?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
net: niu: fix format string overflow warning:
[linux-2.6/btrfs-unstable.git] / fs / ceph / addr.c
blob50836280a6f8a3a2d1a82d3c00f3f274e5844f7e
1 #include <linux/ceph/ceph_debug.h>
2
3 #include <linux/backing-dev.h>
4 #include <linux/fs.h>
5 #include <linux/mm.h>
6 #include <linux/pagemap.h>
7 #include <linux/writeback.h> /* generic_writepages */
8 #include <linux/slab.h>
9 #include <linux/pagevec.h>
10 #include <linux/task_io_accounting_ops.h>
11 #include <linux/signal.h>
12
13 #include "super.h"
14 #include "mds_client.h"
15 #include "cache.h"
16 #include <linux/ceph/osd_client.h>
17
18 /*
19 * Ceph address space ops.
20 *
21 * There are a few funny things going on here.
22 *
23 * The page->private field is used to reference a struct
24 * ceph_snap_context for _every_ dirty page. This indicates which
25 * snapshot the page was logically dirtied in, and thus which snap
26 * context needs to be associated with the osd write during writeback.
27 *
28 * Similarly, struct ceph_inode_info maintains a set of counters to
29 * count dirty pages on the inode. In the absence of snapshots,
30 * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count.
31 *
32 * When a snapshot is taken (that is, when the client receives
33 * notification that a snapshot was taken), each inode with caps and
34 * with dirty pages (dirty pages implies there is a cap) gets a new
35 * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending
36 * order, new snaps go to the tail). The i_wrbuffer_ref_head count is
37 * moved to capsnap->dirty. (Unless a sync write is currently in
38 * progress. In that case, the capsnap is said to be "pending", new
39 * writes cannot start, and the capsnap isn't "finalized" until the
40 * write completes (or fails) and a final size/mtime for the inode for
41 * that snap can be settled upon.) i_wrbuffer_ref_head is reset to 0.
42 *
43 * On writeback, we must submit writes to the osd IN SNAP ORDER. So,
44 * we look for the first capsnap in i_cap_snaps and write out pages in
45 * that snap context _only_. Then we move on to the next capsnap,
46 * eventually reaching the "live" or "head" context (i.e., pages that
47 * are not yet snapped) and are writing the most recently dirtied
48 * pages.
49 *
50 * Invalidate and so forth must take care to ensure the dirty page
51 * accounting is preserved.
52 */
53
54 #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10))
55 #define CONGESTION_OFF_THRESH(congestion_kb) \
56 (CONGESTION_ON_THRESH(congestion_kb) - \
57 (CONGESTION_ON_THRESH(congestion_kb) >> 2))
58
59 static inline struct ceph_snap_context *page_snap_context(struct page *page)
60 {
61 if (PagePrivate(page))
62 return (void *)page->private;
63 return NULL;
64 }
65
66 /*
67 * Dirty a page. Optimistically adjust accounting, on the assumption
68 * that we won't race with invalidate. If we do, readjust.
69 */
70 static int ceph_set_page_dirty(struct page *page)
71 {
72 struct address_space *mapping = page->mapping;
73 struct inode *inode;
74 struct ceph_inode_info *ci;
75 struct ceph_snap_context *snapc;
76 int ret;
77
78 if (unlikely(!mapping))
79 return !TestSetPageDirty(page);
80
81 if (PageDirty(page)) {
82 dout("%p set_page_dirty %p idx %lu -- already dirty\n",
83 mapping->host, page, page->index);
84 BUG_ON(!PagePrivate(page));
85 return 0;
86 }
87
88 inode = mapping->host;
89 ci = ceph_inode(inode);
90
91 /* dirty the head */
92 spin_lock(&ci->i_ceph_lock);
93 BUG_ON(ci->i_wr_ref == 0); // caller should hold Fw reference
94 if (__ceph_have_pending_cap_snap(ci)) {
95 struct ceph_cap_snap *capsnap =
96 list_last_entry(&ci->i_cap_snaps,
97 struct ceph_cap_snap,
98 ci_item);
99 snapc = ceph_get_snap_context(capsnap->context);
100 capsnap->dirty_pages++;
101 } else {
102 BUG_ON(!ci->i_head_snapc);
103 snapc = ceph_get_snap_context(ci->i_head_snapc);
104 ++ci->i_wrbuffer_ref_head;
105 }
106 if (ci->i_wrbuffer_ref == 0)
107 ihold(inode);
108 ++ci->i_wrbuffer_ref;
109 dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d "
110 "snapc %p seq %lld (%d snaps)\n",
111 mapping->host, page, page->index,
112 ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1,
113 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
114 snapc, snapc->seq, snapc->num_snaps);
115 spin_unlock(&ci->i_ceph_lock);
116
117 /*
118 * Reference snap context in page->private. Also set
119 * PagePrivate so that we get invalidatepage callback.
120 */
121 BUG_ON(PagePrivate(page));
122 page->private = (unsigned long)snapc;
123 SetPagePrivate(page);
124
125 ret = __set_page_dirty_nobuffers(page);
126 WARN_ON(!PageLocked(page));
127 WARN_ON(!page->mapping);
128
129 return ret;
130 }
131
132 /*
133 * If we are truncating the full page (i.e. offset == 0), adjust the
134 * dirty page counters appropriately. Only called if there is private
135 * data on the page.
136 */
137 static void ceph_invalidatepage(struct page *page, unsigned int offset,
138 unsigned int length)
139 {
140 struct inode *inode;
141 struct ceph_inode_info *ci;
142 struct ceph_snap_context *snapc = page_snap_context(page);
143
144 inode = page->mapping->host;
145 ci = ceph_inode(inode);
146
147 if (offset != 0 || length != PAGE_SIZE) {
148 dout("%p invalidatepage %p idx %lu partial dirty page %u~%u\n",
149 inode, page, page->index, offset, length);
150 return;
151 }
152
153 ceph_invalidate_fscache_page(inode, page);
154
155 if (!PagePrivate(page))
156 return;
157
158 /*
159 * We can get non-dirty pages here due to races between
160 * set_page_dirty and truncate_complete_page; just spit out a
161 * warning, in case we end up with accounting problems later.
162 */
163 if (!PageDirty(page))
164 pr_err("%p invalidatepage %p page not dirty\n", inode, page);
165
166 ClearPageChecked(page);
167
168 dout("%p invalidatepage %p idx %lu full dirty page\n",
169 inode, page, page->index);
170
171 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
172 ceph_put_snap_context(snapc);
173 page->private = 0;
174 ClearPagePrivate(page);
175 }
176
177 static int ceph_releasepage(struct page *page, gfp_t g)
178 {
179 dout("%p releasepage %p idx %lu (%sdirty)\n", page->mapping->host,
180 page, page->index, PageDirty(page) ? "" : "not ");
181
182 /* Can we release the page from the cache? */
183 if (!ceph_release_fscache_page(page, g))
184 return 0;
185
186 return !PagePrivate(page);
187 }
188
189 /*
190 * read a single page, without unlocking it.
191 */
192 static int readpage_nounlock(struct file *filp, struct page *page)
193 {
194 struct inode *inode = file_inode(filp);
195 struct ceph_inode_info *ci = ceph_inode(inode);
196 struct ceph_osd_client *osdc =
197 &ceph_inode_to_client(inode)->client->osdc;
198 int err = 0;
199 u64 off = page_offset(page);
200 u64 len = PAGE_SIZE;
201
202 if (off >= i_size_read(inode)) {
203 zero_user_segment(page, 0, PAGE_SIZE);
204 SetPageUptodate(page);
205 return 0;
206 }
207
208 if (ci->i_inline_version != CEPH_INLINE_NONE) {
209 /*
210 * Uptodate inline data should have been added
211 * into page cache while getting Fcr caps.
212 */
213 if (off == 0)
214 return -EINVAL;
215 zero_user_segment(page, 0, PAGE_SIZE);
216 SetPageUptodate(page);
217 return 0;
218 }
219
220 err = ceph_readpage_from_fscache(inode, page);
221 if (err == 0)
222 goto out;
223
224 dout("readpage inode %p file %p page %p index %lu\n",
225 inode, filp, page, page->index);
226 err = ceph_osdc_readpages(osdc, ceph_vino(inode), &ci->i_layout,
227 off, &len,
228 ci->i_truncate_seq, ci->i_truncate_size,
229 &page, 1, 0);
230 if (err == -ENOENT)
231 err = 0;
232 if (err < 0) {
233 SetPageError(page);
234 ceph_fscache_readpage_cancel(inode, page);
235 goto out;
236 }
237 if (err < PAGE_SIZE)
238 /* zero fill remainder of page */
239 zero_user_segment(page, err, PAGE_SIZE);
240 else
241 flush_dcache_page(page);
242
243 SetPageUptodate(page);
244 ceph_readpage_to_fscache(inode, page);
245
246 out:
247 return err < 0 ? err : 0;
248 }
249
250 static int ceph_readpage(struct file *filp, struct page *page)
251 {
252 int r = readpage_nounlock(filp, page);
253 unlock_page(page);
254 return r;
255 }
256
257 /*
258 * Finish an async read(ahead) op.
