?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 4.19.133
[linux/fpc-iii.git] / fs / ceph / addr.c
blob476728bdae8c668d3e3eb3eed90f3a892edae9bf
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/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
14 #include "super.h"
15 #include "mds_client.h"
16 #include "cache.h"
17 #include <linux/ceph/osd_client.h>
18 #include <linux/ceph/striper.h>
19
20 /*
21 * Ceph address space ops.
22 *
23 * There are a few funny things going on here.
24 *
25 * The page->private field is used to reference a struct
26 * ceph_snap_context for _every_ dirty page. This indicates which
27 * snapshot the page was logically dirtied in, and thus which snap
28 * context needs to be associated with the osd write during writeback.
29 *
30 * Similarly, struct ceph_inode_info maintains a set of counters to
31 * count dirty pages on the inode. In the absence of snapshots,
32 * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count.
33 *
34 * When a snapshot is taken (that is, when the client receives
35 * notification that a snapshot was taken), each inode with caps and
36 * with dirty pages (dirty pages implies there is a cap) gets a new
37 * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending
38 * order, new snaps go to the tail). The i_wrbuffer_ref_head count is
39 * moved to capsnap->dirty. (Unless a sync write is currently in
40 * progress. In that case, the capsnap is said to be "pending", new
41 * writes cannot start, and the capsnap isn't "finalized" until the
42 * write completes (or fails) and a final size/mtime for the inode for
43 * that snap can be settled upon.) i_wrbuffer_ref_head is reset to 0.
44 *
45 * On writeback, we must submit writes to the osd IN SNAP ORDER. So,
46 * we look for the first capsnap in i_cap_snaps and write out pages in
47 * that snap context _only_. Then we move on to the next capsnap,
48 * eventually reaching the "live" or "head" context (i.e., pages that
49 * are not yet snapped) and are writing the most recently dirtied
50 * pages.
51 *
52 * Invalidate and so forth must take care to ensure the dirty page
53 * accounting is preserved.
54 */
55
56 #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10))
57 #define CONGESTION_OFF_THRESH(congestion_kb) \
58 (CONGESTION_ON_THRESH(congestion_kb) - \
59 (CONGESTION_ON_THRESH(congestion_kb) >> 2))
60
61 static inline struct ceph_snap_context *page_snap_context(struct page *page)
62 {
63 if (PagePrivate(page))
64 return (void *)page->private;
65 return NULL;
66 }
67
68 /*
69 * Dirty a page. Optimistically adjust accounting, on the assumption
70 * that we won't race with invalidate. If we do, readjust.
71 */
72 static int ceph_set_page_dirty(struct page *page)
73 {
74 struct address_space *mapping = page->mapping;
75 struct inode *inode;
76 struct ceph_inode_info *ci;
77 struct ceph_snap_context *snapc;
78 int ret;
79
80 if (unlikely(!mapping))
81 return !TestSetPageDirty(page);
82
83 if (PageDirty(page)) {
84 dout("%p set_page_dirty %p idx %lu -- already dirty\n",
85 mapping->host, page, page->index);
86 BUG_ON(!PagePrivate(page));
87 return 0;
88 }
89
90 inode = mapping->host;
91 ci = ceph_inode(inode);
92
93 /* dirty the head */
94 spin_lock(&ci->i_ceph_lock);
95 BUG_ON(ci->i_wr_ref == 0); // caller should hold Fw reference
96 if (__ceph_have_pending_cap_snap(ci)) {
97 struct ceph_cap_snap *capsnap =
98 list_last_entry(&ci->i_cap_snaps,
99 struct ceph_cap_snap,
100 ci_item);
101 snapc = ceph_get_snap_context(capsnap->context);
102 capsnap->dirty_pages++;
103 } else {
104 BUG_ON(!ci->i_head_snapc);
105 snapc = ceph_get_snap_context(ci->i_head_snapc);
106 ++ci->i_wrbuffer_ref_head;
107 }
108 if (ci->i_wrbuffer_ref == 0)
109 ihold(inode);
110 ++ci->i_wrbuffer_ref;
111 dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d "
112 "snapc %p seq %lld (%d snaps)\n",
113 mapping->host, page, page->index,
114 ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1,
115 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
116 snapc, snapc->seq, snapc->num_snaps);
117 spin_unlock(&ci->i_ceph_lock);
118
119 /*
120 * Reference snap context in page->private. Also set
121 * PagePrivate so that we get invalidatepage callback.
122 */
123 BUG_ON(PagePrivate(page));
124 page->private = (unsigned long)snapc;
125 SetPagePrivate(page);
126
127 ret = __set_page_dirty_nobuffers(page);
128 WARN_ON(!PageLocked(page));
129 WARN_ON(!page->mapping);
130
131 return ret;
132 }
133
134 /*
135 * If we are truncating the full page (i.e. offset == 0), adjust the
136 * dirty page counters appropriately. Only called if there is private
137 * data on the page.
138 */
139 static void ceph_invalidatepage(struct page *page, unsigned int offset,
140 unsigned int length)
141 {
142 struct inode *inode;
143 struct ceph_inode_info *ci;
144 struct ceph_snap_context *snapc = page_snap_context(page);
145
146 inode = page->mapping->host;
147 ci = ceph_inode(inode);
148
149 if (offset != 0 || length != PAGE_SIZE) {
150 dout("%p invalidatepage %p idx %lu partial dirty page %u~%u\n",
151 inode, page, page->index, offset, length);
152 return;
153 }
154
155 ceph_invalidate_fscache_page(inode, page);
156
157 WARN_ON(!PageLocked(page));
158 if (!PagePrivate(page))
159 return;
160
161 ClearPageChecked(page);
162
163 dout("%p invalidatepage %p idx %lu full dirty page\n",
164 inode, page, page->index);
165
166 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
167 ceph_put_snap_context(snapc);
168 page->private = 0;
169 ClearPagePrivate(page);
170 }
171
172 static int ceph_releasepage(struct page *page, gfp_t g)
173 {
174 dout("%p releasepage %p idx %lu (%sdirty)\n", page->mapping->host,
175 page, page->index, PageDirty(page) ? "" : "not ");
176
177 /* Can we release the page from the cache? */
178 if (!ceph_release_fscache_page(page, g))
179 return 0;
180
181 return !PagePrivate(page);
182 }
183
184 /*
185 * read a single page, without unlocking it.
186 */
187 static int ceph_do_readpage(struct file *filp, struct page *page)
188 {
189 struct inode *inode = file_inode(filp);
190 struct ceph_inode_info *ci = ceph_inode(inode);
191 struct ceph_osd_client *osdc =
192 &ceph_inode_to_client(inode)->client->osdc;
193 int err = 0;
194 u64 off = page_offset(page);
195 u64 len = PAGE_SIZE;
196
197 if (off >= i_size_read(inode)) {
198 zero_user_segment(page, 0, PAGE_SIZE);
199 SetPageUptodate(page);
200 return 0;
201 }
202
203 if (ci->i_inline_version != CEPH_INLINE_NONE) {
204 /*
205 * Uptodate inline data should have been added
206 * into page cache while getting Fcr caps.
207 */
208 if (off == 0)
209 return -EINVAL;
210 zero_user_segment(page, 0, PAGE_SIZE);
211 SetPageUptodate(page);
212 return 0;
213 }
214
215 err = ceph_readpage_from_fscache(inode, page);
216 if (err == 0)
217 return -EINPROGRESS;
218
219 dout("readpage inode %p file %p page %p index %lu\n",
220 inode, filp, page, page->index);
221 err = ceph_osdc_readpages(osdc, ceph_vino(inode), &ci->i_layout,
222 off, &len,
223 ci->i_truncate_seq, ci->i_truncate_size,
224 &page, 1, 0);
225 if (err == -ENOENT)
226 err = 0;
227 if (err < 0) {
228 SetPageError(page);
229 ceph_fscache_readpage_cancel(inode, page);
230 goto out;
231 }
232 if (err < PAGE_SIZE)
233 /* zero fill remainder of page */
234 zero_user_segment(page, err, PAGE_SIZE);
235 else
236 flush_dcache_page(page);
237
238 SetPageUptodate(page);
239 ceph_readpage_to_fscache(inode, page);
240
241 out:
242 return err < 0 ? err : 0;
243 }
244
245 static int ceph_readpage(struct file *filp, struct page *page)
246 {
247 int r = ceph_do_readpage(filp, page);
248 if (r != -EINPROGRESS)
249 unlock_page(page);
250 else
251 r = 0;
252 return r;
253 }
254
255 /*
256 * Finish an async read(ahead) op.
