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

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

Version: ~ [ linux-5.10-rc5 ] ~ [ linux-5.9.10 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.79 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.159 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.208 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.245 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.245 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ 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-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 // SPDX-License-Identifier: GPL-2.0
  2 #include <linux/ceph/ceph_debug.h>
  3 
  4 #include <linux/backing-dev.h>
  5 #include <linux/fs.h>
  6 #include <linux/mm.h>
  7 #include <linux/pagemap.h>
  8 #include <linux/writeback.h>    /* generic_writepages */
  9 #include <linux/slab.h>
 10 #include <linux/pagevec.h>
 11 #include <linux/task_io_accounting_ops.h>
 12 #include <linux/signal.h>
 13 
 14 #include "super.h"
 15 #include "mds_client.h"
 16 #include "cache.h"
 17 #include <linux/ceph/osd_client.h>
 18 
 19 /*
 20  * Ceph address space ops.
 21  *
 22  * There are a few funny things going on here.
 23  *
 24  * The page->private field is used to reference a struct
 25  * ceph_snap_context for _every_ dirty page.  This indicates which
 26  * snapshot the page was logically dirtied in, and thus which snap
 27  * context needs to be associated with the osd write during writeback.
 28  *
 29  * Similarly, struct ceph_inode_info maintains a set of counters to
 30  * count dirty pages on the inode.  In the absence of snapshots,
 31  * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count.
 32  *
 33  * When a snapshot is taken (that is, when the client receives
 34  * notification that a snapshot was taken), each inode with caps and
 35  * with dirty pages (dirty pages implies there is a cap) gets a new
 36  * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending
 37  * order, new snaps go to the tail).  The i_wrbuffer_ref_head count is
 38  * moved to capsnap->dirty. (Unless a sync write is currently in
 39  * progress.  In that case, the capsnap is said to be "pending", new
 40  * writes cannot start, and the capsnap isn't "finalized" until the
 41  * write completes (or fails) and a final size/mtime for the inode for
 42  * that snap can be settled upon.)  i_wrbuffer_ref_head is reset to 0.
 43  *
 44  * On writeback, we must submit writes to the osd IN SNAP ORDER.  So,
 45  * we look for the first capsnap in i_cap_snaps and write out pages in
 46  * that snap context _only_.  Then we move on to the next capsnap,
 47  * eventually reaching the "live" or "head" context (i.e., pages that
 48  * are not yet snapped) and are writing the most recently dirtied
 49  * pages.
 50  *
 51  * Invalidate and so forth must take care to ensure the dirty page
 52  * accounting is preserved.
 53  */
 54 
 55 #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10))
 56 #define CONGESTION_OFF_THRESH(congestion_kb)                            \
 57         (CONGESTION_ON_THRESH(congestion_kb) -                          \
 58          (CONGESTION_ON_THRESH(congestion_kb) >> 2))
 59 
 60 static inline struct ceph_snap_context *page_snap_context(struct page *page)
 61 {
 62         if (PagePrivate(page))
 63                 return (void *)page->private;
 64         return NULL;
 65 }
 66 
 67 /*
 68  * Dirty a page.  Optimistically adjust accounting, on the assumption
 69  * that we won't race with invalidate.  If we do, readjust.
 70  */
 71 static int ceph_set_page_dirty(struct page *page)
 72 {
 73         struct address_space *mapping = page->mapping;
 74         struct inode *inode;
 75         struct ceph_inode_info *ci;
 76         struct ceph_snap_context *snapc;
 77         int ret;
 78 
 79         if (unlikely(!mapping))
 80                 return !TestSetPageDirty(page);
 81 
 82         if (PageDirty(page)) {
 83                 dout("%p set_page_dirty %p idx %lu -- already dirty\n",
 84                      mapping->host, page, page->index);
 85                 BUG_ON(!PagePrivate(page));
 86                 return 0;
 87         }
 88 
 89         inode = mapping->host;
 90         ci = ceph_inode(inode);
 91 
 92         /* dirty the head */
 93         spin_lock(&ci->i_ceph_lock);
 94         BUG_ON(ci->i_wr_ref == 0); // caller should hold Fw reference
 95         if (__ceph_have_pending_cap_snap(ci)) {
 96                 struct ceph_cap_snap *capsnap =
 97                                 list_last_entry(&ci->i_cap_snaps,
 98                                                 struct ceph_cap_snap,
 99                                                 ci_item);
100                 snapc = ceph_get_snap_context(capsnap->context);
101                 capsnap->dirty_pages++;
102         } else {
103                 BUG_ON(!ci->i_head_snapc);
104                 snapc = ceph_get_snap_context(ci->i_head_snapc);
105                 ++ci->i_wrbuffer_ref_head;
106         }
107         if (ci->i_wrbuffer_ref == 0)
108                 ihold(inode);
109         ++ci->i_wrbuffer_ref;
110         dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d "
111              "snapc %p seq %lld (%d snaps)\n",
112              mapping->host, page, page->index,
113              ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1,
114              ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
115              snapc, snapc->seq, snapc->num_snaps);
116         spin_unlock(&ci->i_ceph_lock);
117 
118         /*
119          * Reference snap context in page->private.  Also set
120          * PagePrivate so that we get invalidatepage callback.
121          */
122         BUG_ON(PagePrivate(page));
123         page->private = (unsigned long)snapc;
124         SetPagePrivate(page);
125 
126         ret = __set_page_dirty_nobuffers(page);
127         WARN_ON(!PageLocked(page));
128         WARN_ON(!page->mapping);
129 
130         return ret;
131 }
132 
133 /*
134  * If we are truncating the full page (i.e. offset == 0), adjust the
135  * dirty page counters appropriately.  Only called if there is private
136  * data on the page.
137  */
138 static void ceph_invalidatepage(struct page *page, unsigned int offset,
139                                 unsigned int length)
140 {
141         struct inode *inode;
142         struct ceph_inode_info *ci;
143         struct ceph_snap_context *snapc = page_snap_context(page);
144 
145         inode = page->mapping->host;
146         ci = ceph_inode(inode);
147 
148         if (offset != 0 || length != PAGE_SIZE) {
149                 dout("%p invalidatepage %p idx %lu partial dirty page %u~%u\n",
150                      inode, page, page->index, offset, length);
151                 return;
152         }
153 
154         ceph_invalidate_fscache_page(inode, page);
155 
156         WARN_ON(!PageLocked(page));
157         if (!PagePrivate(page))
158                 return;
159 
160         ClearPageChecked(page);
161 
162         dout("%p invalidatepage %p idx %lu full dirty page\n",
163              inode, page, page->index);
164 
165         ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
166         ceph_put_snap_context(snapc);
167         page->private = 0;
168         ClearPagePrivate(page);
169 }
170 
171 static int ceph_releasepage(struct page *page, gfp_t g)
172 {
173         dout("%p releasepage %p idx %lu (%sdirty)\n", page->mapping->host,
174              page, page->index, PageDirty(page) ? "" : "not ");
175 
176         /* Can we release the page from the cache? */
177         if (!ceph_release_fscache_page(page, g))
178                 return 0;
179 
180         return !PagePrivate(page);
181 }
182 
183 /*
184  * read a single page, without unlocking it.
185  */
186 static int ceph_do_readpage(struct file *filp, struct page *page)
187 {
188         struct inode *inode = file_inode(filp);
189         struct ceph_inode_info *ci = ceph_inode(inode);
190         struct ceph_osd_client *osdc =
191                 &ceph_inode_to_client(inode)->client->osdc;
192         int err = 0;
193         u64 off = page_offset(page);
194         u64 len = PAGE_SIZE;
195 
196         if (off >= i_size_read(inode)) {
197                 zero_user_segment(page, 0, PAGE_SIZE);
198                 SetPageUptodate(page);
199                 return 0;
200         }
201 
202         if (ci->i_inline_version != CEPH_INLINE_NONE) {
203                 /*
204                  * Uptodate inline data should have been added
205                  * into page cache while getting Fcr caps.
