1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
4 #include <linux/backing-dev.h>
7 #include <linux/pagemap.h>
8 #include <linux/writeback.h> /* generic_writepages */
9 #include <linux/slab.h>
10 #include <linux/pagevec.h>
11 #include <linux/task_io_accounting_ops.h>
12 #include <linux/signal.h>
13 #include <linux/iversion.h>
14 #include <linux/ktime.h>
15 #include <linux/netfs.h>
18 #include "mds_client.h"
21 #include <linux/ceph/osd_client.h>
22 #include <linux/ceph/striper.h>
25 * Ceph address space ops.
27 * There are a few funny things going on here.
29 * The page->private field is used to reference a struct
30 * ceph_snap_context for _every_ dirty page. This indicates which
31 * snapshot the page was logically dirtied in, and thus which snap
32 * context needs to be associated with the osd write during writeback.
34 * Similarly, struct ceph_inode_info maintains a set of counters to
35 * count dirty pages on the inode. In the absence of snapshots,
36 * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count.
38 * When a snapshot is taken (that is, when the client receives
39 * notification that a snapshot was taken), each inode with caps and
40 * with dirty pages (dirty pages implies there is a cap) gets a new
41 * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending
42 * order, new snaps go to the tail). The i_wrbuffer_ref_head count is
43 * moved to capsnap->dirty. (Unless a sync write is currently in
44 * progress. In that case, the capsnap is said to be "pending", new
45 * writes cannot start, and the capsnap isn't "finalized" until the
46 * write completes (or fails) and a final size/mtime for the inode for
47 * that snap can be settled upon.) i_wrbuffer_ref_head is reset to 0.
49 * On writeback, we must submit writes to the osd IN SNAP ORDER. So,
50 * we look for the first capsnap in i_cap_snaps and write out pages in
51 * that snap context _only_. Then we move on to the next capsnap,
52 * eventually reaching the "live" or "head" context (i.e., pages that
53 * are not yet snapped) and are writing the most recently dirtied
56 * Invalidate and so forth must take care to ensure the dirty page
57 * accounting is preserved.
60 #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10))
61 #define CONGESTION_OFF_THRESH(congestion_kb) \
62 (CONGESTION_ON_THRESH(congestion_kb) - \
63 (CONGESTION_ON_THRESH(congestion_kb) >> 2))
65 static int ceph_netfs_check_write_begin(struct file *file, loff_t pos, unsigned int len,
66 struct page *page, void **_fsdata);
68 static inline struct ceph_snap_context *page_snap_context(struct page *page)
70 if (PagePrivate(page))
71 return (void *)page->private;
76 * Dirty a page. Optimistically adjust accounting, on the assumption
77 * that we won't race with invalidate. If we do, readjust.
79 static int ceph_set_page_dirty(struct page *page)
81 struct address_space *mapping = page->mapping;
83 struct ceph_inode_info *ci;
84 struct ceph_snap_context *snapc;
87 if (unlikely(!mapping))
88 return !TestSetPageDirty(page);
90 if (PageDirty(page)) {
91 dout("%p set_page_dirty %p idx %lu -- already dirty\n",
92 mapping->host, page, page->index);
93 BUG_ON(!PagePrivate(page));
97 inode = mapping->host;
98 ci = ceph_inode(inode);
101 spin_lock(&ci->i_ceph_lock);
102 BUG_ON(ci->i_wr_ref == 0); // caller should hold Fw reference
103 if (__ceph_have_pending_cap_snap(ci)) {
104 struct ceph_cap_snap *capsnap =
105 list_last_entry(&ci->i_cap_snaps,
106 struct ceph_cap_snap,
108 snapc = ceph_get_snap_context(capsnap->context);
109 capsnap->dirty_pages++;
111 BUG_ON(!ci->i_head_snapc);
112 snapc = ceph_get_snap_context(ci->i_head_snapc);
113 ++ci->i_wrbuffer_ref_head;
115 if (ci->i_wrbuffer_ref == 0)
117 ++ci->i_wrbuffer_ref;
118 dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d "
119 "snapc %p seq %lld (%d snaps)\n",
120 mapping->host, page, page->index,
121 ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1,
122 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
123 snapc, snapc->seq, snapc->num_snaps);
124 spin_unlock(&ci->i_ceph_lock);
127 * Reference snap context in page->private. Also set
128 * PagePrivate so that we get invalidatepage callback.
130 BUG_ON(PagePrivate(page));
131 attach_page_private(page, snapc);
133 ret = __set_page_dirty_nobuffers(page);
134 WARN_ON(!PageLocked(page));
135 WARN_ON(!page->mapping);
141 * If we are truncating the full page (i.e. offset == 0), adjust the
142 * dirty page counters appropriately. Only called if there is private
145 static void ceph_invalidatepage(struct page *page, unsigned int offset,
149 struct ceph_inode_info *ci;
150 struct ceph_snap_context *snapc;
152 wait_on_page_fscache(page);
154 inode = page->mapping->host;
155 ci = ceph_inode(inode);
157 if (offset != 0 || length != thp_size(page)) {
158 dout("%p invalidatepage %p idx %lu partial dirty page %u~%u\n",
159 inode, page, page->index, offset, length);
163 WARN_ON(!PageLocked(page));
164 if (!PagePrivate(page))
167 dout("%p invalidatepage %p idx %lu full dirty page\n",
168 inode, page, page->index);
170 snapc = detach_page_private(page);
171 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
172 ceph_put_snap_context(snapc);
175 static int ceph_releasepage(struct page *page, gfp_t gfp)
177 dout("%p releasepage %p idx %lu (%sdirty)\n", page->mapping->host,
178 page, page->index, PageDirty(page) ? "" : "not ");
180 if (PageFsCache(page)) {
181 if (!(gfp & __GFP_DIRECT_RECLAIM) || !(gfp & __GFP_FS))
183 wait_on_page_fscache(page);
185 return !PagePrivate(page);
188 static void ceph_netfs_expand_readahead(struct netfs_read_request *rreq)
190 struct inode *inode = rreq->mapping->host;
191 struct ceph_inode_info *ci = ceph_inode(inode);
192 struct ceph_file_layout *lo = &ci->i_layout;
196 /* Expand the start downward */
197 blockno = div_u64_rem(rreq->start, lo->stripe_unit, &blockoff);
198 rreq->start = blockno * lo->stripe_unit;
199 rreq->len += blockoff;
201 /* Now, round up the length to the next block */
202 rreq->len = roundup(rreq->len, lo->stripe_unit);
205 static bool ceph_netfs_clamp_length(struct netfs_read_subrequest *subreq)
207 struct inode *inode = subreq->rreq->mapping->host;
208 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
209 struct ceph_inode_info *ci = ceph_inode(inode);
213 /* Truncate the extent at the end of the current block */
214 ceph_calc_file_object_mapping(&ci->i_layout, subreq->start, subreq->len,
215 &objno, &objoff, &xlen);
216 subreq->len = min(xlen, fsc->mount_options->rsize);
220 static void finish_netfs_read(struct ceph_osd_request *req)
222 struct ceph_fs_client *fsc = ceph_inode_to_client(req->r_inode);
223 struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0);
224 struct netfs_read_subrequest *subreq = req->r_priv;
226 int err = req->r_result;
228 ceph_update_read_metrics(&fsc->mdsc->metric, req->r_start_latency,
229 req->r_end_latency, err);
231 dout("%s: result %d subreq->len=%zu i_size=%lld\n", __func__, req->r_result,
232 subreq->len, i_size_read(req->r_inode));
234 /* no object means success but no data */
237 else if (err == -EBLOCKLISTED)
238 fsc->blocklisted = true;
240 if (err >= 0 && err < subreq->len)
241 __set_bit(NETFS_SREQ_CLEAR_TAIL, &subreq->flags);
243 netfs_subreq_terminated(subreq, err, true);
245 num_pages = calc_pages_for(osd_data->alignment, osd_data->length);
246 ceph_put_page_vector(osd_data->pages, num_pages, false);
250 static void ceph_netfs_issue_op(struct netfs_read_subrequest *subreq)
252 struct netfs_read_request *rreq = subreq->rreq;
253 struct inode *inode = rreq->mapping->host;
254 struct ceph_inode_info *ci = ceph_inode(inode);
255 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
256 struct ceph_osd_request *req;
257 struct ceph_vino vino = ceph_vino(inode);
258 struct iov_iter iter;
262 u64 len = subreq->len;
264 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, vino, subreq->start, &len,
265 0, 1, CEPH_OSD_OP_READ,
266 CEPH_OSD_FLAG_READ | fsc->client->osdc.client->options->read_from_replica,
267 NULL, ci->i_truncate_seq, ci->i_truncate_size, false);
274 dout("%s: pos=%llu orig_len=%zu len=%llu\n", __func__, subreq->start, subreq->len, len);
275 iov_iter_xarray(&iter, READ, &rreq->mapping->i_pages, subreq->start, len);
276 err = iov_iter_get_pages_alloc(&iter, &pages, len, &page_off);
278 dout("%s: iov_ter_get_pages_alloc returned %d\n", __func__, err);
282 /* should always give us a page-aligned read */
283 WARN_ON_ONCE(page_off);
286 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false);
287 req->r_callback = finish_netfs_read;
288 req->r_priv = subreq;
289 req->r_inode = inode;
292 err = ceph_osdc_start_request(req->r_osdc, req, false);
296 ceph_osdc_put_request(req);
298 netfs_subreq_terminated(subreq, err, false);
299 dout("%s: result %d\n", __func__, err);
302 static void ceph_init_rreq(struct netfs_read_request *rreq, struct file *file)
306 static void ceph_readahead_cleanup(struct address_space *mapping, void *priv)
308 struct inode *inode = mapping->host;
309 struct ceph_inode_info *ci = ceph_inode(inode);
310 int got = (uintptr_t)priv;
313 ceph_put_cap_refs(ci, got);
316 const struct netfs_read_request_ops ceph_netfs_read_ops = {
317 .init_rreq = ceph_init_rreq,
318 .is_cache_enabled = ceph_is_cache_enabled,
319 .begin_cache_operation = ceph_begin_cache_operation,
320 .issue_op = ceph_netfs_issue_op,
321 .expand_readahead = ceph_netfs_expand_readahead,
322 .clamp_length = ceph_netfs_clamp_length,
323 .check_write_begin = ceph_netfs_check_write_begin,
324 .cleanup = ceph_readahead_cleanup,
327 /* read a single page, without unlocking it. */
328 static int ceph_readpage(struct file *file, struct page *page)
330 struct inode *inode = file_inode(file);
331 struct ceph_inode_info *ci = ceph_inode(inode);
332 struct ceph_vino vino = ceph_vino(inode);
333 u64 off = page_offset(page);
334 u64 len = thp_size(page);
336 if (ci->i_inline_version != CEPH_INLINE_NONE) {
338 * Uptodate inline data should have been added
339 * into page cache while getting Fcr caps.
345 zero_user_segment(page, 0, thp_size(page));
346 SetPageUptodate(page);
351 dout("readpage ino %llx.%llx file %p off %llu len %llu page %p index %lu\n",
352 vino.ino, vino.snap, file, off, len, page, page->index);
354 return netfs_readpage(file, page, &ceph_netfs_read_ops, NULL);
357 static void ceph_readahead(struct readahead_control *ractl)
359 struct inode *inode = file_inode(ractl->file);
360 struct ceph_file_info *fi = ractl->file->private_data;
361 struct ceph_rw_context *rw_ctx;
365 if (ceph_inode(inode)->i_inline_version != CEPH_INLINE_NONE)
368 rw_ctx = ceph_find_rw_context(fi);
371 * readahead callers do not necessarily hold Fcb caps
372 * (e.g. fadvise, madvise).
374 int want = CEPH_CAP_FILE_CACHE;
376 ret = ceph_try_get_caps(inode, CEPH_CAP_FILE_RD, want, true, &got);
378 dout("start_read %p, error getting cap\n", inode);
379 else if (!(got & want))
380 dout("start_read %p, no cache cap\n", inode);
385 netfs_readahead(ractl, &ceph_netfs_read_ops, (void *)(uintptr_t)got);
388 struct ceph_writeback_ctl
398 * Get ref for the oldest snapc for an inode with dirty data... that is, the
399 * only snap context we are allowed to write back.
401 static struct ceph_snap_context *
402 get_oldest_context(struct inode *inode, struct ceph_writeback_ctl *ctl,
403 struct ceph_snap_context *page_snapc)
405 struct ceph_inode_info *ci = ceph_inode(inode);
406 struct ceph_snap_context *snapc = NULL;
407 struct ceph_cap_snap *capsnap = NULL;
409 spin_lock(&ci->i_ceph_lock);
410 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
411 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap,
412 capsnap->context, capsnap->dirty_pages);
413 if (!capsnap->dirty_pages)
416 /* get i_size, truncate_{seq,size} for page_snapc? */
417 if (snapc && capsnap->context != page_snapc)
421 if (capsnap->writing) {
422 ctl->i_size = i_size_read(inode);
423 ctl->size_stable = false;
425 ctl->i_size = capsnap->size;
426 ctl->size_stable = true;
428 ctl->truncate_size = capsnap->truncate_size;
429 ctl->truncate_seq = capsnap->truncate_seq;
430 ctl->head_snapc = false;
436 snapc = ceph_get_snap_context(capsnap->context);
438 page_snapc == snapc ||
439 page_snapc->seq > snapc->seq)
442 if (!snapc && ci->i_wrbuffer_ref_head) {
443 snapc = ceph_get_snap_context(ci->i_head_snapc);
444 dout(" head snapc %p has %d dirty pages\n",
445 snapc, ci->i_wrbuffer_ref_head);
447 ctl->i_size = i_size_read(inode);
448 ctl->truncate_size = ci->i_truncate_size;
449 ctl->truncate_seq = ci->i_truncate_seq;
450 ctl->size_stable = false;
451 ctl->head_snapc = true;
454 spin_unlock(&ci->i_ceph_lock);
458 static u64 get_writepages_data_length(struct inode *inode,
459 struct page *page, u64 start)
461 struct ceph_inode_info *ci = ceph_inode(inode);
462 struct ceph_snap_context *snapc = page_snap_context(page);
463 struct ceph_cap_snap *capsnap = NULL;
464 u64 end = i_size_read(inode);
466 if (snapc != ci->i_head_snapc) {
468 spin_lock(&ci->i_ceph_lock);
469 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
470 if (capsnap->context == snapc) {
471 if (!capsnap->writing)
477 spin_unlock(&ci->i_ceph_lock);
480 if (end > page_offset(page) + thp_size(page))
481 end = page_offset(page) + thp_size(page);
482 return end > start ? end - start : 0;
486 * Write a single page, but leave the page locked.
488 * If we get a write error, mark the mapping for error, but still adjust the
489 * dirty page accounting (i.e., page is no longer dirty).
491 static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
493 struct inode *inode = page->mapping->host;
494 struct ceph_inode_info *ci = ceph_inode(inode);
495 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
496 struct ceph_snap_context *snapc, *oldest;
497 loff_t page_off = page_offset(page);
499 loff_t len = thp_size(page);
500 struct ceph_writeback_ctl ceph_wbc;
501 struct ceph_osd_client *osdc = &fsc->client->osdc;
502 struct ceph_osd_request *req;
504 dout("writepage %p idx %lu\n", page, page->index);
506 /* verify this is a writeable snap context */
507 snapc = page_snap_context(page);
509 dout("writepage %p page %p not dirty?\n", inode, page);
512 oldest = get_oldest_context(inode, &ceph_wbc, snapc);
513 if (snapc->seq > oldest->seq) {
514 dout("writepage %p page %p snapc %p not writeable - noop\n",
516 /* we should only noop if called by kswapd */
517 WARN_ON(!(current->flags & PF_MEMALLOC));
518 ceph_put_snap_context(oldest);
519 redirty_page_for_writepage(wbc, page);
522 ceph_put_snap_context(oldest);
524 /* is this a partial page at end of file? */
525 if (page_off >= ceph_wbc.i_size) {
526 dout("%p page eof %llu\n", page, ceph_wbc.i_size);
527 page->mapping->a_ops->invalidatepage(page, 0, thp_size(page));
531 if (ceph_wbc.i_size < page_off + len)
532 len = ceph_wbc.i_size - page_off;
534 dout("writepage %p page %p index %lu on %llu~%llu snapc %p seq %lld\n",
535 inode, page, page->index, page_off, len, snapc, snapc->seq);
537 if (atomic_long_inc_return(&fsc->writeback_count) >
538 CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb))
539 set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC);
541 set_page_writeback(page);
542 req = ceph_osdc_new_request(osdc, &ci->i_layout, ceph_vino(inode), page_off, &len, 0, 1,
543 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE, snapc,
544 ceph_wbc.truncate_seq, ceph_wbc.truncate_size,
547 redirty_page_for_writepage(wbc, page);
548 end_page_writeback(page);
552 /* it may be a short write due to an object boundary */
553 WARN_ON_ONCE(len > thp_size(page));
554 osd_req_op_extent_osd_data_pages(req, 0, &page, len, 0, false, false);
555 dout("writepage %llu~%llu (%llu bytes)\n", page_off, len, len);
557 req->r_mtime = inode->i_mtime;
558 err = ceph_osdc_start_request(osdc, req, true);
560 err = ceph_osdc_wait_request(osdc, req);
562 ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
563 req->r_end_latency, err);
565 ceph_osdc_put_request(req);
570 struct writeback_control tmp_wbc;
573 if (err == -ERESTARTSYS) {
574 /* killed by SIGKILL */
575 dout("writepage interrupted page %p\n", page);
576 redirty_page_for_writepage(wbc, page);
577 end_page_writeback(page);
580 if (err == -EBLOCKLISTED)
581 fsc->blocklisted = true;
582 dout("writepage setting page/mapping error %d %p\n",
584 mapping_set_error(&inode->i_data, err);
585 wbc->pages_skipped++;
587 dout("writepage cleaned page %p\n", page);
588 err = 0; /* vfs expects us to return 0 */
590 oldest = detach_page_private(page);
591 WARN_ON_ONCE(oldest != snapc);
592 end_page_writeback(page);
593 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
594 ceph_put_snap_context(snapc); /* page's reference */
596 if (atomic_long_dec_return(&fsc->writeback_count) <
597 CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb))
598 clear_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC);
603 static int ceph_writepage(struct page *page, struct writeback_control *wbc)
606 struct inode *inode = page->mapping->host;
609 err = writepage_nounlock(page, wbc);
610 if (err == -ERESTARTSYS) {
611 /* direct memory reclaimer was killed by SIGKILL. return 0
612 * to prevent caller from setting mapping/page error */
621 * async writeback completion handler.
623 * If we get an error, set the mapping error bit, but not the individual
626 static void writepages_finish(struct ceph_osd_request *req)
628 struct inode *inode = req->r_inode;
629 struct ceph_inode_info *ci = ceph_inode(inode);
630 struct ceph_osd_data *osd_data;
632 int num_pages, total_pages = 0;
634 int rc = req->r_result;
635 struct ceph_snap_context *snapc = req->r_snapc;
636 struct address_space *mapping = inode->i_mapping;
637 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
640 dout("writepages_finish %p rc %d\n", inode, rc);
642 mapping_set_error(mapping, rc);
643 ceph_set_error_write(ci);
644 if (rc == -EBLOCKLISTED)
645 fsc->blocklisted = true;
647 ceph_clear_error_write(ci);
650 ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
651 req->r_end_latency, rc);
654 * We lost the cache cap, need to truncate the page before
655 * it is unlocked, otherwise we'd truncate it later in the
656 * page truncation thread, possibly losing some data that
659 remove_page = !(ceph_caps_issued(ci) &
660 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO));
662 /* clean all pages */
663 for (i = 0; i < req->r_num_ops; i++) {
664 if (req->r_ops[i].op != CEPH_OSD_OP_WRITE)
667 osd_data = osd_req_op_extent_osd_data(req, i);
668 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
669 num_pages = calc_pages_for((u64)osd_data->alignment,
670 (u64)osd_data->length);
671 total_pages += num_pages;
672 for (j = 0; j < num_pages; j++) {
673 page = osd_data->pages[j];
675 WARN_ON(!PageUptodate(page));
677 if (atomic_long_dec_return(&fsc->writeback_count) <
678 CONGESTION_OFF_THRESH(
679 fsc->mount_options->congestion_kb))
680 clear_bdi_congested(inode_to_bdi(inode),
683 ceph_put_snap_context(detach_page_private(page));
684 end_page_writeback(page);
685 dout("unlocking %p\n", page);
688 generic_error_remove_page(inode->i_mapping,
693 dout("writepages_finish %p wrote %llu bytes cleaned %d pages\n",
694 inode, osd_data->length, rc >= 0 ? num_pages : 0);
696 release_pages(osd_data->pages, num_pages);
699 ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc);
701 osd_data = osd_req_op_extent_osd_data(req, 0);
702 if (osd_data->pages_from_pool)
703 mempool_free(osd_data->pages, ceph_wb_pagevec_pool);
705 kfree(osd_data->pages);
706 ceph_osdc_put_request(req);
710 * initiate async writeback
712 static int ceph_writepages_start(struct address_space *mapping,
713 struct writeback_control *wbc)
715 struct inode *inode = mapping->host;
716 struct ceph_inode_info *ci = ceph_inode(inode);
717 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
718 struct ceph_vino vino = ceph_vino(inode);
719 pgoff_t index, start_index, end = -1;
720 struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc;
723 unsigned int wsize = i_blocksize(inode);
724 struct ceph_osd_request *req = NULL;
725 struct ceph_writeback_ctl ceph_wbc;
726 bool should_loop, range_whole = false;
729 dout("writepages_start %p (mode=%s)\n", inode,
730 wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
731 (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
733 if (READ_ONCE(fsc->mount_state) >= CEPH_MOUNT_SHUTDOWN) {
734 if (ci->i_wrbuffer_ref > 0) {
736 "writepage_start %p %lld forced umount\n",
737 inode, ceph_ino(inode));
739 mapping_set_error(mapping, -EIO);
740 return -EIO; /* we're in a forced umount, don't write! */
742 if (fsc->mount_options->wsize < wsize)
743 wsize = fsc->mount_options->wsize;
747 start_index = wbc->range_cyclic ? mapping->writeback_index : 0;
751 /* find oldest snap context with dirty data */
752 snapc = get_oldest_context(inode, &ceph_wbc, NULL);
754 /* hmm, why does writepages get called when there
756 dout(" no snap context with dirty data?\n");
759 dout(" oldest snapc is %p seq %lld (%d snaps)\n",
760 snapc, snapc->seq, snapc->num_snaps);
763 if (ceph_wbc.head_snapc && snapc != last_snapc) {
764 /* where to start/end? */
765 if (wbc->range_cyclic) {
770 dout(" cyclic, start at %lu\n", index);
772 index = wbc->range_start >> PAGE_SHIFT;
773 end = wbc->range_end >> PAGE_SHIFT;
774 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
776 dout(" not cyclic, %lu to %lu\n", index, end);
778 } else if (!ceph_wbc.head_snapc) {
779 /* Do not respect wbc->range_{start,end}. Dirty pages
780 * in that range can be associated with newer snapc.
781 * They are not writeable until we write all dirty pages
782 * associated with 'snapc' get written */
785 dout(" non-head snapc, range whole\n");
788 ceph_put_snap_context(last_snapc);
791 while (!done && index <= end) {
792 int num_ops = 0, op_idx;
793 unsigned i, pvec_pages, max_pages, locked_pages = 0;
794 struct page **pages = NULL, **data_pages;
796 pgoff_t strip_unit_end = 0;
797 u64 offset = 0, len = 0;
798 bool from_pool = false;
800 max_pages = wsize >> PAGE_SHIFT;
803 pvec_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
804 end, PAGECACHE_TAG_DIRTY);
805 dout("pagevec_lookup_range_tag got %d\n", pvec_pages);
806 if (!pvec_pages && !locked_pages)
808 for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) {
809 page = pvec.pages[i];
810 dout("? %p idx %lu\n", page, page->index);
811 if (locked_pages == 0)
812 lock_page(page); /* first page */
813 else if (!trylock_page(page))
816 /* only dirty pages, or our accounting breaks */
817 if (unlikely(!PageDirty(page)) ||
818 unlikely(page->mapping != mapping)) {
819 dout("!dirty or !mapping %p\n", page);
823 /* only if matching snap context */
824 pgsnapc = page_snap_context(page);
825 if (pgsnapc != snapc) {
826 dout("page snapc %p %lld != oldest %p %lld\n",
827 pgsnapc, pgsnapc->seq, snapc, snapc->seq);
829 !ceph_wbc.head_snapc &&
830 wbc->sync_mode != WB_SYNC_NONE)
835 if (page_offset(page) >= ceph_wbc.i_size) {
836 dout("%p page eof %llu\n",
837 page, ceph_wbc.i_size);
838 if ((ceph_wbc.size_stable ||
839 page_offset(page) >= i_size_read(inode)) &&
840 clear_page_dirty_for_io(page))
841 mapping->a_ops->invalidatepage(page,
846 if (strip_unit_end && (page->index > strip_unit_end)) {
847 dout("end of strip unit %p\n", page);
851 if (PageWriteback(page)) {
852 if (wbc->sync_mode == WB_SYNC_NONE) {
853 dout("%p under writeback\n", page);
857 dout("waiting on writeback %p\n", page);
858 wait_on_page_writeback(page);
861 if (!clear_page_dirty_for_io(page)) {
862 dout("%p !clear_page_dirty_for_io\n", page);
868 * We have something to write. If this is
869 * the first locked page this time through,
870 * calculate max possinle write size and
871 * allocate a page array
873 if (locked_pages == 0) {
878 /* prepare async write request */
879 offset = (u64)page_offset(page);
880 ceph_calc_file_object_mapping(&ci->i_layout,
887 strip_unit_end = page->index +
888 ((len - 1) >> PAGE_SHIFT);
891 max_pages = calc_pages_for(0, (u64)len);
892 pages = kmalloc_array(max_pages,
897 pages = mempool_alloc(ceph_wb_pagevec_pool, GFP_NOFS);
902 } else if (page->index !=
903 (offset + len) >> PAGE_SHIFT) {
904 if (num_ops >= (from_pool ? CEPH_OSD_SLAB_OPS :
906 redirty_page_for_writepage(wbc, page);
912 offset = (u64)page_offset(page);
916 /* note position of first page in pvec */
917 dout("%p will write page %p idx %lu\n",
918 inode, page, page->index);
920 if (atomic_long_inc_return(&fsc->writeback_count) >
921 CONGESTION_ON_THRESH(
922 fsc->mount_options->congestion_kb)) {
923 set_bdi_congested(inode_to_bdi(inode),
928 pages[locked_pages++] = page;
929 pvec.pages[i] = NULL;
931 len += thp_size(page);
934 /* did we get anything? */
936 goto release_pvec_pages;
939 /* shift unused page to beginning of pvec */
940 for (j = 0; j < pvec_pages; j++) {
944 pvec.pages[n] = pvec.pages[j];
949 if (pvec_pages && i == pvec_pages &&
950 locked_pages < max_pages) {
951 dout("reached end pvec, trying for more\n");
952 pagevec_release(&pvec);
958 offset = page_offset(pages[0]);
961 req = ceph_osdc_new_request(&fsc->client->osdc,
963 offset, &len, 0, num_ops,
964 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
965 snapc, ceph_wbc.truncate_seq,
966 ceph_wbc.truncate_size, false);
968 req = ceph_osdc_new_request(&fsc->client->osdc,
975 snapc, ceph_wbc.truncate_seq,
976 ceph_wbc.truncate_size, true);
979 BUG_ON(len < page_offset(pages[locked_pages - 1]) +
980 thp_size(page) - offset);
982 req->r_callback = writepages_finish;
983 req->r_inode = inode;
985 /* Format the osd request message and submit the write */
989 for (i = 0; i < locked_pages; i++) {
990 u64 cur_offset = page_offset(pages[i]);
991 if (offset + len != cur_offset) {
992 if (op_idx + 1 == req->r_num_ops)
994 osd_req_op_extent_dup_last(req, op_idx,
995 cur_offset - offset);
996 dout("writepages got pages at %llu~%llu\n",
998 osd_req_op_extent_osd_data_pages(req, op_idx,
1001 osd_req_op_extent_update(req, op_idx, len);
1004 offset = cur_offset;
1005 data_pages = pages + i;
1009 set_page_writeback(pages[i]);
1010 len += thp_size(page);
1013 if (ceph_wbc.size_stable) {
1014 len = min(len, ceph_wbc.i_size - offset);
1015 } else if (i == locked_pages) {
1016 /* writepages_finish() clears writeback pages
1017 * according to the data length, so make sure
1018 * data length covers all locked pages */
1019 u64 min_len = len + 1 - thp_size(page);
1020 len = get_writepages_data_length(inode, pages[i - 1],
1022 len = max(len, min_len);
1024 dout("writepages got pages at %llu~%llu\n", offset, len);
1026 osd_req_op_extent_osd_data_pages(req, op_idx, data_pages, len,
1027 0, from_pool, false);
1028 osd_req_op_extent_update(req, op_idx, len);
1030 BUG_ON(op_idx + 1 != req->r_num_ops);
1033 if (i < locked_pages) {
1034 BUG_ON(num_ops <= req->r_num_ops);
1035 num_ops -= req->r_num_ops;
1038 /* allocate new pages array for next request */
1040 pages = kmalloc_array(locked_pages, sizeof(*pages),
1044 pages = mempool_alloc(ceph_wb_pagevec_pool, GFP_NOFS);
1047 memcpy(pages, data_pages + i,
1048 locked_pages * sizeof(*pages));
1049 memset(data_pages + i, 0,
1050 locked_pages * sizeof(*pages));
1052 BUG_ON(num_ops != req->r_num_ops);
1053 index = pages[i - 1]->index + 1;
1054 /* request message now owns the pages array */
1058 req->r_mtime = inode->i_mtime;
1059 rc = ceph_osdc_start_request(&fsc->client->osdc, req, true);
1063 wbc->nr_to_write -= i;
1068 * We stop writing back only if we are not doing
1069 * integrity sync. In case of integrity sync we have to
1070 * keep going until we have written all the pages
1071 * we tagged for writeback prior to entering this loop.
1073 if (wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE)
1077 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr,
1078 pvec.nr ? pvec.pages[0] : NULL);
1079 pagevec_release(&pvec);
1082 if (should_loop && !done) {
1083 /* more to do; loop back to beginning of file */
1084 dout("writepages looping back to beginning of file\n");
1085 end = start_index - 1; /* OK even when start_index == 0 */
1087 /* to write dirty pages associated with next snapc,
1088 * we need to wait until current writes complete */
1089 if (wbc->sync_mode != WB_SYNC_NONE &&
1090 start_index == 0 && /* all dirty pages were checked */
1091 !ceph_wbc.head_snapc) {
1095 while ((index <= end) &&
1096 (nr = pagevec_lookup_tag(&pvec, mapping, &index,
1097 PAGECACHE_TAG_WRITEBACK))) {
1098 for (i = 0; i < nr; i++) {
1099 page = pvec.pages[i];
1100 if (page_snap_context(page) != snapc)
1102 wait_on_page_writeback(page);
1104 pagevec_release(&pvec);
1114 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1115 mapping->writeback_index = index;
1118 ceph_osdc_put_request(req);
1119 ceph_put_snap_context(last_snapc);
1120 dout("writepages dend - startone, rc = %d\n", rc);
1127 * See if a given @snapc is either writeable, or already written.
1129 static int context_is_writeable_or_written(struct inode *inode,
1130 struct ceph_snap_context *snapc)
1132 struct ceph_snap_context *oldest = get_oldest_context(inode, NULL, NULL);
1133 int ret = !oldest || snapc->seq <= oldest->seq;
1135 ceph_put_snap_context(oldest);
1140 * ceph_find_incompatible - find an incompatible context and return it
1141 * @page: page being dirtied
1143 * We are only allowed to write into/dirty a page if the page is
1144 * clean, or already dirty within the same snap context. Returns a
1145 * conflicting context if there is one, NULL if there isn't, or a
1146 * negative error code on other errors.
1148 * Must be called with page lock held.
1150 static struct ceph_snap_context *
1151 ceph_find_incompatible(struct page *page)
1153 struct inode *inode = page->mapping->host;
1154 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1155 struct ceph_inode_info *ci = ceph_inode(inode);
1157 if (READ_ONCE(fsc->mount_state) >= CEPH_MOUNT_SHUTDOWN) {
1158 dout(" page %p forced umount\n", page);
1159 return ERR_PTR(-EIO);
1163 struct ceph_snap_context *snapc, *oldest;
1165 wait_on_page_writeback(page);
1167 snapc = page_snap_context(page);
1168 if (!snapc || snapc == ci->i_head_snapc)
1172 * this page is already dirty in another (older) snap
1173 * context! is it writeable now?
1175 oldest = get_oldest_context(inode, NULL, NULL);
1176 if (snapc->seq > oldest->seq) {
1177 /* not writeable -- return it for the caller to deal with */
1178 ceph_put_snap_context(oldest);
1179 dout(" page %p snapc %p not current or oldest\n", page, snapc);
1180 return ceph_get_snap_context(snapc);
1182 ceph_put_snap_context(oldest);
1184 /* yay, writeable, do it now (without dropping page lock) */
1185 dout(" page %p snapc %p not current, but oldest\n", page, snapc);
1186 if (clear_page_dirty_for_io(page)) {
1187 int r = writepage_nounlock(page, NULL);
1195 static int ceph_netfs_check_write_begin(struct file *file, loff_t pos, unsigned int len,
1196 struct page *page, void **_fsdata)
1198 struct inode *inode = file_inode(file);
1199 struct ceph_inode_info *ci = ceph_inode(inode);
1200 struct ceph_snap_context *snapc;
1202 snapc = ceph_find_incompatible(page);
1209 return PTR_ERR(snapc);
1211 ceph_queue_writeback(inode);
1212 r = wait_event_killable(ci->i_cap_wq,
1213 context_is_writeable_or_written(inode, snapc));
1214 ceph_put_snap_context(snapc);
1215 return r == 0 ? -EAGAIN : r;
1221 * We are only allowed to write into/dirty the page if the page is
1222 * clean, or already dirty within the same snap context.
