IPVS: fix netns if reading ip_vs_* procfs entries
[linux-2.6/linux-mips.git] / fs / ceph / addr.c
blobe159c529fd2bbc845bbd893f645ac7633d91b9eb
1 #include <linux/ceph/ceph_debug.h>
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>
12 #include "super.h"
13 #include "mds_client.h"
14 #include <linux/ceph/osd_client.h>
17 * Ceph address space ops.
19 * There are a few funny things going on here.
21 * The page->private field is used to reference a struct
22 * ceph_snap_context for _every_ dirty page. This indicates which
23 * snapshot the page was logically dirtied in, and thus which snap
24 * context needs to be associated with the osd write during writeback.
26 * Similarly, struct ceph_inode_info maintains a set of counters to
27 * count dirty pages on the inode. In the absence of snapshots,
28 * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count.
30 * When a snapshot is taken (that is, when the client receives
31 * notification that a snapshot was taken), each inode with caps and
32 * with dirty pages (dirty pages implies there is a cap) gets a new
33 * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending
34 * order, new snaps go to the tail). The i_wrbuffer_ref_head count is
35 * moved to capsnap->dirty. (Unless a sync write is currently in
36 * progress. In that case, the capsnap is said to be "pending", new
37 * writes cannot start, and the capsnap isn't "finalized" until the
38 * write completes (or fails) and a final size/mtime for the inode for
39 * that snap can be settled upon.) i_wrbuffer_ref_head is reset to 0.
41 * On writeback, we must submit writes to the osd IN SNAP ORDER. So,
42 * we look for the first capsnap in i_cap_snaps and write out pages in
43 * that snap context _only_. Then we move on to the next capsnap,
44 * eventually reaching the "live" or "head" context (i.e., pages that
45 * are not yet snapped) and are writing the most recently dirtied
46 * pages.
48 * Invalidate and so forth must take care to ensure the dirty page
49 * accounting is preserved.
52 #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10))
53 #define CONGESTION_OFF_THRESH(congestion_kb) \
54 (CONGESTION_ON_THRESH(congestion_kb) - \
55 (CONGESTION_ON_THRESH(congestion_kb) >> 2))
60 * Dirty a page. Optimistically adjust accounting, on the assumption
61 * that we won't race with invalidate. If we do, readjust.
63 static int ceph_set_page_dirty(struct page *page)
65 struct address_space *mapping = page->mapping;
66 struct inode *inode;
67 struct ceph_inode_info *ci;
68 int undo = 0;
69 struct ceph_snap_context *snapc;
71 if (unlikely(!mapping))
72 return !TestSetPageDirty(page);
74 if (TestSetPageDirty(page)) {
75 dout("%p set_page_dirty %p idx %lu -- already dirty\n",
76 mapping->host, page, page->index);
77 return 0;
80 inode = mapping->host;
81 ci = ceph_inode(inode);
84 * Note that we're grabbing a snapc ref here without holding
85 * any locks!
87 snapc = ceph_get_snap_context(ci->i_snap_realm->cached_context);
89 /* dirty the head */
90 spin_lock(&inode->i_lock);
91 if (ci->i_head_snapc == NULL)
92 ci->i_head_snapc = ceph_get_snap_context(snapc);
93 ++ci->i_wrbuffer_ref_head;
94 if (ci->i_wrbuffer_ref == 0)
95 ihold(inode);
96 ++ci->i_wrbuffer_ref;
97 dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d "
98 "snapc %p seq %lld (%d snaps)\n",
99 mapping->host, page, page->index,
100 ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1,
101 ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
102 snapc, snapc->seq, snapc->num_snaps);
103 spin_unlock(&inode->i_lock);
105 /* now adjust page */
106 spin_lock_irq(&mapping->tree_lock);
107 if (page->mapping) { /* Race with truncate? */
108 WARN_ON_ONCE(!PageUptodate(page));
109 account_page_dirtied(page, page->mapping);
110 radix_tree_tag_set(&mapping->page_tree,
111 page_index(page), PAGECACHE_TAG_DIRTY);
114 * Reference snap context in page->private. Also set
115 * PagePrivate so that we get invalidatepage callback.
117 page->private = (unsigned long)snapc;
118 SetPagePrivate(page);
119 } else {
120 dout("ANON set_page_dirty %p (raced truncate?)\n", page);
121 undo = 1;
124 spin_unlock_irq(&mapping->tree_lock);
126 if (undo)
127 /* whoops, we failed to dirty the page */
128 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
130 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
132 BUG_ON(!PageDirty(page));
133 return 1;
137 * If we are truncating the full page (i.e. offset == 0), adjust the
138 * dirty page counters appropriately. Only called if there is private
139 * data on the page.
