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 *folio, 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 invalidate_folio callback.
126 BUG_ON(PagePrivate(page));
127 attach_page_private(page, snapc);
129 return ceph_fscache_set_page_dirty(page);
133 * If we are truncating the full folio (i.e. offset == 0), adjust the
134 * dirty folio counters appropriately. Only called if there is private
137 static void ceph_invalidate_folio(struct folio *folio, size_t offset,
141 struct ceph_inode_info *ci;
142 struct ceph_snap_context *snapc;
144 inode = folio->mapping->host;
145 ci = ceph_inode(inode);
147 if (offset != 0 || length != folio_size(folio)) {
148 dout("%p invalidate_folio idx %lu partial dirty page %zu~%zu\n",
149 inode, folio->index, offset, length);
153 WARN_ON(!folio_test_locked(folio));
154 if (folio_get_private(folio)) {
155 dout("%p invalidate_folio idx %lu full dirty page\n",
156 inode, folio->index);
158 snapc = folio_detach_private(folio);
159 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
160 ceph_put_snap_context(snapc);
163 folio_wait_fscache(folio);
166 static int ceph_releasepage(struct page *page, gfp_t gfp)
168 struct inode *inode = page->mapping->host;
170 dout("%llx:%llx releasepage %p idx %lu (%sdirty)\n",
171 ceph_vinop(inode), page,
172 page->index, PageDirty(page) ? "" : "not ");
174 if (PagePrivate(page))
177 if (PageFsCache(page)) {
178 if (current_is_kswapd() || !(gfp & __GFP_FS))
180 wait_on_page_fscache(page);
182 ceph_fscache_note_page_release(inode);
186 static void ceph_netfs_expand_readahead(struct netfs_read_request *rreq)
188 struct inode *inode = rreq->mapping->host;
189 struct ceph_inode_info *ci = ceph_inode(inode);
190 struct ceph_file_layout *lo = &ci->i_layout;
194 /* Expand the start downward */
195 blockno = div_u64_rem(rreq->start, lo->stripe_unit, &blockoff);
196 rreq->start = blockno * lo->stripe_unit;
197 rreq->len += blockoff;
199 /* Now, round up the length to the next block */
200 rreq->len = roundup(rreq->len, lo->stripe_unit);
203 static bool ceph_netfs_clamp_length(struct netfs_read_subrequest *subreq)
205 struct inode *inode = subreq->rreq->mapping->host;
206 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
207 struct ceph_inode_info *ci = ceph_inode(inode);
211 /* Truncate the extent at the end of the current block */
212 ceph_calc_file_object_mapping(&ci->i_layout, subreq->start, subreq->len,
213 &objno, &objoff, &xlen);
214 subreq->len = min(xlen, fsc->mount_options->rsize);
218 static void finish_netfs_read(struct ceph_osd_request *req)
220 struct ceph_fs_client *fsc = ceph_inode_to_client(req->r_inode);
221 struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0);
222 struct netfs_read_subrequest *subreq = req->r_priv;
224 int err = req->r_result;
226 ceph_update_read_metrics(&fsc->mdsc->metric, req->r_start_latency,
227 req->r_end_latency, osd_data->length, err);
229 dout("%s: result %d subreq->len=%zu i_size=%lld\n", __func__, req->r_result,
230 subreq->len, i_size_read(req->r_inode));
232 /* no object means success but no data */
235 else if (err == -EBLOCKLISTED)
236 fsc->blocklisted = true;
238 if (err >= 0 && err < subreq->len)
239 __set_bit(NETFS_SREQ_CLEAR_TAIL, &subreq->flags);
241 netfs_subreq_terminated(subreq, err, true);
243 num_pages = calc_pages_for(osd_data->alignment, osd_data->length);
244 ceph_put_page_vector(osd_data->pages, num_pages, false);
248 static void ceph_netfs_issue_op(struct netfs_read_subrequest *subreq)
250 struct netfs_read_request *rreq = subreq->rreq;
251 struct inode *inode = rreq->mapping->host;
252 struct ceph_inode_info *ci = ceph_inode(inode);
253 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
254 struct ceph_osd_request *req;
255 struct ceph_vino vino = ceph_vino(inode);
256 struct iov_iter iter;
260 u64 len = subreq->len;
262 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, vino, subreq->start, &len,
263 0, 1, CEPH_OSD_OP_READ,
264 CEPH_OSD_FLAG_READ | fsc->client->osdc.client->options->read_from_replica,
265 NULL, ci->i_truncate_seq, ci->i_truncate_size, false);
272 dout("%s: pos=%llu orig_len=%zu len=%llu\n", __func__, subreq->start, subreq->len, len);
273 iov_iter_xarray(&iter, READ, &rreq->mapping->i_pages, subreq->start, len);
274 err = iov_iter_get_pages_alloc(&iter, &pages, len, &page_off);
276 dout("%s: iov_ter_get_pages_alloc returned %d\n", __func__, err);
280 /* should always give us a page-aligned read */
281 WARN_ON_ONCE(page_off);
284 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false);
285 req->r_callback = finish_netfs_read;
286 req->r_priv = subreq;
287 req->r_inode = inode;
290 err = ceph_osdc_start_request(req->r_osdc, req, false);
294 ceph_osdc_put_request(req);
296 netfs_subreq_terminated(subreq, err, false);
297 dout("%s: result %d\n", __func__, err);
300 static void ceph_readahead_cleanup(struct address_space *mapping, void *priv)
302 struct inode *inode = mapping->host;
303 struct ceph_inode_info *ci = ceph_inode(inode);
304 int got = (uintptr_t)priv;
307 ceph_put_cap_refs(ci, got);
310 static const struct netfs_read_request_ops ceph_netfs_read_ops = {
311 .is_cache_enabled = ceph_is_cache_enabled,
312 .begin_cache_operation = ceph_begin_cache_operation,
313 .issue_op = ceph_netfs_issue_op,
314 .expand_readahead = ceph_netfs_expand_readahead,
315 .clamp_length = ceph_netfs_clamp_length,
316 .check_write_begin = ceph_netfs_check_write_begin,
317 .cleanup = ceph_readahead_cleanup,
320 /* read a single page, without unlocking it. */
321 static int ceph_readpage(struct file *file, struct page *subpage)
323 struct folio *folio = page_folio(subpage);
324 struct inode *inode = file_inode(file);
325 struct ceph_inode_info *ci = ceph_inode(inode);
326 struct ceph_vino vino = ceph_vino(inode);
327 size_t len = folio_size(folio);
328 u64 off = folio_file_pos(folio);
330 if (ci->i_inline_version != CEPH_INLINE_NONE) {
332 * Uptodate inline data should have been added
333 * into page cache while getting Fcr caps.
