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>
15 #include <linux/netfs.h>
18 #include "mds_client.h"
21 #include <linux/ceph/osd_client.h>
22 #include <linux/ceph/striper.h>
25 * Ceph address space ops.
27 * There are a few funny things going on here.
29 * The page->private field is used to reference a struct
30 * ceph_snap_context for _every_ dirty page. This indicates which
31 * snapshot the page was logically dirtied in, and thus which snap
32 * context needs to be associated with the osd write during writeback.
34 * Similarly, struct ceph_inode_info maintains a set of counters to
35 * count dirty pages on the inode. In the absence of snapshots,
36 * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count.
38 * When a snapshot is taken (that is, when the client receives
39 * notification that a snapshot was taken), each inode with caps and
40 * with dirty pages (dirty pages implies there is a cap) gets a new
41 * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending
42 * order, new snaps go to the tail). The i_wrbuffer_ref_head count is
43 * moved to capsnap->dirty. (Unless a sync write is currently in
44 * progress. In that case, the capsnap is said to be "pending", new
45 * writes cannot start, and the capsnap isn't "finalized" until the
46 * write completes (or fails) and a final size/mtime for the inode for
47 * that snap can be settled upon.) i_wrbuffer_ref_head is reset to 0.
49 * On writeback, we must submit writes to the osd IN SNAP ORDER. So,
50 * we look for the first capsnap in i_cap_snaps and write out pages in
51 * that snap context _only_. Then we move on to the next capsnap,
52 * eventually reaching the "live" or "head" context (i.e., pages that
53 * are not yet snapped) and are writing the most recently dirtied
56 * Invalidate and so forth must take care to ensure the dirty page
57 * accounting is preserved.
60 #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10))
61 #define CONGESTION_OFF_THRESH(congestion_kb) \
62 (CONGESTION_ON_THRESH(congestion_kb) - \
63 (CONGESTION_ON_THRESH(congestion_kb) >> 2))
65 static int ceph_netfs_check_write_begin(struct file *file, loff_t pos, unsigned int len,
66 struct page *page, void **_fsdata);
68 static inline struct ceph_snap_context *page_snap_context(struct page *page)
70 if (PagePrivate(page))
71 return (void *)page->private;
76 * Dirty a page. Optimistically adjust accounting, on the assumption
77 * that we won't race with invalidate. If we do, readjust.
79 static int ceph_set_page_dirty(struct page *page)
81 struct address_space *mapping = page->mapping;
83 struct ceph_inode_info *ci;
84 struct ceph_snap_context *snapc;
86 if (PageDirty(page)) {
87 dout("%p set_page_dirty %p idx %lu -- already dirty\n",
88 mapping->host, page, page->index);
89 BUG_ON(!PagePrivate(page));
93 inode = mapping->host;
94 ci = ceph_inode(inode);
97 spin_lock(&ci->i_ceph_lock);
98 BUG_ON(ci->i_wr_ref == 0); // caller should hold Fw reference
99 if (__ceph_have_pending_cap_snap(ci)) {
100 struct ceph_cap_snap *capsnap =
101 list_last_entry(&ci->i_cap_snaps,
102 struct ceph_cap_snap,
104 snapc = ceph_get_snap_context(capsnap->context);
105 capsnap->dirty_pages++;
107 BUG_ON(!ci->i_head_snapc);
108 snapc = ceph_get_snap_context(ci->i_head_snapc);
109 ++ci->i_wrbuffer_ref_head;
111 if (ci->i_wrbuffer_ref == 0)
113 ++ci->i_wrbuffer_ref;
114 dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d "
115 "snapc %p seq %lld (%d snaps)\n",
116 mapping->host, page, page->index,
117 ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1,
118 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
119 snapc, snapc->seq, snapc->num_snaps);
120 spin_unlock(&ci->i_ceph_lock);
123 * Reference snap context in page->private. Also set
124 * PagePrivate so that we get invalidatepage callback.
126 BUG_ON(PagePrivate(page));
127 attach_page_private(page, snapc);
129 return __set_page_dirty_nobuffers(page);
133 * If we are truncating the full page (i.e. offset == 0), adjust the
134 * dirty page counters appropriately. Only called if there is private
137 static void ceph_invalidatepage(struct page *page, unsigned int offset,
141 struct ceph_inode_info *ci;
142 struct ceph_snap_context *snapc;
144 wait_on_page_fscache(page);
146 inode = page->mapping->host;
147 ci = ceph_inode(inode);
149 if (offset != 0 || length != thp_size(page)) {
150 dout("%p invalidatepage %p idx %lu partial dirty page %u~%u\n",
151 inode, page, page->index, offset, length);
155 WARN_ON(!PageLocked(page));
156 if (!PagePrivate(page))
159 dout("%p invalidatepage %p idx %lu full dirty page\n",
160 inode, page, page->index);
162 snapc = detach_page_private(page);
163 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
164 ceph_put_snap_context(snapc);
167 static int ceph_releasepage(struct page *page, gfp_t gfp)
169 dout("%p releasepage %p idx %lu (%sdirty)\n", page->mapping->host,
170 page, page->index, PageDirty(page) ? "" : "not ");
172 if (PageFsCache(page)) {
173 if (!(gfp & __GFP_DIRECT_RECLAIM) || !(gfp & __GFP_FS))
175 wait_on_page_fscache(page);
177 return !PagePrivate(page);
180 static void ceph_netfs_expand_readahead(struct netfs_read_request *rreq)
182 struct inode *inode = rreq->mapping->host;
183 struct ceph_inode_info *ci = ceph_inode(inode);
184 struct ceph_file_layout *lo = &ci->i_layout;
188 /* Expand the start downward */
189 blockno = div_u64_rem(rreq->start, lo->stripe_unit, &blockoff);
190 rreq->start = blockno * lo->stripe_unit;
191 rreq->len += blockoff;
193 /* Now, round up the length to the next block */
194 rreq->len = roundup(rreq->len, lo->stripe_unit);
197 static bool ceph_netfs_clamp_length(struct netfs_read_subrequest *subreq)
199 struct inode *inode = subreq->rreq->mapping->host;
200 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
201 struct ceph_inode_info *ci = ceph_inode(inode);
205 /* Truncate the extent at the end of the current block */
206 ceph_calc_file_object_mapping(&ci->i_layout, subreq->start, subreq->len,
207 &objno, &objoff, &xlen);
208 subreq->len = min(xlen, fsc->mount_options->rsize);
212 static void finish_netfs_read(struct ceph_osd_request *req)
214 struct ceph_fs_client *fsc = ceph_inode_to_client(req->r_inode);
215 struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0);
216 struct netfs_read_subrequest *subreq = req->r_priv;
218 int err = req->r_result;
220 ceph_update_read_metrics(&fsc->mdsc->metric, req->r_start_latency,
221 req->r_end_latency, osd_data->length, err);