141 static void ceph_invalidatepage(struct page *page, unsigned long offset)
143 struct inode *inode;
144 struct ceph_inode_info *ci;
145 struct ceph_snap_context *snapc = (void *)page->private;
147 BUG_ON(!PageLocked(page));
148 BUG_ON(!page->private);
149 BUG_ON(!PagePrivate(page));
150 BUG_ON(!page->mapping);
152 inode = page->mapping->host;
155 * We can get non-dirty pages here due to races between
156 * set_page_dirty and truncate_complete_page; just spit out a
157 * warning, in case we end up with accounting problems later.
159 if (!PageDirty(page))
160 pr_err("%p invalidatepage %p page not dirty\n", inode, page);
162 if (offset == 0)
163 ClearPageChecked(page);
165 ci = ceph_inode(inode);
166 if (offset == 0) {
167 dout("%p invalidatepage %p idx %lu full dirty page %lu\n",
168 inode, page, page->index, offset);
169 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
170 ceph_put_snap_context(snapc);
171 page->private = 0;
172 ClearPagePrivate(page);
173 } else {
174 dout("%p invalidatepage %p idx %lu partial dirty page\n",
175 inode, page, page->index);
179 /* just a sanity check */
180 static int ceph_releasepage(struct page *page, gfp_t g)
182 struct inode *inode = page->mapping ? page->mapping->host : NULL;
183 dout("%p releasepage %p idx %lu\n", inode, page, page->index);
184 WARN_ON(PageDirty(page));
185 WARN_ON(page->private);
186 WARN_ON(PagePrivate(page));
187 return 0;
191 * read a single page, without unlocking it.
193 static int readpage_nounlock(struct file *filp, struct page *page)
195 struct inode *inode = filp->f_dentry->d_inode;
196 struct ceph_inode_info *ci = ceph_inode(inode);
197 struct ceph_osd_client *osdc =
198 &ceph_inode_to_client(inode)->client->osdc;
199 int err = 0;
200 u64 len = PAGE_CACHE_SIZE;
202 dout("readpage inode %p file %p page %p index %lu\n",
203 inode, filp, page, page->index);
204 err = ceph_osdc_readpages(osdc, ceph_vino(inode), &ci->i_layout,
205 page->index << PAGE_CACHE_SHIFT, &len,
206 ci->i_truncate_seq, ci->i_truncate_size,
207 &page, 1, 0);
208 if (err == -ENOENT)
209 err = 0;
210 if (err < 0) {
211 SetPageError(page);
212 goto out;
213 } else if (err < PAGE_CACHE_SIZE) {
214 /* zero fill remainder of page */
215 zero_user_segment(page, err, PAGE_CACHE_SIZE);
217 SetPageUptodate(page);
219 out:
220 return err < 0 ? err : 0;
223 static int ceph_readpage(struct file *filp, struct page *page)
225 int r = readpage_nounlock(filp, page);
226 unlock_page(page);
227 return r;
231 * Build a vector of contiguous pages from the provided page list.
233 static struct page **page_vector_from_list(struct list_head *page_list,
234 unsigned *nr_pages)
236 struct page **pages;
237 struct page *page;
238 int next_index, contig_pages = 0;
240 /* build page vector */
241 pages = kmalloc(sizeof(*pages) * *nr_pages, GFP_NOFS);
242 if (!pages)
243 return ERR_PTR(-ENOMEM);
245 BUG_ON(list_empty(page_list));
246 next_index = list_entry(page_list->prev, struct page, lru)->index;
247 list_for_each_entry_reverse(page, page_list, lru) {
248 if (page->index == next_index) {
249 dout("readpages page %d %p\n", contig_pages, page);
250 pages[contig_pages] = page;
251 contig_pages++;
252 next_index++;
253 } else {
254 break;
257 *nr_pages = contig_pages;
258 return pages;
262 * Read multiple pages. Leave pages we don't read + unlock in page_list;
263 * the caller (VM) cleans them up.