206                  */
207                 if (off == 0)
208                         return -EINVAL;
209                 zero_user_segment(page, 0, PAGE_SIZE);
210                 SetPageUptodate(page);
211                 return 0;
212         }
213 
214         err = ceph_readpage_from_fscache(inode, page);
215         if (err == 0)
216                 return -EINPROGRESS;
217 
218         dout("readpage inode %p file %p page %p index %lu\n",
219              inode, filp, page, page->index);
220         err = ceph_osdc_readpages(osdc, ceph_vino(inode), &ci->i_layout,
221                                   off, &len,
222                                   ci->i_truncate_seq, ci->i_truncate_size,
223                                   &page, 1, 0);
224         if (err == -ENOENT)
225                 err = 0;
226         if (err < 0) {
227                 SetPageError(page);
228                 ceph_fscache_readpage_cancel(inode, page);
229                 goto out;
230         }
231         if (err < PAGE_SIZE)
232                 /* zero fill remainder of page */
233                 zero_user_segment(page, err, PAGE_SIZE);
234         else
235                 flush_dcache_page(page);
236 
237         SetPageUptodate(page);
238         ceph_readpage_to_fscache(inode, page);
239 
240 out:
241         return err < 0 ? err : 0;
242 }
243 
244 static int ceph_readpage(struct file *filp, struct page *page)
245 {
246         int r = ceph_do_readpage(filp, page);
247         if (r != -EINPROGRESS)
248                 unlock_page(page);
249         else
250                 r = 0;
251         return r;
252 }
253 
254 /*
255  * Finish an async read(ahead) op.
256  */
257 static void finish_read(struct ceph_osd_request *req)
258 {
259         struct inode *inode = req->r_inode;
260         struct ceph_osd_data *osd_data;
261         int rc = req->r_result <= 0 ? req->r_result : 0;
262         int bytes = req->r_result >= 0 ? req->r_result : 0;
263         int num_pages;
264         int i;
265 
266         dout("finish_read %p req %p rc %d bytes %d\n", inode, req, rc, bytes);
267 
268         /* unlock all pages, zeroing any data we didn't read */
269         osd_data = osd_req_op_extent_osd_data(req, 0);
270         BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
271         num_pages = calc_pages_for((u64)osd_data->alignment,
272                                         (u64)osd_data->length);
273         for (i = 0; i < num_pages; i++) {
274                 struct page *page = osd_data->pages[i];
275 
276                 if (rc < 0 && rc != -ENOENT) {
277                         ceph_fscache_readpage_cancel(inode, page);
278                         goto unlock;
279                 }
280                 if (bytes < (int)PAGE_SIZE) {
281                         /* zero (remainder of) page */
282                         int s = bytes < 0 ? 0 : bytes;
283                         zero_user_segment(page, s, PAGE_SIZE);
284                 }
285                 dout("finish_read %p uptodate %p idx %lu\n", inode, page,
286                      page->index);
287                 flush_dcache_page(page);
288                 SetPageUptodate(page);
289                 ceph_readpage_to_fscache(inode, page);
290 unlock:
291                 unlock_page(page);
292                 put_page(page);
293                 bytes -= PAGE_SIZE;
294         }
295         kfree(osd_data->pages);
296 }
297 
298 /*
299  * start an async read(ahead) operation.  return nr_pages we submitted
300  * a read for on success, or negative error code.
301  */
302 static int start_read(struct inode *inode, struct list_head *page_list, int max)
303 {
304         struct ceph_osd_client *osdc =
305                 &ceph_inode_to_client(inode)->client->osdc;
306         struct ceph_inode_info *ci = ceph_inode(inode);
307         struct page *page = list_entry(page_list->prev, struct page, lru);
308         struct ceph_vino vino;
309         struct ceph_osd_request *req;
310         u64 off;
311         u64 len;
312         int i;
313         struct page **pages;
314         pgoff_t next_index;
315         int nr_pages = 0;
316         int got = 0;
317         int ret = 0;
318 
319         if (!current->journal_info) {
320                 /* caller of readpages does not hold buffer and read caps
321                  * (fadvise, madvise and readahead cases) */
322                 int want = CEPH_CAP_FILE_CACHE;
323                 ret = ceph_try_get_caps(ci, CEPH_CAP_FILE_RD, want, &got);
324                 if (ret < 0) {
325                         dout("start_read %p, error getting cap\n", inode);
326                 } else if (!(got & want)) {
327                         dout("start_read %p, no cache cap\n", inode);
328                         ret = 0;
329                 }
330                 if (ret <= 0) {
331                         if (got)
332                                 ceph_put_cap_refs(ci, got);
333                         while (!list_empty(page_list)) {
334                                 page = list_entry(page_list->prev,
335                                                   struct page, lru);
336                                 list_del(&page->lru);
337                                 put_page(page);
338                         }
339                         return ret;
340                 }
341         }
342 
343         off = (u64) page_offset(page);
344 
345         /* count pages */
346         next_index = page->index;
347         list_for_each_entry_reverse(page, page_list, lru) {
348                 if (page->index != next_index)
349                         break;
350                 nr_pages++;
351                 next_index++;
352                 if (max && nr_pages == max)
353                         break;
354         }
355         len = nr_pages << PAGE_SHIFT;
356         dout("start_read %p nr_pages %d is %lld~%lld\n", inode, nr_pages,
357              off, len);
358         vino = ceph_vino(inode);
359         req = ceph_osdc_new_request(osdc, &ci->i_layout, vino, off, &len,
360                                     0, 1, CEPH_OSD_OP_READ,
361                                     CEPH_OSD_FLAG_READ, NULL,
362                                     ci->i_truncate_seq, ci->i_truncate_size,
363                                     false);
364         if (IS_ERR(req)) {
365                 ret = PTR_ERR(req);
366                 goto out;
367         }
368 
369         /* build page vector */
370         nr_pages = calc_pages_for(0, len);
371         pages = kmalloc(sizeof(*pages) * nr_pages, GFP_KERNEL);
372         if (!pages) {
373                 ret = -ENOMEM;
374                 goto out_put;
375         }
376         for (i = 0; i < nr_pages; ++i) {
377                 page = list_entry(page_list->prev, struct page, lru);
378                 BUG_ON(PageLocked(page));
379                 list_del(&page->lru);
380 
381                 dout("start_read %p adding %p idx %lu\n", inode, page,
382                      page->index);
383                 if (add_to_page_cache_lru(page, &inode->i_data, page->index,
384                                           GFP_KERNEL)) {
385                         ceph_fscache_uncache_page(inode, page);
386                         put_page(page);
387                         dout("start_read %p add_to_page_cache failed %p\n",
388                              inode, page);
389                         nr_pages = i;
390                         if (nr_pages > 0) {
391                                 len = nr_pages << PAGE_SHIFT;
392                                 osd_req_op_extent_update(req, 0, len);
393                                 break;
394                         }
395                         goto out_pages;
396                 }
397                 pages[i] = page;
398         }
399         osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false);
400         req->r_callback = finish_read;
401         req->r_inode = inode;
402 
403         dout("start_read %p starting %p %lld~%lld\n", inode, req, off, len);
404         ret = ceph_osdc_start_request(osdc, req, false);
405         if (ret < 0)
406                 goto out_pages;
407         ceph_osdc_put_request(req);
408 
409         /* After adding locked pages to page cache, the inode holds cache cap.