257 */
258 static void finish_read(struct ceph_osd_request *req)
259 {
260 struct inode *inode = req->r_inode;
261 struct ceph_osd_data *osd_data;
262 int rc = req->r_result <= 0 ? req->r_result : 0;
263 int bytes = req->r_result >= 0 ? req->r_result : 0;
264 int num_pages;
265 int i;
266
267 dout("finish_read %p req %p rc %d bytes %d\n", inode, req, rc, bytes);
268
269 /* unlock all pages, zeroing any data we didn't read */
270 osd_data = osd_req_op_extent_osd_data(req, 0);
271 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
272 num_pages = calc_pages_for((u64)osd_data->alignment,
273 (u64)osd_data->length);
274 for (i = 0; i < num_pages; i++) {
275 struct page *page = osd_data->pages[i];
276
277 if (rc < 0 && rc != -ENOENT) {
278 ceph_fscache_readpage_cancel(inode, page);
279 goto unlock;
280 }
281 if (bytes < (int)PAGE_SIZE) {
282 /* zero (remainder of) page */
283 int s = bytes < 0 ? 0 : bytes;
284 zero_user_segment(page, s, PAGE_SIZE);
285 }
286 dout("finish_read %p uptodate %p idx %lu\n", inode, page,
287 page->index);
288 flush_dcache_page(page);
289 SetPageUptodate(page);
290 ceph_readpage_to_fscache(inode, page);
291 unlock:
292 unlock_page(page);
293 put_page(page);
294 bytes -= PAGE_SIZE;
295 }
296 kfree(osd_data->pages);
297 }
298
299 /*
300 * start an async read(ahead) operation. return nr_pages we submitted
301 * a read for on success, or negative error code.
302 */
303 static int start_read(struct inode *inode, struct ceph_rw_context *rw_ctx,
304 struct list_head *page_list, int max)
305 {
306 struct ceph_osd_client *osdc =
307 &ceph_inode_to_client(inode)->client->osdc;
308 struct ceph_inode_info *ci = ceph_inode(inode);
309 struct page *page = list_entry(page_list->prev, struct page, lru);
310 struct ceph_vino vino;
311 struct ceph_osd_request *req;
312 u64 off;
313 u64 len;
314 int i;
315 struct page **pages;
316 pgoff_t next_index;
317 int nr_pages = 0;
318 int got = 0;
319 int ret = 0;
320
321 if (!rw_ctx) {
322 /* caller of readpages does not hold buffer and read caps
323 * (fadvise, madvise and readahead cases) */
324 int want = CEPH_CAP_FILE_CACHE;
325 ret = ceph_try_get_caps(ci, CEPH_CAP_FILE_RD, want, &got);
326 if (ret < 0) {
327 dout("start_read %p, error getting cap\n", inode);
328 } else if (!(got & want)) {
329 dout("start_read %p, no cache cap\n", inode);
330 ret = 0;
331 }
332 if (ret <= 0) {
333 if (got)
334 ceph_put_cap_refs(ci, got);
335 while (!list_empty(page_list)) {
336 page = list_entry(page_list->prev,
337 struct page, lru);
338 list_del(&page->lru);
339 put_page(page);
340 }
341 return ret;
342 }
343 }
344
345 off = (u64) page_offset(page);
346
347 /* count pages */
348 next_index = page->index;
349 list_for_each_entry_reverse(page, page_list, lru) {
350 if (page->index != next_index)
351 break;
352 nr_pages++;
353 next_index++;
354 if (max && nr_pages == max)
355 break;
356 }
357 len = nr_pages << PAGE_SHIFT;
358 dout("start_read %p nr_pages %d is %lld~%lld\n", inode, nr_pages,
359 off, len);
360 vino = ceph_vino(inode);
361 req = ceph_osdc_new_request(osdc, &ci->i_layout, vino, off, &len,
362 0, 1, CEPH_OSD_OP_READ,
363 CEPH_OSD_FLAG_READ, NULL,
364 ci->i_truncate_seq, ci->i_truncate_size,
365 false);
366 if (IS_ERR(req)) {
367 ret = PTR_ERR(req);
368 goto out;
369 }
370
371 /* build page vector */
372 nr_pages = calc_pages_for(0, len);
373 pages = kmalloc_array(nr_pages, sizeof(*pages), GFP_KERNEL);
374 if (!pages) {
375 ret = -ENOMEM;
376 goto out_put;
377 }
378 for (i = 0; i < nr_pages; ++i) {
379 page = list_entry(page_list->prev, struct page, lru);
380 BUG_ON(PageLocked(page));
381 list_del(&page->lru);
382
383 dout("start_read %p adding %p idx %lu\n", inode, page,
384 page->index);
385 if (add_to_page_cache_lru(page, &inode->i_data, page->index,
386 GFP_KERNEL)) {
387 ceph_fscache_uncache_page(inode, page);
388 put_page(page);
389 dout("start_read %p add_to_page_cache failed %p\n",
390 inode, page);
391 nr_pages = i;
392 if (nr_pages > 0) {
393 len = nr_pages << PAGE_SHIFT;
394 osd_req_op_extent_update(req, 0, len);
395 break;
396 }
397 goto out_pages;
398 }
399 pages[i] = page;
400 }
401 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false);
402 req->r_callback = finish_read;
403 req->r_inode = inode;
404
405 dout("start_read %p starting %p %lld~%lld\n", inode, req, off, len);
406 ret = ceph_osdc_start_request(osdc, req, false);
407 if (ret < 0)
408 goto out_pages;
409 ceph_osdc_put_request(req);
410
411 /* After adding locked pages to page cache, the inode holds cache cap.
412 * So we can drop our cap refs. */
413 if (got)
414 ceph_put_cap_refs(ci, got);
415
416 return nr_pages;
417
418 out_pages:
419 for (i = 0; i < nr_pages; ++i) {
420 ceph_fscache_readpage_cancel(inode, pages[i]);
421 unlock_page(pages[i]);
422 }
423 ceph_put_page_vector(pages, nr_pages, false);
424 out_put:
425 ceph_osdc_put_request(req);
426 out:
427 if (got)
428 ceph_put_cap_refs(ci, got);
429 return ret;
430 }
431
432
433 /*
434 * Read multiple pages. Leave pages we don't read + unlock in page_list;
435 * the caller (VM) cleans them up.
436 */
437 static int ceph_readpages(struct file *file, struct address_space *mapping,
438 struct list_head *page_list, unsigned nr_pages)
439 {
440 struct inode *inode = file_inode(file);
441 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
442 struct ceph_file_info *fi = file->private_data;
443 struct ceph_rw_context *rw_ctx;
444 int rc = 0;
445 int max = 0;
446
447 if (ceph_inode(inode)->i_inline_version != CEPH_INLINE_NONE)
448 return -EINVAL;
449
450 rc = ceph_readpages_from_fscache(mapping->host, mapping, page_list,
451 &nr_pages);
452
453 if (rc == 0)
454 goto out;
455
456 rw_ctx = ceph_find_rw_context(fi);
457 max = fsc->mount_options->rsize >> PAGE_SHIFT;
458 dout("readpages %p file %p ctx %p nr_pages %d max %d\n",
459 inode, file, rw_ctx, nr_pages, max);
460 while (!list_empty(page_list)) {
461 rc = start_read(inode, rw_ctx, page_list, max);
462 if (rc < 0)
463 goto out;
464 }
465 out:
466 ceph_fscache_readpages_cancel(inode, page_list);
467
468 dout("readpages %p file %p ret %d\n", inode, file, rc);
469 return rc;
470 }
471
472 struct ceph_writeback_ctl
473 {
474 loff_t i_size;
475 u64 truncate_size;
476 u32 truncate_seq;
477 bool size_stable;
478 bool head_snapc;
479 };
480
481 /*
482 * Get ref for the oldest snapc for an inode with dirty data... that is, the
483 * only snap context we are allowed to write back.
484 */
485 static struct ceph_snap_context *
486 get_oldest_context(struct inode *inode, struct ceph_writeback_ctl *ctl,
487 struct ceph_snap_context *page_snapc)
488 {
489 struct ceph_inode_info *ci = ceph_inode(inode);
490 struct ceph_snap_context *snapc = NULL;
491 struct ceph_cap_snap *capsnap = NULL;
492
493 spin_lock(&ci->i_ceph_lock);
494 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
495 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap,
496 capsnap->context, capsnap->dirty_pages);
497 if (!capsnap->dirty_pages)
498 continue;
499
500 /* get i_size, truncate_{seq,size} for page_snapc? */
501 if (snapc && capsnap->context != page_snapc)
502 continue;
503
504 if (ctl) {
505 if (capsnap->writing) {
506 ctl->i_size = i_size_read(inode);
507 ctl->size_stable = false;
508 } else {
509 ctl->i_size = capsnap->size;
510 ctl->size_stable = true;
511 }
512 ctl->truncate_size = capsnap->truncate_size;
513 ctl->truncate_seq = capsnap->truncate_seq;
514 ctl->head_snapc = false;
515 }
516
517 if (snapc)
518 break;
519
520 snapc = ceph_get_snap_context(capsnap->context);
521 if (!page_snapc ||
522 page_snapc == snapc ||
523 page_snapc->seq > snapc->seq)
524 break;
525 }
526 if (!snapc && ci->i_wrbuffer_ref_head) {
527 snapc = ceph_get_snap_context(ci->i_head_snapc);
528 dout(" head snapc %p has %d dirty pages\n",
529 snapc, ci->i_wrbuffer_ref_head);
530 if (ctl) {
531 ctl->i_size = i_size_read(inode);
532 ctl->truncate_size = ci->i_truncate_size;
533 ctl->truncate_seq = ci->i_truncate_seq;
534 ctl->size_stable = false;
535 ctl->head_snapc = true;
536 }
537 }
538 spin_unlock(&ci->i_ceph_lock);
539 return snapc;
540 }
541
542 static u64 get_writepages_data_length(struct inode *inode,
543 struct page *page, u64 start)
544 {
545 struct ceph_inode_info *ci = ceph_inode(inode);
546 struct ceph_snap_context *snapc = page_snap_context(page);
547 struct ceph_cap_snap *capsnap = NULL;
548 u64 end = i_size_read(inode);
549
550 if (snapc != ci->i_head_snapc) {
551 bool found = false;
552 spin_lock(&ci->i_ceph_lock);
553 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
554 if (capsnap->context == snapc) {
555 if (!capsnap->writing)
556 end = capsnap->size;
557 found = true;
558 break;
559 }
560 }
561 spin_unlock(&ci->i_ceph_lock);
562 WARN_ON(!found);
563 }
564 if (end > page_offset(page) + PAGE_SIZE)
565 end = page_offset(page) + PAGE_SIZE;
566 return end > start ? end - start : 0;
567 }
568
569 /*
570 * Write a single page, but leave the page locked.