265 static int ceph_readpages(struct file *file, struct address_space *mapping,
266 struct list_head *page_list, unsigned nr_pages)
268 struct inode *inode = file->f_dentry->d_inode;
269 struct ceph_inode_info *ci = ceph_inode(inode);
270 struct ceph_osd_client *osdc =
271 &ceph_inode_to_client(inode)->client->osdc;
272 int rc = 0;
273 struct page **pages;
274 loff_t offset;
275 u64 len;
277 dout("readpages %p file %p nr_pages %d\n",
278 inode, file, nr_pages);
280 pages = page_vector_from_list(page_list, &nr_pages);
281 if (IS_ERR(pages))
282 return PTR_ERR(pages);
284 /* guess read extent */
285 offset = pages[0]->index << PAGE_CACHE_SHIFT;
286 len = nr_pages << PAGE_CACHE_SHIFT;
287 rc = ceph_osdc_readpages(osdc, ceph_vino(inode), &ci->i_layout,
288 offset, &len,
289 ci->i_truncate_seq, ci->i_truncate_size,
290 pages, nr_pages, 0);
291 if (rc == -ENOENT)
292 rc = 0;
293 if (rc < 0)
294 goto out;
296 for (; !list_empty(page_list) && len > 0;
297 rc -= PAGE_CACHE_SIZE, len -= PAGE_CACHE_SIZE) {
298 struct page *page =
299 list_entry(page_list->prev, struct page, lru);
301 list_del(&page->lru);
303 if (rc < (int)PAGE_CACHE_SIZE) {
304 /* zero (remainder of) page */
305 int s = rc < 0 ? 0 : rc;
306 zero_user_segment(page, s, PAGE_CACHE_SIZE);
309 if (add_to_page_cache_lru(page, mapping, page->index,
310 GFP_NOFS)) {
311 page_cache_release(page);
312 dout("readpages %p add_to_page_cache failed %p\n",
313 inode, page);
314 continue;
316 dout("readpages %p adding %p idx %lu\n", inode, page,
317 page->index);
318 flush_dcache_page(page);
319 SetPageUptodate(page);
320 unlock_page(page);
321 page_cache_release(page);
323 rc = 0;
325 out:
326 kfree(pages);
327 return rc;
331 * Get ref for the oldest snapc for an inode with dirty data... that is, the
332 * only snap context we are allowed to write back.
334 static struct ceph_snap_context *get_oldest_context(struct inode *inode,
335 u64 *snap_size)
337 struct ceph_inode_info *ci = ceph_inode(inode);
338 struct ceph_snap_context *snapc = NULL;
339 struct ceph_cap_snap *capsnap = NULL;
341 spin_lock(&inode->i_lock);
342 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
343 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap,
344 capsnap->context, capsnap->dirty_pages);
345 if (capsnap->dirty_pages) {
346 snapc = ceph_get_snap_context(capsnap->context);
347 if (snap_size)
348 *snap_size = capsnap->size;
349 break;
352 if (!snapc && ci->i_wrbuffer_ref_head) {
353 snapc = ceph_get_snap_context(ci->i_head_snapc);
354 dout(" head snapc %p has %d dirty pages\n",
355 snapc, ci->i_wrbuffer_ref_head);
357 spin_unlock(&inode->i_lock);
358 return snapc;
362 * Write a single page, but leave the page locked.
364 * If we get a write error, set the page error bit, but still adjust the
365 * dirty page accounting (i.e., page is no longer dirty).
367 static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
369 struct inode *inode;
370 struct ceph_inode_info *ci;
371 struct ceph_fs_client *fsc;
372 struct ceph_osd_client *osdc;
373 loff_t page_off = page->index << PAGE_CACHE_SHIFT;
374 int len = PAGE_CACHE_SIZE;
375 loff_t i_size;
376 int err = 0;
377 struct ceph_snap_context *snapc, *oldest;
378 u64 snap_size = 0;
379 long writeback_stat;
381 dout("writepage %p idx %lu\n", page, page->index);
383 if (!page->mapping || !page->mapping->host) {
384 dout("writepage %p - no mapping\n", page);
385 return -EFAULT;
387 inode = page->mapping->host;
388 ci = ceph_inode(inode);
389 fsc = ceph_inode_to_client(inode);
390 osdc = &fsc->client->osdc;
392 /* verify this is a writeable snap context */
393 snapc = (void *)page->private;
394 if (snapc == NULL) {
395 dout("writepage %p page %p not dirty?\n", inode, page);
396 goto out;
398 oldest = get_oldest_context(inode, &snap_size);
399 if (snapc->seq > oldest->seq) {
400 dout("writepage %p page %p snapc %p not writeable - noop\n",
401 inode, page, (void *)page->private);
402 /* we should only noop if called by kswapd */
403 WARN_ON((current->flags & PF_MEMALLOC) == 0);
404 ceph_put_snap_context(oldest);
405 goto out;
407 ceph_put_snap_context(oldest);
409 /* is this a partial page at end of file? */
410 if (snap_size)
411 i_size = snap_size;
412 else
413 i_size = i_size_read(inode);
414 if (i_size < page_off + len)
415 len = i_size - page_off;
417 dout("writepage %p page %p index %lu on %llu~%u snapc %p\n",
418 inode, page, page->index, page_off, len, snapc);
420 writeback_stat = atomic_long_inc_return(&fsc->writeback_count);
421 if (writeback_stat >
422 CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb))
423 set_bdi_congested(&fsc->backing_dev_info, BLK_RW_ASYNC);
425 set_page_writeback(page);
426 err = ceph_osdc_writepages(osdc, ceph_vino(inode),
427 &ci->i_layout, snapc,
428 page_off, len,
429 ci->i_truncate_seq, ci->i_truncate_size,
430 &inode->i_mtime,
431 &page, 1, 0, 0, true);
432 if (err < 0) {
433 dout("writepage setting page/mapping error %d %p\n", err, page);
434 SetPageError(page);
435 mapping_set_error(&inode->i_data, err);
436 if (wbc)
437 wbc->pages_skipped++;
438 } else {
439 dout("writepage cleaned page %p\n", page);
440 err = 0; /* vfs expects us to return 0 */
442 page->private = 0;
443 ClearPagePrivate(page);
444 end_page_writeback(page);
445 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
446 ceph_put_snap_context(snapc); /* page's reference */
447 out:
448 return err;
451 static int ceph_writepage(struct page *page, struct writeback_control *wbc)
453 int err;
454 struct inode *inode = page->mapping->host;
455 BUG_ON(!inode);
456 igrab(inode);
457 err = writepage_nounlock(page, wbc);
458 unlock_page(page);
459 iput(inode);
460 return err;
465 * lame release_pages helper. release_pages() isn't exported to
466 * modules.