223 dout("%s: result %d subreq->len=%zu i_size=%lld\n", __func__, req->r_result,
224 subreq->len, i_size_read(req->r_inode));
226 /* no object means success but no data */
229 else if (err == -EBLOCKLISTED)
230 fsc->blocklisted = true;
232 if (err >= 0 && err < subreq->len)
233 __set_bit(NETFS_SREQ_CLEAR_TAIL, &subreq->flags);
235 netfs_subreq_terminated(subreq, err, true);
237 num_pages = calc_pages_for(osd_data->alignment, osd_data->length);
238 ceph_put_page_vector(osd_data->pages, num_pages, false);
242 static void ceph_netfs_issue_op(struct netfs_read_subrequest *subreq)
244 struct netfs_read_request *rreq = subreq->rreq;
245 struct inode *inode = rreq->mapping->host;
246 struct ceph_inode_info *ci = ceph_inode(inode);
247 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
248 struct ceph_osd_request *req;
249 struct ceph_vino vino = ceph_vino(inode);
250 struct iov_iter iter;
254 u64 len = subreq->len;
256 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, vino, subreq->start, &len,
257 0, 1, CEPH_OSD_OP_READ,
258 CEPH_OSD_FLAG_READ | fsc->client->osdc.client->options->read_from_replica,
259 NULL, ci->i_truncate_seq, ci->i_truncate_size, false);
266 dout("%s: pos=%llu orig_len=%zu len=%llu\n", __func__, subreq->start, subreq->len, len);
267 iov_iter_xarray(&iter, READ, &rreq->mapping->i_pages, subreq->start, len);
268 err = iov_iter_get_pages_alloc(&iter, &pages, len, &page_off);
270 dout("%s: iov_ter_get_pages_alloc returned %d\n", __func__, err);
274 /* should always give us a page-aligned read */
275 WARN_ON_ONCE(page_off);
278 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false);
279 req->r_callback = finish_netfs_read;
280 req->r_priv = subreq;
281 req->r_inode = inode;
284 err = ceph_osdc_start_request(req->r_osdc, req, false);
288 ceph_osdc_put_request(req);
290 netfs_subreq_terminated(subreq, err, false);
291 dout("%s: result %d\n", __func__, err);
294 static void ceph_init_rreq(struct netfs_read_request *rreq, struct file *file)
298 static void ceph_readahead_cleanup(struct address_space *mapping, void *priv)
300 struct inode *inode = mapping->host;
301 struct ceph_inode_info *ci = ceph_inode(inode);
302 int got = (uintptr_t)priv;
305 ceph_put_cap_refs(ci, got);
308 static const struct netfs_read_request_ops ceph_netfs_read_ops = {
309 .init_rreq = ceph_init_rreq,
310 .is_cache_enabled = ceph_is_cache_enabled,
311 .begin_cache_operation = ceph_begin_cache_operation,
312 .issue_op = ceph_netfs_issue_op,
313 .expand_readahead = ceph_netfs_expand_readahead,
314 .clamp_length = ceph_netfs_clamp_length,
315 .check_write_begin = ceph_netfs_check_write_begin,
316 .cleanup = ceph_readahead_cleanup,
319 /* read a single page, without unlocking it. */
320 static int ceph_readpage(struct file *file, struct page *page)
322 struct inode *inode = file_inode(file);
323 struct ceph_inode_info *ci = ceph_inode(inode);
324 struct ceph_vino vino = ceph_vino(inode);
325 u64 off = page_offset(page);
326 u64 len = thp_size(page);
328 if (ci->i_inline_version != CEPH_INLINE_NONE) {
330 * Uptodate inline data should have been added
331 * into page cache while getting Fcr caps.
337 zero_user_segment(page, 0, thp_size(page));
338 SetPageUptodate(page);
343 dout("readpage ino %llx.%llx file %p off %llu len %llu page %p index %lu\n",
344 vino.ino, vino.snap, file, off, len, page, page->index);
346 return netfs_readpage(file, page, &ceph_netfs_read_ops, NULL);
349 static void ceph_readahead(struct readahead_control *ractl)
351 struct inode *inode = file_inode(ractl->file);
352 struct ceph_file_info *fi = ractl->file->private_data;
353 struct ceph_rw_context *rw_ctx;
357 if (ceph_inode(inode)->i_inline_version != CEPH_INLINE_NONE)
360 rw_ctx = ceph_find_rw_context(fi);
363 * readahead callers do not necessarily hold Fcb caps
364 * (e.g. fadvise, madvise).
366 int want = CEPH_CAP_FILE_CACHE;
368 ret = ceph_try_get_caps(inode, CEPH_CAP_FILE_RD, want, true, &got);
370 dout("start_read %p, error getting cap\n", inode);
371 else if (!(got & want))
372 dout("start_read %p, no cache cap\n", inode);
377 netfs_readahead(ractl, &ceph_netfs_read_ops, (void *)(uintptr_t)got);
380 struct ceph_writeback_ctl
390 * Get ref for the oldest snapc for an inode with dirty data... that is, the
391 * only snap context we are allowed to write back.
393 static struct ceph_snap_context *
394 get_oldest_context(struct inode *inode, struct ceph_writeback_ctl *ctl,
395 struct ceph_snap_context *page_snapc)
397 struct ceph_inode_info *ci = ceph_inode(inode);
398 struct ceph_snap_context *snapc = NULL;
399 struct ceph_cap_snap *capsnap = NULL;
401 spin_lock(&ci->i_ceph_lock);
402 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
403 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap,
404 capsnap->context, capsnap->dirty_pages);
405 if (!capsnap->dirty_pages)
408 /* get i_size, truncate_{seq,size} for page_snapc? */
409 if (snapc && capsnap->context != page_snapc)
413 if (capsnap->writing) {
414 ctl->i_size = i_size_read(inode);
415 ctl->size_stable = false;
417 ctl->i_size = capsnap->size;
418 ctl->size_stable = true;
420 ctl->truncate_size = capsnap->truncate_size;
421 ctl->truncate_seq = capsnap->truncate_seq;
422 ctl->head_snapc = false;
428 snapc = ceph_get_snap_context(capsnap->context);
430 page_snapc == snapc ||
431 page_snapc->seq > snapc->seq)
434 if (!snapc && ci->i_wrbuffer_ref_head) {
435 snapc = ceph_get_snap_context(ci->i_head_snapc);
436 dout(" head snapc %p has %d dirty pages\n",
437 snapc, ci->i_wrbuffer_ref_head);
439 ctl->i_size = i_size_read(inode);
440 ctl->truncate_size = ci->i_truncate_size;
441 ctl->truncate_seq = ci->i_truncate_seq;
442 ctl->size_stable = false;
443 ctl->head_snapc = true;
446 spin_unlock(&ci->i_ceph_lock);
450 static u64 get_writepages_data_length(struct inode *inode,
451 struct page *page, u64 start)
453 struct ceph_inode_info *ci = ceph_inode(inode);
454 struct ceph_snap_context *snapc = page_snap_context(page);
455 struct ceph_cap_snap *capsnap = NULL;
456 u64 end = i_size_read(inode);
458 if (snapc != ci->i_head_snapc) {
460 spin_lock(&ci->i_ceph_lock);
461 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
462 if (capsnap->context == snapc) {
463 if (!capsnap->writing)
469 spin_unlock(&ci->i_ceph_lock);
472 if (end > page_offset(page) + thp_size(page))
473 end = page_offset(page) + thp_size(page);
474 return end > start ? end - start : 0;
478 * Write a single page, but leave the page locked.