571 *
572 * If we get a write error, set the page error bit, but still adjust the
573 * dirty page accounting (i.e., page is no longer dirty).
574 */
575 static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
576 {
577 struct inode *inode;
578 struct ceph_inode_info *ci;
579 struct ceph_fs_client *fsc;
580 struct ceph_snap_context *snapc, *oldest;
581 loff_t page_off = page_offset(page);
582 int err, len = PAGE_SIZE;
583 struct ceph_writeback_ctl ceph_wbc;
584
585 dout("writepage %p idx %lu\n", page, page->index);
586
587 inode = page->mapping->host;
588 ci = ceph_inode(inode);
589 fsc = ceph_inode_to_client(inode);
590
591 /* verify this is a writeable snap context */
592 snapc = page_snap_context(page);
593 if (!snapc) {
594 dout("writepage %p page %p not dirty?\n", inode, page);
595 return 0;
596 }
597 oldest = get_oldest_context(inode, &ceph_wbc, snapc);
598 if (snapc->seq > oldest->seq) {
599 dout("writepage %p page %p snapc %p not writeable - noop\n",
600 inode, page, snapc);
601 /* we should only noop if called by kswapd */
602 WARN_ON(!(current->flags & PF_MEMALLOC));
603 ceph_put_snap_context(oldest);
604 redirty_page_for_writepage(wbc, page);
605 return 0;
606 }
607 ceph_put_snap_context(oldest);
608
609 /* is this a partial page at end of file? */
610 if (page_off >= ceph_wbc.i_size) {
611 dout("%p page eof %llu\n", page, ceph_wbc.i_size);
612 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
613 return 0;
614 }
615
616 if (ceph_wbc.i_size < page_off + len)
617 len = ceph_wbc.i_size - page_off;
618
619 dout("writepage %p page %p index %lu on %llu~%u snapc %p seq %lld\n",
620 inode, page, page->index, page_off, len, snapc, snapc->seq);
621
622 if (atomic_long_inc_return(&fsc->writeback_count) >
623 CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb))
624 set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC);
625
626 set_page_writeback(page);
627 err = ceph_osdc_writepages(&fsc->client->osdc, ceph_vino(inode),
628 &ci->i_layout, snapc, page_off, len,
629 ceph_wbc.truncate_seq,
630 ceph_wbc.truncate_size,
631 &inode->i_mtime, &page, 1);
632 if (err < 0) {
633 struct writeback_control tmp_wbc;
634 if (!wbc)
635 wbc = &tmp_wbc;
636 if (err == -ERESTARTSYS) {
637 /* killed by SIGKILL */
638 dout("writepage interrupted page %p\n", page);
639 redirty_page_for_writepage(wbc, page);
640 end_page_writeback(page);
641 return err;
642 }
643 dout("writepage setting page/mapping error %d %p\n",
644 err, page);
645 SetPageError(page);
646 mapping_set_error(&inode->i_data, err);
647 wbc->pages_skipped++;
648 } else {
649 dout("writepage cleaned page %p\n", page);
650 err = 0; /* vfs expects us to return 0 */
651 }
652 page->private = 0;
653 ClearPagePrivate(page);
654 end_page_writeback(page);
655 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
656 ceph_put_snap_context(snapc); /* page's reference */
657
658 if (atomic_long_dec_return(&fsc->writeback_count) <
659 CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb))
660 clear_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC);
661
662 return err;
663 }
664
665 static int ceph_writepage(struct page *page, struct writeback_control *wbc)
666 {
667 int err;
668 struct inode *inode = page->mapping->host;
669 BUG_ON(!inode);
670 ihold(inode);
671 err = writepage_nounlock(page, wbc);
672 if (err == -ERESTARTSYS) {
673 /* direct memory reclaimer was killed by SIGKILL. return 0
674 * to prevent caller from setting mapping/page error */
675 err = 0;
676 }
677 unlock_page(page);
678 iput(inode);
679 return err;
680 }
681
682 /*
683 * lame release_pages helper. release_pages() isn't exported to
684 * modules.
685 */
686 static void ceph_release_pages(struct page **pages, int num)
687 {
688 struct pagevec pvec;
689 int i;
690
691 pagevec_init(&pvec);
692 for (i = 0; i < num; i++) {
693 if (pagevec_add(&pvec, pages[i]) == 0)
694 pagevec_release(&pvec);
695 }
696 pagevec_release(&pvec);
697 }
698
699 /*
700 * async writeback completion handler.
701 *
702 * If we get an error, set the mapping error bit, but not the individual
703 * page error bits.
704 */
705 static void writepages_finish(struct ceph_osd_request *req)
706 {
707 struct inode *inode = req->r_inode;
708 struct ceph_inode_info *ci = ceph_inode(inode);
709 struct ceph_osd_data *osd_data;
710 struct page *page;
711 int num_pages, total_pages = 0;
712 int i, j;
713 int rc = req->r_result;
714 struct ceph_snap_context *snapc = req->r_snapc;
715 struct address_space *mapping = inode->i_mapping;
716 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
717 bool remove_page;
718
719 dout("writepages_finish %p rc %d\n", inode, rc);
720 if (rc < 0) {
721 mapping_set_error(mapping, rc);
722 ceph_set_error_write(ci);
723 } else {
724 ceph_clear_error_write(ci);
725 }
726
727 /*
728 * We lost the cache cap, need to truncate the page before
729 * it is unlocked, otherwise we'd truncate it later in the
730 * page truncation thread, possibly losing some data that
731 * raced its way in
732 */
733 remove_page = !(ceph_caps_issued(ci) &
734 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO));
735
736 /* clean all pages */
737 for (i = 0; i < req->r_num_ops; i++) {
738 if (req->r_ops[i].op != CEPH_OSD_OP_WRITE)
739 break;
740
741 osd_data = osd_req_op_extent_osd_data(req, i);
742 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
743 num_pages = calc_pages_for((u64)osd_data->alignment,
744 (u64)osd_data->length);
745 total_pages += num_pages;
746 for (j = 0; j < num_pages; j++) {
747 page = osd_data->pages[j];
748 BUG_ON(!page);
749 WARN_ON(!PageUptodate(page));
750
751 if (atomic_long_dec_return(&fsc->writeback_count) <
752 CONGESTION_OFF_THRESH(
753 fsc->mount_options->congestion_kb))
754 clear_bdi_congested(inode_to_bdi(inode),
755 BLK_RW_ASYNC);
756
757 ceph_put_snap_context(page_snap_context(page));
758 page->private = 0;
759 ClearPagePrivate(page);
760 dout("unlocking %p\n", page);
761 end_page_writeback(page);
762
763 if (remove_page)
764 generic_error_remove_page(inode->i_mapping,
765 page);
766
767 unlock_page(page);
768 }
769 dout("writepages_finish %p wrote %llu bytes cleaned %d pages\n",
770 inode, osd_data->length, rc >= 0 ? num_pages : 0);
771
772 ceph_release_pages(osd_data->pages, num_pages);
773 }
774
775 ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc);
776
777 osd_data = osd_req_op_extent_osd_data(req, 0);
778 if (osd_data->pages_from_pool)
779 mempool_free(osd_data->pages,
780 ceph_sb_to_client(inode->i_sb)->wb_pagevec_pool);
781 else
782 kfree(osd_data->pages);
783 ceph_osdc_put_request(req);
784 }
785
786 /*
787 * initiate async writeback
788 */
789 static int ceph_writepages_start(struct address_space *mapping,
790 struct writeback_control *wbc)
791 {
792 struct inode *inode = mapping->host;
793 struct ceph_inode_info *ci = ceph_inode(inode);
794 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
795 struct ceph_vino vino = ceph_vino(inode);
796 pgoff_t index, start_index, end = -1;
797 struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc;
798 struct pagevec pvec;
799 int rc = 0;
800 unsigned int wsize = i_blocksize(inode);
801 struct ceph_osd_request *req = NULL;
802 struct ceph_writeback_ctl ceph_wbc;
803 bool should_loop, range_whole = false;
804 bool done = false;
805
806 dout("writepages_start %p (mode=%s)\n", inode,
807 wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
808 (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
809
810 if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
811 if (ci->i_wrbuffer_ref > 0) {
812 pr_warn_ratelimited(
813 "writepage_start %p %lld forced umount\n",
814 inode, ceph_ino(inode));
815 }
816 mapping_set_error(mapping, -EIO);
817 return -EIO; /* we're in a forced umount, don't write! */
818 }
819 if (fsc->mount_options->wsize < wsize)
820 wsize = fsc->mount_options->wsize;
821
822 pagevec_init(&pvec);
823
824 start_index = wbc->range_cyclic ? mapping->writeback_index : 0;
825 index = start_index;
826
827 retry:
828 /* find oldest snap context with dirty data */
829 snapc = get_oldest_context(inode, &ceph_wbc, NULL);
830 if (!snapc) {
831 /* hmm, why does writepages get called when there
832 is no dirty data? */
833 dout(" no snap context with dirty data?\n");
834 goto out;
835 }
836 dout(" oldest snapc is %p seq %lld (%d snaps)\n",
837 snapc, snapc->seq, snapc->num_snaps);
838
839 should_loop = false;
840 if (ceph_wbc.head_snapc && snapc != last_snapc) {
841 /* where to start/end? */
842 if (wbc->range_cyclic) {
843 index = start_index;
844 end = -1;
845 if (index > 0)
846 should_loop = true;
847 dout(" cyclic, start at %lu\n", index);
848 } else {
849 index = wbc->range_start >> PAGE_SHIFT;
850 end = wbc->range_end >> PAGE_SHIFT;
851 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
852 range_whole = true;
853 dout(" not cyclic, %lu to %lu\n", index, end);
854 }
855 } else if (!ceph_wbc.head_snapc) {
856 /* Do not respect wbc->range_{start,end}. Dirty pages
857 * in that range can be associated with newer snapc.
