~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/ceph/addr.c

Version: ~ [ linux-5.1-rc5 ] ~ [ linux-5.0.7 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.34 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.111 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.168 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.178 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.138 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.65 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ linux-2.6.39.4 ] ~ [ linux-2.6.38.8 ] ~ [ linux-2.6.37.6 ] ~ [ linux-2.6.36.4 ] ~ [ linux-2.6.35.14 ] ~ [ linux-2.6.34.15 ] ~ [ linux-2.6.33.20 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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 
 12 #include "super.h"
 13 #include "mds_client.h"
 14 #include <linux/ceph/osd_client.h>
 15 
 16 /*
 17  * Ceph address space ops.
 18  *
 19  * There are a few funny things going on here.
 20  *
 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.
 25  *
 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.
 29  *
 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.
 40  *
 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.
 47  *
 48  * Invalidate and so forth must take care to ensure the dirty page
 49  * accounting is preserved.
 50  */
 51 
 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))
 56 
 57 
 58 
 59 /*
 60  * Dirty a page.  Optimistically adjust accounting, on the assumption
 61  * that we won't race with invalidate.  If we do, readjust.
 62  */
 63 static int ceph_set_page_dirty(struct page *page)
 64 {
 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;
 70 
 71         if (unlikely(!mapping))
 72                 return !TestSetPageDirty(page);
 73 
 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;
 78         }
 79 
 80         inode = mapping->host;
 81         ci = ceph_inode(inode);
 82 
 83         /*
 84          * Note that we're grabbing a snapc ref here without holding
 85          * any locks!
 86          */
 87         snapc = ceph_get_snap_context(ci->i_snap_realm->cached_context);
 88 
 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);
104 
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);
112 
113                 /*
114                  * Reference snap context in page->private.  Also set
115                  * PagePrivate so that we get invalidatepage callback.
116                  */
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;
122         }
123 
124         spin_unlock_irq(&mapping->tree_lock);
125 
126         if (undo)
127                 /* whoops, we failed to dirty the page */
128                 ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
129 
130         __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
131 
132         BUG_ON(!PageDirty(page));
133         return 1;
134 }
135 
136 /*
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.
140  */
141 static void ceph_invalidatepage(struct page *page, unsigned long offset)
142 {
143         struct inode *inode;
144         struct ceph_inode_info *ci;
145         struct ceph_snap_context *snapc = (void *)page->private;
146 
147         BUG_ON(!PageLocked(page));
148         BUG_ON(!page->private);
149         BUG_ON(!PagePrivate(page));
150         BUG_ON(!page->mapping);
151 
152         inode = page->mapping->host;
153 
154         /*
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.
158          */
159         if (!PageDirty(page))
160                 pr_err("%p invalidatepage %p page not dirty\n", inode, page);
161 
162         if (offset == 0)
163                 ClearPageChecked(page);
164 
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);
176         }
177 }
178 
179 /* just a sanity check */
180 static int ceph_releasepage(struct page *page, gfp_t g)
181 {
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;
188 }
189 
190 /*
191  * read a single page, without unlocking it.
192  */
193 static int readpage_nounlock(struct file *filp, struct page *page)
194 {
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;
201 
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);
216         }
217         SetPageUptodate(page);
218 
219 out:
220         return err < 0 ? err : 0;
221 }
222 
223 static int ceph_readpage(struct file *filp, struct page *page)
224 {
225         int r = readpage_nounlock(filp, page);
226         unlock_page(page);
227         return r;
228 }
229 
230 /*
231  * Build a vector of contiguous pages from the provided page list.
232  */
233 static struct page **page_vector_from_list(struct list_head *page_list,
234                                            unsigned *nr_pages)
235 {
236         struct page **pages;
237         struct page *page;
238         int next_index, contig_pages = 0;
239 
240         /* build page vector */
241         pages = kmalloc(sizeof(*pages) * *nr_pages, GFP_NOFS);
242         if (!pages)
243                 return ERR_PTR(-ENOMEM);
244 
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;
255                 }
256         }
257         *nr_pages = contig_pages;
258         return pages;
259 }
260 
261 /*
262  * Read multiple pages.  Leave pages we don't read + unlock in page_list;
263  * the caller (VM) cleans them up.
