1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
4 #include <linux/backing-dev.h>
7 #include <linux/swap.h>
8 #include <linux/pagemap.h>
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 folio **foliop, 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 bool ceph_dirty_folio(struct address_space *mapping, struct folio *folio)
82 struct ceph_inode_info *ci;
83 struct ceph_snap_context *snapc;
85 if (folio_test_dirty(folio)) {
86 dout("%p dirty_folio %p idx %lu -- already dirty\n",
87 mapping->host, folio, folio->index);
88 VM_BUG_ON_FOLIO(!folio_test_private(folio), folio);
92 inode = mapping->host;
93 ci = ceph_inode(inode);
96 spin_lock(&ci->i_ceph_lock);
97 BUG_ON(ci->i_wr_ref == 0); // caller should hold Fw reference
98 if (__ceph_have_pending_cap_snap(ci)) {
99 struct ceph_cap_snap *capsnap =
100 list_last_entry(&ci->i_cap_snaps,
101 struct ceph_cap_snap,
103 snapc = ceph_get_snap_context(capsnap->context);
104 capsnap->dirty_pages++;
106 BUG_ON(!ci->i_head_snapc);
107 snapc = ceph_get_snap_context(ci->i_head_snapc);
108 ++ci->i_wrbuffer_ref_head;
110 if (ci->i_wrbuffer_ref == 0)
112 ++ci->i_wrbuffer_ref;
113 dout("%p dirty_folio %p idx %lu head %d/%d -> %d/%d "
114 "snapc %p seq %lld (%d snaps)\n",
115 mapping->host, folio, folio->index,
116 ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1,
117 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
118 snapc, snapc->seq, snapc->num_snaps);
119 spin_unlock(&ci->i_ceph_lock);
122 * Reference snap context in folio->private. Also set
123 * PagePrivate so that we get invalidate_folio callback.
125 VM_WARN_ON_FOLIO(folio->private, folio);
126 folio_attach_private(folio, snapc);
128 return ceph_fscache_dirty_folio(mapping, folio);
132 * If we are truncating the full folio (i.e. offset == 0), adjust the
133 * dirty folio counters appropriately. Only called if there is private
136 static void ceph_invalidate_folio(struct folio *folio, size_t offset,
140 struct ceph_inode_info *ci;
141 struct ceph_snap_context *snapc;
143 inode = folio->mapping->host;
144 ci = ceph_inode(inode);
146 if (offset != 0 || length != folio_size(folio)) {
147 dout("%p invalidate_folio idx %lu partial dirty page %zu~%zu\n",
148 inode, folio->index, offset, length);
152 WARN_ON(!folio_test_locked(folio));
153 if (folio_test_private(folio)) {
154 dout("%p invalidate_folio idx %lu full dirty page\n",
155 inode, folio->index);
157 snapc = folio_detach_private(folio);
158 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
159 ceph_put_snap_context(snapc);
162 folio_wait_fscache(folio);
165 static bool ceph_release_folio(struct folio *folio, gfp_t gfp)
167 struct inode *inode = folio->mapping->host;
169 dout("%llx:%llx release_folio idx %lu (%sdirty)\n",
171 folio->index, folio_test_dirty(folio) ? "" : "not ");
173 if (folio_test_private(folio))
176 if (folio_test_fscache(folio)) {
177 if (current_is_kswapd() || !(gfp & __GFP_FS))
179 folio_wait_fscache(folio);
181 ceph_fscache_note_page_release(inode);
185 static void ceph_netfs_expand_readahead(struct netfs_io_request *rreq)
187 struct inode *inode = rreq->inode;
188 struct ceph_inode_info *ci = ceph_inode(inode);
189 struct ceph_file_layout *lo = &ci->i_layout;
193 /* Expand the start downward */
194 blockno = div_u64_rem(rreq->start, lo->stripe_unit, &blockoff);
195 rreq->start = blockno * lo->stripe_unit;
196 rreq->len += blockoff;
198 /* Now, round up the length to the next block */
199 rreq->len = roundup(rreq->len, lo->stripe_unit);
202 static bool ceph_netfs_clamp_length(struct netfs_io_subrequest *subreq)
204 struct inode *inode = subreq->rreq->inode;
205 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
206 struct ceph_inode_info *ci = ceph_inode(inode);
210 /* Truncate the extent at the end of the current block */
211 ceph_calc_file_object_mapping(&ci->i_layout, subreq->start, subreq->len,
212 &objno, &objoff, &xlen);
213 subreq->len = min(xlen, fsc->mount_options->rsize);
217 static void finish_netfs_read(struct ceph_osd_request *req)
219 struct ceph_fs_client *fsc = ceph_inode_to_client(req->r_inode);
220 struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0);
221 struct netfs_io_subrequest *subreq = req->r_priv;
223 int err = req->r_result;
225 ceph_update_read_metrics(&fsc->mdsc->metric, req->r_start_latency,
226 req->r_end_latency, osd_data->length, err);
228 dout("%s: result %d subreq->len=%zu i_size=%lld\n", __func__, req->r_result,
229 subreq->len, i_size_read(req->r_inode));
231 /* no object means success but no data */
234 else if (err == -EBLOCKLISTED)
235 fsc->blocklisted = true;
237 if (err >= 0 && err < subreq->len)
238 __set_bit(NETFS_SREQ_CLEAR_TAIL, &subreq->flags);
240 netfs_subreq_terminated(subreq, err, false);
242 num_pages = calc_pages_for(osd_data->alignment, osd_data->length);
243 ceph_put_page_vector(osd_data->pages, num_pages, false);
247 static bool ceph_netfs_issue_op_inline(struct netfs_io_subrequest *subreq)
249 struct netfs_io_request *rreq = subreq->rreq;
250 struct inode *inode = rreq->inode;
251 struct ceph_mds_reply_info_parsed *rinfo;
252 struct ceph_mds_reply_info_in *iinfo;
253 struct ceph_mds_request *req;
254 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
255 struct ceph_inode_info *ci = ceph_inode(inode);
256 struct iov_iter iter;
261 __set_bit(NETFS_SREQ_CLEAR_TAIL, &subreq->flags);
262 __clear_bit(NETFS_SREQ_COPY_TO_CACHE, &subreq->flags);
264 if (subreq->start >= inode->i_size)
267 /* We need to fetch the inline data. */
268 mode = ceph_try_to_choose_auth_mds(inode, CEPH_STAT_CAP_INLINE_DATA);
269 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
274 req->r_ino1 = ci->i_vino;
275 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INLINE_DATA);
278 err = ceph_mdsc_do_request(mdsc, NULL, req);
282 rinfo = &req->r_reply_info;
283 iinfo = &rinfo->targeti;
284 if (iinfo->inline_version == CEPH_INLINE_NONE) {
285 /* The data got uninlined */
286 ceph_mdsc_put_request(req);
290 len = min_t(size_t, iinfo->inline_len - subreq->start, subreq->len);
291 iov_iter_xarray(&iter, ITER_DEST, &rreq->mapping->i_pages, subreq->start, len);