480 * If we get a write error, mark the mapping for error, but still adjust the
481 * dirty page accounting (i.e., page is no longer dirty).
483 static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
485 struct inode *inode = page->mapping->host;
486 struct ceph_inode_info *ci = ceph_inode(inode);
487 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
488 struct ceph_snap_context *snapc, *oldest;
489 loff_t page_off = page_offset(page);
491 loff_t len = thp_size(page);
492 struct ceph_writeback_ctl ceph_wbc;
493 struct ceph_osd_client *osdc = &fsc->client->osdc;
494 struct ceph_osd_request *req;
496 dout("writepage %p idx %lu\n", page, page->index);
498 /* verify this is a writeable snap context */
499 snapc = page_snap_context(page);
501 dout("writepage %p page %p not dirty?\n", inode, page);
504 oldest = get_oldest_context(inode, &ceph_wbc, snapc);
505 if (snapc->seq > oldest->seq) {
506 dout("writepage %p page %p snapc %p not writeable - noop\n",
508 /* we should only noop if called by kswapd */
509 WARN_ON(!(current->flags & PF_MEMALLOC));
510 ceph_put_snap_context(oldest);
511 redirty_page_for_writepage(wbc, page);
514 ceph_put_snap_context(oldest);
516 /* is this a partial page at end of file? */
517 if (page_off >= ceph_wbc.i_size) {
518 dout("%p page eof %llu\n", page, ceph_wbc.i_size);
519 page->mapping->a_ops->invalidatepage(page, 0, thp_size(page));
523 if (ceph_wbc.i_size < page_off + len)
524 len = ceph_wbc.i_size - page_off;
526 dout("writepage %p page %p index %lu on %llu~%llu snapc %p seq %lld\n",
527 inode, page, page->index, page_off, len, snapc, snapc->seq);
529 if (atomic_long_inc_return(&fsc->writeback_count) >
530 CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb))
531 set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC);
533 set_page_writeback(page);
534 req = ceph_osdc_new_request(osdc, &ci->i_layout, ceph_vino(inode), page_off, &len, 0, 1,
535 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, snapc,
536 ceph_wbc.truncate_seq, ceph_wbc.truncate_size,
539 redirty_page_for_writepage(wbc, page);
540 end_page_writeback(page);
544 /* it may be a short write due to an object boundary */
545 WARN_ON_ONCE(len > thp_size(page));
546 osd_req_op_extent_osd_data_pages(req, 0, &page, len, 0, false, false);
547 dout("writepage %llu~%llu (%llu bytes)\n", page_off, len, len);
549 req->r_mtime = inode->i_mtime;
550 err = ceph_osdc_start_request(osdc, req, true);
552 err = ceph_osdc_wait_request(osdc, req);
554 ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
555 req->r_end_latency, len, err);
557 ceph_osdc_put_request(req);
562 struct writeback_control tmp_wbc;
565 if (err == -ERESTARTSYS) {
566 /* killed by SIGKILL */
567 dout("writepage interrupted page %p\n", page);
568 redirty_page_for_writepage(wbc, page);
569 end_page_writeback(page);
572 if (err == -EBLOCKLISTED)
573 fsc->blocklisted = true;
574 dout("writepage setting page/mapping error %d %p\n",
576 mapping_set_error(&inode->i_data, err);
577 wbc->pages_skipped++;
579 dout("writepage cleaned page %p\n", page);
580 err = 0; /* vfs expects us to return 0 */
582 oldest = detach_page_private(page);
583 WARN_ON_ONCE(oldest != snapc);
584 end_page_writeback(page);
585 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
586 ceph_put_snap_context(snapc); /* page's reference */
588 if (atomic_long_dec_return(&fsc->writeback_count) <
589 CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb))
590 clear_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC);
595 static int ceph_writepage(struct page *page, struct writeback_control *wbc)
598 struct inode *inode = page->mapping->host;
601 err = writepage_nounlock(page, wbc);
602 if (err == -ERESTARTSYS) {
603 /* direct memory reclaimer was killed by SIGKILL. return 0
604 * to prevent caller from setting mapping/page error */
613 * async writeback completion handler.
615 * If we get an error, set the mapping error bit, but not the individual
618 static void writepages_finish(struct ceph_osd_request *req)
620 struct inode *inode = req->r_inode;
621 struct ceph_inode_info *ci = ceph_inode(inode);
622 struct ceph_osd_data *osd_data;
624 int num_pages, total_pages = 0;
626 int rc = req->r_result;
627 struct ceph_snap_context *snapc = req->r_snapc;
628 struct address_space *mapping = inode->i_mapping;
629 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
630 unsigned int len = 0;
633 dout("writepages_finish %p rc %d\n", inode, rc);
635 mapping_set_error(mapping, rc);
636 ceph_set_error_write(ci);
637 if (rc == -EBLOCKLISTED)
638 fsc->blocklisted = true;
640 ceph_clear_error_write(ci);
644 * We lost the cache cap, need to truncate the page before
645 * it is unlocked, otherwise we'd truncate it later in the
646 * page truncation thread, possibly losing some data that
649 remove_page = !(ceph_caps_issued(ci) &
650 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO));
652 /* clean all pages */
653 for (i = 0; i < req->r_num_ops; i++) {
654 if (req->r_ops[i].op != CEPH_OSD_OP_WRITE)
657 osd_data = osd_req_op_extent_osd_data(req, i);
658 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
659 len += osd_data->length;
660 num_pages = calc_pages_for((u64)osd_data->alignment,
661 (u64)osd_data->length);
662 total_pages += num_pages;
663 for (j = 0; j < num_pages; j++) {
664 page = osd_data->pages[j];
666 WARN_ON(!PageUptodate(page));
668 if (atomic_long_dec_return(&fsc->writeback_count) <
669 CONGESTION_OFF_THRESH(
670 fsc->mount_options->congestion_kb))
671 clear_bdi_congested(inode_to_bdi(inode),
674 ceph_put_snap_context(detach_page_private(page));
675 end_page_writeback(page);
676 dout("unlocking %p\n", page);
679 generic_error_remove_page(inode->i_mapping,
684 dout("writepages_finish %p wrote %llu bytes cleaned %d pages\n",
685 inode, osd_data->length, rc >= 0 ? num_pages : 0);
687 release_pages(osd_data->pages, num_pages);
690 ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
691 req->r_end_latency, len, rc);
693 ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc);