858 * They are not writeable until we write all dirty pages
859 * associated with 'snapc' get written */
860 if (index > 0)
861 should_loop = true;
862 dout(" non-head snapc, range whole\n");
863 }
864
865 ceph_put_snap_context(last_snapc);
866 last_snapc = snapc;
867
868 while (!done && index <= end) {
869 int num_ops = 0, op_idx;
870 unsigned i, pvec_pages, max_pages, locked_pages = 0;
871 struct page **pages = NULL, **data_pages;
872 mempool_t *pool = NULL; /* Becomes non-null if mempool used */
873 struct page *page;
874 pgoff_t strip_unit_end = 0;
875 u64 offset = 0, len = 0;
876
877 max_pages = wsize >> PAGE_SHIFT;
878
879 get_more_pages:
880 pvec_pages = pagevec_lookup_range_nr_tag(&pvec, mapping, &index,
881 end, PAGECACHE_TAG_DIRTY,
882 max_pages - locked_pages);
883 dout("pagevec_lookup_range_tag got %d\n", pvec_pages);
884 if (!pvec_pages && !locked_pages)
885 break;
886 for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) {
887 page = pvec.pages[i];
888 dout("? %p idx %lu\n", page, page->index);
889 if (locked_pages == 0)
890 lock_page(page); /* first page */
891 else if (!trylock_page(page))
892 break;
893
894 /* only dirty pages, or our accounting breaks */
895 if (unlikely(!PageDirty(page)) ||
896 unlikely(page->mapping != mapping)) {
897 dout("!dirty or !mapping %p\n", page);
898 unlock_page(page);
899 continue;
900 }
901 /* only if matching snap context */
902 pgsnapc = page_snap_context(page);
903 if (pgsnapc != snapc) {
904 dout("page snapc %p %lld != oldest %p %lld\n",
905 pgsnapc, pgsnapc->seq, snapc, snapc->seq);
906 if (!should_loop &&
907 !ceph_wbc.head_snapc &&
908 wbc->sync_mode != WB_SYNC_NONE)
909 should_loop = true;
910 unlock_page(page);
911 continue;
912 }
913 if (page_offset(page) >= ceph_wbc.i_size) {
914 dout("%p page eof %llu\n",
915 page, ceph_wbc.i_size);
916 if ((ceph_wbc.size_stable ||
917 page_offset(page) >= i_size_read(inode)) &&
918 clear_page_dirty_for_io(page))
919 mapping->a_ops->invalidatepage(page,
920 0, PAGE_SIZE);
921 unlock_page(page);
922 continue;
923 }
924 if (strip_unit_end && (page->index > strip_unit_end)) {
925 dout("end of strip unit %p\n", page);
926 unlock_page(page);
927 break;
928 }
929 if (PageWriteback(page)) {
930 if (wbc->sync_mode == WB_SYNC_NONE) {
931 dout("%p under writeback\n", page);
932 unlock_page(page);
933 continue;
934 }
935 dout("waiting on writeback %p\n", page);
936 wait_on_page_writeback(page);
937 }
938
939 if (!clear_page_dirty_for_io(page)) {
940 dout("%p !clear_page_dirty_for_io\n", page);
941 unlock_page(page);
942 continue;
943 }
944
945 /*
946 * We have something to write. If this is
947 * the first locked page this time through,
948 * calculate max possinle write size and
949 * allocate a page array
950 */
951 if (locked_pages == 0) {
952 u64 objnum;
953 u64 objoff;
954 u32 xlen;
955
956 /* prepare async write request */
957 offset = (u64)page_offset(page);
958 ceph_calc_file_object_mapping(&ci->i_layout,
959 offset, wsize,
960 &objnum, &objoff,
961 &xlen);
962 len = xlen;
963
964 num_ops = 1;
965 strip_unit_end = page->index +
966 ((len - 1) >> PAGE_SHIFT);
967
968 BUG_ON(pages);
969 max_pages = calc_pages_for(0, (u64)len);
970 pages = kmalloc_array(max_pages,
971 sizeof(*pages),
972 GFP_NOFS);
973 if (!pages) {
974 pool = fsc->wb_pagevec_pool;
975 pages = mempool_alloc(pool, GFP_NOFS);
976 BUG_ON(!pages);
977 }
978
979 len = 0;
980 } else if (page->index !=
981 (offset + len) >> PAGE_SHIFT) {
982 if (num_ops >= (pool ? CEPH_OSD_SLAB_OPS :
983 CEPH_OSD_MAX_OPS)) {
984 redirty_page_for_writepage(wbc, page);
985 unlock_page(page);
986 break;
987 }
988
989 num_ops++;
990 offset = (u64)page_offset(page);
991 len = 0;
992 }
993
994 /* note position of first page in pvec */
995 dout("%p will write page %p idx %lu\n",
996 inode, page, page->index);
997
998 if (atomic_long_inc_return(&fsc->writeback_count) >
999 CONGESTION_ON_THRESH(
1000 fsc->mount_options->congestion_kb)) {
1001 set_bdi_congested(inode_to_bdi(inode),
1002 BLK_RW_ASYNC);
1003 }
1004
1005
1006 pages[locked_pages++] = page;
1007 pvec.pages[i] = NULL;
1008
1009 len += PAGE_SIZE;
1010 }
1011
1012 /* did we get anything? */
1013 if (!locked_pages)
1014 goto release_pvec_pages;
1015 if (i) {
1016 unsigned j, n = 0;
1017 /* shift unused page to beginning of pvec */
1018 for (j = 0; j < pvec_pages; j++) {
1019 if (!pvec.pages[j])
1020 continue;
1021 if (n < j)
1022 pvec.pages[n] = pvec.pages[j];
1023 n++;
1024 }
1025 pvec.nr = n;
1026
1027 if (pvec_pages && i == pvec_pages &&
1028 locked_pages < max_pages) {
1029 dout("reached end pvec, trying for more\n");
1030 pagevec_release(&pvec);
1031 goto get_more_pages;
1032 }
1033 }
1034
1035 new_request:
1036 offset = page_offset(pages[0]);
1037 len = wsize;
1038
1039 req = ceph_osdc_new_request(&fsc->client->osdc,
1040 &ci->i_layout, vino,
1041 offset, &len, 0, num_ops,
1042 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
1043 snapc, ceph_wbc.truncate_seq,
1044 ceph_wbc.truncate_size, false);
1045 if (IS_ERR(req)) {
1046 req = ceph_osdc_new_request(&fsc->client->osdc,
1047 &ci->i_layout, vino,
1048 offset, &len, 0,
1049 min(num_ops,
1050 CEPH_OSD_SLAB_OPS),
1051 CEPH_OSD_OP_WRITE,
1052 CEPH_OSD_FLAG_WRITE,
1053 snapc, ceph_wbc.truncate_seq,
1054 ceph_wbc.truncate_size, true);
1055 BUG_ON(IS_ERR(req));
1056 }
1057 BUG_ON(len < page_offset(pages[locked_pages - 1]) +
1058 PAGE_SIZE - offset);
1059
1060 req->r_callback = writepages_finish;
1061 req->r_inode = inode;
1062
1063 /* Format the osd request message and submit the write */
1064 len = 0;
1065 data_pages = pages;
1066 op_idx = 0;
1067 for (i = 0; i < locked_pages; i++) {
1068 u64 cur_offset = page_offset(pages[i]);
1069 if (offset + len != cur_offset) {
1070 if (op_idx + 1 == req->r_num_ops)
1071 break;
1072 osd_req_op_extent_dup_last(req, op_idx,
1073 cur_offset - offset);
1074 dout("writepages got pages at %llu~%llu\n",
1075 offset, len);
1076 osd_req_op_extent_osd_data_pages(req, op_idx,
1077 data_pages, len, 0,
1078 !!pool, false);
1079 osd_req_op_extent_update(req, op_idx, len);
1080
1081 len = 0;
1082 offset = cur_offset;
1083 data_pages = pages + i;
1084 op_idx++;
1085 }
1086
1087 set_page_writeback(pages[i]);
1088 len += PAGE_SIZE;
1089 }
1090
1091 if (ceph_wbc.size_stable) {
1092 len = min(len, ceph_wbc.i_size - offset);
1093 } else if (i == locked_pages) {
1094 /* writepages_finish() clears writeback pages
1095 * according to the data length, so make sure
1096 * data length covers all locked pages */
1097 u64 min_len = len + 1 - PAGE_SIZE;
1098 len = get_writepages_data_length(inode, pages[i - 1],
1099 offset);