259 */
260 static void finish_read(struct ceph_osd_request *req)
261 {
262 struct inode *inode = req->r_inode;
263 struct ceph_osd_data *osd_data;
264 int rc = req->r_result <= 0 ? req->r_result : 0;
265 int bytes = req->r_result >= 0 ? req->r_result : 0;
266 int num_pages;
267 int i;
268
269 dout("finish_read %p req %p rc %d bytes %d\n", inode, req, rc, bytes);
270
271 /* unlock all pages, zeroing any data we didn't read */
272 osd_data = osd_req_op_extent_osd_data(req, 0);
273 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
274 num_pages = calc_pages_for((u64)osd_data->alignment,
275 (u64)osd_data->length);
276 for (i = 0; i < num_pages; i++) {
277 struct page *page = osd_data->pages[i];
278
279 if (rc < 0 && rc != -ENOENT) {
280 ceph_fscache_readpage_cancel(inode, page);
281 goto unlock;
282 }
283 if (bytes < (int)PAGE_SIZE) {
284 /* zero (remainder of) page */
285 int s = bytes < 0 ? 0 : bytes;
286 zero_user_segment(page, s, PAGE_SIZE);
287 }
288 dout("finish_read %p uptodate %p idx %lu\n", inode, page,
289 page->index);
290 flush_dcache_page(page);
291 SetPageUptodate(page);
292 ceph_readpage_to_fscache(inode, page);
293 unlock:
294 unlock_page(page);
295 put_page(page);
296 bytes -= PAGE_SIZE;
297 }
298 kfree(osd_data->pages);
299 }
300
301 /*
302 * start an async read(ahead) operation. return nr_pages we submitted
303 * a read for on success, or negative error code.
304 */
305 static int start_read(struct inode *inode, struct list_head *page_list, int max)
306 {
307 struct ceph_osd_client *osdc =
308 &ceph_inode_to_client(inode)->client->osdc;
309 struct ceph_inode_info *ci = ceph_inode(inode);
310 struct page *page = list_entry(page_list->prev, struct page, lru);
311 struct ceph_vino vino;
312 struct ceph_osd_request *req;
313 u64 off;
314 u64 len;
315 int i;
316 struct page **pages;
317 pgoff_t next_index;
318 int nr_pages = 0;
319 int got = 0;
320 int ret = 0;
321
322 if (!current->journal_info) {
323 /* caller of readpages does not hold buffer and read caps
324 * (fadvise, madvise and readahead cases) */
325 int want = CEPH_CAP_FILE_CACHE;
326 ret = ceph_try_get_caps(ci, CEPH_CAP_FILE_RD, want, &got);
327 if (ret < 0) {
328 dout("start_read %p, error getting cap\n", inode);
329 } else if (!(got & want)) {
330 dout("start_read %p, no cache cap\n", inode);
331 ret = 0;
332 }
333 if (ret <= 0) {
334 if (got)
335 ceph_put_cap_refs(ci, got);
336 while (!list_empty(page_list)) {
337 page = list_entry(page_list->prev,
338 struct page, lru);
339 list_del(&page->lru);
340 put_page(page);
341 }
342 return ret;
343 }
344 }
345
346 off = (u64) page_offset(page);
347
348 /* count pages */
349 next_index = page->index;
350 list_for_each_entry_reverse(page, page_list, lru) {
351 if (page->index != next_index)
352 break;
353 nr_pages++;
354 next_index++;
355 if (max && nr_pages == max)
356 break;
357 }
358 len = nr_pages << PAGE_SHIFT;
359 dout("start_read %p nr_pages %d is %lld~%lld\n", inode, nr_pages,
360 off, len);
361 vino = ceph_vino(inode);
362 req = ceph_osdc_new_request(osdc, &ci->i_layout, vino, off, &len,
363 0, 1, CEPH_OSD_OP_READ,
364 CEPH_OSD_FLAG_READ, NULL,
365 ci->i_truncate_seq, ci->i_truncate_size,
366 false);
367 if (IS_ERR(req)) {
368 ret = PTR_ERR(req);
369 goto out;
370 }
371
372 /* build page vector */
373 nr_pages = calc_pages_for(0, len);
374 pages = kmalloc(sizeof(*pages) * nr_pages, GFP_KERNEL);
375 if (!pages) {
376 ret = -ENOMEM;
377 goto out_put;
378 }
379 for (i = 0; i < nr_pages; ++i) {
380 page = list_entry(page_list->prev, struct page, lru);
381 BUG_ON(PageLocked(page));
382 list_del(&page->lru);
383
384 dout("start_read %p adding %p idx %lu\n", inode, page,
385 page->index);
386 if (add_to_page_cache_lru(page, &inode->i_data, page->index,
387 GFP_KERNEL)) {
388 ceph_fscache_uncache_page(inode, page);
389 put_page(page);
390 dout("start_read %p add_to_page_cache failed %p\n",
391 inode, page);
392 nr_pages = i;
393 if (nr_pages > 0) {
394 len = nr_pages << PAGE_SHIFT;
395 osd_req_op_extent_update(req, 0, len);
396 break;
397 }
398 goto out_pages;
399 }
400 pages[i] = page;
401 }
402 osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false);
403 req->r_callback = finish_read;
404 req->r_inode = inode;
405
406 dout("start_read %p starting %p %lld~%lld\n", inode, req, off, len);
407 ret = ceph_osdc_start_request(osdc, req, false);
408 if (ret < 0)
409 goto out_pages;
410 ceph_osdc_put_request(req);
411
412 /* After adding locked pages to page cache, the inode holds cache cap.
413 * So we can drop our cap refs. */
414 if (got)
415 ceph_put_cap_refs(ci, got);
416
417 return nr_pages;
418
419 out_pages:
420 for (i = 0; i < nr_pages; ++i) {
421 ceph_fscache_readpage_cancel(inode, pages[i]);
422 unlock_page(pages[i]);
423 }
424 ceph_put_page_vector(pages, nr_pages, false);
425 out_put:
426 ceph_osdc_put_request(req);
427 out:
428 if (got)
429 ceph_put_cap_refs(ci, got);
430 return ret;
431 }
432
433
434 /*
435 * Read multiple pages. Leave pages we don't read + unlock in page_list;
436 * the caller (VM) cleans them up.
437 */
438 static int ceph_readpages(struct file *file, struct address_space *mapping,
439 struct list_head *page_list, unsigned nr_pages)
440 {
441 struct inode *inode = file_inode(file);
442 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
443 int rc = 0;
444 int max = 0;
445
446 if (ceph_inode(inode)->i_inline_version != CEPH_INLINE_NONE)
447 return -EINVAL;
448
449 rc = ceph_readpages_from_fscache(mapping->host, mapping, page_list,
450 &nr_pages);
451
452 if (rc == 0)
453 goto out;
454
455 if (fsc->mount_options->rsize >= PAGE_SIZE)
456 max = (fsc->mount_options->rsize + PAGE_SIZE - 1)
457 >> PAGE_SHIFT;
458
459 dout("readpages %p file %p nr_pages %d max %d\n", inode,
460 file, nr_pages,
461 max);
462 while (!list_empty(page_list)) {
463 rc = start_read(inode, page_list, max);
464 if (rc < 0)
465 goto out;
466 }
467 out:
468 ceph_fscache_readpages_cancel(inode, page_list);
469
470 dout("readpages %p file %p ret %d\n", inode, file, rc);
471 return rc;
472 }
473
474 /*
475 * Get ref for the oldest snapc for an inode with dirty data... that is, the
476 * only snap context we are allowed to write back.
477 */
478 static struct ceph_snap_context *get_oldest_context(struct inode *inode,
479 loff_t *snap_size,
480 u64 *truncate_size,
481 u32 *truncate_seq)
482 {
483 struct ceph_inode_info *ci = ceph_inode(inode);
484 struct ceph_snap_context *snapc = NULL;
485 struct ceph_cap_snap *capsnap = NULL;
486
487 spin_lock(&ci->i_ceph_lock);
488 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
489 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap,
490 capsnap->context, capsnap->dirty_pages);
491 if (capsnap->dirty_pages) {
492 snapc = ceph_get_snap_context(capsnap->context);
493 if (snap_size)
494 *snap_size = capsnap->size;
495 if (truncate_size)
496 *truncate_size = capsnap->truncate_size;
497 if (truncate_seq)
498 *truncate_seq = capsnap->truncate_seq;
499 break;
500 }
501 }
502 if (!snapc && ci->i_wrbuffer_ref_head) {
503 snapc = ceph_get_snap_context(ci->i_head_snapc);
504 dout(" head snapc %p has %d dirty pages\n",
505 snapc, ci->i_wrbuffer_ref_head);
506 if (truncate_size)
507 *truncate_size = ci->i_truncate_size;
508 if (truncate_seq)
509 *truncate_seq = ci->i_truncate_seq;
510 }
511 spin_unlock(&ci->i_ceph_lock);
512 return snapc;
513 }
514
515 /*
516 * Write a single page, but leave the page locked.
