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>
13 #include <linux/iversion.h>
14 #include <linux/ktime.h>
17 #include "mds_client.h"
20 #include <linux/ceph/osd_client.h>
21 #include <linux/ceph/striper.h>
24 * Ceph address space ops.
26 * There are a few funny things going on here.
28 * The page->private field is used to reference a struct
29 * ceph_snap_context for _every_ dirty page. This indicates which
30 * snapshot the page was logically dirtied in, and thus which snap
31 * context needs to be associated with the osd write during writeback.
33 * Similarly, struct ceph_inode_info maintains a set of counters to
34 * count dirty pages on the inode. In the absence of snapshots,
35 * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count.
37 * When a snapshot is taken (that is, when the client receives
38 * notification that a snapshot was taken), each inode with caps and
39 * with dirty pages (dirty pages implies there is a cap) gets a new
40 * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending
41 * order, new snaps go to the tail). The i_wrbuffer_ref_head count is
42 * moved to capsnap->dirty. (Unless a sync write is currently in
43 * progress. In that case, the capsnap is said to be "pending", new
44 * writes cannot start, and the capsnap isn't "finalized" until the
45 * write completes (or fails) and a final size/mtime for the inode for
46 * that snap can be settled upon.) i_wrbuffer_ref_head is reset to 0.
48 * On writeback, we must submit writes to the osd IN SNAP ORDER. So,
49 * we look for the first capsnap in i_cap_snaps and write out pages in
50 * that snap context _only_. Then we move on to the next capsnap,
51 * eventually reaching the "live" or "head" context (i.e., pages that
52 * are not yet snapped) and are writing the most recently dirtied
55 * Invalidate and so forth must take care to ensure the dirty page
56 * accounting is preserved.
59 #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10))
60 #define CONGESTION_OFF_THRESH(congestion_kb) \
61 (CONGESTION_ON_THRESH(congestion_kb) - \
62 (CONGESTION_ON_THRESH(congestion_kb) >> 2))
64 static inline struct ceph_snap_context *page_snap_context(struct page *page)
66 if (PagePrivate(page))
67 return (void *)page->private;
72 * Dirty a page. Optimistically adjust accounting, on the assumption
73 * that we won't race with invalidate. If we do, readjust.
75 static int ceph_set_page_dirty(struct page *page)
77 struct address_space *mapping = page->mapping;
79 struct ceph_inode_info *ci;
80 struct ceph_snap_context *snapc;
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 return __set_page_dirty_nobuffers(page);
130 * If we are truncating the full page (i.e. offset == 0), adjust the
131 * dirty page counters appropriately. Only called if there is private
134 static void ceph_invalidatepage(struct page *page, unsigned int offset,
138 struct ceph_inode_info *ci;
139 struct ceph_snap_context *snapc = page_snap_context(page);
141 inode = page->mapping->host;
142 ci = ceph_inode(inode);
144 if (offset != 0 || length != PAGE_SIZE) {
145 dout("%p invalidatepage %p idx %lu partial dirty page %u~%u\n",
146 inode, page, page->index, offset, length);
150 ceph_invalidate_fscache_page(inode, page);
152 WARN_ON(!PageLocked(page));
153 if (!PagePrivate(page))
156 dout("%p invalidatepage %p idx %lu full dirty page\n",
157 inode, page, page->index);
159 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
160 ceph_put_snap_context(snapc);
162 ClearPagePrivate(page);
165 static int ceph_releasepage(struct page *page, gfp_t g)
167 dout("%p releasepage %p idx %lu (%sdirty)\n", page->mapping->host,
168 page, page->index, PageDirty(page) ? "" : "not ");
170 /* Can we release the page from the cache? */
171 if (!ceph_release_fscache_page(page, g))
174 return !PagePrivate(page);
177 /* read a single page, without unlocking it. */
178 static int ceph_do_readpage(struct file *filp, struct page *page)
180 struct inode *inode = file_inode(filp);
181 struct ceph_inode_info *ci = ceph_inode(inode);
182 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
183 struct ceph_osd_client *osdc = &fsc->client->osdc;
184 struct ceph_osd_request *req;
185 struct ceph_vino vino = ceph_vino(inode);
187 u64 off = page_offset(page);
190 if (off >= i_size_read(inode)) {
191 zero_user_segment(page, 0, PAGE_SIZE);
192 SetPageUptodate(page);
196 if (ci->i_inline_version != CEPH_INLINE_NONE) {
198 * Uptodate inline data should have been added
199 * into page cache while getting Fcr caps.
203 zero_user_segment(page, 0, PAGE_SIZE);
204 SetPageUptodate(page);
208 err = ceph_readpage_from_fscache(inode, page);
212 dout("readpage ino %llx.%llx file %p off %llu len %llu page %p index %lu\n",
213 vino.ino, vino.snap, filp, off, len, page, page->index);
214 req = ceph_osdc_new_request(osdc, &ci->i_layout, vino, off, &len, 0, 1,
215 CEPH_OSD_OP_READ, CEPH_OSD_FLAG_READ, NULL,
216 ci->i_truncate_seq, ci->i_truncate_size,
221 osd_req_op_extent_osd_data_pages(req, 0, &page, len, 0, false, false);
223 err = ceph_osdc_start_request(osdc, req, false);
225 err = ceph_osdc_wait_request(osdc, req);
227 ceph_update_read_latency(&fsc->mdsc->metric, req->r_start_latency,
228 req->r_end_latency, err);
230 ceph_osdc_put_request(req);
231 dout("readpage result %d\n", err);
236 ceph_fscache_readpage_cancel(inode, page);
237 if (err == -EBLOCKLISTED)
238 fsc->blocklisted = true;
242 /* zero fill remainder of page */
243 zero_user_segment(page, err, PAGE_SIZE);
245 flush_dcache_page(page);
247 SetPageUptodate(page);
248 ceph_readpage_to_fscache(inode, page);
251 return err < 0 ? err : 0;
254 static int ceph_readpage(struct file *filp, struct page *page)
256 int r = ceph_do_readpage(filp, page);
257 if (r != -EINPROGRESS)
265 * Finish an async read(ahead) op.
267 static void finish_read(struct ceph_osd_request *req)
269 struct inode *inode = req->r_inode;
270 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
271 struct ceph_osd_data *osd_data;
272 int rc = req->r_result <= 0 ? req->r_result : 0;
273 int bytes = req->r_result >= 0 ? req->r_result : 0;
277 dout("finish_read %p req %p rc %d bytes %d\n", inode, req, rc, bytes);
278 if (rc == -EBLOCKLISTED)
279 ceph_inode_to_client(inode)->blocklisted = true;
281 /* unlock all pages, zeroing any data we didn't read */
282 osd_data = osd_req_op_extent_osd_data(req, 0);
283 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
284 num_pages = calc_pages_for((u64)osd_data->alignment,
285 (u64)osd_data->length);
286 for (i = 0; i < num_pages; i++) {
287 struct page *page = osd_data->pages[i];
289 if (rc < 0 && rc != -ENOENT) {
290 ceph_fscache_readpage_cancel(inode, page);
293 if (bytes < (int)PAGE_SIZE) {
294 /* zero (remainder of) page */
295 int s = bytes < 0 ? 0 : bytes;
296 zero_user_segment(page, s, PAGE_SIZE);
298 dout("finish_read %p uptodate %p idx %lu\n", inode, page,
300 flush_dcache_page(page);
301 SetPageUptodate(page);
302 ceph_readpage_to_fscache(inode, page);
309 ceph_update_read_latency(&fsc->mdsc->metric, req->r_start_latency,
310 req->r_end_latency, rc);
312 kfree(osd_data->pages);
316 * start an async read(ahead) operation. return nr_pages we submitted
317 * a read for on success, or negative error code.
319 static int start_read(struct inode *inode, struct ceph_rw_context *rw_ctx,
320 struct list_head *page_list, int max)
322 struct ceph_osd_client *osdc =
323 &ceph_inode_to_client(inode)->client->osdc;
324 struct ceph_inode_info *ci = ceph_inode(inode);
325 struct page *page = lru_to_page(page_list);
326 struct ceph_vino vino;
327 struct ceph_osd_request *req;
338 /* caller of readpages does not hold buffer and read caps
339 * (fadvise, madvise and readahead cases) */
340 int want = CEPH_CAP_FILE_CACHE;
341 ret = ceph_try_get_caps(inode, CEPH_CAP_FILE_RD, want,
344 dout("start_read %p, error getting cap\n", inode);
345 } else if (!(got & want)) {
346 dout("start_read %p, no cache cap\n", inode);
351 ceph_put_cap_refs(ci, got);
352 while (!list_empty(page_list)) {
353 page = lru_to_page(page_list);
354 list_del(&page->lru);
361 off = (u64) page_offset(page);
364 next_index = page->index;
365 list_for_each_entry_reverse(page, page_list, lru) {
366 if (page->index != next_index)
370 if (max && nr_pages == max)
373 len = nr_pages << PAGE_SHIFT;
374 dout("start_read %p nr_pages %d is %lld~%lld\n", inode, nr_pages,
376 vino = ceph_vino(inode);
377 req = ceph_osdc_new_request(osdc, &ci->i_layout, vino, off, &len,
378 0, 1, CEPH_OSD_OP_READ,
379 CEPH_OSD_FLAG_READ, NULL,
380 ci->i_truncate_seq, ci->i_truncate_size,
387 /* build page vector */
388 nr_pages = calc_pages_for(0, len);
389 pages = kmalloc_array(nr_pages, sizeof(*pages), GFP_KERNEL);
394 for (i = 0; i < nr_pages; ++i) {
395 page = list_entry(page_list->prev, struct page, lru);
396 BUG_ON(PageLocked(page));
397 list_del(&page->lru);
399 dout("start_read %p adding %p idx %lu\n", inode, page,
401 if (add_to_page_cache_lru(page, &inode->i_data, page->index,
403 ceph_fscache_uncache_page(inode, page);
405 dout("start_read %p add_to_page_cache failed %p\n",
409 len = nr_pages << PAGE_SHIFT;
410 osd_req_op_extent_update(req, 0, len);
417 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false);
418 req->r_callback = finish_read;
419 req->r_inode = inode;
421 dout("start_read %p starting %p %lld~%lld\n", inode, req, off, len);
422 ret = ceph_osdc_start_request(osdc, req, false);
425 ceph_osdc_put_request(req);
427 /* After adding locked pages to page cache, the inode holds cache cap.
428 * So we can drop our cap refs. */
430 ceph_put_cap_refs(ci, got);
435 for (i = 0; i < nr_pages; ++i) {
436 ceph_fscache_readpage_cancel(inode, pages[i]);
437 unlock_page(pages[i]);
439 ceph_put_page_vector(pages, nr_pages, false);
441 ceph_osdc_put_request(req);
444 ceph_put_cap_refs(ci, got);
450 * Read multiple pages. Leave pages we don't read + unlock in page_list;
451 * the caller (VM) cleans them up.
453 static int ceph_readpages(struct file *file, struct address_space *mapping,
454 struct list_head *page_list, unsigned nr_pages)
456 struct inode *inode = file_inode(file);
457 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
458 struct ceph_file_info *fi = file->private_data;
459 struct ceph_rw_context *rw_ctx;
463 if (ceph_inode(inode)->i_inline_version != CEPH_INLINE_NONE)
466 rc = ceph_readpages_from_fscache(mapping->host, mapping, page_list,
472 rw_ctx = ceph_find_rw_context(fi);
473 max = fsc->mount_options->rsize >> PAGE_SHIFT;
474 dout("readpages %p file %p ctx %p nr_pages %d max %d\n",
475 inode, file, rw_ctx, nr_pages, max);
476 while (!list_empty(page_list)) {
477 rc = start_read(inode, rw_ctx, page_list, max);
482 ceph_fscache_readpages_cancel(inode, page_list);
484 dout("readpages %p file %p ret %d\n", inode, file, rc);
488 struct ceph_writeback_ctl
498 * Get ref for the oldest snapc for an inode with dirty data... that is, the
499 * only snap context we are allowed to write back.
501 static struct ceph_snap_context *
502 get_oldest_context(struct inode *inode, struct ceph_writeback_ctl *ctl,
503 struct ceph_snap_context *page_snapc)
505 struct ceph_inode_info *ci = ceph_inode(inode);
506 struct ceph_snap_context *snapc = NULL;
507 struct ceph_cap_snap *capsnap = NULL;
509 spin_lock(&ci->i_ceph_lock);
510 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
511 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap,
512 capsnap->context, capsnap->dirty_pages);
513 if (!capsnap->dirty_pages)
516 /* get i_size, truncate_{seq,size} for page_snapc? */
517 if (snapc && capsnap->context != page_snapc)
521 if (capsnap->writing) {
522 ctl->i_size = i_size_read(inode);
523 ctl->size_stable = false;
525 ctl->i_size = capsnap->size;
526 ctl->size_stable = true;
528 ctl->truncate_size = capsnap->truncate_size;
529 ctl->truncate_seq = capsnap->truncate_seq;
530 ctl->head_snapc = false;
536 snapc = ceph_get_snap_context(capsnap->context);
538 page_snapc == snapc ||
539 page_snapc->seq > snapc->seq)
542 if (!snapc && ci->i_wrbuffer_ref_head) {
543 snapc = ceph_get_snap_context(ci->i_head_snapc);
544 dout(" head snapc %p has %d dirty pages\n",
545 snapc, ci->i_wrbuffer_ref_head);
547 ctl->i_size = i_size_read(inode);
548 ctl->truncate_size = ci->i_truncate_size;
549 ctl->truncate_seq = ci->i_truncate_seq;
550 ctl->size_stable = false;
551 ctl->head_snapc = true;
554 spin_unlock(&ci->i_ceph_lock);
558 static u64 get_writepages_data_length(struct inode *inode,
559 struct page *page, u64 start)
561 struct ceph_inode_info *ci = ceph_inode(inode);
562 struct ceph_snap_context *snapc = page_snap_context(page);
563 struct ceph_cap_snap *capsnap = NULL;
564 u64 end = i_size_read(inode);
566 if (snapc != ci->i_head_snapc) {
568 spin_lock(&ci->i_ceph_lock);
569 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
570 if (capsnap->context == snapc) {
571 if (!capsnap->writing)
577 spin_unlock(&ci->i_ceph_lock);
580 if (end > page_offset(page) + PAGE_SIZE)
581 end = page_offset(page) + PAGE_SIZE;
582 return end > start ? end - start : 0;
586 * Write a single page, but leave the page locked.
