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;
76 if (PageDirty(page)) {
77 dout("%p set_page_dirty %p idx %lu -- already dirty\n",
78 mapping->host, page, page->index);
79 BUG_ON(!PagePrivate(page));
83 inode = mapping->host;
84 ci = ceph_inode(inode);
87 spin_lock(&ci->i_ceph_lock);
88 BUG_ON(ci->i_wr_ref == 0); // caller should hold Fw reference
89 if (__ceph_have_pending_cap_snap(ci)) {
90 struct ceph_cap_snap *capsnap =
91 list_last_entry(&ci->i_cap_snaps,
94 snapc = ceph_get_snap_context(capsnap->context);
95 capsnap->dirty_pages++;
97 BUG_ON(!ci->i_head_snapc);
98 snapc = ceph_get_snap_context(ci->i_head_snapc);
99 ++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 return __set_page_dirty_nobuffers(page);
124 * If we are truncating the full page (i.e. offset == 0), adjust the
125 * dirty page counters appropriately. Only called if there is private
128 static void ceph_invalidatepage(struct page *page, unsigned int offset,
132 struct ceph_inode_info *ci;
133 struct ceph_snap_context *snapc = page_snap_context(page);
135 inode = page->mapping->host;
136 ci = ceph_inode(inode);
138 if (offset != 0 || length != PAGE_SIZE) {
139 dout("%p invalidatepage %p idx %lu partial dirty page %u~%u\n",
140 inode, page, page->index, offset, length);
144 ceph_invalidate_fscache_page(inode, page);
146 if (!PagePrivate(page))
150 * We can get non-dirty pages here due to races between
151 * set_page_dirty and truncate_complete_page; just spit out a
152 * warning, in case we end up with accounting problems later.
154 if (!PageDirty(page))
155 pr_err("%p invalidatepage %p page not dirty\n", inode, page);
157 ClearPageChecked(page);
159 dout("%p invalidatepage %p idx %lu full dirty page\n",
160 inode, page, page->index);
162 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
163 ceph_put_snap_context(snapc);
165 ClearPagePrivate(page);
168 static int ceph_releasepage(struct page *page, gfp_t g)
170 dout("%p releasepage %p idx %lu (%sdirty)\n", page->mapping->host,
171 page, page->index, PageDirty(page) ? "" : "not ");
173 /* Can we release the page from the cache? */
174 if (!ceph_release_fscache_page(page, g))
177 return !PagePrivate(page);
181 * read a single page, without unlocking it.
183 static int ceph_do_readpage(struct file *filp, struct page *page)
185 struct inode *inode = file_inode(filp);
186 struct ceph_inode_info *ci = ceph_inode(inode);
187 struct ceph_osd_client *osdc =
188 &ceph_inode_to_client(inode)->client->osdc;
190 u64 off = page_offset(page);
193 if (off >= i_size_read(inode)) {
194 zero_user_segment(page, 0, PAGE_SIZE);
195 SetPageUptodate(page);
199 if (ci->i_inline_version != CEPH_INLINE_NONE) {
201 * Uptodate inline data should have been added
202 * into page cache while getting Fcr caps.
206 zero_user_segment(page, 0, PAGE_SIZE);
207 SetPageUptodate(page);
211 err = ceph_readpage_from_fscache(inode, page);
215 dout("readpage inode %p file %p page %p index %lu\n",
216 inode, filp, page, page->index);
217 err = ceph_osdc_readpages(osdc, ceph_vino(inode), &ci->i_layout,
219 ci->i_truncate_seq, ci->i_truncate_size,
225 ceph_fscache_readpage_cancel(inode, page);
229 /* zero fill remainder of page */
230 zero_user_segment(page, err, PAGE_SIZE);
232 flush_dcache_page(page);
234 SetPageUptodate(page);
235 ceph_readpage_to_fscache(inode, page);
238 return err < 0 ? err : 0;
241 static int ceph_readpage(struct file *filp, struct page *page)
243 int r = ceph_do_readpage(filp, page);
244 if (r != -EINPROGRESS)
252 * Finish an async read(ahead) op.
254 static void finish_read(struct ceph_osd_request *req)
256 struct inode *inode = req->r_inode;
257 struct ceph_osd_data *osd_data;
258 int rc = req->r_result <= 0 ? req->r_result : 0;
259 int bytes = req->r_result >= 0 ? req->r_result : 0;
263 dout("finish_read %p req %p rc %d bytes %d\n", inode, req, rc, bytes);
265 /* unlock all pages, zeroing any data we didn't read */
266 osd_data = osd_req_op_extent_osd_data(req, 0);
267 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
268 num_pages = calc_pages_for((u64)osd_data->alignment,
269 (u64)osd_data->length);
270 for (i = 0; i < num_pages; i++) {
271 struct page *page = osd_data->pages[i];
273 if (rc < 0 && rc != -ENOENT) {
274 ceph_fscache_readpage_cancel(inode, page);
277 if (bytes < (int)PAGE_SIZE) {
278 /* zero (remainder of) page */
279 int s = bytes < 0 ? 0 : bytes;
280 zero_user_segment(page, s, PAGE_SIZE);
282 dout("finish_read %p uptodate %p idx %lu\n", inode, page,
284 flush_dcache_page(page);
285 SetPageUptodate(page);
286 ceph_readpage_to_fscache(inode, page);
292 kfree(osd_data->pages);
296 * start an async read(ahead) operation. return nr_pages we submitted
297 * a read for on success, or negative error code.