410          * So we can drop our cap refs. */
411         if (got)
412                 ceph_put_cap_refs(ci, got);
413 
414         return nr_pages;
415 
416 out_pages:
417         for (i = 0; i < nr_pages; ++i) {
418                 ceph_fscache_readpage_cancel(inode, pages[i]);
419                 unlock_page(pages[i]);
420         }
421         ceph_put_page_vector(pages, nr_pages, false);
422 out_put:
423         ceph_osdc_put_request(req);
424 out:
425         if (got)
426                 ceph_put_cap_refs(ci, got);
427         return ret;
428 }
429 
430 
431 /*
432  * Read multiple pages.  Leave pages we don't read + unlock in page_list;
433  * the caller (VM) cleans them up.
434  */
435 static int ceph_readpages(struct file *file, struct address_space *mapping,
436                           struct list_head *page_list, unsigned nr_pages)
437 {
438         struct inode *inode = file_inode(file);
439         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
440         int rc = 0;
441         int max = 0;
442 
443         if (ceph_inode(inode)->i_inline_version != CEPH_INLINE_NONE)
444                 return -EINVAL;
445 
446         rc = ceph_readpages_from_fscache(mapping->host, mapping, page_list,
447                                          &nr_pages);
448 
449         if (rc == 0)
450                 goto out;
451 
452         max = fsc->mount_options->rsize >> PAGE_SHIFT;
453         dout("readpages %p file %p nr_pages %d max %d\n",
454              inode, file, nr_pages, max);
455         while (!list_empty(page_list)) {
456                 rc = start_read(inode, page_list, max);
457                 if (rc < 0)
458                         goto out;
459         }
460 out:
461         ceph_fscache_readpages_cancel(inode, page_list);
462 
463         dout("readpages %p file %p ret %d\n", inode, file, rc);
464         return rc;
465 }
466 
467 struct ceph_writeback_ctl
468 {
469         loff_t i_size;
470         u64 truncate_size;
471         u32 truncate_seq;
472         bool size_stable;
473         bool head_snapc;
474 };
475 
476 /*
477  * Get ref for the oldest snapc for an inode with dirty data... that is, the
478  * only snap context we are allowed to write back.
479  */
480 static struct ceph_snap_context *
481 get_oldest_context(struct inode *inode, struct ceph_writeback_ctl *ctl,
482                    struct ceph_snap_context *page_snapc)
483 {
484         struct ceph_inode_info *ci = ceph_inode(inode);
485         struct ceph_snap_context *snapc = NULL;
486         struct ceph_cap_snap *capsnap = NULL;
487 
488         spin_lock(&ci->i_ceph_lock);
489         list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
490                 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap,
491                      capsnap->context, capsnap->dirty_pages);
492                 if (!capsnap->dirty_pages)
493                         continue;
494 
495                 /* get i_size, truncate_{seq,size} for page_snapc? */
496                 if (snapc && capsnap->context != page_snapc)
497                         continue;
498 
499                 if (ctl) {
500                         if (capsnap->writing) {
501                                 ctl->i_size = i_size_read(inode);
502                                 ctl->size_stable = false;
503                         } else {
504                                 ctl->i_size = capsnap->size;
505                                 ctl->size_stable = true;
506                         }
507                         ctl->truncate_size = capsnap->truncate_size;
508                         ctl->truncate_seq = capsnap->truncate_seq;
509                         ctl->head_snapc = false;
510                 }
511 
512                 if (snapc)
513                         break;
514 
515                 snapc = ceph_get_snap_context(capsnap->context);
516                 if (!page_snapc ||
517                     page_snapc == snapc ||
518                     page_snapc->seq > snapc->seq)
519                         break;
520         }
521         if (!snapc && ci->i_wrbuffer_ref_head) {
522                 snapc = ceph_get_snap_context(ci->i_head_snapc);
523                 dout(" head snapc %p has %d dirty pages\n",
524                      snapc, ci->i_wrbuffer_ref_head);
525                 if (ctl) {
526                         ctl->i_size = i_size_read(inode);
527                         ctl->truncate_size = ci->i_truncate_size;
528                         ctl->truncate_seq = ci->i_truncate_seq;
529                         ctl->size_stable = false;
530                         ctl->head_snapc = true;
531                 }
532         }
533         spin_unlock(&ci->i_ceph_lock);
534         return snapc;
535 }
536 
537 static u64 get_writepages_data_length(struct inode *inode,
538                                       struct page *page, u64 start)
539 {
540         struct ceph_inode_info *ci = ceph_inode(inode);
541         struct ceph_snap_context *snapc = page_snap_context(page);
542         struct ceph_cap_snap *capsnap = NULL;
543         u64 end = i_size_read(inode);
544 
545         if (snapc != ci->i_head_snapc) {
546                 bool found = false;
547                 spin_lock(&ci->i_ceph_lock);
548                 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
549                         if (capsnap->context == snapc) {
550                                 if (!capsnap->writing)
551                                         end = capsnap->size;
552                                 found = true;
553                                 break;
554                         }
555                 }
556                 spin_unlock(&ci->i_ceph_lock);
557                 WARN_ON(!found);
558         }
559         if (end > page_offset(page) + PAGE_SIZE)
560                 end = page_offset(page) + PAGE_SIZE;
561         return end > start ? end - start : 0;
562 }
563 
564 /*
565  * Write a single page, but leave the page locked.
566  *
567  * If we get a write error, set the page error bit, but still adjust the
568  * dirty page accounting (i.e., page is no longer dirty).
569  */
570 static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
571 {
572         struct inode *inode;
573         struct ceph_inode_info *ci;
574         struct ceph_fs_client *fsc;
575         struct ceph_snap_context *snapc, *oldest;
576         loff_t page_off = page_offset(page);
577         long writeback_stat;
578         int err, len = PAGE_SIZE;
579         struct ceph_writeback_ctl ceph_wbc;
580 
581         dout("writepage %p idx %lu\n", page, page->index);
582 
583         inode = page->mapping->host;
584         ci = ceph_inode(inode);
585         fsc = ceph_inode_to_client(inode);
586 
587         /* verify this is a writeable snap context */
588         snapc = page_snap_context(page);
589         if (!snapc) {
590                 dout("writepage %p page %p not dirty?\n", inode, page);
591                 return 0;
592         }
593         oldest = get_oldest_context(inode, &ceph_wbc, snapc);
594         if (snapc->seq > oldest->seq) {
595                 dout("writepage %p page %p snapc %p not writeable - noop\n",
596                      inode, page, snapc);
597                 /* we should only noop if called by kswapd */
598                 WARN_ON(!(current->flags & PF_MEMALLOC));
599                 ceph_put_snap_context(oldest);
600                 redirty_page_for_writepage(wbc, page);
601                 return 0;
602         }
603         ceph_put_snap_context(oldest);
604 
605         /* is this a partial page at end of file? */
606         if (page_off >= ceph_wbc.i_size) {
607                 dout("%p page eof %llu\n", page, ceph_wbc.i_size);
608                 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
609                 return 0;
610         }
611 
612         if (ceph_wbc.i_size < page_off + len)
613                 len = ceph_wbc.i_size - page_off;
614 
615         dout("writepage %p page %p index %lu on %llu~%u snapc %p seq %lld\n",
616              inode, page, page->index, page_off, len, snapc, snapc->seq);
617 
618         writeback_stat = atomic_long_inc_return(&fsc->writeback_count);
619         if (writeback_stat >
620             CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb))
621                 set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC);
622 
623         set_page_writeback(page);
624         err = ceph_osdc_writepages(&fsc->client->osdc, ceph_vino(inode),
625                                    &ci->i_layout, snapc, page_off, len,
626                                    ceph_wbc.truncate_seq,
627                                    ceph_wbc.truncate_size,
628                                    &inode->i_mtime, &page, 1);
629         if (err < 0) {
630                 struct writeback_control tmp_wbc;
631                 if (!wbc)
632                         wbc = &tmp_wbc;
633                 if (err == -ERESTARTSYS) {
634                         /* killed by SIGKILL */
635                         dout("writepage interrupted page %p\n", page);
636                         redirty_page_for_writepage(wbc, page);
637                         end_page_writeback(page);
638                         return err;
639                 }
640                 dout("writepage setting page/mapping error %d %p\n",
641                      err, page);
642                 SetPageError(page);
643                 mapping_set_error(&inode->i_data, err);
644                 wbc->pages_skipped++;
645         } else {
646                 dout("writepage cleaned page %p\n", page);
647                 err = 0;  /* vfs expects us to return 0 */
648         }
649         page->private = 0;
650         ClearPagePrivate(page);
651         end_page_writeback(page);
652         ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
653         ceph_put_snap_context(snapc);  /* page's reference */
654         return err;
655 }
656 
657 static int ceph_writepage(struct page *page, struct writeback_control *wbc)
658 {
659         int err;
660         struct inode *inode = page->mapping->host;
661         BUG_ON(!inode);
662         ihold(inode);
663         err = writepage_nounlock(page, wbc);
664         if (err == -ERESTARTSYS) {
665                 /* direct memory reclaimer was killed by SIGKILL. return 0
666                  * to prevent caller from setting mapping/page error */
667                 err = 0;
668         }
669         unlock_page(page);
670         iput(inode);
671         return err;
672 }
673 
674 /*
675  * lame release_pages helper.  release_pages() isn't exported to
676  * modules.
677  */
678 static void ceph_release_pages(struct page **pages, int num)
679 {
680         struct pagevec pvec;
681         int i;
682 
683         pagevec_init(&pvec);
684         for (i = 0; i < num; i++) {
685                 if (pagevec_add(&pvec, pages[i]) == 0)
686                         pagevec_release(&pvec);
687         }
688         pagevec_release(&pvec);
689 }
690 
691 /*
692  * async writeback completion handler.
693  *
694  * If we get an error, set the mapping error bit, but not the individual
695  * page error bits.
696  */
697 static void writepages_finish(struct ceph_osd_request *req)
698 {
699         struct inode *inode = req->r_inode;
700         struct ceph_inode_info *ci = ceph_inode(inode);
701         struct ceph_osd_data *osd_data;
702         struct page *page;
703         int num_pages, total_pages = 0;
704         int i, j;
705         int rc = req->r_result;
706         struct ceph_snap_context *snapc = req->r_snapc;
707         struct address_space *mapping = inode->i_mapping;
708         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
709         bool remove_page;
710 
711         dout("writepages_finish %p rc %d\n", inode, rc);
712         if (rc < 0) {
713                 mapping_set_error(mapping, rc);
714                 ceph_set_error_write(ci);
715         } else {
716                 ceph_clear_error_write(ci);
717         }
718 
719         /*
720          * We lost the cache cap, need to truncate the page before
721          * it is unlocked, otherwise we'd truncate it later in the
722          * page truncation thread, possibly losing some data that
723          * raced its way in
724          */
725         remove_page = !(ceph_caps_issued(ci) &
726                         (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO));
727 
728         /* clean all pages */
729         for (i = 0; i < req->r_num_ops; i++) {
730                 if (req->r_ops[i].op != CEPH_OSD_OP_WRITE)
731                         break;
732 
733                 osd_data = osd_req_op_extent_osd_data(req, i);
734                 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
735                 num_pages = calc_pages_for((u64)osd_data->alignment,
736                                            (u64)osd_data->length);
737                 total_pages += num_pages;
738                 for (j = 0; j < num_pages; j++) {
739                         page = osd_data->pages[j];
740                         BUG_ON(!page);
741                         WARN_ON(!PageUptodate(page));
742 
743                         if (atomic_long_dec_return(&fsc->writeback_count) <
744                              CONGESTION_OFF_THRESH(
745                                         fsc->mount_options->congestion_kb))
746                                 clear_bdi_congested(inode_to_bdi(inode),
747                                                     BLK_RW_ASYNC);
748 
749                         ceph_put_snap_context(page_snap_context(page));
750                         page->private = 0;
751                         ClearPagePrivate(page);
752                         dout("unlocking %p\n", page);
753                         end_page_writeback(page);
754 
755                         if (remove_page)
756                                 generic_error_remove_page(inode->i_mapping,
757                                                           page);
758 
759                         unlock_page(page);
760                 }
761                 dout("writepages_finish %p wrote %llu bytes cleaned %d pages\n",
762                      inode, osd_data->length, rc >= 0 ? num_pages : 0);
763 
764                 ceph_release_pages(osd_data->pages, num_pages);
765         }
766 
767         ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc);
768 
769         osd_data = osd_req_op_extent_osd_data(req, 0);
770         if (osd_data->pages_from_pool)
771                 mempool_free(osd_data->pages,
772                              ceph_sb_to_client(inode->i_sb)->wb_pagevec_pool);
773         else
774                 kfree(osd_data->pages);
775         ceph_osdc_put_request(req);
776 }
777 
778 /*
779  * initiate async writeback
780  */
781 static int ceph_writepages_start(struct address_space *mapping,
782                                  struct writeback_control *wbc)
783 {
784         struct inode *inode = mapping->host;
785         struct ceph_inode_info *ci = ceph_inode(inode);
786         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
787         struct ceph_vino vino = ceph_vino(inode);
788         pgoff_t index, start_index, end = -1;
789         struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc;
790         struct pagevec pvec;
791         int rc = 0;
792         unsigned int wsize = i_blocksize(inode);
793         struct ceph_osd_request *req = NULL;
794         struct ceph_writeback_ctl ceph_wbc;
795         bool should_loop, range_whole = false;
796         bool stop, done = false;
797 
798         dout("writepages_start %p (mode=%s)\n", inode,
799              wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
800              (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
801 
802         if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
803                 if (ci->i_wrbuffer_ref > 0) {
804                         pr_warn_ratelimited(
805                                 "writepage_start %p %lld forced umount\n",
806                                 inode, ceph_ino(inode));
807                 }
808                 mapping_set_error(mapping, -EIO);
809                 return -EIO; /* we're in a forced umount, don't write! */
810         }
811         if (fsc->mount_options->wsize < wsize)
812                 wsize = fsc->mount_options->wsize;
813 
814         pagevec_init(&pvec);
815 
816         start_index = wbc->range_cyclic ? mapping->writeback_index : 0;
817         index = start_index;
818 
819 retry:
820         /* find oldest snap context with dirty data */
821         snapc = get_oldest_context(inode, &ceph_wbc, NULL);
822         if (!snapc) {
823                 /* hmm, why does writepages get called when there
824                    is no dirty data? */
825                 dout(" no snap context with dirty data?\n");
826                 goto out;
827         }
828         dout(" oldest snapc is %p seq %lld (%d snaps)\n",
829              snapc, snapc->seq, snapc->num_snaps);
830 
831         should_loop = false;
832         if (ceph_wbc.head_snapc && snapc != last_snapc) {
833                 /* where to start/end? */
834                 if (wbc->range_cyclic) {
835                         index = start_index;
836                         end = -1;
837                         if (index > 0)
838                                 should_loop = true;
839                         dout(" cyclic, start at %lu\n", index);
840                 } else {
841                         index = wbc->range_start >> PAGE_SHIFT;
842                         end = wbc->range_end >> PAGE_SHIFT;
843                         if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
844                                 range_whole = true;
845                         dout(" not cyclic, %lu to %lu\n", index, end);
846                 }
847         } else if (!ceph_wbc.head_snapc) {
848                 /* Do not respect wbc->range_{start,end}. Dirty pages
849                  * in that range can be associated with newer snapc.
850                  * They are not writeable until we write all dirty pages
851                  * associated with 'snapc' get written */
852                 if (index > 0 || wbc->sync_mode != WB_SYNC_NONE)
853                         should_loop = true;
854                 dout(" non-head snapc, range whole\n");
855         }
856 
857         ceph_put_snap_context(last_snapc);
858         last_snapc = snapc;
859 
860         stop = false;
861         while (!stop && index <= end) {
862                 int num_ops = 0, op_idx;
863                 unsigned i, pvec_pages, max_pages, locked_pages = 0;
864                 struct page **pages = NULL, **data_pages;
865                 mempool_t *pool = NULL; /* Becomes non-null if mempool used */
866                 struct page *page;
867                 pgoff_t strip_unit_end = 0;
868                 u64 offset = 0, len = 0;
869 
870                 max_pages = wsize >> PAGE_SHIFT;
871 
872 get_more_pages:
873                 pvec_pages = pagevec_lookup_range_nr_tag(&pvec, mapping, &index,
874                                                 end, PAGECACHE_TAG_DIRTY,
875                                                 max_pages - locked_pages);
876                 dout("pagevec_lookup_range_tag got %d\n", pvec_pages);
877                 if (!pvec_pages && !locked_pages)
878                         break;
879                 for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) {
880                         page = pvec.pages[i];
881                         dout("? %p idx %lu\n", page, page->index);
882                         if (locked_pages == 0)
883                                 lock_page(page);  /* first page */
884                         else if (!trylock_page(page))
885                                 break;
886 
887                         /* only dirty pages, or our accounting breaks */
888                         if (unlikely(!PageDirty(page)) ||
889                             unlikely(page->mapping != mapping)) {
890                                 dout("!dirty or !mapping %p\n", page);
891                                 unlock_page(page);
892                                 continue;
893                         }
894                         if (strip_unit_end && (page->index > strip_unit_end)) {
895                                 dout("end of strip unit %p\n", page);
896                                 unlock_page(page);
897                                 break;
898                         }
899                         if (page_offset(page) >= ceph_wbc.i_size) {
900                                 dout("%p page eof %llu\n",
901                                      page, ceph_wbc.i_size);
902                                 /* not done if range_cyclic */
903                                 stop = true;
904                                 unlock_page(page);
905                                 break;
906                         }
907                         if (PageWriteback(page)) {
908                                 if (wbc->sync_mode == WB_SYNC_NONE) {
909                                         dout("%p under writeback\n", page);
910                                         unlock_page(page);
911                                         continue;
912                                 }
913                                 dout("waiting on writeback %p\n", page);
914                                 wait_on_page_writeback(page);
915                         }
916 
917                         /* only if matching snap context */
918                         pgsnapc = page_snap_context(page);
919                         if (pgsnapc != snapc) {
920                                 dout("page snapc %p %lld != oldest %p %lld\n",
921                                      pgsnapc, pgsnapc->seq, snapc, snapc->seq);
922                                 unlock_page(page);
923                                 continue;
924                         }
925 
926                         if (!clear_page_dirty_for_io(page)) {
927                                 dout("%p !clear_page_dirty_for_io\n", page);
928                                 unlock_page(page);
929                                 continue;
930                         }
931 
932                         /*
933                          * We have something to write.  If this is
934                          * the first locked page this time through,
935                          * calculate max possinle write size and
936                          * allocate a page array
937                          */
938                         if (locked_pages == 0) {
939                                 u64 objnum;
940                                 u64 objoff;
941 
942                                 /* prepare async write request */
943                                 offset = (u64)page_offset(page);
944                                 len = wsize;
945 
946                                 rc = ceph_calc_file_object_mapping(&ci->i_layout,
947                                                                 offset, len,
948                                                                 &objnum, &objoff,
949                                                                 &len);
950                                 if (rc < 0) {
951                                         unlock_page(page);
952                                         break;
953                                 }
954 
955                                 num_ops = 1;
956                                 strip_unit_end = page->index +
957                                         ((len - 1) >> PAGE_SHIFT);
958 
959                                 BUG_ON(pages);
960                                 max_pages = calc_pages_for(0, (u64)len);
961                                 pages = kmalloc(max_pages * sizeof (*pages),
962                                                 GFP_NOFS);
963                                 if (!pages) {
964                                         pool = fsc->wb_pagevec_pool;
965                                         pages = mempool_alloc(pool, GFP_NOFS);
966                                         BUG_ON(!pages);
967                                 }
968 
969                                 len = 0;
970                         } else if (page->index !=
971                                    (offset + len) >> PAGE_SHIFT) {
972                                 if (num_ops >= (pool ?  CEPH_OSD_SLAB_OPS :
973                                                         CEPH_OSD_MAX_OPS)) {
974                                         redirty_page_for_writepage(wbc, page);
975                                         unlock_page(page);
976                                         break;
977                                 }
978 
979                                 num_ops++;
980                                 offset = (u64)page_offset(page);
981                                 len = 0;
982                         }
983 
984                         /* note position of first page in pvec */
985                         dout("%p will write page %p idx %lu\n",
986                              inode, page, page->index);
987 
988                         if (atomic_long_inc_return(&fsc->writeback_count) >
989                             CONGESTION_ON_THRESH(
990                                     fsc->mount_options->congestion_kb)) {
991                                 set_bdi_congested(inode_to_bdi(inode),
992                                                   BLK_RW_ASYNC);
993                         }
994 
995 
996                         pages[locked_pages++] = page;
997                         pvec.pages[i] = NULL;
998 
999                         len += PAGE_SIZE;
1000                 }
1001 
1002                 /* did we get anything? */
1003                 if (!locked_pages)
1004                         goto release_pvec_pages;
1005                 if (i) {
1006                         unsigned j, n = 0;
1007                         /* shift unused page to beginning of pvec */
1008                         for (j = 0; j < pvec_pages; j++) {
1009                                 if (!pvec.pages[j])
1010                                         continue;
1011                                 if (n < j)
1012                                         pvec.pages[n] = pvec.pages[j];
1013                                 n++;
1014                         }
1015                         pvec.nr = n;
1016 
1017                         if (pvec_pages && i == pvec_pages &&
1018                             locked_pages < max_pages) {
1019                                 dout("reached end pvec, trying for more\n");
1020                                 pagevec_release(&pvec);
1021                                 goto get_more_pages;
1022                         }
1023                 }
1024 
1025 new_request:
1026                 offset = page_offset(pages[0]);
1027                 len = wsize;
1028 
1029                 req = ceph_osdc_new_request(&fsc->client->osdc,
1030                                         &ci->i_layout, vino,
1031                                         offset, &len, 0, num_ops,
1032                                         CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
1033                                         snapc, ceph_wbc.truncate_seq,
1034                                         ceph_wbc.truncate_size, false);
1035                 if (IS_ERR(req)) {
1036                         req = ceph_osdc_new_request(&fsc->client->osdc,
1037                                                 &ci->i_layout, vino,
1038                                                 offset, &len, 0,
1039                                                 min(num_ops,
1040                                                     CEPH_OSD_SLAB_OPS),
1041                                                 CEPH_OSD_OP_WRITE,
1042                                                 CEPH_OSD_FLAG_WRITE,
1043                                                 snapc, ceph_wbc.truncate_seq,
1044                                                 ceph_wbc.truncate_size, true);
1045                         BUG_ON(IS_ERR(req));
1046                 }
1047                 BUG_ON(len < page_offset(pages[locked_pages - 1]) +
1048                              PAGE_SIZE - offset);
1049 
1050                 req->r_callback = writepages_finish;
1051                 req->r_inode = inode;
1052 
1053                 /* Format the osd request message and submit the write */
1054                 len = 0;
1055                 data_pages = pages;
1056                 op_idx = 0;
1057                 for (i = 0; i < locked_pages; i++) {
1058                         u64 cur_offset = page_offset(pages[i]);
1059                         if (offset + len != cur_offset) {
1060                                 if (op_idx + 1 == req->r_num_ops)
1061                                         break;
1062                                 osd_req_op_extent_dup_last(req, op_idx,
1063                                                            cur_offset - offset);
1064                                 dout("writepages got pages at %llu~%llu\n",
1065                                      offset, len);
1066                                 osd_req_op_extent_osd_data_pages(req, op_idx,
1067                                                         data_pages, len, 0,
1068                                                         !!pool, false);
1069                                 osd_req_op_extent_update(req, op_idx, len);
1070 
1071                                 len = 0;
1072                                 offset = cur_offset; 
1073                                 data_pages = pages + i;
1074                                 op_idx++;
1075                         }
1076 
1077                         set_page_writeback(pages[i]);
1078                         len += PAGE_SIZE;
1079                 }
1080 
1081                 if (ceph_wbc.size_stable) {
1082                         len = min(len, ceph_wbc.i_size - offset);
1083                 } else if (i == locked_pages) {
1084                         /* writepages_finish() clears writeback pages
1085                          * according to the data length, so make sure
1086                          * data length covers all locked pages */
1087                         u64 min_len = len + 1 - PAGE_SIZE;
1088                         len = get_writepages_data_length(inode, pages[i - 1],
1089                                                          offset);
1090                         len = max(len, min_len);
1091                 }
1092                 dout("writepages got pages at %llu~%llu\n", offset, len);
1093 
1094                 osd_req_op_extent_osd_data_pages(req, op_idx, data_pages, len,
1095                                                  0, !!pool, false);
1096                 osd_req_op_extent_update(req, op_idx, len);
1097 
1098                 BUG_ON(op_idx + 1 != req->r_num_ops);
1099 
1100                 pool = NULL;
1101                 if (i < locked_pages) {
1102                         BUG_ON(num_ops <= req->r_num_ops);
1103                         num_ops -= req->r_num_ops;
1104                         locked_pages -= i;
1105 
1106                         /* allocate new pages array for next request */
1107                         data_pages = pages;
1108                         pages = kmalloc(locked_pages * sizeof (*pages),
1109                                         GFP_NOFS);
1110                         if (!pages) {
1111                                 pool = fsc->wb_pagevec_pool;
1112                                 pages = mempool_alloc(pool, GFP_NOFS);
1113                                 BUG_ON(!pages);
1114                         }
1115                         memcpy(pages, data_pages + i,
1116                                locked_pages * sizeof(*pages));
1117                         memset(data_pages + i, 0,
1118                                locked_pages * sizeof(*pages));
1119                 } else {
1120                         BUG_ON(num_ops != req->r_num_ops);
1121                         index = pages[i - 1]->index + 1;
1122                         /* request message now owns the pages array */
1123                         pages = NULL;
1124                 }
1125 
1126                 req->r_mtime = inode->i_mtime;
1127                 rc = ceph_osdc_start_request(&fsc->client->osdc, req, true);
1128                 BUG_ON(rc);
1129                 req = NULL;
1130 
1131                 wbc->nr_to_write -= i;
1132                 if (pages)
1133                         goto new_request;
1134 
1135                 /*
1136                  * We stop writing back only if we are not doing
1137                  * integrity sync. In case of integrity sync we have to
1138                  * keep going until we have written all the pages
1139                  * we tagged for writeback prior to entering this loop.
1140                  */
1141                 if (wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE)
1142                         done = stop = true;
1143 
1144 release_pvec_pages:
1145                 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr,
1146                      pvec.nr ? pvec.pages[0] : NULL);
1147                 pagevec_release(&pvec);
1148         }
1149 
1150         if (should_loop && !done) {
1151                 /* more to do; loop back to beginning of file */
1152                 dout("writepages looping back to beginning of file\n");
1153                 end = start_index - 1; /* OK even when start_index == 0 */
1154 
1155                 /* to write dirty pages associated with next snapc,
1156                  * we need to wait until current writes complete */
1157                 if (wbc->sync_mode != WB_SYNC_NONE &&
1158                     start_index == 0 && /* all dirty pages were checked */
1159                     !ceph_wbc.head_snapc) {
1160                         struct page *page;
1161                         unsigned i, nr;
1162                         index = 0;
1163                         while ((index <= end) &&
1164                                (nr = pagevec_lookup_tag(&pvec, mapping, &index,
1165                                                 PAGECACHE_TAG_WRITEBACK))) {
1166                                 for (i = 0; i < nr; i++) {
1167                                         page = pvec.pages[i];
1168                                         if (page_snap_context(page) != snapc)
1169                                                 continue;
1170                                         wait_on_page_writeback(page);
1171                                 }
1172                                 pagevec_release(&pvec);
1173                                 cond_resched();
1174                         }
1175                 }
1176 
1177                 start_index = 0;
1178                 index = 0;
1179                 goto retry;
1180         }
1181 
1182         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1183                 mapping->writeback_index = index;
1184 
1185 out:
1186         ceph_osdc_put_request(req);
1187         ceph_put_snap_context(last_snapc);
1188         dout("writepages dend - startone, rc = %d\n", rc);
1189         return rc;
1190 }
1191 
1192 
1193 
1194 /*
1195  * See if a given @snapc is either writeable, or already written.
1196  */
1197 static int context_is_writeable_or_written(struct inode *inode,
1198                                            struct ceph_snap_context *snapc)
1199 {
1200         struct ceph_snap_context *oldest = get_oldest_context(inode, NULL, NULL);
1201         int ret = !oldest || snapc->seq <= oldest->seq;
1202 
1203         ceph_put_snap_context(oldest);
1204         return ret;
1205 }
1206 
1207 /*
1208  * We are only allowed to write into/dirty the page if the page is
1209  * clean, or already dirty within the same snap context.
1210  *
1211  * called with page locked.
1212  * return success with page locked,
1213  * or any failure (incl -EAGAIN) with page unlocked.
1214  */
1215 static int ceph_update_writeable_page(struct file *file,
1216                             loff_t pos, unsigned len,
1217                             struct page *page)
1218 {
1219         struct inode *inode = file_inode(file);
1220         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1221         struct ceph_inode_info *ci = ceph_inode(inode);
1222         loff_t page_off = pos & PAGE_MASK;
1223         int pos_in_page = pos & ~PAGE_MASK;
1224         int end_in_page = pos_in_page + len;
1225         loff_t i_size;
1226         int r;
1227         struct ceph_snap_context *snapc, *oldest;
1228 
1229         if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1230                 dout(" page %p forced umount\n", page);
1231                 unlock_page(page);
1232                 return -EIO;
1233         }
1234 
1235 retry_locked:
1236         /* writepages currently holds page lock, but if we change that later, */
1237         wait_on_page_writeback(page);
1238 
1239         snapc = page_snap_context(page);
1240         if (snapc && snapc != ci->i_head_snapc) {
1241                 /*
1242                  * this page is already dirty in another (older) snap
1243                  * context!  is it writeable now?
1244                  */
1245                 oldest = get_oldest_context(inode, NULL, NULL);
1246                 if (snapc->seq > oldest->seq) {
1247                         ceph_put_snap_context(oldest);
1248                         dout(" page %p snapc %p not current or oldest\n",
1249                              page, snapc);
1250                         /*
1251                          * queue for writeback, and wait for snapc to
1252                          * be writeable or written
1253                          */
1254                         snapc = ceph_get_snap_context(snapc);
1255                         unlock_page(page);
1256                         ceph_queue_writeback(inode);
1257                         r = wait_event_killable(ci->i_cap_wq,
1258                                context_is_writeable_or_written(inode, snapc));
1259                         ceph_put_snap_context(snapc);
1260                         if (r == -ERESTARTSYS)
1261                                 return r;
1262                         return -EAGAIN;
1263                 }
1264                 ceph_put_snap_context(oldest);
1265 
1266                 /* yay, writeable, do it now (without dropping page lock) */
1267                 dout(" page %p snapc %p not current, but oldest\n",
1268                      page, snapc);
1269                 if (!clear_page_dirty_for_io(page))
1270                         goto retry_locked;
1271                 r = writepage_nounlock(page, NULL);
1272                 if (r < 0)
1273                         goto fail_unlock;
1274                 goto retry_locked;
1275         }
1276 
1277         if (PageUptodate(page)) {
1278                 dout(" page %p already uptodate\n", page);
1279                 return 0;
1280         }
1281 
1282         /* full page? */
1283         if (pos_in_page == 0 && len == PAGE_SIZE)
1284                 return 0;
1285 
1286         /* past end of file? */
1287         i_size = i_size_read(inode);
1288 
1289         if (page_off >= i_size ||
1290             (pos_in_page == 0 && (pos+len) >= i_size &&
1291              end_in_page - pos_in_page != PAGE_SIZE)) {
1292                 dout(" zeroing %p 0 - %d and %d - %d\n",
1293                      page, pos_in_page, end_in_page, (int)PAGE_SIZE);
1294                 zero_user_segments(page,
1295                                    0, pos_in_page,
1296                                    end_in_page, PAGE_SIZE);
1297                 return 0;
1298         }
1299 
1300         /* we need to read it. */
1301         r = ceph_do_readpage(file, page);
1302         if (r < 0) {
1303                 if (r == -EINPROGRESS)
1304                         return -EAGAIN;
1305                 goto fail_unlock;
1306         }
1307         goto retry_locked;
1308 fail_unlock:
1309         unlock_page(page);
1310         return r;
1311 }
1312 
1313 /*
1314  * We are only allowed to write into/dirty the page if the page is
1315  * clean, or already dirty within the same snap context.
1316  */
1317 static int ceph_write_begin(struct file *file, struct address_space *mapping,
1318                             loff_t pos, unsigned len, unsigned flags,
1319                             struct page **pagep, void **fsdata)
1320 {
1321         struct inode *inode = file_inode(file);
1322         struct page *page;
1323         pgoff_t index = pos >> PAGE_SHIFT;
1324         int r;
1325 
1326         do {
1327                 /* get a page */
1328                 page = grab_cache_page_write_begin(mapping, index, 0);
1329                 if (!page)
1330                         return -ENOMEM;
1331 
1332                 dout("write_begin file %p inode %p page %p %d~%d\n", file,
1333                      inode, page, (int)pos, (int)len);
1334 
1335                 r = ceph_update_writeable_page(file, pos, len, page);
1336                 if (r < 0)
1337                         put_page(page);
1338                 else
1339                         *pagep = page;
1340         } while (r == -EAGAIN);
1341 
1342         return r;
1343 }
1344 
1345 /*
1346  * we don't do anything in here that simple_write_end doesn't do
1347  * except adjust dirty page accounting
1348  */
1349 static int ceph_write_end(struct file *file, struct address_space *mapping,
1350                           loff_t pos, unsigned len, unsigned copied,
1351                           struct page *page, void *fsdata)
1352 {
1353         struct inode *inode = file_inode(file);
1354         bool check_cap = false;
1355 
1356         dout("write_end file %p inode %p page %p %d~%d (%d)\n", file,
1357              inode, page, (int)pos, (int)copied, (int)len);
1358 
1359         /* zero the stale part of the page if we did a short copy */
1360         if (!PageUptodate(page)) {
1361                 if (copied < len) {
1362                         copied = 0;
1363                         goto out;
1364                 }
1365                 SetPageUptodate(page);
1366         }
1367 
1368         /* did file size increase? */
1369         if (pos+copied > i_size_read(inode))
1370                 check_cap = ceph_inode_set_size(inode, pos+copied);
1371 
1372         set_page_dirty(page);
1373 
1374 out:
1375         unlock_page(page);
1376         put_page(page);
1377 
1378         if (check_cap)
1379                 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL);
1380 
1381         return copied;
1382 }
1383 
1384 /*
1385  * we set .direct_IO to indicate direct io is supported, but since we
1386  * intercept O_DIRECT reads and writes early, this function should
1387  * never get called.
1388  */
1389 static ssize_t ceph_direct_io(struct kiocb *iocb, struct iov_iter *iter)
1390 {
1391         WARN_ON(1);
1392         return -EINVAL;
1393 }
1394 
1395 const struct address_space_operations ceph_aops = {
1396         .readpage = ceph_readpage,
1397         .readpages = ceph_readpages,
1398         .writepage = ceph_writepage,
1399         .writepages = ceph_writepages_start,
1400         .write_begin = ceph_write_begin,
1401         .write_end = ceph_write_end,
1402         .set_page_dirty = ceph_set_page_dirty,
1403         .invalidatepage = ceph_invalidatepage,
1404         .releasepage = ceph_releasepage,
1405         .direct_IO = ceph_direct_io,
1406 };
1407 
1408 static void ceph_block_sigs(sigset_t *oldset)
1409 {
1410         sigset_t mask;
1411         siginitsetinv(&mask, sigmask(SIGKILL));
1412         sigprocmask(SIG_BLOCK, &mask, oldset);
1413 }
1414 
1415 static void ceph_restore_sigs(sigset_t *oldset)
1416 {
1417         sigprocmask(SIG_SETMASK, oldset, NULL);
1418 }
1419 
1420 /*
1421  * vm ops
1422  */
1423 static int ceph_filemap_fault(struct vm_fault *vmf)
1424 {
1425         struct vm_area_struct *vma = vmf->vma;
1426         struct inode *inode = file_inode(vma->vm_file);
1427         struct ceph_inode_info *ci = ceph_inode(inode);
1428         struct ceph_file_info *fi = vma->vm_file->private_data;
1429         struct page *pinned_page = NULL;
1430         loff_t off = vmf->pgoff << PAGE_SHIFT;
1431         int want, got, ret;
1432         sigset_t oldset;
1433 
1434         ceph_block_sigs(&oldset);
1435 
1436         dout("filemap_fault %p %llx.%llx %llu~%zd trying to get caps\n",
1437              inode, ceph_vinop(inode), off, (size_t)PAGE_SIZE);
1438         if (fi->fmode & CEPH_FILE_MODE_LAZY)
1439                 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1440         else
1441                 want = CEPH_CAP_FILE_CACHE;
1442 
1443         got = 0;
1444         ret = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page);
1445         if (ret < 0)
1446                 goto out_restore;
1447 
1448         dout("filemap_fault %p %llu~%zd got cap refs on %s\n",
1449              inode, off, (size_t)PAGE_SIZE, ceph_cap_string(got));
1450 
1451         if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) ||
1452             ci->i_inline_version == CEPH_INLINE_NONE) {
1453                 current->journal_info = vma->vm_file;
1454                 ret = filemap_fault(vmf);
1455                 current->journal_info = NULL;
1456         } else
1457                 ret = -EAGAIN;
1458 
1459         dout("filemap_fault %p %llu~%zd dropping cap refs on %s ret %d\n",
1460              inode, off, (size_t)PAGE_SIZE, ceph_cap_string(got), ret);
1461         if (pinned_page)
1462                 put_page(pinned_page);
1463         ceph_put_cap_refs(ci, got);
1464 
1465         if (ret != -EAGAIN)
1466                 goto out_restore;
1467 
1468         /* read inline data */
1469         if (off >= PAGE_SIZE) {
1470                 /* does not support inline data > PAGE_SIZE */
1471                 ret = VM_FAULT_SIGBUS;
1472         } else {
1473                 int ret1;
1474                 struct address_space *mapping = inode->i_mapping;
1475                 struct page *page = find_or_create_page(mapping, 0,
1476                                                 mapping_gfp_constraint(mapping,
1477                                                 ~__GFP_FS));
1478                 if (!page) {
1479                         ret = VM_FAULT_OOM;
1480                         goto out_inline;
1481                 }
1482                 ret1 = __ceph_do_getattr(inode, page,
1483                                          CEPH_STAT_CAP_INLINE_DATA, true);
1484                 if (ret1 < 0 || off >= i_size_read(inode)) {
1485                         unlock_page(page);
1486                         put_page(page);
1487                         if (ret1 < 0)
1488                                 ret = ret1;
1489                         else
1490                                 ret = VM_FAULT_SIGBUS;
1491                         goto out_inline;
1492                 }
1493                 if (ret1 < PAGE_SIZE)
1494                         zero_user_segment(page, ret1, PAGE_SIZE);
1495                 else
1496                         flush_dcache_page(page);
1497                 SetPageUptodate(page);
1498                 vmf->page = page;
1499                 ret = VM_FAULT_MAJOR | VM_FAULT_LOCKED;
1500 out_inline:
1501                 dout("filemap_fault %p %llu~%zd read inline data ret %d\n",
1502                      inode, off, (size_t)PAGE_SIZE, ret);
1503         }
1504 out_restore:
1505         ceph_restore_sigs(&oldset);
1506         if (ret < 0)
1507                 ret = (ret == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS;
1508 
1509         return ret;
1510 }
1511 
1512 /*
1513  * Reuse write_begin here for simplicity.