1224 static int ceph_write_begin(struct file *file, struct address_space *mapping,
1225 loff_t pos, unsigned len, unsigned flags,
1226 struct page **pagep, void **fsdata)
1228 struct inode *inode = file_inode(file);
1229 struct ceph_inode_info *ci = ceph_inode(inode);
1230 struct page *page = NULL;
1231 pgoff_t index = pos >> PAGE_SHIFT;
1235 * Uninlining should have already been done and everything updated, EXCEPT
1236 * for inline_version sent to the MDS.
1238 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1239 page = grab_cache_page_write_begin(mapping, index, flags);
1244 * The inline_version on a new inode is set to 1. If that's the
1245 * case, then the page is brand new and isn't yet Uptodate.
1248 if (index == 0 && ci->i_inline_version != 1) {
1249 if (!PageUptodate(page)) {
1250 WARN_ONCE(1, "ceph: write_begin called on still-inlined inode (inline_version %llu)!\n",
1251 ci->i_inline_version);
1256 zero_user_segment(page, 0, thp_size(page));
1257 SetPageUptodate(page);
1261 r = netfs_write_begin(file, inode->i_mapping, pos, len, 0, &page, NULL,
1262 &ceph_netfs_read_ops, NULL);
1265 wait_on_page_fscache(page);
1270 WARN_ON_ONCE(!PageLocked(page));
1277 * we don't do anything in here that simple_write_end doesn't do
1278 * except adjust dirty page accounting
1280 static int ceph_write_end(struct file *file, struct address_space *mapping,
1281 loff_t pos, unsigned len, unsigned copied,
1282 struct page *page, void *fsdata)
1284 struct inode *inode = file_inode(file);
1285 bool check_cap = false;
1287 dout("write_end file %p inode %p page %p %d~%d (%d)\n", file,
1288 inode, page, (int)pos, (int)copied, (int)len);
1290 /* zero the stale part of the page if we did a short copy */
1291 if (!PageUptodate(page)) {
1296 SetPageUptodate(page);
1299 /* did file size increase? */
1300 if (pos+copied > i_size_read(inode))
1301 check_cap = ceph_inode_set_size(inode, pos+copied);
1303 set_page_dirty(page);
1310 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL);
1316 * we set .direct_IO to indicate direct io is supported, but since we
1317 * intercept O_DIRECT reads and writes early, this function should
1320 static ssize_t ceph_direct_io(struct kiocb *iocb, struct iov_iter *iter)
1326 const struct address_space_operations ceph_aops = {
1327 .readpage = ceph_readpage,
1328 .readahead = ceph_readahead,
1329 .writepage = ceph_writepage,
1330 .writepages = ceph_writepages_start,
1331 .write_begin = ceph_write_begin,
1332 .write_end = ceph_write_end,
1333 .set_page_dirty = ceph_set_page_dirty,
1334 .invalidatepage = ceph_invalidatepage,
1335 .releasepage = ceph_releasepage,
1336 .direct_IO = ceph_direct_io,
1339 static void ceph_block_sigs(sigset_t *oldset)
1342 siginitsetinv(&mask, sigmask(SIGKILL));
1343 sigprocmask(SIG_BLOCK, &mask, oldset);
1346 static void ceph_restore_sigs(sigset_t *oldset)
1348 sigprocmask(SIG_SETMASK, oldset, NULL);
1354 static vm_fault_t ceph_filemap_fault(struct vm_fault *vmf)
1356 struct vm_area_struct *vma = vmf->vma;
1357 struct inode *inode = file_inode(vma->vm_file);
1358 struct ceph_inode_info *ci = ceph_inode(inode);
1359 struct ceph_file_info *fi = vma->vm_file->private_data;
1360 loff_t off = (loff_t)vmf->pgoff << PAGE_SHIFT;
1363 vm_fault_t ret = VM_FAULT_SIGBUS;
1365 ceph_block_sigs(&oldset);
1367 dout("filemap_fault %p %llx.%llx %llu trying to get caps\n",
1368 inode, ceph_vinop(inode), off);
1369 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1370 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1372 want = CEPH_CAP_FILE_CACHE;
1375 err = ceph_get_caps(vma->vm_file, CEPH_CAP_FILE_RD, want, -1, &got);
1379 dout("filemap_fault %p %llu got cap refs on %s\n",
1380 inode, off, ceph_cap_string(got));
1382 if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) ||
1383 ci->i_inline_version == CEPH_INLINE_NONE) {
1384 CEPH_DEFINE_RW_CONTEXT(rw_ctx, got);
1385 ceph_add_rw_context(fi, &rw_ctx);
1386 ret = filemap_fault(vmf);
1387 ceph_del_rw_context(fi, &rw_ctx);
1388 dout("filemap_fault %p %llu drop cap refs %s ret %x\n",
1389 inode, off, ceph_cap_string(got), ret);
1393 ceph_put_cap_refs(ci, got);
1398 /* read inline data */
1399 if (off >= PAGE_SIZE) {
1400 /* does not support inline data > PAGE_SIZE */
1401 ret = VM_FAULT_SIGBUS;
1403 struct address_space *mapping = inode->i_mapping;
1404 struct page *page = find_or_create_page(mapping, 0,
1405 mapping_gfp_constraint(mapping,
1411 err = __ceph_do_getattr(inode, page,
1412 CEPH_STAT_CAP_INLINE_DATA, true);
1413 if (err < 0 || off >= i_size_read(inode)) {
1416 ret = vmf_error(err);
1419 if (err < PAGE_SIZE)
1420 zero_user_segment(page, err, PAGE_SIZE);
1422 flush_dcache_page(page);
1423 SetPageUptodate(page);
1425 ret = VM_FAULT_MAJOR | VM_FAULT_LOCKED;
1427 dout("filemap_fault %p %llu read inline data ret %x\n",
1431 ceph_restore_sigs(&oldset);
1433 ret = vmf_error(err);
1438 static vm_fault_t ceph_page_mkwrite(struct vm_fault *vmf)
1440 struct vm_area_struct *vma = vmf->vma;
1441 struct inode *inode = file_inode(vma->vm_file);
1442 struct ceph_inode_info *ci = ceph_inode(inode);
1443 struct ceph_file_info *fi = vma->vm_file->private_data;
1444 struct ceph_cap_flush *prealloc_cf;
1445 struct page *page = vmf->page;
1446 loff_t off = page_offset(page);
1447 loff_t size = i_size_read(inode);
1451 vm_fault_t ret = VM_FAULT_SIGBUS;
1453 prealloc_cf = ceph_alloc_cap_flush();
1455 return VM_FAULT_OOM;
1457 sb_start_pagefault(inode->i_sb);
1458 ceph_block_sigs(&oldset);