299 static int start_read(struct inode *inode, struct list_head *page_list, int max)
301 struct ceph_osd_client *osdc =
302 &ceph_inode_to_client(inode)->client->osdc;
303 struct ceph_inode_info *ci = ceph_inode(inode);
304 struct page *page = list_entry(page_list->prev, struct page, lru);
305 struct ceph_vino vino;
306 struct ceph_osd_request *req;
316 if (!current->journal_info) {
317 /* caller of readpages does not hold buffer and read caps
318 * (fadvise, madvise and readahead cases) */
319 int want = CEPH_CAP_FILE_CACHE;
320 ret = ceph_try_get_caps(ci, CEPH_CAP_FILE_RD, want, &got);
322 dout("start_read %p, error getting cap\n", inode);
323 } else if (!(got & want)) {
324 dout("start_read %p, no cache cap\n", inode);
329 ceph_put_cap_refs(ci, got);
330 while (!list_empty(page_list)) {
331 page = list_entry(page_list->prev,
333 list_del(&page->lru);
340 off = (u64) page_offset(page);
343 next_index = page->index;
344 list_for_each_entry_reverse(page, page_list, lru) {
345 if (page->index != next_index)
349 if (max && nr_pages == max)
352 len = nr_pages << PAGE_SHIFT;
353 dout("start_read %p nr_pages %d is %lld~%lld\n", inode, nr_pages,
355 vino = ceph_vino(inode);
356 req = ceph_osdc_new_request(osdc, &ci->i_layout, vino, off, &len,
357 0, 1, CEPH_OSD_OP_READ,
358 CEPH_OSD_FLAG_READ, NULL,
359 ci->i_truncate_seq, ci->i_truncate_size,
366 /* build page vector */
367 nr_pages = calc_pages_for(0, len);
368 pages = kmalloc(sizeof(*pages) * nr_pages, GFP_KERNEL);
373 for (i = 0; i < nr_pages; ++i) {
374 page = list_entry(page_list->prev, struct page, lru);
375 BUG_ON(PageLocked(page));
376 list_del(&page->lru);
378 dout("start_read %p adding %p idx %lu\n", inode, page,
380 if (add_to_page_cache_lru(page, &inode->i_data, page->index,
382 ceph_fscache_uncache_page(inode, page);
384 dout("start_read %p add_to_page_cache failed %p\n",
388 len = nr_pages << PAGE_SHIFT;
389 osd_req_op_extent_update(req, 0, len);
396 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false);
397 req->r_callback = finish_read;
398 req->r_inode = inode;
400 dout("start_read %p starting %p %lld~%lld\n", inode, req, off, len);
401 ret = ceph_osdc_start_request(osdc, req, false);
404 ceph_osdc_put_request(req);
406 /* After adding locked pages to page cache, the inode holds cache cap.
407 * So we can drop our cap refs. */
409 ceph_put_cap_refs(ci, got);
414 for (i = 0; i < nr_pages; ++i) {
415 ceph_fscache_readpage_cancel(inode, pages[i]);
416 unlock_page(pages[i]);
418 ceph_put_page_vector(pages, nr_pages, false);
420 ceph_osdc_put_request(req);
423 ceph_put_cap_refs(ci, got);
429 * Read multiple pages. Leave pages we don't read + unlock in page_list;
430 * the caller (VM) cleans them up.
432 static int ceph_readpages(struct file *file, struct address_space *mapping,
433 struct list_head *page_list, unsigned nr_pages)
435 struct inode *inode = file_inode(file);
436 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
440 if (ceph_inode(inode)->i_inline_version != CEPH_INLINE_NONE)
443 rc = ceph_readpages_from_fscache(mapping->host, mapping, page_list,
449 if (fsc->mount_options->rsize >= PAGE_SIZE)
450 max = (fsc->mount_options->rsize + PAGE_SIZE - 1)
453 dout("readpages %p file %p nr_pages %d max %d\n", inode,
456 while (!list_empty(page_list)) {
457 rc = start_read(inode, page_list, max);
462 ceph_fscache_readpages_cancel(inode, page_list);
464 dout("readpages %p file %p ret %d\n", inode, file, rc);
469 * Get ref for the oldest snapc for an inode with dirty data... that is, the
470 * only snap context we are allowed to write back.
472 static struct ceph_snap_context *get_oldest_context(struct inode *inode,
475 struct ceph_inode_info *ci = ceph_inode(inode);
476 struct ceph_snap_context *snapc = NULL;
477 struct ceph_cap_snap *capsnap = NULL;
479 spin_lock(&ci->i_ceph_lock);
480 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
481 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap,
482 capsnap->context, capsnap->dirty_pages);
483 if (capsnap->dirty_pages) {
484 snapc = ceph_get_snap_context(capsnap->context);
486 *snap_size = capsnap->size;
490 if (!snapc && ci->i_wrbuffer_ref_head) {
491 snapc = ceph_get_snap_context(ci->i_head_snapc);
492 dout(" head snapc %p has %d dirty pages\n",
493 snapc, ci->i_wrbuffer_ref_head);
495 spin_unlock(&ci->i_ceph_lock);
500 * Write a single page, but leave the page locked.
502 * If we get a write error, set the page error bit, but still adjust the
503 * dirty page accounting (i.e., page is no longer dirty).