339 zero_user_segment(&folio->page, 0, folio_size(folio));
340 folio_mark_uptodate(folio);
345 dout("readpage ino %llx.%llx file %p off %llu len %zu folio %p index %lu\n",
346 vino.ino, vino.snap, file, off, len, folio, folio_index(folio));
348 return netfs_readpage(file, folio, &ceph_netfs_read_ops, NULL);
351 static void ceph_readahead(struct readahead_control *ractl)
353 struct inode *inode = file_inode(ractl->file);
354 struct ceph_file_info *fi = ractl->file->private_data;
355 struct ceph_rw_context *rw_ctx;
359 if (ceph_inode(inode)->i_inline_version != CEPH_INLINE_NONE)
362 rw_ctx = ceph_find_rw_context(fi);
365 * readahead callers do not necessarily hold Fcb caps
366 * (e.g. fadvise, madvise).
368 int want = CEPH_CAP_FILE_CACHE;
370 ret = ceph_try_get_caps(inode, CEPH_CAP_FILE_RD, want, true, &got);
372 dout("start_read %p, error getting cap\n", inode);
373 else if (!(got & want))
374 dout("start_read %p, no cache cap\n", inode);
379 netfs_readahead(ractl, &ceph_netfs_read_ops, (void *)(uintptr_t)got);
382 #ifdef CONFIG_CEPH_FSCACHE
383 static void ceph_set_page_fscache(struct page *page)
385 set_page_fscache(page);
388 static void ceph_fscache_write_terminated(void *priv, ssize_t error, bool was_async)
390 struct inode *inode = priv;
392 if (IS_ERR_VALUE(error) && error != -ENOBUFS)
393 ceph_fscache_invalidate(inode, false);
396 static void ceph_fscache_write_to_cache(struct inode *inode, u64 off, u64 len, bool caching)
398 struct ceph_inode_info *ci = ceph_inode(inode);
399 struct fscache_cookie *cookie = ceph_fscache_cookie(ci);
401 fscache_write_to_cache(cookie, inode->i_mapping, off, len, i_size_read(inode),
402 ceph_fscache_write_terminated, inode, caching);
405 static inline void ceph_set_page_fscache(struct page *page)
409 static inline void ceph_fscache_write_to_cache(struct inode *inode, u64 off, u64 len, bool caching)
412 #endif /* CONFIG_CEPH_FSCACHE */
414 struct ceph_writeback_ctl
424 * Get ref for the oldest snapc for an inode with dirty data... that is, the
425 * only snap context we are allowed to write back.
427 static struct ceph_snap_context *
428 get_oldest_context(struct inode *inode, struct ceph_writeback_ctl *ctl,
429 struct ceph_snap_context *page_snapc)
431 struct ceph_inode_info *ci = ceph_inode(inode);
432 struct ceph_snap_context *snapc = NULL;
433 struct ceph_cap_snap *capsnap = NULL;
435 spin_lock(&ci->i_ceph_lock);
436 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
437 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap,
438 capsnap->context, capsnap->dirty_pages);
439 if (!capsnap->dirty_pages)
442 /* get i_size, truncate_{seq,size} for page_snapc? */
443 if (snapc && capsnap->context != page_snapc)
447 if (capsnap->writing) {
448 ctl->i_size = i_size_read(inode);
449 ctl->size_stable = false;
451 ctl->i_size = capsnap->size;
452 ctl->size_stable = true;
454 ctl->truncate_size = capsnap->truncate_size;
455 ctl->truncate_seq = capsnap->truncate_seq;
456 ctl->head_snapc = false;
462 snapc = ceph_get_snap_context(capsnap->context);
464 page_snapc == snapc ||
465 page_snapc->seq > snapc->seq)
468 if (!snapc && ci->i_wrbuffer_ref_head) {
469 snapc = ceph_get_snap_context(ci->i_head_snapc);
470 dout(" head snapc %p has %d dirty pages\n",
471 snapc, ci->i_wrbuffer_ref_head);
473 ctl->i_size = i_size_read(inode);
474 ctl->truncate_size = ci->i_truncate_size;
475 ctl->truncate_seq = ci->i_truncate_seq;
476 ctl->size_stable = false;
477 ctl->head_snapc = true;
480 spin_unlock(&ci->i_ceph_lock);
484 static u64 get_writepages_data_length(struct inode *inode,
485 struct page *page, u64 start)
487 struct ceph_inode_info *ci = ceph_inode(inode);
488 struct ceph_snap_context *snapc = page_snap_context(page);
489 struct ceph_cap_snap *capsnap = NULL;
490 u64 end = i_size_read(inode);
492 if (snapc != ci->i_head_snapc) {
494 spin_lock(&ci->i_ceph_lock);
495 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
496 if (capsnap->context == snapc) {
497 if (!capsnap->writing)
503 spin_unlock(&ci->i_ceph_lock);
506 if (end > page_offset(page) + thp_size(page))
507 end = page_offset(page) + thp_size(page);
508 return end > start ? end - start : 0;
512 * Write a single page, but leave the page locked.
514 * If we get a write error, mark the mapping for error, but still adjust the
515 * dirty page accounting (i.e., page is no longer dirty).
517 static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
519 struct folio *folio = page_folio(page);
520 struct inode *inode = page->mapping->host;
521 struct ceph_inode_info *ci = ceph_inode(inode);
522 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
523 struct ceph_snap_context *snapc, *oldest;
524 loff_t page_off = page_offset(page);
526 loff_t len = thp_size(page);
527 struct ceph_writeback_ctl ceph_wbc;
528 struct ceph_osd_client *osdc = &fsc->client->osdc;
529 struct ceph_osd_request *req;
530 bool caching = ceph_is_cache_enabled(inode);
532 dout("writepage %p idx %lu\n", page, page->index);
534 /* verify this is a writeable snap context */
535 snapc = page_snap_context(page);
537 dout("writepage %p page %p not dirty?\n", inode, page);
540 oldest = get_oldest_context(inode, &ceph_wbc, snapc);
541 if (snapc->seq > oldest->seq) {
542 dout("writepage %p page %p snapc %p not writeable - noop\n",
544 /* we should only noop if called by kswapd */
545 WARN_ON(!(current->flags & PF_MEMALLOC));
546 ceph_put_snap_context(oldest);
547 redirty_page_for_writepage(wbc, page);
550 ceph_put_snap_context(oldest);
552 /* is this a partial page at end of file? */
553 if (page_off >= ceph_wbc.i_size) {
554 dout("folio at %lu beyond eof %llu\n", folio->index,
556 folio_invalidate(folio, 0, folio_size(folio));
560 if (ceph_wbc.i_size < page_off + len)
561 len = ceph_wbc.i_size - page_off;
563 dout("writepage %p page %p index %lu on %llu~%llu snapc %p seq %lld\n",
564 inode, page, page->index, page_off, len, snapc, snapc->seq);
566 if (atomic_long_inc_return(&fsc->writeback_count) >
567 CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb))
568 set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC);
570 req = ceph_osdc_new_request(osdc, &ci->i_layout, ceph_vino(inode), page_off, &len, 0, 1,
571 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, snapc,
572 ceph_wbc.truncate_seq, ceph_wbc.truncate_size,
577 set_page_writeback(page);
579 ceph_set_page_fscache(page);
580 ceph_fscache_write_to_cache(inode, page_off, len, caching);
582 /* it may be a short write due to an object boundary */
583 WARN_ON_ONCE(len > thp_size(page));
584 osd_req_op_extent_osd_data_pages(req, 0, &page, len, 0, false, false);
585 dout("writepage %llu~%llu (%llu bytes)\n", page_off, len, len);
587 req->r_mtime = inode->i_mtime;
588 err = ceph_osdc_start_request(osdc, req, true);
590 err = ceph_osdc_wait_request(osdc, req);
592 ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
593 req->r_end_latency, len, err);
595 ceph_osdc_put_request(req);
600 struct writeback_control tmp_wbc;
603 if (err == -ERESTARTSYS) {
604 /* killed by SIGKILL */
605 dout("writepage interrupted page %p\n", page);
606 redirty_page_for_writepage(wbc, page);
607 end_page_writeback(page);
610 if (err == -EBLOCKLISTED)
611 fsc->blocklisted = true;
612 dout("writepage setting page/mapping error %d %p\n",
614 mapping_set_error(&inode->i_data, err);
615 wbc->pages_skipped++;
617 dout("writepage cleaned page %p\n", page);
618 err = 0; /* vfs expects us to return 0 */
620 oldest = detach_page_private(page);
621 WARN_ON_ONCE(oldest != snapc);
622 end_page_writeback(page);
623 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
624 ceph_put_snap_context(snapc); /* page's reference */
626 if (atomic_long_dec_return(&fsc->writeback_count) <
627 CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb))
628 clear_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC);
633 static int ceph_writepage(struct page *page, struct writeback_control *wbc)
636 struct inode *inode = page->mapping->host;
640 wait_on_page_fscache(page);
642 err = writepage_nounlock(page, wbc);
643 if (err == -ERESTARTSYS) {
644 /* direct memory reclaimer was killed by SIGKILL. return 0
645 * to prevent caller from setting mapping/page error */
654 * async writeback completion handler.
656 * If we get an error, set the mapping error bit, but not the individual
659 static void writepages_finish(struct ceph_osd_request *req)
661 struct inode *inode = req->r_inode;
662 struct ceph_inode_info *ci = ceph_inode(inode);
663 struct ceph_osd_data *osd_data;
665 int num_pages, total_pages = 0;
667 int rc = req->r_result;
668 struct ceph_snap_context *snapc = req->r_snapc;
669 struct address_space *mapping = inode->i_mapping;
670 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
671 unsigned int len = 0;
674 dout("writepages_finish %p rc %d\n", inode, rc);
676 mapping_set_error(mapping, rc);
677 ceph_set_error_write(ci);
678 if (rc == -EBLOCKLISTED)
679 fsc->blocklisted = true;
681 ceph_clear_error_write(ci);
685 * We lost the cache cap, need to truncate the page before
686 * it is unlocked, otherwise we'd truncate it later in the
687 * page truncation thread, possibly losing some data that
690 remove_page = !(ceph_caps_issued(ci) &
691 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO));
693 /* clean all pages */
694 for (i = 0; i < req->r_num_ops; i++) {
695 if (req->r_ops[i].op != CEPH_OSD_OP_WRITE)
698 osd_data = osd_req_op_extent_osd_data(req, i);
699 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
700 len += osd_data->length;
701 num_pages = calc_pages_for((u64)osd_data->alignment,
702 (u64)osd_data->length);
703 total_pages += num_pages;
704 for (j = 0; j < num_pages; j++) {
705 page = osd_data->pages[j];
707 WARN_ON(!PageUptodate(page));
709 if (atomic_long_dec_return(&fsc->writeback_count) <
710 CONGESTION_OFF_THRESH(
711 fsc->mount_options->congestion_kb))
712 clear_bdi_congested(inode_to_bdi(inode),
715 ceph_put_snap_context(detach_page_private(page));
716 end_page_writeback(page);
717 dout("unlocking %p\n", page);
720 generic_error_remove_page(inode->i_mapping,
725 dout("writepages_finish %p wrote %llu bytes cleaned %d pages\n",
726 inode, osd_data->length, rc >= 0 ? num_pages : 0);
728 release_pages(osd_data->pages, num_pages);
731 ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
732 req->r_end_latency, len, rc);
734 ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc);
736 osd_data = osd_req_op_extent_osd_data(req, 0);
737 if (osd_data->pages_from_pool)
738 mempool_free(osd_data->pages, ceph_wb_pagevec_pool);
740 kfree(osd_data->pages);
741 ceph_osdc_put_request(req);
745 * initiate async writeback
747 static int ceph_writepages_start(struct address_space *mapping,
748 struct writeback_control *wbc)
750 struct inode *inode = mapping->host;
751 struct ceph_inode_info *ci = ceph_inode(inode);
752 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
753 struct ceph_vino vino = ceph_vino(inode);
754 pgoff_t index, start_index, end = -1;
755 struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc;
758 unsigned int wsize = i_blocksize(inode);
759 struct ceph_osd_request *req = NULL;
760 struct ceph_writeback_ctl ceph_wbc;
761 bool should_loop, range_whole = false;
763 bool caching = ceph_is_cache_enabled(inode);
765 dout("writepages_start %p (mode=%s)\n", inode,
766 wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
767 (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
769 if (ceph_inode_is_shutdown(inode)) {
770 if (ci->i_wrbuffer_ref > 0) {
772 "writepage_start %p %lld forced umount\n",
773 inode, ceph_ino(inode));
775 mapping_set_error(mapping, -EIO);
776 return -EIO; /* we're in a forced umount, don't write! */
778 if (fsc->mount_options->wsize < wsize)
779 wsize = fsc->mount_options->wsize;
783 start_index = wbc->range_cyclic ? mapping->writeback_index : 0;
787 /* find oldest snap context with dirty data */
788 snapc = get_oldest_context(inode, &ceph_wbc, NULL);
790 /* hmm, why does writepages get called when there
792 dout(" no snap context with dirty data?\n");
795 dout(" oldest snapc is %p seq %lld (%d snaps)\n",
796 snapc, snapc->seq, snapc->num_snaps);
799 if (ceph_wbc.head_snapc && snapc != last_snapc) {
800 /* where to start/end? */
801 if (wbc->range_cyclic) {
806 dout(" cyclic, start at %lu\n", index);
808 index = wbc->range_start >> PAGE_SHIFT;
809 end = wbc->range_end >> PAGE_SHIFT;
810 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
812 dout(" not cyclic, %lu to %lu\n", index, end);
814 } else if (!ceph_wbc.head_snapc) {
815 /* Do not respect wbc->range_{start,end}. Dirty pages
816 * in that range can be associated with newer snapc.
817 * They are not writeable until we write all dirty pages
818 * associated with 'snapc' get written */
821 dout(" non-head snapc, range whole\n");
824 ceph_put_snap_context(last_snapc);
827 while (!done && index <= end) {
828 int num_ops = 0, op_idx;
829 unsigned i, pvec_pages, max_pages, locked_pages = 0;
830 struct page **pages = NULL, **data_pages;
832 pgoff_t strip_unit_end = 0;
833 u64 offset = 0, len = 0;
834 bool from_pool = false;
836 max_pages = wsize >> PAGE_SHIFT;
839 pvec_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
840 end, PAGECACHE_TAG_DIRTY);
841 dout("pagevec_lookup_range_tag got %d\n", pvec_pages);
842 if (!pvec_pages && !locked_pages)
844 for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) {
845 page = pvec.pages[i];
846 dout("? %p idx %lu\n", page, page->index);
847 if (locked_pages == 0)
848 lock_page(page); /* first page */
849 else if (!trylock_page(page))
852 /* only dirty pages, or our accounting breaks */
853 if (unlikely(!PageDirty(page)) ||
854 unlikely(page->mapping != mapping)) {
855 dout("!dirty or !mapping %p\n", page);
859 /* only if matching snap context */
860 pgsnapc = page_snap_context(page);
861 if (pgsnapc != snapc) {
862 dout("page snapc %p %lld != oldest %p %lld\n",
863 pgsnapc, pgsnapc->seq, snapc, snapc->seq);
865 !ceph_wbc.head_snapc &&
866 wbc->sync_mode != WB_SYNC_NONE)
871 if (page_offset(page) >= ceph_wbc.i_size) {
872 struct folio *folio = page_folio(page);
874 dout("folio at %lu beyond eof %llu\n",
875 folio->index, ceph_wbc.i_size);
876 if ((ceph_wbc.size_stable ||
877 folio_pos(folio) >= i_size_read(inode)) &&
878 folio_clear_dirty_for_io(folio))
879 folio_invalidate(folio, 0,
884 if (strip_unit_end && (page->index > strip_unit_end)) {
885 dout("end of strip unit %p\n", page);
889 if (PageWriteback(page) || PageFsCache(page)) {
890 if (wbc->sync_mode == WB_SYNC_NONE) {
891 dout("%p under writeback\n", page);
895 dout("waiting on writeback %p\n", page);
896 wait_on_page_writeback(page);
897 wait_on_page_fscache(page);
900 if (!clear_page_dirty_for_io(page)) {
901 dout("%p !clear_page_dirty_for_io\n", page);
907 * We have something to write. If this is
908 * the first locked page this time through,
909 * calculate max possinle write size and
910 * allocate a page array
912 if (locked_pages == 0) {
917 /* prepare async write request */
918 offset = (u64)page_offset(page);
919 ceph_calc_file_object_mapping(&ci->i_layout,
926 strip_unit_end = page->index +
927 ((len - 1) >> PAGE_SHIFT);
930 max_pages = calc_pages_for(0, (u64)len);
931 pages = kmalloc_array(max_pages,
936 pages = mempool_alloc(ceph_wb_pagevec_pool, GFP_NOFS);
941 } else if (page->index !=
942 (offset + len) >> PAGE_SHIFT) {
943 if (num_ops >= (from_pool ? CEPH_OSD_SLAB_OPS :
945 redirty_page_for_writepage(wbc, page);
951 offset = (u64)page_offset(page);
955 /* note position of first page in pvec */
956 dout("%p will write page %p idx %lu\n",
957 inode, page, page->index);
959 if (atomic_long_inc_return(&fsc->writeback_count) >
960 CONGESTION_ON_THRESH(
961 fsc->mount_options->congestion_kb)) {
962 set_bdi_congested(inode_to_bdi(inode),
967 pages[locked_pages++] = page;
968 pvec.pages[i] = NULL;
970 len += thp_size(page);
973 /* did we get anything? */
975 goto release_pvec_pages;
978 /* shift unused page to beginning of pvec */
979 for (j = 0; j < pvec_pages; j++) {
983 pvec.pages[n] = pvec.pages[j];
988 if (pvec_pages && i == pvec_pages &&
989 locked_pages < max_pages) {
990 dout("reached end pvec, trying for more\n");
991 pagevec_release(&pvec);
997 offset = page_offset(pages[0]);
1000 req = ceph_osdc_new_request(&fsc->client->osdc,
1001 &ci->i_layout, vino,
1002 offset, &len, 0, num_ops,
1003 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
1004 snapc, ceph_wbc.truncate_seq,
1005 ceph_wbc.truncate_size, false);
1007 req = ceph_osdc_new_request(&fsc->client->osdc,
1008 &ci->i_layout, vino,
1013 CEPH_OSD_FLAG_WRITE,
1014 snapc, ceph_wbc.truncate_seq,
1015 ceph_wbc.truncate_size, true);
1016 BUG_ON(IS_ERR(req));
1018 BUG_ON(len < page_offset(pages[locked_pages - 1]) +
1019 thp_size(page) - offset);
1021 req->r_callback = writepages_finish;
1022 req->r_inode = inode;
1024 /* Format the osd request message and submit the write */
1028 for (i = 0; i < locked_pages; i++) {
1029 u64 cur_offset = page_offset(pages[i]);
1031 * Discontinuity in page range? Ceph can handle that by just passing
1032 * multiple extents in the write op.