695 osd_data = osd_req_op_extent_osd_data(req, 0);
696 if (osd_data->pages_from_pool)
697 mempool_free(osd_data->pages, ceph_wb_pagevec_pool);
699 kfree(osd_data->pages);
700 ceph_osdc_put_request(req);
704 * initiate async writeback
706 static int ceph_writepages_start(struct address_space *mapping,
707 struct writeback_control *wbc)
709 struct inode *inode = mapping->host;
710 struct ceph_inode_info *ci = ceph_inode(inode);
711 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
712 struct ceph_vino vino = ceph_vino(inode);
713 pgoff_t index, start_index, end = -1;
714 struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc;
717 unsigned int wsize = i_blocksize(inode);
718 struct ceph_osd_request *req = NULL;
719 struct ceph_writeback_ctl ceph_wbc;
720 bool should_loop, range_whole = false;
723 dout("writepages_start %p (mode=%s)\n", inode,
724 wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
725 (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
727 if (READ_ONCE(fsc->mount_state) >= CEPH_MOUNT_SHUTDOWN) {
728 if (ci->i_wrbuffer_ref > 0) {
730 "writepage_start %p %lld forced umount\n",
731 inode, ceph_ino(inode));
733 mapping_set_error(mapping, -EIO);
734 return -EIO; /* we're in a forced umount, don't write! */
736 if (fsc->mount_options->wsize < wsize)
737 wsize = fsc->mount_options->wsize;
741 start_index = wbc->range_cyclic ? mapping->writeback_index : 0;
745 /* find oldest snap context with dirty data */
746 snapc = get_oldest_context(inode, &ceph_wbc, NULL);
748 /* hmm, why does writepages get called when there
750 dout(" no snap context with dirty data?\n");
753 dout(" oldest snapc is %p seq %lld (%d snaps)\n",
754 snapc, snapc->seq, snapc->num_snaps);
757 if (ceph_wbc.head_snapc && snapc != last_snapc) {
758 /* where to start/end? */
759 if (wbc->range_cyclic) {
764 dout(" cyclic, start at %lu\n", index);
766 index = wbc->range_start >> PAGE_SHIFT;
767 end = wbc->range_end >> PAGE_SHIFT;
768 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
770 dout(" not cyclic, %lu to %lu\n", index, end);
772 } else if (!ceph_wbc.head_snapc) {
773 /* Do not respect wbc->range_{start,end}. Dirty pages
774 * in that range can be associated with newer snapc.
775 * They are not writeable until we write all dirty pages
776 * associated with 'snapc' get written */
779 dout(" non-head snapc, range whole\n");
782 ceph_put_snap_context(last_snapc);
785 while (!done && index <= end) {
786 int num_ops = 0, op_idx;
787 unsigned i, pvec_pages, max_pages, locked_pages = 0;
788 struct page **pages = NULL, **data_pages;
790 pgoff_t strip_unit_end = 0;
791 u64 offset = 0, len = 0;
792 bool from_pool = false;
794 max_pages = wsize >> PAGE_SHIFT;
797 pvec_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
798 end, PAGECACHE_TAG_DIRTY);
799 dout("pagevec_lookup_range_tag got %d\n", pvec_pages);
800 if (!pvec_pages && !locked_pages)
802 for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) {
803 page = pvec.pages[i];
804 dout("? %p idx %lu\n", page, page->index);
805 if (locked_pages == 0)
806 lock_page(page); /* first page */
807 else if (!trylock_page(page))
810 /* only dirty pages, or our accounting breaks */
811 if (unlikely(!PageDirty(page)) ||
812 unlikely(page->mapping != mapping)) {
813 dout("!dirty or !mapping %p\n", page);
817 /* only if matching snap context */
818 pgsnapc = page_snap_context(page);
819 if (pgsnapc != snapc) {
820 dout("page snapc %p %lld != oldest %p %lld\n",
821 pgsnapc, pgsnapc->seq, snapc, snapc->seq);
823 !ceph_wbc.head_snapc &&
824 wbc->sync_mode != WB_SYNC_NONE)
829 if (page_offset(page) >= ceph_wbc.i_size) {
830 dout("%p page eof %llu\n",
831 page, ceph_wbc.i_size);
832 if ((ceph_wbc.size_stable ||
833 page_offset(page) >= i_size_read(inode)) &&
834 clear_page_dirty_for_io(page))
835 mapping->a_ops->invalidatepage(page,
840 if (strip_unit_end && (page->index > strip_unit_end)) {
841 dout("end of strip unit %p\n", page);
845 if (PageWriteback(page)) {
846 if (wbc->sync_mode == WB_SYNC_NONE) {
847 dout("%p under writeback\n", page);
851 dout("waiting on writeback %p\n", page);
852 wait_on_page_writeback(page);
855 if (!clear_page_dirty_for_io(page)) {
856 dout("%p !clear_page_dirty_for_io\n", page);
862 * We have something to write. If this is
863 * the first locked page this time through,
864 * calculate max possinle write size and
865 * allocate a page array
867 if (locked_pages == 0) {
872 /* prepare async write request */
873 offset = (u64)page_offset(page);
874 ceph_calc_file_object_mapping(&ci->i_layout,
881 strip_unit_end = page->index +
882 ((len - 1) >> PAGE_SHIFT);
885 max_pages = calc_pages_for(0, (u64)len);
886 pages = kmalloc_array(max_pages,
891 pages = mempool_alloc(ceph_wb_pagevec_pool, GFP_NOFS);
896 } else if (page->index !=
897 (offset + len) >> PAGE_SHIFT) {
898 if (num_ops >= (from_pool ? CEPH_OSD_SLAB_OPS :
900 redirty_page_for_writepage(wbc, page);
906 offset = (u64)page_offset(page);
910 /* note position of first page in pvec */
911 dout("%p will write page %p idx %lu\n",
912 inode, page, page->index);
914 if (atomic_long_inc_return(&fsc->writeback_count) >
915 CONGESTION_ON_THRESH(
916 fsc->mount_options->congestion_kb)) {
917 set_bdi_congested(inode_to_bdi(inode),
922 pages[locked_pages++] = page;
923 pvec.pages[i] = NULL;
925 len += thp_size(page);
928 /* did we get anything? */
930 goto release_pvec_pages;
933 /* shift unused page to beginning of pvec */
934 for (j = 0; j < pvec_pages; j++) {
938 pvec.pages[n] = pvec.pages[j];
943 if (pvec_pages && i == pvec_pages &&
944 locked_pages < max_pages) {
945 dout("reached end pvec, trying for more\n");
946 pagevec_release(&pvec);
952 offset = page_offset(pages[0]);
955 req = ceph_osdc_new_request(&fsc->client->osdc,
957 offset, &len, 0, num_ops,
958 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
959 snapc, ceph_wbc.truncate_seq,
960 ceph_wbc.truncate_size, false);
962 req = ceph_osdc_new_request(&fsc->client->osdc,
969 snapc, ceph_wbc.truncate_seq,
970 ceph_wbc.truncate_size, true);
