1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/ceph/ceph_debug.h>
3
4 #include <linux/backing-dev.h>
5 #include <linux/fs.h>
6 #include <linux/mm.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>
16
17 #include "super.h"
18 #include "mds_client.h"
19 #include "cache.h"
20 #include "metric.h"
21 #include "crypto.h"
22 #include <linux/ceph/osd_client.h>
23 #include <linux/ceph/striper.h>
24
25 /*
26 * Ceph address space ops.
27 *
28 * There are a few funny things going on here.
29 *
30 * The page->private field is used to reference a struct
31 * ceph_snap_context for _every_ dirty page. This indicates which
32 * snapshot the page was logically dirtied in, and thus which snap
33 * context needs to be associated with the osd write during writeback.
34 *
35 * Similarly, struct ceph_inode_info maintains a set of counters to
36 * count dirty pages on the inode. In the absence of snapshots,
37 * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count.
38 *
39 * When a snapshot is taken (that is, when the client receives
40 * notification that a snapshot was taken), each inode with caps and
41 * with dirty pages (dirty pages implies there is a cap) gets a new
42 * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending
43 * order, new snaps go to the tail). The i_wrbuffer_ref_head count is
44 * moved to capsnap->dirty. (Unless a sync write is currently in
45 * progress. In that case, the capsnap is said to be "pending", new
46 * writes cannot start, and the capsnap isn't "finalized" until the
47 * write completes (or fails) and a final size/mtime for the inode for
48 * that snap can be settled upon.) i_wrbuffer_ref_head is reset to 0.
49 *
50 * On writeback, we must submit writes to the osd IN SNAP ORDER. So,
51 * we look for the first capsnap in i_cap_snaps and write out pages in
52 * that snap context _only_. Then we move on to the next capsnap,
53 * eventually reaching the "live" or "head" context (i.e., pages that
54 * are not yet snapped) and are writing the most recently dirtied
55 * pages.
56 *
57 * Invalidate and so forth must take care to ensure the dirty page
58 * accounting is preserved.
59 */
60
61 #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10))
62 #define CONGESTION_OFF_THRESH(congestion_kb) \
63 (CONGESTION_ON_THRESH(congestion_kb) - \
64 (CONGESTION_ON_THRESH(congestion_kb) >> 2))
65
66 static int ceph_netfs_check_write_begin(struct file *file, loff_t pos, unsigned int len,
67 struct folio **foliop, void **_fsdata);
68
69 static inline struct ceph_snap_context *page_snap_context(struct page *page)
70 {
71 if (PagePrivate(page))
72 return (void *)page->private;
73 return NULL;
74 }
75
76 /*
77 * Dirty a page. Optimistically adjust accounting, on the assumption
78 * that we won't race with invalidate. If we do, readjust.
79 */
80 static bool ceph_dirty_folio(struct address_space *mapping, struct folio *folio)
81 {
82 struct inode *inode = mapping->host;
83 struct ceph_client *cl = ceph_inode_to_client(inode);
84 struct ceph_inode_info *ci;
85 struct ceph_snap_context *snapc;
86
87 if (folio_test_dirty(folio)) {
88 doutc(cl, "%llx.%llx %p idx %lu -- already dirty\n",
89 ceph_vinop(inode), folio, folio->index);
90 VM_BUG_ON_FOLIO(!folio_test_private(folio), folio);
91 return false;
92 }
93
94 ci = ceph_inode(inode);
95
96 /* dirty the head */
97 spin_lock(&ci->i_ceph_lock);
98 if (__ceph_have_pending_cap_snap(ci)) {
99 struct ceph_cap_snap *capsnap =
100 list_last_entry(&ci->i_cap_snaps,
101 struct ceph_cap_snap,
102 ci_item);
103 snapc = ceph_get_snap_context(capsnap->context);
104 capsnap->dirty_pages++;
105 } else {
106 BUG_ON(!ci->i_head_snapc);
107 snapc = ceph_get_snap_context(ci->i_head_snapc);
108 ++ci->i_wrbuffer_ref_head;
109 }
110 if (ci->i_wrbuffer_ref == 0)
111 ihold(inode);
112 ++ci->i_wrbuffer_ref;
113 doutc(cl, "%llx.%llx %p idx %lu head %d/%d -> %d/%d "
114 "snapc %p seq %lld (%d snaps)\n",
115 ceph_vinop(inode), folio, folio->index,
116 ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1,
117 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
118 snapc, snapc->seq, snapc->num_snaps);
119 spin_unlock(&ci->i_ceph_lock);
120
121 /*
122 * Reference snap context in folio->private. Also set
123 * PagePrivate so that we get invalidate_folio callback.
124 */
125 VM_WARN_ON_FOLIO(folio->private, folio);
126 folio_attach_private(folio, snapc);
127
128 return ceph_fscache_dirty_folio(mapping, folio);
129 }
130
131 /*
132 * If we are truncating the full folio (i.e. offset == 0), adjust the
133 * dirty folio counters appropriately. Only called if there is private
134 * data on the folio.
135 */
136 static void ceph_invalidate_folio(struct folio *folio, size_t offset,
137 size_t length)
138 {
139 struct inode *inode = folio->mapping->host;
140 struct ceph_client *cl = ceph_inode_to_client(inode);
141 struct ceph_inode_info *ci = ceph_inode(inode);
142 struct ceph_snap_context *snapc;
143
144
145 if (offset != 0 || length != folio_size(folio)) {
146 doutc(cl, "%llx.%llx idx %lu partial dirty page %zu~%zu\n",
147 ceph_vinop(inode), folio->index, offset, length);
148 return;
149 }
150
151 WARN_ON(!folio_test_locked(folio));
152 if (folio_test_private(folio)) {
153 doutc(cl, "%llx.%llx idx %lu full dirty page\n",
154 ceph_vinop(inode), folio->index);
155
156 snapc = folio_detach_private(folio);
157 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
158 ceph_put_snap_context(snapc);
159 }
160
161 netfs_invalidate_folio(folio, offset, length);
162 }
163
164 static void ceph_netfs_expand_readahead(struct netfs_io_request *rreq)
165 {
166 struct inode *inode = rreq->inode;
167 struct ceph_inode_info *ci = ceph_inode(inode);
168 struct ceph_file_layout *lo = &ci->i_layout;
169 unsigned long max_pages = inode->i_sb->s_bdi->ra_pages;
170 loff_t end = rreq->start + rreq->len, new_end;
171 struct ceph_netfs_request_data *priv = rreq->netfs_priv;
172 unsigned long max_len;
173 u32 blockoff;
174
175 if (priv) {
176 /* Readahead is disabled by posix_fadvise POSIX_FADV_RANDOM */
177 if (priv->file_ra_disabled)
178 max_pages = 0;
179 else
180 max_pages = priv->file_ra_pages;
181
182 }
183
184 /* Readahead is disabled */
185 if (!max_pages)
186 return;
187
188 max_len = max_pages << PAGE_SHIFT;
189
190 /*
191 * Try to expand the length forward by rounding up it to the next
192 * block, but do not exceed the file size, unless the original
193 * request already exceeds it.
194 */
195 new_end = umin(round_up(end, lo->stripe_unit), rreq->i_size);
196 if (new_end > end && new_end <= rreq->start + max_len)
197 rreq->len = new_end - rreq->start;
198
199 /* Try to expand the start downward */
200 div_u64_rem(rreq->start, lo->stripe_unit, &blockoff);
201 if (rreq->len + blockoff <= max_len) {
202 rreq->start -= blockoff;
203 rreq->len += blockoff;
204 }
205 }
206
207 static bool ceph_netfs_clamp_length(struct netfs_io_subrequest *subreq)
208 {
209 struct inode *inode = subreq->rreq->inode;
210 struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
211 struct ceph_inode_info *ci = ceph_inode(inode);
212 u64 objno, objoff;
213 u32 xlen;
214
215 /* Truncate the extent at the end of the current block */
216 ceph_calc_file_object_mapping(&ci->i_layout, subreq->start, subreq->len,
217 &objno, &objoff, &xlen);
218 subreq->len = min(xlen, fsc->mount_options->rsize);
219 return true;
220 }
221
222 static void finish_netfs_read(struct ceph_osd_request *req)
223 {
224 struct inode *inode = req->r_inode;
225 struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
226 struct ceph_client *cl = fsc->client;
227 struct ceph_osd_data *osd_data = osd_req_op_extent_osd_data(req, 0);
228 struct netfs_io_subrequest *subreq = req->r_priv;
229 struct ceph_osd_req_op *op = &req->r_ops[0];
230 int err = req->r_result;
231 bool sparse = (op->op == CEPH_OSD_OP_SPARSE_READ);
232
233 ceph_update_read_metrics(&fsc->mdsc->metric, req->r_start_latency,
234 req->r_end_latency, osd_data->length, err);
235
236 doutc(cl, "result %d subreq->len=%zu i_size=%lld\n", req->r_result,
237 subreq->len, i_size_read(req->r_inode));
238
239 /* no object means success but no data */
240 if (err == -ENOENT)
241 err = 0;
242 else if (err == -EBLOCKLISTED)
243 fsc->blocklisted = true;
244
245 if (err >= 0) {
246 if (sparse && err > 0)
247 err = ceph_sparse_ext_map_end(op);
248 if (err < subreq->len &&
249 subreq->rreq->origin != NETFS_DIO_READ)
250 __set_bit(NETFS_SREQ_CLEAR_TAIL, &subreq->flags);
251 if (IS_ENCRYPTED(inode) && err > 0) {
252 err = ceph_fscrypt_decrypt_extents(inode,
253 osd_data->pages, subreq->start,
254 op->extent.sparse_ext,