1034 if (offset + len != cur_offset) {
1035 /* If it's full, stop here */
1036 if (op_idx + 1 == req->r_num_ops)
1039 /* Kick off an fscache write with what we have so far. */
1040 ceph_fscache_write_to_cache(inode, offset, len, caching);
1042 /* Start a new extent */
1043 osd_req_op_extent_dup_last(req, op_idx,
1044 cur_offset - offset);
1045 dout("writepages got pages at %llu~%llu\n",
1047 osd_req_op_extent_osd_data_pages(req, op_idx,
1050 osd_req_op_extent_update(req, op_idx, len);
1053 offset = cur_offset;
1054 data_pages = pages + i;
1058 set_page_writeback(pages[i]);
1060 ceph_set_page_fscache(pages[i]);
1061 len += thp_size(page);
1063 ceph_fscache_write_to_cache(inode, offset, len, caching);
1065 if (ceph_wbc.size_stable) {
1066 len = min(len, ceph_wbc.i_size - offset);
1067 } else if (i == locked_pages) {
1068 /* writepages_finish() clears writeback pages
1069 * according to the data length, so make sure
1070 * data length covers all locked pages */
1071 u64 min_len = len + 1 - thp_size(page);
1072 len = get_writepages_data_length(inode, pages[i - 1],
1074 len = max(len, min_len);
1076 dout("writepages got pages at %llu~%llu\n", offset, len);
1078 osd_req_op_extent_osd_data_pages(req, op_idx, data_pages, len,
1079 0, from_pool, false);
1080 osd_req_op_extent_update(req, op_idx, len);
1082 BUG_ON(op_idx + 1 != req->r_num_ops);
1085 if (i < locked_pages) {
1086 BUG_ON(num_ops <= req->r_num_ops);
1087 num_ops -= req->r_num_ops;
1090 /* allocate new pages array for next request */
1092 pages = kmalloc_array(locked_pages, sizeof(*pages),
1096 pages = mempool_alloc(ceph_wb_pagevec_pool, GFP_NOFS);
1099 memcpy(pages, data_pages + i,
1100 locked_pages * sizeof(*pages));
1101 memset(data_pages + i, 0,
1102 locked_pages * sizeof(*pages));
1104 BUG_ON(num_ops != req->r_num_ops);
1105 index = pages[i - 1]->index + 1;
1106 /* request message now owns the pages array */
1110 req->r_mtime = inode->i_mtime;
1111 rc = ceph_osdc_start_request(&fsc->client->osdc, req, true);
1115 wbc->nr_to_write -= i;
1120 * We stop writing back only if we are not doing
1121 * integrity sync. In case of integrity sync we have to
1122 * keep going until we have written all the pages
1123 * we tagged for writeback prior to entering this loop.
1125 if (wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE)
1129 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr,
1130 pvec.nr ? pvec.pages[0] : NULL);
1131 pagevec_release(&pvec);
1134 if (should_loop && !done) {
1135 /* more to do; loop back to beginning of file */
1136 dout("writepages looping back to beginning of file\n");
1137 end = start_index - 1; /* OK even when start_index == 0 */
1139 /* to write dirty pages associated with next snapc,
1140 * we need to wait until current writes complete */
1141 if (wbc->sync_mode != WB_SYNC_NONE &&
1142 start_index == 0 && /* all dirty pages were checked */
1143 !ceph_wbc.head_snapc) {
1147 while ((index <= end) &&
1148 (nr = pagevec_lookup_tag(&pvec, mapping, &index,
1149 PAGECACHE_TAG_WRITEBACK))) {
1150 for (i = 0; i < nr; i++) {
1151 page = pvec.pages[i];
1152 if (page_snap_context(page) != snapc)
1154 wait_on_page_writeback(page);
1156 pagevec_release(&pvec);
1166 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1167 mapping->writeback_index = index;
1170 ceph_osdc_put_request(req);
1171 ceph_put_snap_context(last_snapc);
1172 dout("writepages dend - startone, rc = %d\n", rc);
1179 * See if a given @snapc is either writeable, or already written.
1181 static int context_is_writeable_or_written(struct inode *inode,
1182 struct ceph_snap_context *snapc)
1184 struct ceph_snap_context *oldest = get_oldest_context(inode, NULL, NULL);
1185 int ret = !oldest || snapc->seq <= oldest->seq;
1187 ceph_put_snap_context(oldest);
1192 * ceph_find_incompatible - find an incompatible context and return it
1193 * @page: page being dirtied
1195 * We are only allowed to write into/dirty a page if the page is
1196 * clean, or already dirty within the same snap context. Returns a
1197 * conflicting context if there is one, NULL if there isn't, or a
1198 * negative error code on other errors.
1200 * Must be called with page lock held.
1202 static struct ceph_snap_context *
1203 ceph_find_incompatible(struct page *page)
1205 struct inode *inode = page->mapping->host;
1206 struct ceph_inode_info *ci = ceph_inode(inode);
1208 if (ceph_inode_is_shutdown(inode)) {
1209 dout(" page %p %llx:%llx is shutdown\n", page,
1211 return ERR_PTR(-ESTALE);
1215 struct ceph_snap_context *snapc, *oldest;
1217 wait_on_page_writeback(page);
1219 snapc = page_snap_context(page);
1220 if (!snapc || snapc == ci->i_head_snapc)
1224 * this page is already dirty in another (older) snap
1225 * context! is it writeable now?
1227 oldest = get_oldest_context(inode, NULL, NULL);
1228 if (snapc->seq > oldest->seq) {
1229 /* not writeable -- return it for the caller to deal with */
1230 ceph_put_snap_context(oldest);
1231 dout(" page %p snapc %p not current or oldest\n", page, snapc);
1232 return ceph_get_snap_context(snapc);
1234 ceph_put_snap_context(oldest);
1236 /* yay, writeable, do it now (without dropping page lock) */
1237 dout(" page %p snapc %p not current, but oldest\n", page, snapc);
1238 if (clear_page_dirty_for_io(page)) {
1239 int r = writepage_nounlock(page, NULL);
1247 static int ceph_netfs_check_write_begin(struct file *file, loff_t pos, unsigned int len,
1248 struct folio *folio, void **_fsdata)
1250 struct inode *inode = file_inode(file);
1251 struct ceph_inode_info *ci = ceph_inode(inode);
1252 struct ceph_snap_context *snapc;
1254 snapc = ceph_find_incompatible(folio_page(folio, 0));
1258 folio_unlock(folio);
1261 return PTR_ERR(snapc);
1263 ceph_queue_writeback(inode);
1264 r = wait_event_killable(ci->i_cap_wq,
1265 context_is_writeable_or_written(inode, snapc));
1266 ceph_put_snap_context(snapc);
1267 return r == 0 ? -EAGAIN : r;
1273 * We are only allowed to write into/dirty the page if the page is
1274 * clean, or already dirty within the same snap context.
1276 static int ceph_write_begin(struct file *file, struct address_space *mapping,
1277 loff_t pos, unsigned len, unsigned aop_flags,
1278 struct page **pagep, void **fsdata)
1280 struct inode *inode = file_inode(file);
1281 struct ceph_inode_info *ci = ceph_inode(inode);
1282 struct folio *folio = NULL;
1283 pgoff_t index = pos >> PAGE_SHIFT;
1287 * Uninlining should have already been done and everything updated, EXCEPT
1288 * for inline_version sent to the MDS.
1290 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1291 unsigned int fgp_flags = FGP_LOCK | FGP_WRITE | FGP_CREAT | FGP_STABLE;
1292 if (aop_flags & AOP_FLAG_NOFS)
1293 fgp_flags |= FGP_NOFS;
1294 folio = __filemap_get_folio(mapping, index, fgp_flags,
1295 mapping_gfp_mask(mapping));
1300 * The inline_version on a new inode is set to 1. If that's the
1301 * case, then the folio is brand new and isn't yet Uptodate.
1304 if (index == 0 && ci->i_inline_version != 1) {
1305 if (!folio_test_uptodate(folio)) {
1306 WARN_ONCE(1, "ceph: write_begin called on still-inlined inode (inline_version %llu)!\n",
1307 ci->i_inline_version);
1312 zero_user_segment(&folio->page, 0, folio_size(folio));
1313 folio_mark_uptodate(folio);
1317 r = netfs_write_begin(file, inode->i_mapping, pos, len, 0, &folio, NULL,
1318 &ceph_netfs_read_ops, NULL);
1321 folio_wait_fscache(folio);
1326 WARN_ON_ONCE(!folio_test_locked(folio));
1327 *pagep = &folio->page;
1333 * we don't do anything in here that simple_write_end doesn't do
1334 * except adjust dirty page accounting
1336 static int ceph_write_end(struct file *file, struct address_space *mapping,
1337 loff_t pos, unsigned len, unsigned copied,
1338 struct page *subpage, void *fsdata)
1340 struct folio *folio = page_folio(subpage);
1341 struct inode *inode = file_inode(file);
1342 bool check_cap = false;
1344 dout("write_end file %p inode %p folio %p %d~%d (%d)\n", file,
1345 inode, folio, (int)pos, (int)copied, (int)len);
1347 if (!folio_test_uptodate(folio)) {
1348 /* just return that nothing was copied on a short copy */
1353 folio_mark_uptodate(folio);
1356 /* did file size increase? */
1357 if (pos+copied > i_size_read(inode))
1358 check_cap = ceph_inode_set_size(inode, pos+copied);
1360 folio_mark_dirty(folio);
1363 folio_unlock(folio);
1367 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL);
1372 const struct address_space_operations ceph_aops = {
1373 .readpage = ceph_readpage,
1374 .readahead = ceph_readahead,
1375 .writepage = ceph_writepage,
1376 .writepages = ceph_writepages_start,
1377 .write_begin = ceph_write_begin,
1378 .write_end = ceph_write_end,
1379 .set_page_dirty = ceph_set_page_dirty,
1380 .invalidate_folio = ceph_invalidate_folio,
1381 .releasepage = ceph_releasepage,
1382 .direct_IO = noop_direct_IO,
1385 static void ceph_block_sigs(sigset_t *oldset)
1388 siginitsetinv(&mask, sigmask(SIGKILL));
1389 sigprocmask(SIG_BLOCK, &mask, oldset);
1392 static void ceph_restore_sigs(sigset_t *oldset)
1394 sigprocmask(SIG_SETMASK, oldset, NULL);
1400 static vm_fault_t ceph_filemap_fault(struct vm_fault *vmf)
1402 struct vm_area_struct *vma = vmf->vma;
1403 struct inode *inode = file_inode(vma->vm_file);
1404 struct ceph_inode_info *ci = ceph_inode(inode);
1405 struct ceph_file_info *fi = vma->vm_file->private_data;
1406 loff_t off = (loff_t)vmf->pgoff << PAGE_SHIFT;
1409 vm_fault_t ret = VM_FAULT_SIGBUS;
1411 if (ceph_inode_is_shutdown(inode))
1414 ceph_block_sigs(&oldset);
1416 dout("filemap_fault %p %llx.%llx %llu trying to get caps\n",
1417 inode, ceph_vinop(inode), off);
1418 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1419 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1421 want = CEPH_CAP_FILE_CACHE;
1424 err = ceph_get_caps(vma->vm_file, CEPH_CAP_FILE_RD, want, -1, &got);
1428 dout("filemap_fault %p %llu got cap refs on %s\n",
1429 inode, off, ceph_cap_string(got));
1431 if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) ||
1432 ci->i_inline_version == CEPH_INLINE_NONE) {
1433 CEPH_DEFINE_RW_CONTEXT(rw_ctx, got);
1434 ceph_add_rw_context(fi, &rw_ctx);
1435 ret = filemap_fault(vmf);
1436 ceph_del_rw_context(fi, &rw_ctx);
1437 dout("filemap_fault %p %llu drop cap refs %s ret %x\n",
1438 inode, off, ceph_cap_string(got), ret);
1442 ceph_put_cap_refs(ci, got);
1447 /* read inline data */
1448 if (off >= PAGE_SIZE) {
1449 /* does not support inline data > PAGE_SIZE */
1450 ret = VM_FAULT_SIGBUS;
1452 struct address_space *mapping = inode->i_mapping;
1455 filemap_invalidate_lock_shared(mapping);
1456 page = find_or_create_page(mapping, 0,
1457 mapping_gfp_constraint(mapping, ~__GFP_FS));
1462 err = __ceph_do_getattr(inode, page,
1463 CEPH_STAT_CAP_INLINE_DATA, true);
1464 if (err < 0 || off >= i_size_read(inode)) {
1467 ret = vmf_error(err);
1470 if (err < PAGE_SIZE)
1471 zero_user_segment(page, err, PAGE_SIZE);
1473 flush_dcache_page(page);
1474 SetPageUptodate(page);
1476 ret = VM_FAULT_MAJOR | VM_FAULT_LOCKED;
1478 filemap_invalidate_unlock_shared(mapping);
1479 dout("filemap_fault %p %llu read inline data ret %x\n",
1483 ceph_restore_sigs(&oldset);
1485 ret = vmf_error(err);
1490 static vm_fault_t ceph_page_mkwrite(struct vm_fault *vmf)
1492 struct vm_area_struct *vma = vmf->vma;
1493 struct inode *inode = file_inode(vma->vm_file);
1494 struct ceph_inode_info *ci = ceph_inode(inode);
1495 struct ceph_file_info *fi = vma->vm_file->private_data;
1496 struct ceph_cap_flush *prealloc_cf;
1497 struct page *page = vmf->page;
1498 loff_t off = page_offset(page);
1499 loff_t size = i_size_read(inode);
1503 vm_fault_t ret = VM_FAULT_SIGBUS;
1505 if (ceph_inode_is_shutdown(inode))
1508 prealloc_cf = ceph_alloc_cap_flush();
1510 return VM_FAULT_OOM;
1512 sb_start_pagefault(inode->i_sb);
1513 ceph_block_sigs(&oldset);
1515 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1516 struct page *locked_page = NULL;
1521 err = ceph_uninline_data(vma->vm_file, locked_page);
1523 unlock_page(locked_page);
1528 if (off + thp_size(page) <= size)
1529 len = thp_size(page);
1531 len = offset_in_thp(page, size);
1533 dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n",
1534 inode, ceph_vinop(inode), off, len, size);