468 static void ceph_release_pages(struct page **pages, int num)
470 struct pagevec pvec;
471 int i;
473 pagevec_init(&pvec, 0);
474 for (i = 0; i < num; i++) {
475 if (pagevec_add(&pvec, pages[i]) == 0)
476 pagevec_release(&pvec);
478 pagevec_release(&pvec);
483 * async writeback completion handler.
485 * If we get an error, set the mapping error bit, but not the individual
486 * page error bits.
488 static void writepages_finish(struct ceph_osd_request *req,
489 struct ceph_msg *msg)
491 struct inode *inode = req->r_inode;
492 struct ceph_osd_reply_head *replyhead;
493 struct ceph_osd_op *op;
494 struct ceph_inode_info *ci = ceph_inode(inode);
495 unsigned wrote;
496 struct page *page;
497 int i;
498 struct ceph_snap_context *snapc = req->r_snapc;
499 struct address_space *mapping = inode->i_mapping;
500 __s32 rc = -EIO;
501 u64 bytes = 0;
502 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
503 long writeback_stat;
504 unsigned issued = ceph_caps_issued(ci);
506 /* parse reply */
507 replyhead = msg->front.iov_base;
508 WARN_ON(le32_to_cpu(replyhead->num_ops) == 0);
509 op = (void *)(replyhead + 1);
510 rc = le32_to_cpu(replyhead->result);
511 bytes = le64_to_cpu(op->extent.length);
513 if (rc >= 0) {
515 * Assume we wrote the pages we originally sent. The
516 * osd might reply with fewer pages if our writeback
517 * raced with a truncation and was adjusted at the osd,
518 * so don't believe the reply.
520 wrote = req->r_num_pages;
521 } else {
522 wrote = 0;
523 mapping_set_error(mapping, rc);
525 dout("writepages_finish %p rc %d bytes %llu wrote %d (pages)\n",
526 inode, rc, bytes, wrote);
528 /* clean all pages */
529 for (i = 0; i < req->r_num_pages; i++) {
530 page = req->r_pages[i];
531 BUG_ON(!page);
532 WARN_ON(!PageUptodate(page));
534 writeback_stat =
535 atomic_long_dec_return(&fsc->writeback_count);
536 if (writeback_stat <
537 CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb))
538 clear_bdi_congested(&fsc->backing_dev_info,
539 BLK_RW_ASYNC);
541 ceph_put_snap_context((void *)page->private);
542 page->private = 0;
543 ClearPagePrivate(page);
544 dout("unlocking %d %p\n", i, page);
545 end_page_writeback(page);
548 * We lost the cache cap, need to truncate the page before
549 * it is unlocked, otherwise we'd truncate it later in the
550 * page truncation thread, possibly losing some data that
551 * raced its way in
553 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0)
554 generic_error_remove_page(inode->i_mapping, page);
556 unlock_page(page);
558 dout("%p wrote+cleaned %d pages\n", inode, wrote);
559 ceph_put_wrbuffer_cap_refs(ci, req->r_num_pages, snapc);
561 ceph_release_pages(req->r_pages, req->r_num_pages);
562 if (req->r_pages_from_pool)
563 mempool_free(req->r_pages,
564 ceph_sb_to_client(inode->i_sb)->wb_pagevec_pool);
565 else
566 kfree(req->r_pages);
567 ceph_osdc_put_request(req);
571 * allocate a page vec, either directly, or if necessary, via a the
572 * mempool. we avoid the mempool if we can because req->r_num_pages
573 * may be less than the maximum write size.
575 static void alloc_page_vec(struct ceph_fs_client *fsc,
576 struct ceph_osd_request *req)
578 req->r_pages = kmalloc(sizeof(struct page *) * req->r_num_pages,
579 GFP_NOFS);
580 if (!req->r_pages) {
581 req->r_pages = mempool_alloc(fsc->wb_pagevec_pool, GFP_NOFS);
582 req->r_pages_from_pool = 1;
583 WARN_ON(!req->r_pages);
588 * initiate async writeback
590 static int ceph_writepages_start(struct address_space *mapping,
591 struct writeback_control *wbc)
593 struct inode *inode = mapping->host;
594 struct ceph_inode_info *ci = ceph_inode(inode);
595 struct ceph_fs_client *fsc;
596 pgoff_t index, start, end;
597 int range_whole = 0;
598 int should_loop = 1;
599 pgoff_t max_pages = 0, max_pages_ever = 0;
600 struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc;
601 struct pagevec pvec;
602 int done = 0;
603 int rc = 0;
604 unsigned wsize = 1 << inode->i_blkbits;
605 struct ceph_osd_request *req = NULL;
606 int do_sync;
607 u64 snap_size = 0;
610 * Include a 'sync' in the OSD request if this is a data
611 * integrity write (e.g., O_SYNC write or fsync()), or if our
612 * cap is being revoked.
614 do_sync = wbc->sync_mode == WB_SYNC_ALL;
615 if (ceph_caps_revoking(ci, CEPH_CAP_FILE_BUFFER))
616 do_sync = 1;
617 dout("writepages_start %p dosync=%d (mode=%s)\n",
618 inode, do_sync,
619 wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
620 (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
622 fsc = ceph_inode_to_client(inode);
623 if (fsc->mount_state == CEPH_MOUNT_SHUTDOWN) {
624 pr_warning("writepage_start %p on forced umount\n", inode);
625 return -EIO; /* we're in a forced umount, don't write! */
627 if (fsc->mount_options->wsize && fsc->mount_options->wsize < wsize)
628 wsize = fsc->mount_options->wsize;
629 if (wsize < PAGE_CACHE_SIZE)
630 wsize = PAGE_CACHE_SIZE;
631 max_pages_ever = wsize >> PAGE_CACHE_SHIFT;
633 pagevec_init(&pvec, 0);
635 /* where to start/end? */
636 if (wbc->range_cyclic) {
637 start = mapping->writeback_index; /* Start from prev offset */
638 end = -1;
639 dout(" cyclic, start at %lu\n", start);
640 } else {
641 start = wbc->range_start >> PAGE_CACHE_SHIFT;
642 end = wbc->range_end >> PAGE_CACHE_SHIFT;
643 if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
644 range_whole = 1;
645 should_loop = 0;
646 dout(" not cyclic, %lu to %lu\n", start, end);
648 index = start;
650 retry:
651 /* find oldest snap context with dirty data */
652 ceph_put_snap_context(snapc);
653 snapc = get_oldest_context(inode, &snap_size);
654 if (!snapc) {
655 /* hmm, why does writepages get called when there
656 is no dirty data? */
657 dout(" no snap context with dirty data?\n");
658 goto out;
660 dout(" oldest snapc is %p seq %lld (%d snaps)\n",
661 snapc, snapc->seq, snapc->num_snaps);
662 if (last_snapc && snapc != last_snapc) {
663 /* if we switched to a newer snapc, restart our scan at the
664 * start of the original file range. */
665 dout(" snapc differs from last pass, restarting at %lu\n",
666 index);
667 index = start;
669 last_snapc = snapc;
671 while (!done && index <= end) {
672 unsigned i;
673 int first;
674 pgoff_t next;
675 int pvec_pages, locked_pages;
676 struct page *page;
677 int want;
678 u64 offset, len;
679 struct ceph_osd_request_head *reqhead;
680 struct ceph_osd_op *op;
681 long writeback_stat;
683 next = 0;
684 locked_pages = 0;
685 max_pages = max_pages_ever;
687 get_more_pages:
688 first = -1;
689 want = min(end - index,
690 min((pgoff_t)PAGEVEC_SIZE,
691 max_pages - (pgoff_t)locked_pages) - 1)
692 + 1;
693 pvec_pages = pagevec_lookup_tag(&pvec, mapping, &index,
694 PAGECACHE_TAG_DIRTY,
695 want);
696 dout("pagevec_lookup_tag got %d\n", pvec_pages);
697 if (!pvec_pages && !locked_pages)
698 break;
699 for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) {
700 page = pvec.pages[i];
701 dout("? %p idx %lu\n", page, page->index);
702 if (locked_pages == 0)
703 lock_page(page); /* first page */
704 else if (!trylock_page(page))
705 break;
707 /* only dirty pages, or our accounting breaks */
708 if (unlikely(!PageDirty(page)) ||
709 unlikely(page->mapping != mapping)) {
710 dout("!dirty or !mapping %p\n", page);
711 unlock_page(page);
712 break;
714 if (!wbc->range_cyclic && page->index > end) {
715 dout("end of range %p\n", page);
716 done = 1;
717 unlock_page(page);
718 break;
720 if (next && (page->index != next)) {
721 dout("not consecutive %p\n", page);
722 unlock_page(page);
723 break;
725 if (wbc->sync_mode != WB_SYNC_NONE) {
726 dout("waiting on writeback %p\n", page);
727 wait_on_page_writeback(page);
729 if ((snap_size && page_offset(page) > snap_size) ||
730 (!snap_size &&
731 page_offset(page) > i_size_read(inode))) {
732 dout("%p page eof %llu\n", page, snap_size ?
733 snap_size : i_size_read(inode));
734 done = 1;
735 unlock_page(page);
736 break;
738 if (PageWriteback(page)) {
739 dout("%p under writeback\n", page);
740 unlock_page(page);
741 break;
744 /* only if matching snap context */
745 pgsnapc = (void *)page->private;
746 if (pgsnapc->seq > snapc->seq) {
747 dout("page snapc %p %lld > oldest %p %lld\n",
748 pgsnapc, pgsnapc->seq, snapc, snapc->seq);
749 unlock_page(page);
750 if (!locked_pages)
751 continue; /* keep looking for snap */
752 break;
755 if (!clear_page_dirty_for_io(page)) {
756 dout("%p !clear_page_dirty_for_io\n", page);
757 unlock_page(page);
758 break;
761 /* ok */
762 if (locked_pages == 0) {
763 /* prepare async write request */
764 offset = (unsigned long long)page->index
765 << PAGE_CACHE_SHIFT;
766 len = wsize;
767 req = ceph_osdc_new_request(&fsc->client->osdc,
768 &ci->i_layout,
769 ceph_vino(inode),
770 offset, &len,
771 CEPH_OSD_OP_WRITE,
772 CEPH_OSD_FLAG_WRITE |
773 CEPH_OSD_FLAG_ONDISK,
774 snapc, do_sync,
775 ci->i_truncate_seq,
776 ci->i_truncate_size,
777 &inode->i_mtime, true, 1, 0);
778 max_pages = req->r_num_pages;
780 alloc_page_vec(fsc, req);
781 req->r_callback = writepages_finish;
782 req->r_inode = inode;
785 /* note position of first page in pvec */
786 if (first < 0)
787 first = i;
788 dout("%p will write page %p idx %lu\n",
789 inode, page, page->index);
791 writeback_stat =
792 atomic_long_inc_return(&fsc->writeback_count);
793 if (writeback_stat > CONGESTION_ON_THRESH(
794 fsc->mount_options->congestion_kb)) {
795 set_bdi_congested(&fsc->backing_dev_info,
796 BLK_RW_ASYNC);
799 set_page_writeback(page);
800 req->r_pages[locked_pages] = page;
801 locked_pages++;
802 next = page->index + 1;
805 /* did we get anything? */
806 if (!locked_pages)
807 goto release_pvec_pages;
808 if (i) {
809 int j;
810 BUG_ON(!locked_pages || first < 0);
812 if (pvec_pages && i == pvec_pages &&
813 locked_pages < max_pages) {
814 dout("reached end pvec, trying for more\n");
815 pagevec_reinit(&pvec);
816 goto get_more_pages;
819 /* shift unused pages over in the pvec... we
820 * will need to release them below. */
821 for (j = i; j < pvec_pages; j++) {
822 dout(" pvec leftover page %p\n",
823 pvec.pages[j]);
824 pvec.pages[j-i+first] = pvec.pages[j];
826 pvec.nr -= i-first;
829 /* submit the write */
830 offset = req->r_pages[0]->index << PAGE_CACHE_SHIFT;
831 len = min((snap_size ? snap_size : i_size_read(inode)) - offset,
832 (u64)locked_pages << PAGE_CACHE_SHIFT);
833 dout("writepages got %d pages at %llu~%llu\n",
834 locked_pages, offset, len);
836 /* revise final length, page count */
837 req->r_num_pages = locked_pages;
838 reqhead = req->r_request->front.iov_base;
839 op = (void *)(reqhead + 1);
840 op->extent.length = cpu_to_le64(len);
841 op->payload_len = cpu_to_le32(len);
842 req->r_request->hdr.data_len = cpu_to_le32(len);
844 ceph_osdc_start_request(&fsc->client->osdc, req, true);
845 req = NULL;
847 /* continue? */
848 index = next;
849 wbc->nr_to_write -= locked_pages;
850 if (wbc->nr_to_write <= 0)
851 done = 1;
853 release_pvec_pages:
854 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr,
855 pvec.nr ? pvec.pages[0] : NULL);
856 pagevec_release(&pvec);
858 if (locked_pages && !done)
859 goto retry;
862 if (should_loop && !done) {
863 /* more to do; loop back to beginning of file */
864 dout("writepages looping back to beginning of file\n");
865 should_loop = 0;
866 index = 0;
867 goto retry;
870 if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
871 mapping->writeback_index = index;
873 out:
874 if (req)
875 ceph_osdc_put_request(req);
876 if (rc > 0)
877 rc = 0; /* vfs expects us to return 0 */
878 ceph_put_snap_context(snapc);
879 dout("writepages done, rc = %d\n", rc);
880 return rc;
886 * See if a given @snapc is either writeable, or already written.
888 static int context_is_writeable_or_written(struct inode *inode,
889 struct ceph_snap_context *snapc)
891 struct ceph_snap_context *oldest = get_oldest_context(inode, NULL);
892 int ret = !oldest || snapc->seq <= oldest->seq;
894 ceph_put_snap_context(oldest);
895 return ret;
899 * We are only allowed to write into/dirty the page if the page is
900 * clean, or already dirty within the same snap context.
902 * called with page locked.
903 * return success with page locked,
904 * or any failure (incl -EAGAIN) with page unlocked.
906 static int ceph_update_writeable_page(struct file *file,
907 loff_t pos, unsigned len,
908 struct page *page)
910 struct inode *inode = file->f_dentry->d_inode;
911 struct ceph_inode_info *ci = ceph_inode(inode);
912 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
913 loff_t page_off = pos & PAGE_CACHE_MASK;
914 int pos_in_page = pos & ~PAGE_CACHE_MASK;
915 int end_in_page = pos_in_page + len;
916 loff_t i_size;
917 int r;
918 struct ceph_snap_context *snapc, *oldest;
920 retry_locked:
921 /* writepages currently holds page lock, but if we change that later, */
922 wait_on_page_writeback(page);
924 /* check snap context */
925 BUG_ON(!ci->i_snap_realm);
926 down_read(&mdsc->snap_rwsem);
927 BUG_ON(!ci->i_snap_realm->cached_context);
928 snapc = (void *)page->private;
929 if (snapc && snapc != ci->i_head_snapc) {
931 * this page is already dirty in another (older) snap
932 * context! is it writeable now?
934 oldest = get_oldest_context(inode, NULL);
935 up_read(&mdsc->snap_rwsem);
937 if (snapc->seq > oldest->seq) {
938 ceph_put_snap_context(oldest);
939 dout(" page %p snapc %p not current or oldest\n",
940 page, snapc);
942 * queue for writeback, and wait for snapc to
943 * be writeable or written
945 snapc = ceph_get_snap_context(snapc);
946 unlock_page(page);
947 ceph_queue_writeback(inode);
948 r = wait_event_interruptible(ci->i_cap_wq,
949 context_is_writeable_or_written(inode, snapc));
950 ceph_put_snap_context(snapc);
951 if (r == -ERESTARTSYS)
952 return r;
953 return -EAGAIN;
955 ceph_put_snap_context(oldest);
957 /* yay, writeable, do it now (without dropping page lock) */
958 dout(" page %p snapc %p not current, but oldest\n",
959 page, snapc);
960 if (!clear_page_dirty_for_io(page))
961 goto retry_locked;
962 r = writepage_nounlock(page, NULL);
963 if (r < 0)
964 goto fail_nosnap;
965 goto retry_locked;
968 if (PageUptodate(page)) {
969 dout(" page %p already uptodate\n", page);
970 return 0;
973 /* full page? */
974 if (pos_in_page == 0 && len == PAGE_CACHE_SIZE)
975 return 0;
977 /* past end of file? */
978 i_size = inode->i_size; /* caller holds i_mutex */
980 if (i_size + len > inode->i_sb->s_maxbytes) {
981 /* file is too big */
982 r = -EINVAL;
983 goto fail;
986 if (page_off >= i_size ||
987 (pos_in_page == 0 && (pos+len) >= i_size &&
988 end_in_page - pos_in_page != PAGE_CACHE_SIZE)) {
989 dout(" zeroing %p 0 - %d and %d - %d\n",
990 page, pos_in_page, end_in_page, (int)PAGE_CACHE_SIZE);
991 zero_user_segments(page,
992 0, pos_in_page,
993 end_in_page, PAGE_CACHE_SIZE);
994 return 0;
997 /* we need to read it. */
998 up_read(&mdsc->snap_rwsem);
999 r = readpage_nounlock(file, page);
1000 if (r < 0)
1001 goto fail_nosnap;
1002 goto retry_locked;
1004 fail:
1005 up_read(&mdsc->snap_rwsem);
1006 fail_nosnap:
1007 unlock_page(page);
1008 return r;
1012 * We are only allowed to write into/dirty the page if the page is
1013 * clean, or already dirty within the same snap context.
1015 static int ceph_write_begin(struct file *file, struct address_space *mapping,
1016 loff_t pos, unsigned len, unsigned flags,
1017 struct page **pagep, void **fsdata)
1019 struct inode *inode = file->f_dentry->d_inode;
1020 struct page *page;
1021 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1022 int r;
1024 do {
1025 /* get a page */
1026 page = grab_cache_page_write_begin(mapping, index, 0);
1027 if (!page)
1028 return -ENOMEM;
1029 *pagep = page;
1031 dout("write_begin file %p inode %p page %p %d~%d\n", file,
1032 inode, page, (int)pos, (int)len);
1034 r = ceph_update_writeable_page(file, pos, len, page);
1035 } while (r == -EAGAIN);
1037 return r;
1041 * we don't do anything in here that simple_write_end doesn't do
1042 * except adjust dirty page accounting and drop read lock on
1043 * mdsc->snap_rwsem.
1045 static int ceph_write_end(struct file *file, struct address_space *mapping,
1046 loff_t pos, unsigned len, unsigned copied,
1047 struct page *page, void *fsdata)
1049 struct inode *inode = file->f_dentry->d_inode;
1050 struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1051 struct ceph_mds_client *mdsc = fsc->mdsc;
1052 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1053 int check_cap = 0;
1055 dout("write_end file %p inode %p page %p %d~%d (%d)\n", file,
1056 inode, page, (int)pos, (int)copied, (int)len);
1058 /* zero the stale part of the page if we did a short copy */
1059 if (copied < len)
1060 zero_user_segment(page, from+copied, len);
1062 /* did file size increase? */
1063 /* (no need for i_size_read(); we caller holds i_mutex */
1064 if (pos+copied > inode->i_size)
1065 check_cap = ceph_inode_set_size(inode, pos+copied);
1067 if (!PageUptodate(page))
1068 SetPageUptodate(page);
1070 set_page_dirty(page);
1072 unlock_page(page);
1073 up_read(&mdsc->snap_rwsem);
1074 page_cache_release(page);
1076 if (check_cap)
1077 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL);
1079 return copied;
1083 * we set .direct_IO to indicate direct io is supported, but since we
1084 * intercept O_DIRECT reads and writes early, this function should
1085 * never get called.
1087 static ssize_t ceph_direct_io(int rw, struct kiocb *iocb,
1088 const struct iovec *iov,
1089 loff_t pos, unsigned long nr_segs)
1091 WARN_ON(1);
1092 return -EINVAL;
1095 const struct address_space_operations ceph_aops = {
1096 .readpage = ceph_readpage,
1097 .readpages = ceph_readpages,
1098 .writepage = ceph_writepage,
1099 .writepages = ceph_writepages_start,
1100 .write_begin = ceph_write_begin,
1101 .write_end = ceph_write_end,
1102 .set_page_dirty = ceph_set_page_dirty,
1103 .invalidatepage = ceph_invalidatepage,
1104 .releasepage = ceph_releasepage,
1105 .direct_IO = ceph_direct_io,
1110 * vm ops
1114 * Reuse write_begin here for simplicity.
1116 static int ceph_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1118 struct inode *inode = vma->vm_file->f_dentry->d_inode;
1119 struct page *page = vmf->page;
1120 struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
1121 loff_t off = page->index << PAGE_CACHE_SHIFT;
1122 loff_t size, len;
1123 int ret;
1125 size = i_size_read(inode);
1126 if (off + PAGE_CACHE_SIZE <= size)
1127 len = PAGE_CACHE_SIZE;
1128 else
1129 len = size & ~PAGE_CACHE_MASK;
1131 dout("page_mkwrite %p %llu~%llu page %p idx %lu\n", inode,
1132 off, len, page, page->index);
1134 lock_page(page);
1136 ret = VM_FAULT_NOPAGE;
1137 if ((off > size) ||
1138 (page->mapping != inode->i_mapping))
1139 goto out;
1141 ret = ceph_update_writeable_page(vma->vm_file, off, len, page);
1142 if (ret == 0) {
1143 /* success. we'll keep the page locked. */
1144 set_page_dirty(page);
1145 up_read(&mdsc->snap_rwsem);
1146 ret = VM_FAULT_LOCKED;
1147 } else {
1148 if (ret == -ENOMEM)
1149 ret = VM_FAULT_OOM;
1150 else
1151 ret = VM_FAULT_SIGBUS;
1153 out:
1154 dout("page_mkwrite %p %llu~%llu = %d\n", inode, off, len, ret);
1155 if (ret != VM_FAULT_LOCKED)
1156 unlock_page(page);
1157 return ret;
1160 static struct vm_operations_struct ceph_vmops = {
1161 .fault = filemap_fault,
1162 .page_mkwrite = ceph_page_mkwrite,
1165 int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1167 struct address_space *mapping = file->f_mapping;
1169 if (!mapping->a_ops->readpage)
1170 return -ENOEXEC;
1171 file_accessed(file);
1172 vma->vm_ops = &ceph_vmops;
1173 vma->vm_flags |= VM_CAN_NONLINEAR;
1174 return 0;