973 BUG_ON(len < page_offset(pages[locked_pages - 1]) +
974 thp_size(page) - offset);
976 req->r_callback = writepages_finish;
977 req->r_inode = inode;
979 /* Format the osd request message and submit the write */
983 for (i = 0; i < locked_pages; i++) {
984 u64 cur_offset = page_offset(pages[i]);
985 if (offset + len != cur_offset) {
986 if (op_idx + 1 == req->r_num_ops)
988 osd_req_op_extent_dup_last(req, op_idx,
989 cur_offset - offset);
990 dout("writepages got pages at %llu~%llu\n",
992 osd_req_op_extent_osd_data_pages(req, op_idx,
995 osd_req_op_extent_update(req, op_idx, len);
999 data_pages = pages + i;
1003 set_page_writeback(pages[i]);
1004 len += thp_size(page);
1007 if (ceph_wbc.size_stable) {
1008 len = min(len, ceph_wbc.i_size - offset);
1009 } else if (i == locked_pages) {
1010 /* writepages_finish() clears writeback pages
1011 * according to the data length, so make sure
1012 * data length covers all locked pages */
1013 u64 min_len = len + 1 - thp_size(page);
1014 len = get_writepages_data_length(inode, pages[i - 1],
1016 len = max(len, min_len);
1018 dout("writepages got pages at %llu~%llu\n", offset, len);
1020 osd_req_op_extent_osd_data_pages(req, op_idx, data_pages, len,
1021 0, from_pool, false);
1022 osd_req_op_extent_update(req, op_idx, len);
1024 BUG_ON(op_idx + 1 != req->r_num_ops);
1027 if (i < locked_pages) {
1028 BUG_ON(num_ops <= req->r_num_ops);
1029 num_ops -= req->r_num_ops;
1032 /* allocate new pages array for next request */
1034 pages = kmalloc_array(locked_pages, sizeof(*pages),
1038 pages = mempool_alloc(ceph_wb_pagevec_pool, GFP_NOFS);
1041 memcpy(pages, data_pages + i,
1042 locked_pages * sizeof(*pages));
1043 memset(data_pages + i, 0,
1044 locked_pages * sizeof(*pages));
1046 BUG_ON(num_ops != req->r_num_ops);
1047 index = pages[i - 1]->index + 1;
1048 /* request message now owns the pages array */
1052 req->r_mtime = inode->i_mtime;
1053 rc = ceph_osdc_start_request(&fsc->client->osdc, req, true);
1057 wbc->nr_to_write -= i;
1062 * We stop writing back only if we are not doing
1063 * integrity sync. In case of integrity sync we have to
1064 * keep going until we have written all the pages
1065 * we tagged for writeback prior to entering this loop.
1067 if (wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE)
1071 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr,
1072 pvec.nr ? pvec.pages[0] : NULL);
1073 pagevec_release(&pvec);
1076 if (should_loop && !done) {
1077 /* more to do; loop back to beginning of file */
1078 dout("writepages looping back to beginning of file\n");
1079 end = start_index - 1; /* OK even when start_index == 0 */
1081 /* to write dirty pages associated with next snapc,
1082 * we need to wait until current writes complete */
1083 if (wbc->sync_mode != WB_SYNC_NONE &&
1084 start_index == 0 && /* all dirty pages were checked */
1085 !ceph_wbc.head_snapc) {
1089 while ((index <= end) &&
1090 (nr = pagevec_lookup_tag(&pvec, mapping, &index,
1091 PAGECACHE_TAG_WRITEBACK))) {
1092 for (i = 0; i < nr; i++) {
1093 page = pvec.pages[i];
1094 if (page_snap_context(page) != snapc)
1096 wait_on_page_writeback(page);
1098 pagevec_release(&pvec);
1108 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1109 mapping->writeback_index = index;
1112 ceph_osdc_put_request(req);
1113 ceph_put_snap_context(last_snapc);
1114 dout("writepages dend - startone, rc = %d\n", rc);
1121 * See if a given @snapc is either writeable, or already written.
1123 static int context_is_writeable_or_written(struct inode *inode,
1124 struct ceph_snap_context *snapc)
1126 struct ceph_snap_context *oldest = get_oldest_context(inode, NULL, NULL);
1127 int ret = !oldest || snapc->seq <= oldest->seq;
1129 ceph_put_snap_context(oldest);
1134 * ceph_find_incompatible - find an incompatible context and return it
1135 * @page: page being dirtied
1137 * We are only allowed to write into/dirty a page if the page is
1138 * clean, or already dirty within the same snap context. Returns a
1139 * conflicting context if there is one, NULL if there isn't, or a
1140 * negative error code on other errors.
1142 * Must be called with page lock held.
1144 static struct ceph_snap_context *
1145 ceph_find_incompatible(struct page *page)
1147 struct inode *inode = page->mapping->host;
1148 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1149 struct ceph_inode_info *ci = ceph_inode(inode);
1151 if (READ_ONCE(fsc->mount_state) >= CEPH_MOUNT_SHUTDOWN) {
1152 dout(" page %p forced umount\n", page);
1153 return ERR_PTR(-EIO);
1157 struct ceph_snap_context *snapc, *oldest;
1159 wait_on_page_writeback(page);
1161 snapc = page_snap_context(page);
1162 if (!snapc || snapc == ci->i_head_snapc)
1166 * this page is already dirty in another (older) snap
1167 * context! is it writeable now?
1169 oldest = get_oldest_context(inode, NULL, NULL);
1170 if (snapc->seq > oldest->seq) {
1171 /* not writeable -- return it for the caller to deal with */
1172 ceph_put_snap_context(oldest);
1173 dout(" page %p snapc %p not current or oldest\n", page, snapc);
1174 return ceph_get_snap_context(snapc);
1176 ceph_put_snap_context(oldest);
1178 /* yay, writeable, do it now (without dropping page lock) */
1179 dout(" page %p snapc %p not current, but oldest\n", page, snapc);
1180 if (clear_page_dirty_for_io(page)) {
1181 int r = writepage_nounlock(page, NULL);
1189 static int ceph_netfs_check_write_begin(struct file *file, loff_t pos, unsigned int len,
1190 struct page *page, void **_fsdata)
1192 struct inode *inode = file_inode(file);
1193 struct ceph_inode_info *ci = ceph_inode(inode);
1194 struct ceph_snap_context *snapc;
1196 snapc = ceph_find_incompatible(page);
1203 return PTR_ERR(snapc);
1205 ceph_queue_writeback(inode);
1206 r = wait_event_killable(ci->i_cap_wq,
1207 context_is_writeable_or_written(inode, snapc));
1208 ceph_put_snap_context(snapc);
1209 return r == 0 ? -EAGAIN : r;
1215 * We are only allowed to write into/dirty the page if the page is
1216 * clean, or already dirty within the same snap context.
1218 static int ceph_write_begin(struct file *file, struct address_space *mapping,
1219 loff_t pos, unsigned len, unsigned flags,
1220 struct page **pagep, void **fsdata)
1222 struct inode *inode = file_inode(file);
1223 struct ceph_inode_info *ci = ceph_inode(inode);
1224 struct page *page = NULL;
1225 pgoff_t index = pos >> PAGE_SHIFT;
1229 * Uninlining should have already been done and everything updated, EXCEPT
1230 * for inline_version sent to the MDS.
1232 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1233 page = grab_cache_page_write_begin(mapping, index, flags);
1238 * The inline_version on a new inode is set to 1. If that's the
1239 * case, then the page is brand new and isn't yet Uptodate.
1242 if (index == 0 && ci->i_inline_version != 1) {
1243 if (!PageUptodate(page)) {
1244 WARN_ONCE(1, "ceph: write_begin called on still-inlined inode (inline_version %llu)!\n",
1245 ci->i_inline_version);
1250 zero_user_segment(page, 0, thp_size(page));
1251 SetPageUptodate(page);
1255 r = netfs_write_begin(file, inode->i_mapping, pos, len, 0, &page, NULL,
1256 &ceph_netfs_read_ops, NULL);
1259 wait_on_page_fscache(page);
1264 WARN_ON_ONCE(!PageLocked(page));
1271 * we don't do anything in here that simple_write_end doesn't do
1272 * except adjust dirty page accounting
1274 static int ceph_write_end(struct file *file, struct address_space *mapping,
1275 loff_t pos, unsigned len, unsigned copied,
1276 struct page *page, void *fsdata)
1278 struct inode *inode = file_inode(file);
1279 bool check_cap = false;
1281 dout("write_end file %p inode %p page %p %d~%d (%d)\n", file,
1282 inode, page, (int)pos, (int)copied, (int)len);
1284 if (!PageUptodate(page)) {
1285 /* just return that nothing was copied on a short copy */
1290 SetPageUptodate(page);
1293 /* did file size increase? */
1294 if (pos+copied > i_size_read(inode))
1295 check_cap = ceph_inode_set_size(inode, pos+copied);
1297 set_page_dirty(page);
1304 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL);
1310 * we set .direct_IO to indicate direct io is supported, but since we
1311 * intercept O_DIRECT reads and writes early, this function should
1314 static ssize_t ceph_direct_io(struct kiocb *iocb, struct iov_iter *iter)
1320 const struct address_space_operations ceph_aops = {
1321 .readpage = ceph_readpage,
1322 .readahead = ceph_readahead,
1323 .writepage = ceph_writepage,
1324 .writepages = ceph_writepages_start,
1325 .write_begin = ceph_write_begin,
1326 .write_end = ceph_write_end,
1327 .set_page_dirty = ceph_set_page_dirty,
1328 .invalidatepage = ceph_invalidatepage,
1329 .releasepage = ceph_releasepage,
1330 .direct_IO = ceph_direct_io,
1333 static void ceph_block_sigs(sigset_t *oldset)
1336 siginitsetinv(&mask, sigmask(SIGKILL));
1337 sigprocmask(SIG_BLOCK, &mask, oldset);
1340 static void ceph_restore_sigs(sigset_t *oldset)
1342 sigprocmask(SIG_SETMASK, oldset, NULL);
1348 static vm_fault_t ceph_filemap_fault(struct vm_fault *vmf)
1350 struct vm_area_struct *vma = vmf->vma;
1351 struct inode *inode = file_inode(vma->vm_file);
1352 struct ceph_inode_info *ci = ceph_inode(inode);
1353 struct ceph_file_info *fi = vma->vm_file->private_data;
1354 loff_t off = (loff_t)vmf->pgoff << PAGE_SHIFT;
1357 vm_fault_t ret = VM_FAULT_SIGBUS;
1359 ceph_block_sigs(&oldset);
1361 dout("filemap_fault %p %llx.%llx %llu trying to get caps\n",
1362 inode, ceph_vinop(inode), off);
1363 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1364 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1366 want = CEPH_CAP_FILE_CACHE;
1369 err = ceph_get_caps(vma->vm_file, CEPH_CAP_FILE_RD, want, -1, &got);
1373 dout("filemap_fault %p %llu got cap refs on %s\n",
1374 inode, off, ceph_cap_string(got));
1376 if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) ||
1377 ci->i_inline_version == CEPH_INLINE_NONE) {
1378 CEPH_DEFINE_RW_CONTEXT(rw_ctx, got);
1379 ceph_add_rw_context(fi, &rw_ctx);
1380 ret = filemap_fault(vmf);
1381 ceph_del_rw_context(fi, &rw_ctx);
1382 dout("filemap_fault %p %llu drop cap refs %s ret %x\n",
1383 inode, off, ceph_cap_string(got), ret);
1387 ceph_put_cap_refs(ci, got);
1392 /* read inline data */
1393 if (off >= PAGE_SIZE) {
1394 /* does not support inline data > PAGE_SIZE */
1395 ret = VM_FAULT_SIGBUS;
1397 struct address_space *mapping = inode->i_mapping;
1400 filemap_invalidate_lock_shared(mapping);
1401 page = find_or_create_page(mapping, 0,
1402 mapping_gfp_constraint(mapping, ~__GFP_FS));
1407 err = __ceph_do_getattr(inode, page,
1408 CEPH_STAT_CAP_INLINE_DATA, true);
1409 if (err < 0 || off >= i_size_read(inode)) {
1412 ret = vmf_error(err);
1415 if (err < PAGE_SIZE)
1416 zero_user_segment(page, err, PAGE_SIZE);
1418 flush_dcache_page(page);
1419 SetPageUptodate(page);
1421 ret = VM_FAULT_MAJOR | VM_FAULT_LOCKED;
1423 filemap_invalidate_unlock_shared(mapping);
1424 dout("filemap_fault %p %llu read inline data ret %x\n",
1428 ceph_restore_sigs(&oldset);
1430 ret = vmf_error(err);
1435 static vm_fault_t ceph_page_mkwrite(struct vm_fault *vmf)
1437 struct vm_area_struct *vma = vmf->vma;
1438 struct inode *inode = file_inode(vma->vm_file);
1439 struct ceph_inode_info *ci = ceph_inode(inode);
1440 struct ceph_file_info *fi = vma->vm_file->private_data;
1441 struct ceph_cap_flush *prealloc_cf;
1442 struct page *page = vmf->page;
1443 loff_t off = page_offset(page);
1444 loff_t size = i_size_read(inode);
1448 vm_fault_t ret = VM_FAULT_SIGBUS;
1450 prealloc_cf = ceph_alloc_cap_flush();
1452 return VM_FAULT_OOM;
1454 sb_start_pagefault(inode->i_sb);
