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 spin_lock(&ci->i_ceph_lock);
92 BUG_ON(ci->i_wr_ref == 0); // caller should hold Fw reference
93 if (__ceph_have_pending_cap_snap(ci)) {
94 struct ceph_cap_snap *capsnap =
95 list_last_entry(&ci->i_cap_snaps,
98 snapc = ceph_get_snap_context(capsnap->context);
99 capsnap->dirty_pages++;
101 BUG_ON(!ci->i_head_snapc);
102 snapc = ceph_get_snap_context(ci->i_head_snapc);
103 ++ci->i_wrbuffer_ref_head;
105 if (ci->i_wrbuffer_ref == 0)
107 ++ci->i_wrbuffer_ref;
108 dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d "
109 "snapc %p seq %lld (%d snaps)\n",
110 mapping->host, page, page->index,
111 ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1,
112 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
113 snapc, snapc->seq, snapc->num_snaps);
114 spin_unlock(&ci->i_ceph_lock);
117 * Reference snap context in page->private. Also set
118 * PagePrivate so that we get invalidatepage callback.
120 BUG_ON(PagePrivate(page));
121 page->private = (unsigned long)snapc;
122 SetPagePrivate(page);
124 ret = __set_page_dirty_nobuffers(page);
125 WARN_ON(!PageLocked(page));
126 WARN_ON(!page->mapping);
132 * If we are truncating the full page (i.e. offset == 0), adjust the
133 * dirty page counters appropriately. Only called if there is private
136 static void ceph_invalidatepage(struct page *page, unsigned int offset,
140 struct ceph_inode_info *ci;
141 struct ceph_snap_context *snapc = page_snap_context(page);
143 inode = page->mapping->host;
144 ci = ceph_inode(inode);
146 if (offset != 0 || length != PAGE_SIZE) {
147 dout("%p invalidatepage %p idx %lu partial dirty page %u~%u\n",
148 inode, page, page->index, offset, length);
152 ceph_invalidate_fscache_page(inode, page);
154 if (!PagePrivate(page))
158 * We can get non-dirty pages here due to races between
159 * set_page_dirty and truncate_complete_page; just spit out a
160 * warning, in case we end up with accounting problems later.
162 if (!PageDirty(page))
163 pr_err("%p invalidatepage %p page not dirty\n", inode, page);
165 ClearPageChecked(page);
167 dout("%p invalidatepage %p idx %lu full dirty page\n",
168 inode, page, page->index);
170 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
171 ceph_put_snap_context(snapc);
173 ClearPagePrivate(page);
176 static int ceph_releasepage(struct page *page, gfp_t g)
178 dout("%p releasepage %p idx %lu (%sdirty)\n", page->mapping->host,
179 page, page->index, PageDirty(page) ? "" : "not ");
181 /* Can we release the page from the cache? */
182 if (!ceph_release_fscache_page(page, g))
185 return !PagePrivate(page);
189 * read a single page, without unlocking it.
191 static int readpage_nounlock(struct file *filp, struct page *page)
193 struct inode *inode = file_inode(filp);
194 struct ceph_inode_info *ci = ceph_inode(inode);
195 struct ceph_osd_client *osdc =
196 &ceph_inode_to_client(inode)->client->osdc;
198 u64 off = page_offset(page);
201 if (off >= i_size_read(inode)) {
202 zero_user_segment(page, 0, PAGE_SIZE);
203 SetPageUptodate(page);
207 if (ci->i_inline_version != CEPH_INLINE_NONE) {
209 * Uptodate inline data should have been added
210 * into page cache while getting Fcr caps.
214 zero_user_segment(page, 0, PAGE_SIZE);
215 SetPageUptodate(page);
219 err = ceph_readpage_from_fscache(inode, page);
223 dout("readpage inode %p file %p page %p index %lu\n",
224 inode, filp, page, page->index);
225 err = ceph_osdc_readpages(osdc, ceph_vino(inode), &ci->i_layout,
227 ci->i_truncate_seq, ci->i_truncate_size,
233 ceph_fscache_readpage_cancel(inode, page);
237 /* zero fill remainder of page */
238 zero_user_segment(page, err, PAGE_SIZE);
240 flush_dcache_page(page);
242 SetPageUptodate(page);
243 ceph_readpage_to_fscache(inode, page);
246 return err < 0 ? err : 0;
249 static int ceph_readpage(struct file *filp, struct page *page)
251 int r = readpage_nounlock(filp, page);
257 * Finish an async read(ahead) op.
259 static void finish_read(struct ceph_osd_request *req)
261 struct inode *inode = req->r_inode;
262 struct ceph_osd_data *osd_data;
263 int rc = req->r_result <= 0 ? req->r_result : 0;
264 int bytes = req->r_result >= 0 ? req->r_result : 0;
268 dout("finish_read %p req %p rc %d bytes %d\n", inode, req, rc, bytes);
270 /* unlock all pages, zeroing any data we didn't read */
271 osd_data = osd_req_op_extent_osd_data(req, 0);
272 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
273 num_pages = calc_pages_for((u64)osd_data->alignment,
274 (u64)osd_data->length);
275 for (i = 0; i < num_pages; i++) {
276 struct page *page = osd_data->pages[i];
278 if (rc < 0 && rc != -ENOENT) {
279 ceph_fscache_readpage_cancel(inode, page);
282 if (bytes < (int)PAGE_SIZE) {
283 /* zero (remainder of) page */
284 int s = bytes < 0 ? 0 : bytes;
285 zero_user_segment(page, s, PAGE_SIZE);
287 dout("finish_read %p uptodate %p idx %lu\n", inode, page,
289 flush_dcache_page(page);
290 SetPageUptodate(page);
291 ceph_readpage_to_fscache(inode, page);
297 kfree(osd_data->pages);
301 * start an async read(ahead) operation. return nr_pages we submitted
302 * a read for on success, or negative error code.