292 err = copy_to_iter(iinfo->inline_data + subreq->start, len, &iter);
296 ceph_mdsc_put_request(req);
298 netfs_subreq_terminated(subreq, err, false);
302 static void ceph_netfs_issue_read(struct netfs_io_subrequest *subreq)
304 struct netfs_io_request *rreq = subreq->rreq;
305 struct inode *inode = rreq->inode;
306 struct ceph_inode_info *ci = ceph_inode(inode);
307 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
308 struct ceph_osd_request *req = NULL;
309 struct ceph_vino vino = ceph_vino(inode);
310 struct iov_iter iter;
314 u64 len = subreq->len;
316 if (ceph_inode_is_shutdown(inode)) {
321 if (ceph_has_inline_data(ci) && ceph_netfs_issue_op_inline(subreq))
324 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, vino, subreq->start, &len,
325 0, 1, CEPH_OSD_OP_READ,
326 CEPH_OSD_FLAG_READ | fsc->client->osdc.client->options->read_from_replica,
327 NULL, ci->i_truncate_seq, ci->i_truncate_size, false);
334 dout("%s: pos=%llu orig_len=%zu len=%llu\n", __func__, subreq->start, subreq->len, len);
335 iov_iter_xarray(&iter, ITER_DEST, &rreq->mapping->i_pages, subreq->start, len);
336 err = iov_iter_get_pages_alloc2(&iter, &pages, len, &page_off);
338 dout("%s: iov_ter_get_pages_alloc returned %d\n", __func__, err);
342 /* should always give us a page-aligned read */
343 WARN_ON_ONCE(page_off);
347 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false);
348 req->r_callback = finish_netfs_read;
349 req->r_priv = subreq;
350 req->r_inode = inode;
353 ceph_osdc_start_request(req->r_osdc, req);
355 ceph_osdc_put_request(req);
357 netfs_subreq_terminated(subreq, err, false);
358 dout("%s: result %d\n", __func__, err);
361 static int ceph_init_request(struct netfs_io_request *rreq, struct file *file)
363 struct inode *inode = rreq->inode;
364 int got = 0, want = CEPH_CAP_FILE_CACHE;
367 if (rreq->origin != NETFS_READAHEAD)
371 struct ceph_rw_context *rw_ctx;
372 struct ceph_file_info *fi = file->private_data;
374 rw_ctx = ceph_find_rw_context(fi);
380 * readahead callers do not necessarily hold Fcb caps
381 * (e.g. fadvise, madvise).
383 ret = ceph_try_get_caps(inode, CEPH_CAP_FILE_RD, want, true, &got);
385 dout("start_read %p, error getting cap\n", inode);
390 dout("start_read %p, no cache cap\n", inode);
396 rreq->netfs_priv = (void *)(uintptr_t)got;
400 static void ceph_netfs_free_request(struct netfs_io_request *rreq)
402 struct ceph_inode_info *ci = ceph_inode(rreq->inode);
403 int got = (uintptr_t)rreq->netfs_priv;
406 ceph_put_cap_refs(ci, got);
409 const struct netfs_request_ops ceph_netfs_ops = {
410 .init_request = ceph_init_request,
411 .free_request = ceph_netfs_free_request,
412 .begin_cache_operation = ceph_begin_cache_operation,
413 .issue_read = ceph_netfs_issue_read,
414 .expand_readahead = ceph_netfs_expand_readahead,
415 .clamp_length = ceph_netfs_clamp_length,
416 .check_write_begin = ceph_netfs_check_write_begin,
419 #ifdef CONFIG_CEPH_FSCACHE
420 static void ceph_set_page_fscache(struct page *page)
422 set_page_fscache(page);
425 static void ceph_fscache_write_terminated(void *priv, ssize_t error, bool was_async)
427 struct inode *inode = priv;
429 if (IS_ERR_VALUE(error) && error != -ENOBUFS)
430 ceph_fscache_invalidate(inode, false);
433 static void ceph_fscache_write_to_cache(struct inode *inode, u64 off, u64 len, bool caching)
435 struct ceph_inode_info *ci = ceph_inode(inode);
436 struct fscache_cookie *cookie = ceph_fscache_cookie(ci);
438 fscache_write_to_cache(cookie, inode->i_mapping, off, len, i_size_read(inode),
439 ceph_fscache_write_terminated, inode, caching);
442 static inline void ceph_set_page_fscache(struct page *page)
446 static inline void ceph_fscache_write_to_cache(struct inode *inode, u64 off, u64 len, bool caching)
449 #endif /* CONFIG_CEPH_FSCACHE */
451 struct ceph_writeback_ctl
461 * Get ref for the oldest snapc for an inode with dirty data... that is, the
462 * only snap context we are allowed to write back.
464 static struct ceph_snap_context *
465 get_oldest_context(struct inode *inode, struct ceph_writeback_ctl *ctl,
466 struct ceph_snap_context *page_snapc)
468 struct ceph_inode_info *ci = ceph_inode(inode);
469 struct ceph_snap_context *snapc = NULL;
470 struct ceph_cap_snap *capsnap = NULL;
472 spin_lock(&ci->i_ceph_lock);
473 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
474 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap,
475 capsnap->context, capsnap->dirty_pages);
476 if (!capsnap->dirty_pages)
479 /* get i_size, truncate_{seq,size} for page_snapc? */
480 if (snapc && capsnap->context != page_snapc)
484 if (capsnap->writing) {
485 ctl->i_size = i_size_read(inode);
486 ctl->size_stable = false;
488 ctl->i_size = capsnap->size;
489 ctl->size_stable = true;
491 ctl->truncate_size = capsnap->truncate_size;
492 ctl->truncate_seq = capsnap->truncate_seq;
493 ctl->head_snapc = false;
499 snapc = ceph_get_snap_context(capsnap->context);
501 page_snapc == snapc ||
502 page_snapc->seq > snapc->seq)
505 if (!snapc && ci->i_wrbuffer_ref_head) {
506 snapc = ceph_get_snap_context(ci->i_head_snapc);
507 dout(" head snapc %p has %d dirty pages\n",
508 snapc, ci->i_wrbuffer_ref_head);
510 ctl->i_size = i_size_read(inode);
511 ctl->truncate_size = ci->i_truncate_size;
512 ctl->truncate_seq = ci->i_truncate_seq;
513 ctl->size_stable = false;
514 ctl->head_snapc = true;
517 spin_unlock(&ci->i_ceph_lock);
521 static u64 get_writepages_data_length(struct inode *inode,
522 struct page *page, u64 start)
524 struct ceph_inode_info *ci = ceph_inode(inode);
525 struct ceph_snap_context *snapc = page_snap_context(page);
526 struct ceph_cap_snap *capsnap = NULL;
527 u64 end = i_size_read(inode);
529 if (snapc != ci->i_head_snapc) {
531 spin_lock(&ci->i_ceph_lock);
532 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
533 if (capsnap->context == snapc) {
534 if (!capsnap->writing)
540 spin_unlock(&ci->i_ceph_lock);
543 if (end > page_offset(page) + thp_size(page))
544 end = page_offset(page) + thp_size(page);
545 return end > start ? end - start : 0;
549 * Write a single page, but leave the page locked.
551 * If we get a write error, mark the mapping for error, but still adjust the
552 * dirty page accounting (i.e., page is no longer dirty).
554 static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
556 struct folio *folio = page_folio(page);
557 struct inode *inode = page->mapping->host;
558 struct ceph_inode_info *ci = ceph_inode(inode);
559 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
560 struct ceph_snap_context *snapc, *oldest;
561 loff_t page_off = page_offset(page);
563 loff_t len = thp_size(page);
564 struct ceph_writeback_ctl ceph_wbc;
565 struct ceph_osd_client *osdc = &fsc->client->osdc;
566 struct ceph_osd_request *req;
567 bool caching = ceph_is_cache_enabled(inode);
569 dout("writepage %p idx %lu\n", page, page->index);
571 if (ceph_inode_is_shutdown(inode))
574 /* verify this is a writeable snap context */
575 snapc = page_snap_context(page);
577 dout("writepage %p page %p not dirty?\n", inode, page);
580 oldest = get_oldest_context(inode, &ceph_wbc, snapc);
581 if (snapc->seq > oldest->seq) {
582 dout("writepage %p page %p snapc %p not writeable - noop\n",
584 /* we should only noop if called by kswapd */
585 WARN_ON(!(current->flags & PF_MEMALLOC));
586 ceph_put_snap_context(oldest);
587 redirty_page_for_writepage(wbc, page);
590 ceph_put_snap_context(oldest);
592 /* is this a partial page at end of file? */
593 if (page_off >= ceph_wbc.i_size) {
594 dout("folio at %lu beyond eof %llu\n", folio->index,
596 folio_invalidate(folio, 0, folio_size(folio));
600 if (ceph_wbc.i_size < page_off + len)
601 len = ceph_wbc.i_size - page_off;
603 dout("writepage %p page %p index %lu on %llu~%llu snapc %p seq %lld\n",
604 inode, page, page->index, page_off, len, snapc, snapc->seq);
606 if (atomic_long_inc_return(&fsc->writeback_count) >
607 CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb))
608 fsc->write_congested = true;
610 req = ceph_osdc_new_request(osdc, &ci->i_layout, ceph_vino(inode), page_off, &len, 0, 1,
611 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, snapc,
612 ceph_wbc.truncate_seq, ceph_wbc.truncate_size,
615 redirty_page_for_writepage(wbc, page);
619 set_page_writeback(page);
621 ceph_set_page_fscache(page);
622 ceph_fscache_write_to_cache(inode, page_off, len, caching);
624 /* it may be a short write due to an object boundary */
625 WARN_ON_ONCE(len > thp_size(page));
626 osd_req_op_extent_osd_data_pages(req, 0, &page, len, 0, false, false);
627 dout("writepage %llu~%llu (%llu bytes)\n", page_off, len, len);
629 req->r_mtime = inode->i_mtime;
630 ceph_osdc_start_request(osdc, req);
631 err = ceph_osdc_wait_request(osdc, req);
633 ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
634 req->r_end_latency, len, err);
636 ceph_osdc_put_request(req);
641 struct writeback_control tmp_wbc;
644 if (err == -ERESTARTSYS) {
645 /* killed by SIGKILL */
646 dout("writepage interrupted page %p\n", page);
647 redirty_page_for_writepage(wbc, page);
648 end_page_writeback(page);
651 if (err == -EBLOCKLISTED)
652 fsc->blocklisted = true;
653 dout("writepage setting page/mapping error %d %p\n",
655 mapping_set_error(&inode->i_data, err);
656 wbc->pages_skipped++;
658 dout("writepage cleaned page %p\n", page);
659 err = 0; /* vfs expects us to return 0 */
661 oldest = detach_page_private(page);
662 WARN_ON_ONCE(oldest != snapc);
663 end_page_writeback(page);
664 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
665 ceph_put_snap_context(snapc); /* page's reference */
667 if (atomic_long_dec_return(&fsc->writeback_count) <
668 CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb))
669 fsc->write_congested = false;
674 static int ceph_writepage(struct page *page, struct writeback_control *wbc)
677 struct inode *inode = page->mapping->host;
681 if (wbc->sync_mode == WB_SYNC_NONE &&
682 ceph_inode_to_client(inode)->write_congested)
683 return AOP_WRITEPAGE_ACTIVATE;
685 wait_on_page_fscache(page);
687 err = writepage_nounlock(page, wbc);
688 if (err == -ERESTARTSYS) {
689 /* direct memory reclaimer was killed by SIGKILL. return 0
690 * to prevent caller from setting mapping/page error */
699 * async writeback completion handler.
701 * If we get an error, set the mapping error bit, but not the individual
704 static void writepages_finish(struct ceph_osd_request *req)
706 struct inode *inode = req->r_inode;
707 struct ceph_inode_info *ci = ceph_inode(inode);
708 struct ceph_osd_data *osd_data;
710 int num_pages, total_pages = 0;
712 int rc = req->r_result;
713 struct ceph_snap_context *snapc = req->r_snapc;
714 struct address_space *mapping = inode->i_mapping;
715 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
716 unsigned int len = 0;
719 dout("writepages_finish %p rc %d\n", inode, rc);
721 mapping_set_error(mapping, rc);
722 ceph_set_error_write(ci);
723 if (rc == -EBLOCKLISTED)
724 fsc->blocklisted = true;
726 ceph_clear_error_write(ci);
730 * We lost the cache cap, need to truncate the page before
731 * it is unlocked, otherwise we'd truncate it later in the
732 * page truncation thread, possibly losing some data that
735 remove_page = !(ceph_caps_issued(ci) &
736 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO));
738 /* clean all pages */
739 for (i = 0; i < req->r_num_ops; i++) {
740 if (req->r_ops[i].op != CEPH_OSD_OP_WRITE) {
741 pr_warn("%s incorrect op %d req %p index %d tid %llu\n",
742 __func__, req->r_ops[i].op, req, i, req->r_tid);
746 osd_data = osd_req_op_extent_osd_data(req, i);
747 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
748 len += osd_data->length;
749 num_pages = calc_pages_for((u64)osd_data->alignment,
750 (u64)osd_data->length);
751 total_pages += num_pages;
752 for (j = 0; j < num_pages; j++) {
753 page = osd_data->pages[j];
755 WARN_ON(!PageUptodate(page));
757 if (atomic_long_dec_return(&fsc->writeback_count) <
758 CONGESTION_OFF_THRESH(
759 fsc->mount_options->congestion_kb))
760 fsc->write_congested = false;
762 ceph_put_snap_context(detach_page_private(page));
763 end_page_writeback(page);
764 dout("unlocking %p\n", page);
767 generic_error_remove_page(inode->i_mapping,
772 dout("writepages_finish %p wrote %llu bytes cleaned %d pages\n",
773 inode, osd_data->length, rc >= 0 ? num_pages : 0);
775 release_pages(osd_data->pages, num_pages);
778 ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
779 req->r_end_latency, len, rc);
781 ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc);
783 osd_data = osd_req_op_extent_osd_data(req, 0);
784 if (osd_data->pages_from_pool)
785 mempool_free(osd_data->pages, ceph_wb_pagevec_pool);
787 kfree(osd_data->pages);
788 ceph_osdc_put_request(req);
792 * initiate async writeback
794 static int ceph_writepages_start(struct address_space *mapping,
795 struct writeback_control *wbc)
797 struct inode *inode = mapping->host;
798 struct ceph_inode_info *ci = ceph_inode(inode);
799 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
800 struct ceph_vino vino = ceph_vino(inode);
801 pgoff_t index, start_index, end = -1;
802 struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc;
803 struct folio_batch fbatch;
805 unsigned int wsize = i_blocksize(inode);
806 struct ceph_osd_request *req = NULL;
807 struct ceph_writeback_ctl ceph_wbc;
808 bool should_loop, range_whole = false;
810 bool caching = ceph_is_cache_enabled(inode);
813 if (wbc->sync_mode == WB_SYNC_NONE &&
814 fsc->write_congested)
817 dout("writepages_start %p (mode=%s)\n", inode,
818 wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
819 (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
821 if (ceph_inode_is_shutdown(inode)) {
822 if (ci->i_wrbuffer_ref > 0) {
824 "writepage_start %p %lld forced umount\n",
825 inode, ceph_ino(inode));
827 mapping_set_error(mapping, -EIO);
828 return -EIO; /* we're in a forced umount, don't write! */
830 if (fsc->mount_options->wsize < wsize)
831 wsize = fsc->mount_options->wsize;
833 folio_batch_init(&fbatch);
835 start_index = wbc->range_cyclic ? mapping->writeback_index : 0;
838 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) {
839 tag = PAGECACHE_TAG_TOWRITE;
841 tag = PAGECACHE_TAG_DIRTY;
844 /* find oldest snap context with dirty data */
845 snapc = get_oldest_context(inode, &ceph_wbc, NULL);
847 /* hmm, why does writepages get called when there
849 dout(" no snap context with dirty data?\n");
852 dout(" oldest snapc is %p seq %lld (%d snaps)\n",
853 snapc, snapc->seq, snapc->num_snaps);
856 if (ceph_wbc.head_snapc && snapc != last_snapc) {
857 /* where to start/end? */
858 if (wbc->range_cyclic) {
863 dout(" cyclic, start at %lu\n", index);
865 index = wbc->range_start >> PAGE_SHIFT;
866 end = wbc->range_end >> PAGE_SHIFT;
867 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
869 dout(" not cyclic, %lu to %lu\n", index, end);
871 } else if (!ceph_wbc.head_snapc) {
872 /* Do not respect wbc->range_{start,end}. Dirty pages
873 * in that range can be associated with newer snapc.
874 * They are not writeable until we write all dirty pages
875 * associated with 'snapc' get written */
878 dout(" non-head snapc, range whole\n");
881 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
882 tag_pages_for_writeback(mapping, index, end);
884 ceph_put_snap_context(last_snapc);
887 while (!done && index <= end) {
888 int num_ops = 0, op_idx;
889 unsigned i, nr_folios, max_pages, locked_pages = 0;
890 struct page **pages = NULL, **data_pages;
892 pgoff_t strip_unit_end = 0;
893 u64 offset = 0, len = 0;
894 bool from_pool = false;
896 max_pages = wsize >> PAGE_SHIFT;
899 nr_folios = filemap_get_folios_tag(mapping, &index,
901 dout("pagevec_lookup_range_tag got %d\n", nr_folios);
902 if (!nr_folios && !locked_pages)
904 for (i = 0; i < nr_folios && locked_pages < max_pages; i++) {
905 page = &fbatch.folios[i]->page;
906 dout("? %p idx %lu\n", page, page->index);
907 if (locked_pages == 0)
908 lock_page(page); /* first page */
909 else if (!trylock_page(page))
912 /* only dirty pages, or our accounting breaks */
913 if (unlikely(!PageDirty(page)) ||
914 unlikely(page->mapping != mapping)) {
915 dout("!dirty or !mapping %p\n", page);
919 /* only if matching snap context */
920 pgsnapc = page_snap_context(page);
921 if (pgsnapc != snapc) {
922 dout("page snapc %p %lld != oldest %p %lld\n",
923 pgsnapc, pgsnapc->seq, snapc, snapc->seq);
925 !ceph_wbc.head_snapc &&
926 wbc->sync_mode != WB_SYNC_NONE)
931 if (page_offset(page) >= ceph_wbc.i_size) {
932 struct folio *folio = page_folio(page);
934 dout("folio at %lu beyond eof %llu\n",
935 folio->index, ceph_wbc.i_size);
936 if ((ceph_wbc.size_stable ||
937 folio_pos(folio) >= i_size_read(inode)) &&
938 folio_clear_dirty_for_io(folio))
939 folio_invalidate(folio, 0,
944 if (strip_unit_end && (page->index > strip_unit_end)) {
945 dout("end of strip unit %p\n", page);
949 if (PageWriteback(page) || PageFsCache(page)) {
950 if (wbc->sync_mode == WB_SYNC_NONE) {
951 dout("%p under writeback\n", page);
955 dout("waiting on writeback %p\n", page);
956 wait_on_page_writeback(page);
957 wait_on_page_fscache(page);
960 if (!clear_page_dirty_for_io(page)) {
961 dout("%p !clear_page_dirty_for_io\n", page);
967 * We have something to write. If this is
968 * the first locked page this time through,
969 * calculate max possinle write size and
970 * allocate a page array
972 if (locked_pages == 0) {
977 /* prepare async write request */
978 offset = (u64)page_offset(page);
979 ceph_calc_file_object_mapping(&ci->i_layout,
986 strip_unit_end = page->index +
987 ((len - 1) >> PAGE_SHIFT);
990 max_pages = calc_pages_for(0, (u64)len);
991 pages = kmalloc_array(max_pages,
996 pages = mempool_alloc(ceph_wb_pagevec_pool, GFP_NOFS);
1001 } else if (page->index !=
1002 (offset + len) >> PAGE_SHIFT) {
1003 if (num_ops >= (from_pool ? CEPH_OSD_SLAB_OPS :
1004 CEPH_OSD_MAX_OPS)) {
1005 redirty_page_for_writepage(wbc, page);
1011 offset = (u64)page_offset(page);
1015 /* note position of first page in fbatch */
1016 dout("%p will write page %p idx %lu\n",
1017 inode, page, page->index);
1019 if (atomic_long_inc_return(&fsc->writeback_count) >
1020 CONGESTION_ON_THRESH(
1021 fsc->mount_options->congestion_kb))
1022 fsc->write_congested = true;
1024 pages[locked_pages++] = page;
1025 fbatch.folios[i] = NULL;
1027 len += thp_size(page);
1030 /* did we get anything? */
1032 goto release_folios;
1035 /* shift unused page to beginning of fbatch */
1036 for (j = 0; j < nr_folios; j++) {
1037 if (!fbatch.folios[j])
1040 fbatch.folios[n] = fbatch.folios[j];
1045 if (nr_folios && i == nr_folios &&
1046 locked_pages < max_pages) {
1047 dout("reached end fbatch, trying for more\n");
1048 folio_batch_release(&fbatch);
1049 goto get_more_pages;
1054 offset = page_offset(pages[0]);
1057 req = ceph_osdc_new_request(&fsc->client->osdc,
1058 &ci->i_layout, vino,
1059 offset, &len, 0, num_ops,
1060 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
1061 snapc, ceph_wbc.truncate_seq,
1062 ceph_wbc.truncate_size, false);
1064 req = ceph_osdc_new_request(&fsc->client->osdc,
1065 &ci->i_layout, vino,
1070 CEPH_OSD_FLAG_WRITE,
1071 snapc, ceph_wbc.truncate_seq,
1072 ceph_wbc.truncate_size, true);
1073 BUG_ON(IS_ERR(req));
1075 BUG_ON(len < page_offset(pages[locked_pages - 1]) +
1076 thp_size(page) - offset);
1078 req->r_callback = writepages_finish;
1079 req->r_inode = inode;
1081 /* Format the osd request message and submit the write */
1085 for (i = 0; i < locked_pages; i++) {
1086 u64 cur_offset = page_offset(pages[i]);
1088 * Discontinuity in page range? Ceph can handle that by just passing
1089 * multiple extents in the write op.
1091 if (offset + len != cur_offset) {
1092 /* If it's full, stop here */
1093 if (op_idx + 1 == req->r_num_ops)
1096 /* Kick off an fscache write with what we have so far. */
1097 ceph_fscache_write_to_cache(inode, offset, len, caching);
1099 /* Start a new extent */
1100 osd_req_op_extent_dup_last(req, op_idx,
1101 cur_offset - offset);
1102 dout("writepages got pages at %llu~%llu\n",
1104 osd_req_op_extent_osd_data_pages(req, op_idx,
1107 osd_req_op_extent_update(req, op_idx, len);
1110 offset = cur_offset;
1111 data_pages = pages + i;
1115 set_page_writeback(pages[i]);
1117 ceph_set_page_fscache(pages[i]);
1118 len += thp_size(page);
1120 ceph_fscache_write_to_cache(inode, offset, len, caching);
1122 if (ceph_wbc.size_stable) {
1123 len = min(len, ceph_wbc.i_size - offset);
1124 } else if (i == locked_pages) {
1125 /* writepages_finish() clears writeback pages
1126 * according to the data length, so make sure
1127 * data length covers all locked pages */
1128 u64 min_len = len + 1 - thp_size(page);
1129 len = get_writepages_data_length(inode, pages[i - 1],
1131 len = max(len, min_len);
1133 dout("writepages got pages at %llu~%llu\n", offset, len);
1135 osd_req_op_extent_osd_data_pages(req, op_idx, data_pages, len,
1136 0, from_pool, false);
1137 osd_req_op_extent_update(req, op_idx, len);
1139 BUG_ON(op_idx + 1 != req->r_num_ops);
1142 if (i < locked_pages) {
1143 BUG_ON(num_ops <= req->r_num_ops);
1144 num_ops -= req->r_num_ops;
1147 /* allocate new pages array for next request */
1149 pages = kmalloc_array(locked_pages, sizeof(*pages),
1153 pages = mempool_alloc(ceph_wb_pagevec_pool, GFP_NOFS);
1156 memcpy(pages, data_pages + i,
1157 locked_pages * sizeof(*pages));
1158 memset(data_pages + i, 0,
1159 locked_pages * sizeof(*pages));
1161 BUG_ON(num_ops != req->r_num_ops);
1162 index = pages[i - 1]->index + 1;
1163 /* request message now owns the pages array */
1167 req->r_mtime = inode->i_mtime;
1168 ceph_osdc_start_request(&fsc->client->osdc, req);
1171 wbc->nr_to_write -= i;
1176 * We stop writing back only if we are not doing
1177 * integrity sync. In case of integrity sync we have to
1178 * keep going until we have written all the pages
1179 * we tagged for writeback prior to entering this loop.
1181 if (wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE)
1185 dout("folio_batch release on %d folios (%p)\n", (int)fbatch.nr,
1186 fbatch.nr ? fbatch.folios[0] : NULL);
1187 folio_batch_release(&fbatch);
1190 if (should_loop && !done) {
1191 /* more to do; loop back to beginning of file */
1192 dout("writepages looping back to beginning of file\n");
1193 end = start_index - 1; /* OK even when start_index == 0 */
1195 /* to write dirty pages associated with next snapc,
1196 * we need to wait until current writes complete */
1197 if (wbc->sync_mode != WB_SYNC_NONE &&
1198 start_index == 0 && /* all dirty pages were checked */
1199 !ceph_wbc.head_snapc) {
1203 while ((index <= end) &&
1204 (nr = filemap_get_folios_tag(mapping, &index,
1206 PAGECACHE_TAG_WRITEBACK,
1208 for (i = 0; i < nr; i++) {
1209 page = &fbatch.folios[i]->page;
1210 if (page_snap_context(page) != snapc)
1212 wait_on_page_writeback(page);
1214 folio_batch_release(&fbatch);
1224 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1225 mapping->writeback_index = index;
1228 ceph_osdc_put_request(req);
1229 ceph_put_snap_context(last_snapc);
1230 dout("writepages dend - startone, rc = %d\n", rc);
1237 * See if a given @snapc is either writeable, or already written.
1239 static int context_is_writeable_or_written(struct inode *inode,
1240 struct ceph_snap_context *snapc)
1242 struct ceph_snap_context *oldest = get_oldest_context(inode, NULL, NULL);
1243 int ret = !oldest || snapc->seq <= oldest->seq;
1245 ceph_put_snap_context(oldest);
1250 * ceph_find_incompatible - find an incompatible context and return it
1251 * @page: page being dirtied
1253 * We are only allowed to write into/dirty a page if the page is
1254 * clean, or already dirty within the same snap context. Returns a
1255 * conflicting context if there is one, NULL if there isn't, or a
1256 * negative error code on other errors.
1258 * Must be called with page lock held.
1260 static struct ceph_snap_context *
1261 ceph_find_incompatible(struct page *page)
1263 struct inode *inode = page->mapping->host;
1264 struct ceph_inode_info *ci = ceph_inode(inode);
1266 if (ceph_inode_is_shutdown(inode)) {
1267 dout(" page %p %llx:%llx is shutdown\n", page,
1269 return ERR_PTR(-ESTALE);
1273 struct ceph_snap_context *snapc, *oldest;
1275 wait_on_page_writeback(page);
1277 snapc = page_snap_context(page);
1278 if (!snapc || snapc == ci->i_head_snapc)
1282 * this page is already dirty in another (older) snap
1283 * context! is it writeable now?
1285 oldest = get_oldest_context(inode, NULL, NULL);
1286 if (snapc->seq > oldest->seq) {
1287 /* not writeable -- return it for the caller to deal with */
1288 ceph_put_snap_context(oldest);
1289 dout(" page %p snapc %p not current or oldest\n", page, snapc);
1290 return ceph_get_snap_context(snapc);
1292 ceph_put_snap_context(oldest);
1294 /* yay, writeable, do it now (without dropping page lock) */
1295 dout(" page %p snapc %p not current, but oldest\n", page, snapc);
1296 if (clear_page_dirty_for_io(page)) {
1297 int r = writepage_nounlock(page, NULL);
1305 static int ceph_netfs_check_write_begin(struct file *file, loff_t pos, unsigned int len,
1306 struct folio **foliop, void **_fsdata)
1308 struct inode *inode = file_inode(file);
1309 struct ceph_inode_info *ci = ceph_inode(inode);
1310 struct ceph_snap_context *snapc;
1312 snapc = ceph_find_incompatible(folio_page(*foliop, 0));
1316 folio_unlock(*foliop);
1320 return PTR_ERR(snapc);
1322 ceph_queue_writeback(inode);
1323 r = wait_event_killable(ci->i_cap_wq,
1324 context_is_writeable_or_written(inode, snapc));
1325 ceph_put_snap_context(snapc);
1326 return r == 0 ? -EAGAIN : r;
1332 * We are only allowed to write into/dirty the page if the page is
1333 * clean, or already dirty within the same snap context.
1335 static int ceph_write_begin(struct file *file, struct address_space *mapping,
1336 loff_t pos, unsigned len,
1337 struct page **pagep, void **fsdata)
1339 struct inode *inode = file_inode(file);
1340 struct ceph_inode_info *ci = ceph_inode(inode);
1341 struct folio *folio = NULL;
1344 r = netfs_write_begin(&ci->netfs, file, inode->i_mapping, pos, len, &folio, NULL);
1348 folio_wait_fscache(folio);
1349 WARN_ON_ONCE(!folio_test_locked(folio));
1350 *pagep = &folio->page;
1355 * we don't do anything in here that simple_write_end doesn't do
1356 * except adjust dirty page accounting
1358 static int ceph_write_end(struct file *file, struct address_space *mapping,
1359 loff_t pos, unsigned len, unsigned copied,
1360 struct page *subpage, void *fsdata)
1362 struct folio *folio = page_folio(subpage);
1363 struct inode *inode = file_inode(file);
1364 bool check_cap = false;
1366 dout("write_end file %p inode %p folio %p %d~%d (%d)\n", file,
1367 inode, folio, (int)pos, (int)copied, (int)len);
1369 if (!folio_test_uptodate(folio)) {
1370 /* just return that nothing was copied on a short copy */
1375 folio_mark_uptodate(folio);
1378 /* did file size increase? */
1379 if (pos+copied > i_size_read(inode))
1380 check_cap = ceph_inode_set_size(inode, pos+copied);
1382 folio_mark_dirty(folio);
1385 folio_unlock(folio);
1389 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY);
1394 const struct address_space_operations ceph_aops = {
1395 .read_folio = netfs_read_folio,
1396 .readahead = netfs_readahead,
1397 .writepage = ceph_writepage,
1398 .writepages = ceph_writepages_start,
1399 .write_begin = ceph_write_begin,
1400 .write_end = ceph_write_end,
1401 .dirty_folio = ceph_dirty_folio,
1402 .invalidate_folio = ceph_invalidate_folio,
1403 .release_folio = ceph_release_folio,
1404 .direct_IO = noop_direct_IO,
1407 static void ceph_block_sigs(sigset_t *oldset)
1410 siginitsetinv(&mask, sigmask(SIGKILL));
1411 sigprocmask(SIG_BLOCK, &mask, oldset);
1414 static void ceph_restore_sigs(sigset_t *oldset)
1416 sigprocmask(SIG_SETMASK, oldset, NULL);
1422 static vm_fault_t ceph_filemap_fault(struct vm_fault *vmf)
1424 struct vm_area_struct *vma = vmf->vma;
1425 struct inode *inode = file_inode(vma->vm_file);
1426 struct ceph_inode_info *ci = ceph_inode(inode);
1427 struct ceph_file_info *fi = vma->vm_file->private_data;
1428 loff_t off = (loff_t)vmf->pgoff << PAGE_SHIFT;
1431 vm_fault_t ret = VM_FAULT_SIGBUS;
1433 if (ceph_inode_is_shutdown(inode))
1436 ceph_block_sigs(&oldset);
1438 dout("filemap_fault %p %llx.%llx %llu trying to get caps\n",
1439 inode, ceph_vinop(inode), off);
1440 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1441 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1443 want = CEPH_CAP_FILE_CACHE;
1446 err = ceph_get_caps(vma->vm_file, CEPH_CAP_FILE_RD, want, -1, &got);
1450 dout("filemap_fault %p %llu got cap refs on %s\n",
1451 inode, off, ceph_cap_string(got));
1453 if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) ||
1454 !ceph_has_inline_data(ci)) {
1455 CEPH_DEFINE_RW_CONTEXT(rw_ctx, got);
1456 ceph_add_rw_context(fi, &rw_ctx);
1457 ret = filemap_fault(vmf);
1458 ceph_del_rw_context(fi, &rw_ctx);
1459 dout("filemap_fault %p %llu drop cap refs %s ret %x\n",
1460 inode, off, ceph_cap_string(got), ret);
1464 ceph_put_cap_refs(ci, got);
1469 /* read inline data */
1470 if (off >= PAGE_SIZE) {
1471 /* does not support inline data > PAGE_SIZE */
1472 ret = VM_FAULT_SIGBUS;
1474 struct address_space *mapping = inode->i_mapping;
1477 filemap_invalidate_lock_shared(mapping);
1478 page = find_or_create_page(mapping, 0,
1479 mapping_gfp_constraint(mapping, ~__GFP_FS));
1484 err = __ceph_do_getattr(inode, page,
1485 CEPH_STAT_CAP_INLINE_DATA, true);
1486 if (err < 0 || off >= i_size_read(inode)) {
1489 ret = vmf_error(err);
1492 if (err < PAGE_SIZE)
1493 zero_user_segment(page, err, PAGE_SIZE);
1495 flush_dcache_page(page);
1496 SetPageUptodate(page);
1498 ret = VM_FAULT_MAJOR | VM_FAULT_LOCKED;
1500 filemap_invalidate_unlock_shared(mapping);
1501 dout("filemap_fault %p %llu read inline data ret %x\n",
1505 ceph_restore_sigs(&oldset);
1507 ret = vmf_error(err);
1512 static vm_fault_t ceph_page_mkwrite(struct vm_fault *vmf)
1514 struct vm_area_struct *vma = vmf->vma;
1515 struct inode *inode = file_inode(vma->vm_file);
1516 struct ceph_inode_info *ci = ceph_inode(inode);
1517 struct ceph_file_info *fi = vma->vm_file->private_data;
1518 struct ceph_cap_flush *prealloc_cf;
1519 struct page *page = vmf->page;
1520 loff_t off = page_offset(page);
1521 loff_t size = i_size_read(inode);
1525 vm_fault_t ret = VM_FAULT_SIGBUS;
1527 if (ceph_inode_is_shutdown(inode))
1530 prealloc_cf = ceph_alloc_cap_flush();
1532 return VM_FAULT_OOM;
1534 sb_start_pagefault(inode->i_sb);
1535 ceph_block_sigs(&oldset);
1537 if (off + thp_size(page) <= size)
1538 len = thp_size(page);
1540 len = offset_in_thp(page, size);
1542 dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n",
1543 inode, ceph_vinop(inode), off, len, size);
1544 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1545 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1547 want = CEPH_CAP_FILE_BUFFER;
1550 err = ceph_get_caps(vma->vm_file, CEPH_CAP_FILE_WR, want, off + len, &got);
1554 dout("page_mkwrite %p %llu~%zd got cap refs on %s\n",
1555 inode, off, len, ceph_cap_string(got));
1557 /* Update time before taking page lock */
1558 file_update_time(vma->vm_file);
1559 inode_inc_iversion_raw(inode);
1562 struct ceph_snap_context *snapc;
1566 if (page_mkwrite_check_truncate(page, inode) < 0) {
1568 ret = VM_FAULT_NOPAGE;
1572 snapc = ceph_find_incompatible(page);
1574 /* success. we'll keep the page locked. */
1575 set_page_dirty(page);
1576 ret = VM_FAULT_LOCKED;
1582 if (IS_ERR(snapc)) {
1583 ret = VM_FAULT_SIGBUS;
1587 ceph_queue_writeback(inode);
1588 err = wait_event_killable(ci->i_cap_wq,
1589 context_is_writeable_or_written(inode, snapc));
1590 ceph_put_snap_context(snapc);
1593 if (ret == VM_FAULT_LOCKED) {
1595 spin_lock(&ci->i_ceph_lock);
1596 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1598 spin_unlock(&ci->i_ceph_lock);
1600 __mark_inode_dirty(inode, dirty);
1603 dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %x\n",
1604 inode, off, len, ceph_cap_string(got), ret);
1605 ceph_put_cap_refs_async(ci, got);
1607 ceph_restore_sigs(&oldset);
1608 sb_end_pagefault(inode->i_sb);
1609 ceph_free_cap_flush(prealloc_cf);
1611 ret = vmf_error(err);
1615 void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
1616 char *data, size_t len)
1618 struct address_space *mapping = inode->i_mapping;
1624 if (i_size_read(inode) == 0)
1626 page = find_or_create_page(mapping, 0,
1627 mapping_gfp_constraint(mapping,
1631 if (PageUptodate(page)) {
1638 dout("fill_inline_data %p %llx.%llx len %zu locked_page %p\n",
1639 inode, ceph_vinop(inode), len, locked_page);
1642 void *kaddr = kmap_atomic(page);
1643 memcpy(kaddr, data, len);
1644 kunmap_atomic(kaddr);
1647 if (page != locked_page) {
1648 if (len < PAGE_SIZE)
1649 zero_user_segment(page, len, PAGE_SIZE);
1651 flush_dcache_page(page);
1653 SetPageUptodate(page);
1659 int ceph_uninline_data(struct file *file)
1661 struct inode *inode = file_inode(file);
1662 struct ceph_inode_info *ci = ceph_inode(inode);
1663 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1664 struct ceph_osd_request *req = NULL;
1665 struct ceph_cap_flush *prealloc_cf = NULL;
1666 struct folio *folio = NULL;
1667 u64 inline_version = CEPH_INLINE_NONE;
1668 struct page *pages[1];
1672 spin_lock(&ci->i_ceph_lock);
1673 inline_version = ci->i_inline_version;
1674 spin_unlock(&ci->i_ceph_lock);
1676 dout("uninline_data %p %llx.%llx inline_version %llu\n",
1677 inode, ceph_vinop(inode), inline_version);
1679 if (ceph_inode_is_shutdown(inode)) {
1684 if (inline_version == CEPH_INLINE_NONE)
1687 prealloc_cf = ceph_alloc_cap_flush();
1691 if (inline_version == 1) /* initial version, no data */
1694 folio = read_mapping_folio(inode->i_mapping, 0, file);
1695 if (IS_ERR(folio)) {
1696 err = PTR_ERR(folio);
1702 len = i_size_read(inode);
1703 if (len > folio_size(folio))
1704 len = folio_size(folio);
1706 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1707 ceph_vino(inode), 0, &len, 0, 1,
1708 CEPH_OSD_OP_CREATE, CEPH_OSD_FLAG_WRITE,
1715 req->r_mtime = inode->i_mtime;
1716 ceph_osdc_start_request(&fsc->client->osdc, req);
1717 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1718 ceph_osdc_put_request(req);
1722 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1723 ceph_vino(inode), 0, &len, 1, 3,
1724 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
1725 NULL, ci->i_truncate_seq,
1726 ci->i_truncate_size, false);
1732 pages[0] = folio_page(folio, 0);
1733 osd_req_op_extent_osd_data_pages(req, 1, pages, len, 0, false, false);
1736 __le64 xattr_buf = cpu_to_le64(inline_version);
1737 err = osd_req_op_xattr_init(req, 0, CEPH_OSD_OP_CMPXATTR,
1738 "inline_version", &xattr_buf,
1740 CEPH_OSD_CMPXATTR_OP_GT,
1741 CEPH_OSD_CMPXATTR_MODE_U64);
1748 int xattr_len = snprintf(xattr_buf, sizeof(xattr_buf),
1749 "%llu", inline_version);
1750 err = osd_req_op_xattr_init(req, 2, CEPH_OSD_OP_SETXATTR,
1752 xattr_buf, xattr_len, 0, 0);
1757 req->r_mtime = inode->i_mtime;
1758 ceph_osdc_start_request(&fsc->client->osdc, req);
1759 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1761 ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
1762 req->r_end_latency, len, err);
1768 /* Set to CAP_INLINE_NONE and dirty the caps */
1769 down_read(&fsc->mdsc->snap_rwsem);
1770 spin_lock(&ci->i_ceph_lock);
1771 ci->i_inline_version = CEPH_INLINE_NONE;
1772 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, &prealloc_cf);
1773 spin_unlock(&ci->i_ceph_lock);
1774 up_read(&fsc->mdsc->snap_rwsem);
1776 __mark_inode_dirty(inode, dirty);
1779 ceph_osdc_put_request(req);
1780 if (err == -ECANCELED)
1784 folio_unlock(folio);
1788 ceph_free_cap_flush(prealloc_cf);
1789 dout("uninline_data %p %llx.%llx inline_version %llu = %d\n",
1790 inode, ceph_vinop(inode), inline_version, err);
1794 static const struct vm_operations_struct ceph_vmops = {
1795 .fault = ceph_filemap_fault,
1796 .page_mkwrite = ceph_page_mkwrite,
1799 int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1801 struct address_space *mapping = file->f_mapping;
1803 if (!mapping->a_ops->read_folio)
1805 vma->vm_ops = &ceph_vmops;
1814 static int __ceph_pool_perm_get(struct ceph_inode_info *ci,
1815 s64 pool, struct ceph_string *pool_ns)
1817 struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->netfs.inode);
1818 struct ceph_mds_client *mdsc = fsc->mdsc;
1819 struct ceph_osd_request *rd_req = NULL, *wr_req = NULL;
1820 struct rb_node **p, *parent;
1821 struct ceph_pool_perm *perm;
1822 struct page **pages;
1824 int err = 0, err2 = 0, have = 0;
1826 down_read(&mdsc->pool_perm_rwsem);
1827 p = &mdsc->pool_perm_tree.rb_node;
1829 perm = rb_entry(*p, struct ceph_pool_perm, node);
1830 if (pool < perm->pool)
1832 else if (pool > perm->pool)
1833 p = &(*p)->rb_right;
1835 int ret = ceph_compare_string(pool_ns,
1841 p = &(*p)->rb_right;
1848 up_read(&mdsc->pool_perm_rwsem);
1853 dout("__ceph_pool_perm_get pool %lld ns %.*s no perm cached\n",
1854 pool, (int)pool_ns->len, pool_ns->str);
1856 dout("__ceph_pool_perm_get pool %lld no perm cached\n", pool);
1858 down_write(&mdsc->pool_perm_rwsem);
1859 p = &mdsc->pool_perm_tree.rb_node;
1863 perm = rb_entry(parent, struct ceph_pool_perm, node);
1864 if (pool < perm->pool)
1866 else if (pool > perm->pool)
1867 p = &(*p)->rb_right;
1869 int ret = ceph_compare_string(pool_ns,
1875 p = &(*p)->rb_right;
1883 up_write(&mdsc->pool_perm_rwsem);
1887 rd_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1888 1, false, GFP_NOFS);
1894 rd_req->r_flags = CEPH_OSD_FLAG_READ;
1895 osd_req_op_init(rd_req, 0, CEPH_OSD_OP_STAT, 0);
1896 rd_req->r_base_oloc.pool = pool;
1898 rd_req->r_base_oloc.pool_ns = ceph_get_string(pool_ns);
1899 ceph_oid_printf(&rd_req->r_base_oid, "%llx.00000000", ci->i_vino.ino);
1901 err = ceph_osdc_alloc_messages(rd_req, GFP_NOFS);
1905 wr_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1906 1, false, GFP_NOFS);
1912 wr_req->r_flags = CEPH_OSD_FLAG_WRITE;
1913 osd_req_op_init(wr_req, 0, CEPH_OSD_OP_CREATE, CEPH_OSD_OP_FLAG_EXCL);
1914 ceph_oloc_copy(&wr_req->r_base_oloc, &rd_req->r_base_oloc);
1915 ceph_oid_copy(&wr_req->r_base_oid, &rd_req->r_base_oid);
1917 err = ceph_osdc_alloc_messages(wr_req, GFP_NOFS);
1921 /* one page should be large enough for STAT data */
1922 pages = ceph_alloc_page_vector(1, GFP_KERNEL);
1923 if (IS_ERR(pages)) {
1924 err = PTR_ERR(pages);
1928 osd_req_op_raw_data_in_pages(rd_req, 0, pages, PAGE_SIZE,
1930 ceph_osdc_start_request(&fsc->client->osdc, rd_req);
1932 wr_req->r_mtime = ci->netfs.inode.i_mtime;
1933 ceph_osdc_start_request(&fsc->client->osdc, wr_req);
1935 err = ceph_osdc_wait_request(&fsc->client->osdc, rd_req);
1936 err2 = ceph_osdc_wait_request(&fsc->client->osdc, wr_req);
1938 if (err >= 0 || err == -ENOENT)
1940 else if (err != -EPERM) {
1941 if (err == -EBLOCKLISTED)
1942 fsc->blocklisted = true;
1946 if (err2 == 0 || err2 == -EEXIST)
1948 else if (err2 != -EPERM) {
1949 if (err2 == -EBLOCKLISTED)
1950 fsc->blocklisted = true;
1955 pool_ns_len = pool_ns ? pool_ns->len : 0;
1956 perm = kmalloc(sizeof(*perm) + pool_ns_len + 1, GFP_NOFS);
1964 perm->pool_ns_len = pool_ns_len;
1965 if (pool_ns_len > 0)
1966 memcpy(perm->pool_ns, pool_ns->str, pool_ns_len);
1967 perm->pool_ns[pool_ns_len] = 0;
1969 rb_link_node(&perm->node, parent, p);
1970 rb_insert_color(&perm->node, &mdsc->pool_perm_tree);
1973 up_write(&mdsc->pool_perm_rwsem);
1975 ceph_osdc_put_request(rd_req);
1976 ceph_osdc_put_request(wr_req);
1981 dout("__ceph_pool_perm_get pool %lld ns %.*s result = %d\n",
1982 pool, (int)pool_ns->len, pool_ns->str, err);
1984 dout("__ceph_pool_perm_get pool %lld result = %d\n", pool, err);
1988 int ceph_pool_perm_check(struct inode *inode, int need)
1990 struct ceph_inode_info *ci = ceph_inode(inode);
1991 struct ceph_string *pool_ns;
1995 /* Only need to do this for regular files */
1996 if (!S_ISREG(inode->i_mode))
1999 if (ci->i_vino.snap != CEPH_NOSNAP) {
2001 * Pool permission check needs to write to the first object.
2002 * But for snapshot, head of the first object may have alread
2003 * been deleted. Skip check to avoid creating orphan object.
2008 if (ceph_test_mount_opt(ceph_inode_to_client(inode),
2012 spin_lock(&ci->i_ceph_lock);
2013 flags = ci->i_ceph_flags;
2014 pool = ci->i_layout.pool_id;
2015 spin_unlock(&ci->i_ceph_lock);
2017 if (flags & CEPH_I_POOL_PERM) {
2018 if ((need & CEPH_CAP_FILE_RD) && !(flags & CEPH_I_POOL_RD)) {
2019 dout("ceph_pool_perm_check pool %lld no read perm\n",
2023 if ((need & CEPH_CAP_FILE_WR) && !(flags & CEPH_I_POOL_WR)) {
2024 dout("ceph_pool_perm_check pool %lld no write perm\n",
2031 pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
2032 ret = __ceph_pool_perm_get(ci, pool, pool_ns);
2033 ceph_put_string(pool_ns);
2037 flags = CEPH_I_POOL_PERM;
2038 if (ret & POOL_READ)
2039 flags |= CEPH_I_POOL_RD;
2040 if (ret & POOL_WRITE)
2041 flags |= CEPH_I_POOL_WR;
2043 spin_lock(&ci->i_ceph_lock);
2044 if (pool == ci->i_layout.pool_id &&
2045 pool_ns == rcu_dereference_raw(ci->i_layout.pool_ns)) {
2046 ci->i_ceph_flags |= flags;
2048 pool = ci->i_layout.pool_id;
2049 flags = ci->i_ceph_flags;
2051 spin_unlock(&ci->i_ceph_lock);
2055 void ceph_pool_perm_destroy(struct ceph_mds_client *mdsc)
2057 struct ceph_pool_perm *perm;
2060 while (!RB_EMPTY_ROOT(&mdsc->pool_perm_tree)) {
2061 n = rb_first(&mdsc->pool_perm_tree);
2062 perm = rb_entry(n, struct ceph_pool_perm, node);
2063 rb_erase(n, &mdsc->pool_perm_tree);