255 op->extent.sparse_ext_cnt);
256 if (err > subreq->len)
257 err = subreq->len;
258 }
259 }
260
261 if (osd_data->type == CEPH_OSD_DATA_TYPE_PAGES) {
262 ceph_put_page_vector(osd_data->pages,
263 calc_pages_for(osd_data->alignment,
264 osd_data->length), false);
265 }
266 netfs_subreq_terminated(subreq, err, false);
267 iput(req->r_inode);
268 ceph_dec_osd_stopping_blocker(fsc->mdsc);
269 }
270
271 static bool ceph_netfs_issue_op_inline(struct netfs_io_subrequest *subreq)
272 {
273 struct netfs_io_request *rreq = subreq->rreq;
274 struct inode *inode = rreq->inode;
275 struct ceph_mds_reply_info_parsed *rinfo;
276 struct ceph_mds_reply_info_in *iinfo;
277 struct ceph_mds_request *req;
278 struct ceph_mds_client *mdsc = ceph_sb_to_mdsc(inode->i_sb);
279 struct ceph_inode_info *ci = ceph_inode(inode);
280 struct iov_iter iter;
281 ssize_t err = 0;
282 size_t len;
283 int mode;
284
285 if (rreq->origin != NETFS_DIO_READ)
286 __set_bit(NETFS_SREQ_CLEAR_TAIL, &subreq->flags);
287 __clear_bit(NETFS_SREQ_COPY_TO_CACHE, &subreq->flags);
288
289 if (subreq->start >= inode->i_size)
290 goto out;
291
292 /* We need to fetch the inline data. */
293 mode = ceph_try_to_choose_auth_mds(inode, CEPH_STAT_CAP_INLINE_DATA);
294 req = ceph_mdsc_create_request(mdsc, CEPH_MDS_OP_GETATTR, mode);
295 if (IS_ERR(req)) {
296 err = PTR_ERR(req);
297 goto out;
298 }
299 req->r_ino1 = ci->i_vino;
300 req->r_args.getattr.mask = cpu_to_le32(CEPH_STAT_CAP_INLINE_DATA);
301 req->r_num_caps = 2;
302
303 err = ceph_mdsc_do_request(mdsc, NULL, req);
304 if (err < 0)
305 goto out;
306
307 rinfo = &req->r_reply_info;
308 iinfo = &rinfo->targeti;
309 if (iinfo->inline_version == CEPH_INLINE_NONE) {
310 /* The data got uninlined */
311 ceph_mdsc_put_request(req);
312 return false;
313 }
314
315 len = min_t(size_t, iinfo->inline_len - subreq->start, subreq->len);
316 iov_iter_xarray(&iter, ITER_DEST, &rreq->mapping->i_pages, subreq->start, len);
317 err = copy_to_iter(iinfo->inline_data + subreq->start, len, &iter);
318 if (err == 0)
319 err = -EFAULT;
320
321 ceph_mdsc_put_request(req);
322 out:
323 netfs_subreq_terminated(subreq, err, false);
324 return true;
325 }
326
327 static void ceph_netfs_issue_read(struct netfs_io_subrequest *subreq)
328 {
329 struct netfs_io_request *rreq = subreq->rreq;
330 struct inode *inode = rreq->inode;
331 struct ceph_inode_info *ci = ceph_inode(inode);
332 struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
333 struct ceph_client *cl = fsc->client;
334 struct ceph_osd_request *req = NULL;
335 struct ceph_vino vino = ceph_vino(inode);
336 struct iov_iter iter;
337 int err = 0;
338 u64 len = subreq->len;
339 bool sparse = IS_ENCRYPTED(inode) || ceph_test_mount_opt(fsc, SPARSEREAD);
340 u64 off = subreq->start;
341 int extent_cnt;
342
343 if (ceph_inode_is_shutdown(inode)) {
344 err = -EIO;
345 goto out;
346 }
347
348 if (ceph_has_inline_data(ci) && ceph_netfs_issue_op_inline(subreq))
349 return;
350
351 ceph_fscrypt_adjust_off_and_len(inode, &off, &len);
352
353 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout, vino,
354 off, &len, 0, 1, sparse ? CEPH_OSD_OP_SPARSE_READ : CEPH_OSD_OP_READ,
355 CEPH_OSD_FLAG_READ, NULL, ci->i_truncate_seq,
356 ci->i_truncate_size, false);
357 if (IS_ERR(req)) {
358 err = PTR_ERR(req);
359 req = NULL;
360 goto out;
361 }
362
363 if (sparse) {
364 extent_cnt = __ceph_sparse_read_ext_count(inode, len);
365 err = ceph_alloc_sparse_ext_map(&req->r_ops[0], extent_cnt);
366 if (err)
367 goto out;
368 }
369
370 doutc(cl, "%llx.%llx pos=%llu orig_len=%zu len=%llu\n",
371 ceph_vinop(inode), subreq->start, subreq->len, len);
372
373 iov_iter_xarray(&iter, ITER_DEST, &rreq->mapping->i_pages, subreq->start, len);
374
375 /*
376 * FIXME: For now, use CEPH_OSD_DATA_TYPE_PAGES instead of _ITER for
377 * encrypted inodes. We'd need infrastructure that handles an iov_iter
378 * instead of page arrays, and we don't have that as of yet. Once the
379 * dust settles on the write helpers and encrypt/decrypt routines for
380 * netfs, we should be able to rework this.
381 */
382 if (IS_ENCRYPTED(inode)) {
383 struct page **pages;
384 size_t page_off;
385
386 err = iov_iter_get_pages_alloc2(&iter, &pages, len, &page_off);
387 if (err < 0) {
388 doutc(cl, "%llx.%llx failed to allocate pages, %d\n",
389 ceph_vinop(inode), err);
390 goto out;
391 }
392
393 /* should always give us a page-aligned read */
394 WARN_ON_ONCE(page_off);
395 len = err;
396 err = 0;
397
398 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false,
399 false);
400 } else {
401 osd_req_op_extent_osd_iter(req, 0, &iter);
402 }
403 if (!ceph_inc_osd_stopping_blocker(fsc->mdsc)) {
404 err = -EIO;
405 goto out;
406 }
407 req->r_callback = finish_netfs_read;
408 req->r_priv = subreq;
409 req->r_inode = inode;
410 ihold(inode);
411
412 ceph_osdc_start_request(req->r_osdc, req);
413 out:
414 ceph_osdc_put_request(req);
415 if (err)
416 netfs_subreq_terminated(subreq, err, false);
417 doutc(cl, "%llx.%llx result %d\n", ceph_vinop(inode), err);
418 }
419
420 static int ceph_init_request(struct netfs_io_request *rreq, struct file *file)
421 {
422 struct inode *inode = rreq->inode;
423 struct ceph_client *cl = ceph_inode_to_client(inode);
424 int got = 0, want = CEPH_CAP_FILE_CACHE;
425 struct ceph_netfs_request_data *priv;
426 int ret = 0;
427
428 /* [DEPRECATED] Use PG_private_2 to mark folio being written to the cache. */
429 __set_bit(NETFS_RREQ_USE_PGPRIV2, &rreq->flags);
430
431 if (rreq->origin != NETFS_READAHEAD)
432 return 0;
433
434 priv = kzalloc(sizeof(*priv), GFP_NOFS);
435 if (!priv)
436 return -ENOMEM;
437
438 if (file) {
439 struct ceph_rw_context *rw_ctx;
440 struct ceph_file_info *fi = file->private_data;
441
442 priv->file_ra_pages = file->f_ra.ra_pages;
443 priv->file_ra_disabled = file->f_mode & FMODE_RANDOM;
444
445 rw_ctx = ceph_find_rw_context(fi);
446 if (rw_ctx) {
447 rreq->netfs_priv = priv;
448 return 0;
449 }
450 }
451
452 /*
453 * readahead callers do not necessarily hold Fcb caps
454 * (e.g. fadvise, madvise).
455 */
456 ret = ceph_try_get_caps(inode, CEPH_CAP_FILE_RD, want, true, &got);
457 if (ret < 0) {
458 doutc(cl, "%llx.%llx, error getting cap\n", ceph_vinop(inode));
459 goto out;
460 }
461
462 if (!(got & want)) {
463 doutc(cl, "%llx.%llx, no cache cap\n", ceph_vinop(inode));
464 ret = -EACCES;
465 goto out;
466 }
467 if (ret == 0) {
468 ret = -EACCES;
469 goto out;
470 }
471
472 priv->caps = got;
473 rreq->netfs_priv = priv;
474
475 out:
476 if (ret < 0) {
477 if (got)
478 ceph_put_cap_refs(ceph_inode(inode), got);
479 kfree(priv);
480 }
481
482 return ret;
483 }
484
485 static void ceph_netfs_free_request(struct netfs_io_request *rreq)
486 {
487 struct ceph_netfs_request_data *priv = rreq->netfs_priv;
488
489 if (!priv)
490 return;
491
492 if (priv->caps)
493 ceph_put_cap_refs(ceph_inode(rreq->inode), priv->caps);
494 kfree(priv);
495 rreq->netfs_priv = NULL;
496 }
497
498 const struct netfs_request_ops ceph_netfs_ops = {
499 .init_request = ceph_init_request,
500 .free_request = ceph_netfs_free_request,
501 .issue_read = ceph_netfs_issue_read,
502 .expand_readahead = ceph_netfs_expand_readahead,
503 .clamp_length = ceph_netfs_clamp_length,
504 .check_write_begin = ceph_netfs_check_write_begin,
505 };
506
507 #ifdef CONFIG_CEPH_FSCACHE
508 static void ceph_set_page_fscache(struct page *page)
509 {
510 folio_start_private_2(page_folio(page)); /* [DEPRECATED] */
511 }
512
513 static void ceph_fscache_write_terminated(void *priv, ssize_t error, bool was_async)
514 {
515 struct inode *inode = priv;
516
517 if (IS_ERR_VALUE(error) && error != -ENOBUFS)
518 ceph_fscache_invalidate(inode, false);
519 }
520
521 static void ceph_fscache_write_to_cache(struct inode *inode, u64 off, u64 len, bool caching)
522 {
523 struct ceph_inode_info *ci = ceph_inode(inode);
524 struct fscache_cookie *cookie = ceph_fscache_cookie(ci);
525
526 fscache_write_to_cache(cookie, inode->i_mapping, off, len, i_size_read(inode),
527 ceph_fscache_write_terminated, inode, true, caching);
528 }
529 #else
530 static inline void ceph_set_page_fscache(struct page *page)
531 {
532 }
533
534 static inline void ceph_fscache_write_to_cache(struct inode *inode, u64 off, u64 len, bool caching)
535 {
536 }
537 #endif /* CONFIG_CEPH_FSCACHE */
538
539 struct ceph_writeback_ctl
540 {
541 loff_t i_size;
542 u64 truncate_size;
543 u32 truncate_seq;
544 bool size_stable;
545 bool head_snapc;
546 };
547
548 /*
549 * Get ref for the oldest snapc for an inode with dirty data... that is, the
550 * only snap context we are allowed to write back.