304 static int start_read(struct inode *inode, struct list_head *page_list, int max)
306 struct ceph_osd_client *osdc =
307 &ceph_inode_to_client(inode)->client->osdc;
308 struct ceph_inode_info *ci = ceph_inode(inode);
309 struct page *page = list_entry(page_list->prev, struct page, lru);
310 struct ceph_vino vino;
311 struct ceph_osd_request *req;
321 if (!current->journal_info) {
322 /* caller of readpages does not hold buffer and read caps
323 * (fadvise, madvise and readahead cases) */
324 int want = CEPH_CAP_FILE_CACHE;
325 ret = ceph_try_get_caps(ci, CEPH_CAP_FILE_RD, want, &got);
327 dout("start_read %p, error getting cap\n", inode);
328 } else if (!(got & want)) {
329 dout("start_read %p, no cache cap\n", inode);
334 ceph_put_cap_refs(ci, got);
335 while (!list_empty(page_list)) {
336 page = list_entry(page_list->prev,
338 list_del(&page->lru);
345 off = (u64) page_offset(page);
348 next_index = page->index;
349 list_for_each_entry_reverse(page, page_list, lru) {
350 if (page->index != next_index)
354 if (max && nr_pages == max)
357 len = nr_pages << PAGE_SHIFT;
358 dout("start_read %p nr_pages %d is %lld~%lld\n", inode, nr_pages,
360 vino = ceph_vino(inode);
361 req = ceph_osdc_new_request(osdc, &ci->i_layout, vino, off, &len,
362 0, 1, CEPH_OSD_OP_READ,
363 CEPH_OSD_FLAG_READ, NULL,
364 ci->i_truncate_seq, ci->i_truncate_size,
371 /* build page vector */
372 nr_pages = calc_pages_for(0, len);
373 pages = kmalloc(sizeof(*pages) * nr_pages, GFP_KERNEL);
378 for (i = 0; i < nr_pages; ++i) {
379 page = list_entry(page_list->prev, struct page, lru);
380 BUG_ON(PageLocked(page));
381 list_del(&page->lru);
383 dout("start_read %p adding %p idx %lu\n", inode, page,
385 if (add_to_page_cache_lru(page, &inode->i_data, page->index,
387 ceph_fscache_uncache_page(inode, page);
389 dout("start_read %p add_to_page_cache failed %p\n",
393 len = nr_pages << PAGE_SHIFT;
400 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false);
401 req->r_callback = finish_read;
402 req->r_inode = inode;
404 dout("start_read %p starting %p %lld~%lld\n", inode, req, off, len);
405 ret = ceph_osdc_start_request(osdc, req, false);
408 ceph_osdc_put_request(req);
410 /* After adding locked pages to page cache, the inode holds cache cap.
411 * So we can drop our cap refs. */
413 ceph_put_cap_refs(ci, got);
418 for (i = 0; i < nr_pages; ++i) {
419 ceph_fscache_readpage_cancel(inode, pages[i]);
420 unlock_page(pages[i]);
422 ceph_put_page_vector(pages, nr_pages, false);
424 ceph_osdc_put_request(req);
427 ceph_put_cap_refs(ci, got);
433 * Read multiple pages. Leave pages we don't read + unlock in page_list;
434 * the caller (VM) cleans them up.
436 static int ceph_readpages(struct file *file, struct address_space *mapping,
437 struct list_head *page_list, unsigned nr_pages)
439 struct inode *inode = file_inode(file);
440 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
444 if (ceph_inode(inode)->i_inline_version != CEPH_INLINE_NONE)
447 rc = ceph_readpages_from_fscache(mapping->host, mapping, page_list,
453 if (fsc->mount_options->rsize >= PAGE_SIZE)
454 max = (fsc->mount_options->rsize + PAGE_SIZE - 1)
457 dout("readpages %p file %p nr_pages %d max %d\n", inode,
460 while (!list_empty(page_list)) {
461 rc = start_read(inode, page_list, max);
466 ceph_fscache_readpages_cancel(inode, page_list);
468 dout("readpages %p file %p ret %d\n", inode, file, rc);
473 * Get ref for the oldest snapc for an inode with dirty data... that is, the
474 * only snap context we are allowed to write back.
476 static struct ceph_snap_context *get_oldest_context(struct inode *inode,
479 struct ceph_inode_info *ci = ceph_inode(inode);
480 struct ceph_snap_context *snapc = NULL;
481 struct ceph_cap_snap *capsnap = NULL;
483 spin_lock(&ci->i_ceph_lock);
484 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
485 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap,
486 capsnap->context, capsnap->dirty_pages);
487 if (capsnap->dirty_pages) {
488 snapc = ceph_get_snap_context(capsnap->context);
490 *snap_size = capsnap->size;
494 if (!snapc && ci->i_wrbuffer_ref_head) {
495 snapc = ceph_get_snap_context(ci->i_head_snapc);
496 dout(" head snapc %p has %d dirty pages\n",
497 snapc, ci->i_wrbuffer_ref_head);
499 spin_unlock(&ci->i_ceph_lock);
504 * Write a single page, but leave the page locked.
506 * If we get a write error, set the page error bit, but still adjust the
507 * dirty page accounting (i.e., page is no longer dirty).