1455 ceph_block_sigs(&oldset);
1457 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1458 struct page *locked_page = NULL;
1463 err = ceph_uninline_data(vma->vm_file, locked_page);
1465 unlock_page(locked_page);
1470 if (off + thp_size(page) <= size)
1471 len = thp_size(page);
1473 len = offset_in_thp(page, size);
1475 dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n",
1476 inode, ceph_vinop(inode), off, len, size);
1477 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1478 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1480 want = CEPH_CAP_FILE_BUFFER;
1483 err = ceph_get_caps(vma->vm_file, CEPH_CAP_FILE_WR, want, off + len, &got);
1487 dout("page_mkwrite %p %llu~%zd got cap refs on %s\n",
1488 inode, off, len, ceph_cap_string(got));
1490 /* Update time before taking page lock */
1491 file_update_time(vma->vm_file);
1492 inode_inc_iversion_raw(inode);
1495 struct ceph_snap_context *snapc;
1499 if (page_mkwrite_check_truncate(page, inode) < 0) {
1501 ret = VM_FAULT_NOPAGE;
1505 snapc = ceph_find_incompatible(page);
1507 /* success. we'll keep the page locked. */
1508 set_page_dirty(page);
1509 ret = VM_FAULT_LOCKED;
1515 if (IS_ERR(snapc)) {
1516 ret = VM_FAULT_SIGBUS;
1520 ceph_queue_writeback(inode);
1521 err = wait_event_killable(ci->i_cap_wq,
1522 context_is_writeable_or_written(inode, snapc));
1523 ceph_put_snap_context(snapc);
1526 if (ret == VM_FAULT_LOCKED ||
1527 ci->i_inline_version != CEPH_INLINE_NONE) {
1529 spin_lock(&ci->i_ceph_lock);
1530 ci->i_inline_version = CEPH_INLINE_NONE;
1531 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1533 spin_unlock(&ci->i_ceph_lock);
1535 __mark_inode_dirty(inode, dirty);
1538 dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %x\n",
1539 inode, off, len, ceph_cap_string(got), ret);
1540 ceph_put_cap_refs_async(ci, got);
1542 ceph_restore_sigs(&oldset);
1543 sb_end_pagefault(inode->i_sb);
1544 ceph_free_cap_flush(prealloc_cf);
1546 ret = vmf_error(err);
1550 void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
1551 char *data, size_t len)
1553 struct address_space *mapping = inode->i_mapping;
1559 if (i_size_read(inode) == 0)
1561 page = find_or_create_page(mapping, 0,
1562 mapping_gfp_constraint(mapping,
1566 if (PageUptodate(page)) {
1573 dout("fill_inline_data %p %llx.%llx len %zu locked_page %p\n",
1574 inode, ceph_vinop(inode), len, locked_page);
1577 void *kaddr = kmap_atomic(page);
1578 memcpy(kaddr, data, len);
1579 kunmap_atomic(kaddr);
1582 if (page != locked_page) {
1583 if (len < PAGE_SIZE)
1584 zero_user_segment(page, len, PAGE_SIZE);
1586 flush_dcache_page(page);
1588 SetPageUptodate(page);
1594 int ceph_uninline_data(struct file *filp, struct page *locked_page)
1596 struct inode *inode = file_inode(filp);
1597 struct ceph_inode_info *ci = ceph_inode(inode);
1598 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1599 struct ceph_osd_request *req;
1600 struct page *page = NULL;
1601 u64 len, inline_version;
1603 bool from_pagecache = false;
1605 spin_lock(&ci->i_ceph_lock);
1606 inline_version = ci->i_inline_version;
1607 spin_unlock(&ci->i_ceph_lock);
1609 dout("uninline_data %p %llx.%llx inline_version %llu\n",
1610 inode, ceph_vinop(inode), inline_version);
1612 if (inline_version == 1 || /* initial version, no data */
1613 inline_version == CEPH_INLINE_NONE)
1618 WARN_ON(!PageUptodate(page));
1619 } else if (ceph_caps_issued(ci) &
1620 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) {
1621 page = find_get_page(inode->i_mapping, 0);
1623 if (PageUptodate(page)) {
1624 from_pagecache = true;
1634 len = i_size_read(inode);
1635 if (len > PAGE_SIZE)
1638 page = __page_cache_alloc(GFP_NOFS);
1643 err = __ceph_do_getattr(inode, page,
1644 CEPH_STAT_CAP_INLINE_DATA, true);
1646 /* no inline data */
1647 if (err == -ENODATA)
1654 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1655 ceph_vino(inode), 0, &len, 0, 1,
1656 CEPH_OSD_OP_CREATE, CEPH_OSD_FLAG_WRITE,
1663 req->r_mtime = inode->i_mtime;
1664 err = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1666 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1667 ceph_osdc_put_request(req);
1671 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1672 ceph_vino(inode), 0, &len, 1, 3,
1673 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
1674 NULL, ci->i_truncate_seq,
1675 ci->i_truncate_size, false);
1681 osd_req_op_extent_osd_data_pages(req, 1, &page, len, 0, false, false);
1684 __le64 xattr_buf = cpu_to_le64(inline_version);
1685 err = osd_req_op_xattr_init(req, 0, CEPH_OSD_OP_CMPXATTR,
1686 "inline_version", &xattr_buf,
1688 CEPH_OSD_CMPXATTR_OP_GT,
1689 CEPH_OSD_CMPXATTR_MODE_U64);
1696 int xattr_len = snprintf(xattr_buf, sizeof(xattr_buf),
1697 "%llu", inline_version);
1698 err = osd_req_op_xattr_init(req, 2, CEPH_OSD_OP_SETXATTR,
1700 xattr_buf, xattr_len, 0, 0);
1705 req->r_mtime = inode->i_mtime;
1706 err = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1708 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1710 ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
1711 req->r_end_latency, len, err);
1714 ceph_osdc_put_request(req);
1715 if (err == -ECANCELED)
1718 if (page && page != locked_page) {
1719 if (from_pagecache) {
1723 __free_pages(page, 0);
1726 dout("uninline_data %p %llx.%llx inline_version %llu = %d\n",
1727 inode, ceph_vinop(inode), inline_version, err);
1731 static const struct vm_operations_struct ceph_vmops = {
1732 .fault = ceph_filemap_fault,
1733 .page_mkwrite = ceph_page_mkwrite,