517 *
518 * If we get a write error, set the page error bit, but still adjust the
519 * dirty page accounting (i.e., page is no longer dirty).
520 */
521 static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
522 {
523 struct inode *inode;
524 struct ceph_inode_info *ci;
525 struct ceph_fs_client *fsc;
526 struct ceph_osd_client *osdc;
527 struct ceph_snap_context *snapc, *oldest;
528 loff_t page_off = page_offset(page);
529 loff_t snap_size = -1;
530 long writeback_stat;
531 u64 truncate_size;
532 u32 truncate_seq;
533 int err, len = PAGE_SIZE;
534
535 dout("writepage %p idx %lu\n", page, page->index);
536
537 inode = page->mapping->host;
538 ci = ceph_inode(inode);
539 fsc = ceph_inode_to_client(inode);
540 osdc = &fsc->client->osdc;
541
542 /* verify this is a writeable snap context */
543 snapc = page_snap_context(page);
544 if (snapc == NULL) {
545 dout("writepage %p page %p not dirty?\n", inode, page);
546 return 0;
547 }
548 oldest = get_oldest_context(inode, &snap_size,
549 &truncate_size, &truncate_seq);
550 if (snapc->seq > oldest->seq) {
551 dout("writepage %p page %p snapc %p not writeable - noop\n",
552 inode, page, snapc);
553 /* we should only noop if called by kswapd */
554 WARN_ON(!(current->flags & PF_MEMALLOC));
555 ceph_put_snap_context(oldest);
556 redirty_page_for_writepage(wbc, page);
557 return 0;
558 }
559 ceph_put_snap_context(oldest);
560
561 if (snap_size == -1)
562 snap_size = i_size_read(inode);
563
564 /* is this a partial page at end of file? */
565 if (page_off >= snap_size) {
566 dout("%p page eof %llu\n", page, snap_size);
567 return 0;
568 }
569
570 if (snap_size < page_off + len)
571 len = snap_size - page_off;
572
573 dout("writepage %p page %p index %lu on %llu~%u snapc %p\n",
574 inode, page, page->index, page_off, len, snapc);
575
576 writeback_stat = atomic_long_inc_return(&fsc->writeback_count);
577 if (writeback_stat >
578 CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb))
579 set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC);
580
581 set_page_writeback(page);
582 err = ceph_osdc_writepages(osdc, ceph_vino(inode),
583 &ci->i_layout, snapc,
584 page_off, len,
585 truncate_seq, truncate_size,
586 &inode->i_mtime, &page, 1);
587 if (err < 0) {
588 struct writeback_control tmp_wbc;
589 if (!wbc)
590 wbc = &tmp_wbc;
591 if (err == -ERESTARTSYS) {
592 /* killed by SIGKILL */
593 dout("writepage interrupted page %p\n", page);
594 redirty_page_for_writepage(wbc, page);
595 end_page_writeback(page);
596 return err;
597 }
598 dout("writepage setting page/mapping error %d %p\n",
599 err, page);
600 SetPageError(page);
601 mapping_set_error(&inode->i_data, err);
602 wbc->pages_skipped++;
603 } else {
604 dout("writepage cleaned page %p\n", page);
605 err = 0; /* vfs expects us to return 0 */
606 }
607 page->private = 0;
608 ClearPagePrivate(page);
609 end_page_writeback(page);
610 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
611 ceph_put_snap_context(snapc); /* page's reference */
612 return err;
613 }
614
615 static int ceph_writepage(struct page *page, struct writeback_control *wbc)
616 {
617 int err;
618 struct inode *inode = page->mapping->host;
619 BUG_ON(!inode);
620 ihold(inode);
621 err = writepage_nounlock(page, wbc);
622 if (err == -ERESTARTSYS) {
623 /* direct memory reclaimer was killed by SIGKILL. return 0
624 * to prevent caller from setting mapping/page error */
625 err = 0;
626 }
627 unlock_page(page);
628 iput(inode);
629 return err;
630 }
631
632 /*
633 * lame release_pages helper. release_pages() isn't exported to
634 * modules.
635 */
636 static void ceph_release_pages(struct page **pages, int num)
637 {
638 struct pagevec pvec;
639 int i;
640
641 pagevec_init(&pvec, 0);
642 for (i = 0; i < num; i++) {
643 if (pagevec_add(&pvec, pages[i]) == 0)
644 pagevec_release(&pvec);
645 }
646 pagevec_release(&pvec);
647 }
648
649 /*
650 * async writeback completion handler.
651 *
652 * If we get an error, set the mapping error bit, but not the individual
653 * page error bits.
654 */
655 static void writepages_finish(struct ceph_osd_request *req)
656 {
657 struct inode *inode = req->r_inode;
658 struct ceph_inode_info *ci = ceph_inode(inode);
659 struct ceph_osd_data *osd_data;
660 struct page *page;
661 int num_pages, total_pages = 0;
662 int i, j;
663 int rc = req->r_result;
664 struct ceph_snap_context *snapc = req->r_snapc;
665 struct address_space *mapping = inode->i_mapping;
666 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
667 bool remove_page;
668
669 dout("writepages_finish %p rc %d\n", inode, rc);
670 if (rc < 0) {
671 mapping_set_error(mapping, rc);
672 ceph_set_error_write(ci);
673 } else {
674 ceph_clear_error_write(ci);
675 }
676
677 /*
678 * We lost the cache cap, need to truncate the page before
679 * it is unlocked, otherwise we'd truncate it later in the
680 * page truncation thread, possibly losing some data that
681 * raced its way in
682 */
683 remove_page = !(ceph_caps_issued(ci) &
684 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO));
685
686 /* clean all pages */
687 for (i = 0; i < req->r_num_ops; i++) {
688 if (req->r_ops[i].op != CEPH_OSD_OP_WRITE)
689 break;
690
691 osd_data = osd_req_op_extent_osd_data(req, i);
692 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
693 num_pages = calc_pages_for((u64)osd_data->alignment,
694 (u64)osd_data->length);
695 total_pages += num_pages;
696 for (j = 0; j < num_pages; j++) {
697 page = osd_data->pages[j];
698 BUG_ON(!page);
699 WARN_ON(!PageUptodate(page));
700
701 if (atomic_long_dec_return(&fsc->writeback_count) <
702 CONGESTION_OFF_THRESH(
703 fsc->mount_options->congestion_kb))
704 clear_bdi_congested(inode_to_bdi(inode),
705 BLK_RW_ASYNC);
706
707 ceph_put_snap_context(page_snap_context(page));
708 page->private = 0;
709 ClearPagePrivate(page);
710 dout("unlocking %p\n", page);
711 end_page_writeback(page);
712
713 if (remove_page)
714 generic_error_remove_page(inode->i_mapping,
715 page);
716
717 unlock_page(page);
718 }
719 dout("writepages_finish %p wrote %llu bytes cleaned %d pages\n",
720 inode, osd_data->length, rc >= 0 ? num_pages : 0);
721
722 ceph_release_pages(osd_data->pages, num_pages);
723 }
724
725 ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc);
726
727 osd_data = osd_req_op_extent_osd_data(req, 0);
728 if (osd_data->pages_from_pool)
729 mempool_free(osd_data->pages,
730 ceph_sb_to_client(inode->i_sb)->wb_pagevec_pool);
731 else
732 kfree(osd_data->pages);
733 ceph_osdc_put_request(req);
734 }
735
736 /*
737 * initiate async writeback
738 */
739 static int ceph_writepages_start(struct address_space *mapping,
740 struct writeback_control *wbc)
741 {
742 struct inode *inode = mapping->host;
743 struct ceph_inode_info *ci = ceph_inode(inode);
744 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
745 struct ceph_vino vino = ceph_vino(inode);
746 pgoff_t index, start, end;
747 int range_whole = 0;
748 int should_loop = 1;
749 pgoff_t max_pages = 0, max_pages_ever = 0;
750 struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc;
751 struct pagevec pvec;
752 int done = 0;
753 int rc = 0;
754 unsigned int wsize = i_blocksize(inode);
755 struct ceph_osd_request *req = NULL;
756 int do_sync = 0;
757 loff_t snap_size, i_size;
758 u64 truncate_size;
759 u32 truncate_seq;
760
761 /*
762 * Include a 'sync' in the OSD request if this is a data
763 * integrity write (e.g., O_SYNC write or fsync()), or if our
764 * cap is being revoked.