1460 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1461 struct page *locked_page = NULL;
1466 err = ceph_uninline_data(vma->vm_file, locked_page);
1468 unlock_page(locked_page);
1473 if (off + thp_size(page) <= size)
1474 len = thp_size(page);
1476 len = offset_in_thp(page, size);
1478 dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n",
1479 inode, ceph_vinop(inode), off, len, size);
1480 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1481 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1483 want = CEPH_CAP_FILE_BUFFER;
1486 err = ceph_get_caps(vma->vm_file, CEPH_CAP_FILE_WR, want, off + len, &got);
1490 dout("page_mkwrite %p %llu~%zd got cap refs on %s\n",
1491 inode, off, len, ceph_cap_string(got));
1493 /* Update time before taking page lock */
1494 file_update_time(vma->vm_file);
1495 inode_inc_iversion_raw(inode);
1498 struct ceph_snap_context *snapc;
1502 if (page_mkwrite_check_truncate(page, inode) < 0) {
1504 ret = VM_FAULT_NOPAGE;
1508 snapc = ceph_find_incompatible(page);
1510 /* success. we'll keep the page locked. */
1511 set_page_dirty(page);
1512 ret = VM_FAULT_LOCKED;
1518 if (IS_ERR(snapc)) {
1519 ret = VM_FAULT_SIGBUS;
1523 ceph_queue_writeback(inode);
1524 err = wait_event_killable(ci->i_cap_wq,
1525 context_is_writeable_or_written(inode, snapc));
1526 ceph_put_snap_context(snapc);
1529 if (ret == VM_FAULT_LOCKED ||
1530 ci->i_inline_version != CEPH_INLINE_NONE) {
1532 spin_lock(&ci->i_ceph_lock);
1533 ci->i_inline_version = CEPH_INLINE_NONE;
1534 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1536 spin_unlock(&ci->i_ceph_lock);
1538 __mark_inode_dirty(inode, dirty);
1541 dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %x\n",
1542 inode, off, len, ceph_cap_string(got), ret);
1543 ceph_put_cap_refs_async(ci, got);
1545 ceph_restore_sigs(&oldset);
1546 sb_end_pagefault(inode->i_sb);
1547 ceph_free_cap_flush(prealloc_cf);
1549 ret = vmf_error(err);
1553 void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
1554 char *data, size_t len)
1556 struct address_space *mapping = inode->i_mapping;
1562 if (i_size_read(inode) == 0)
1564 page = find_or_create_page(mapping, 0,
1565 mapping_gfp_constraint(mapping,
1569 if (PageUptodate(page)) {
1576 dout("fill_inline_data %p %llx.%llx len %zu locked_page %p\n",
1577 inode, ceph_vinop(inode), len, locked_page);
1580 void *kaddr = kmap_atomic(page);
1581 memcpy(kaddr, data, len);
1582 kunmap_atomic(kaddr);
1585 if (page != locked_page) {
1586 if (len < PAGE_SIZE)
1587 zero_user_segment(page, len, PAGE_SIZE);
1589 flush_dcache_page(page);
1591 SetPageUptodate(page);
1597 int ceph_uninline_data(struct file *filp, struct page *locked_page)
1599 struct inode *inode = file_inode(filp);
1600 struct ceph_inode_info *ci = ceph_inode(inode);
1601 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1602 struct ceph_osd_request *req;
1603 struct page *page = NULL;
1604 u64 len, inline_version;
1606 bool from_pagecache = false;
1608 spin_lock(&ci->i_ceph_lock);
1609 inline_version = ci->i_inline_version;
1610 spin_unlock(&ci->i_ceph_lock);
1612 dout("uninline_data %p %llx.%llx inline_version %llu\n",
1613 inode, ceph_vinop(inode), inline_version);
1615 if (inline_version == 1 || /* initial version, no data */
1616 inline_version == CEPH_INLINE_NONE)
1621 WARN_ON(!PageUptodate(page));
1622 } else if (ceph_caps_issued(ci) &
1623 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) {
1624 page = find_get_page(inode->i_mapping, 0);
1626 if (PageUptodate(page)) {
1627 from_pagecache = true;
1637 len = i_size_read(inode);
1638 if (len > PAGE_SIZE)
1641 page = __page_cache_alloc(GFP_NOFS);
1646 err = __ceph_do_getattr(inode, page,
1647 CEPH_STAT_CAP_INLINE_DATA, true);
1649 /* no inline data */
1650 if (err == -ENODATA)
1657 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1658 ceph_vino(inode), 0, &len, 0, 1,
1659 CEPH_OSD_OP_CREATE, CEPH_OSD_FLAG_WRITE,
1666 req->r_mtime = inode->i_mtime;
1667 err = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1669 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1670 ceph_osdc_put_request(req);
1674 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1675 ceph_vino(inode), 0, &len, 1, 3,
1676 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
1677 NULL, ci->i_truncate_seq,
1678 ci->i_truncate_size, false);
1684 osd_req_op_extent_osd_data_pages(req, 1, &page, len, 0, false, false);
1687 __le64 xattr_buf = cpu_to_le64(inline_version);
1688 err = osd_req_op_xattr_init(req, 0, CEPH_OSD_OP_CMPXATTR,
1689 "inline_version", &xattr_buf,
1691 CEPH_OSD_CMPXATTR_OP_GT,
1692 CEPH_OSD_CMPXATTR_MODE_U64);
1699 int xattr_len = snprintf(xattr_buf, sizeof(xattr_buf),
1700 "%llu", inline_version);
1701 err = osd_req_op_xattr_init(req, 2, CEPH_OSD_OP_SETXATTR,
1703 xattr_buf, xattr_len, 0, 0);
1708 req->r_mtime = inode->i_mtime;
1709 err = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1711 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1713 ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
1714 req->r_end_latency, err);
1717 ceph_osdc_put_request(req);
1718 if (err == -ECANCELED)
1721 if (page && page != locked_page) {
1722 if (from_pagecache) {
1726 __free_pages(page, 0);
1729 dout("uninline_data %p %llx.%llx inline_version %llu = %d\n",
1730 inode, ceph_vinop(inode), inline_version, err);
1734 static const struct vm_operations_struct ceph_vmops = {
1735 .fault = ceph_filemap_fault,
1736 .page_mkwrite = ceph_page_mkwrite,