551 */
552 static struct ceph_snap_context *
553 get_oldest_context(struct inode *inode, struct ceph_writeback_ctl *ctl,
554 struct ceph_snap_context *page_snapc)
555 {
556 struct ceph_inode_info *ci = ceph_inode(inode);
557 struct ceph_client *cl = ceph_inode_to_client(inode);
558 struct ceph_snap_context *snapc = NULL;
559 struct ceph_cap_snap *capsnap = NULL;
560
561 spin_lock(&ci->i_ceph_lock);
562 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
563 doutc(cl, " capsnap %p snapc %p has %d dirty pages\n",
564 capsnap, capsnap->context, capsnap->dirty_pages);
565 if (!capsnap->dirty_pages)
566 continue;
567
568 /* get i_size, truncate_{seq,size} for page_snapc? */
569 if (snapc && capsnap->context != page_snapc)
570 continue;
571
572 if (ctl) {
573 if (capsnap->writing) {
574 ctl->i_size = i_size_read(inode);
575 ctl->size_stable = false;
576 } else {
577 ctl->i_size = capsnap->size;
578 ctl->size_stable = true;
579 }
580 ctl->truncate_size = capsnap->truncate_size;
581 ctl->truncate_seq = capsnap->truncate_seq;
582 ctl->head_snapc = false;
583 }
584
585 if (snapc)
586 break;
587
588 snapc = ceph_get_snap_context(capsnap->context);
589 if (!page_snapc ||
590 page_snapc == snapc ||
591 page_snapc->seq > snapc->seq)
592 break;
593 }
594 if (!snapc && ci->i_wrbuffer_ref_head) {
595 snapc = ceph_get_snap_context(ci->i_head_snapc);
596 doutc(cl, " head snapc %p has %d dirty pages\n", snapc,
597 ci->i_wrbuffer_ref_head);
598 if (ctl) {
599 ctl->i_size = i_size_read(inode);
600 ctl->truncate_size = ci->i_truncate_size;
601 ctl->truncate_seq = ci->i_truncate_seq;
602 ctl->size_stable = false;
603 ctl->head_snapc = true;
604 }
605 }
606 spin_unlock(&ci->i_ceph_lock);
607 return snapc;
608 }
609
610 static u64 get_writepages_data_length(struct inode *inode,
611 struct page *page, u64 start)
612 {
613 struct ceph_inode_info *ci = ceph_inode(inode);
614 struct ceph_snap_context *snapc;
615 struct ceph_cap_snap *capsnap = NULL;
616 u64 end = i_size_read(inode);
617 u64 ret;
618
619 snapc = page_snap_context(ceph_fscrypt_pagecache_page(page));
620 if (snapc != ci->i_head_snapc) {
621 bool found = false;
622 spin_lock(&ci->i_ceph_lock);
623 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
624 if (capsnap->context == snapc) {
625 if (!capsnap->writing)
626 end = capsnap->size;
627 found = true;
628 break;
629 }
630 }
631 spin_unlock(&ci->i_ceph_lock);
632 WARN_ON(!found);
633 }
634 if (end > ceph_fscrypt_page_offset(page) + thp_size(page))
635 end = ceph_fscrypt_page_offset(page) + thp_size(page);
636 ret = end > start ? end - start : 0;
637 if (ret && fscrypt_is_bounce_page(page))
638 ret = round_up(ret, CEPH_FSCRYPT_BLOCK_SIZE);
639 return ret;
640 }
641
642 /*
643 * Write a single page, but leave the page locked.
644 *
645 * If we get a write error, mark the mapping for error, but still adjust the
646 * dirty page accounting (i.e., page is no longer dirty).
647 */
648 static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
649 {
650 struct folio *folio = page_folio(page);
651 struct inode *inode = page->mapping->host;
652 struct ceph_inode_info *ci = ceph_inode(inode);
653 struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
654 struct ceph_client *cl = fsc->client;
655 struct ceph_snap_context *snapc, *oldest;
656 loff_t page_off = page_offset(page);
657 int err;
658 loff_t len = thp_size(page);
659 loff_t wlen;
660 struct ceph_writeback_ctl ceph_wbc;
661 struct ceph_osd_client *osdc = &fsc->client->osdc;
662 struct ceph_osd_request *req;
663 bool caching = ceph_is_cache_enabled(inode);
664 struct page *bounce_page = NULL;
665
666 doutc(cl, "%llx.%llx page %p idx %lu\n", ceph_vinop(inode), page,
667 page->index);
668
669 if (ceph_inode_is_shutdown(inode))
670 return -EIO;
671
672 /* verify this is a writeable snap context */
673 snapc = page_snap_context(page);
674 if (!snapc) {
675 doutc(cl, "%llx.%llx page %p not dirty?\n", ceph_vinop(inode),
676 page);
677 return 0;
678 }
679 oldest = get_oldest_context(inode, &ceph_wbc, snapc);
680 if (snapc->seq > oldest->seq) {
681 doutc(cl, "%llx.%llx page %p snapc %p not writeable - noop\n",
682 ceph_vinop(inode), page, snapc);
683 /* we should only noop if called by kswapd */
684 WARN_ON(!(current->flags & PF_MEMALLOC));
685 ceph_put_snap_context(oldest);
686 redirty_page_for_writepage(wbc, page);
687 return 0;
688 }
689 ceph_put_snap_context(oldest);
690
691 /* is this a partial page at end of file? */
692 if (page_off >= ceph_wbc.i_size) {
693 doutc(cl, "%llx.%llx folio at %lu beyond eof %llu\n",
694 ceph_vinop(inode), folio->index, ceph_wbc.i_size);
695 folio_invalidate(folio, 0, folio_size(folio));
696 return 0;
697 }
698
699 if (ceph_wbc.i_size < page_off + len)
700 len = ceph_wbc.i_size - page_off;
701
702 wlen = IS_ENCRYPTED(inode) ? round_up(len, CEPH_FSCRYPT_BLOCK_SIZE) : len;
703 doutc(cl, "%llx.%llx page %p index %lu on %llu~%llu snapc %p seq %lld\n",
704 ceph_vinop(inode), page, page->index, page_off, wlen, snapc,
705 snapc->seq);
706
707 if (atomic_long_inc_return(&fsc->writeback_count) >
708 CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb))
709 fsc->write_congested = true;
710
711 req = ceph_osdc_new_request(osdc, &ci->i_layout, ceph_vino(inode),
712 page_off, &wlen, 0, 1, CEPH_OSD_OP_WRITE,
713 CEPH_OSD_FLAG_WRITE, snapc,
714 ceph_wbc.truncate_seq,
715 ceph_wbc.truncate_size, true);
716 if (IS_ERR(req)) {
717 redirty_page_for_writepage(wbc, page);
718 return PTR_ERR(req);
719 }
720
721 if (wlen < len)
722 len = wlen;
723
724 set_page_writeback(page);
725 if (caching)
726 ceph_set_page_fscache(page);
727 ceph_fscache_write_to_cache(inode, page_off, len, caching);
728
729 if (IS_ENCRYPTED(inode)) {
730 bounce_page = fscrypt_encrypt_pagecache_blocks(page,
731 CEPH_FSCRYPT_BLOCK_SIZE, 0,
732 GFP_NOFS);
733 if (IS_ERR(bounce_page)) {
734 redirty_page_for_writepage(wbc, page);
735 end_page_writeback(page);
736 ceph_osdc_put_request(req);
737 return PTR_ERR(bounce_page);
738 }
739 }
740
741 /* it may be a short write due to an object boundary */
742 WARN_ON_ONCE(len > thp_size(page));
743 osd_req_op_extent_osd_data_pages(req, 0,
744 bounce_page ? &bounce_page : &page, wlen, 0,
745 false, false);
746 doutc(cl, "%llx.%llx %llu~%llu (%llu bytes, %sencrypted)\n",
747 ceph_vinop(inode), page_off, len, wlen,
748 IS_ENCRYPTED(inode) ? "" : "not ");
749
750 req->r_mtime = inode_get_mtime(inode);
751 ceph_osdc_start_request(osdc, req);
752 err = ceph_osdc_wait_request(osdc, req);
753
754 ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
755 req->r_end_latency, len, err);
756 fscrypt_free_bounce_page(bounce_page);
757 ceph_osdc_put_request(req);
758 if (err == 0)
759 err = len;
760
761 if (err < 0) {
762 struct writeback_control tmp_wbc;
763 if (!wbc)
764 wbc = &tmp_wbc;
765 if (err == -ERESTARTSYS) {
766 /* killed by SIGKILL */
767 doutc(cl, "%llx.%llx interrupted page %p\n",
768 ceph_vinop(inode), page);
769 redirty_page_for_writepage(wbc, page);
770 end_page_writeback(page);
771 return err;
772 }
773 if (err == -EBLOCKLISTED)
774 fsc->blocklisted = true;
775 doutc(cl, "%llx.%llx setting page/mapping error %d %p\n",
776 ceph_vinop(inode), err, page);
777 mapping_set_error(&inode->i_data, err);
778 wbc->pages_skipped++;
779 } else {
780 doutc(cl, "%llx.%llx cleaned page %p\n",
781 ceph_vinop(inode), page);
782 err = 0; /* vfs expects us to return 0 */
783 }
784 oldest = detach_page_private(page);
785 WARN_ON_ONCE(oldest != snapc);
786 end_page_writeback(page);
787 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
788 ceph_put_snap_context(snapc); /* page's reference */
789
790 if (atomic_long_dec_return(&fsc->writeback_count) <
791 CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb))
792 fsc->write_congested = false;
793
794 return err;
795 }
796
797 static int ceph_writepage(struct page *page, struct writeback_control *wbc)
798 {
799 int err;
800 struct inode *inode = page->mapping->host;
801 BUG_ON(!inode);
802 ihold(inode);
803
804 if (wbc->sync_mode == WB_SYNC_NONE &&
805 ceph_inode_to_fs_client(inode)->write_congested) {
806 redirty_page_for_writepage(wbc, page);
807 return AOP_WRITEPAGE_ACTIVATE;
808 }
809
810 folio_wait_private_2(page_folio(page)); /* [DEPRECATED] */
811
812 err = writepage_nounlock(page, wbc);
813 if (err == -ERESTARTSYS) {
814 /* direct memory reclaimer was killed by SIGKILL. return 0
815 * to prevent caller from setting mapping/page error */
816 err = 0;
817 }
818 unlock_page(page);
819 iput(inode);
820 return err;
821 }
822
823 /*
824 * async writeback completion handler.
825 *
826 * If we get an error, set the mapping error bit, but not the individual
827 * page error bits.