264  */
265 static int ceph_readpages(struct file *file, struct address_space *mapping,
266                           struct list_head *page_list, unsigned nr_pages)
267 {
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;
276 
277         dout("readpages %p file %p nr_pages %d\n",
278              inode, file, nr_pages);
279 
280         pages = page_vector_from_list(page_list, &nr_pages);
281         if (IS_ERR(pages))
282                 return PTR_ERR(pages);
283 
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;
295 
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);
300 
301                 list_del(&page->lru);
302 
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);
307                 }
308 
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;
315                 }
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);
322         }
323         rc = 0;
324 
325 out:
326         kfree(pages);
327         return rc;
328 }
329 
330 /*
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.
333  */
334 static struct ceph_snap_context *get_oldest_context(struct inode *inode,
335                                                     u64 *snap_size)
336 {
337         struct ceph_inode_info *ci = ceph_inode(inode);
338         struct ceph_snap_context *snapc = NULL;
339         struct ceph_cap_snap *capsnap = NULL;
340 
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;
350                 }
351         }
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);
356         }
357         spin_unlock(&inode->i_lock);
358         return snapc;
359 }
360 
361 /*
362  * Write a single page, but leave the page locked.
363  *
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).
366  */
367 static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
368 {
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;
380 
381         dout("writepage %p idx %lu\n", page, page->index);
382 
383         if (!page->mapping || !page->mapping->host) {
384                 dout("writepage %p - no mapping\n", page);
385                 return -EFAULT;
386         }
387         inode = page->mapping->host;
388         ci = ceph_inode(inode);
389         fsc = ceph_inode_to_client(inode);
390         osdc = &fsc->client->osdc;
391 
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;
397         }
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;
406         }
407         ceph_put_snap_context(oldest);
408 
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;
416 
417         dout("writepage %p page %p index %lu on %llu~%u snapc %p\n",
418              inode, page, page->index, page_off, len, snapc);
419 
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);
424 
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 */
441         }
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;
449 }
450 
451 static int ceph_writepage(struct page *page, struct writeback_control *wbc)
452 {
453         int err;
454         struct inode *inode = page->mapping->host;
455         BUG_ON(!inode);
456         ihold(inode);
457         err = writepage_nounlock(page, wbc);
458         unlock_page(page);
459         iput(inode);
460         return err;
461 }
462 
463 
464 /*
465  * lame release_pages helper.  release_pages() isn't exported to
466  * modules.
467  */
468 static void ceph_release_pages(struct page **pages, int num)
469 {
470         struct pagevec pvec;
471         int i;
472 
473         pagevec_init(&pvec, 0);
474         for (i = 0; i < num; i++) {
475                 if (pagevec_add(&pvec, pages[i]) == 0)
476                         pagevec_release(&pvec);
477         }
478         pagevec_release(&pvec);
479 }
480 
481 
482 /*
483  * async writeback completion handler.
484  *
485  * If we get an error, set the mapping error bit, but not the individual
486  * page error bits.
487  */
488 static void writepages_finish(struct ceph_osd_request *req,
489                               struct ceph_msg *msg)
490 {
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);
505 
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);
512 
513         if (rc >= 0) {
514                 /*
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.
519                  */
520                 wrote = req->r_num_pages;
521         } else {
522                 wrote = 0;
523                 mapping_set_error(mapping, rc);
524         }
525         dout("writepages_finish %p rc %d bytes %llu wrote %d (pages)\n",
526              inode, rc, bytes, wrote);
527 
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));
533 
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);
540 
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);
546 
547                 /*
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
552                  */
553                 if ((issued & (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) == 0)
554                         generic_error_remove_page(inode->i_mapping, page);
555 
556                 unlock_page(page);
557         }
558         dout("%p wrote+cleaned %d pages\n", inode, wrote);
559         ceph_put_wrbuffer_cap_refs(ci, req->r_num_pages, snapc);
560 
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);
568 }
569 
570 /*
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.