1100 len = max(len, min_len);
1101 }
1102 dout("writepages got pages at %llu~%llu\n", offset, len);
1103
1104 osd_req_op_extent_osd_data_pages(req, op_idx, data_pages, len,
1105 0, !!pool, false);
1106 osd_req_op_extent_update(req, op_idx, len);
1107
1108 BUG_ON(op_idx + 1 != req->r_num_ops);
1109
1110 pool = NULL;
1111 if (i < locked_pages) {
1112 BUG_ON(num_ops <= req->r_num_ops);
1113 num_ops -= req->r_num_ops;
1114 locked_pages -= i;
1115
1116 /* allocate new pages array for next request */
1117 data_pages = pages;
1118 pages = kmalloc_array(locked_pages, sizeof(*pages),
1119 GFP_NOFS);
1120 if (!pages) {
1121 pool = fsc->wb_pagevec_pool;
1122 pages = mempool_alloc(pool, GFP_NOFS);
1123 BUG_ON(!pages);
1124 }
1125 memcpy(pages, data_pages + i,
1126 locked_pages * sizeof(*pages));
1127 memset(data_pages + i, 0,
1128 locked_pages * sizeof(*pages));
1129 } else {
1130 BUG_ON(num_ops != req->r_num_ops);
1131 index = pages[i - 1]->index + 1;
1132 /* request message now owns the pages array */
1133 pages = NULL;
1134 }
1135
1136 req->r_mtime = inode->i_mtime;
1137 rc = ceph_osdc_start_request(&fsc->client->osdc, req, true);
1138 BUG_ON(rc);
1139 req = NULL;
1140
1141 wbc->nr_to_write -= i;
1142 if (pages)
1143 goto new_request;
1144
1145 /*
1146 * We stop writing back only if we are not doing
1147 * integrity sync. In case of integrity sync we have to
1148 * keep going until we have written all the pages
1149 * we tagged for writeback prior to entering this loop.
1150 */
1151 if (wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE)
1152 done = true;
1153
1154 release_pvec_pages:
1155 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr,
1156 pvec.nr ? pvec.pages[0] : NULL);
1157 pagevec_release(&pvec);
1158 }
1159
1160 if (should_loop && !done) {
1161 /* more to do; loop back to beginning of file */
1162 dout("writepages looping back to beginning of file\n");
1163 end = start_index - 1; /* OK even when start_index == 0 */
1164
1165 /* to write dirty pages associated with next snapc,
1166 * we need to wait until current writes complete */
1167 if (wbc->sync_mode != WB_SYNC_NONE &&
1168 start_index == 0 && /* all dirty pages were checked */
1169 !ceph_wbc.head_snapc) {
1170 struct page *page;
1171 unsigned i, nr;
1172 index = 0;
1173 while ((index <= end) &&
1174 (nr = pagevec_lookup_tag(&pvec, mapping, &index,
1175 PAGECACHE_TAG_WRITEBACK))) {
1176 for (i = 0; i < nr; i++) {
1177 page = pvec.pages[i];
1178 if (page_snap_context(page) != snapc)
1179 continue;
1180 wait_on_page_writeback(page);
1181 }
1182 pagevec_release(&pvec);
1183 cond_resched();
1184 }
1185 }
1186
1187 start_index = 0;
1188 index = 0;
1189 goto retry;
1190 }
1191
1192 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1193 mapping->writeback_index = index;
1194
1195 out:
1196 ceph_osdc_put_request(req);
1197 ceph_put_snap_context(last_snapc);
1198 dout("writepages dend - startone, rc = %d\n", rc);
1199 return rc;
1200 }
1201
1202
1203
1204 /*
1205 * See if a given @snapc is either writeable, or already written.
1206 */
1207 static int context_is_writeable_or_written(struct inode *inode,
1208 struct ceph_snap_context *snapc)
1209 {
1210 struct ceph_snap_context *oldest = get_oldest_context(inode, NULL, NULL);
1211 int ret = !oldest || snapc->seq <= oldest->seq;
1212
1213 ceph_put_snap_context(oldest);
1214 return ret;
1215 }
1216
1217 /*
1218 * We are only allowed to write into/dirty the page if the page is
1219 * clean, or already dirty within the same snap context.
1220 *
1221 * called with page locked.
1222 * return success with page locked,
1223 * or any failure (incl -EAGAIN) with page unlocked.
1224 */
1225 static int ceph_update_writeable_page(struct file *file,
1226 loff_t pos, unsigned len,
1227 struct page *page)
1228 {
1229 struct inode *inode = file_inode(file);
1230 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1231 struct ceph_inode_info *ci = ceph_inode(inode);
1232 loff_t page_off = pos & PAGE_MASK;
1233 int pos_in_page = pos & ~PAGE_MASK;
1234 int end_in_page = pos_in_page + len;
1235 loff_t i_size;
1236 int r;
1237 struct ceph_snap_context *snapc, *oldest;
1238
1239 if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1240 dout(" page %p forced umount\n", page);
1241 unlock_page(page);
1242 return -EIO;
1243 }
1244
1245 retry_locked:
1246 /* writepages currently holds page lock, but if we change that later, */
1247 wait_on_page_writeback(page);
1248
1249 snapc = page_snap_context(page);
1250 if (snapc && snapc != ci->i_head_snapc) {
1251 /*
1252 * this page is already dirty in another (older) snap
1253 * context! is it writeable now?
1254 */
1255 oldest = get_oldest_context(inode, NULL, NULL);
1256 if (snapc->seq > oldest->seq) {
1257 ceph_put_snap_context(oldest);
1258 dout(" page %p snapc %p not current or oldest\n",
1259 page, snapc);
1260 /*
1261 * queue for writeback, and wait for snapc to
1262 * be writeable or written
1263 */
1264 snapc = ceph_get_snap_context(snapc);
1265 unlock_page(page);
1266 ceph_queue_writeback(inode);
1267 r = wait_event_killable(ci->i_cap_wq,
1268 context_is_writeable_or_written(inode, snapc));
1269 ceph_put_snap_context(snapc);
1270 if (r == -ERESTARTSYS)
1271 return r;
1272 return -EAGAIN;
1273 }
1274 ceph_put_snap_context(oldest);
1275
1276 /* yay, writeable, do it now (without dropping page lock) */
1277 dout(" page %p snapc %p not current, but oldest\n",
1278 page, snapc);
1279 if (!clear_page_dirty_for_io(page))
1280 goto retry_locked;
1281 r = writepage_nounlock(page, NULL);
1282 if (r < 0)
1283 goto fail_unlock;
1284 goto retry_locked;
1285 }
1286
1287 if (PageUptodate(page)) {
1288 dout(" page %p already uptodate\n", page);
1289 return 0;
1290 }
1291
1292 /* full page? */
1293 if (pos_in_page == 0 && len == PAGE_SIZE)
1294 return 0;
1295
1296 /* past end of file? */
1297 i_size = i_size_read(inode);
1298
1299 if (page_off >= i_size ||
1300 (pos_in_page == 0 && (pos+len) >= i_size &&
1301 end_in_page - pos_in_page != PAGE_SIZE)) {
1302 dout(" zeroing %p 0 - %d and %d - %d\n",
1303 page, pos_in_page, end_in_page, (int)PAGE_SIZE);
1304 zero_user_segments(page,
1305 0, pos_in_page,
1306 end_in_page, PAGE_SIZE);
1307 return 0;
1308 }
1309
1310 /* we need to read it. */
1311 r = ceph_do_readpage(file, page);
1312 if (r < 0) {
1313 if (r == -EINPROGRESS)
1314 return -EAGAIN;
1315 goto fail_unlock;
1316 }
1317 goto retry_locked;
1318 fail_unlock:
1319 unlock_page(page);
1320 return r;
1321 }
1322
1323 /*
1324 * We are only allowed to write into/dirty the page if the page is
1325 * clean, or already dirty within the same snap context.