828 */
829 static void writepages_finish(struct ceph_osd_request *req)
830 {
831 struct inode *inode = req->r_inode;
832 struct ceph_inode_info *ci = ceph_inode(inode);
833 struct ceph_client *cl = ceph_inode_to_client(inode);
834 struct ceph_osd_data *osd_data;
835 struct page *page;
836 int num_pages, total_pages = 0;
837 int i, j;
838 int rc = req->r_result;
839 struct ceph_snap_context *snapc = req->r_snapc;
840 struct address_space *mapping = inode->i_mapping;
841 struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
842 unsigned int len = 0;
843 bool remove_page;
844
845 doutc(cl, "%llx.%llx rc %d\n", ceph_vinop(inode), rc);
846 if (rc < 0) {
847 mapping_set_error(mapping, rc);
848 ceph_set_error_write(ci);
849 if (rc == -EBLOCKLISTED)
850 fsc->blocklisted = true;
851 } else {
852 ceph_clear_error_write(ci);
853 }
854
855 /*
856 * We lost the cache cap, need to truncate the page before
857 * it is unlocked, otherwise we'd truncate it later in the
858 * page truncation thread, possibly losing some data that
859 * raced its way in
860 */
861 remove_page = !(ceph_caps_issued(ci) &
862 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO));
863
864 /* clean all pages */
865 for (i = 0; i < req->r_num_ops; i++) {
866 if (req->r_ops[i].op != CEPH_OSD_OP_WRITE) {
867 pr_warn_client(cl,
868 "%llx.%llx incorrect op %d req %p index %d tid %llu\n",
869 ceph_vinop(inode), req->r_ops[i].op, req, i,
870 req->r_tid);
871 break;
872 }
873
874 osd_data = osd_req_op_extent_osd_data(req, i);
875 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
876 len += osd_data->length;
877 num_pages = calc_pages_for((u64)osd_data->alignment,
878 (u64)osd_data->length);
879 total_pages += num_pages;
880 for (j = 0; j < num_pages; j++) {
881 page = osd_data->pages[j];
882 if (fscrypt_is_bounce_page(page)) {
883 page = fscrypt_pagecache_page(page);
884 fscrypt_free_bounce_page(osd_data->pages[j]);
885 osd_data->pages[j] = page;
886 }
887 BUG_ON(!page);
888 WARN_ON(!PageUptodate(page));
889
890 if (atomic_long_dec_return(&fsc->writeback_count) <
891 CONGESTION_OFF_THRESH(
892 fsc->mount_options->congestion_kb))
893 fsc->write_congested = false;
894
895 ceph_put_snap_context(detach_page_private(page));
896 end_page_writeback(page);
897 doutc(cl, "unlocking %p\n", page);
898
899 if (remove_page)
900 generic_error_remove_folio(inode->i_mapping,
901 page_folio(page));
902
903 unlock_page(page);
904 }
905 doutc(cl, "%llx.%llx wrote %llu bytes cleaned %d pages\n",
906 ceph_vinop(inode), osd_data->length,
907 rc >= 0 ? num_pages : 0);
908
909 release_pages(osd_data->pages, num_pages);
910 }
911
912 ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
913 req->r_end_latency, len, rc);
914
915 ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc);
916
917 osd_data = osd_req_op_extent_osd_data(req, 0);
918 if (osd_data->pages_from_pool)
919 mempool_free(osd_data->pages, ceph_wb_pagevec_pool);
920 else
921 kfree(osd_data->pages);
922 ceph_osdc_put_request(req);
923 ceph_dec_osd_stopping_blocker(fsc->mdsc);
924 }
925
926 /*
927 * initiate async writeback
928 */
929 static int ceph_writepages_start(struct address_space *mapping,
930 struct writeback_control *wbc)
931 {
932 struct inode *inode = mapping->host;
933 struct ceph_inode_info *ci = ceph_inode(inode);
934 struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
935 struct ceph_client *cl = fsc->client;
936 struct ceph_vino vino = ceph_vino(inode);
937 pgoff_t index, start_index, end = -1;
938 struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc;
939 struct folio_batch fbatch;
940 int rc = 0;
941 unsigned int wsize = i_blocksize(inode);
942 struct ceph_osd_request *req = NULL;
943 struct ceph_writeback_ctl ceph_wbc;
944 bool should_loop, range_whole = false;
945 bool done = false;
946 bool caching = ceph_is_cache_enabled(inode);
947 xa_mark_t tag;
948
949 if (wbc->sync_mode == WB_SYNC_NONE &&
950 fsc->write_congested)
951 return 0;
952
953 doutc(cl, "%llx.%llx (mode=%s)\n", ceph_vinop(inode),
954 wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
955 (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
956
957 if (ceph_inode_is_shutdown(inode)) {
958 if (ci->i_wrbuffer_ref > 0) {
959 pr_warn_ratelimited_client(cl,
960 "%llx.%llx %lld forced umount\n",
961 ceph_vinop(inode), ceph_ino(inode));
962 }
963 mapping_set_error(mapping, -EIO);
964 return -EIO; /* we're in a forced umount, don't write! */
965 }
966 if (fsc->mount_options->wsize < wsize)
967 wsize = fsc->mount_options->wsize;
968
969 folio_batch_init(&fbatch);
970
971 start_index = wbc->range_cyclic ? mapping->writeback_index : 0;
972 index = start_index;
973
974 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) {
975 tag = PAGECACHE_TAG_TOWRITE;
976 } else {
977 tag = PAGECACHE_TAG_DIRTY;
978 }
979 retry:
980 /* find oldest snap context with dirty data */
981 snapc = get_oldest_context(inode, &ceph_wbc, NULL);
982 if (!snapc) {
983 /* hmm, why does writepages get called when there
984 is no dirty data? */
985 doutc(cl, " no snap context with dirty data?\n");
986 goto out;
987 }
988 doutc(cl, " oldest snapc is %p seq %lld (%d snaps)\n", snapc,
989 snapc->seq, snapc->num_snaps);
990
991 should_loop = false;
992 if (ceph_wbc.head_snapc && snapc != last_snapc) {
993 /* where to start/end? */
994 if (wbc->range_cyclic) {
995 index = start_index;
996 end = -1;
997 if (index > 0)
998 should_loop = true;
999 doutc(cl, " cyclic, start at %lu\n", index);
1000 } else {
1001 index = wbc->range_start >> PAGE_SHIFT;
1002 end = wbc->range_end >> PAGE_SHIFT;
1003 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
1004 range_whole = true;
1005 doutc(cl, " not cyclic, %lu to %lu\n", index, end);
1006 }
1007 } else if (!ceph_wbc.head_snapc) {
1008 /* Do not respect wbc->range_{start,end}. Dirty pages
1009 * in that range can be associated with newer snapc.
1010 * They are not writeable until we write all dirty pages
1011 * associated with 'snapc' get written */
1012 if (index > 0)
1013 should_loop = true;
1014 doutc(cl, " non-head snapc, range whole\n");
1015 }
1016
1017 if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)
1018 tag_pages_for_writeback(mapping, index, end);
1019
1020 ceph_put_snap_context(last_snapc);
1021 last_snapc = snapc;
1022
1023 while (!done && index <= end) {
1024 int num_ops = 0, op_idx;
1025 unsigned i, nr_folios, max_pages, locked_pages = 0;
1026 struct page **pages = NULL, **data_pages;
1027 struct page *page;
1028 pgoff_t strip_unit_end = 0;
1029 u64 offset = 0, len = 0;
1030 bool from_pool = false;
1031
1032 max_pages = wsize >> PAGE_SHIFT;
1033
1034 get_more_pages:
1035 nr_folios = filemap_get_folios_tag(mapping, &index,
1036 end, tag, &fbatch);
1037 doutc(cl, "pagevec_lookup_range_tag got %d\n", nr_folios);
1038 if (!nr_folios && !locked_pages)
1039 break;
1040 for (i = 0; i < nr_folios && locked_pages < max_pages; i++) {
1041 page = &fbatch.folios[i]->page;
1042 doutc(cl, "? %p idx %lu\n", page, page->index);
1043 if (locked_pages == 0)
1044 lock_page(page); /* first page */
1045 else if (!trylock_page(page))
1046 break;
1047
1048 /* only dirty pages, or our accounting breaks */
1049 if (unlikely(!PageDirty(page)) ||
1050 unlikely(page->mapping != mapping)) {
1051 doutc(cl, "!dirty or !mapping %p\n", page);
1052 unlock_page(page);
1053 continue;
1054 }
1055 /* only if matching snap context */
1056 pgsnapc = page_snap_context(page);
1057 if (pgsnapc != snapc) {
1058 doutc(cl, "page snapc %p %lld != oldest %p %lld\n",
1059 pgsnapc, pgsnapc->seq, snapc, snapc->seq);
1060 if (!should_loop &&
1061 !ceph_wbc.head_snapc &&
1062 wbc->sync_mode != WB_SYNC_NONE)
1063 should_loop = true;
1064 unlock_page(page);
1065 continue;
1066 }
1067 if (page_offset(page) >= ceph_wbc.i_size) {
1068 struct folio *folio = page_folio(page);
1069
1070 doutc(cl, "folio at %lu beyond eof %llu\n",
1071 folio->index, ceph_wbc.i_size);
1072 if ((ceph_wbc.size_stable ||
1073 folio_pos(folio) >= i_size_read(inode)) &&
1074 folio_clear_dirty_for_io(folio))
1075 folio_invalidate(folio, 0,
1076 folio_size(folio));
1077 folio_unlock(folio);
1078 continue;
1079 }
1080 if (strip_unit_end && (page->index > strip_unit_end)) {
1081 doutc(cl, "end of strip unit %p\n", page);
1082 unlock_page(page);
1083 break;
1084 }
1085 if (PageWriteback(page) ||
1086 PagePrivate2(page) /* [DEPRECATED] */) {
1087 if (wbc->sync_mode == WB_SYNC_NONE) {
1088 doutc(cl, "%p under writeback\n", page);
1089 unlock_page(page);
1090 continue;
1091 }
1092 doutc(cl, "waiting on writeback %p\n", page);
1093 wait_on_page_writeback(page);
1094 folio_wait_private_2(page_folio(page)); /* [DEPRECATED] */
1095 }
1096
1097 if (!clear_page_dirty_for_io(page)) {
1098 doutc(cl, "%p !clear_page_dirty_for_io\n", page);
1099 unlock_page(page);