1535 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1536 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1538 want = CEPH_CAP_FILE_BUFFER;
1541 err = ceph_get_caps(vma->vm_file, CEPH_CAP_FILE_WR, want, off + len, &got);
1545 dout("page_mkwrite %p %llu~%zd got cap refs on %s\n",
1546 inode, off, len, ceph_cap_string(got));
1548 /* Update time before taking page lock */
1549 file_update_time(vma->vm_file);
1550 inode_inc_iversion_raw(inode);
1553 struct ceph_snap_context *snapc;
1557 if (page_mkwrite_check_truncate(page, inode) < 0) {
1559 ret = VM_FAULT_NOPAGE;
1563 snapc = ceph_find_incompatible(page);
1565 /* success. we'll keep the page locked. */
1566 set_page_dirty(page);
1567 ret = VM_FAULT_LOCKED;
1573 if (IS_ERR(snapc)) {
1574 ret = VM_FAULT_SIGBUS;
1578 ceph_queue_writeback(inode);
1579 err = wait_event_killable(ci->i_cap_wq,
1580 context_is_writeable_or_written(inode, snapc));
1581 ceph_put_snap_context(snapc);
1584 if (ret == VM_FAULT_LOCKED ||
1585 ci->i_inline_version != CEPH_INLINE_NONE) {
1587 spin_lock(&ci->i_ceph_lock);
1588 ci->i_inline_version = CEPH_INLINE_NONE;
1589 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1591 spin_unlock(&ci->i_ceph_lock);
1593 __mark_inode_dirty(inode, dirty);
1596 dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %x\n",
1597 inode, off, len, ceph_cap_string(got), ret);
1598 ceph_put_cap_refs_async(ci, got);
1600 ceph_restore_sigs(&oldset);
1601 sb_end_pagefault(inode->i_sb);
1602 ceph_free_cap_flush(prealloc_cf);
1604 ret = vmf_error(err);
1608 void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
1609 char *data, size_t len)
1611 struct address_space *mapping = inode->i_mapping;
1617 if (i_size_read(inode) == 0)
1619 page = find_or_create_page(mapping, 0,
1620 mapping_gfp_constraint(mapping,
1624 if (PageUptodate(page)) {
1631 dout("fill_inline_data %p %llx.%llx len %zu locked_page %p\n",
1632 inode, ceph_vinop(inode), len, locked_page);
1635 void *kaddr = kmap_atomic(page);
1636 memcpy(kaddr, data, len);
1637 kunmap_atomic(kaddr);
1640 if (page != locked_page) {
1641 if (len < PAGE_SIZE)
1642 zero_user_segment(page, len, PAGE_SIZE);
1644 flush_dcache_page(page);
1646 SetPageUptodate(page);
1652 int ceph_uninline_data(struct file *filp, struct page *locked_page)
1654 struct inode *inode = file_inode(filp);
1655 struct ceph_inode_info *ci = ceph_inode(inode);
1656 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1657 struct ceph_osd_request *req;
1658 struct page *page = NULL;
1659 u64 len, inline_version;
1661 bool from_pagecache = false;
1663 spin_lock(&ci->i_ceph_lock);
1664 inline_version = ci->i_inline_version;
1665 spin_unlock(&ci->i_ceph_lock);
1667 dout("uninline_data %p %llx.%llx inline_version %llu\n",
1668 inode, ceph_vinop(inode), inline_version);
1670 if (inline_version == 1 || /* initial version, no data */
1671 inline_version == CEPH_INLINE_NONE)
1676 WARN_ON(!PageUptodate(page));
1677 } else if (ceph_caps_issued(ci) &
1678 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) {
1679 page = find_get_page(inode->i_mapping, 0);
1681 if (PageUptodate(page)) {
1682 from_pagecache = true;
1692 len = i_size_read(inode);
1693 if (len > PAGE_SIZE)
1696 page = __page_cache_alloc(GFP_NOFS);
1701 err = __ceph_do_getattr(inode, page,
1702 CEPH_STAT_CAP_INLINE_DATA, true);
1704 /* no inline data */
1705 if (err == -ENODATA)
1712 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1713 ceph_vino(inode), 0, &len, 0, 1,
1714 CEPH_OSD_OP_CREATE, CEPH_OSD_FLAG_WRITE,
1721 req->r_mtime = inode->i_mtime;
1722 err = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1724 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1725 ceph_osdc_put_request(req);
1729 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1730 ceph_vino(inode), 0, &len, 1, 3,
1731 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
1732 NULL, ci->i_truncate_seq,
1733 ci->i_truncate_size, false);
1739 osd_req_op_extent_osd_data_pages(req, 1, &page, len, 0, false, false);
1742 __le64 xattr_buf = cpu_to_le64(inline_version);
1743 err = osd_req_op_xattr_init(req, 0, CEPH_OSD_OP_CMPXATTR,
1744 "inline_version", &xattr_buf,
1746 CEPH_OSD_CMPXATTR_OP_GT,
1747 CEPH_OSD_CMPXATTR_MODE_U64);
1754 int xattr_len = snprintf(xattr_buf, sizeof(xattr_buf),
1755 "%llu", inline_version);
1756 err = osd_req_op_xattr_init(req, 2, CEPH_OSD_OP_SETXATTR,
1758 xattr_buf, xattr_len, 0, 0);
1763 req->r_mtime = inode->i_mtime;
1764 err = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1766 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1768 ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
1769 req->r_end_latency, len, err);
1772 ceph_osdc_put_request(req);
1773 if (err == -ECANCELED)
1776 if (page && page != locked_page) {
1777 if (from_pagecache) {
1781 __free_pages(page, 0);
1784 dout("uninline_data %p %llx.%llx inline_version %llu = %d\n",
1785 inode, ceph_vinop(inode), inline_version, err);
1789 static const struct vm_operations_struct ceph_vmops = {
1790 .fault = ceph_filemap_fault,
1791 .page_mkwrite = ceph_page_mkwrite,
1794 int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1796 struct address_space *mapping = file->f_mapping;
1798 if (!mapping->a_ops->readpage)
1800 file_accessed(file);
1801 vma->vm_ops = &ceph_vmops;
1810 static int __ceph_pool_perm_get(struct ceph_inode_info *ci,
1811 s64 pool, struct ceph_string *pool_ns)
1813 struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode);
1814 struct ceph_mds_client *mdsc = fsc->mdsc;
1815 struct ceph_osd_request *rd_req = NULL, *wr_req = NULL;
1816 struct rb_node **p, *parent;
1817 struct ceph_pool_perm *perm;
1818 struct page **pages;
1820 int err = 0, err2 = 0, have = 0;
1822 down_read(&mdsc->pool_perm_rwsem);
1823 p = &mdsc->pool_perm_tree.rb_node;
1825 perm = rb_entry(*p, struct ceph_pool_perm, node);
1826 if (pool < perm->pool)
1828 else if (pool > perm->pool)
1829 p = &(*p)->rb_right;
1831 int ret = ceph_compare_string(pool_ns,
1837 p = &(*p)->rb_right;
1844 up_read(&mdsc->pool_perm_rwsem);
1849 dout("__ceph_pool_perm_get pool %lld ns %.*s no perm cached\n",
1850 pool, (int)pool_ns->len, pool_ns->str);
1852 dout("__ceph_pool_perm_get pool %lld no perm cached\n", pool);
1854 down_write(&mdsc->pool_perm_rwsem);
1855 p = &mdsc->pool_perm_tree.rb_node;
1859 perm = rb_entry(parent, struct ceph_pool_perm, node);
1860 if (pool < perm->pool)
1862 else if (pool > perm->pool)
1863 p = &(*p)->rb_right;
1865 int ret = ceph_compare_string(pool_ns,
1871 p = &(*p)->rb_right;
1879 up_write(&mdsc->pool_perm_rwsem);
1883 rd_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1884 1, false, GFP_NOFS);
1890 rd_req->r_flags = CEPH_OSD_FLAG_READ;
1891 osd_req_op_init(rd_req, 0, CEPH_OSD_OP_STAT, 0);
1892 rd_req->r_base_oloc.pool = pool;
1894 rd_req->r_base_oloc.pool_ns = ceph_get_string(pool_ns);
1895 ceph_oid_printf(&rd_req->r_base_oid, "%llx.00000000", ci->i_vino.ino);
1897 err = ceph_osdc_alloc_messages(rd_req, GFP_NOFS);
1901 wr_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1902 1, false, GFP_NOFS);
1908 wr_req->r_flags = CEPH_OSD_FLAG_WRITE;
1909 osd_req_op_init(wr_req, 0, CEPH_OSD_OP_CREATE, CEPH_OSD_OP_FLAG_EXCL);
1910 ceph_oloc_copy(&wr_req->r_base_oloc, &rd_req->r_base_oloc);
1911 ceph_oid_copy(&wr_req->r_base_oid, &rd_req->r_base_oid);
1913 err = ceph_osdc_alloc_messages(wr_req, GFP_NOFS);
1917 /* one page should be large enough for STAT data */
1918 pages = ceph_alloc_page_vector(1, GFP_KERNEL);
1919 if (IS_ERR(pages)) {
1920 err = PTR_ERR(pages);
1924 osd_req_op_raw_data_in_pages(rd_req, 0, pages, PAGE_SIZE,
1926 err = ceph_osdc_start_request(&fsc->client->osdc, rd_req, false);
1928 wr_req->r_mtime = ci->vfs_inode.i_mtime;
1929 err2 = ceph_osdc_start_request(&fsc->client->osdc, wr_req, false);
1932 err = ceph_osdc_wait_request(&fsc->client->osdc, rd_req);
1934 err2 = ceph_osdc_wait_request(&fsc->client->osdc, wr_req);
1936 if (err >= 0 || err == -ENOENT)
1938 else if (err != -EPERM) {
1939 if (err == -EBLOCKLISTED)
1940 fsc->blocklisted = true;
1944 if (err2 == 0 || err2 == -EEXIST)
1946 else if (err2 != -EPERM) {
1947 if (err2 == -EBLOCKLISTED)
1948 fsc->blocklisted = true;
1953 pool_ns_len = pool_ns ? pool_ns->len : 0;
1954 perm = kmalloc(sizeof(*perm) + pool_ns_len + 1, GFP_NOFS);
1962 perm->pool_ns_len = pool_ns_len;
1963 if (pool_ns_len > 0)
1964 memcpy(perm->pool_ns, pool_ns->str, pool_ns_len);
1965 perm->pool_ns[pool_ns_len] = 0;
1967 rb_link_node(&perm->node, parent, p);
1968 rb_insert_color(&perm->node, &mdsc->pool_perm_tree);
1971 up_write(&mdsc->pool_perm_rwsem);
1973 ceph_osdc_put_request(rd_req);
1974 ceph_osdc_put_request(wr_req);
1979 dout("__ceph_pool_perm_get pool %lld ns %.*s result = %d\n",
1980 pool, (int)pool_ns->len, pool_ns->str, err);
1982 dout("__ceph_pool_perm_get pool %lld result = %d\n", pool, err);
1986 int ceph_pool_perm_check(struct inode *inode, int need)
1988 struct ceph_inode_info *ci = ceph_inode(inode);
1989 struct ceph_string *pool_ns;
1993 /* Only need to do this for regular files */
1994 if (!S_ISREG(inode->i_mode))
1997 if (ci->i_vino.snap != CEPH_NOSNAP) {
1999 * Pool permission check needs to write to the first object.
2000 * But for snapshot, head of the first object may have alread
2001 * been deleted. Skip check to avoid creating orphan object.
2006 if (ceph_test_mount_opt(ceph_inode_to_client(inode),
2010 spin_lock(&ci->i_ceph_lock);
2011 flags = ci->i_ceph_flags;
2012 pool = ci->i_layout.pool_id;
2013 spin_unlock(&ci->i_ceph_lock);
2015 if (flags & CEPH_I_POOL_PERM) {
2016 if ((need & CEPH_CAP_FILE_RD) && !(flags & CEPH_I_POOL_RD)) {
2017 dout("ceph_pool_perm_check pool %lld no read perm\n",
2021 if ((need & CEPH_CAP_FILE_WR) && !(flags & CEPH_I_POOL_WR)) {
2022 dout("ceph_pool_perm_check pool %lld no write perm\n",
2029 pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
2030 ret = __ceph_pool_perm_get(ci, pool, pool_ns);
2031 ceph_put_string(pool_ns);
2035 flags = CEPH_I_POOL_PERM;
2036 if (ret & POOL_READ)
2037 flags |= CEPH_I_POOL_RD;
2038 if (ret & POOL_WRITE)
2039 flags |= CEPH_I_POOL_WR;
2041 spin_lock(&ci->i_ceph_lock);
2042 if (pool == ci->i_layout.pool_id &&
2043 pool_ns == rcu_dereference_raw(ci->i_layout.pool_ns)) {
2044 ci->i_ceph_flags |= flags;
2046 pool = ci->i_layout.pool_id;
2047 flags = ci->i_ceph_flags;
2049 spin_unlock(&ci->i_ceph_lock);
2053 void ceph_pool_perm_destroy(struct ceph_mds_client *mdsc)
2055 struct ceph_pool_perm *perm;
2058 while (!RB_EMPTY_ROOT(&mdsc->pool_perm_tree)) {
2059 n = rb_first(&mdsc->pool_perm_tree);
2060 perm = rb_entry(n, struct ceph_pool_perm, node);
2061 rb_erase(n, &mdsc->pool_perm_tree);