1326 */
1327 static int ceph_write_begin(struct file *file, struct address_space *mapping,
1328 loff_t pos, unsigned len, unsigned flags,
1329 struct page **pagep, void **fsdata)
1330 {
1331 struct inode *inode = file_inode(file);
1332 struct page *page;
1333 pgoff_t index = pos >> PAGE_SHIFT;
1334 int r;
1335
1336 do {
1337 /* get a page */
1338 page = grab_cache_page_write_begin(mapping, index, 0);
1339 if (!page)
1340 return -ENOMEM;
1341
1342 dout("write_begin file %p inode %p page %p %d~%d\n", file,
1343 inode, page, (int)pos, (int)len);
1344
1345 r = ceph_update_writeable_page(file, pos, len, page);
1346 if (r < 0)
1347 put_page(page);
1348 else
1349 *pagep = page;
1350 } while (r == -EAGAIN);
1351
1352 return r;
1353 }
1354
1355 /*
1356 * we don't do anything in here that simple_write_end doesn't do
1357 * except adjust dirty page accounting
1358 */
1359 static int ceph_write_end(struct file *file, struct address_space *mapping,
1360 loff_t pos, unsigned len, unsigned copied,
1361 struct page *page, void *fsdata)
1362 {
1363 struct inode *inode = file_inode(file);
1364 bool check_cap = false;
1365
1366 dout("write_end file %p inode %p page %p %d~%d (%d)\n", file,
1367 inode, page, (int)pos, (int)copied, (int)len);
1368
1369 /* zero the stale part of the page if we did a short copy */
1370 if (!PageUptodate(page)) {
1371 if (copied < len) {
1372 copied = 0;
1373 goto out;
1374 }
1375 SetPageUptodate(page);
1376 }
1377
1378 /* did file size increase? */
1379 if (pos+copied > i_size_read(inode))
1380 check_cap = ceph_inode_set_size(inode, pos+copied);
1381
1382 set_page_dirty(page);
1383
1384 out:
1385 unlock_page(page);
1386 put_page(page);
1387
1388 if (check_cap)
1389 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL);
1390
1391 return copied;
1392 }
1393
1394 /*
1395 * we set .direct_IO to indicate direct io is supported, but since we
1396 * intercept O_DIRECT reads and writes early, this function should
1397 * never get called.
1398 */
1399 static ssize_t ceph_direct_io(struct kiocb *iocb, struct iov_iter *iter)
1400 {
1401 WARN_ON(1);
1402 return -EINVAL;
1403 }
1404
1405 const struct address_space_operations ceph_aops = {
1406 .readpage = ceph_readpage,
1407 .readpages = ceph_readpages,
1408 .writepage = ceph_writepage,
1409 .writepages = ceph_writepages_start,
1410 .write_begin = ceph_write_begin,
1411 .write_end = ceph_write_end,
1412 .set_page_dirty = ceph_set_page_dirty,
1413 .invalidatepage = ceph_invalidatepage,
1414 .releasepage = ceph_releasepage,
1415 .direct_IO = ceph_direct_io,
1416 };
1417
1418 static void ceph_block_sigs(sigset_t *oldset)
1419 {
1420 sigset_t mask;
1421 siginitsetinv(&mask, sigmask(SIGKILL));
1422 sigprocmask(SIG_BLOCK, &mask, oldset);
1423 }
1424
1425 static void ceph_restore_sigs(sigset_t *oldset)
1426 {
1427 sigprocmask(SIG_SETMASK, oldset, NULL);
1428 }
1429
1430 /*
1431 * vm ops
1432 */
1433 static vm_fault_t ceph_filemap_fault(struct vm_fault *vmf)
1434 {
1435 struct vm_area_struct *vma = vmf->vma;
1436 struct inode *inode = file_inode(vma->vm_file);
1437 struct ceph_inode_info *ci = ceph_inode(inode);
1438 struct ceph_file_info *fi = vma->vm_file->private_data;
1439 struct page *pinned_page = NULL;
1440 loff_t off = vmf->pgoff << PAGE_SHIFT;
1441 int want, got, err;
1442 sigset_t oldset;
1443 vm_fault_t ret = VM_FAULT_SIGBUS;
1444
1445 ceph_block_sigs(&oldset);
1446
1447 dout("filemap_fault %p %llx.%llx %llu~%zd trying to get caps\n",
1448 inode, ceph_vinop(inode), off, (size_t)PAGE_SIZE);
1449 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1450 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1451 else
1452 want = CEPH_CAP_FILE_CACHE;
1453
1454 got = 0;
1455 err = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page);
1456 if (err < 0)
1457 goto out_restore;
1458
1459 dout("filemap_fault %p %llu~%zd got cap refs on %s\n",
1460 inode, off, (size_t)PAGE_SIZE, ceph_cap_string(got));
1461
1462 if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) ||
1463 ci->i_inline_version == CEPH_INLINE_NONE) {
1464 CEPH_DEFINE_RW_CONTEXT(rw_ctx, got);
1465 ceph_add_rw_context(fi, &rw_ctx);
1466 ret = filemap_fault(vmf);
1467 ceph_del_rw_context(fi, &rw_ctx);
1468 dout("filemap_fault %p %llu~%zd drop cap refs %s ret %x\n",
1469 inode, off, (size_t)PAGE_SIZE,
1470 ceph_cap_string(got), ret);
1471 } else
1472 err = -EAGAIN;
1473
1474 if (pinned_page)
1475 put_page(pinned_page);
1476 ceph_put_cap_refs(ci, got);
1477
1478 if (err != -EAGAIN)
1479 goto out_restore;
1480
1481 /* read inline data */
1482 if (off >= PAGE_SIZE) {
1483 /* does not support inline data > PAGE_SIZE */
1484 ret = VM_FAULT_SIGBUS;
1485 } else {
1486 struct address_space *mapping = inode->i_mapping;
1487 struct page *page = find_or_create_page(mapping, 0,
1488 mapping_gfp_constraint(mapping,
1489 ~__GFP_FS));
1490 if (!page) {
1491 ret = VM_FAULT_OOM;
1492 goto out_inline;
1493 }
1494 err = __ceph_do_getattr(inode, page,
1495 CEPH_STAT_CAP_INLINE_DATA, true);
1496 if (err < 0 || off >= i_size_read(inode)) {
1497 unlock_page(page);
1498 put_page(page);
1499 if (err == -ENOMEM)
1500 ret = VM_FAULT_OOM;
1501 else
1502 ret = VM_FAULT_SIGBUS;
1503 goto out_inline;
1504 }
1505 if (err < PAGE_SIZE)
1506 zero_user_segment(page, err, PAGE_SIZE);
1507 else
1508 flush_dcache_page(page);
1509 SetPageUptodate(page);
1510 vmf->page = page;
1511 ret = VM_FAULT_MAJOR | VM_FAULT_LOCKED;
1512 out_inline:
1513 dout("filemap_fault %p %llu~%zd read inline data ret %x\n",
1514 inode, off, (size_t)PAGE_SIZE, ret);
1515 }
1516 out_restore:
1517 ceph_restore_sigs(&oldset);
1518 if (err < 0)
1519 ret = vmf_error(err);
1520
1521 return ret;
1522 }
1523
1524 /*
1525 * Reuse write_begin here for simplicity.
1526 */
1527 static vm_fault_t ceph_page_mkwrite(struct vm_fault *vmf)
1528 {
1529 struct vm_area_struct *vma = vmf->vma;
1530 struct inode *inode = file_inode(vma->vm_file);
1531 struct ceph_inode_info *ci = ceph_inode(inode);
1532 struct ceph_file_info *fi = vma->vm_file->private_data;
1533 struct ceph_cap_flush *prealloc_cf;
1534 struct page *page = vmf->page;
1535 loff_t off = page_offset(page);
1536 loff_t size = i_size_read(inode);
1537 size_t len;
1538 int want, got, err;
1539 sigset_t oldset;
1540 vm_fault_t ret = VM_FAULT_SIGBUS;
1541
1542 prealloc_cf = ceph_alloc_cap_flush();
1543 if (!prealloc_cf)
1544 return VM_FAULT_OOM;
1545
1546 ceph_block_sigs(&oldset);
1547
1548 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1549 struct page *locked_page = NULL;
1550 if (off == 0) {
1551 lock_page(page);
1552 locked_page = page;
1553 }
1554 err = ceph_uninline_data(vma->vm_file, locked_page);
1555 if (locked_page)
1556 unlock_page(locked_page);
1557 if (err < 0)
1558 goto out_free;
1559 }
1560
1561 if (off + PAGE_SIZE <= size)
1562 len = PAGE_SIZE;
1563 else
1564 len = size & ~PAGE_MASK;
1565
1566 dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n",
1567 inode, ceph_vinop(inode), off, len, size);
1568 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1569 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1570 else
1571 want = CEPH_CAP_FILE_BUFFER;
1572
1573 got = 0;
1574 err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, off + len,
1575 &got, NULL);
1576 if (err < 0)
1577 goto out_free;
1578
1579 dout("page_mkwrite %p %llu~%zd got cap refs on %s\n",
1580 inode, off, len, ceph_cap_string(got));
1581
1582 /* Update time before taking page lock */
1583 file_update_time(vma->vm_file);
1584
1585 do {
1586 lock_page(page);
1587
1588 if ((off > size) || (page->mapping != inode->i_mapping)) {
1589 unlock_page(page);
1590 ret = VM_FAULT_NOPAGE;
1591 break;
1592 }
1593
1594 err = ceph_update_writeable_page(vma->vm_file, off, len, page);
1595 if (err >= 0) {
1596 /* success. we'll keep the page locked. */
1597 set_page_dirty(page);
1598 ret = VM_FAULT_LOCKED;
1599 }
1600 } while (err == -EAGAIN);
1601
1602 if (ret == VM_FAULT_LOCKED ||
1603 ci->i_inline_version != CEPH_INLINE_NONE) {
1604 int dirty;
1605 spin_lock(&ci->i_ceph_lock);
1606 ci->i_inline_version = CEPH_INLINE_NONE;
1607 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1608 &prealloc_cf);
1609 spin_unlock(&ci->i_ceph_lock);
1610 if (dirty)
1611 __mark_inode_dirty(inode, dirty);
1612 }
1613
1614 dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %x\n",
1615 inode, off, len, ceph_cap_string(got), ret);
1616 ceph_put_cap_refs(ci, got);
1617 out_free:
1618 ceph_restore_sigs(&oldset);
1619 ceph_free_cap_flush(prealloc_cf);
1620 if (err < 0)
1621 ret = vmf_error(err);
1622 return ret;
1623 }
1624
1625 void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
1626 char *data, size_t len)
1627 {
1628 struct address_space *mapping = inode->i_mapping;
1629 struct page *page;
1630
1631 if (locked_page) {
1632 page = locked_page;
1633 } else {
1634 if (i_size_read(inode) == 0)
1635 return;
1636 page = find_or_create_page(mapping, 0,
1637 mapping_gfp_constraint(mapping,
1638 ~__GFP_FS));
1639 if (!page)
1640 return;
1641 if (PageUptodate(page)) {
1642 unlock_page(page);
1643 put_page(page);
1644 return;
1645 }
1646 }
1647
1648 dout("fill_inline_data %p %llx.%llx len %zu locked_page %p\n",
1649 inode, ceph_vinop(inode), len, locked_page);
1650
1651 if (len > 0) {
1652 void *kaddr = kmap_atomic(page);
1653 memcpy(kaddr, data, len);
1654 kunmap_atomic(kaddr);
1655 }
1656
1657 if (page != locked_page) {
1658 if (len < PAGE_SIZE)
1659 zero_user_segment(page, len, PAGE_SIZE);
1660 else
1661 flush_dcache_page(page);
1662
1663 SetPageUptodate(page);
1664 unlock_page(page);
1665 put_page(page);
1666 }
1667 }
1668
1669 int ceph_uninline_data(struct file *filp, struct page *locked_page)
1670 {
1671 struct inode *inode = file_inode(filp);
1672 struct ceph_inode_info *ci = ceph_inode(inode);
1673 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1674 struct ceph_osd_request *req;
1675 struct page *page = NULL;
1676 u64 len, inline_version;
1677 int err = 0;
1678 bool from_pagecache = false;
1679
1680 spin_lock(&ci->i_ceph_lock);
1681 inline_version = ci->i_inline_version;
1682 spin_unlock(&ci->i_ceph_lock);
1683
1684 dout("uninline_data %p %llx.%llx inline_version %llu\n",
1685 inode, ceph_vinop(inode), inline_version);
1686
1687 if (inline_version == 1 || /* initial version, no data */
1688 inline_version == CEPH_INLINE_NONE)
1689 goto out;
1690
1691 if (locked_page) {
1692 page = locked_page;
1693 WARN_ON(!PageUptodate(page));
1694 } else if (ceph_caps_issued(ci) &
1695 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) {
1696 page = find_get_page(inode->i_mapping, 0);
1697 if (page) {
1698 if (PageUptodate(page)) {
1699 from_pagecache = true;
1700 lock_page(page);
1701 } else {
1702 put_page(page);
1703 page = NULL;
1704 }
1705 }
1706 }
1707
1708 if (page) {
1709 len = i_size_read(inode);
1710 if (len > PAGE_SIZE)
1711 len = PAGE_SIZE;
1712 } else {
1713 page = __page_cache_alloc(GFP_NOFS);
1714 if (!page) {
1715 err = -ENOMEM;
1716 goto out;
1717 }
1718 err = __ceph_do_getattr(inode, page,
1719 CEPH_STAT_CAP_INLINE_DATA, true);
1720 if (err < 0) {
1721 /* no inline data */
1722 if (err == -ENODATA)
1723 err = 0;
1724 goto out;
1725 }
1726 len = err;
1727 }
1728
1729 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1730 ceph_vino(inode), 0, &len, 0, 1,
1731 CEPH_OSD_OP_CREATE, CEPH_OSD_FLAG_WRITE,
1732 NULL, 0, 0, false);
1733 if (IS_ERR(req)) {
1734 err = PTR_ERR(req);
1735 goto out;
1736 }
1737
1738 req->r_mtime = inode->i_mtime;
1739 err = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1740 if (!err)
1741 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1742 ceph_osdc_put_request(req);
1743 if (err < 0)
1744 goto out;
1745
1746 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1747 ceph_vino(inode), 0, &len, 1, 3,
1748 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
1749 NULL, ci->i_truncate_seq,
1750 ci->i_truncate_size, false);
1751 if (IS_ERR(req)) {
1752 err = PTR_ERR(req);
1753 goto out;
1754 }
1755
1756 osd_req_op_extent_osd_data_pages(req, 1, &page, len, 0, false, false);
1757
1758 {
1759 __le64 xattr_buf = cpu_to_le64(inline_version);
1760 err = osd_req_op_xattr_init(req, 0, CEPH_OSD_OP_CMPXATTR,
1761 "inline_version", &xattr_buf,
1762 sizeof(xattr_buf),
1763 CEPH_OSD_CMPXATTR_OP_GT,
1764 CEPH_OSD_CMPXATTR_MODE_U64);
1765 if (err)
1766 goto out_put;
1767 }
1768
1769 {
1770 char xattr_buf[32];
1771 int xattr_len = snprintf(xattr_buf, sizeof(xattr_buf),
1772 "%llu", inline_version);
1773 err = osd_req_op_xattr_init(req, 2, CEPH_OSD_OP_SETXATTR,
1774 "inline_version",
1775 xattr_buf, xattr_len, 0, 0);
1776 if (err)
1777 goto out_put;
1778 }
1779
1780 req->r_mtime = inode->i_mtime;
1781 err = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1782 if (!err)
1783 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1784 out_put:
1785 ceph_osdc_put_request(req);
1786 if (err == -ECANCELED)
1787 err = 0;
1788 out:
1789 if (page && page != locked_page) {
1790 if (from_pagecache) {
1791 unlock_page(page);
1792 put_page(page);
1793 } else
1794 __free_pages(page, 0);
1795 }
1796
1797 dout("uninline_data %p %llx.%llx inline_version %llu = %d\n",
1798 inode, ceph_vinop(inode), inline_version, err);