1100 continue;
1101 }
1102
1103 /*
1104 * We have something to write. If this is
1105 * the first locked page this time through,
1106 * calculate max possinle write size and
1107 * allocate a page array
1108 */
1109 if (locked_pages == 0) {
1110 u64 objnum;
1111 u64 objoff;
1112 u32 xlen;
1113
1114 /* prepare async write request */
1115 offset = (u64)page_offset(page);
1116 ceph_calc_file_object_mapping(&ci->i_layout,
1117 offset, wsize,
1118 &objnum, &objoff,
1119 &xlen);
1120 len = xlen;
1121
1122 num_ops = 1;
1123 strip_unit_end = page->index +
1124 ((len - 1) >> PAGE_SHIFT);
1125
1126 BUG_ON(pages);
1127 max_pages = calc_pages_for(0, (u64)len);
1128 pages = kmalloc_array(max_pages,
1129 sizeof(*pages),
1130 GFP_NOFS);
1131 if (!pages) {
1132 from_pool = true;
1133 pages = mempool_alloc(ceph_wb_pagevec_pool, GFP_NOFS);
1134 BUG_ON(!pages);
1135 }
1136
1137 len = 0;
1138 } else if (page->index !=
1139 (offset + len) >> PAGE_SHIFT) {
1140 if (num_ops >= (from_pool ? CEPH_OSD_SLAB_OPS :
1141 CEPH_OSD_MAX_OPS)) {
1142 redirty_page_for_writepage(wbc, page);
1143 unlock_page(page);
1144 break;
1145 }
1146
1147 num_ops++;
1148 offset = (u64)page_offset(page);
1149 len = 0;
1150 }
1151
1152 /* note position of first page in fbatch */
1153 doutc(cl, "%llx.%llx will write page %p idx %lu\n",
1154 ceph_vinop(inode), page, page->index);
1155
1156 if (atomic_long_inc_return(&fsc->writeback_count) >
1157 CONGESTION_ON_THRESH(
1158 fsc->mount_options->congestion_kb))
1159 fsc->write_congested = true;
1160
1161 if (IS_ENCRYPTED(inode)) {
1162 pages[locked_pages] =
1163 fscrypt_encrypt_pagecache_blocks(page,
1164 PAGE_SIZE, 0,
1165 locked_pages ? GFP_NOWAIT : GFP_NOFS);
1166 if (IS_ERR(pages[locked_pages])) {
1167 if (PTR_ERR(pages[locked_pages]) == -EINVAL)
1168 pr_err_client(cl,
1169 "inode->i_blkbits=%hhu\n",
1170 inode->i_blkbits);
1171 /* better not fail on first page! */
1172 BUG_ON(locked_pages == 0);
1173 pages[locked_pages] = NULL;
1174 redirty_page_for_writepage(wbc, page);
1175 unlock_page(page);
1176 break;
1177 }
1178 ++locked_pages;
1179 } else {
1180 pages[locked_pages++] = page;
1181 }
1182
1183 fbatch.folios[i] = NULL;
1184 len += thp_size(page);
1185 }
1186
1187 /* did we get anything? */
1188 if (!locked_pages)
1189 goto release_folios;
1190 if (i) {
1191 unsigned j, n = 0;
1192 /* shift unused page to beginning of fbatch */
1193 for (j = 0; j < nr_folios; j++) {
1194 if (!fbatch.folios[j])
1195 continue;
1196 if (n < j)
1197 fbatch.folios[n] = fbatch.folios[j];
1198 n++;
1199 }
1200 fbatch.nr = n;
1201
1202 if (nr_folios && i == nr_folios &&
1203 locked_pages < max_pages) {
1204 doutc(cl, "reached end fbatch, trying for more\n");
1205 folio_batch_release(&fbatch);
1206 goto get_more_pages;
1207 }
1208 }
1209
1210 new_request:
1211 offset = ceph_fscrypt_page_offset(pages[0]);
1212 len = wsize;
1213
1214 req = ceph_osdc_new_request(&fsc->client->osdc,
1215 &ci->i_layout, vino,
1216 offset, &len, 0, num_ops,
1217 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
1218 snapc, ceph_wbc.truncate_seq,
1219 ceph_wbc.truncate_size, false);
1220 if (IS_ERR(req)) {
1221 req = ceph_osdc_new_request(&fsc->client->osdc,
1222 &ci->i_layout, vino,
1223 offset, &len, 0,
1224 min(num_ops,
1225 CEPH_OSD_SLAB_OPS),
1226 CEPH_OSD_OP_WRITE,
1227 CEPH_OSD_FLAG_WRITE,
1228 snapc, ceph_wbc.truncate_seq,
1229 ceph_wbc.truncate_size, true);
1230 BUG_ON(IS_ERR(req));
1231 }
1232 BUG_ON(len < ceph_fscrypt_page_offset(pages[locked_pages - 1]) +
1233 thp_size(pages[locked_pages - 1]) - offset);
1234
1235 if (!ceph_inc_osd_stopping_blocker(fsc->mdsc)) {
1236 rc = -EIO;
1237 goto release_folios;
1238 }
1239 req->r_callback = writepages_finish;
1240 req->r_inode = inode;
1241
1242 /* Format the osd request message and submit the write */
1243 len = 0;
1244 data_pages = pages;
1245 op_idx = 0;
1246 for (i = 0; i < locked_pages; i++) {
1247 struct page *page = ceph_fscrypt_pagecache_page(pages[i]);
1248
1249 u64 cur_offset = page_offset(page);
1250 /*
1251 * Discontinuity in page range? Ceph can handle that by just passing
1252 * multiple extents in the write op.
1253 */
1254 if (offset + len != cur_offset) {
1255 /* If it's full, stop here */
1256 if (op_idx + 1 == req->r_num_ops)
1257 break;
1258
1259 /* Kick off an fscache write with what we have so far. */
1260 ceph_fscache_write_to_cache(inode, offset, len, caching);
1261
1262 /* Start a new extent */
1263 osd_req_op_extent_dup_last(req, op_idx,
1264 cur_offset - offset);
1265 doutc(cl, "got pages at %llu~%llu\n", offset,
1266 len);
1267 osd_req_op_extent_osd_data_pages(req, op_idx,
1268 data_pages, len, 0,
1269 from_pool, false);
1270 osd_req_op_extent_update(req, op_idx, len);
1271
1272 len = 0;
1273 offset = cur_offset;
1274 data_pages = pages + i;
1275 op_idx++;
1276 }
1277
1278 set_page_writeback(page);
1279 if (caching)
1280 ceph_set_page_fscache(page);
1281 len += thp_size(page);
1282 }
1283 ceph_fscache_write_to_cache(inode, offset, len, caching);
1284
1285 if (ceph_wbc.size_stable) {
1286 len = min(len, ceph_wbc.i_size - offset);
1287 } else if (i == locked_pages) {
1288 /* writepages_finish() clears writeback pages
1289 * according to the data length, so make sure
1290 * data length covers all locked pages */
1291 u64 min_len = len + 1 - thp_size(page);
1292 len = get_writepages_data_length(inode, pages[i - 1],
1293 offset);
1294 len = max(len, min_len);
1295 }
1296 if (IS_ENCRYPTED(inode))
1297 len = round_up(len, CEPH_FSCRYPT_BLOCK_SIZE);
1298
1299 doutc(cl, "got pages at %llu~%llu\n", offset, len);
1300
1301 if (IS_ENCRYPTED(inode) &&
1302 ((offset | len) & ~CEPH_FSCRYPT_BLOCK_MASK))
1303 pr_warn_client(cl,
1304 "bad encrypted write offset=%lld len=%llu\n",
1305 offset, len);
1306
1307 osd_req_op_extent_osd_data_pages(req, op_idx, data_pages, len,
1308 0, from_pool, false);
1309 osd_req_op_extent_update(req, op_idx, len);
1310
1311 BUG_ON(op_idx + 1 != req->r_num_ops);
1312
1313 from_pool = false;
1314 if (i < locked_pages) {
1315 BUG_ON(num_ops <= req->r_num_ops);
1316 num_ops -= req->r_num_ops;
1317 locked_pages -= i;
1318
1319 /* allocate new pages array for next request */
1320 data_pages = pages;
1321 pages = kmalloc_array(locked_pages, sizeof(*pages),
1322 GFP_NOFS);
1323 if (!pages) {
1324 from_pool = true;
1325 pages = mempool_alloc(ceph_wb_pagevec_pool, GFP_NOFS);
1326 BUG_ON(!pages);
1327 }
1328 memcpy(pages, data_pages + i,
1329 locked_pages * sizeof(*pages));
1330 memset(data_pages + i, 0,
1331 locked_pages * sizeof(*pages));
1332 } else {
1333 BUG_ON(num_ops != req->r_num_ops);
1334 index = pages[i - 1]->index + 1;
1335 /* request message now owns the pages array */
1336 pages = NULL;
1337 }
1338
1339 req->r_mtime = inode_get_mtime(inode);
1340 ceph_osdc_start_request(&fsc->client->osdc, req);
1341 req = NULL;
1342
1343 wbc->nr_to_write -= i;
1344 if (pages)
1345 goto new_request;
1346
1347 /*
1348 * We stop writing back only if we are not doing
1349 * integrity sync. In case of integrity sync we have to
1350 * keep going until we have written all the pages
1351 * we tagged for writeback prior to entering this loop.
1352 */
1353 if (wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE)
1354 done = true;
1355
1356 release_folios:
1357 doutc(cl, "folio_batch release on %d folios (%p)\n",
1358 (int)fbatch.nr, fbatch.nr ? fbatch.folios[0] : NULL);
1359 folio_batch_release(&fbatch);
1360 }
1361
1362 if (should_loop && !done) {
1363 /* more to do; loop back to beginning of file */
1364 doutc(cl, "looping back to beginning of file\n");
1365 end = start_index - 1; /* OK even when start_index == 0 */
1366
1367 /* to write dirty pages associated with next snapc,
1368 * we need to wait until current writes complete */
1369 if (wbc->sync_mode != WB_SYNC_NONE &&
1370 start_index == 0 && /* all dirty pages were checked */
1371 !ceph_wbc.head_snapc) {
1372 struct page *page;
1373 unsigned i, nr;
1374 index = 0;
1375 while ((index <= end) &&
1376 (nr = filemap_get_folios_tag(mapping, &index,
1377 (pgoff_t)-1,
1378 PAGECACHE_TAG_WRITEBACK,
1379 &fbatch))) {
1380 for (i = 0; i < nr; i++) {
1381 page = &fbatch.folios[i]->page;
1382 if (page_snap_context(page) != snapc)
1383 continue;
1384 wait_on_page_writeback(page);
1385 }
1386 folio_batch_release(&fbatch);
1387 cond_resched();
1388 }
1389 }
1390
1391 start_index = 0;
1392 index = 0;
1393 goto retry;
1394 }
1395
1396 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1397 mapping->writeback_index = index;
1398
1399 out:
1400 ceph_osdc_put_request(req);
1401 ceph_put_snap_context(last_snapc);
1402 doutc(cl, "%llx.%llx dend - startone, rc = %d\n", ceph_vinop(inode),
1403 rc);
1404 return rc;
1405 }
1406
1407
1408
1409 /*
1410 * See if a given @snapc is either writeable, or already written.