1736 int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1738 struct address_space *mapping = file->f_mapping;
1740 if (!mapping->a_ops->readpage)
1742 file_accessed(file);
1743 vma->vm_ops = &ceph_vmops;
1752 static int __ceph_pool_perm_get(struct ceph_inode_info *ci,
1753 s64 pool, struct ceph_string *pool_ns)
1755 struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode);
1756 struct ceph_mds_client *mdsc = fsc->mdsc;
1757 struct ceph_osd_request *rd_req = NULL, *wr_req = NULL;
1758 struct rb_node **p, *parent;
1759 struct ceph_pool_perm *perm;
1760 struct page **pages;
1762 int err = 0, err2 = 0, have = 0;
1764 down_read(&mdsc->pool_perm_rwsem);
1765 p = &mdsc->pool_perm_tree.rb_node;
1767 perm = rb_entry(*p, struct ceph_pool_perm, node);
1768 if (pool < perm->pool)
1770 else if (pool > perm->pool)
1771 p = &(*p)->rb_right;
1773 int ret = ceph_compare_string(pool_ns,
1779 p = &(*p)->rb_right;
1786 up_read(&mdsc->pool_perm_rwsem);
1791 dout("__ceph_pool_perm_get pool %lld ns %.*s no perm cached\n",
1792 pool, (int)pool_ns->len, pool_ns->str);
1794 dout("__ceph_pool_perm_get pool %lld no perm cached\n", pool);
1796 down_write(&mdsc->pool_perm_rwsem);
1797 p = &mdsc->pool_perm_tree.rb_node;
1801 perm = rb_entry(parent, struct ceph_pool_perm, node);
1802 if (pool < perm->pool)
1804 else if (pool > perm->pool)
1805 p = &(*p)->rb_right;
1807 int ret = ceph_compare_string(pool_ns,
1813 p = &(*p)->rb_right;
1821 up_write(&mdsc->pool_perm_rwsem);
1825 rd_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1826 1, false, GFP_NOFS);
1832 rd_req->r_flags = CEPH_OSD_FLAG_READ;
1833 osd_req_op_init(rd_req, 0, CEPH_OSD_OP_STAT, 0);
1834 rd_req->r_base_oloc.pool = pool;
1836 rd_req->r_base_oloc.pool_ns = ceph_get_string(pool_ns);
1837 ceph_oid_printf(&rd_req->r_base_oid, "%llx.00000000", ci->i_vino.ino);
1839 err = ceph_osdc_alloc_messages(rd_req, GFP_NOFS);
1843 wr_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1844 1, false, GFP_NOFS);
1850 wr_req->r_flags = CEPH_OSD_FLAG_WRITE;
1851 osd_req_op_init(wr_req, 0, CEPH_OSD_OP_CREATE, CEPH_OSD_OP_FLAG_EXCL);
1852 ceph_oloc_copy(&wr_req->r_base_oloc, &rd_req->r_base_oloc);
1853 ceph_oid_copy(&wr_req->r_base_oid, &rd_req->r_base_oid);
1855 err = ceph_osdc_alloc_messages(wr_req, GFP_NOFS);
1859 /* one page should be large enough for STAT data */
1860 pages = ceph_alloc_page_vector(1, GFP_KERNEL);
1861 if (IS_ERR(pages)) {
1862 err = PTR_ERR(pages);
1866 osd_req_op_raw_data_in_pages(rd_req, 0, pages, PAGE_SIZE,
1868 err = ceph_osdc_start_request(&fsc->client->osdc, rd_req, false);
1870 wr_req->r_mtime = ci->vfs_inode.i_mtime;
1871 err2 = ceph_osdc_start_request(&fsc->client->osdc, wr_req, false);
1874 err = ceph_osdc_wait_request(&fsc->client->osdc, rd_req);
1876 err2 = ceph_osdc_wait_request(&fsc->client->osdc, wr_req);
1878 if (err >= 0 || err == -ENOENT)
1880 else if (err != -EPERM) {
1881 if (err == -EBLOCKLISTED)
1882 fsc->blocklisted = true;
1886 if (err2 == 0 || err2 == -EEXIST)
1888 else if (err2 != -EPERM) {
1889 if (err2 == -EBLOCKLISTED)
1890 fsc->blocklisted = true;
1895 pool_ns_len = pool_ns ? pool_ns->len : 0;
1896 perm = kmalloc(sizeof(*perm) + pool_ns_len + 1, GFP_NOFS);
1904 perm->pool_ns_len = pool_ns_len;
1905 if (pool_ns_len > 0)
1906 memcpy(perm->pool_ns, pool_ns->str, pool_ns_len);
1907 perm->pool_ns[pool_ns_len] = 0;
1909 rb_link_node(&perm->node, parent, p);
1910 rb_insert_color(&perm->node, &mdsc->pool_perm_tree);
1913 up_write(&mdsc->pool_perm_rwsem);
1915 ceph_osdc_put_request(rd_req);
1916 ceph_osdc_put_request(wr_req);
1921 dout("__ceph_pool_perm_get pool %lld ns %.*s result = %d\n",
1922 pool, (int)pool_ns->len, pool_ns->str, err);
1924 dout("__ceph_pool_perm_get pool %lld result = %d\n", pool, err);
1928 int ceph_pool_perm_check(struct inode *inode, int need)
1930 struct ceph_inode_info *ci = ceph_inode(inode);
1931 struct ceph_string *pool_ns;
1935 /* Only need to do this for regular files */
1936 if (!S_ISREG(inode->i_mode))
1939 if (ci->i_vino.snap != CEPH_NOSNAP) {
1941 * Pool permission check needs to write to the first object.
1942 * But for snapshot, head of the first object may have alread
1943 * been deleted. Skip check to avoid creating orphan object.
1948 if (ceph_test_mount_opt(ceph_inode_to_client(inode),
1952 spin_lock(&ci->i_ceph_lock);
1953 flags = ci->i_ceph_flags;
1954 pool = ci->i_layout.pool_id;
1955 spin_unlock(&ci->i_ceph_lock);
1957 if (flags & CEPH_I_POOL_PERM) {
1958 if ((need & CEPH_CAP_FILE_RD) && !(flags & CEPH_I_POOL_RD)) {
1959 dout("ceph_pool_perm_check pool %lld no read perm\n",
1963 if ((need & CEPH_CAP_FILE_WR) && !(flags & CEPH_I_POOL_WR)) {
1964 dout("ceph_pool_perm_check pool %lld no write perm\n",
1971 pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
1972 ret = __ceph_pool_perm_get(ci, pool, pool_ns);
1973 ceph_put_string(pool_ns);
1977 flags = CEPH_I_POOL_PERM;
1978 if (ret & POOL_READ)
1979 flags |= CEPH_I_POOL_RD;
1980 if (ret & POOL_WRITE)
1981 flags |= CEPH_I_POOL_WR;
1983 spin_lock(&ci->i_ceph_lock);
1984 if (pool == ci->i_layout.pool_id &&
1985 pool_ns == rcu_dereference_raw(ci->i_layout.pool_ns)) {
1986 ci->i_ceph_flags |= flags;
1988 pool = ci->i_layout.pool_id;
1989 flags = ci->i_ceph_flags;
1991 spin_unlock(&ci->i_ceph_lock);
1995 void ceph_pool_perm_destroy(struct ceph_mds_client *mdsc)
1997 struct ceph_pool_perm *perm;
2000 while (!RB_EMPTY_ROOT(&mdsc->pool_perm_tree)) {
2001 n = rb_first(&mdsc->pool_perm_tree);
2002 perm = rb_entry(n, struct ceph_pool_perm, node);
2003 rb_erase(n, &mdsc->pool_perm_tree);