574  */
575 static void alloc_page_vec(struct ceph_fs_client *fsc,
576                            struct ceph_osd_request *req)
577 {
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);
584         }
585 }
586 
587 /*
588  * initiate async writeback
589  */
590 static int ceph_writepages_start(struct address_space *mapping,
591                                  struct writeback_control *wbc)
592 {
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;
608 
609         /*
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.
613          */
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"));
621 
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! */
626         }
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;
632 
633         pagevec_init(&pvec, 0);
634 
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);
647         }
648         index = start;
649 
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;
659         }
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;
668         }
669         last_snapc = snapc;
670 
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;
682 
683                 next = 0;
684                 locked_pages = 0;
685                 max_pages = max_pages_ever;
686 
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;
706 
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;
713                         }
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;
719                         }
720                         if (next && (page->index != next)) {
721                                 dout("not consecutive %p\n", page);
722                                 unlock_page(page);
723                                 break;
724                         }
725                         if (wbc->sync_mode != WB_SYNC_NONE) {
726                                 dout("waiting on writeback %p\n", page);
727                                 wait_on_page_writeback(page);
728                         }
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;
737                         }
738                         if (PageWriteback(page)) {
739                                 dout("%p under writeback\n", page);
740                                 unlock_page(page);
741                                 break;
742                         }
743 
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;
753                         }
754 
755                         if (!clear_page_dirty_for_io(page)) {
756                                 dout("%p !clear_page_dirty_for_io\n", page);
757                                 unlock_page(page);
758                                 break;
759                         }
760 
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 
779                                 if (!req) {
780                                         rc = -ENOMEM;
781                                         unlock_page(page);
782                                         break;
783                                 }
784 
785                                 max_pages = req->r_num_pages;
786 
787                                 alloc_page_vec(fsc, req);
788                                 req->r_callback = writepages_finish;
789                                 req->r_inode = inode;
790                         }
791 
792                         /* note position of first page in pvec */
793                         if (first < 0)
794                                 first = i;
795                         dout("%p will write page %p idx %lu\n",
796                              inode, page, page->index);
797 
798                         writeback_stat =
799                                atomic_long_inc_return(&fsc->writeback_count);
800                         if (writeback_stat > CONGESTION_ON_THRESH(
801                                     fsc->mount_options->congestion_kb)) {
802                                 set_bdi_congested(&fsc->backing_dev_info,
803                                                   BLK_RW_ASYNC);
804                         }
805 
806                         set_page_writeback(page);
807                         req->r_pages[locked_pages] = page;
808                         locked_pages++;
809                         next = page->index + 1;
810                 }
811 
812                 /* did we get anything? */
813                 if (!locked_pages)
814                         goto release_pvec_pages;
815                 if (i) {
816                         int j;
817                         BUG_ON(!locked_pages || first < 0);
818 
819                         if (pvec_pages && i == pvec_pages &&
820                             locked_pages < max_pages) {
821                                 dout("reached end pvec, trying for more\n");