1739 int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1741 struct address_space *mapping = file->f_mapping;
1743 if (!mapping->a_ops->readpage)
1745 file_accessed(file);
1746 vma->vm_ops = &ceph_vmops;
1755 static int __ceph_pool_perm_get(struct ceph_inode_info *ci,
1756 s64 pool, struct ceph_string *pool_ns)
1758 struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode);
1759 struct ceph_mds_client *mdsc = fsc->mdsc;
1760 struct ceph_osd_request *rd_req = NULL, *wr_req = NULL;
1761 struct rb_node **p, *parent;
1762 struct ceph_pool_perm *perm;
1763 struct page **pages;
1765 int err = 0, err2 = 0, have = 0;
1767 down_read(&mdsc->pool_perm_rwsem);
1768 p = &mdsc->pool_perm_tree.rb_node;
1770 perm = rb_entry(*p, struct ceph_pool_perm, node);
1771 if (pool < perm->pool)
1773 else if (pool > perm->pool)
1774 p = &(*p)->rb_right;
1776 int ret = ceph_compare_string(pool_ns,
1782 p = &(*p)->rb_right;
1789 up_read(&mdsc->pool_perm_rwsem);
1794 dout("__ceph_pool_perm_get pool %lld ns %.*s no perm cached\n",
1795 pool, (int)pool_ns->len, pool_ns->str);
1797 dout("__ceph_pool_perm_get pool %lld no perm cached\n", pool);
1799 down_write(&mdsc->pool_perm_rwsem);
1800 p = &mdsc->pool_perm_tree.rb_node;
1804 perm = rb_entry(parent, struct ceph_pool_perm, node);
1805 if (pool < perm->pool)
1807 else if (pool > perm->pool)
1808 p = &(*p)->rb_right;
1810 int ret = ceph_compare_string(pool_ns,
1816 p = &(*p)->rb_right;
1824 up_write(&mdsc->pool_perm_rwsem);
1828 rd_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1829 1, false, GFP_NOFS);
1835 rd_req->r_flags = CEPH_OSD_FLAG_READ;
1836 osd_req_op_init(rd_req, 0, CEPH_OSD_OP_STAT, 0);
1837 rd_req->r_base_oloc.pool = pool;
1839 rd_req->r_base_oloc.pool_ns = ceph_get_string(pool_ns);
1840 ceph_oid_printf(&rd_req->r_base_oid, "%llx.00000000", ci->i_vino.ino);
1842 err = ceph_osdc_alloc_messages(rd_req, GFP_NOFS);
1846 wr_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1847 1, false, GFP_NOFS);
1853 wr_req->r_flags = CEPH_OSD_FLAG_WRITE;
1854 osd_req_op_init(wr_req, 0, CEPH_OSD_OP_CREATE, CEPH_OSD_OP_FLAG_EXCL);
1855 ceph_oloc_copy(&wr_req->r_base_oloc, &rd_req->r_base_oloc);
1856 ceph_oid_copy(&wr_req->r_base_oid, &rd_req->r_base_oid);
1858 err = ceph_osdc_alloc_messages(wr_req, GFP_NOFS);
1862 /* one page should be large enough for STAT data */
1863 pages = ceph_alloc_page_vector(1, GFP_KERNEL);
1864 if (IS_ERR(pages)) {
1865 err = PTR_ERR(pages);
1869 osd_req_op_raw_data_in_pages(rd_req, 0, pages, PAGE_SIZE,
1871 err = ceph_osdc_start_request(&fsc->client->osdc, rd_req, false);
1873 wr_req->r_mtime = ci->vfs_inode.i_mtime;
1874 err2 = ceph_osdc_start_request(&fsc->client->osdc, wr_req, false);
1877 err = ceph_osdc_wait_request(&fsc->client->osdc, rd_req);
1879 err2 = ceph_osdc_wait_request(&fsc->client->osdc, wr_req);
1881 if (err >= 0 || err == -ENOENT)
1883 else if (err != -EPERM) {
1884 if (err == -EBLOCKLISTED)
1885 fsc->blocklisted = true;
1889 if (err2 == 0 || err2 == -EEXIST)
1891 else if (err2 != -EPERM) {
1892 if (err2 == -EBLOCKLISTED)
1893 fsc->blocklisted = true;
1898 pool_ns_len = pool_ns ? pool_ns->len : 0;
1899 perm = kmalloc(sizeof(*perm) + pool_ns_len + 1, GFP_NOFS);
1907 perm->pool_ns_len = pool_ns_len;
1908 if (pool_ns_len > 0)
1909 memcpy(perm->pool_ns, pool_ns->str, pool_ns_len);
1910 perm->pool_ns[pool_ns_len] = 0;
1912 rb_link_node(&perm->node, parent, p);
1913 rb_insert_color(&perm->node, &mdsc->pool_perm_tree);
1916 up_write(&mdsc->pool_perm_rwsem);
1918 ceph_osdc_put_request(rd_req);
1919 ceph_osdc_put_request(wr_req);
1924 dout("__ceph_pool_perm_get pool %lld ns %.*s result = %d\n",
1925 pool, (int)pool_ns->len, pool_ns->str, err);
1927 dout("__ceph_pool_perm_get pool %lld result = %d\n", pool, err);
1931 int ceph_pool_perm_check(struct inode *inode, int need)
1933 struct ceph_inode_info *ci = ceph_inode(inode);
1934 struct ceph_string *pool_ns;
1938 /* Only need to do this for regular files */
1939 if (!S_ISREG(inode->i_mode))
1942 if (ci->i_vino.snap != CEPH_NOSNAP) {
1944 * Pool permission check needs to write to the first object.
1945 * But for snapshot, head of the first object may have alread
1946 * been deleted. Skip check to avoid creating orphan object.
1951 if (ceph_test_mount_opt(ceph_inode_to_client(inode),
1955 spin_lock(&ci->i_ceph_lock);
1956 flags = ci->i_ceph_flags;
1957 pool = ci->i_layout.pool_id;
1958 spin_unlock(&ci->i_ceph_lock);
1960 if (flags & CEPH_I_POOL_PERM) {
1961 if ((need & CEPH_CAP_FILE_RD) && !(flags & CEPH_I_POOL_RD)) {
1962 dout("ceph_pool_perm_check pool %lld no read perm\n",
1966 if ((need & CEPH_CAP_FILE_WR) && !(flags & CEPH_I_POOL_WR)) {
1967 dout("ceph_pool_perm_check pool %lld no write perm\n",
1974 pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
1975 ret = __ceph_pool_perm_get(ci, pool, pool_ns);
1976 ceph_put_string(pool_ns);
1980 flags = CEPH_I_POOL_PERM;
1981 if (ret & POOL_READ)
1982 flags |= CEPH_I_POOL_RD;
1983 if (ret & POOL_WRITE)
1984 flags |= CEPH_I_POOL_WR;
1986 spin_lock(&ci->i_ceph_lock);
1987 if (pool == ci->i_layout.pool_id &&
1988 pool_ns == rcu_dereference_raw(ci->i_layout.pool_ns)) {
1989 ci->i_ceph_flags |= flags;
1991 pool = ci->i_layout.pool_id;
1992 flags = ci->i_ceph_flags;
1994 spin_unlock(&ci->i_ceph_lock);
1998 void ceph_pool_perm_destroy(struct ceph_mds_client *mdsc)
2000 struct ceph_pool_perm *perm;
2003 while (!RB_EMPTY_ROOT(&mdsc->pool_perm_tree)) {
2004 n = rb_first(&mdsc->pool_perm_tree);
2005 perm = rb_entry(n, struct ceph_pool_perm, node);
2006 rb_erase(n, &mdsc->pool_perm_tree);