505 static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
508 struct ceph_inode_info *ci;
509 struct ceph_fs_client *fsc;
510 struct ceph_osd_client *osdc;
511 struct ceph_snap_context *snapc, *oldest;
512 loff_t page_off = page_offset(page);
513 loff_t snap_size = -1;
517 int err = 0, len = PAGE_SIZE;
519 dout("writepage %p idx %lu\n", page, page->index);
521 if (!page->mapping || !page->mapping->host) {
522 dout("writepage %p - no mapping\n", page);
525 inode = page->mapping->host;
526 ci = ceph_inode(inode);
527 fsc = ceph_inode_to_client(inode);
528 osdc = &fsc->client->osdc;
530 /* verify this is a writeable snap context */
531 snapc = page_snap_context(page);
533 dout("writepage %p page %p not dirty?\n", inode, page);
536 oldest = get_oldest_context(inode, &snap_size);
537 if (snapc->seq > oldest->seq) {
538 dout("writepage %p page %p snapc %p not writeable - noop\n",
540 /* we should only noop if called by kswapd */
541 WARN_ON((current->flags & PF_MEMALLOC) == 0);
542 ceph_put_snap_context(oldest);
545 ceph_put_snap_context(oldest);
547 spin_lock(&ci->i_ceph_lock);
548 truncate_seq = ci->i_truncate_seq;
549 truncate_size = ci->i_truncate_size;
551 snap_size = i_size_read(inode);
552 spin_unlock(&ci->i_ceph_lock);
554 /* is this a partial page at end of file? */
555 if (page_off >= snap_size) {
556 dout("%p page eof %llu\n", page, snap_size);
559 if (snap_size < page_off + len)
560 len = snap_size - page_off;
562 dout("writepage %p page %p index %lu on %llu~%u snapc %p\n",
563 inode, page, page->index, page_off, len, snapc);
565 writeback_stat = atomic_long_inc_return(&fsc->writeback_count);
567 CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb))
568 set_bdi_congested(&fsc->backing_dev_info, BLK_RW_ASYNC);
570 set_page_writeback(page);
571 err = ceph_osdc_writepages(osdc, ceph_vino(inode),
572 &ci->i_layout, snapc,
574 truncate_seq, truncate_size,
575 &inode->i_mtime, &page, 1);
577 struct writeback_control tmp_wbc;
580 if (err == -ERESTARTSYS) {
581 /* killed by SIGKILL */
582 dout("writepage interrupted page %p\n", page);
583 redirty_page_for_writepage(wbc, page);
584 end_page_writeback(page);
587 dout("writepage setting page/mapping error %d %p\n",
590 mapping_set_error(&inode->i_data, err);
591 wbc->pages_skipped++;
593 dout("writepage cleaned page %p\n", page);
594 err = 0; /* vfs expects us to return 0 */
597 ClearPagePrivate(page);
598 end_page_writeback(page);
599 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
600 ceph_put_snap_context(snapc); /* page's reference */
605 static int ceph_writepage(struct page *page, struct writeback_control *wbc)
608 struct inode *inode = page->mapping->host;
611 err = writepage_nounlock(page, wbc);
612 if (err == -ERESTARTSYS) {
613 /* direct memory reclaimer was killed by SIGKILL. return 0
614 * to prevent caller from setting mapping/page error */
623 * lame release_pages helper. release_pages() isn't exported to
626 static void ceph_release_pages(struct page **pages, int num)
631 pagevec_init(&pvec, 0);
632 for (i = 0; i < num; i++) {
633 if (pagevec_add(&pvec, pages[i]) == 0)
634 pagevec_release(&pvec);
636 pagevec_release(&pvec);
640 * async writeback completion handler.
642 * If we get an error, set the mapping error bit, but not the individual
645 static void writepages_finish(struct ceph_osd_request *req)
647 struct inode *inode = req->r_inode;
648 struct ceph_inode_info *ci = ceph_inode(inode);
649 struct ceph_osd_data *osd_data;
651 int num_pages, total_pages = 0;
653 int rc = req->r_result;
654 struct ceph_snap_context *snapc = req->r_snapc;
655 struct address_space *mapping = inode->i_mapping;
656 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
659 dout("writepages_finish %p rc %d\n", inode, rc);
661 mapping_set_error(mapping, rc);
664 * We lost the cache cap, need to truncate the page before
665 * it is unlocked, otherwise we'd truncate it later in the
666 * page truncation thread, possibly losing some data that
669 remove_page = !(ceph_caps_issued(ci) &
670 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO));
672 /* clean all pages */
673 for (i = 0; i < req->r_num_ops; i++) {
674 if (req->r_ops[i].op != CEPH_OSD_OP_WRITE)
677 osd_data = osd_req_op_extent_osd_data(req, i);
678 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
679 num_pages = calc_pages_for((u64)osd_data->alignment,
680 (u64)osd_data->length);
681 total_pages += num_pages;
682 for (j = 0; j < num_pages; j++) {
683 page = osd_data->pages[j];
685 WARN_ON(!PageUptodate(page));
687 if (atomic_long_dec_return(&fsc->writeback_count) <
688 CONGESTION_OFF_THRESH(
689 fsc->mount_options->congestion_kb))
690 clear_bdi_congested(&fsc->backing_dev_info,
696 ceph_put_snap_context(page_snap_context(page));
698 ClearPagePrivate(page);
699 dout("unlocking %p\n", page);
700 end_page_writeback(page);
703 generic_error_remove_page(inode->i_mapping,
708 dout("writepages_finish %p wrote %llu bytes cleaned %d pages\n",
709 inode, osd_data->length, rc >= 0 ? num_pages : 0);
711 ceph_release_pages(osd_data->pages, num_pages);
714 ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc);
716 osd_data = osd_req_op_extent_osd_data(req, 0);
717 if (osd_data->pages_from_pool)
718 mempool_free(osd_data->pages,
719 ceph_sb_to_client(inode->i_sb)->wb_pagevec_pool);
721 kfree(osd_data->pages);
722 ceph_osdc_put_request(req);
726 * initiate async writeback
728 static int ceph_writepages_start(struct address_space *mapping,
729 struct writeback_control *wbc)
731 struct inode *inode = mapping->host;
732 struct ceph_inode_info *ci = ceph_inode(inode);
733 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
734 struct ceph_vino vino = ceph_vino(inode);
735 pgoff_t index, start, end;
738 pgoff_t max_pages = 0, max_pages_ever = 0;
739 struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc;
743 unsigned int wsize = i_blocksize(inode);
744 struct ceph_osd_request *req = NULL;
746 loff_t snap_size, i_size;
751 * Include a 'sync' in the OSD request if this is a data
752 * integrity write (e.g., O_SYNC write or fsync()), or if our
753 * cap is being revoked.
755 if ((wbc->sync_mode == WB_SYNC_ALL) ||
756 ceph_caps_revoking(ci, CEPH_CAP_FILE_BUFFER))
758 dout("writepages_start %p dosync=%d (mode=%s)\n",
760 wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
761 (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
763 if (ACCESS_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
764 if (ci->i_wrbuffer_ref > 0) {
766 "writepage_start %p %lld forced umount\n",
767 inode, ceph_ino(inode));
769 mapping_set_error(mapping, -EIO);
770 return -EIO; /* we're in a forced umount, don't write! */
772 if (fsc->mount_options->wsize && fsc->mount_options->wsize < wsize)
773 wsize = fsc->mount_options->wsize;
774 if (wsize < PAGE_SIZE)
776 max_pages_ever = wsize >> PAGE_SHIFT;
778 pagevec_init(&pvec, 0);
780 /* where to start/end? */
781 if (wbc->range_cyclic) {
782 start = mapping->writeback_index; /* Start from prev offset */
784 dout(" cyclic, start at %lu\n", start);
786 start = wbc->range_start >> PAGE_SHIFT;
787 end = wbc->range_end >> PAGE_SHIFT;
788 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
791 dout(" not cyclic, %lu to %lu\n", start, end);
796 /* find oldest snap context with dirty data */
797 ceph_put_snap_context(snapc);
799 snapc = get_oldest_context(inode, &snap_size);
801 /* hmm, why does writepages get called when there
803 dout(" no snap context with dirty data?\n");
806 dout(" oldest snapc is %p seq %lld (%d snaps)\n",
807 snapc, snapc->seq, snapc->num_snaps);
809 spin_lock(&ci->i_ceph_lock);
810 truncate_seq = ci->i_truncate_seq;
811 truncate_size = ci->i_truncate_size;
812 i_size = i_size_read(inode);
813 spin_unlock(&ci->i_ceph_lock);
815 if (last_snapc && snapc != last_snapc) {
816 /* if we switched to a newer snapc, restart our scan at the
817 * start of the original file range. */
818 dout(" snapc differs from last pass, restarting at %lu\n",
824 while (!done && index <= end) {
827 pgoff_t strip_unit_end = 0;
828 int num_ops = 0, op_idx;
829 int pvec_pages, locked_pages = 0;
830 struct page **pages = NULL, **data_pages;
831 mempool_t *pool = NULL; /* Becomes non-null if mempool used */
834 u64 offset = 0, len = 0;
836 max_pages = max_pages_ever;
840 want = min(end - index,
841 min((pgoff_t)PAGEVEC_SIZE,
842 max_pages - (pgoff_t)locked_pages) - 1)
844 pvec_pages = pagevec_lookup_tag(&pvec, mapping, &index,
847 dout("pagevec_lookup_tag got %d\n", pvec_pages);
848 if (!pvec_pages && !locked_pages)
850 for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) {
851 page = pvec.pages[i];
852 dout("? %p idx %lu\n", page, page->index);
853 if (locked_pages == 0)
854 lock_page(page); /* first page */
855 else if (!trylock_page(page))
858 /* only dirty pages, or our accounting breaks */
859 if (unlikely(!PageDirty(page)) ||
860 unlikely(page->mapping != mapping)) {
861 dout("!dirty or !mapping %p\n", page);
865 if (!wbc->range_cyclic && page->index > end) {
866 dout("end of range %p\n", page);
871 if (strip_unit_end && (page->index > strip_unit_end)) {
872 dout("end of strip unit %p\n", page);
876 if (wbc->sync_mode != WB_SYNC_NONE) {
877 dout("waiting on writeback %p\n", page);
878 wait_on_page_writeback(page);
880 if (page_offset(page) >=
881 (snap_size == -1 ? i_size : snap_size)) {
882 dout("%p page eof %llu\n", page,
883 (snap_size == -1 ? i_size : snap_size));
888 if (PageWriteback(page)) {
889 dout("%p under writeback\n", page);
894 /* only if matching snap context */
895 pgsnapc = page_snap_context(page);
896 if (pgsnapc->seq > snapc->seq) {
897 dout("page snapc %p %lld > oldest %p %lld\n",
898 pgsnapc, pgsnapc->seq, snapc, snapc->seq);
901 continue; /* keep looking for snap */
905 if (!clear_page_dirty_for_io(page)) {
906 dout("%p !clear_page_dirty_for_io\n", page);
912 * We have something to write. If this is
913 * the first locked page this time through,
914 * calculate max possinle write size and
915 * allocate a page array
917 if (locked_pages == 0) {
921 /* prepare async write request */
922 offset = (u64)page_offset(page);
925 rc = ceph_calc_file_object_mapping(&ci->i_layout,
934 num_ops = 1 + do_sync;
935 strip_unit_end = page->index +
936 ((len - 1) >> PAGE_SHIFT);
939 max_pages = calc_pages_for(0, (u64)len);
940 pages = kmalloc(max_pages * sizeof (*pages),
943 pool = fsc->wb_pagevec_pool;
944 pages = mempool_alloc(pool, GFP_NOFS);
949 } else if (page->index !=
950 (offset + len) >> PAGE_SHIFT) {
951 if (num_ops >= (pool ? CEPH_OSD_SLAB_OPS :
953 redirty_page_for_writepage(wbc, page);
959 offset = (u64)page_offset(page);
963 /* note position of first page in pvec */
966 dout("%p will write page %p idx %lu\n",
967 inode, page, page->index);
969 if (atomic_long_inc_return(&fsc->writeback_count) >
970 CONGESTION_ON_THRESH(
971 fsc->mount_options->congestion_kb)) {
972 set_bdi_congested(&fsc->backing_dev_info,
976 pages[locked_pages] = page;
981 /* did we get anything? */
983 goto release_pvec_pages;
986 BUG_ON(!locked_pages || first < 0);
988 if (pvec_pages && i == pvec_pages &&
989 locked_pages < max_pages) {
990 dout("reached end pvec, trying for more\n");
991 pagevec_reinit(&pvec);
995 /* shift unused pages over in the pvec... we
996 * will need to release them below. */
997 for (j = i; j < pvec_pages; j++) {
998 dout(" pvec leftover page %p\n", pvec.pages[j]);
999 pvec.pages[j-i+first] = pvec.pages[j];
1005 offset = page_offset(pages[0]);
1008 req = ceph_osdc_new_request(&fsc->client->osdc,
1009 &ci->i_layout, vino,
1010 offset, &len, 0, num_ops,
1012 CEPH_OSD_FLAG_WRITE |
1013 CEPH_OSD_FLAG_ONDISK,
1014 snapc, truncate_seq,
1015 truncate_size, false);
1017 req = ceph_osdc_new_request(&fsc->client->osdc,
1018 &ci->i_layout, vino,
1023 CEPH_OSD_FLAG_WRITE |
1024 CEPH_OSD_FLAG_ONDISK,
1025 snapc, truncate_seq,
1026 truncate_size, true);
1027 BUG_ON(IS_ERR(req));
1029 BUG_ON(len < page_offset(pages[locked_pages - 1]) +
1030 PAGE_SIZE - offset);
1032 req->r_callback = writepages_finish;
1033 req->r_inode = inode;
1035 /* Format the osd request message and submit the write */
1039 for (i = 0; i < locked_pages; i++) {
1040 u64 cur_offset = page_offset(pages[i]);
1041 if (offset + len != cur_offset) {
1042 if (op_idx + do_sync + 1 == req->r_num_ops)
1044 osd_req_op_extent_dup_last(req, op_idx,
1045 cur_offset - offset);
1046 dout("writepages got pages at %llu~%llu\n",
1048 osd_req_op_extent_osd_data_pages(req, op_idx,
1051 osd_req_op_extent_update(req, op_idx, len);
1054 offset = cur_offset;
1055 data_pages = pages + i;
1059 set_page_writeback(pages[i]);
1063 if (snap_size != -1) {
1064 len = min(len, snap_size - offset);
1065 } else if (i == locked_pages) {
1066 /* writepages_finish() clears writeback pages
1067 * according to the data length, so make sure
1068 * data length covers all locked pages */
1069 u64 min_len = len + 1 - PAGE_SIZE;
1070 len = min(len, (u64)i_size_read(inode) - offset);
1071 len = max(len, min_len);
1073 dout("writepages got pages at %llu~%llu\n", offset, len);
1075 osd_req_op_extent_osd_data_pages(req, op_idx, data_pages, len,
1077 osd_req_op_extent_update(req, op_idx, len);
1081 osd_req_op_init(req, op_idx, CEPH_OSD_OP_STARTSYNC, 0);
1083 BUG_ON(op_idx + 1 != req->r_num_ops);
1086 if (i < locked_pages) {
1087 BUG_ON(num_ops <= req->r_num_ops);
1088 num_ops -= req->r_num_ops;
1092 /* allocate new pages array for next request */
1094 pages = kmalloc(locked_pages * sizeof (*pages),
1097 pool = fsc->wb_pagevec_pool;
1098 pages = mempool_alloc(pool, GFP_NOFS);
1101 memcpy(pages, data_pages + i,
1102 locked_pages * sizeof(*pages));
1103 memset(data_pages + i, 0,
1104 locked_pages * sizeof(*pages));
1106 BUG_ON(num_ops != req->r_num_ops);
1107 index = pages[i - 1]->index + 1;
1108 /* request message now owns the pages array */
1112 req->r_mtime = inode->i_mtime;
1113 rc = ceph_osdc_start_request(&fsc->client->osdc, req, true);
1117 wbc->nr_to_write -= i;
1121 if (wbc->nr_to_write <= 0)
1125 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr,
1126 pvec.nr ? pvec.pages[0] : NULL);
1127 pagevec_release(&pvec);
1129 if (locked_pages && !done)
1133 if (should_loop && !done) {
1134 /* more to do; loop back to beginning of file */
1135 dout("writepages looping back to beginning of file\n");
1141 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1142 mapping->writeback_index = index;
1145 ceph_osdc_put_request(req);
1146 ceph_put_snap_context(snapc);
1147 dout("writepages done, rc = %d\n", rc);
1154 * See if a given @snapc is either writeable, or already written.
1156 static int context_is_writeable_or_written(struct inode *inode,
1157 struct ceph_snap_context *snapc)
1159 struct ceph_snap_context *oldest = get_oldest_context(inode, NULL);
1160 int ret = !oldest || snapc->seq <= oldest->seq;
1162 ceph_put_snap_context(oldest);
1167 * We are only allowed to write into/dirty the page if the page is
1168 * clean, or already dirty within the same snap context.
1170 * called with page locked.
1171 * return success with page locked,
1172 * or any failure (incl -EAGAIN) with page unlocked.
1174 static int ceph_update_writeable_page(struct file *file,
1175 loff_t pos, unsigned len,
1178 struct inode *inode = file_inode(file);
1179 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1180 struct ceph_inode_info *ci = ceph_inode(inode);
1181 loff_t page_off = pos & PAGE_MASK;
1182 int pos_in_page = pos & ~PAGE_MASK;
1183 int end_in_page = pos_in_page + len;
1186 struct ceph_snap_context *snapc, *oldest;
1188 if (ACCESS_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1189 dout(" page %p forced umount\n", page);
1195 /* writepages currently holds page lock, but if we change that later, */
1196 wait_on_page_writeback(page);
1198 snapc = page_snap_context(page);
1199 if (snapc && snapc != ci->i_head_snapc) {
1201 * this page is already dirty in another (older) snap
1202 * context! is it writeable now?
1204 oldest = get_oldest_context(inode, NULL);
1206 if (snapc->seq > oldest->seq) {
1207 ceph_put_snap_context(oldest);
1208 dout(" page %p snapc %p not current or oldest\n",
1211 * queue for writeback, and wait for snapc to
1212 * be writeable or written
1214 snapc = ceph_get_snap_context(snapc);
1216 ceph_queue_writeback(inode);
1217 r = wait_event_killable(ci->i_cap_wq,
1218 context_is_writeable_or_written(inode, snapc));
1219 ceph_put_snap_context(snapc);
1220 if (r == -ERESTARTSYS)
1224 ceph_put_snap_context(oldest);
1226 /* yay, writeable, do it now (without dropping page lock) */
1227 dout(" page %p snapc %p not current, but oldest\n",
1229 if (!clear_page_dirty_for_io(page))
1231 r = writepage_nounlock(page, NULL);
1237 if (PageUptodate(page)) {
1238 dout(" page %p already uptodate\n", page);
1243 if (pos_in_page == 0 && len == PAGE_SIZE)
1246 /* past end of file? */
1247 i_size = i_size_read(inode);
1249 if (page_off >= i_size ||
1250 (pos_in_page == 0 && (pos+len) >= i_size &&
1251 end_in_page - pos_in_page != PAGE_SIZE)) {
1252 dout(" zeroing %p 0 - %d and %d - %d\n",
1253 page, pos_in_page, end_in_page, (int)PAGE_SIZE);
1254 zero_user_segments(page,
1256 end_in_page, PAGE_SIZE);
1260 /* we need to read it. */
1261 r = ceph_do_readpage(file, page);
1263 if (r == -EINPROGRESS)
1274 * We are only allowed to write into/dirty the page if the page is
1275 * clean, or already dirty within the same snap context.
1277 static int ceph_write_begin(struct file *file, struct address_space *mapping,
1278 loff_t pos, unsigned len, unsigned flags,
1279 struct page **pagep, void **fsdata)
1281 struct inode *inode = file_inode(file);
1283 pgoff_t index = pos >> PAGE_SHIFT;
1288 page = grab_cache_page_write_begin(mapping, index, 0);
1292 dout("write_begin file %p inode %p page %p %d~%d\n", file,
1293 inode, page, (int)pos, (int)len);
1295 r = ceph_update_writeable_page(file, pos, len, page);
1300 } while (r == -EAGAIN);
1306 * we don't do anything in here that simple_write_end doesn't do
1307 * except adjust dirty page accounting
1309 static int ceph_write_end(struct file *file, struct address_space *mapping,
1310 loff_t pos, unsigned len, unsigned copied,
1311 struct page *page, void *fsdata)
1313 struct inode *inode = file_inode(file);
1314 unsigned from = pos & (PAGE_SIZE - 1);
1317 dout("write_end file %p inode %p page %p %d~%d (%d)\n", file,
1318 inode, page, (int)pos, (int)copied, (int)len);
1320 /* zero the stale part of the page if we did a short copy */
1322 zero_user_segment(page, from+copied, len);
1324 /* did file size increase? */
1325 if (pos+copied > i_size_read(inode))
1326 check_cap = ceph_inode_set_size(inode, pos+copied);
1328 if (!PageUptodate(page))
1329 SetPageUptodate(page);
1331 set_page_dirty(page);
1337 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL);
1343 * we set .direct_IO to indicate direct io is supported, but since we
1344 * intercept O_DIRECT reads and writes early, this function should
1347 static ssize_t ceph_direct_io(struct kiocb *iocb, struct iov_iter *iter)
1353 const struct address_space_operations ceph_aops = {
1354 .readpage = ceph_readpage,
1355 .readpages = ceph_readpages,
1356 .writepage = ceph_writepage,
1357 .writepages = ceph_writepages_start,
1358 .write_begin = ceph_write_begin,
1359 .write_end = ceph_write_end,
1360 .set_page_dirty = ceph_set_page_dirty,
1361 .invalidatepage = ceph_invalidatepage,
1362 .releasepage = ceph_releasepage,
1363 .direct_IO = ceph_direct_io,
1366 static void ceph_block_sigs(sigset_t *oldset)
1369 siginitsetinv(&mask, sigmask(SIGKILL));
1370 sigprocmask(SIG_BLOCK, &mask, oldset);
1373 static void ceph_restore_sigs(sigset_t *oldset)
1375 sigprocmask(SIG_SETMASK, oldset, NULL);
1381 static int ceph_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1383 struct inode *inode = file_inode(vma->vm_file);
1384 struct ceph_inode_info *ci = ceph_inode(inode);
1385 struct ceph_file_info *fi = vma->vm_file->private_data;
1386 struct page *pinned_page = NULL;
1387 loff_t off = (loff_t)vmf->pgoff << PAGE_SHIFT;
1391 ceph_block_sigs(&oldset);
1393 dout("filemap_fault %p %llx.%llx %llu~%zd trying to get caps\n",
1394 inode, ceph_vinop(inode), off, (size_t)PAGE_SIZE);
1395 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1396 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1398 want = CEPH_CAP_FILE_CACHE;
1401 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page);
1405 dout("filemap_fault %p %llu~%zd got cap refs on %s\n",
1406 inode, off, (size_t)PAGE_SIZE, ceph_cap_string(got));
1408 if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) ||
1409 ci->i_inline_version == CEPH_INLINE_NONE) {
1410 current->journal_info = vma->vm_file;
1411 ret = filemap_fault(vma, vmf);
1412 current->journal_info = NULL;
1416 dout("filemap_fault %p %llu~%zd dropping cap refs on %s ret %d\n",
1417 inode, off, (size_t)PAGE_SIZE, ceph_cap_string(got), ret);
1419 put_page(pinned_page);
1420 ceph_put_cap_refs(ci, got);
1425 /* read inline data */
1426 if (off >= PAGE_SIZE) {
1427 /* does not support inline data > PAGE_SIZE */
1428 ret = VM_FAULT_SIGBUS;
1431 struct address_space *mapping = inode->i_mapping;
1432 struct page *page = find_or_create_page(mapping, 0,
1433 mapping_gfp_constraint(mapping,
1439 ret1 = __ceph_do_getattr(inode, page,
1440 CEPH_STAT_CAP_INLINE_DATA, true);
1441 if (ret1 < 0 || off >= i_size_read(inode)) {
1447 ret = VM_FAULT_SIGBUS;
1450 if (ret1 < PAGE_SIZE)
1451 zero_user_segment(page, ret1, PAGE_SIZE);
1453 flush_dcache_page(page);
1454 SetPageUptodate(page);
1456 ret = VM_FAULT_MAJOR | VM_FAULT_LOCKED;
1458 dout("filemap_fault %p %llu~%zd read inline data ret %d\n",
1459 inode, off, (size_t)PAGE_SIZE, ret);
1462 ceph_restore_sigs(&oldset);
1464 ret = (ret == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS;
1470 * Reuse write_begin here for simplicity.
1472 static int ceph_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1474 struct inode *inode = file_inode(vma->vm_file);
1475 struct ceph_inode_info *ci = ceph_inode(inode);
1476 struct ceph_file_info *fi = vma->vm_file->private_data;
1477 struct ceph_cap_flush *prealloc_cf;
1478 struct page *page = vmf->page;
1479 loff_t off = page_offset(page);
1480 loff_t size = i_size_read(inode);
1485 prealloc_cf = ceph_alloc_cap_flush();
1487 return VM_FAULT_OOM;
1489 ceph_block_sigs(&oldset);
1491 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1492 struct page *locked_page = NULL;
1497 ret = ceph_uninline_data(vma->vm_file, locked_page);
1499 unlock_page(locked_page);
1504 if (off + PAGE_SIZE <= size)
1507 len = size & ~PAGE_MASK;
1509 dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n",
1510 inode, ceph_vinop(inode), off, len, size);
1511 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1512 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1514 want = CEPH_CAP_FILE_BUFFER;
1517 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, off + len,
1522 dout("page_mkwrite %p %llu~%zd got cap refs on %s\n",
1523 inode, off, len, ceph_cap_string(got));
1525 /* Update time before taking page lock */
1526 file_update_time(vma->vm_file);
1531 if ((off > size) || (page->mapping != inode->i_mapping)) {
1533 ret = VM_FAULT_NOPAGE;
1537 ret = ceph_update_writeable_page(vma->vm_file, off, len, page);
1539 /* success. we'll keep the page locked. */
1540 set_page_dirty(page);
1541 ret = VM_FAULT_LOCKED;
1543 } while (ret == -EAGAIN);
1545 if (ret == VM_FAULT_LOCKED ||
1546 ci->i_inline_version != CEPH_INLINE_NONE) {
1548 spin_lock(&ci->i_ceph_lock);
1549 ci->i_inline_version = CEPH_INLINE_NONE;
1550 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1552 spin_unlock(&ci->i_ceph_lock);
1554 __mark_inode_dirty(inode, dirty);
1557 dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %d\n",
1558 inode, off, len, ceph_cap_string(got), ret);
1559 ceph_put_cap_refs(ci, got);
1561 ceph_restore_sigs(&oldset);
1562 ceph_free_cap_flush(prealloc_cf);
1564 ret = (ret == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS;
1568 void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
1569 char *data, size_t len)
1571 struct address_space *mapping = inode->i_mapping;
1577 if (i_size_read(inode) == 0)
1579 page = find_or_create_page(mapping, 0,
1580 mapping_gfp_constraint(mapping,
1584 if (PageUptodate(page)) {
1591 dout("fill_inline_data %p %llx.%llx len %zu locked_page %p\n",
1592 inode, ceph_vinop(inode), len, locked_page);
1595 void *kaddr = kmap_atomic(page);
1596 memcpy(kaddr, data, len);
1597 kunmap_atomic(kaddr);
1600 if (page != locked_page) {
1601 if (len < PAGE_SIZE)
1602 zero_user_segment(page, len, PAGE_SIZE);
1604 flush_dcache_page(page);
1606 SetPageUptodate(page);
1612 int ceph_uninline_data(struct file *filp, struct page *locked_page)
1614 struct inode *inode = file_inode(filp);
1615 struct ceph_inode_info *ci = ceph_inode(inode);
1616 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1617 struct ceph_osd_request *req;
1618 struct page *page = NULL;
1619 u64 len, inline_version;
1621 bool from_pagecache = false;
1623 spin_lock(&ci->i_ceph_lock);
1624 inline_version = ci->i_inline_version;
1625 spin_unlock(&ci->i_ceph_lock);
1627 dout("uninline_data %p %llx.%llx inline_version %llu\n",
1628 inode, ceph_vinop(inode), inline_version);
1630 if (inline_version == 1 || /* initial version, no data */
1631 inline_version == CEPH_INLINE_NONE)
1636 WARN_ON(!PageUptodate(page));
1637 } else if (ceph_caps_issued(ci) &
1638 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) {
1639 page = find_get_page(inode->i_mapping, 0);
1641 if (PageUptodate(page)) {
1642 from_pagecache = true;
1652 len = i_size_read(inode);
1653 if (len > PAGE_SIZE)
1656 page = __page_cache_alloc(GFP_NOFS);
1661 err = __ceph_do_getattr(inode, page,
1662 CEPH_STAT_CAP_INLINE_DATA, true);
1664 /* no inline data */
1665 if (err == -ENODATA)
1672 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1673 ceph_vino(inode), 0, &len, 0, 1,
1675 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
1682 req->r_mtime = inode->i_mtime;
1683 err = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1685 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1686 ceph_osdc_put_request(req);
1690 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1691 ceph_vino(inode), 0, &len, 1, 3,
1693 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
1694 NULL, ci->i_truncate_seq,
1695 ci->i_truncate_size, false);
1701 osd_req_op_extent_osd_data_pages(req, 1, &page, len, 0, false, false);
1704 __le64 xattr_buf = cpu_to_le64(inline_version);
1705 err = osd_req_op_xattr_init(req, 0, CEPH_OSD_OP_CMPXATTR,
1706 "inline_version", &xattr_buf,
1708 CEPH_OSD_CMPXATTR_OP_GT,
1709 CEPH_OSD_CMPXATTR_MODE_U64);
1716 int xattr_len = snprintf(xattr_buf, sizeof(xattr_buf),
1717 "%llu", inline_version);
1718 err = osd_req_op_xattr_init(req, 2, CEPH_OSD_OP_SETXATTR,
1720 xattr_buf, xattr_len, 0, 0);
1725 req->r_mtime = inode->i_mtime;
1726 err = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1728 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1730 ceph_osdc_put_request(req);
1731 if (err == -ECANCELED)
1734 if (page && page != locked_page) {
1735 if (from_pagecache) {
1739 __free_pages(page, 0);
1742 dout("uninline_data %p %llx.%llx inline_version %llu = %d\n",
1743 inode, ceph_vinop(inode), inline_version, err);
1747 static const struct vm_operations_struct ceph_vmops = {
1748 .fault = ceph_filemap_fault,
1749 .page_mkwrite = ceph_page_mkwrite,
1752 int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1754 struct address_space *mapping = file->f_mapping;
1756 if (!mapping->a_ops->readpage)
1758 file_accessed(file);
1759 vma->vm_ops = &ceph_vmops;
1768 static int __ceph_pool_perm_get(struct ceph_inode_info *ci,
1769 s64 pool, struct ceph_string *pool_ns)
1771 struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode);
1772 struct ceph_mds_client *mdsc = fsc->mdsc;
1773 struct ceph_osd_request *rd_req = NULL, *wr_req = NULL;
1774 struct rb_node **p, *parent;
1775 struct ceph_pool_perm *perm;
1776 struct page **pages;
1778 int err = 0, err2 = 0, have = 0;
1780 down_read(&mdsc->pool_perm_rwsem);
1781 p = &mdsc->pool_perm_tree.rb_node;
1783 perm = rb_entry(*p, struct ceph_pool_perm, node);
1784 if (pool < perm->pool)
1786 else if (pool > perm->pool)
1787 p = &(*p)->rb_right;
1789 int ret = ceph_compare_string(pool_ns,
1795 p = &(*p)->rb_right;
1802 up_read(&mdsc->pool_perm_rwsem);
1807 dout("__ceph_pool_perm_get pool %lld ns %.*s no perm cached\n",
1808 pool, (int)pool_ns->len, pool_ns->str);
1810 dout("__ceph_pool_perm_get pool %lld no perm cached\n", pool);
1812 down_write(&mdsc->pool_perm_rwsem);
1813 p = &mdsc->pool_perm_tree.rb_node;
1817 perm = rb_entry(parent, struct ceph_pool_perm, node);
1818 if (pool < perm->pool)
1820 else if (pool > perm->pool)
1821 p = &(*p)->rb_right;
1823 int ret = ceph_compare_string(pool_ns,
1829 p = &(*p)->rb_right;
1837 up_write(&mdsc->pool_perm_rwsem);
1841 rd_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1842 1, false, GFP_NOFS);
1848 rd_req->r_flags = CEPH_OSD_FLAG_READ;
1849 osd_req_op_init(rd_req, 0, CEPH_OSD_OP_STAT, 0);
1850 rd_req->r_base_oloc.pool = pool;
1852 rd_req->r_base_oloc.pool_ns = ceph_get_string(pool_ns);
1853 ceph_oid_printf(&rd_req->r_base_oid, "%llx.00000000", ci->i_vino.ino);
1855 err = ceph_osdc_alloc_messages(rd_req, GFP_NOFS);
1859 wr_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1860 1, false, GFP_NOFS);
1866 wr_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ACK;
1867 osd_req_op_init(wr_req, 0, CEPH_OSD_OP_CREATE, CEPH_OSD_OP_FLAG_EXCL);
1868 ceph_oloc_copy(&wr_req->r_base_oloc, &rd_req->r_base_oloc);
1869 ceph_oid_copy(&wr_req->r_base_oid, &rd_req->r_base_oid);
1871 err = ceph_osdc_alloc_messages(wr_req, GFP_NOFS);
1875 /* one page should be large enough for STAT data */
1876 pages = ceph_alloc_page_vector(1, GFP_KERNEL);
1877 if (IS_ERR(pages)) {
1878 err = PTR_ERR(pages);
1882 osd_req_op_raw_data_in_pages(rd_req, 0, pages, PAGE_SIZE,
1884 err = ceph_osdc_start_request(&fsc->client->osdc, rd_req, false);
1886 wr_req->r_mtime = ci->vfs_inode.i_mtime;
1887 err2 = ceph_osdc_start_request(&fsc->client->osdc, wr_req, false);
1890 err = ceph_osdc_wait_request(&fsc->client->osdc, rd_req);
1892 err2 = ceph_osdc_wait_request(&fsc->client->osdc, wr_req);
1894 if (err >= 0 || err == -ENOENT)
1896 else if (err != -EPERM)
1899 if (err2 == 0 || err2 == -EEXIST)
1901 else if (err2 != -EPERM) {
1906 pool_ns_len = pool_ns ? pool_ns->len : 0;
1907 perm = kmalloc(sizeof(*perm) + pool_ns_len + 1, GFP_NOFS);
1915 perm->pool_ns_len = pool_ns_len;
1916 if (pool_ns_len > 0)
1917 memcpy(perm->pool_ns, pool_ns->str, pool_ns_len);
1918 perm->pool_ns[pool_ns_len] = 0;
1920 rb_link_node(&perm->node, parent, p);
1921 rb_insert_color(&perm->node, &mdsc->pool_perm_tree);
1924 up_write(&mdsc->pool_perm_rwsem);
1926 ceph_osdc_put_request(rd_req);
1927 ceph_osdc_put_request(wr_req);
1932 dout("__ceph_pool_perm_get pool %lld ns %.*s result = %d\n",
1933 pool, (int)pool_ns->len, pool_ns->str, err);
1935 dout("__ceph_pool_perm_get pool %lld result = %d\n", pool, err);
1939 int ceph_pool_perm_check(struct ceph_inode_info *ci, int need)
1942 struct ceph_string *pool_ns;
1945 if (ceph_test_mount_opt(ceph_inode_to_client(&ci->vfs_inode),
1949 spin_lock(&ci->i_ceph_lock);
1950 flags = ci->i_ceph_flags;
1951 pool = ci->i_layout.pool_id;
1952 spin_unlock(&ci->i_ceph_lock);
1954 if (flags & CEPH_I_POOL_PERM) {
1955 if ((need & CEPH_CAP_FILE_RD) && !(flags & CEPH_I_POOL_RD)) {
1956 dout("ceph_pool_perm_check pool %lld no read perm\n",
1960 if ((need & CEPH_CAP_FILE_WR) && !(flags & CEPH_I_POOL_WR)) {
1961 dout("ceph_pool_perm_check pool %lld no write perm\n",
1968 pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
1969 ret = __ceph_pool_perm_get(ci, pool, pool_ns);
1970 ceph_put_string(pool_ns);
1974 flags = CEPH_I_POOL_PERM;
1975 if (ret & POOL_READ)
1976 flags |= CEPH_I_POOL_RD;
1977 if (ret & POOL_WRITE)
1978 flags |= CEPH_I_POOL_WR;
1980 spin_lock(&ci->i_ceph_lock);
1981 if (pool == ci->i_layout.pool_id &&
1982 pool_ns == rcu_dereference_raw(ci->i_layout.pool_ns)) {
1983 ci->i_ceph_flags |= flags;
1985 pool = ci->i_layout.pool_id;
1986 flags = ci->i_ceph_flags;
1988 spin_unlock(&ci->i_ceph_lock);
1992 void ceph_pool_perm_destroy(struct ceph_mds_client *mdsc)
1994 struct ceph_pool_perm *perm;
1997 while (!RB_EMPTY_ROOT(&mdsc->pool_perm_tree)) {
1998 n = rb_first(&mdsc->pool_perm_tree);
1999 perm = rb_entry(n, struct ceph_pool_perm, node);
2000 rb_erase(n, &mdsc->pool_perm_tree);