822                                 pagevec_reinit(&pvec);
823                                 goto get_more_pages;
824                         }
825 
826                         /* shift unused pages over in the pvec...  we
827                          * will need to release them below. */
828                         for (j = i; j < pvec_pages; j++) {
829                                 dout(" pvec leftover page %p\n",
830                                      pvec.pages[j]);
831                                 pvec.pages[j-i+first] = pvec.pages[j];
832                         }
833                         pvec.nr -= i-first;
834                 }
835 
836                 /* submit the write */
837                 offset = req->r_pages[0]->index << PAGE_CACHE_SHIFT;
838                 len = min((snap_size ? snap_size : i_size_read(inode)) - offset,
839                           (u64)locked_pages << PAGE_CACHE_SHIFT);
840                 dout("writepages got %d pages at %llu~%llu\n",
841                      locked_pages, offset, len);
842 
843                 /* revise final length, page count */
844                 req->r_num_pages = locked_pages;
845                 reqhead = req->r_request->front.iov_base;
846                 op = (void *)(reqhead + 1);
847                 op->extent.length = cpu_to_le64(len);
848                 op->payload_len = cpu_to_le32(len);
849                 req->r_request->hdr.data_len = cpu_to_le32(len);
850 
851                 rc = ceph_osdc_start_request(&fsc->client->osdc, req, true);
852                 BUG_ON(rc);
853                 req = NULL;
854 
855                 /* continue? */
856                 index = next;
857                 wbc->nr_to_write -= locked_pages;
858                 if (wbc->nr_to_write <= 0)
859                         done = 1;
860 
861 release_pvec_pages:
862                 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr,
863                      pvec.nr ? pvec.pages[0] : NULL);
864                 pagevec_release(&pvec);
865 
866                 if (locked_pages && !done)
867                         goto retry;
868         }
869 
870         if (should_loop && !done) {
871                 /* more to do; loop back to beginning of file */
872                 dout("writepages looping back to beginning of file\n");
873                 should_loop = 0;
874                 index = 0;
875                 goto retry;
876         }
877 
878         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
879                 mapping->writeback_index = index;
880 
881 out:
882         if (req)
883                 ceph_osdc_put_request(req);
884         ceph_put_snap_context(snapc);
885         dout("writepages done, rc = %d\n", rc);
886         return rc;
887 }
888 
889 
890 
891 /*
892  * See if a given @snapc is either writeable, or already written.
893  */
894 static int context_is_writeable_or_written(struct inode *inode,
895                                            struct ceph_snap_context *snapc)
896 {
897         struct ceph_snap_context *oldest = get_oldest_context(inode, NULL);
898         int ret = !oldest || snapc->seq <= oldest->seq;
899 
900         ceph_put_snap_context(oldest);
901         return ret;
902 }
903 
904 /*
905  * We are only allowed to write into/dirty the page if the page is
906  * clean, or already dirty within the same snap context.
907  *
908  * called with page locked.
909  * return success with page locked,
910  * or any failure (incl -EAGAIN) with page unlocked.
911  */
912 static int ceph_update_writeable_page(struct file *file,
913                             loff_t pos, unsigned len,
914                             struct page *page)
915 {
916         struct inode *inode = file->f_dentry->d_inode;
917         struct ceph_inode_info *ci = ceph_inode(inode);
918         struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
919         loff_t page_off = pos & PAGE_CACHE_MASK;
920         int pos_in_page = pos & ~PAGE_CACHE_MASK;
921         int end_in_page = pos_in_page + len;
922         loff_t i_size;
923         int r;
924         struct ceph_snap_context *snapc, *oldest;
925 
926 retry_locked:
927         /* writepages currently holds page lock, but if we change that later, */
928         wait_on_page_writeback(page);
929 
930         /* check snap context */
931         BUG_ON(!ci->i_snap_realm);
932         down_read(&mdsc->snap_rwsem);
933         BUG_ON(!ci->i_snap_realm->cached_context);
934         snapc = (void *)page->private;
935         if (snapc && snapc != ci->i_head_snapc) {
936                 /*
937                  * this page is already dirty in another (older) snap
938                  * context!  is it writeable now?