509 static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
512 struct ceph_inode_info *ci;
513 struct ceph_fs_client *fsc;
514 struct ceph_osd_client *osdc;
515 struct ceph_snap_context *snapc, *oldest;
516 loff_t page_off = page_offset(page);
517 loff_t snap_size = -1;
521 int err = 0, len = PAGE_SIZE;
523 dout("writepage %p idx %lu\n", page, page->index);
525 if (!page->mapping || !page->mapping->host) {
526 dout("writepage %p - no mapping\n", page);
529 inode = page->mapping->host;
530 ci = ceph_inode(inode);
531 fsc = ceph_inode_to_client(inode);
532 osdc = &fsc->client->osdc;
534 /* verify this is a writeable snap context */
535 snapc = page_snap_context(page);
537 dout("writepage %p page %p not dirty?\n", inode, page);
540 oldest = get_oldest_context(inode, &snap_size);
541 if (snapc->seq > oldest->seq) {
542 dout("writepage %p page %p snapc %p not writeable - noop\n",
544 /* we should only noop if called by kswapd */
545 WARN_ON((current->flags & PF_MEMALLOC) == 0);
546 ceph_put_snap_context(oldest);
549 ceph_put_snap_context(oldest);
551 spin_lock(&ci->i_ceph_lock);
552 truncate_seq = ci->i_truncate_seq;
553 truncate_size = ci->i_truncate_size;
555 snap_size = i_size_read(inode);
556 spin_unlock(&ci->i_ceph_lock);
558 /* is this a partial page at end of file? */
559 if (page_off >= snap_size) {
560 dout("%p page eof %llu\n", page, snap_size);
563 if (snap_size < page_off + len)
564 len = snap_size - page_off;
566 dout("writepage %p page %p index %lu on %llu~%u snapc %p\n",
567 inode, page, page->index, page_off, len, snapc);
569 writeback_stat = atomic_long_inc_return(&fsc->writeback_count);
571 CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb))
572 set_bdi_congested(&fsc->backing_dev_info, BLK_RW_ASYNC);
574 set_page_writeback(page);
575 err = ceph_osdc_writepages(osdc, ceph_vino(inode),
576 &ci->i_layout, snapc,
578 truncate_seq, truncate_size,
579 &inode->i_mtime, &page, 1);
581 struct writeback_control tmp_wbc;
584 if (err == -ERESTARTSYS) {
585 /* killed by SIGKILL */
586 dout("writepage interrupted page %p\n", page);
587 redirty_page_for_writepage(wbc, page);
588 end_page_writeback(page);
591 dout("writepage setting page/mapping error %d %p\n",
594 mapping_set_error(&inode->i_data, err);
595 wbc->pages_skipped++;
597 dout("writepage cleaned page %p\n", page);
598 err = 0; /* vfs expects us to return 0 */
601 ClearPagePrivate(page);
602 end_page_writeback(page);
603 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
604 ceph_put_snap_context(snapc); /* page's reference */
609 static int ceph_writepage(struct page *page, struct writeback_control *wbc)
612 struct inode *inode = page->mapping->host;
615 err = writepage_nounlock(page, wbc);
616 if (err == -ERESTARTSYS) {
617 /* direct memory reclaimer was killed by SIGKILL. return 0
618 * to prevent caller from setting mapping/page error */
627 * lame release_pages helper. release_pages() isn't exported to
630 static void ceph_release_pages(struct page **pages, int num)
635 pagevec_init(&pvec, 0);
636 for (i = 0; i < num; i++) {
637 if (pagevec_add(&pvec, pages[i]) == 0)
638 pagevec_release(&pvec);
640 pagevec_release(&pvec);
644 * async writeback completion handler.
646 * If we get an error, set the mapping error bit, but not the individual
649 static void writepages_finish(struct ceph_osd_request *req)
651 struct inode *inode = req->r_inode;
652 struct ceph_inode_info *ci = ceph_inode(inode);
653 struct ceph_osd_data *osd_data;
655 int num_pages, total_pages = 0;
657 int rc = req->r_result;
658 struct ceph_snap_context *snapc = req->r_snapc;
659 struct address_space *mapping = inode->i_mapping;
660 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
663 dout("writepages_finish %p rc %d\n", inode, rc);
665 mapping_set_error(mapping, rc);
668 * We lost the cache cap, need to truncate the page before
669 * it is unlocked, otherwise we'd truncate it later in the
670 * page truncation thread, possibly losing some data that
673 remove_page = !(ceph_caps_issued(ci) &
674 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO));
676 /* clean all pages */
677 for (i = 0; i < req->r_num_ops; i++) {
678 if (req->r_ops[i].op != CEPH_OSD_OP_WRITE)
681 osd_data = osd_req_op_extent_osd_data(req, i);
682 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
683 num_pages = calc_pages_for((u64)osd_data->alignment,
684 (u64)osd_data->length);
685 total_pages += num_pages;
686 for (j = 0; j < num_pages; j++) {
687 page = osd_data->pages[j];
689 WARN_ON(!PageUptodate(page));
691 if (atomic_long_dec_return(&fsc->writeback_count) <
692 CONGESTION_OFF_THRESH(
693 fsc->mount_options->congestion_kb))
694 clear_bdi_congested(&fsc->backing_dev_info,
700 ceph_put_snap_context(page_snap_context(page));
702 ClearPagePrivate(page);
703 dout("unlocking %p\n", page);
704 end_page_writeback(page);
707 generic_error_remove_page(inode->i_mapping,
712 dout("writepages_finish %p wrote %llu bytes cleaned %d pages\n",
713 inode, osd_data->length, rc >= 0 ? num_pages : 0);
715 ceph_release_pages(osd_data->pages, num_pages);
718 ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc);
720 osd_data = osd_req_op_extent_osd_data(req, 0);
721 if (osd_data->pages_from_pool)
722 mempool_free(osd_data->pages,
723 ceph_sb_to_client(inode->i_sb)->wb_pagevec_pool);
725 kfree(osd_data->pages);
726 ceph_osdc_put_request(req);
730 * initiate async writeback
732 static int ceph_writepages_start(struct address_space *mapping,
733 struct writeback_control *wbc)
735 struct inode *inode = mapping->host;
736 struct ceph_inode_info *ci = ceph_inode(inode);
737 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
738 struct ceph_vino vino = ceph_vino(inode);
739 pgoff_t index, start, end;
742 pgoff_t max_pages = 0, max_pages_ever = 0;
743 struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc;
747 unsigned wsize = 1 << inode->i_blkbits;
748 struct ceph_osd_request *req = NULL;
750 loff_t snap_size, i_size;
755 * Include a 'sync' in the OSD request if this is a data
756 * integrity write (e.g., O_SYNC write or fsync()), or if our
757 * cap is being revoked.
759 if ((wbc->sync_mode == WB_SYNC_ALL) ||
760 ceph_caps_revoking(ci, CEPH_CAP_FILE_BUFFER))
762 dout("writepages_start %p dosync=%d (mode=%s)\n",
764 wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
765 (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
767 if (ACCESS_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
768 if (ci->i_wrbuffer_ref > 0) {
770 "writepage_start %p %lld forced umount\n",
771 inode, ceph_ino(inode));
773 mapping_set_error(mapping, -EIO);
774 return -EIO; /* we're in a forced umount, don't write! */
776 if (fsc->mount_options->wsize && fsc->mount_options->wsize < wsize)
777 wsize = fsc->mount_options->wsize;
778 if (wsize < PAGE_SIZE)
780 max_pages_ever = wsize >> PAGE_SHIFT;
782 pagevec_init(&pvec, 0);
784 /* where to start/end? */
785 if (wbc->range_cyclic) {
786 start = mapping->writeback_index; /* Start from prev offset */
788 dout(" cyclic, start at %lu\n", start);
790 start = wbc->range_start >> PAGE_SHIFT;
791 end = wbc->range_end >> PAGE_SHIFT;
792 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
795 dout(" not cyclic, %lu to %lu\n", start, end);
800 /* find oldest snap context with dirty data */
801 ceph_put_snap_context(snapc);
803 snapc = get_oldest_context(inode, &snap_size);
805 /* hmm, why does writepages get called when there
807 dout(" no snap context with dirty data?\n");
810 dout(" oldest snapc is %p seq %lld (%d snaps)\n",
811 snapc, snapc->seq, snapc->num_snaps);
813 spin_lock(&ci->i_ceph_lock);
814 truncate_seq = ci->i_truncate_seq;
815 truncate_size = ci->i_truncate_size;
816 i_size = i_size_read(inode);
817 spin_unlock(&ci->i_ceph_lock);
819 if (last_snapc && snapc != last_snapc) {
820 /* if we switched to a newer snapc, restart our scan at the
821 * start of the original file range. */
822 dout(" snapc differs from last pass, restarting at %lu\n",
828 while (!done && index <= end) {
831 pgoff_t strip_unit_end = 0;
832 int num_ops = 0, op_idx;
833 int pvec_pages, locked_pages = 0;
834 struct page **pages = NULL, **data_pages;
835 mempool_t *pool = NULL; /* Becomes non-null if mempool used */
838 u64 offset = 0, len = 0;
840 max_pages = max_pages_ever;
844 want = min(end - index,
845 min((pgoff_t)PAGEVEC_SIZE,
846 max_pages - (pgoff_t)locked_pages) - 1)
848 pvec_pages = pagevec_lookup_tag(&pvec, mapping, &index,
851 dout("pagevec_lookup_tag got %d\n", pvec_pages);
852 if (!pvec_pages && !locked_pages)
854 for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) {
855 page = pvec.pages[i];
856 dout("? %p idx %lu\n", page, page->index);
857 if (locked_pages == 0)
858 lock_page(page); /* first page */
859 else if (!trylock_page(page))
862 /* only dirty pages, or our accounting breaks */
863 if (unlikely(!PageDirty(page)) ||
864 unlikely(page->mapping != mapping)) {
865 dout("!dirty or !mapping %p\n", page);
869 if (!wbc->range_cyclic && page->index > end) {
870 dout("end of range %p\n", page);
875 if (strip_unit_end && (page->index > strip_unit_end)) {
876 dout("end of strip unit %p\n", page);
880 if (wbc->sync_mode != WB_SYNC_NONE) {
881 dout("waiting on writeback %p\n", page);
882 wait_on_page_writeback(page);
884 if (page_offset(page) >=
885 (snap_size == -1 ? i_size : snap_size)) {
886 dout("%p page eof %llu\n", page,
887 (snap_size == -1 ? i_size : snap_size));
892 if (PageWriteback(page)) {
893 dout("%p under writeback\n", page);
898 /* only if matching snap context */
899 pgsnapc = page_snap_context(page);
900 if (pgsnapc->seq > snapc->seq) {
901 dout("page snapc %p %lld > oldest %p %lld\n",
902 pgsnapc, pgsnapc->seq, snapc, snapc->seq);
905 continue; /* keep looking for snap */
909 if (!clear_page_dirty_for_io(page)) {
910 dout("%p !clear_page_dirty_for_io\n", page);
916 * We have something to write. If this is
917 * the first locked page this time through,
918 * calculate max possinle write size and
919 * allocate a page array
921 if (locked_pages == 0) {
925 /* prepare async write request */
926 offset = (u64)page_offset(page);
929 rc = ceph_calc_file_object_mapping(&ci->i_layout,
938 num_ops = 1 + do_sync;
939 strip_unit_end = page->index +
940 ((len - 1) >> PAGE_SHIFT);
943 max_pages = calc_pages_for(0, (u64)len);
944 pages = kmalloc(max_pages * sizeof (*pages),
947 pool = fsc->wb_pagevec_pool;
948 pages = mempool_alloc(pool, GFP_NOFS);
953 } else if (page->index !=
954 (offset + len) >> PAGE_SHIFT) {
955 if (num_ops >= (pool ? CEPH_OSD_SLAB_OPS :
957 redirty_page_for_writepage(wbc, page);
963 offset = (u64)page_offset(page);
967 /* note position of first page in pvec */
970 dout("%p will write page %p idx %lu\n",
971 inode, page, page->index);
973 if (atomic_long_inc_return(&fsc->writeback_count) >
974 CONGESTION_ON_THRESH(
975 fsc->mount_options->congestion_kb)) {
976 set_bdi_congested(&fsc->backing_dev_info,
980 pages[locked_pages] = page;
985 /* did we get anything? */
987 goto release_pvec_pages;
990 BUG_ON(!locked_pages || first < 0);
992 if (pvec_pages && i == pvec_pages &&
993 locked_pages < max_pages) {
994 dout("reached end pvec, trying for more\n");
995 pagevec_reinit(&pvec);
999 /* shift unused pages over in the pvec... we
1000 * will need to release them below. */
1001 for (j = i; j < pvec_pages; j++) {
1002 dout(" pvec leftover page %p\n", pvec.pages[j]);
1003 pvec.pages[j-i+first] = pvec.pages[j];
1009 offset = page_offset(pages[0]);
1012 req = ceph_osdc_new_request(&fsc->client->osdc,
1013 &ci->i_layout, vino,
1014 offset, &len, 0, num_ops,
1016 CEPH_OSD_FLAG_WRITE |
1017 CEPH_OSD_FLAG_ONDISK,
1018 snapc, truncate_seq,
1019 truncate_size, false);
1021 req = ceph_osdc_new_request(&fsc->client->osdc,
1022 &ci->i_layout, vino,
1027 CEPH_OSD_FLAG_WRITE |
1028 CEPH_OSD_FLAG_ONDISK,
1029 snapc, truncate_seq,
1030 truncate_size, true);
1031 BUG_ON(IS_ERR(req));
1033 BUG_ON(len < page_offset(pages[locked_pages - 1]) +
1034 PAGE_SIZE - offset);
1036 req->r_callback = writepages_finish;
1037 req->r_inode = inode;
1039 /* Format the osd request message and submit the write */
1043 for (i = 0; i < locked_pages; i++) {
1044 u64 cur_offset = page_offset(pages[i]);
1045 if (offset + len != cur_offset) {
1046 if (op_idx + do_sync + 1 == req->r_num_ops)
1048 osd_req_op_extent_dup_last(req, op_idx,
1049 cur_offset - offset);
1050 dout("writepages got pages at %llu~%llu\n",
1052 osd_req_op_extent_osd_data_pages(req, op_idx,
1055 osd_req_op_extent_update(req, op_idx, len);
1058 offset = cur_offset;
1059 data_pages = pages + i;
1063 set_page_writeback(pages[i]);
1067 if (snap_size != -1) {
1068 len = min(len, snap_size - offset);
1069 } else if (i == locked_pages) {
1070 /* writepages_finish() clears writeback pages
1071 * according to the data length, so make sure
1072 * data length covers all locked pages */
1073 u64 min_len = len + 1 - PAGE_SIZE;
1074 len = min(len, (u64)i_size_read(inode) - offset);
1075 len = max(len, min_len);
1077 dout("writepages got pages at %llu~%llu\n", offset, len);
1079 osd_req_op_extent_osd_data_pages(req, op_idx, data_pages, len,
1081 osd_req_op_extent_update(req, op_idx, len);
1085 osd_req_op_init(req, op_idx, CEPH_OSD_OP_STARTSYNC, 0);
1087 BUG_ON(op_idx + 1 != req->r_num_ops);
1090 if (i < locked_pages) {
1091 BUG_ON(num_ops <= req->r_num_ops);
1092 num_ops -= req->r_num_ops;
1096 /* allocate new pages array for next request */
1098 pages = kmalloc(locked_pages * sizeof (*pages),
1101 pool = fsc->wb_pagevec_pool;
1102 pages = mempool_alloc(pool, GFP_NOFS);
1105 memcpy(pages, data_pages + i,
1106 locked_pages * sizeof(*pages));
1107 memset(data_pages + i, 0,
1108 locked_pages * sizeof(*pages));
1110 BUG_ON(num_ops != req->r_num_ops);
1111 index = pages[i - 1]->index + 1;
1112 /* request message now owns the pages array */
1116 req->r_mtime = inode->i_mtime;
1117 rc = ceph_osdc_start_request(&fsc->client->osdc, req, true);
1121 wbc->nr_to_write -= i;
1125 if (wbc->nr_to_write <= 0)
1129 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr,
1130 pvec.nr ? pvec.pages[0] : NULL);
1131 pagevec_release(&pvec);
1133 if (locked_pages && !done)
1137 if (should_loop && !done) {
1138 /* more to do; loop back to beginning of file */
1139 dout("writepages looping back to beginning of file\n");
1145 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1146 mapping->writeback_index = index;
1149 ceph_osdc_put_request(req);
1150 ceph_put_snap_context(snapc);
1151 dout("writepages done, rc = %d\n", rc);
1158 * See if a given @snapc is either writeable, or already written.
1160 static int context_is_writeable_or_written(struct inode *inode,
1161 struct ceph_snap_context *snapc)
1163 struct ceph_snap_context *oldest = get_oldest_context(inode, NULL);
1164 int ret = !oldest || snapc->seq <= oldest->seq;
1166 ceph_put_snap_context(oldest);
1171 * We are only allowed to write into/dirty the page if the page is
1172 * clean, or already dirty within the same snap context.
1174 * called with page locked.
1175 * return success with page locked,
1176 * or any failure (incl -EAGAIN) with page unlocked.
1178 static int ceph_update_writeable_page(struct file *file,
1179 loff_t pos, unsigned len,
1182 struct inode *inode = file_inode(file);
1183 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1184 struct ceph_inode_info *ci = ceph_inode(inode);
1185 loff_t page_off = pos & PAGE_MASK;
1186 int pos_in_page = pos & ~PAGE_MASK;
1187 int end_in_page = pos_in_page + len;
1190 struct ceph_snap_context *snapc, *oldest;
1192 if (ACCESS_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1193 dout(" page %p forced umount\n", page);
1199 /* writepages currently holds page lock, but if we change that later, */
1200 wait_on_page_writeback(page);
1202 snapc = page_snap_context(page);
1203 if (snapc && snapc != ci->i_head_snapc) {
1205 * this page is already dirty in another (older) snap
1206 * context! is it writeable now?
1208 oldest = get_oldest_context(inode, NULL);
1210 if (snapc->seq > oldest->seq) {
1211 ceph_put_snap_context(oldest);
1212 dout(" page %p snapc %p not current or oldest\n",
1215 * queue for writeback, and wait for snapc to
1216 * be writeable or written
1218 snapc = ceph_get_snap_context(snapc);
1220 ceph_queue_writeback(inode);
1221 r = wait_event_killable(ci->i_cap_wq,
1222 context_is_writeable_or_written(inode, snapc));
1223 ceph_put_snap_context(snapc);
1224 if (r == -ERESTARTSYS)
1228 ceph_put_snap_context(oldest);
1230 /* yay, writeable, do it now (without dropping page lock) */
1231 dout(" page %p snapc %p not current, but oldest\n",
1233 if (!clear_page_dirty_for_io(page))
1235 r = writepage_nounlock(page, NULL);
1241 if (PageUptodate(page)) {
1242 dout(" page %p already uptodate\n", page);
1247 if (pos_in_page == 0 && len == PAGE_SIZE)
1250 /* past end of file? */
1251 i_size = i_size_read(inode);
1253 if (page_off >= i_size ||
1254 (pos_in_page == 0 && (pos+len) >= i_size &&
1255 end_in_page - pos_in_page != PAGE_SIZE)) {
1256 dout(" zeroing %p 0 - %d and %d - %d\n",
1257 page, pos_in_page, end_in_page, (int)PAGE_SIZE);
1258 zero_user_segments(page,
1260 end_in_page, PAGE_SIZE);
1264 /* we need to read it. */
1265 r = readpage_nounlock(file, page);
1275 * We are only allowed to write into/dirty the page if the page is
1276 * clean, or already dirty within the same snap context.
1278 static int ceph_write_begin(struct file *file, struct address_space *mapping,
1279 loff_t pos, unsigned len, unsigned flags,
1280 struct page **pagep, void **fsdata)
1282 struct inode *inode = file_inode(file);
1284 pgoff_t index = pos >> PAGE_SHIFT;
1289 page = grab_cache_page_write_begin(mapping, index, 0);
1293 dout("write_begin file %p inode %p page %p %d~%d\n", file,
1294 inode, page, (int)pos, (int)len);
1296 r = ceph_update_writeable_page(file, pos, len, page);
1301 } while (r == -EAGAIN);
1307 * we don't do anything in here that simple_write_end doesn't do
1308 * except adjust dirty page accounting
1310 static int ceph_write_end(struct file *file, struct address_space *mapping,
1311 loff_t pos, unsigned len, unsigned copied,
1312 struct page *page, void *fsdata)
1314 struct inode *inode = file_inode(file);
1315 unsigned from = pos & (PAGE_SIZE - 1);
1318 dout("write_end file %p inode %p page %p %d~%d (%d)\n", file,
1319 inode, page, (int)pos, (int)copied, (int)len);
1321 /* zero the stale part of the page if we did a short copy */
1323 zero_user_segment(page, from+copied, len);
1325 /* did file size increase? */
1326 if (pos+copied > i_size_read(inode))
1327 check_cap = ceph_inode_set_size(inode, pos+copied);
1329 if (!PageUptodate(page))
1330 SetPageUptodate(page);
1332 set_page_dirty(page);
1338 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL);
1344 * we set .direct_IO to indicate direct io is supported, but since we
1345 * intercept O_DIRECT reads and writes early, this function should
1348 static ssize_t ceph_direct_io(struct kiocb *iocb, struct iov_iter *iter)
1354 const struct address_space_operations ceph_aops = {
1355 .readpage = ceph_readpage,
1356 .readpages = ceph_readpages,
1357 .writepage = ceph_writepage,
1358 .writepages = ceph_writepages_start,
1359 .write_begin = ceph_write_begin,
1360 .write_end = ceph_write_end,
1361 .set_page_dirty = ceph_set_page_dirty,
1362 .invalidatepage = ceph_invalidatepage,
1363 .releasepage = ceph_releasepage,
1364 .direct_IO = ceph_direct_io,
1367 static void ceph_block_sigs(sigset_t *oldset)
1370 siginitsetinv(&mask, sigmask(SIGKILL));
1371 sigprocmask(SIG_BLOCK, &mask, oldset);
1374 static void ceph_restore_sigs(sigset_t *oldset)
1376 sigprocmask(SIG_SETMASK, oldset, NULL);
1382 static int ceph_filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1384 struct inode *inode = file_inode(vma->vm_file);
1385 struct ceph_inode_info *ci = ceph_inode(inode);
1386 struct ceph_file_info *fi = vma->vm_file->private_data;
1387 struct page *pinned_page = NULL;
1388 loff_t off = vmf->pgoff << PAGE_SHIFT;
1392 ceph_block_sigs(&oldset);
1394 dout("filemap_fault %p %llx.%llx %llu~%zd trying to get caps\n",
1395 inode, ceph_vinop(inode), off, (size_t)PAGE_SIZE);
1396 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1397 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1399 want = CEPH_CAP_FILE_CACHE;
1402 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page);
1406 dout("filemap_fault %p %llu~%zd got cap refs on %s\n",
1407 inode, off, (size_t)PAGE_SIZE, ceph_cap_string(got));
1409 if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) ||
1410 ci->i_inline_version == CEPH_INLINE_NONE) {
1411 current->journal_info = vma->vm_file;
1412 ret = filemap_fault(vma, vmf);
1413 current->journal_info = NULL;
1417 dout("filemap_fault %p %llu~%zd dropping cap refs on %s ret %d\n",
1418 inode, off, (size_t)PAGE_SIZE, ceph_cap_string(got), ret);
1420 put_page(pinned_page);
1421 ceph_put_cap_refs(ci, got);
1426 /* read inline data */
1427 if (off >= PAGE_SIZE) {
1428 /* does not support inline data > PAGE_SIZE */
1429 ret = VM_FAULT_SIGBUS;
1432 struct address_space *mapping = inode->i_mapping;
1433 struct page *page = find_or_create_page(mapping, 0,
1434 mapping_gfp_constraint(mapping,
1440 ret1 = __ceph_do_getattr(inode, page,
1441 CEPH_STAT_CAP_INLINE_DATA, true);
1442 if (ret1 < 0 || off >= i_size_read(inode)) {
1448 ret = VM_FAULT_SIGBUS;
1451 if (ret1 < PAGE_SIZE)
1452 zero_user_segment(page, ret1, PAGE_SIZE);
1454 flush_dcache_page(page);
1455 SetPageUptodate(page);
1457 ret = VM_FAULT_MAJOR | VM_FAULT_LOCKED;
1459 dout("filemap_fault %p %llu~%zd read inline data ret %d\n",
1460 inode, off, (size_t)PAGE_SIZE, ret);
1463 ceph_restore_sigs(&oldset);
1465 ret = (ret == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS;
1471 * Reuse write_begin here for simplicity.
1473 static int ceph_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1475 struct inode *inode = file_inode(vma->vm_file);
1476 struct ceph_inode_info *ci = ceph_inode(inode);
1477 struct ceph_file_info *fi = vma->vm_file->private_data;
1478 struct ceph_cap_flush *prealloc_cf;
1479 struct page *page = vmf->page;
1480 loff_t off = page_offset(page);
1481 loff_t size = i_size_read(inode);
1486 prealloc_cf = ceph_alloc_cap_flush();
1488 return VM_FAULT_OOM;
1490 ceph_block_sigs(&oldset);
1492 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1493 struct page *locked_page = NULL;
1498 ret = ceph_uninline_data(vma->vm_file, locked_page);
1500 unlock_page(locked_page);
1505 if (off + PAGE_SIZE <= size)
1508 len = size & ~PAGE_MASK;
1510 dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n",
1511 inode, ceph_vinop(inode), off, len, size);
1512 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1513 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1515 want = CEPH_CAP_FILE_BUFFER;
1518 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, off + len,
1523 dout("page_mkwrite %p %llu~%zd got cap refs on %s\n",
1524 inode, off, len, ceph_cap_string(got));
1526 /* Update time before taking page lock */
1527 file_update_time(vma->vm_file);
1532 if ((off > size) || (page->mapping != inode->i_mapping)) {
1534 ret = VM_FAULT_NOPAGE;
1538 ret = ceph_update_writeable_page(vma->vm_file, off, len, page);
1540 /* success. we'll keep the page locked. */
1541 set_page_dirty(page);
1542 ret = VM_FAULT_LOCKED;
1544 } while (ret == -EAGAIN);
1546 if (ret == VM_FAULT_LOCKED ||
1547 ci->i_inline_version != CEPH_INLINE_NONE) {
1549 spin_lock(&ci->i_ceph_lock);
1550 ci->i_inline_version = CEPH_INLINE_NONE;
1551 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1553 spin_unlock(&ci->i_ceph_lock);
1555 __mark_inode_dirty(inode, dirty);
1558 dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %d\n",
1559 inode, off, len, ceph_cap_string(got), ret);
1560 ceph_put_cap_refs(ci, got);
1562 ceph_restore_sigs(&oldset);
1563 ceph_free_cap_flush(prealloc_cf);
1565 ret = (ret == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS;
1569 void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
1570 char *data, size_t len)
1572 struct address_space *mapping = inode->i_mapping;
1578 if (i_size_read(inode) == 0)
1580 page = find_or_create_page(mapping, 0,
1581 mapping_gfp_constraint(mapping,
1585 if (PageUptodate(page)) {
1592 dout("fill_inline_data %p %llx.%llx len %zu locked_page %p\n",
1593 inode, ceph_vinop(inode), len, locked_page);
1596 void *kaddr = kmap_atomic(page);
1597 memcpy(kaddr, data, len);
1598 kunmap_atomic(kaddr);
1601 if (page != locked_page) {
1602 if (len < PAGE_SIZE)
1603 zero_user_segment(page, len, PAGE_SIZE);
1605 flush_dcache_page(page);
1607 SetPageUptodate(page);
1613 int ceph_uninline_data(struct file *filp, struct page *locked_page)
1615 struct inode *inode = file_inode(filp);
1616 struct ceph_inode_info *ci = ceph_inode(inode);
1617 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1618 struct ceph_osd_request *req;
1619 struct page *page = NULL;
1620 u64 len, inline_version;
1622 bool from_pagecache = false;
1624 spin_lock(&ci->i_ceph_lock);
1625 inline_version = ci->i_inline_version;
1626 spin_unlock(&ci->i_ceph_lock);
1628 dout("uninline_data %p %llx.%llx inline_version %llu\n",
1629 inode, ceph_vinop(inode), inline_version);
1631 if (inline_version == 1 || /* initial version, no data */
1632 inline_version == CEPH_INLINE_NONE)
1637 WARN_ON(!PageUptodate(page));
1638 } else if (ceph_caps_issued(ci) &
1639 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) {
1640 page = find_get_page(inode->i_mapping, 0);
1642 if (PageUptodate(page)) {
1643 from_pagecache = true;
1653 len = i_size_read(inode);
1654 if (len > PAGE_SIZE)
1657 page = __page_cache_alloc(GFP_NOFS);
1662 err = __ceph_do_getattr(inode, page,
1663 CEPH_STAT_CAP_INLINE_DATA, true);
1665 /* no inline data */
1666 if (err == -ENODATA)
1673 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1674 ceph_vino(inode), 0, &len, 0, 1,
1676 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
1683 req->r_mtime = inode->i_mtime;
1684 err = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1686 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1687 ceph_osdc_put_request(req);
1691 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1692 ceph_vino(inode), 0, &len, 1, 3,
1694 CEPH_OSD_FLAG_ONDISK | CEPH_OSD_FLAG_WRITE,
1695 NULL, ci->i_truncate_seq,
1696 ci->i_truncate_size, false);
1702 osd_req_op_extent_osd_data_pages(req, 1, &page, len, 0, false, false);
1705 __le64 xattr_buf = cpu_to_le64(inline_version);
1706 err = osd_req_op_xattr_init(req, 0, CEPH_OSD_OP_CMPXATTR,
1707 "inline_version", &xattr_buf,
1709 CEPH_OSD_CMPXATTR_OP_GT,
1710 CEPH_OSD_CMPXATTR_MODE_U64);
1717 int xattr_len = snprintf(xattr_buf, sizeof(xattr_buf),
1718 "%llu", inline_version);
1719 err = osd_req_op_xattr_init(req, 2, CEPH_OSD_OP_SETXATTR,
1721 xattr_buf, xattr_len, 0, 0);
1726 req->r_mtime = inode->i_mtime;
1727 err = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1729 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1731 ceph_osdc_put_request(req);
1732 if (err == -ECANCELED)
1735 if (page && page != locked_page) {
1736 if (from_pagecache) {
1740 __free_pages(page, 0);
1743 dout("uninline_data %p %llx.%llx inline_version %llu = %d\n",
1744 inode, ceph_vinop(inode), inline_version, err);
1748 static const struct vm_operations_struct ceph_vmops = {
1749 .fault = ceph_filemap_fault,
1750 .page_mkwrite = ceph_page_mkwrite,
1753 int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1755 struct address_space *mapping = file->f_mapping;
1757 if (!mapping->a_ops->readpage)
1759 file_accessed(file);
1760 vma->vm_ops = &ceph_vmops;
1769 static int __ceph_pool_perm_get(struct ceph_inode_info *ci,
1770 s64 pool, struct ceph_string *pool_ns)
1772 struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode);
1773 struct ceph_mds_client *mdsc = fsc->mdsc;
1774 struct ceph_osd_request *rd_req = NULL, *wr_req = NULL;
1775 struct rb_node **p, *parent;
1776 struct ceph_pool_perm *perm;
1777 struct page **pages;
1779 int err = 0, err2 = 0, have = 0;
1781 down_read(&mdsc->pool_perm_rwsem);
1782 p = &mdsc->pool_perm_tree.rb_node;
1784 perm = rb_entry(*p, struct ceph_pool_perm, node);
1785 if (pool < perm->pool)
1787 else if (pool > perm->pool)
1788 p = &(*p)->rb_right;
1790 int ret = ceph_compare_string(pool_ns,
1796 p = &(*p)->rb_right;
1803 up_read(&mdsc->pool_perm_rwsem);
1808 dout("__ceph_pool_perm_get pool %lld ns %.*s no perm cached\n",
1809 pool, (int)pool_ns->len, pool_ns->str);
1811 dout("__ceph_pool_perm_get pool %lld no perm cached\n", pool);
1813 down_write(&mdsc->pool_perm_rwsem);
1814 p = &mdsc->pool_perm_tree.rb_node;
1818 perm = rb_entry(parent, struct ceph_pool_perm, node);
1819 if (pool < perm->pool)
1821 else if (pool > perm->pool)
1822 p = &(*p)->rb_right;
1824 int ret = ceph_compare_string(pool_ns,
1830 p = &(*p)->rb_right;
1838 up_write(&mdsc->pool_perm_rwsem);
1842 rd_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1843 1, false, GFP_NOFS);
1849 rd_req->r_flags = CEPH_OSD_FLAG_READ;
1850 osd_req_op_init(rd_req, 0, CEPH_OSD_OP_STAT, 0);
1851 rd_req->r_base_oloc.pool = pool;
1853 rd_req->r_base_oloc.pool_ns = ceph_get_string(pool_ns);
1854 ceph_oid_printf(&rd_req->r_base_oid, "%llx.00000000", ci->i_vino.ino);
1856 err = ceph_osdc_alloc_messages(rd_req, GFP_NOFS);
1860 wr_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1861 1, false, GFP_NOFS);
1867 wr_req->r_flags = CEPH_OSD_FLAG_WRITE | CEPH_OSD_FLAG_ACK;
1868 osd_req_op_init(wr_req, 0, CEPH_OSD_OP_CREATE, CEPH_OSD_OP_FLAG_EXCL);
1869 ceph_oloc_copy(&wr_req->r_base_oloc, &rd_req->r_base_oloc);
1870 ceph_oid_copy(&wr_req->r_base_oid, &rd_req->r_base_oid);
1872 err = ceph_osdc_alloc_messages(wr_req, GFP_NOFS);
1876 /* one page should be large enough for STAT data */
1877 pages = ceph_alloc_page_vector(1, GFP_KERNEL);
1878 if (IS_ERR(pages)) {
1879 err = PTR_ERR(pages);
1883 osd_req_op_raw_data_in_pages(rd_req, 0, pages, PAGE_SIZE,
1885 err = ceph_osdc_start_request(&fsc->client->osdc, rd_req, false);
1887 wr_req->r_mtime = ci->vfs_inode.i_mtime;
1888 err2 = ceph_osdc_start_request(&fsc->client->osdc, wr_req, false);
1891 err = ceph_osdc_wait_request(&fsc->client->osdc, rd_req);
1893 err2 = ceph_osdc_wait_request(&fsc->client->osdc, wr_req);
1895 if (err >= 0 || err == -ENOENT)
1897 else if (err != -EPERM)
1900 if (err2 == 0 || err2 == -EEXIST)
1902 else if (err2 != -EPERM) {
1907 pool_ns_len = pool_ns ? pool_ns->len : 0;
1908 perm = kmalloc(sizeof(*perm) + pool_ns_len + 1, GFP_NOFS);
1916 perm->pool_ns_len = pool_ns_len;
1917 if (pool_ns_len > 0)
1918 memcpy(perm->pool_ns, pool_ns->str, pool_ns_len);
1919 perm->pool_ns[pool_ns_len] = 0;
1921 rb_link_node(&perm->node, parent, p);
1922 rb_insert_color(&perm->node, &mdsc->pool_perm_tree);
1925 up_write(&mdsc->pool_perm_rwsem);
1927 ceph_osdc_put_request(rd_req);
1928 ceph_osdc_put_request(wr_req);
1933 dout("__ceph_pool_perm_get pool %lld ns %.*s result = %d\n",
1934 pool, (int)pool_ns->len, pool_ns->str, err);
1936 dout("__ceph_pool_perm_get pool %lld result = %d\n", pool, err);
1940 int ceph_pool_perm_check(struct ceph_inode_info *ci, int need)
1943 struct ceph_string *pool_ns;
1946 if (ceph_test_mount_opt(ceph_inode_to_client(&ci->vfs_inode),
1950 spin_lock(&ci->i_ceph_lock);
1951 flags = ci->i_ceph_flags;
1952 pool = ci->i_layout.pool_id;
1953 spin_unlock(&ci->i_ceph_lock);
1955 if (flags & CEPH_I_POOL_PERM) {
1956 if ((need & CEPH_CAP_FILE_RD) && !(flags & CEPH_I_POOL_RD)) {
1957 dout("ceph_pool_perm_check pool %lld no read perm\n",
1961 if ((need & CEPH_CAP_FILE_WR) && !(flags & CEPH_I_POOL_WR)) {
1962 dout("ceph_pool_perm_check pool %lld no write perm\n",
1969 pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
1970 ret = __ceph_pool_perm_get(ci, pool, pool_ns);
1971 ceph_put_string(pool_ns);
1975 flags = CEPH_I_POOL_PERM;
1976 if (ret & POOL_READ)
1977 flags |= CEPH_I_POOL_RD;
1978 if (ret & POOL_WRITE)
1979 flags |= CEPH_I_POOL_WR;
1981 spin_lock(&ci->i_ceph_lock);
1982 if (pool == ci->i_layout.pool_id &&
1983 pool_ns == rcu_dereference_raw(ci->i_layout.pool_ns)) {
1984 ci->i_ceph_flags |= flags;
1986 pool = ci->i_layout.pool_id;
1987 flags = ci->i_ceph_flags;
1989 spin_unlock(&ci->i_ceph_lock);
1993 void ceph_pool_perm_destroy(struct ceph_mds_client *mdsc)
1995 struct ceph_pool_perm *perm;
1998 while (!RB_EMPTY_ROOT(&mdsc->pool_perm_tree)) {
1999 n = rb_first(&mdsc->pool_perm_tree);
2000 perm = rb_entry(n, struct ceph_pool_perm, node);
2001 rb_erase(n, &mdsc->pool_perm_tree);