1411 */
1412 static int context_is_writeable_or_written(struct inode *inode,
1413 struct ceph_snap_context *snapc)
1414 {
1415 struct ceph_snap_context *oldest = get_oldest_context(inode, NULL, NULL);
1416 int ret = !oldest || snapc->seq <= oldest->seq;
1417
1418 ceph_put_snap_context(oldest);
1419 return ret;
1420 }
1421
1422 /**
1423 * ceph_find_incompatible - find an incompatible context and return it
1424 * @page: page being dirtied
1425 *
1426 * We are only allowed to write into/dirty a page if the page is
1427 * clean, or already dirty within the same snap context. Returns a
1428 * conflicting context if there is one, NULL if there isn't, or a
1429 * negative error code on other errors.
1430 *
1431 * Must be called with page lock held.
1432 */
1433 static struct ceph_snap_context *
1434 ceph_find_incompatible(struct page *page)
1435 {
1436 struct inode *inode = page->mapping->host;
1437 struct ceph_client *cl = ceph_inode_to_client(inode);
1438 struct ceph_inode_info *ci = ceph_inode(inode);
1439
1440 if (ceph_inode_is_shutdown(inode)) {
1441 doutc(cl, " %llx.%llx page %p is shutdown\n",
1442 ceph_vinop(inode), page);
1443 return ERR_PTR(-ESTALE);
1444 }
1445
1446 for (;;) {
1447 struct ceph_snap_context *snapc, *oldest;
1448
1449 wait_on_page_writeback(page);
1450
1451 snapc = page_snap_context(page);
1452 if (!snapc || snapc == ci->i_head_snapc)
1453 break;
1454
1455 /*
1456 * this page is already dirty in another (older) snap
1457 * context! is it writeable now?
1458 */
1459 oldest = get_oldest_context(inode, NULL, NULL);
1460 if (snapc->seq > oldest->seq) {
1461 /* not writeable -- return it for the caller to deal with */
1462 ceph_put_snap_context(oldest);
1463 doutc(cl, " %llx.%llx page %p snapc %p not current or oldest\n",
1464 ceph_vinop(inode), page, snapc);
1465 return ceph_get_snap_context(snapc);
1466 }
1467 ceph_put_snap_context(oldest);
1468
1469 /* yay, writeable, do it now (without dropping page lock) */
1470 doutc(cl, " %llx.%llx page %p snapc %p not current, but oldest\n",
1471 ceph_vinop(inode), page, snapc);
1472 if (clear_page_dirty_for_io(page)) {
1473 int r = writepage_nounlock(page, NULL);
1474 if (r < 0)
1475 return ERR_PTR(r);
1476 }
1477 }
1478 return NULL;
1479 }
1480
1481 static int ceph_netfs_check_write_begin(struct file *file, loff_t pos, unsigned int len,
1482 struct folio **foliop, void **_fsdata)
1483 {
1484 struct inode *inode = file_inode(file);
1485 struct ceph_inode_info *ci = ceph_inode(inode);
1486 struct ceph_snap_context *snapc;
1487
1488 snapc = ceph_find_incompatible(folio_page(*foliop, 0));
1489 if (snapc) {
1490 int r;
1491
1492 folio_unlock(*foliop);
1493 folio_put(*foliop);
1494 *foliop = NULL;
1495 if (IS_ERR(snapc))
1496 return PTR_ERR(snapc);
1497
1498 ceph_queue_writeback(inode);
1499 r = wait_event_killable(ci->i_cap_wq,
1500 context_is_writeable_or_written(inode, snapc));
1501 ceph_put_snap_context(snapc);
1502 return r == 0 ? -EAGAIN : r;
1503 }
1504 return 0;
1505 }
1506
1507 /*
1508 * We are only allowed to write into/dirty the page if the page is
1509 * clean, or already dirty within the same snap context.
1510 */
1511 static int ceph_write_begin(struct file *file, struct address_space *mapping,
1512 loff_t pos, unsigned len,
1513 struct page **pagep, void **fsdata)
1514 {
1515 struct inode *inode = file_inode(file);
1516 struct ceph_inode_info *ci = ceph_inode(inode);
1517 struct folio *folio = NULL;
1518 int r;
1519
1520 r = netfs_write_begin(&ci->netfs, file, inode->i_mapping, pos, len, &folio, NULL);
1521 if (r < 0)
1522 return r;
1523
1524 folio_wait_private_2(folio); /* [DEPRECATED] */
1525 WARN_ON_ONCE(!folio_test_locked(folio));
1526 *pagep = &folio->page;
1527 return 0;
1528 }
1529
1530 /*
1531 * we don't do anything in here that simple_write_end doesn't do
1532 * except adjust dirty page accounting
1533 */
1534 static int ceph_write_end(struct file *file, struct address_space *mapping,
1535 loff_t pos, unsigned len, unsigned copied,
1536 struct page *subpage, void *fsdata)
1537 {
1538 struct folio *folio = page_folio(subpage);
1539 struct inode *inode = file_inode(file);
1540 struct ceph_client *cl = ceph_inode_to_client(inode);
1541 bool check_cap = false;
1542
1543 doutc(cl, "%llx.%llx file %p folio %p %d~%d (%d)\n", ceph_vinop(inode),
1544 file, folio, (int)pos, (int)copied, (int)len);
1545
1546 if (!folio_test_uptodate(folio)) {
1547 /* just return that nothing was copied on a short copy */
1548 if (copied < len) {
1549 copied = 0;
1550 goto out;
1551 }
1552 folio_mark_uptodate(folio);
1553 }
1554
1555 /* did file size increase? */
1556 if (pos+copied > i_size_read(inode))
1557 check_cap = ceph_inode_set_size(inode, pos+copied);
1558
1559 folio_mark_dirty(folio);
1560
1561 out:
1562 folio_unlock(folio);
1563 folio_put(folio);
1564
1565 if (check_cap)
1566 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY);
1567
1568 return copied;
1569 }
1570
1571 const struct address_space_operations ceph_aops = {
1572 .read_folio = netfs_read_folio,
1573 .readahead = netfs_readahead,
1574 .writepage = ceph_writepage,
1575 .writepages = ceph_writepages_start,
1576 .write_begin = ceph_write_begin,
1577 .write_end = ceph_write_end,
1578 .dirty_folio = ceph_dirty_folio,
1579 .invalidate_folio = ceph_invalidate_folio,
1580 .release_folio = netfs_release_folio,
1581 .direct_IO = noop_direct_IO,
1582 };
1583
1584 static void ceph_block_sigs(sigset_t *oldset)
1585 {
1586 sigset_t mask;
1587 siginitsetinv(&mask, sigmask(SIGKILL));
1588 sigprocmask(SIG_BLOCK, &mask, oldset);
1589 }
1590
1591 static void ceph_restore_sigs(sigset_t *oldset)
1592 {
1593 sigprocmask(SIG_SETMASK, oldset, NULL);
1594 }
1595
1596 /*
1597 * vm ops
1598 */
1599 static vm_fault_t ceph_filemap_fault(struct vm_fault *vmf)
1600 {
1601 struct vm_area_struct *vma = vmf->vma;
1602 struct inode *inode = file_inode(vma->vm_file);
1603 struct ceph_inode_info *ci = ceph_inode(inode);
1604 struct ceph_client *cl = ceph_inode_to_client(inode);
1605 struct ceph_file_info *fi = vma->vm_file->private_data;
1606 loff_t off = (loff_t)vmf->pgoff << PAGE_SHIFT;
1607 int want, got, err;
1608 sigset_t oldset;
1609 vm_fault_t ret = VM_FAULT_SIGBUS;
1610
1611 if (ceph_inode_is_shutdown(inode))
1612 return ret;
1613
1614 ceph_block_sigs(&oldset);
1615
1616 doutc(cl, "%llx.%llx %llu trying to get caps\n",
1617 ceph_vinop(inode), off);
1618 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1619 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1620 else
1621 want = CEPH_CAP_FILE_CACHE;
1622
1623 got = 0;
1624 err = ceph_get_caps(vma->vm_file, CEPH_CAP_FILE_RD, want, -1, &got);
1625 if (err < 0)
1626 goto out_restore;
1627
1628 doutc(cl, "%llx.%llx %llu got cap refs on %s\n", ceph_vinop(inode),
1629 off, ceph_cap_string(got));
1630
1631 if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) ||
1632 !ceph_has_inline_data(ci)) {
1633 CEPH_DEFINE_RW_CONTEXT(rw_ctx, got);
1634 ceph_add_rw_context(fi, &rw_ctx);
1635 ret = filemap_fault(vmf);
1636 ceph_del_rw_context(fi, &rw_ctx);
1637 doutc(cl, "%llx.%llx %llu drop cap refs %s ret %x\n",
1638 ceph_vinop(inode), off, ceph_cap_string(got), ret);
1639 } else
1640 err = -EAGAIN;
1641
1642 ceph_put_cap_refs(ci, got);
1643
1644 if (err != -EAGAIN)
1645 goto out_restore;
1646
1647 /* read inline data */
1648 if (off >= PAGE_SIZE) {
1649 /* does not support inline data > PAGE_SIZE */
1650 ret = VM_FAULT_SIGBUS;
1651 } else {
1652 struct address_space *mapping = inode->i_mapping;
1653 struct page *page;
1654
1655 filemap_invalidate_lock_shared(mapping);
1656 page = find_or_create_page(mapping, 0,
1657 mapping_gfp_constraint(mapping, ~__GFP_FS));
1658 if (!page) {
1659 ret = VM_FAULT_OOM;
1660 goto out_inline;
1661 }
1662 err = __ceph_do_getattr(inode, page,
1663 CEPH_STAT_CAP_INLINE_DATA, true);
1664 if (err < 0 || off >= i_size_read(inode)) {
1665 unlock_page(page);
1666 put_page(page);
1667 ret = vmf_error(err);
1668 goto out_inline;
1669 }
1670 if (err < PAGE_SIZE)
1671 zero_user_segment(page, err, PAGE_SIZE);
1672 else
1673 flush_dcache_page(page);
1674 SetPageUptodate(page);
1675 vmf->page = page;
1676 ret = VM_FAULT_MAJOR | VM_FAULT_LOCKED;
1677 out_inline:
1678 filemap_invalidate_unlock_shared(mapping);
1679 doutc(cl, "%llx.%llx %llu read inline data ret %x\n",
1680 ceph_vinop(inode), off, ret);
1681 }
1682 out_restore:
1683 ceph_restore_sigs(&oldset);
1684 if (err < 0)
1685 ret = vmf_error(err);
1686
1687 return ret;
1688 }
1689
1690 static vm_fault_t ceph_page_mkwrite(struct vm_fault *vmf)
1691 {
1692 struct vm_area_struct *vma = vmf->vma;
1693 struct inode *inode = file_inode(vma->vm_file);
1694 struct ceph_client *cl = ceph_inode_to_client(inode);
1695 struct ceph_inode_info *ci = ceph_inode(inode);
1696 struct ceph_file_info *fi = vma->vm_file->private_data;
1697 struct ceph_cap_flush *prealloc_cf;
1698 struct page *page = vmf->page;
1699 loff_t off = page_offset(page);
1700 loff_t size = i_size_read(inode);
1701 size_t len;
1702 int want, got, err;
1703 sigset_t oldset;
1704 vm_fault_t ret = VM_FAULT_SIGBUS;
1705
1706 if (ceph_inode_is_shutdown(inode))
1707 return ret;
1708
1709 prealloc_cf = ceph_alloc_cap_flush();
1710 if (!prealloc_cf)
1711 return VM_FAULT_OOM;
1712
1713 sb_start_pagefault(inode->i_sb);
1714 ceph_block_sigs(&oldset);
1715
1716 if (off + thp_size(page) <= size)
1717 len = thp_size(page);
1718 else
1719 len = offset_in_thp(page, size);
1720
1721 doutc(cl, "%llx.%llx %llu~%zd getting caps i_size %llu\n",
1722 ceph_vinop(inode), off, len, size);
1723 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1724 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1725 else
1726 want = CEPH_CAP_FILE_BUFFER;
1727
1728 got = 0;
1729 err = ceph_get_caps(vma->vm_file, CEPH_CAP_FILE_WR, want, off + len, &got);
1730 if (err < 0)
1731 goto out_free;
1732
1733 doutc(cl, "%llx.%llx %llu~%zd got cap refs on %s\n", ceph_vinop(inode),
1734 off, len, ceph_cap_string(got));
1735
1736 /* Update time before taking page lock */
1737 file_update_time(vma->vm_file);
1738 inode_inc_iversion_raw(inode);
1739
1740 do {
1741 struct ceph_snap_context *snapc;
1742
1743 lock_page(page);
1744
1745 if (page_mkwrite_check_truncate(page, inode) < 0) {
1746 unlock_page(page);
1747 ret = VM_FAULT_NOPAGE;
1748 break;
1749 }
1750
1751 snapc = ceph_find_incompatible(page);
1752 if (!snapc) {
1753 /* success. we'll keep the page locked. */
1754 set_page_dirty(page);
1755 ret = VM_FAULT_LOCKED;
1756 break;
1757 }
1758
1759 unlock_page(page);
1760
1761 if (IS_ERR(snapc)) {
1762 ret = VM_FAULT_SIGBUS;
1763 break;
1764 }
1765
1766 ceph_queue_writeback(inode);
1767 err = wait_event_killable(ci->i_cap_wq,
1768 context_is_writeable_or_written(inode, snapc));
1769 ceph_put_snap_context(snapc);
1770 } while (err == 0);
1771
1772 if (ret == VM_FAULT_LOCKED) {
1773 int dirty;
1774 spin_lock(&ci->i_ceph_lock);
1775 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1776 &prealloc_cf);
1777 spin_unlock(&ci->i_ceph_lock);
1778 if (dirty)
1779 __mark_inode_dirty(inode, dirty);
1780 }
1781
1782 doutc(cl, "%llx.%llx %llu~%zd dropping cap refs on %s ret %x\n",
1783 ceph_vinop(inode), off, len, ceph_cap_string(got), ret);
1784 ceph_put_cap_refs_async(ci, got);
1785 out_free:
1786 ceph_restore_sigs(&oldset);
1787 sb_end_pagefault(inode->i_sb);
1788 ceph_free_cap_flush(prealloc_cf);
1789 if (err < 0)
1790 ret = vmf_error(err);
1791 return ret;
1792 }
1793
1794 void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
1795 char *data, size_t len)
1796 {
1797 struct ceph_client *cl = ceph_inode_to_client(inode);
1798 struct address_space *mapping = inode->i_mapping;
1799 struct page *page;
1800
1801 if (locked_page) {
1802 page = locked_page;
1803 } else {
1804 if (i_size_read(inode) == 0)
1805 return;
1806 page = find_or_create_page(mapping, 0,
1807 mapping_gfp_constraint(mapping,
1808 ~__GFP_FS));
1809 if (!page)
1810 return;
1811 if (PageUptodate(page)) {
1812 unlock_page(page);
1813 put_page(page);
1814 return;
1815 }
1816 }
1817
1818 doutc(cl, "%p %llx.%llx len %zu locked_page %p\n", inode,
1819 ceph_vinop(inode), len, locked_page);
1820
1821 if (len > 0) {
1822 void *kaddr = kmap_atomic(page);
1823 memcpy(kaddr, data, len);
1824 kunmap_atomic(kaddr);
1825 }
1826
1827 if (page != locked_page) {
1828 if (len < PAGE_SIZE)
1829 zero_user_segment(page, len, PAGE_SIZE);
1830 else
1831 flush_dcache_page(page);
1832
1833 SetPageUptodate(page);
1834 unlock_page(page);
1835 put_page(page);
1836 }
1837 }
1838
1839 int ceph_uninline_data(struct file *file)
1840 {
1841 struct inode *inode = file_inode(file);
1842 struct ceph_inode_info *ci = ceph_inode(inode);
1843 struct ceph_fs_client *fsc = ceph_inode_to_fs_client(inode);
1844 struct ceph_client *cl = fsc->client;
1845 struct ceph_osd_request *req = NULL;
1846 struct ceph_cap_flush *prealloc_cf = NULL;
1847 struct folio *folio = NULL;
1848 u64 inline_version = CEPH_INLINE_NONE;
1849 struct page *pages[1];
1850 int err = 0;
1851 u64 len;
1852
1853 spin_lock(&ci->i_ceph_lock);
1854 inline_version = ci->i_inline_version;
1855 spin_unlock(&ci->i_ceph_lock);
1856
1857 doutc(cl, "%llx.%llx inline_version %llu\n", ceph_vinop(inode),
1858 inline_version);
1859
1860 if (ceph_inode_is_shutdown(inode)) {
1861 err = -EIO;
1862 goto out;
1863 }
1864
1865 if (inline_version == CEPH_INLINE_NONE)
1866 return 0;
1867
1868 prealloc_cf = ceph_alloc_cap_flush();
1869 if (!prealloc_cf)
1870 return -ENOMEM;
1871
1872 if (inline_version == 1) /* initial version, no data */
1873 goto out_uninline;
1874
1875 folio = read_mapping_folio(inode->i_mapping, 0, file);
1876 if (IS_ERR(folio)) {
1877 err = PTR_ERR(folio);
1878 goto out;
1879 }
1880
1881 folio_lock(folio);
1882
1883 len = i_size_read(inode);
1884 if (len > folio_size(folio))
1885 len = folio_size(folio);
1886
1887 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1888 ceph_vino(inode), 0, &len, 0, 1,
1889 CEPH_OSD_OP_CREATE, CEPH_OSD_FLAG_WRITE,
1890 NULL, 0, 0, false);
1891 if (IS_ERR(req)) {
1892 err = PTR_ERR(req);
1893 goto out_unlock;
1894 }
1895
1896 req->r_mtime = inode_get_mtime(inode);
1897 ceph_osdc_start_request(&fsc->client->osdc, req);
1898 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1899 ceph_osdc_put_request(req);
1900 if (err < 0)
1901 goto out_unlock;
1902
1903 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1904 ceph_vino(inode), 0, &len, 1, 3,
1905 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
1906 NULL, ci->i_truncate_seq,
1907 ci->i_truncate_size, false);
1908 if (IS_ERR(req)) {
1909 err = PTR_ERR(req);
1910 goto out_unlock;
1911 }
1912
1913 pages[0] = folio_page(folio, 0);
1914 osd_req_op_extent_osd_data_pages(req, 1, pages, len, 0, false, false);
1915
1916 {
1917 __le64 xattr_buf = cpu_to_le64(inline_version);
1918 err = osd_req_op_xattr_init(req, 0, CEPH_OSD_OP_CMPXATTR,
1919 "inline_version", &xattr_buf,
1920 sizeof(xattr_buf),
1921 CEPH_OSD_CMPXATTR_OP_GT,
1922 CEPH_OSD_CMPXATTR_MODE_U64);
1923 if (err)
1924 goto out_put_req;
1925 }
1926
1927 {
1928 char xattr_buf[32];
1929 int xattr_len = snprintf(xattr_buf, sizeof(xattr_buf),
1930 "%llu", inline_version);
1931 err = osd_req_op_xattr_init(req, 2, CEPH_OSD_OP_SETXATTR,
1932 "inline_version",
1933 xattr_buf, xattr_len, 0, 0);
1934 if (err)
1935 goto out_put_req;
1936 }
1937
1938 req->r_mtime = inode_get_mtime(inode);
1939 ceph_osdc_start_request(&fsc->client->osdc, req);
1940 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1941
1942 ceph_update_write_metrics(&fsc->mdsc->metric, req->r_start_latency,
1943 req->r_end_latency, len, err);
1944
1945 out_uninline:
1946 if (!err) {
1947 int dirty;
1948
1949 /* Set to CAP_INLINE_NONE and dirty the caps */
1950 down_read(&fsc->mdsc->snap_rwsem);
1951 spin_lock(&ci->i_ceph_lock);
1952 ci->i_inline_version = CEPH_INLINE_NONE;
1953 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR, &prealloc_cf);
1954 spin_unlock(&ci->i_ceph_lock);
1955 up_read(&fsc->mdsc->snap_rwsem);
1956 if (dirty)
1957 __mark_inode_dirty(inode, dirty);
1958 }
1959 out_put_req:
1960 ceph_osdc_put_request(req);
1961 if (err == -ECANCELED)
1962 err = 0;
1963 out_unlock:
1964 if (folio) {
1965 folio_unlock(folio);
1966 folio_put(folio);
1967 }
1968 out:
1969 ceph_free_cap_flush(prealloc_cf);
1970 doutc(cl, "%llx.%llx inline_version %llu = %d\n",
1971 ceph_vinop(inode), inline_version, err);
1972 return err;
1973 }
1974
1975 static const struct vm_operations_struct ceph_vmops = {
1976 .fault = ceph_filemap_fault,
1977 .page_mkwrite = ceph_page_mkwrite,
1978 };
1979
1980 int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1981 {
1982 struct address_space *mapping = file->f_mapping;
1983
1984 if (!mapping->a_ops->read_folio)
1985 return -ENOEXEC;
1986 vma->vm_ops = &ceph_vmops;
1987 return 0;
1988 }
1989
1990 enum {
1991 POOL_READ = 1,
1992 POOL_WRITE = 2,
1993 };
1994
1995 static int __ceph_pool_perm_get(struct ceph_inode_info *ci,
1996 s64 pool, struct ceph_string *pool_ns)
1997 {
1998 struct ceph_fs_client *fsc = ceph_inode_to_fs_client(&ci->netfs.inode);
1999 struct ceph_mds_client *mdsc = fsc->mdsc;
2000 struct ceph_client *cl = fsc->client;
2001 struct ceph_osd_request *rd_req = NULL, *wr_req = NULL;
2002 struct rb_node **p, *parent;
2003 struct ceph_pool_perm *perm;
2004 struct page **pages;
2005 size_t pool_ns_len;
2006 int err = 0, err2 = 0, have = 0;
2007
2008 down_read(&mdsc->pool_perm_rwsem);
2009 p = &mdsc->pool_perm_tree.rb_node;
2010 while (*p) {
2011 perm = rb_entry(*p, struct ceph_pool_perm, node);
2012 if (pool < perm->pool)
2013 p = &(*p)->rb_left;
2014 else if (pool > perm->pool)
2015 p = &(*p)->rb_right;
2016 else {
2017 int ret = ceph_compare_string(pool_ns,
2018 perm->pool_ns,
2019 perm->pool_ns_len);
2020 if (ret < 0)
2021 p = &(*p)->rb_left;
2022 else if (ret > 0)
2023 p = &(*p)->rb_right;
2024 else {
2025 have = perm->perm;
2026 break;
2027 }
2028 }
2029 }
2030 up_read(&mdsc->pool_perm_rwsem);
2031 if (*p)
2032 goto out;
2033
2034 if (pool_ns)
2035 doutc(cl, "pool %lld ns %.*s no perm cached\n", pool,
2036 (int)pool_ns->len, pool_ns->str);
2037 else
2038 doutc(cl, "pool %lld no perm cached\n", pool);
2039
2040 down_write(&mdsc->pool_perm_rwsem);
2041 p = &mdsc->pool_perm_tree.rb_node;
2042 parent = NULL;
2043 while (*p) {
2044 parent = *p;
2045 perm = rb_entry(parent, struct ceph_pool_perm, node);
2046 if (pool < perm->pool)
2047 p = &(*p)->rb_left;
2048 else if (pool > perm->pool)
2049 p = &(*p)->rb_right;
2050 else {
2051 int ret = ceph_compare_string(pool_ns,
2052 perm->pool_ns,
2053 perm->pool_ns_len);
2054 if (ret < 0)
2055 p = &(*p)->rb_left;
2056 else if (ret > 0)
2057 p = &(*p)->rb_right;
2058 else {
2059 have = perm->perm;
2060 break;
2061 }
2062 }
2063 }
2064 if (*p) {
2065 up_write(&mdsc->pool_perm_rwsem);
2066 goto out;
2067 }
2068
2069 rd_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
2070 1, false, GFP_NOFS);
2071 if (!rd_req) {
2072 err = -ENOMEM;
2073 goto out_unlock;
2074 }
2075
2076 rd_req->r_flags = CEPH_OSD_FLAG_READ;
2077 osd_req_op_init(rd_req, 0, CEPH_OSD_OP_STAT, 0);
2078 rd_req->r_base_oloc.pool = pool;
2079 if (pool_ns)
2080 rd_req->r_base_oloc.pool_ns = ceph_get_string(pool_ns);
2081 ceph_oid_printf(&rd_req->r_base_oid, "%llx.00000000", ci->i_vino.ino);
2082
2083 err = ceph_osdc_alloc_messages(rd_req, GFP_NOFS);
2084 if (err)
2085 goto out_unlock;
2086
2087 wr_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
2088 1, false, GFP_NOFS);
2089 if (!wr_req) {
2090 err = -ENOMEM;
2091 goto out_unlock;
2092 }
2093
2094 wr_req->r_flags = CEPH_OSD_FLAG_WRITE;
2095 osd_req_op_init(wr_req, 0, CEPH_OSD_OP_CREATE, CEPH_OSD_OP_FLAG_EXCL);
2096 ceph_oloc_copy(&wr_req->r_base_oloc, &rd_req->r_base_oloc);
2097 ceph_oid_copy(&wr_req->r_base_oid, &rd_req->r_base_oid);
2098
2099 err = ceph_osdc_alloc_messages(wr_req, GFP_NOFS);
2100 if (err)
2101 goto out_unlock;
2102
2103 /* one page should be large enough for STAT data */
2104 pages = ceph_alloc_page_vector(1, GFP_KERNEL);
2105 if (IS_ERR(pages)) {
2106 err = PTR_ERR(pages);
2107 goto out_unlock;
2108 }
2109
2110 osd_req_op_raw_data_in_pages(rd_req, 0, pages, PAGE_SIZE,
2111 0, false, true);
2112 ceph_osdc_start_request(&fsc->client->osdc, rd_req);
2113
2114 wr_req->r_mtime = inode_get_mtime(&ci->netfs.inode);
2115 ceph_osdc_start_request(&fsc->client->osdc, wr_req);
2116
2117 err = ceph_osdc_wait_request(&fsc->client->osdc, rd_req);
2118 err2 = ceph_osdc_wait_request(&fsc->client->osdc, wr_req);
2119
2120 if (err >= 0 || err == -ENOENT)
2121 have |= POOL_READ;
2122 else if (err != -EPERM) {
2123 if (err == -EBLOCKLISTED)
2124 fsc->blocklisted = true;
2125 goto out_unlock;
2126 }
2127
2128 if (err2 == 0 || err2 == -EEXIST)
2129 have |= POOL_WRITE;
2130 else if (err2 != -EPERM) {
2131 if (err2 == -EBLOCKLISTED)
2132 fsc->blocklisted = true;
2133 err = err2;
2134 goto out_unlock;
2135 }
2136
2137 pool_ns_len = pool_ns ? pool_ns->len : 0;
2138 perm = kmalloc(sizeof(*perm) + pool_ns_len + 1, GFP_NOFS);
2139 if (!perm) {
2140 err = -ENOMEM;
2141 goto out_unlock;
2142 }
2143
2144 perm->pool = pool;
2145 perm->perm = have;
2146 perm->pool_ns_len = pool_ns_len;
2147 if (pool_ns_len > 0)
2148 memcpy(perm->pool_ns, pool_ns->str, pool_ns_len);
2149 perm->pool_ns[pool_ns_len] = 0;
2150
2151 rb_link_node(&perm->node, parent, p);
2152 rb_insert_color(&perm->node, &mdsc->pool_perm_tree);
2153 err = 0;
2154 out_unlock:
2155 up_write(&mdsc->pool_perm_rwsem);
2156
2157 ceph_osdc_put_request(rd_req);
2158 ceph_osdc_put_request(wr_req);
2159 out:
2160 if (!err)
2161 err = have;
2162 if (pool_ns)
2163 doutc(cl, "pool %lld ns %.*s result = %d\n", pool,
2164 (int)pool_ns->len, pool_ns->str, err);
2165 else
2166 doutc(cl, "pool %lld result = %d\n", pool, err);
2167 return err;
2168 }
2169
2170 int ceph_pool_perm_check(struct inode *inode, int need)
2171 {
2172 struct ceph_client *cl = ceph_inode_to_client(inode);
2173 struct ceph_inode_info *ci = ceph_inode(inode);
2174 struct ceph_string *pool_ns;
2175 s64 pool;
2176 int ret, flags;
2177
2178 /* Only need to do this for regular files */
2179 if (!S_ISREG(inode->i_mode))
2180 return 0;
2181
2182 if (ci->i_vino.snap != CEPH_NOSNAP) {
2183 /*
2184 * Pool permission check needs to write to the first object.
2185 * But for snapshot, head of the first object may have alread
2186 * been deleted. Skip check to avoid creating orphan object.
2187 */
2188 return 0;
2189 }
2190
2191 if (ceph_test_mount_opt(ceph_inode_to_fs_client(inode),
2192 NOPOOLPERM))
2193 return 0;
2194
2195 spin_lock(&ci->i_ceph_lock);
2196 flags = ci->i_ceph_flags;
2197 pool = ci->i_layout.pool_id;
2198 spin_unlock(&ci->i_ceph_lock);
2199 check:
2200 if (flags & CEPH_I_POOL_PERM) {
2201 if ((need & CEPH_CAP_FILE_RD) && !(flags & CEPH_I_POOL_RD)) {
2202 doutc(cl, "pool %lld no read perm\n", pool);
2203 return -EPERM;
2204 }
2205 if ((need & CEPH_CAP_FILE_WR) && !(flags & CEPH_I_POOL_WR)) {
2206 doutc(cl, "pool %lld no write perm\n", pool);
2207 return -EPERM;
2208 }
2209 return 0;
2210 }
2211
2212 pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
2213 ret = __ceph_pool_perm_get(ci, pool, pool_ns);
2214 ceph_put_string(pool_ns);
2215 if (ret < 0)
2216 return ret;
2217
2218 flags = CEPH_I_POOL_PERM;
2219 if (ret & POOL_READ)
2220 flags |= CEPH_I_POOL_RD;
2221 if (ret & POOL_WRITE)
2222 flags |= CEPH_I_POOL_WR;
2223
2224 spin_lock(&ci->i_ceph_lock);
2225 if (pool == ci->i_layout.pool_id &&
2226 pool_ns == rcu_dereference_raw(ci->i_layout.pool_ns)) {
2227 ci->i_ceph_flags |= flags;
2228 } else {
2229 pool = ci->i_layout.pool_id;
2230 flags = ci->i_ceph_flags;
2231 }
2232 spin_unlock(&ci->i_ceph_lock);
2233 goto check;
2234 }
2235
2236 void ceph_pool_perm_destroy(struct ceph_mds_client *mdsc)
2237 {
2238 struct ceph_pool_perm *perm;
2239 struct rb_node *n;
2240
2241 while (!RB_EMPTY_ROOT(&mdsc->pool_perm_tree)) {
2242 n = rb_first(&mdsc->pool_perm_tree);
2243 perm = rb_entry(n, struct ceph_pool_perm, node);
2244 rb_erase(n, &mdsc->pool_perm_tree);
2245 kfree(perm);
2246 }
2247 }
2248
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.