765 */
766 if ((wbc->sync_mode == WB_SYNC_ALL) ||
767 ceph_caps_revoking(ci, CEPH_CAP_FILE_BUFFER))
768 do_sync = 1;
769 dout("writepages_start %p dosync=%d (mode=%s)\n",
770 inode, do_sync,
771 wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
772 (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
773
774 if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
775 if (ci->i_wrbuffer_ref > 0) {
776 pr_warn_ratelimited(
777 "writepage_start %p %lld forced umount\n",
778 inode, ceph_ino(inode));
779 }
780 mapping_set_error(mapping, -EIO);
781 return -EIO; /* we're in a forced umount, don't write! */
782 }
783 if (fsc->mount_options->wsize && fsc->mount_options->wsize < wsize)
784 wsize = fsc->mount_options->wsize;
785 if (wsize < PAGE_SIZE)
786 wsize = PAGE_SIZE;
787 max_pages_ever = wsize >> PAGE_SHIFT;
788
789 pagevec_init(&pvec, 0);
790
791 /* where to start/end? */
792 if (wbc->range_cyclic) {
793 start = mapping->writeback_index; /* Start from prev offset */
794 end = -1;
795 dout(" cyclic, start at %lu\n", start);
796 } else {
797 start = wbc->range_start >> PAGE_SHIFT;
798 end = wbc->range_end >> PAGE_SHIFT;
799 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
800 range_whole = 1;
801 should_loop = 0;
802 dout(" not cyclic, %lu to %lu\n", start, end);
803 }
804 index = start;
805
806 retry:
807 /* find oldest snap context with dirty data */
808 ceph_put_snap_context(snapc);
809 snap_size = -1;
810 snapc = get_oldest_context(inode, &snap_size,
811 &truncate_size, &truncate_seq);
812 if (!snapc) {
813 /* hmm, why does writepages get called when there
814 is no dirty data? */
815 dout(" no snap context with dirty data?\n");
816 goto out;
817 }
818 dout(" oldest snapc is %p seq %lld (%d snaps)\n",
819 snapc, snapc->seq, snapc->num_snaps);
820
821 i_size = i_size_read(inode);
822
823 if (last_snapc && snapc != last_snapc) {
824 /* if we switched to a newer snapc, restart our scan at the
825 * start of the original file range. */
826 dout(" snapc differs from last pass, restarting at %lu\n",
827 index);
828 index = start;
829 }
830 last_snapc = snapc;
831
832 while (!done && index <= end) {
833 unsigned i;
834 int first;
835 pgoff_t strip_unit_end = 0;
836 int num_ops = 0, op_idx;
837 int pvec_pages, locked_pages = 0;
838 struct page **pages = NULL, **data_pages;
839 mempool_t *pool = NULL; /* Becomes non-null if mempool used */
840 struct page *page;
841 int want;
842 u64 offset = 0, len = 0;
843
844 max_pages = max_pages_ever;
845
846 get_more_pages:
847 first = -1;
848 want = min(end - index,
849 min((pgoff_t)PAGEVEC_SIZE,
850 max_pages - (pgoff_t)locked_pages) - 1)
851 + 1;
852 pvec_pages = pagevec_lookup_tag(&pvec, mapping, &index,
853 PAGECACHE_TAG_DIRTY,
854 want);
855 dout("pagevec_lookup_tag got %d\n", pvec_pages);
856 if (!pvec_pages && !locked_pages)
857 break;
858 for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) {
859 page = pvec.pages[i];
860 dout("? %p idx %lu\n", page, page->index);
861 if (locked_pages == 0)
862 lock_page(page); /* first page */
863 else if (!trylock_page(page))
864 break;
865
866 /* only dirty pages, or our accounting breaks */
867 if (unlikely(!PageDirty(page)) ||
868 unlikely(page->mapping != mapping)) {
869 dout("!dirty or !mapping %p\n", page);
870 unlock_page(page);
871 break;
872 }
873 if (!wbc->range_cyclic && page->index > end) {
874 dout("end of range %p\n", page);
875 done = 1;
876 unlock_page(page);
877 break;
878 }
879 if (strip_unit_end && (page->index > strip_unit_end)) {
880 dout("end of strip unit %p\n", page);
881 unlock_page(page);
882 break;
883 }
884 if (wbc->sync_mode != WB_SYNC_NONE) {
885 dout("waiting on writeback %p\n", page);
886 wait_on_page_writeback(page);
887 }
888 if (page_offset(page) >=
889 (snap_size == -1 ? i_size : snap_size)) {
890 dout("%p page eof %llu\n", page,
891 (snap_size == -1 ? i_size : snap_size));
892 done = 1;
893 unlock_page(page);
894 break;
895 }
896 if (PageWriteback(page)) {
897 dout("%p under writeback\n", page);
898 unlock_page(page);
899 break;
900 }
901
902 /* only if matching snap context */
903 pgsnapc = page_snap_context(page);
904 if (pgsnapc->seq > snapc->seq) {
905 dout("page snapc %p %lld > oldest %p %lld\n",
906 pgsnapc, pgsnapc->seq, snapc, snapc->seq);
907 unlock_page(page);
908 if (!locked_pages)
909 continue; /* keep looking for snap */
910 break;
911 }
912
913 if (!clear_page_dirty_for_io(page)) {
914 dout("%p !clear_page_dirty_for_io\n", page);
915 unlock_page(page);
916 break;
917 }
918
919 /*
920 * We have something to write. If this is
921 * the first locked page this time through,
922 * calculate max possinle write size and
923 * allocate a page array
924 */
925 if (locked_pages == 0) {
926 u64 objnum;
927 u64 objoff;
928
929 /* prepare async write request */
930 offset = (u64)page_offset(page);
931 len = wsize;
932
933 rc = ceph_calc_file_object_mapping(&ci->i_layout,
934 offset, len,
935 &objnum, &objoff,
936 &len);
937 if (rc < 0) {
938 unlock_page(page);
939 break;
940 }
941
942 num_ops = 1 + do_sync;
943 strip_unit_end = page->index +
944 ((len - 1) >> PAGE_SHIFT);
945
946 BUG_ON(pages);
947 max_pages = calc_pages_for(0, (u64)len);
948 pages = kmalloc(max_pages * sizeof (*pages),
949 GFP_NOFS);
950 if (!pages) {
951 pool = fsc->wb_pagevec_pool;
952 pages = mempool_alloc(pool, GFP_NOFS);
953 BUG_ON(!pages);
954 }
955
956 len = 0;
957 } else if (page->index !=
958 (offset + len) >> PAGE_SHIFT) {
959 if (num_ops >= (pool ? CEPH_OSD_SLAB_OPS :
960 CEPH_OSD_MAX_OPS)) {
961 redirty_page_for_writepage(wbc, page);
962 unlock_page(page);
963 break;
964 }
965
966 num_ops++;
967 offset = (u64)page_offset(page);
968 len = 0;
969 }
970
971 /* note position of first page in pvec */
972 if (first < 0)
973 first = i;
974 dout("%p will write page %p idx %lu\n",
975 inode, page, page->index);
976
977 if (atomic_long_inc_return(&fsc->writeback_count) >
978 CONGESTION_ON_THRESH(
979 fsc->mount_options->congestion_kb)) {
980 set_bdi_congested(inode_to_bdi(inode),
981 BLK_RW_ASYNC);
982 }
983
984 pages[locked_pages] = page;
985 locked_pages++;
986 len += PAGE_SIZE;
987 }
988
989 /* did we get anything? */
990 if (!locked_pages)
991 goto release_pvec_pages;
992 if (i) {
993 int j;
994 BUG_ON(!locked_pages || first < 0);
995
996 if (pvec_pages && i == pvec_pages &&
997 locked_pages < max_pages) {
998 dout("reached end pvec, trying for more\n");
999 pagevec_reinit(&pvec);
1000 goto get_more_pages;
1001 }
1002
1003 /* shift unused pages over in the pvec... we
1004 * will need to release them below. */
1005 for (j = i; j < pvec_pages; j++) {
1006 dout(" pvec leftover page %p\n", pvec.pages[j]);
1007 pvec.pages[j-i+first] = pvec.pages[j];
1008 }
1009 pvec.nr -= i-first;
1010 }
1011
1012 new_request:
1013 offset = page_offset(pages[0]);
1014 len = wsize;
1015
1016 req = ceph_osdc_new_request(&fsc->client->osdc,
1017 &ci->i_layout, vino,
1018 offset, &len, 0, num_ops,
1019 CEPH_OSD_OP_WRITE,
1020 CEPH_OSD_FLAG_WRITE,
1021 snapc, truncate_seq,
1022 truncate_size, false);
1023 if (IS_ERR(req)) {
1024 req = ceph_osdc_new_request(&fsc->client->osdc,
1025 &ci->i_layout, vino,
1026 offset, &len, 0,
1027 min(num_ops,