939                  */
940                 oldest = get_oldest_context(inode, NULL);
941                 up_read(&mdsc->snap_rwsem);
942 
943                 if (snapc->seq > oldest->seq) {
944                         ceph_put_snap_context(oldest);
945                         dout(" page %p snapc %p not current or oldest\n",
946                              page, snapc);
947                         /*
948                          * queue for writeback, and wait for snapc to
949                          * be writeable or written
950                          */
951                         snapc = ceph_get_snap_context(snapc);
952                         unlock_page(page);
953                         ceph_queue_writeback(inode);
954                         r = wait_event_interruptible(ci->i_cap_wq,
955                                context_is_writeable_or_written(inode, snapc));
956                         ceph_put_snap_context(snapc);
957                         if (r == -ERESTARTSYS)
958                                 return r;
959                         return -EAGAIN;
960                 }
961                 ceph_put_snap_context(oldest);
962 
963                 /* yay, writeable, do it now (without dropping page lock) */
964                 dout(" page %p snapc %p not current, but oldest\n",
965                      page, snapc);
966                 if (!clear_page_dirty_for_io(page))
967                         goto retry_locked;
968                 r = writepage_nounlock(page, NULL);
969                 if (r < 0)
970                         goto fail_nosnap;
971                 goto retry_locked;
972         }
973 
974         if (PageUptodate(page)) {
975                 dout(" page %p already uptodate\n", page);
976                 return 0;
977         }
978 
979         /* full page? */
980         if (pos_in_page == 0 && len == PAGE_CACHE_SIZE)
981                 return 0;
982 
983         /* past end of file? */
984         i_size = inode->i_size;   /* caller holds i_mutex */
985 
986         if (i_size + len > inode->i_sb->s_maxbytes) {
987                 /* file is too big */
988                 r = -EINVAL;
989                 goto fail;
990         }
991 
992         if (page_off >= i_size ||
993             (pos_in_page == 0 && (pos+len) >= i_size &&
994              end_in_page - pos_in_page != PAGE_CACHE_SIZE)) {
995                 dout(" zeroing %p 0 - %d and %d - %d\n",
996                      page, pos_in_page, end_in_page, (int)PAGE_CACHE_SIZE);
997                 zero_user_segments(page,
998                                    0, pos_in_page,
999                                    end_in_page, PAGE_CACHE_SIZE);
1000                 return 0;
1001         }
1002 
1003         /* we need to read it. */
1004         up_read(&mdsc->snap_rwsem);
1005         r = readpage_nounlock(file, page);
1006         if (r < 0)
1007                 goto fail_nosnap;
1008         goto retry_locked;
1009 
1010 fail:
1011         up_read(&mdsc->snap_rwsem);
1012 fail_nosnap:
1013         unlock_page(page);
1014         return r;
1015 }
1016 
1017 /*
1018  * We are only allowed to write into/dirty the page if the page is
1019  * clean, or already dirty within the same snap context.
1020  */
1021 static int ceph_write_begin(struct file *file, struct address_space *mapping,
1022                             loff_t pos, unsigned len, unsigned flags,
1023                             struct page **pagep, void **fsdata)
1024 {
1025         struct inode *inode = file->f_dentry->d_inode;
1026         struct page *page;
1027         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1028         int r;
1029 
1030         do {
1031                 /* get a page */
1032                 page = grab_cache_page_write_begin(mapping, index, 0);
1033                 if (!page)
1034                         return -ENOMEM;
1035                 *pagep = page;
1036 
1037                 dout("write_begin file %p inode %p page %p %d~%d\n", file,
1038                      inode, page, (int)pos, (int)len);
1039 
1040                 r = ceph_update_writeable_page(file, pos, len, page);
1041         } while (r == -EAGAIN);
1042 
1043         return r;
1044 }
1045 
1046 /*
1047  * we don't do anything in here that simple_write_end doesn't do
1048  * except adjust dirty page accounting and drop read lock on
1049  * mdsc->snap_rwsem.
1050  */
1051 static int ceph_write_end(struct file *file, struct address_space *mapping,
1052                           loff_t pos, unsigned len, unsigned copied,
1053                           struct page *page, void *fsdata)
1054 {
1055         struct inode *inode = file->f_dentry->d_inode;
1056         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1057         struct ceph_mds_client *mdsc = fsc->mdsc;
1058         unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1059         int check_cap = 0;
1060 
1061         dout("write_end file %p inode %p page %p %d~%d (%d)\n", file,
1062              inode, page, (int)pos, (int)copied, (int)len);
1063 
1064         /* zero the stale part of the page if we did a short copy */
1065         if (copied < len)
1066                 zero_user_segment(page, from+copied, len);
1067 
1068         /* did file size increase? */
1069         /* (no need for i_size_read(); we caller holds i_mutex */
1070         if (pos+copied > inode->i_size)
1071                 check_cap = ceph_inode_set_size(inode, pos+copied);
1072 
1073         if (!PageUptodate(page))
1074                 SetPageUptodate(page);
1075 
1076         set_page_dirty(page);
1077 
1078         unlock_page(page);
1079         up_read(&mdsc->snap_rwsem);
1080         page_cache_release(page);
1081 
1082         if (check_cap)
1083                 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL);
1084 
1085         return copied;
1086 }
1087 
1088 /*
1089  * we set .direct_IO to indicate direct io is supported, but since we
1090  * intercept O_DIRECT reads and writes early, this function should
1091  * never get called.