1799 return err;
1800 }
1801
1802 static const struct vm_operations_struct ceph_vmops = {
1803 .fault = ceph_filemap_fault,
1804 .page_mkwrite = ceph_page_mkwrite,
1805 };
1806
1807 int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1808 {
1809 struct address_space *mapping = file->f_mapping;
1810
1811 if (!mapping->a_ops->readpage)
1812 return -ENOEXEC;
1813 file_accessed(file);
1814 vma->vm_ops = &ceph_vmops;
1815 return 0;
1816 }
1817
1818 enum {
1819 POOL_READ = 1,
1820 POOL_WRITE = 2,
1821 };
1822
1823 static int __ceph_pool_perm_get(struct ceph_inode_info *ci,
1824 s64 pool, struct ceph_string *pool_ns)
1825 {
1826 struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode);
1827 struct ceph_mds_client *mdsc = fsc->mdsc;
1828 struct ceph_osd_request *rd_req = NULL, *wr_req = NULL;
1829 struct rb_node **p, *parent;
1830 struct ceph_pool_perm *perm;
1831 struct page **pages;
1832 size_t pool_ns_len;
1833 int err = 0, err2 = 0, have = 0;
1834
1835 down_read(&mdsc->pool_perm_rwsem);
1836 p = &mdsc->pool_perm_tree.rb_node;
1837 while (*p) {
1838 perm = rb_entry(*p, struct ceph_pool_perm, node);
1839 if (pool < perm->pool)
1840 p = &(*p)->rb_left;
1841 else if (pool > perm->pool)
1842 p = &(*p)->rb_right;
1843 else {
1844 int ret = ceph_compare_string(pool_ns,
1845 perm->pool_ns,
1846 perm->pool_ns_len);
1847 if (ret < 0)
1848 p = &(*p)->rb_left;
1849 else if (ret > 0)
1850 p = &(*p)->rb_right;
1851 else {
1852 have = perm->perm;
1853 break;
1854 }
1855 }
1856 }
1857 up_read(&mdsc->pool_perm_rwsem);
1858 if (*p)
1859 goto out;
1860
1861 if (pool_ns)
1862 dout("__ceph_pool_perm_get pool %lld ns %.*s no perm cached\n",
1863 pool, (int)pool_ns->len, pool_ns->str);
1864 else
1865 dout("__ceph_pool_perm_get pool %lld no perm cached\n", pool);
1866
1867 down_write(&mdsc->pool_perm_rwsem);
1868 p = &mdsc->pool_perm_tree.rb_node;
1869 parent = NULL;
1870 while (*p) {
1871 parent = *p;
1872 perm = rb_entry(parent, struct ceph_pool_perm, node);
1873 if (pool < perm->pool)
1874 p = &(*p)->rb_left;
1875 else if (pool > perm->pool)
1876 p = &(*p)->rb_right;
1877 else {
1878 int ret = ceph_compare_string(pool_ns,
1879 perm->pool_ns,
1880 perm->pool_ns_len);
1881 if (ret < 0)
1882 p = &(*p)->rb_left;
1883 else if (ret > 0)
1884 p = &(*p)->rb_right;
1885 else {
1886 have = perm->perm;
1887 break;
1888 }
1889 }
1890 }
1891 if (*p) {
1892 up_write(&mdsc->pool_perm_rwsem);
1893 goto out;
1894 }
1895
1896 rd_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1897 1, false, GFP_NOFS);
1898 if (!rd_req) {
1899 err = -ENOMEM;
1900 goto out_unlock;
1901 }
1902
1903 rd_req->r_flags = CEPH_OSD_FLAG_READ;
1904 osd_req_op_init(rd_req, 0, CEPH_OSD_OP_STAT, 0);
1905 rd_req->r_base_oloc.pool = pool;
1906 if (pool_ns)
1907 rd_req->r_base_oloc.pool_ns = ceph_get_string(pool_ns);
1908 ceph_oid_printf(&rd_req->r_base_oid, "%llx.00000000", ci->i_vino.ino);
1909
1910 err = ceph_osdc_alloc_messages(rd_req, GFP_NOFS);
1911 if (err)
1912 goto out_unlock;
1913
1914 wr_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1915 1, false, GFP_NOFS);
1916 if (!wr_req) {
1917 err = -ENOMEM;
1918 goto out_unlock;
1919 }
1920
1921 wr_req->r_flags = CEPH_OSD_FLAG_WRITE;
1922 osd_req_op_init(wr_req, 0, CEPH_OSD_OP_CREATE, CEPH_OSD_OP_FLAG_EXCL);
1923 ceph_oloc_copy(&wr_req->r_base_oloc, &rd_req->r_base_oloc);
1924 ceph_oid_copy(&wr_req->r_base_oid, &rd_req->r_base_oid);
1925
1926 err = ceph_osdc_alloc_messages(wr_req, GFP_NOFS);
1927 if (err)
1928 goto out_unlock;
1929
1930 /* one page should be large enough for STAT data */
1931 pages = ceph_alloc_page_vector(1, GFP_KERNEL);
1932 if (IS_ERR(pages)) {
1933 err = PTR_ERR(pages);
1934 goto out_unlock;
1935 }
1936
1937 osd_req_op_raw_data_in_pages(rd_req, 0, pages, PAGE_SIZE,
1938 0, false, true);
1939 err = ceph_osdc_start_request(&fsc->client->osdc, rd_req, false);
1940
1941 wr_req->r_mtime = ci->vfs_inode.i_mtime;
1942 err2 = ceph_osdc_start_request(&fsc->client->osdc, wr_req, false);
1943
1944 if (!err)
1945 err = ceph_osdc_wait_request(&fsc->client->osdc, rd_req);
1946 if (!err2)
1947 err2 = ceph_osdc_wait_request(&fsc->client->osdc, wr_req);
1948
1949 if (err >= 0 || err == -ENOENT)
1950 have |= POOL_READ;
1951 else if (err != -EPERM)
1952 goto out_unlock;
1953
1954 if (err2 == 0 || err2 == -EEXIST)
1955 have |= POOL_WRITE;
1956 else if (err2 != -EPERM) {
1957 err = err2;
1958 goto out_unlock;
1959 }
1960
1961 pool_ns_len = pool_ns ? pool_ns->len : 0;
1962 perm = kmalloc(sizeof(*perm) + pool_ns_len + 1, GFP_NOFS);
1963 if (!perm) {
1964 err = -ENOMEM;
1965 goto out_unlock;
1966 }
1967
1968 perm->pool = pool;
1969 perm->perm = have;
1970 perm->pool_ns_len = pool_ns_len;
1971 if (pool_ns_len > 0)
1972 memcpy(perm->pool_ns, pool_ns->str, pool_ns_len);
1973 perm->pool_ns[pool_ns_len] = 0;
1974
1975 rb_link_node(&perm->node, parent, p);
1976 rb_insert_color(&perm->node, &mdsc->pool_perm_tree);
1977 err = 0;
1978 out_unlock:
1979 up_write(&mdsc->pool_perm_rwsem);
1980
1981 ceph_osdc_put_request(rd_req);
1982 ceph_osdc_put_request(wr_req);
1983 out:
1984 if (!err)
1985 err = have;
1986 if (pool_ns)
1987 dout("__ceph_pool_perm_get pool %lld ns %.*s result = %d\n",
1988 pool, (int)pool_ns->len, pool_ns->str, err);
1989 else
1990 dout("__ceph_pool_perm_get pool %lld result = %d\n", pool, err);
1991 return err;
1992 }
1993
1994 int ceph_pool_perm_check(struct ceph_inode_info *ci, int need)
1995 {
1996 s64 pool;
1997 struct ceph_string *pool_ns;
1998 int ret, flags;
1999
2000 if (ci->i_vino.snap != CEPH_NOSNAP) {
2001 /*
2002 * Pool permission check needs to write to the first object.
2003 * But for snapshot, head of the first object may have alread
2004 * been deleted. Skip check to avoid creating orphan object.
2005 */
2006 return 0;
2007 }
2008
2009 if (ceph_test_mount_opt(ceph_inode_to_client(&ci->vfs_inode),
2010 NOPOOLPERM))
2011 return 0;
2012
2013 spin_lock(&ci->i_ceph_lock);
2014 flags = ci->i_ceph_flags;
2015 pool = ci->i_layout.pool_id;
2016 spin_unlock(&ci->i_ceph_lock);
2017 check:
2018 if (flags & CEPH_I_POOL_PERM) {
2019 if ((need & CEPH_CAP_FILE_RD) && !(flags & CEPH_I_POOL_RD)) {
2020 dout("ceph_pool_perm_check pool %lld no read perm\n",
2021 pool);
2022 return -EPERM;
2023 }
2024 if ((need & CEPH_CAP_FILE_WR) && !(flags & CEPH_I_POOL_WR)) {
2025 dout("ceph_pool_perm_check pool %lld no write perm\n",
2026 pool);
2027 return -EPERM;
2028 }
2029 return 0;
2030 }
2031
2032 pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
2033 ret = __ceph_pool_perm_get(ci, pool, pool_ns);
2034 ceph_put_string(pool_ns);
2035 if (ret < 0)
2036 return ret;
2037
2038 flags = CEPH_I_POOL_PERM;
2039 if (ret & POOL_READ)
2040 flags |= CEPH_I_POOL_RD;
2041 if (ret & POOL_WRITE)
2042 flags |= CEPH_I_POOL_WR;
2043
2044 spin_lock(&ci->i_ceph_lock);
2045 if (pool == ci->i_layout.pool_id &&
2046 pool_ns == rcu_dereference_raw(ci->i_layout.pool_ns)) {
2047 ci->i_ceph_flags |= flags;
2048 } else {
2049 pool = ci->i_layout.pool_id;
2050 flags = ci->i_ceph_flags;
2051 }
2052 spin_unlock(&ci->i_ceph_lock);
2053 goto check;
2054 }
2055
2056 void ceph_pool_perm_destroy(struct ceph_mds_client *mdsc)
2057 {
2058 struct ceph_pool_perm *perm;
2059 struct rb_node *n;
2060
2061 while (!RB_EMPTY_ROOT(&mdsc->pool_perm_tree)) {
2062 n = rb_first(&mdsc->pool_perm_tree);
2063 perm = rb_entry(n, struct ceph_pool_perm, node);
2064 rb_erase(n, &mdsc->pool_perm_tree);
2065 kfree(perm);
2066 }
2067 }
Linux with added FPC-III support