1028 CEPH_OSD_SLAB_OPS),
1029 CEPH_OSD_OP_WRITE,
1030 CEPH_OSD_FLAG_WRITE,
1031 snapc, truncate_seq,
1032 truncate_size, true);
1033 BUG_ON(IS_ERR(req));
1034 }
1035 BUG_ON(len < page_offset(pages[locked_pages - 1]) +
1036 PAGE_SIZE - offset);
1037
1038 req->r_callback = writepages_finish;
1039 req->r_inode = inode;
1040
1041 /* Format the osd request message and submit the write */
1042 len = 0;
1043 data_pages = pages;
1044 op_idx = 0;
1045 for (i = 0; i < locked_pages; i++) {
1046 u64 cur_offset = page_offset(pages[i]);
1047 if (offset + len != cur_offset) {
1048 if (op_idx + do_sync + 1 == req->r_num_ops)
1049 break;
1050 osd_req_op_extent_dup_last(req, op_idx,
1051 cur_offset - offset);
1052 dout("writepages got pages at %llu~%llu\n",
1053 offset, len);
1054 osd_req_op_extent_osd_data_pages(req, op_idx,
1055 data_pages, len, 0,
1056 !!pool, false);
1057 osd_req_op_extent_update(req, op_idx, len);
1058
1059 len = 0;
1060 offset = cur_offset;
1061 data_pages = pages + i;
1062 op_idx++;
1063 }
1064
1065 set_page_writeback(pages[i]);
1066 len += PAGE_SIZE;
1067 }
1068
1069 if (snap_size != -1) {
1070 len = min(len, snap_size - offset);
1071 } else if (i == locked_pages) {
1072 /* writepages_finish() clears writeback pages
1073 * according to the data length, so make sure
1074 * data length covers all locked pages */
1075 u64 min_len = len + 1 - PAGE_SIZE;
1076 len = min(len, (u64)i_size_read(inode) - offset);
1077 len = max(len, min_len);
1078 }
1079 dout("writepages got pages at %llu~%llu\n", offset, len);
1080
1081 osd_req_op_extent_osd_data_pages(req, op_idx, data_pages, len,
1082 0, !!pool, false);
1083 osd_req_op_extent_update(req, op_idx, len);
1084
1085 if (do_sync) {
1086 op_idx++;
1087 osd_req_op_init(req, op_idx, CEPH_OSD_OP_STARTSYNC, 0);
1088 }
1089 BUG_ON(op_idx + 1 != req->r_num_ops);
1090
1091 pool = NULL;
1092 if (i < locked_pages) {
1093 BUG_ON(num_ops <= req->r_num_ops);
1094 num_ops -= req->r_num_ops;
1095 num_ops += do_sync;
1096 locked_pages -= i;
1097
1098 /* allocate new pages array for next request */
1099 data_pages = pages;
1100 pages = kmalloc(locked_pages * sizeof (*pages),
1101 GFP_NOFS);
1102 if (!pages) {
1103 pool = fsc->wb_pagevec_pool;
1104 pages = mempool_alloc(pool, GFP_NOFS);
1105 BUG_ON(!pages);
1106 }
1107 memcpy(pages, data_pages + i,
1108 locked_pages * sizeof(*pages));
1109 memset(data_pages + i, 0,
1110 locked_pages * sizeof(*pages));
1111 } else {
1112 BUG_ON(num_ops != req->r_num_ops);
1113 index = pages[i - 1]->index + 1;
1114 /* request message now owns the pages array */
1115 pages = NULL;
1116 }
1117
1118 req->r_mtime = inode->i_mtime;
1119 rc = ceph_osdc_start_request(&fsc->client->osdc, req, true);
1120 BUG_ON(rc);
1121 req = NULL;
1122
1123 wbc->nr_to_write -= i;
1124 if (pages)
1125 goto new_request;
1126
1127 if (wbc->nr_to_write <= 0)
1128 done = 1;
1129
1130 release_pvec_pages:
1131 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr,
1132 pvec.nr ? pvec.pages[0] : NULL);
1133 pagevec_release(&pvec);
1134
1135 if (locked_pages && !done)
1136 goto retry;
1137 }
1138
1139 if (should_loop && !done) {
1140 /* more to do; loop back to beginning of file */
1141 dout("writepages looping back to beginning of file\n");
1142 should_loop = 0;
1143 index = 0;
1144 goto retry;
1145 }
1146
1147 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1148 mapping->writeback_index = index;
1149
1150 out:
1151 ceph_osdc_put_request(req);
1152 ceph_put_snap_context(snapc);
1153 dout("writepages done, rc = %d\n", rc);
1154 return rc;
1155 }
1156
1157
1158
1159 /*
1160 * See if a given @snapc is either writeable, or already written.
1161 */
1162 static int context_is_writeable_or_written(struct inode *inode,
1163 struct ceph_snap_context *snapc)
1164 {
1165 struct ceph_snap_context *oldest = get_oldest_context(inode, NULL,
1166 NULL, NULL);
1167 int ret = !oldest || snapc->seq <= oldest->seq;
1168
1169 ceph_put_snap_context(oldest);
1170 return ret;
1171 }
1172
1173 /*
1174 * We are only allowed to write into/dirty the page if the page is
1175 * clean, or already dirty within the same snap context.
1176 *
1177 * called with page locked.
1178 * return success with page locked,
1179 * or any failure (incl -EAGAIN) with page unlocked.
1180 */
1181 static int ceph_update_writeable_page(struct file *file,
1182 loff_t pos, unsigned len,
1183 struct page *page)
1184 {
1185 struct inode *inode = file_inode(file);
1186 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1187 struct ceph_inode_info *ci = ceph_inode(inode);
1188 loff_t page_off = pos & PAGE_MASK;
1189 int pos_in_page = pos & ~PAGE_MASK;
1190 int end_in_page = pos_in_page + len;
1191 loff_t i_size;
1192 int r;
1193 struct ceph_snap_context *snapc, *oldest;
1194
1195 if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1196 dout(" page %p forced umount\n", page);
1197 unlock_page(page);
1198 return -EIO;
1199 }
1200
1201 retry_locked:
1202 /* writepages currently holds page lock, but if we change that later, */
1203 wait_on_page_writeback(page);
1204
1205 snapc = page_snap_context(page);
1206 if (snapc && snapc != ci->i_head_snapc) {
1207 /*
1208 * this page is already dirty in another (older) snap
1209 * context! is it writeable now?
1210 */
1211 oldest = get_oldest_context(inode, NULL, NULL, NULL);
1212
1213 if (snapc->seq > oldest->seq) {
1214 ceph_put_snap_context(oldest);
1215 dout(" page %p snapc %p not current or oldest\n",
1216 page, snapc);
1217 /*
1218 * queue for writeback, and wait for snapc to
1219 * be writeable or written
1220 */
1221 snapc = ceph_get_snap_context(snapc);
1222 unlock_page(page);
1223 ceph_queue_writeback(inode);
1224 r = wait_event_killable(ci->i_cap_wq,
1225 context_is_writeable_or_written(inode, snapc));
1226 ceph_put_snap_context(snapc);
1227 if (r == -ERESTARTSYS)
1228 return r;
1229 return -EAGAIN;
1230 }
1231 ceph_put_snap_context(oldest);
1232
1233 /* yay, writeable, do it now (without dropping page lock) */
1234 dout(" page %p snapc %p not current, but oldest\n",
1235 page, snapc);
1236 if (!clear_page_dirty_for_io(page))
1237 goto retry_locked;
1238 r = writepage_nounlock(page, NULL);
1239 if (r < 0)
1240 goto fail_nosnap;
1241 goto retry_locked;
1242 }
1243
1244 if (PageUptodate(page)) {
1245 dout(" page %p already uptodate\n", page);
1246 return 0;
1247 }
1248
1249 /* full page? */
1250 if (pos_in_page == 0 && len == PAGE_SIZE)
1251 return 0;
1252
1253 /* past end of file? */
1254 i_size = i_size_read(inode);
1255
1256 if (page_off >= i_size ||
1257 (pos_in_page == 0 && (pos+len) >= i_size &&
1258 end_in_page - pos_in_page != PAGE_SIZE)) {
1259 dout(" zeroing %p 0 - %d and %d - %d\n",
1260 page, pos_in_page, end_in_page, (int)PAGE_SIZE);
1261 zero_user_segments(page,
1262 0, pos_in_page,
1263 end_in_page, PAGE_SIZE);
1264 return 0;
1265 }
1266
1267 /* we need to read it. */
1268 r = readpage_nounlock(file, page);
1269 if (r < 0)
1270 goto fail_nosnap;
1271 goto retry_locked;
1272 fail_nosnap:
1273 unlock_page(page);
1274 return r;
1275 }
1276
1277 /*
1278 * We are only allowed to write into/dirty the page if the page is
1279 * clean, or already dirty within the same snap context.
1280 */
1281 static int ceph_write_begin(struct file *file, struct address_space *mapping,
1282 loff_t pos, unsigned len, unsigned flags,
1283 struct page **pagep, void **fsdata)
1284 {
1285 struct inode *inode = file_inode(file);
1286 struct page *page;
1287 pgoff_t index = pos >> PAGE_SHIFT;
1288 int r;
1289
1290 do {
1291 /* get a page */
1292 page = grab_cache_page_write_begin(mapping, index, 0);
1293 if (!page)
1294 return -ENOMEM;
1295
1296 dout("write_begin file %p inode %p page %p %d~%d\n", file,
1297 inode, page, (int)pos, (int)len);
1298
1299 r = ceph_update_writeable_page(file, pos, len, page);
1300 if (r < 0)
1301 put_page(page);
1302 else
1303 *pagep = page;
1304 } while (r == -EAGAIN);
1305
1306 return r;
1307 }
1308
1309 /*
1310 * we don't do anything in here that simple_write_end doesn't do
1311 * except adjust dirty page accounting
1312 */
1313 static int ceph_write_end(struct file *file, struct address_space *mapping,
1314 loff_t pos, unsigned len, unsigned copied,
1315 struct page *page, void *fsdata)
1316 {
1317 struct inode *inode = file_inode(file);
1318 bool check_cap = false;
1319
1320 dout("write_end file %p inode %p page %p %d~%d (%d)\n", file,
1321 inode, page, (int)pos, (int)copied, (int)len);
1322
1323 /* zero the stale part of the page if we did a short copy */
1324 if (!PageUptodate(page)) {
1325 if (copied < len) {
1326 copied = 0;
1327 goto out;
1328 }
1329 SetPageUptodate(page);
1330 }
1331
1332 /* did file size increase? */
1333 if (pos+copied > i_size_read(inode))
1334 check_cap = ceph_inode_set_size(inode, pos+copied);
1335
1336 set_page_dirty(page);
1337
1338 out:
1339 unlock_page(page);
1340 put_page(page);
1341
1342 if (check_cap)
1343 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL);
1344
1345 return copied;
1346 }
1347
1348 /*
1349 * we set .direct_IO to indicate direct io is supported, but since we
1350 * intercept O_DIRECT reads and writes early, this function should
1351 * never get called.
1352 */
1353 static ssize_t ceph_direct_io(struct kiocb *iocb, struct iov_iter *iter)
1354 {
1355 WARN_ON(1);
1356 return -EINVAL;
1357 }
1358
1359 const struct address_space_operations ceph_aops = {
1360 .readpage = ceph_readpage,
1361 .readpages = ceph_readpages,
1362 .writepage = ceph_writepage,
1363 .writepages = ceph_writepages_start,
1364 .write_begin = ceph_write_begin,
1365 .write_end = ceph_write_end,
1366 .set_page_dirty = ceph_set_page_dirty,
1367 .invalidatepage = ceph_invalidatepage,
1368 .releasepage = ceph_releasepage,
1369 .direct_IO = ceph_direct_io,
1370 };
1371
1372 static void ceph_block_sigs(sigset_t *oldset)
1373 {
1374 sigset_t mask;
1375 siginitsetinv(&mask, sigmask(SIGKILL));
1376 sigprocmask(SIG_BLOCK, &mask, oldset);
1377 }
1378
1379 static void ceph_restore_sigs(sigset_t *oldset)
1380 {
1381 sigprocmask(SIG_SETMASK, oldset, NULL);
1382 }
1383
1384 /*
1385 * vm ops
1386 */
1387 static int ceph_filemap_fault(struct vm_fault *vmf)
1388 {
1389 struct vm_area_struct *vma = vmf->vma;
1390 struct inode *inode = file_inode(vma->vm_file);
1391 struct ceph_inode_info *ci = ceph_inode(inode);
1392 struct ceph_file_info *fi = vma->vm_file->private_data;
1393 struct page *pinned_page = NULL;
1394 loff_t off = vmf->pgoff << PAGE_SHIFT;
1395 int want, got, ret;
1396 sigset_t oldset;
1397
1398 ceph_block_sigs(&oldset);
1399
1400 dout("filemap_fault %p %llx.%llx %llu~%zd trying to get caps\n",
1401 inode, ceph_vinop(inode), off, (size_t)PAGE_SIZE);
1402 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1403 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1404 else
1405 want = CEPH_CAP_FILE_CACHE;
1406
1407 got = 0;
1408 ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page);
1409 if (ret < 0)
1410 goto out_restore;
1411
1412 dout("filemap_fault %p %llu~%zd got cap refs on %s\n",
1413 inode, off, (size_t)PAGE_SIZE, ceph_cap_string(got));
1414
1415 if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) ||
1416 ci->i_inline_version == CEPH_INLINE_NONE) {
1417 current->journal_info = vma->vm_file;
1418 ret = filemap_fault(vmf);
1419 current->journal_info = NULL;
1420 } else
1421 ret = -EAGAIN;
1422
1423 dout("filemap_fault %p %llu~%zd dropping cap refs on %s ret %d\n",
1424 inode, off, (size_t)PAGE_SIZE, ceph_cap_string(got), ret);
1425 if (pinned_page)
1426 put_page(pinned_page);
1427 ceph_put_cap_refs(ci, got);
1428
1429 if (ret != -EAGAIN)
1430 goto out_restore;
1431
1432 /* read inline data */
1433 if (off >= PAGE_SIZE) {
1434 /* does not support inline data > PAGE_SIZE */
1435 ret = VM_FAULT_SIGBUS;
1436 } else {
1437 int ret1;
1438 struct address_space *mapping = inode->i_mapping;
1439 struct page *page = find_or_create_page(mapping, 0,
1440 mapping_gfp_constraint(mapping,
1441 ~__GFP_FS));
1442 if (!page) {
1443 ret = VM_FAULT_OOM;
1444 goto out_inline;
1445 }
1446 ret1 = __ceph_do_getattr(inode, page,
1447 CEPH_STAT_CAP_INLINE_DATA, true);
1448 if (ret1 < 0 || off >= i_size_read(inode)) {
1449 unlock_page(page);
1450 put_page(page);
1451 if (ret1 < 0)
1452 ret = ret1;
1453 else
1454 ret = VM_FAULT_SIGBUS;
1455 goto out_inline;
1456 }
1457 if (ret1 < PAGE_SIZE)
1458 zero_user_segment(page, ret1, PAGE_SIZE);
1459 else
1460 flush_dcache_page(page);
1461 SetPageUptodate(page);
1462 vmf->page = page;
1463 ret = VM_FAULT_MAJOR | VM_FAULT_LOCKED;
1464 out_inline:
1465 dout("filemap_fault %p %llu~%zd read inline data ret %d\n",
1466 inode, off, (size_t)PAGE_SIZE, ret);
1467 }
1468 out_restore:
1469 ceph_restore_sigs(&oldset);
1470 if (ret < 0)
1471 ret = (ret == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS;
1472
1473 return ret;
1474 }
1475
1476 /*
1477 * Reuse write_begin here for simplicity.
1478 */
1479 static int ceph_page_mkwrite(struct vm_fault *vmf)
1480 {
1481 struct vm_area_struct *vma = vmf->vma;
1482 struct inode *inode = file_inode(vma->vm_file);
1483 struct ceph_inode_info *ci = ceph_inode(inode);
1484 struct ceph_file_info *fi = vma->vm_file->private_data;
1485 struct ceph_cap_flush *prealloc_cf;
1486 struct page *page = vmf->page;
1487 loff_t off = page_offset(page);
1488 loff_t size = i_size_read(inode);
1489 size_t len;
1490 int want, got, ret;
1491 sigset_t oldset;
1492
1493 prealloc_cf = ceph_alloc_cap_flush();
1494 if (!prealloc_cf)
1495 return VM_FAULT_OOM;
1496
1497 ceph_block_sigs(&oldset);
1498
1499 if (ci->i_inline_version != CEPH_INLINE_NONE) {
1500 struct page *locked_page = NULL;
1501 if (off == 0) {
1502 lock_page(page);
1503 locked_page = page;
1504 }
1505 ret = ceph_uninline_data(vma->vm_file, locked_page);
1506 if (locked_page)
1507 unlock_page(locked_page);
1508 if (ret < 0)
1509 goto out_free;
1510 }
1511
1512 if (off + PAGE_SIZE <= size)
1513 len = PAGE_SIZE;
1514 else
1515 len = size & ~PAGE_MASK;
1516
1517 dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n",
1518 inode, ceph_vinop(inode), off, len, size);
1519 if (fi->fmode & CEPH_FILE_MODE_LAZY)
1520 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1521 else
1522 want = CEPH_CAP_FILE_BUFFER;
1523
1524 got = 0;
1525 ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, off + len,
1526 &got, NULL);
1527 if (ret < 0)
1528 goto out_free;
1529
1530 dout("page_mkwrite %p %llu~%zd got cap refs on %s\n",
1531 inode, off, len, ceph_cap_string(got));
1532
1533 /* Update time before taking page lock */
1534 file_update_time(vma->vm_file);
1535
1536 do {
1537 lock_page(page);
1538
1539 if ((off > size) || (page->mapping != inode->i_mapping)) {
1540 unlock_page(page);
1541 ret = VM_FAULT_NOPAGE;
1542 break;
1543 }
1544
1545 ret = ceph_update_writeable_page(vma->vm_file, off, len, page);
1546 if (ret >= 0) {
1547 /* success. we'll keep the page locked. */
1548 set_page_dirty(page);
1549 ret = VM_FAULT_LOCKED;
1550 }
1551 } while (ret == -EAGAIN);
1552
1553 if (ret == VM_FAULT_LOCKED ||
1554 ci->i_inline_version != CEPH_INLINE_NONE) {
1555 int dirty;
1556 spin_lock(&ci->i_ceph_lock);
1557 ci->i_inline_version = CEPH_INLINE_NONE;
1558 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1559 &prealloc_cf);
1560 spin_unlock(&ci->i_ceph_lock);
1561 if (dirty)
1562 __mark_inode_dirty(inode, dirty);
1563 }
1564
1565 dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %d\n",
1566 inode, off, len, ceph_cap_string(got), ret);
1567 ceph_put_cap_refs(ci, got);
1568 out_free:
1569 ceph_restore_sigs(&oldset);
1570 ceph_free_cap_flush(prealloc_cf);
1571 if (ret < 0)
1572 ret = (ret == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS;
1573 return ret;
1574 }
1575
1576 void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
1577 char *data, size_t len)
1578 {
1579 struct address_space *mapping = inode->i_mapping;
1580 struct page *page;
1581
1582 if (locked_page) {
1583 page = locked_page;
1584 } else {
1585 if (i_size_read(inode) == 0)
1586 return;
1587 page = find_or_create_page(mapping, 0,
1588 mapping_gfp_constraint(mapping,
1589 ~__GFP_FS));
1590 if (!page)
1591 return;
1592 if (PageUptodate(page)) {
1593 unlock_page(page);
1594 put_page(page);
1595 return;
1596 }
1597 }
1598
1599 dout("fill_inline_data %p %llx.%llx len %zu locked_page %p\n",
1600 inode, ceph_vinop(inode), len, locked_page);
1601
1602 if (len > 0) {
1603 void *kaddr = kmap_atomic(page);
1604 memcpy(kaddr, data, len);
1605 kunmap_atomic(kaddr);
1606 }
1607
1608 if (page != locked_page) {
1609 if (len < PAGE_SIZE)
1610 zero_user_segment(page, len, PAGE_SIZE);
1611 else
1612 flush_dcache_page(page);
1613
1614 SetPageUptodate(page);
1615 unlock_page(page);
1616 put_page(page);
1617 }
1618 }
1619
1620 int ceph_uninline_data(struct file *filp, struct page *locked_page)
1621 {
1622 struct inode *inode = file_inode(filp);
1623 struct ceph_inode_info *ci = ceph_inode(inode);
1624 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1625 struct ceph_osd_request *req;
1626 struct page *page = NULL;
1627 u64 len, inline_version;
1628 int err = 0;
1629 bool from_pagecache = false;
1630
1631 spin_lock(&ci->i_ceph_lock);
1632 inline_version = ci->i_inline_version;
1633 spin_unlock(&ci->i_ceph_lock);
1634
1635 dout("uninline_data %p %llx.%llx inline_version %llu\n",
1636 inode, ceph_vinop(inode), inline_version);
1637
1638 if (inline_version == 1 || /* initial version, no data */
1639 inline_version == CEPH_INLINE_NONE)
1640 goto out;
1641
1642 if (locked_page) {
1643 page = locked_page;
1644 WARN_ON(!PageUptodate(page));
1645 } else if (ceph_caps_issued(ci) &
1646 (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) {
1647 page = find_get_page(inode->i_mapping, 0);
1648 if (page) {
1649 if (PageUptodate(page)) {
1650 from_pagecache = true;
1651 lock_page(page);
1652 } else {
1653 put_page(page);
1654 page = NULL;
1655 }
1656 }
1657 }
1658
1659 if (page) {
1660 len = i_size_read(inode);
1661 if (len > PAGE_SIZE)
1662 len = PAGE_SIZE;
1663 } else {
1664 page = __page_cache_alloc(GFP_NOFS);
1665 if (!page) {
1666 err = -ENOMEM;
1667 goto out;
1668 }
1669 err = __ceph_do_getattr(inode, page,
1670 CEPH_STAT_CAP_INLINE_DATA, true);
1671 if (err < 0) {
1672 /* no inline data */
1673 if (err == -ENODATA)
1674 err = 0;
1675 goto out;
1676 }
1677 len = err;
1678 }
1679
1680 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1681 ceph_vino(inode), 0, &len, 0, 1,
1682 CEPH_OSD_OP_CREATE, CEPH_OSD_FLAG_WRITE,
1683 NULL, 0, 0, false);
1684 if (IS_ERR(req)) {
1685 err = PTR_ERR(req);
1686 goto out;
1687 }
1688
1689 req->r_mtime = inode->i_mtime;
1690 err = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1691 if (!err)
1692 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1693 ceph_osdc_put_request(req);
1694 if (err < 0)
1695 goto out;
1696
1697 req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1698 ceph_vino(inode), 0, &len, 1, 3,
1699 CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
1700 NULL, ci->i_truncate_seq,
1701 ci->i_truncate_size, false);
1702 if (IS_ERR(req)) {
1703 err = PTR_ERR(req);
1704 goto out;
1705 }
1706
1707 osd_req_op_extent_osd_data_pages(req, 1, &page, len, 0, false, false);
1708
1709 {
1710 __le64 xattr_buf = cpu_to_le64(inline_version);
1711 err = osd_req_op_xattr_init(req, 0, CEPH_OSD_OP_CMPXATTR,
1712 "inline_version", &xattr_buf,
1713 sizeof(xattr_buf),
1714 CEPH_OSD_CMPXATTR_OP_GT,
1715 CEPH_OSD_CMPXATTR_MODE_U64);
1716 if (err)
1717 goto out_put;
1718 }
1719
1720 {
1721 char xattr_buf[32];
1722 int xattr_len = snprintf(xattr_buf, sizeof(xattr_buf),
1723 "%llu", inline_version);
1724 err = osd_req_op_xattr_init(req, 2, CEPH_OSD_OP_SETXATTR,
1725 "inline_version",
1726 xattr_buf, xattr_len, 0, 0);
1727 if (err)
1728 goto out_put;
1729 }
1730
1731 req->r_mtime = inode->i_mtime;
1732 err = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1733 if (!err)
1734 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1735 out_put:
1736 ceph_osdc_put_request(req);
1737 if (err == -ECANCELED)
1738 err = 0;
1739 out:
1740 if (page && page != locked_page) {
1741 if (from_pagecache) {
1742 unlock_page(page);
1743 put_page(page);
1744 } else
1745 __free_pages(page, 0);
1746 }
1747
1748 dout("uninline_data %p %llx.%llx inline_version %llu = %d\n",
1749 inode, ceph_vinop(inode), inline_version, err);
1750 return err;
1751 }
1752
1753 static const struct vm_operations_struct ceph_vmops = {
1754 .fault = ceph_filemap_fault,
1755 .page_mkwrite = ceph_page_mkwrite,
1756 };
1757
1758 int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1759 {
1760 struct address_space *mapping = file->f_mapping;
1761
1762 if (!mapping->a_ops->readpage)
1763 return -ENOEXEC;
1764 file_accessed(file);
1765 vma->vm_ops = &ceph_vmops;
1766 return 0;
1767 }
1768
1769 enum {
1770 POOL_READ = 1,
1771 POOL_WRITE = 2,
1772 };
1773
1774 static int __ceph_pool_perm_get(struct ceph_inode_info *ci,
1775 s64 pool, struct ceph_string *pool_ns)
1776 {
1777 struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode);
1778 struct ceph_mds_client *mdsc = fsc->mdsc;
1779 struct ceph_osd_request *rd_req = NULL, *wr_req = NULL;
1780 struct rb_node **p, *parent;
1781 struct ceph_pool_perm *perm;
1782 struct page **pages;
1783 size_t pool_ns_len;
1784 int err = 0, err2 = 0, have = 0;
1785
1786 down_read(&mdsc->pool_perm_rwsem);
1787 p = &mdsc->pool_perm_tree.rb_node;
1788 while (*p) {
1789 perm = rb_entry(*p, struct ceph_pool_perm, node);
1790 if (pool < perm->pool)
1791 p = &(*p)->rb_left;
1792 else if (pool > perm->pool)
1793 p = &(*p)->rb_right;
1794 else {
1795 int ret = ceph_compare_string(pool_ns,
1796 perm->pool_ns,
1797 perm->pool_ns_len);
1798 if (ret < 0)
1799 p = &(*p)->rb_left;
1800 else if (ret > 0)
1801 p = &(*p)->rb_right;
1802 else {
1803 have = perm->perm;
1804 break;
1805 }
1806 }
1807 }
1808 up_read(&mdsc->pool_perm_rwsem);
1809 if (*p)
1810 goto out;
1811
1812 if (pool_ns)
1813 dout("__ceph_pool_perm_get pool %lld ns %.*s no perm cached\n",
1814 pool, (int)pool_ns->len, pool_ns->str);
1815 else
1816 dout("__ceph_pool_perm_get pool %lld no perm cached\n", pool);
1817
1818 down_write(&mdsc->pool_perm_rwsem);
1819 p = &mdsc->pool_perm_tree.rb_node;
1820 parent = NULL;
1821 while (*p) {
1822 parent = *p;
1823 perm = rb_entry(parent, struct ceph_pool_perm, node);
1824 if (pool < perm->pool)
1825 p = &(*p)->rb_left;
1826 else if (pool > perm->pool)
1827 p = &(*p)->rb_right;
1828 else {
1829 int ret = ceph_compare_string(pool_ns,
1830 perm->pool_ns,
1831 perm->pool_ns_len);
1832 if (ret < 0)
1833 p = &(*p)->rb_left;
1834 else if (ret > 0)
1835 p = &(*p)->rb_right;
1836 else {
1837 have = perm->perm;
1838 break;
1839 }
1840 }
1841 }
1842 if (*p) {
1843 up_write(&mdsc->pool_perm_rwsem);
1844 goto out;
1845 }
1846
1847 rd_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1848 1, false, GFP_NOFS);
1849 if (!rd_req) {
1850 err = -ENOMEM;
1851 goto out_unlock;
1852 }
1853
1854 rd_req->r_flags = CEPH_OSD_FLAG_READ;
1855 osd_req_op_init(rd_req, 0, CEPH_OSD_OP_STAT, 0);
1856 rd_req->r_base_oloc.pool = pool;
1857 if (pool_ns)
1858 rd_req->r_base_oloc.pool_ns = ceph_get_string(pool_ns);
1859 ceph_oid_printf(&rd_req->r_base_oid, "%llx.00000000", ci->i_vino.ino);
1860
1861 err = ceph_osdc_alloc_messages(rd_req, GFP_NOFS);
1862 if (err)
1863 goto out_unlock;
1864
1865 wr_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1866 1, false, GFP_NOFS);
1867 if (!wr_req) {
1868 err = -ENOMEM;
1869 goto out_unlock;
1870 }
1871
1872 wr_req->r_flags = CEPH_OSD_FLAG_WRITE;
1873 osd_req_op_init(wr_req, 0, CEPH_OSD_OP_CREATE, CEPH_OSD_OP_FLAG_EXCL);
1874 ceph_oloc_copy(&wr_req->r_base_oloc, &rd_req->r_base_oloc);
1875 ceph_oid_copy(&wr_req->r_base_oid, &rd_req->r_base_oid);
1876
1877 err = ceph_osdc_alloc_messages(wr_req, GFP_NOFS);
1878 if (err)
1879 goto out_unlock;
1880
1881 /* one page should be large enough for STAT data */
1882 pages = ceph_alloc_page_vector(1, GFP_KERNEL);
1883 if (IS_ERR(pages)) {
1884 err = PTR_ERR(pages);
1885 goto out_unlock;
1886 }
1887
1888 osd_req_op_raw_data_in_pages(rd_req, 0, pages, PAGE_SIZE,
1889 0, false, true);
1890 err = ceph_osdc_start_request(&fsc->client->osdc, rd_req, false);
1891
1892 wr_req->r_mtime = ci->vfs_inode.i_mtime;
1893 wr_req->r_abort_on_full = true;
1894 err2 = ceph_osdc_start_request(&fsc->client->osdc, wr_req, false);
1895
1896 if (!err)
1897 err = ceph_osdc_wait_request(&fsc->client->osdc, rd_req);
1898 if (!err2)
1899 err2 = ceph_osdc_wait_request(&fsc->client->osdc, wr_req);
1900
1901 if (err >= 0 || err == -ENOENT)
1902 have |= POOL_READ;
1903 else if (err != -EPERM)
1904 goto out_unlock;
1905
1906 if (err2 == 0 || err2 == -EEXIST)
1907 have |= POOL_WRITE;
1908 else if (err2 != -EPERM) {
1909 err = err2;
1910 goto out_unlock;
1911 }
1912
1913 pool_ns_len = pool_ns ? pool_ns->len : 0;
1914 perm = kmalloc(sizeof(*perm) + pool_ns_len + 1, GFP_NOFS);
1915 if (!perm) {
1916 err = -ENOMEM;
1917 goto out_unlock;
1918 }
1919
1920 perm->pool = pool;
1921 perm->perm = have;
1922 perm->pool_ns_len = pool_ns_len;
1923 if (pool_ns_len > 0)
1924 memcpy(perm->pool_ns, pool_ns->str, pool_ns_len);
1925 perm->pool_ns[pool_ns_len] = 0;
1926
1927 rb_link_node(&perm->node, parent, p);
1928 rb_insert_color(&perm->node, &mdsc->pool_perm_tree);
1929 err = 0;
1930 out_unlock:
1931 up_write(&mdsc->pool_perm_rwsem);
1932
1933 ceph_osdc_put_request(rd_req);
1934 ceph_osdc_put_request(wr_req);
1935 out:
1936 if (!err)
1937 err = have;
1938 if (pool_ns)
1939 dout("__ceph_pool_perm_get pool %lld ns %.*s result = %d\n",
1940 pool, (int)pool_ns->len, pool_ns->str, err);
1941 else
1942 dout("__ceph_pool_perm_get pool %lld result = %d\n", pool, err);
1943 return err;
1944 }
1945
1946 int ceph_pool_perm_check(struct ceph_inode_info *ci, int need)
1947 {
1948 s64 pool;
1949 struct ceph_string *pool_ns;
1950 int ret, flags;
1951
1952 if (ci->i_vino.snap != CEPH_NOSNAP) {
1953 /*
1954 * Pool permission check needs to write to the first object.
1955 * But for snapshot, head of the first object may have alread
1956 * been deleted. Skip check to avoid creating orphan object.
1957 */
1958 return 0;
1959 }
1960
1961 if (ceph_test_mount_opt(ceph_inode_to_client(&ci->vfs_inode),
1962 NOPOOLPERM))
1963 return 0;
1964
1965 spin_lock(&ci->i_ceph_lock);
1966 flags = ci->i_ceph_flags;
1967 pool = ci->i_layout.pool_id;
1968 spin_unlock(&ci->i_ceph_lock);
1969 check:
1970 if (flags & CEPH_I_POOL_PERM) {
1971 if ((need & CEPH_CAP_FILE_RD) && !(flags & CEPH_I_POOL_RD)) {
1972 dout("ceph_pool_perm_check pool %lld no read perm\n",
1973 pool);
1974 return -EPERM;
1975 }
1976 if ((need & CEPH_CAP_FILE_WR) && !(flags & CEPH_I_POOL_WR)) {
1977 dout("ceph_pool_perm_check pool %lld no write perm\n",
1978 pool);
1979 return -EPERM;
1980 }
1981 return 0;
1982 }
1983
1984 pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
1985 ret = __ceph_pool_perm_get(ci, pool, pool_ns);
1986 ceph_put_string(pool_ns);
1987 if (ret < 0)
1988 return ret;
1989
1990 flags = CEPH_I_POOL_PERM;
1991 if (ret & POOL_READ)
1992 flags |= CEPH_I_POOL_RD;
1993 if (ret & POOL_WRITE)
1994 flags |= CEPH_I_POOL_WR;
1995
1996 spin_lock(&ci->i_ceph_lock);
1997 if (pool == ci->i_layout.pool_id &&
1998 pool_ns == rcu_dereference_raw(ci->i_layout.pool_ns)) {
1999 ci->i_ceph_flags |= flags;
2000 } else {
2001 pool = ci->i_layout.pool_id;
2002 flags = ci->i_ceph_flags;
2003 }
2004 spin_unlock(&ci->i_ceph_lock);
2005 goto check;
2006 }
2007
2008 void ceph_pool_perm_destroy(struct ceph_mds_client *mdsc)
2009 {
2010 struct ceph_pool_perm *perm;
2011 struct rb_node *n;
2012
2013 while (!RB_EMPTY_ROOT(&mdsc->pool_perm_tree)) {
2014 n = rb_first(&mdsc->pool_perm_tree);
2015 perm = rb_entry(n, struct ceph_pool_perm, node);
2016 rb_erase(n, &mdsc->pool_perm_tree);
2017 kfree(perm);
2018 }
2019 }
(deprecated) Btrfs devel tree