1092  */
1093 static ssize_t ceph_direct_io(int rw, struct kiocb *iocb,
1094                               const struct iovec *iov,
1095                               loff_t pos, unsigned long nr_segs)
1096 {
1097         WARN_ON(1);
1098         return -EINVAL;
1099 }
1100 
1101 const struct address_space_operations ceph_aops = {
1102         .readpage = ceph_readpage,
1103         .readpages = ceph_readpages,
1104         .writepage = ceph_writepage,
1105         .writepages = ceph_writepages_start,
1106         .write_begin = ceph_write_begin,
1107         .write_end = ceph_write_end,
1108         .set_page_dirty = ceph_set_page_dirty,
1109         .invalidatepage = ceph_invalidatepage,
1110         .releasepage = ceph_releasepage,
1111         .direct_IO = ceph_direct_io,
1112 };
1113 
1114 
1115 /*
1116  * vm ops
1117  */
1118 
1119 /*
1120  * Reuse write_begin here for simplicity.
1121  */
1122 static int ceph_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
1123 {
1124         struct inode *inode = vma->vm_file->f_dentry->d_inode;
1125         struct page *page = vmf->page;
1126         struct ceph_mds_client *mdsc = ceph_inode_to_client(inode)->mdsc;
1127         loff_t off = page->index << PAGE_CACHE_SHIFT;
1128         loff_t size, len;
1129         int ret;
1130 
1131         size = i_size_read(inode);
1132         if (off + PAGE_CACHE_SIZE <= size)
1133                 len = PAGE_CACHE_SIZE;
1134         else
1135                 len = size & ~PAGE_CACHE_MASK;
1136 
1137         dout("page_mkwrite %p %llu~%llu page %p idx %lu\n", inode,
1138              off, len, page, page->index);
1139 
1140         lock_page(page);
1141 
1142         ret = VM_FAULT_NOPAGE;
1143         if ((off > size) ||
1144             (page->mapping != inode->i_mapping))
1145                 goto out;
1146 
1147         ret = ceph_update_writeable_page(vma->vm_file, off, len, page);
1148         if (ret == 0) {
1149                 /* success.  we'll keep the page locked. */
1150                 set_page_dirty(page);
1151                 up_read(&mdsc->snap_rwsem);
1152                 ret = VM_FAULT_LOCKED;
1153         } else {
1154                 if (ret == -ENOMEM)
1155                         ret = VM_FAULT_OOM;
1156                 else
1157                         ret = VM_FAULT_SIGBUS;
1158         }
1159 out:
1160         dout("page_mkwrite %p %llu~%llu = %d\n", inode, off, len, ret);
1161         if (ret != VM_FAULT_LOCKED)
1162                 unlock_page(page);
1163         return ret;
1164 }
1165 
1166 static struct vm_operations_struct ceph_vmops = {
1167         .fault          = filemap_fault,
1168         .page_mkwrite   = ceph_page_mkwrite,
1169 };
1170 
1171 int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1172 {
1173         struct address_space *mapping = file->f_mapping;
1174 
1175         if (!mapping->a_ops->readpage)
1176                 return -ENOEXEC;
1177         file_accessed(file);
1178         vma->vm_ops = &ceph_vmops;
1179         vma->vm_flags |= VM_CAN_NONLINEAR;
1180         return 0;
1181 }
1182 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp