(got), ret);
1619 ceph_put_cap_refs(ci, got);
1621 ceph_restore_sigs(&oldset);
1622 ceph_free_cap_flush(prealloc_cf);
1624 ret = (ret == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS;
1628 void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
1629 char *data, size_t len)
1631 struct address_space *mapping = inode->i_mapping;
1637 if (i_size_read(inode) == 0)
1639 page = find_or_create_page(mapping, 0,
1640 mapping_gfp_constraint(mapping,
1644 if (PageUptodate(page)) {
1651 dout("fill_inline_data %p %llx.%llx len %zu locked_page %p\n",
1652 inode, ceph_vinop(inode), len, locked_page);
1655 void *kaddr = kmap_atomic(page);
1656 memcpy(kaddr, data, len);
1657 kunmap_atomic(kaddr);
1660 if (page != locked_page) {
1661 if (len < PAGE_SIZE)
1662 zero_user_segment(page, len, PAGE_SIZE);
1664 flush_dcache_page(page);
1666 SetPageUptodate(page);
1672 int ceph_uninline_data(struct file *filp, struct page *locked_page)
1674 struct inode *inode = file_inode(filp);
1675 struct ceph_inode_info *ci = ceph_inode(inode);
1676 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1677 struct ceph_osd_request *req;
1678 struct page *page = NULL;
1679 u64 len, inline_version;
1681 bool from_pagecache = false;
1683 spin_lock(&ci->i_ceph_lock);
1684 inline_version = ci->i_inline_version;
1685 spin_unlock(&ci->i_ceph_lock);
1687 dout("uninline_data %p %llx.%llx inline_version %llu\n",
1688 inode, ceph_vinop(inode), inline_version);
1690 if (inline_version == 1 || /* initial version, no data */
1691 inline_version == CEPH_INLINE_NONE)
1696 WARN_ON(!PageUptodate(page));
1697 } else if (ceph_caps_issued(ci) &
1698 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) {
1699 page = find_get_page(inode->i_mapping, 0);
1701 if (PageUptodate(page)) {
1702 from_pagecache = true;
1712 len = i_size_read(inode);
1713 if (len > PAGE_SIZE)
1716 page = __page_cache_alloc(GFP_NOFS);
1721 err = __ceph_do_getattr(inode, page,
1722 CEPH_STAT_CAP_INLINE_DATA, true);
1724 /* no inline data */
1725 if (err == -ENODATA)
1732 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1733 ceph_vino(inode), 0, &len, 0, 1,
1734 CEPH_OSD_OP_CREATE, CEPH_OSD_FLAG_WRITE,
1741 req->r_mtime = inode->i_mtime;
1742 err = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1744 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1745 ceph_osdc_put_request(req);
1749 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1750 ceph_vino(inode), 0, &len, 1, 3,
1751 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
1752 NULL, ci->i_truncate_seq,
1753 ci->i_truncate_size, false);
1759 osd_req_op_extent_osd_data_pages(req, 1, &page, len, 0, false, false);
1762 __le64 xattr_buf = cpu_to_le64(inline_version);
1763 err = osd_req_op_xattr_init(req, 0, CEPH_OSD_OP_CMPXATTR,
1764 "inline_version", &xattr_buf,
1766 CEPH_OSD_CMPXATTR_OP_GT,
1767 CEPH_OSD_CMPXATTR_MODE_U64);
1774 int xattr_len = snprintf(xattr_buf, sizeof(xattr_buf),
1775 "%llu", inline_version);
1776 err = osd_req_op_xattr_init(req, 2, CEPH_OSD_OP_SETXATTR,
1778 xattr_buf, xattr_len, 0, 0);
1783 req->r_mtime = inode->i_mtime;
1784 err = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1786 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1788 ceph_osdc_put_request(req);
1789 if (err == -ECANCELED)
1792 if (page && page != locked_page) {
1793 if (from_pagecache) {
1797 __free_pages(page, 0);
1800 dout("uninline_data %p %llx.%llx inline_version %llu = %d\n",
1801 inode, ceph_vinop(inode), inline_version, err);
1805 static const struct vm_operations_struct ceph_vmops = {
1806 .fault = ceph_filemap_fault,
1807 .page_mkwrite = ceph_page_mkwrite,
1810 int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1812 struct address_space *mapping = file->f_mapping;
1814 if (!mapping->a_ops->readpage)
1816 file_accessed(file);
1817 vma->vm_ops = &ceph_vmops;
1826 static int __ceph_pool_perm_get(struct ceph_inode_info *ci,
1827 s64 pool, struct ceph_string *pool_ns)
1829 struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode);
1830 struct ceph_mds_client *mdsc = fsc->mdsc;
1831 struct ceph_osd_request *rd_req = NULL, *wr_req = NULL;
1832 struct rb_node **p, *parent;
1833 struct ceph_pool_perm *perm;
1834 struct page **pages;
1836 int err = 0, err2 = 0, have = 0;
1838 down_read(&mdsc->pool_perm_rwsem);
1839 p = &mdsc->pool_perm_tree.rb_node;
1841 perm = rb_entry(*p, struct ceph_pool_perm, node);
1842 if (pool < perm->pool)
1844 else if (pool > perm->pool)
1845 p = &(*p)->rb_right;
1847 int ret = ceph_compare_string(pool_ns,
1853 p = &(*p)->rb_right;
1860 up_read(&mdsc->pool_perm_rwsem);
1865 dout("__ceph_pool_perm_get pool %lld ns %.*s no perm cached\n",
1866 pool, (int)pool_ns->len, pool_ns->str);
1868 dout("__ceph_pool_perm_get pool %lld no perm cached\n", pool);
1870 down_write(&mdsc->pool_perm_rwsem);
1871 p = &mdsc->pool_perm_tree.rb_node;
1875 perm = rb_entry(parent, struct ceph_pool_perm, node);
1876 if (pool < perm->pool)
1878 else if (pool > perm->pool)
1879 p = &(*p)->rb_right;
1881 int ret = ceph_compare_string(pool_ns,
1887 p = &(*p)->rb_right;
1895 up_write(&mdsc->pool_perm_rwsem);
1899 rd_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1900 1, false, GFP_NOFS);
1906 rd_req->r_flags = CEPH_OSD_FLAG_READ;
1907 osd_req_op_init(rd_req, 0, CEPH_OSD_OP_STAT, 0);
1908 rd_req->r_base_oloc.pool = pool;
1910 rd_req->r_base_oloc.pool_ns = ceph_get_string(pool_ns);
1911 ceph_oid_printf(&rd_req->r_base_oid, "%llx.00000000", ci->i_vino.ino);
1913 err = ceph_osdc_alloc_messages(rd_req, GFP_NOFS);
1917 wr_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1918 1, false, GFP_NOFS);
1924 wr_req->r_flags = CEPH_OSD_FLAG_WRITE;
1925 osd_req_op_init(wr_req, 0, CEPH_OSD_OP_CREATE, CEPH_OSD_OP_FLAG_EXCL);
1926 ceph_oloc_copy(&wr_req->r_base_oloc, &rd_req->r_base_oloc);
1927 ceph_oid_copy(&wr_req->r_base_oid, &rd_req->r_base_oid);
1929 err = ceph_osdc_alloc_messages(wr_req, GFP_NOFS);
1933 /* one page should be large enough for STAT data */
1934 pages = ceph_alloc_page_vector(1, GFP_KERNEL);
1935 if (IS_ERR(pages)) {
1936 err = PTR_ERR(pages);
1940 osd_req_op_raw_data_in_pages(rd_req, 0, pages, PAGE_SIZE,
1942 err = ceph_osdc_start_request(&fsc->client->osdc, rd_req, false);
1944 wr_req->r_mtime = ci->vfs_inode.i_mtime;
1945 wr_req->r_abort_on_full = true;
1946 err2 = ceph_osdc_start_request(&fsc->client->osdc, wr_req, false);
1949 err = ceph_osdc_wait_request(&fsc->client->osdc, rd_req);
1951 err2 = ceph_osdc_wait_request(&fsc->client->osdc, wr_req);
1953 if (err >= 0 || err == -ENOENT)
1955 else if (err != -EPERM)
1958 if (err2 == 0 || err2 == -EEXIST)
1960 else if (err2 != -EPERM) {
1965 pool_ns_len = pool_ns ? pool_ns->len : 0;
1966 perm = kmalloc(sizeof(*perm) + pool_ns_len + 1, GFP_NOFS);
1974 perm->pool_ns_len = pool_ns_len;
1975 if (pool_ns_len > 0)
1976 memcpy(perm->pool_ns, pool_ns->str, pool_ns_len);
1977 perm->pool_ns[pool_ns_len] = 0;
1979 rb_link_node(&perm->node, parent, p);
1980 rb_insert_color(&perm->node, &mdsc->pool_perm_tree);
1983 up_write(&mdsc->pool_perm_rwsem);
1985 ceph_osdc_put_request(rd_req);
1986 ceph_osdc_put_request(wr_req);
1991 dout("__ceph_pool_perm_get pool %lld ns %.*s result = %d\n",
1992 pool, (int)pool_ns->len, pool_ns->str, err);
1994 dout("__ceph_pool_perm_get pool %lld result = %d\n", pool, err);
1998 int ceph_pool_perm_check(struct ceph_inode_info *ci, int need)
2001 struct ceph_string *pool_ns;
2004 if (ci->i_vino.snap != CEPH_NOSNAP) {
2006 * Pool permission check needs to write to the first object.
2007 * But for snapshot, head of the first object may have alread
2008 * been deleted. Skip check to avoid creating orphan object.
2013 if (ceph_test_mount_opt(ceph_inode_to_client(&ci->vfs_inode),
2017 spin_lock(&ci->i_ceph_lock);
2018 flags = ci->i_ceph_flags;
2019 pool = ci->i_layout.pool_id;
2020 spin_unlock(&ci->i_ceph_lock);
2022 if (flags & CEPH_I_POOL_PERM) {
2023 if ((need & CEPH_CAP_FILE_RD) && !(flags & CEPH_I_POOL_RD)) {
2024 dout("ceph_pool_perm_check pool %lld no read perm\n",
2028 if ((need & CEPH_CAP_FILE_WR) && !(flags & CEPH_I_POOL_WR)) {
2029 dout("ceph_pool_perm_check pool %lld no write perm\n",
2036 pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
2037 ret = __ceph_pool_perm_get(ci, pool, pool_ns);
2038 ceph_put_string(pool_ns);
2042 flags = CEPH_I_POOL_PERM;
2043 if (ret & POOL_READ)
2044 flags |= CEPH_I_POOL_RD;
2045 if (ret & POOL_WRITE)
2046 flags |= CEPH_I_POOL_WR;
2048 spin_lock(&ci->i_ceph_lock);
2049 if (pool == ci->i_layout.pool_id &&
2050 pool_ns == rcu_dereference_raw(ci->i_layout.pool_ns)) {
2051 ci->i_ceph_flags |= flags;
2053 pool = ci->i_layout.pool_id;
2054 flags = ci->i_ceph_flags;
2056 spin_unlock(&ci->i_ceph_lock);
2060 void ceph_pool_perm_destroy(struct ceph_mds_client *mdsc)
2062 struct ceph_pool_perm *perm;
2065 while (!RB_EMPTY_ROOT(&mdsc->pool_perm_tree)) {
2066 n = rb_first(&mdsc->pool_perm_tree);
2067 perm = rb_entry(n, struct ceph_pool_perm, node);
2068 rb_erase(n, &mdsc->pool_perm_tree);