588 * If we get a write error, mark the mapping for error, but still adjust the
589 * dirty page accounting (i.e., page is no longer dirty).
591 static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
593 struct inode *inode = page->mapping->host;
594 struct ceph_inode_info *ci = ceph_inode(inode);
595 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
596 struct ceph_snap_context *snapc, *oldest;
597 loff_t page_off = page_offset(page);
599 loff_t len = PAGE_SIZE;
600 struct ceph_writeback_ctl ceph_wbc;
601 struct ceph_osd_client *osdc = &fsc->client->osdc;
602 struct ceph_osd_request *req;
604 dout("writepage %p idx %lu\n", page, page->index);
606 /* verify this is a writeable snap context */
607 snapc = page_snap_context(page);
609 dout("writepage %p page %p not dirty?\n", inode, page);
612 oldest = get_oldest_context(inode, &ceph_wbc, snapc);
613 if (snapc->seq > oldest->seq) {
614 dout("writepage %p page %p snapc %p not writeable - noop\n",
616 /* we should only noop if called by kswapd */
617 WARN_ON(!(current->flags & PF_MEMALLOC));
618 ceph_put_snap_context(oldest);
619 redirty_page_for_writepage(wbc, page);
622 ceph_put_snap_context(oldest);
624 /* is this a partial page at end of file? */
625 if (page_off >= ceph_wbc.i_size) {
626 dout("%p page eof %llu\n", page, ceph_wbc.i_size);
627 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
631 if (ceph_wbc.i_size < page_off + len)
632 len = ceph_wbc.i_size - page_off;
634 dout("writepage %p page %p index %lu on %llu~%llu snapc %p seq %lld\n",
635 inode, page, page->index, page_off, len, snapc, snapc->seq);
637 if (atomic_long_inc_return(&fsc->writeback_count) >
638 CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb))
639 set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC);
641 set_page_writeback(page);
642 req = ceph_osdc_new_request(osdc, &ci->i_layout, ceph_vino(inode), page_off, &len, 0, 1,
643 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, snapc,
644 ceph_wbc.truncate_seq, ceph_wbc.truncate_size,
647 redirty_page_for_writepage(wbc, page);
648 end_page_writeback(page);
652 /* it may be a short write due to an object boundary */
653 WARN_ON_ONCE(len > PAGE_SIZE);
654 osd_req_op_extent_osd_data_pages(req, 0, &page, len, 0, false, false);
655 dout("writepage %llu~%llu (%llu bytes)\n", page_off, len, len);
657 req->r_mtime = inode->i_mtime;
658 err = ceph_osdc_start_request(osdc, req, true);
660 err = ceph_osdc_wait_request(osdc, req);
662 ceph_update_write_latency(&fsc->mdsc->metric, req->r_start_latency,
663 req->r_end_latency, err);
665 ceph_osdc_put_request(req);
670 struct writeback_control tmp_wbc;
673 if (err == -ERESTARTSYS) {
674 /* killed by SIGKILL */
675 dout("writepage interrupted page %p\n", page);
676 redirty_page_for_writepage(wbc, page);
677 end_page_writeback(page);
680 if (err == -EBLOCKLISTED)
681 fsc->blocklisted = true;
682 dout("writepage setting page/mapping error %d %p\n",
684 mapping_set_error(&inode->i_data, err);
685 wbc->pages_skipped++;
687 dout("writepage cleaned page %p\n", page);
688 err = 0; /* vfs expects us to return 0 */
691 ClearPagePrivate(page);
692 end_page_writeback(page);
693 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
694 ceph_put_snap_context(snapc); /* page's reference */
696 if (atomic_long_dec_return(&fsc->writeback_count) <
697 CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb))
698 clear_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC);
703 static int ceph_writepage(struct page *page, struct writeback_control *wbc)
706 struct inode *inode = page->mapping->host;
709 err = writepage_nounlock(page, wbc);
710 if (err == -ERESTARTSYS) {
711 /* direct memory reclaimer was killed by SIGKILL. return 0
712 * to prevent caller from setting mapping/page error */
721 * async writeback completion handler.
723 * If we get an error, set the mapping error bit, but not the individual
726 static void writepages_finish(struct ceph_osd_request *req)
728 struct inode *inode = req->r_inode;
729 struct ceph_inode_info *ci = ceph_inode(inode);
730 struct ceph_osd_data *osd_data;
732 int num_pages, total_pages = 0;
734 int rc = req->r_result;
735 struct ceph_snap_context *snapc = req->r_snapc;
736 struct address_space *mapping = inode->i_mapping;
737 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
740 dout("writepages_finish %p rc %d\n", inode, rc);
742 mapping_set_error(mapping, rc);
743 ceph_set_error_write(ci);
744 if (rc == -EBLOCKLISTED)
745 fsc->blocklisted = true;
747 ceph_clear_error_write(ci);
750 ceph_update_write_latency(&fsc->mdsc->metric, req->r_start_latency,
751 req->r_end_latency, rc);
754 * We lost the cache cap, need to truncate the page before
755 * it is unlocked, otherwise we'd truncate it later in the
756 * page truncation thread, possibly losing some data that
759 remove_page = !(ceph_caps_issued(ci) &
760 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO));
762 /* clean all pages */
763 for (i = 0; i < req->r_num_ops; i++) {
764 if (req->r_ops[i].op != CEPH_OSD_OP_WRITE)
767 osd_data = osd_req_op_extent_osd_data(req, i);
768 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
769 num_pages = calc_pages_for((u64)osd_data->alignment,
770 (u64)osd_data->length);
771 total_pages += num_pages;
772 for (j = 0; j < num_pages; j++) {
773 page = osd_data->pages[j];
775 WARN_ON(!PageUptodate(page));
777 if (atomic_long_dec_return(&fsc->writeback_count) <
778 CONGESTION_OFF_THRESH(
779 fsc->mount_options->congestion_kb))
780 clear_bdi_congested(inode_to_bdi(inode),
783 ceph_put_snap_context(page_snap_context(page));
785 ClearPagePrivate(page);
786 dout("unlocking %p\n", page);
787 end_page_writeback(page);
790 generic_error_remove_page(inode->i_mapping,
795 dout("writepages_finish %p wrote %llu bytes cleaned %d pages\n",
796 inode, osd_data->length, rc >= 0 ? num_pages : 0);
798 release_pages(osd_data->pages, num_pages);
801 ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc);
803 osd_data = osd_req_op_extent_osd_data(req, 0);
804 if (osd_data->pages_from_pool)
805 mempool_free(osd_data->pages, ceph_wb_pagevec_pool);
807 kfree(osd_data->pages);
808 ceph_osdc_put_request(req);
812 * initiate async writeback
814 static int ceph_writepages_start(struct address_space *mapping,
815 struct writeback_control *wbc)
817 struct inode *inode = mapping->host;
818 struct ceph_inode_info *ci = ceph_inode(inode);
819 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
820 struct ceph_vino vino = ceph_vino(inode);
821 pgoff_t index, start_index, end = -1;
822 struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc;
825 unsigned int wsize = i_blocksize(inode);
826 struct ceph_osd_request *req = NULL;
827 struct ceph_writeback_ctl ceph_wbc;
828 bool should_loop, range_whole = false;
831 dout("writepages_start %p (mode=%s)\n", inode,
832 wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
833 (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
835 if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
836 if (ci->i_wrbuffer_ref > 0) {
838 "writepage_start %p %lld forced umount\n",
839 inode, ceph_ino(inode));
841 mapping_set_error(mapping, -EIO);
842 return -EIO; /* we're in a forced umount, don't write! */
844 if (fsc->mount_options->wsize < wsize)
845 wsize = fsc->mount_options->wsize;
849 start_index = wbc->range_cyclic ? mapping->writeback_index : 0;
853 /* find oldest snap context with dirty data */
854 snapc = get_oldest_context(inode, &ceph_wbc, NULL);
856 /* hmm, why does writepages get called when there
858 dout(" no snap context with dirty data?\n");
861 dout(" oldest snapc is %p seq %lld (%d snaps)\n",
862 snapc, snapc->seq, snapc->num_snaps);
865 if (ceph_wbc.head_snapc && snapc != last_snapc) {
866 /* where to start/end? */
867 if (wbc->range_cyclic) {
872 dout(" cyclic, start at %lu\n", index);
874 index = wbc->range_start >> PAGE_SHIFT;
875 end = wbc->range_end >> PAGE_SHIFT;
876 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
878 dout(" not cyclic, %lu to %lu\n", index, end);
880 } else if (!ceph_wbc.head_snapc) {
881 /* Do not respect wbc->range_{start,end}. Dirty pages
882 * in that range can be associated with newer snapc.
883 * They are not writeable until we write all dirty pages
884 * associated with 'snapc' get written */
887 dout(" non-head snapc, range whole\n");
890 ceph_put_snap_context(last_snapc);
893 while (!done && index <= end) {
894 int num_ops = 0, op_idx;
895 unsigned i, pvec_pages, max_pages, locked_pages = 0;
896 struct page **pages = NULL, **data_pages;
898 pgoff_t strip_unit_end = 0;
899 u64 offset = 0, len = 0;
900 bool from_pool = false;
902 max_pages = wsize >> PAGE_SHIFT;
905 pvec_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
906 end, PAGECACHE_TAG_DIRTY);
907 dout("pagevec_lookup_range_tag got %d\n", pvec_pages);
908 if (!pvec_pages && !locked_pages)
910 for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) {
911 page = pvec.pages[i];
912 dout("? %p idx %lu\n", page, page->index);
913 if (locked_pages == 0)
914 lock_page(page); /* first page */
915 else if (!trylock_page(page))
918 /* only dirty pages, or our accounting breaks */
919 if (unlikely(!PageDirty(page)) ||
920 unlikely(page->mapping != mapping)) {
921 dout("!dirty or !mapping %p\n", page);
925 /* only if matching snap context */
926 pgsnapc = page_snap_context(page);
927 if (pgsnapc != snapc) {
928 dout("page snapc %p %lld != oldest %p %lld\n",
929 pgsnapc, pgsnapc->seq, snapc, snapc->seq);
931 !ceph_wbc.head_snapc &&
932 wbc->sync_mode != WB_SYNC_NONE)
937 if (page_offset(page) >= ceph_wbc.i_size) {
938 dout("%p page eof %llu\n",
939 page, ceph_wbc.i_size);
940 if ((ceph_wbc.size_stable ||
941 page_offset(page) >= i_size_read(inode)) &&
942 clear_page_dirty_for_io(page))
943 mapping->a_ops->invalidatepage(page,
948 if (strip_unit_end && (page->index > strip_unit_end)) {
949 dout("end of strip unit %p\n", page);
953 if (PageWriteback(page)) {
954 if (wbc->sync_mode == WB_SYNC_NONE) {
955 dout("%p under writeback\n", page);
959 dout("waiting on writeback %p\n", page);
960 wait_on_page_writeback(page);
963 if (!clear_page_dirty_for_io(page)) {
964 dout("%p !clear_page_dirty_for_io\n", page);
970 * We have something to write. If this is
971 * the first locked page this time through,
972 * calculate max possinle write size and
973 * allocate a page array
975 if (locked_pages == 0) {
980 /* prepare async write request */
981 offset = (u64)page_offset(page);
982 ceph_calc_file_object_mapping(&ci->i_layout,
989 strip_unit_end = page->index +
990 ((len - 1) >> PAGE_SHIFT);
993 max_pages = calc_pages_for(0, (u64)len);
994 pages = kmalloc_array(max_pages,
999 pages = mempool_alloc(ceph_wb_pagevec_pool, GFP_NOFS);
1004 } else if (page->index !=
1005 (offset + len) >> PAGE_SHIFT) {
1006 if (num_ops >= (from_pool ? CEPH_OSD_SLAB_OPS :
1007 CEPH_OSD_MAX_OPS)) {
1008 redirty_page_for_writepage(wbc, page);
1014 offset = (u64)page_offset(page);
1018 /* note position of first page in pvec */
1019 dout("%p will write page %p idx %lu\n",
1020 inode, page, page->index);
1022 if (atomic_long_inc_return(&fsc->writeback_count) >
1023 CONGESTION_ON_THRESH(
1024 fsc->mount_options->congestion_kb)) {
1025 set_bdi_congested(inode_to_bdi(inode),
1030 pages[locked_pages++] = page;
1031 pvec.pages[i] = NULL;
1036 /* did we get anything? */
1038 goto release_pvec_pages;
1041 /* shift unused page to beginning of pvec */
1042 for (j = 0; j < pvec_pages; j++) {
1046 pvec.pages[n] = pvec.pages[j];
1051 if (pvec_pages && i == pvec_pages &&
1052 locked_pages < max_pages) {
1053 dout("reached end pvec, trying for more\n");
1054 pagevec_release(&pvec);
1055 goto get_more_pages;
1060 offset = page_offset(pages[0]);
1063 req = ceph_osdc_new_request(&fsc->client->osdc,
1064 &ci->i_layout, vino,
1065 offset, &len, 0, num_ops,
1066 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
1067 snapc, ceph_wbc.truncate_seq,
1068 ceph_wbc.truncate_size, false);
1070 req = ceph_osdc_new_request(&fsc->client->osdc,
1071 &ci->i_layout, vino,
1076 CEPH_OSD_FLAG_WRITE,
1077 snapc, ceph_wbc.truncate_seq,
1078 ceph_wbc.truncate_size, true);
1079 BUG_ON(IS_ERR(req));
1081 BUG_ON(len < page_offset(pages[locked_pages - 1]) +
1082 PAGE_SIZE - offset);
1084 req->r_callback = writepages_finish;
1085 req->r_inode = inode;
1087 /* Format the osd request message and submit the write */
1091 for (i = 0; i < locked_pages; i++) {
1092 u64 cur_offset = page_offset(pages[i]);
1093 if (offset + len != cur_offset) {
1094 if (op_idx + 1 == req->r_num_ops)
1096 osd_req_op_extent_dup_last(req, op_idx,
1097 cur_offset - offset);
1098 dout("writepages got pages at %llu~%llu\n",
1100 osd_req_op_extent_osd_data_pages(req, op_idx,
1103 osd_req_op_extent_update(req, op_idx, len);
1106 offset = cur_offset;
1107 data_pages = pages + i;
1111 set_page_writeback(pages[i]);
1115 if (ceph_wbc.size_stable) {
1116 len = min(len, ceph_wbc.i_size - offset);
1117 } else if (i == locked_pages) {
1118 /* writepages_finish() clears writeback pages
1119 * according to the data length, so make sure
1120 * data length covers all locked pages */
1121 u64 min_len = len + 1 - PAGE_SIZE;
1122 len = get_writepages_data_length(inode, pages[i - 1],
1124 len = max(len, min_len);
1126 dout("writepages got pages at %llu~%llu\n", offset, len);
1128 osd_req_op_extent_osd_data_pages(req, op_idx, data_pages, len,
1129 0, from_pool, false);
1130 osd_req_op_extent_update(req, op_idx, len);
1132 BUG_ON(op_idx + 1 != req->r_num_ops);
1135 if (i < locked_pages) {
1136 BUG_ON(num_ops <= req->r_num_ops);
1137 num_ops -= req->r_num_ops;
1140 /* allocate new pages array for next request */
1142 pages = kmalloc_array(locked_pages, sizeof(*pages),
1146 pages = mempool_alloc(ceph_wb_pagevec_pool, GFP_NOFS);
1149 memcpy(pages, data_pages + i,
1150 locked_pages * sizeof(*pages));
1151 memset(data_pages + i, 0,
1152 locked_pages * sizeof(*pages));
1154 BUG_ON(num_ops != req->r_num_ops);
1155 index = pages[i - 1]->index + 1;
1156 /* request message now owns the pages array */
1160 req->r_mtime = inode->i_mtime;
1161 rc = ceph_osdc_start_request(&fsc->client->osdc, req, true);
1165 wbc->nr_to_write -= i;
1170 * We stop writing back only if we are not doing
1171 * integrity sync. In case of integrity sync we have to
1172 * keep going until we have written all the pages
1173 * we tagged for writeback prior to entering this loop.
1175 if (wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE)
1179 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr,
1180 pvec.nr ? pvec.pages[0] : NULL);
1181 pagevec_release(&pvec);
1184 if (should_loop && !done) {
1185 /* more to do; loop back to beginning of file */
1186 dout("writepages looping back to beginning of file\n");
1187 end = start_index - 1; /* OK even when start_index == 0 */
1189 /* to write dirty pages associated with next snapc,
1190 * we need to wait until current writes complete */
1191 if (wbc->sync_mode != WB_SYNC_NONE &&
1192 start_index == 0 && /* all dirty pages were checked */
1193 !ceph_wbc.head_snapc) {
1197 while ((index <= end) &&
1198 (nr = pagevec_lookup_tag(&pvec, mapping, &index,
1199 PAGECACHE_TAG_WRITEBACK))) {
1200 for (i = 0; i < nr; i++) {
1201 page = pvec.pages[i];
1202 if (page_snap_context(page) != snapc)
1204 wait_on_page_writeback(page);
1206 pagevec_release(&pvec);
1216 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1217 mapping->writeback_index = index;
1220 ceph_osdc_put_request(req);
1221 ceph_put_snap_context(last_snapc);
1222 dout("writepages dend - startone, rc = %d\n", rc);
1229 * See if a given @snapc is either writeable, or already written.
1231 static int context_is_writeable_or_written(struct inode *inode,
1232 struct ceph_snap_context *snapc)
1234 struct ceph_snap_context *oldest = get_oldest_context(inode, NULL, NULL);
1235 int ret = !oldest || snapc->seq <= oldest->seq;
1237 ceph_put_snap_context(oldest);
1242 * ceph_find_incompatible - find an incompatible context and return it
1243 * @page: page being dirtied
1245 * We are only allowed to write into/dirty a page if the page is
1246 * clean, or already dirty within the same snap context. Returns a
1247 * conflicting context if there is one, NULL if there isn't, or a
1248 * negative error code on other errors.
1250 * Must be called with page lock held.
1252 static struct ceph_snap_context *
1253 ceph_find_incompatible(struct page *page)
1255 struct inode *inode = page->mapping->host;
1256 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1257 struct ceph_inode_info *ci = ceph_inode(inode);
1259 if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1260 dout(" page %p forced umount\n", page);
1261 return ERR_PTR(-EIO);
1265 struct ceph_snap_context *snapc, *oldest;
1267 wait_on_page_writeback(page);
1269 snapc = page_snap_context(page);
1270 if (!snapc || snapc == ci->i_head_snapc)
1274 * this page is already dirty in another (older) snap
1275 * context! is it writeable now?
1277 oldest = get_oldest_context(inode, NULL, NULL);
1278 if (snapc->seq > oldest->seq) {
1279 /* not writeable -- return it for the caller to deal with */
1280 ceph_put_snap_context(oldest);
1281 dout(" page %p snapc %p not current or oldest\n", page, snapc);
1282 return ceph_get_snap_context(snapc);
1284 ceph_put_snap_context(oldest);
1286 /* yay, writeable, do it now (without dropping page lock) */
1287 dout(" page %p snapc %p not current, but oldest\n", page, snapc);
1288 if (clear_page_dirty_for_io(page)) {
1289 int r = writepage_nounlock(page, NULL);
1298 * prep_noread_page - prep a page for writing without reading first
1299 * @page: page being prepared
1300 * @pos: starting position for the write
1301 * @len: length of write
1303 * In some cases, write_begin doesn't need to read at all:
1305 * - file is currently zero-length
1306 * - write that lies in a page that is completely beyond EOF
1307 * - write that covers the the page from start to EOF or beyond it
1309 * If any of these criteria are met, then zero out the unwritten parts
1310 * of the page and return true. Otherwise, return false.
1312 static bool skip_page_read(struct page *page, loff_t pos, size_t len)
1314 struct inode *inode = page->mapping->host;
1315 loff_t i_size = i_size_read(inode);
1316 size_t offset = offset_in_page(pos);
1318 /* Full page write */
1319 if (offset == 0 && len >= PAGE_SIZE)
1322 /* pos beyond last page in the file */
1323 if (pos - offset >= i_size)
1326 /* write that covers the whole page from start to EOF or beyond it */
1327 if (offset == 0 && (pos + len) >= i_size)
1332 zero_user_segments(page, 0, offset, offset + len, PAGE_SIZE);
1337 * We are only allowed to write into/dirty the page if the page is
1338 * clean, or already dirty within the same snap context.
1340 static int ceph_write_begin(struct file *file, struct address_space *mapping,
1341 loff_t pos, unsigned len, unsigned flags,
1342 struct page **pagep, void **fsdata)
1344 struct inode *inode = file_inode(file);
1345 struct ceph_inode_info *ci = ceph_inode(inode);
1346 struct ceph_snap_context *snapc;
1347 struct page *page = NULL;
1348 pgoff_t index = pos >> PAGE_SHIFT;
1351 dout("write_begin file %p inode %p page %p %d~%d\n", file, inode, page, (int)pos, (int)len);
1354 page = grab_cache_page_write_begin(mapping, index, 0);
1360 snapc = ceph_find_incompatible(page);
1362 if (IS_ERR(snapc)) {
1369 ceph_queue_writeback(inode);
1370 r = wait_event_killable(ci->i_cap_wq,
1371 context_is_writeable_or_written(inode, snapc));
1372 ceph_put_snap_context(snapc);
1378 if (PageUptodate(page)) {
1379 dout(" page %p already uptodate\n", page);
1383 /* No need to read in some cases */
1384 if (skip_page_read(page, pos, len))
1388 * We need to read it. If we get back -EINPROGRESS, then the page was
1389 * handed off to fscache and it will be unlocked when the read completes.
1390 * Refind the page in that case so we can reacquire the page lock. Otherwise
1391 * we got a hard error or the read was completed synchronously.
1393 r = ceph_do_readpage(file, page);
1394 if (r != -EINPROGRESS)
1410 * we don't do anything in here that simple_write_end doesn't do
1411 * except adjust dirty page accounting
1413 static int ceph_write_end(struct file *file, struct address_space *mapping,
1414 loff_t pos, unsigned len, unsigned copied,
1415 struct page *page, void *fsdata)
1417 struct inode *inode = file_inode(file);
1418 bool check_cap = false;
1420 dout("write_end file %p inode %p page %p %d~%d (%d)\n", file,
1421 inode, page, (int)pos, (int)copied, (int)len);
1423 /* zero the stale part of the page if we did a short copy */
1424 if (!PageUptodate(page)) {
1429 SetPageUptodate(page);
1432 /* did file size increase? */
1433 if (pos+copied > i_size_read(inode))
1434 check_cap = ceph_inode_set_size(inode, pos+copied);
1436 set_page_dirty(page);
1443 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL);
1449 * we set .direct_IO to indicate direct io is supported, but since we
1450 * intercept O_DIRECT reads and writes early, this function should
1453 static ssize_t ceph_direct_io(struct kiocb *iocb, struct iov_iter *iter)
1459 const struct address_space_operations ceph_aops = {
1460 .readpage = ceph_readpage,
1461 .readpages = ceph_readpages,
1462 .writepage = ceph_writepage,
1463 .writepages = ceph_writepages_start,
1464 .write_begin = ceph_write_begin,
1465 .write_end = ceph_write_end,
1466 .set_page_dirty = ceph_set_page_dirty,
1467 .invalidatepage = ceph_invalidatepage,
1468 .releasepage = ceph_releasepage,
1469 .direct_IO = ceph_direct_io,
1472 static void ceph_block_sigs(sigset_t *oldset)
1475 siginitsetinv(&mask, sigmask(SIGKILL));
1476 sigprocmask(SIG_BLOCK, &mask, oldset);
1479 static void ceph_restore_sigs(sigset_t *oldset)
1481 sigprocmask(SIG_SETMASK, oldset, NULL);
1487 static vm_fault_t ceph_filemap_fault(struct vm_fault *vmf)
1489 struct vm_area_struct *vma = vmf->vma;
1490 struct inode *inode = file_inode(vma->vm_file);
1491 struct ceph_inode_info *ci = ceph_inode(inode);
1492 struct ceph_file_info *fi = vma->vm_file->private_data;
1493 struct page *pinned_page = NULL;
1494 loff_t off = (loff_t)vmf->pgoff << PAGE_SHIFT;
1497 vm_fault_t ret = VM_FAULT_SIGBUS;
1499 ceph_block_sigs(&oldset);
1501 dout("filemap_fault %p %llx.%llx %llu~%zd trying to get caps\n",
1502 inode, ceph_vinop(inode), off, (size_t)PAGE_SIZE);
1503 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1504 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1506 want = CEPH_CAP_FILE_CACHE;
1509 err = ceph_get_caps(vma->vm_file, CEPH_CAP_FILE_RD, want, -1,
1510 &got, &pinned_page);
1514 dout("filemap_fault %p %llu~%zd got cap refs on %s\n",
1515 inode, off, (size_t)PAGE_SIZE, ceph_cap_string(got));
1517 if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) ||
1518 ci->i_inline_version == CEPH_INLINE_NONE) {
1519 CEPH_DEFINE_RW_CONTEXT(rw_ctx, got);
1520 ceph_add_rw_context(fi, &rw_ctx);
1521 ret = filemap_fault(vmf);
1522 ceph_del_rw_context(fi, &rw_ctx);
1523 dout("filemap_fault %p %llu~%zd drop cap refs %s ret %x\n",
1524 inode, off, (size_t)PAGE_SIZE,
1525 ceph_cap_string(got), ret);
1530 put_page(pinned_page);
1531 ceph_put_cap_refs(ci, got);
1536 /* read inline data */
1537 if (off >= PAGE_SIZE) {
1538 /* does not support inline data > PAGE_SIZE */
1539 ret = VM_FAULT_SIGBUS;
1541 struct address_space *mapping = inode->i_mapping;
1542 struct page *page = find_or_create_page(mapping, 0,
1543 mapping_gfp_constraint(mapping,
1549 err = __ceph_do_getattr(inode, page,
1550 CEPH_STAT_CAP_INLINE_DATA, true);
1551 if (err < 0 || off >= i_size_read(inode)) {
1554 ret = vmf_error(err);
1557 if (err < PAGE_SIZE)
1558 zero_user_segment(page, err, PAGE_SIZE);
1560 flush_dcache_page(page);
1561 SetPageUptodate(page);
1563 ret = VM_FAULT_MAJOR | VM_FAULT_LOCKED;
1565 dout("filemap_fault %p %llu~%zd read inline data ret %x\n",
1566 inode, off, (size_t)PAGE_SIZE, ret);
1569 ceph_restore_sigs(&oldset);
1571 ret = vmf_error(err);
1577 * Reuse write_begin here for simplicity.
1579 static vm_fault_t ceph_page_mkwrite(struct vm_fault *vmf)
1581 struct vm_area_struct *vma = vmf->vma;
1582 struct inode *inode = file_inode(vma->vm_file);
1583 struct ceph_inode_info *ci = ceph_inode(inode);
1584 struct ceph_file_info *fi = vma->vm_file->private_data;
1585 struct ceph_cap_flush *prealloc_cf;
1586 struct page *page = vmf->page;
1587 loff_t off = page_offset(page);
1588 loff_t size = i_size_read(inode);
1592 vm_fault_t ret = VM_FAULT_SIGBUS;
1594 prealloc_cf = ceph_alloc_cap_flush();
1596 return VM_FAULT_OOM;
1598 sb_start_pagefault(inode->i_sb);
1599 ceph_block_sigs(&oldset);
1601 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1602 struct page *locked_page = NULL;
1607 err = ceph_uninline_data(vma->vm_file, locked_page);
1609 unlock_page(locked_page);
1614 if (off + PAGE_SIZE <= size)
1617 len = size & ~PAGE_MASK;
1619 dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n",
1620 inode, ceph_vinop(inode), off, len, size);
1621 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1622 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1624 want = CEPH_CAP_FILE_BUFFER;
1627 err = ceph_get_caps(vma->vm_file, CEPH_CAP_FILE_WR, want, off + len,
1632 dout("page_mkwrite %p %llu~%zd got cap refs on %s\n",
1633 inode, off, len, ceph_cap_string(got));
1635 /* Update time before taking page lock */
1636 file_update_time(vma->vm_file);
1637 inode_inc_iversion_raw(inode);
1640 struct ceph_snap_context *snapc;
1644 if (page_mkwrite_check_truncate(page, inode) < 0) {
1646 ret = VM_FAULT_NOPAGE;
1650 snapc = ceph_find_incompatible(page);
1652 /* success. we'll keep the page locked. */
1653 set_page_dirty(page);
1654 ret = VM_FAULT_LOCKED;
1660 if (IS_ERR(snapc)) {
1661 ret = VM_FAULT_SIGBUS;
1665 ceph_queue_writeback(inode);
1666 err = wait_event_killable(ci->i_cap_wq,
1667 context_is_writeable_or_written(inode, snapc));
1668 ceph_put_snap_context(snapc);
1671 if (ret == VM_FAULT_LOCKED ||
1672 ci->i_inline_version != CEPH_INLINE_NONE) {
1674 spin_lock(&ci->i_ceph_lock);
1675 ci->i_inline_version = CEPH_INLINE_NONE;
1676 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1678 spin_unlock(&ci->i_ceph_lock);
1680 __mark_inode_dirty(inode, dirty);
1683 dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %x\n",
1684 inode, off, len, ceph_cap_string(got), ret);
1685 ceph_put_cap_refs(ci, got);
1687 ceph_restore_sigs(&oldset);
1688 sb_end_pagefault(inode->i_sb);
1689 ceph_free_cap_flush(prealloc_cf);
1691 ret = vmf_error(err);
1695 void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
1696 char *data, size_t len)
1698 struct address_space *mapping = inode->i_mapping;
1704 if (i_size_read(inode) == 0)
1706 page = find_or_create_page(mapping, 0,
1707 mapping_gfp_constraint(mapping,
1711 if (PageUptodate(page)) {
1718 dout("fill_inline_data %p %llx.%llx len %zu locked_page %p\n",
1719 inode, ceph_vinop(inode), len, locked_page);
1722 void *kaddr = kmap_atomic(page);
1723 memcpy(kaddr, data, len);
1724 kunmap_atomic(kaddr);
1727 if (page != locked_page) {
1728 if (len < PAGE_SIZE)
1729 zero_user_segment(page, len, PAGE_SIZE);
1731 flush_dcache_page(page);
1733 SetPageUptodate(page);
1739 int ceph_uninline_data(struct file *filp, struct page *locked_page)
1741 struct inode *inode = file_inode(filp);
1742 struct ceph_inode_info *ci = ceph_inode(inode);
1743 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1744 struct ceph_osd_request *req;
1745 struct page *page = NULL;
1746 u64 len, inline_version;
1748 bool from_pagecache = false;
1750 spin_lock(&ci->i_ceph_lock);
1751 inline_version = ci->i_inline_version;
1752 spin_unlock(&ci->i_ceph_lock);
1754 dout("uninline_data %p %llx.%llx inline_version %llu\n",
1755 inode, ceph_vinop(inode), inline_version);
1757 if (inline_version == 1 || /* initial version, no data */
1758 inline_version == CEPH_INLINE_NONE)
1763 WARN_ON(!PageUptodate(page));
1764 } else if (ceph_caps_issued(ci) &
1765 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) {
1766 page = find_get_page(inode->i_mapping, 0);
1768 if (PageUptodate(page)) {
1769 from_pagecache = true;
1779 len = i_size_read(inode);
1780 if (len > PAGE_SIZE)
1783 page = __page_cache_alloc(GFP_NOFS);
1788 err = __ceph_do_getattr(inode, page,
1789 CEPH_STAT_CAP_INLINE_DATA, true);
1791 /* no inline data */
1792 if (err == -ENODATA)
1799 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1800 ceph_vino(inode), 0, &len, 0, 1,
1801 CEPH_OSD_OP_CREATE, CEPH_OSD_FLAG_WRITE,
1808 req->r_mtime = inode->i_mtime;
1809 err = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1811 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1812 ceph_osdc_put_request(req);
1816 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1817 ceph_vino(inode), 0, &len, 1, 3,
1818 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
1819 NULL, ci->i_truncate_seq,
1820 ci->i_truncate_size, false);
1826 osd_req_op_extent_osd_data_pages(req, 1, &page, len, 0, false, false);
1829 __le64 xattr_buf = cpu_to_le64(inline_version);
1830 err = osd_req_op_xattr_init(req, 0, CEPH_OSD_OP_CMPXATTR,
1831 "inline_version", &xattr_buf,
1833 CEPH_OSD_CMPXATTR_OP_GT,
1834 CEPH_OSD_CMPXATTR_MODE_U64);
1841 int xattr_len = snprintf(xattr_buf, sizeof(xattr_buf),
1842 "%llu", inline_version);
1843 err = osd_req_op_xattr_init(req, 2, CEPH_OSD_OP_SETXATTR,
1845 xattr_buf, xattr_len, 0, 0);
1850 req->r_mtime = inode->i_mtime;
1851 err = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1853 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1855 ceph_update_write_latency(&fsc->mdsc->metric, req->r_start_latency,
1856 req->r_end_latency, err);
1859 ceph_osdc_put_request(req);
1860 if (err == -ECANCELED)
1863 if (page && page != locked_page) {
1864 if (from_pagecache) {
1868 __free_pages(page, 0);
1871 dout("uninline_data %p %llx.%llx inline_version %llu = %d\n",
1872 inode, ceph_vinop(inode), inline_version, err);
1876 static const struct vm_operations_struct ceph_vmops = {
1877 .fault = ceph_filemap_fault,
1878 .page_mkwrite = ceph_page_mkwrite,
1881 int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1883 struct address_space *mapping = file->f_mapping;
1885 if (!mapping->a_ops->readpage)
1887 file_accessed(file);
1888 vma->vm_ops = &ceph_vmops;
1897 static int __ceph_pool_perm_get(struct ceph_inode_info *ci,
1898 s64 pool, struct ceph_string *pool_ns)
1900 struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode);
1901 struct ceph_mds_client *mdsc = fsc->mdsc;
1902 struct ceph_osd_request *rd_req = NULL, *wr_req = NULL;
1903 struct rb_node **p, *parent;
1904 struct ceph_pool_perm *perm;
1905 struct page **pages;
1907 int err = 0, err2 = 0, have = 0;
1909 down_read(&mdsc->pool_perm_rwsem);
1910 p = &mdsc->pool_perm_tree.rb_node;
1912 perm = rb_entry(*p, struct ceph_pool_perm, node);
1913 if (pool < perm->pool)
1915 else if (pool > perm->pool)
1916 p = &(*p)->rb_right;
1918 int ret = ceph_compare_string(pool_ns,
1924 p = &(*p)->rb_right;
1931 up_read(&mdsc->pool_perm_rwsem);
1936 dout("__ceph_pool_perm_get pool %lld ns %.*s no perm cached\n",
1937 pool, (int)pool_ns->len, pool_ns->str);
1939 dout("__ceph_pool_perm_get pool %lld no perm cached\n", pool);
1941 down_write(&mdsc->pool_perm_rwsem);
1942 p = &mdsc->pool_perm_tree.rb_node;
1946 perm = rb_entry(parent, struct ceph_pool_perm, node);
1947 if (pool < perm->pool)
1949 else if (pool > perm->pool)
1950 p = &(*p)->rb_right;
1952 int ret = ceph_compare_string(pool_ns,
1958 p = &(*p)->rb_right;
1966 up_write(&mdsc->pool_perm_rwsem);
1970 rd_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1971 1, false, GFP_NOFS);
1977 rd_req->r_flags = CEPH_OSD_FLAG_READ;
1978 osd_req_op_init(rd_req, 0, CEPH_OSD_OP_STAT, 0);
1979 rd_req->r_base_oloc.pool = pool;
1981 rd_req->r_base_oloc.pool_ns = ceph_get_string(pool_ns);
1982 ceph_oid_printf(&rd_req->r_base_oid, "%llx.00000000", ci->i_vino.ino);
1984 err = ceph_osdc_alloc_messages(rd_req, GFP_NOFS);
1988 wr_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1989 1, false, GFP_NOFS);
1995 wr_req->r_flags = CEPH_OSD_FLAG_WRITE;
1996 osd_req_op_init(wr_req, 0, CEPH_OSD_OP_CREATE, CEPH_OSD_OP_FLAG_EXCL);
1997 ceph_oloc_copy(&wr_req->r_base_oloc, &rd_req->r_base_oloc);
1998 ceph_oid_copy(&wr_req->r_base_oid, &rd_req->r_base_oid);
2000 err = ceph_osdc_alloc_messages(wr_req, GFP_NOFS);
2004 /* one page should be large enough for STAT data */
2005 pages = ceph_alloc_page_vector(1, GFP_KERNEL);
2006 if (IS_ERR(pages)) {
2007 err = PTR_ERR(pages);
2011 osd_req_op_raw_data_in_pages(rd_req, 0, pages, PAGE_SIZE,
2013 err = ceph_osdc_start_request(&fsc->client->osdc, rd_req, false);
2015 wr_req->r_mtime = ci->vfs_inode.i_mtime;
2016 err2 = ceph_osdc_start_request(&fsc->client->osdc, wr_req, false);
2019 err = ceph_osdc_wait_request(&fsc->client->osdc, rd_req);
2021 err2 = ceph_osdc_wait_request(&fsc->client->osdc, wr_req);
2023 if (err >= 0 || err == -ENOENT)
2025 else if (err != -EPERM) {
2026 if (err == -EBLOCKLISTED)
2027 fsc->blocklisted = true;
2031 if (err2 == 0 || err2 == -EEXIST)
2033 else if (err2 != -EPERM) {
2034 if (err2 == -EBLOCKLISTED)
2035 fsc->blocklisted = true;
2040 pool_ns_len = pool_ns ? pool_ns->len : 0;
2041 perm = kmalloc(sizeof(*perm) + pool_ns_len + 1, GFP_NOFS);
2049 perm->pool_ns_len = pool_ns_len;
2050 if (pool_ns_len > 0)
2051 memcpy(perm->pool_ns, pool_ns->str, pool_ns_len);
2052 perm->pool_ns[pool_ns_len] = 0;
2054 rb_link_node(&perm->node, parent, p);
2055 rb_insert_color(&perm->node, &mdsc->pool_perm_tree);
2058 up_write(&mdsc->pool_perm_rwsem);
2060 ceph_osdc_put_request(rd_req);
2061 ceph_osdc_put_request(wr_req);
2066 dout("__ceph_pool_perm_get pool %lld ns %.*s result = %d\n",
2067 pool, (int)pool_ns->len, pool_ns->str, err);
2069 dout("__ceph_pool_perm_get pool %lld result = %d\n", pool, err);
2073 int ceph_pool_perm_check(struct inode *inode, int need)
2075 struct ceph_inode_info *ci = ceph_inode(inode);
2076 struct ceph_string *pool_ns;
2080 if (ci->i_vino.snap != CEPH_NOSNAP) {
2082 * Pool permission check needs to write to the first object.
2083 * But for snapshot, head of the first object may have alread
2084 * been deleted. Skip check to avoid creating orphan object.
2089 if (ceph_test_mount_opt(ceph_inode_to_client(inode),
2093 spin_lock(&ci->i_ceph_lock);
2094 flags = ci->i_ceph_flags;
2095 pool = ci->i_layout.pool_id;
2096 spin_unlock(&ci->i_ceph_lock);
2098 if (flags & CEPH_I_POOL_PERM) {
2099 if ((need & CEPH_CAP_FILE_RD) && !(flags & CEPH_I_POOL_RD)) {
2100 dout("ceph_pool_perm_check pool %lld no read perm\n",
2104 if ((need & CEPH_CAP_FILE_WR) && !(flags & CEPH_I_POOL_WR)) {
2105 dout("ceph_pool_perm_check pool %lld no write perm\n",
2112 pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
2113 ret = __ceph_pool_perm_get(ci, pool, pool_ns);
2114 ceph_put_string(pool_ns);
2118 flags = CEPH_I_POOL_PERM;
2119 if (ret & POOL_READ)
2120 flags |= CEPH_I_POOL_RD;
2121 if (ret & POOL_WRITE)
2122 flags |= CEPH_I_POOL_WR;
2124 spin_lock(&ci->i_ceph_lock);
2125 if (pool == ci->i_layout.pool_id &&
2126 pool_ns == rcu_dereference_raw(ci->i_layout.pool_ns)) {
2127 ci->i_ceph_flags |= flags;
2129 pool = ci->i_layout.pool_id;
2130 flags = ci->i_ceph_flags;
2132 spin_unlock(&ci->i_ceph_lock);
2136 void ceph_pool_perm_destroy(struct ceph_mds_client *mdsc)
2138 struct ceph_pool_perm *perm;
2141 while (!RB_EMPTY_ROOT(&mdsc->pool_perm_tree)) {
2142 n = rb_first(&mdsc->pool_perm_tree);
2143 perm = rb_entry(n, struct ceph_pool_perm, node);
2144 rb_erase(n, &mdsc->pool_perm_tree);