1514  */
1515 static int ceph_page_mkwrite(struct vm_fault *vmf)
1516 {
1517         struct vm_area_struct *vma = vmf->vma;
1518         struct inode *inode = file_inode(vma->vm_file);
1519         struct ceph_inode_info *ci = ceph_inode(inode);
1520         struct ceph_file_info *fi = vma->vm_file->private_data;
1521         struct ceph_cap_flush *prealloc_cf;
1522         struct page *page = vmf->page;
1523         loff_t off = page_offset(page);
1524         loff_t size = i_size_read(inode);
1525         size_t len;
1526         int want, got, ret;
1527         sigset_t oldset;
1528 
1529         prealloc_cf = ceph_alloc_cap_flush();
1530         if (!prealloc_cf)
1531                 return VM_FAULT_OOM;
1532 
1533         ceph_block_sigs(&oldset);
1534 
1535         if (ci->i_inline_version != CEPH_INLINE_NONE) {
1536                 struct page *locked_page = NULL;
1537                 if (off == 0) {
1538                         lock_page(page);
1539                         locked_page = page;
1540                 }
1541                 ret = ceph_uninline_data(vma->vm_file, locked_page);
1542                 if (locked_page)
1543                         unlock_page(locked_page);
1544                 if (ret < 0)
1545                         goto out_free;
1546         }
1547 
1548         if (off + PAGE_SIZE <= size)
1549                 len = PAGE_SIZE;
1550         else
1551                 len = size & ~PAGE_MASK;
1552 
1553         dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n",
1554              inode, ceph_vinop(inode), off, len, size);
1555         if (fi->fmode & CEPH_FILE_MODE_LAZY)
1556                 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1557         else
1558                 want = CEPH_CAP_FILE_BUFFER;
1559 
1560         got = 0;
1561         ret = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, off + len,
1562                             &got, NULL);
1563         if (ret < 0)
1564                 goto out_free;
1565 
1566         dout("page_mkwrite %p %llu~%zd got cap refs on %s\n",
1567              inode, off, len, ceph_cap_string(got));
1568 
1569         /* Update time before taking page lock */
1570         file_update_time(vma->vm_file);
1571 
1572         do {
1573                 lock_page(page);
1574 
1575                 if ((off > size) || (page->mapping != inode->i_mapping)) {
1576                         unlock_page(page);
1577                         ret = VM_FAULT_NOPAGE;
1578                         break;
1579                 }
1580 
1581                 ret = ceph_update_writeable_page(vma->vm_file, off, len, page);
1582                 if (ret >= 0) {
1583                         /* success.  we'll keep the page locked. */
1584                         set_page_dirty(page);
1585                         ret = VM_FAULT_LOCKED;
1586                 }
1587         } while (ret == -EAGAIN);
1588 
1589         if (ret == VM_FAULT_LOCKED ||
1590             ci->i_inline_version != CEPH_INLINE_NONE) {
1591                 int dirty;
1592                 spin_lock(&ci->i_ceph_lock);
1593                 ci->i_inline_version = CEPH_INLINE_NONE;
1594                 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1595                                                &prealloc_cf);
1596                 spin_unlock(&ci->i_ceph_lock);
1597                 if (dirty)
1598                         __mark_inode_dirty(inode, dirty);
1599         }
1600 
1601         dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %d\n",
1602              inode, off, len, ceph_cap_string(got), ret);
1603         ceph_put_cap_refs(ci, got);
1604 out_free:
1605         ceph_restore_sigs(&oldset);
1606         ceph_free_cap_flush(prealloc_cf);
1607         if (ret < 0)
1608                 ret = (ret == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS;
1609         return ret;
1610 }
1611 
1612 void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
1613                            char *data, size_t len)
1614 {
1615         struct address_space *mapping = inode->i_mapping;
1616         struct page *page;
1617 
1618         if (locked_page) {
1619                 page = locked_page;
1620         } else {
1621                 if (i_size_read(inode) == 0)
1622                         return;
1623                 page = find_or_create_page(mapping, 0,
1624                                            mapping_gfp_constraint(mapping,
1625                                            ~__GFP_FS));
1626                 if (!page)
1627                         return;
1628                 if (PageUptodate(page)) {
1629                         unlock_page(page);
1630                         put_page(page);
1631                         return;
1632                 }
1633         }
1634 
1635         dout("fill_inline_data %p %llx.%llx len %zu locked_page %p\n",
1636              inode, ceph_vinop(inode), len, locked_page);
1637 
1638         if (len > 0) {
1639                 void *kaddr = kmap_atomic(page);
1640                 memcpy(kaddr, data, len);
1641                 kunmap_atomic(kaddr);
1642         }
1643 
1644         if (page != locked_page) {
1645                 if (len < PAGE_SIZE)
1646                         zero_user_segment(page, len, PAGE_SIZE);
1647                 else
1648                         flush_dcache_page(page);
1649 
1650                 SetPageUptodate(page);
1651                 unlock_page(page);
1652                 put_page(page);
1653         }
1654 }
1655 
1656 int ceph_uninline_data(struct file *filp, struct page *locked_page)
1657 {
1658         struct inode *inode = file_inode(filp);
1659         struct ceph_inode_info *ci = ceph_inode(inode);
1660         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1661         struct ceph_osd_request *req;
1662         struct page *page = NULL;
1663         u64 len, inline_version;
1664         int err = 0;
1665         bool from_pagecache = false;
1666 
1667         spin_lock(&ci->i_ceph_lock);
1668         inline_version = ci->i_inline_version;
1669         spin_unlock(&ci->i_ceph_lock);
1670 
1671         dout("uninline_data %p %llx.%llx inline_version %llu\n",
1672              inode, ceph_vinop(inode), inline_version);
1673 
1674         if (inline_version == 1 || /* initial version, no data */
1675             inline_version == CEPH_INLINE_NONE)
1676                 goto out;
1677 
1678         if (locked_page) {
1679                 page = locked_page;
1680                 WARN_ON(!PageUptodate(page));
1681         } else if (ceph_caps_issued(ci) &
1682                    (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) {
1683                 page = find_get_page(inode->i_mapping, 0);
1684                 if (page) {
1685                         if (PageUptodate(page)) {
1686                                 from_pagecache = true;
1687                                 lock_page(page);
1688                         } else {
1689                                 put_page(page);
1690                                 page = NULL;
1691                         }
1692                 }
1693         }
1694 
1695         if (page) {
1696                 len = i_size_read(inode);
1697                 if (len > PAGE_SIZE)
1698                         len = PAGE_SIZE;
1699         } else {
1700                 page = __page_cache_alloc(GFP_NOFS);
1701                 if (!page) {
1702                         err = -ENOMEM;
1703                         goto out;
1704                 }
1705                 err = __ceph_do_getattr(inode, page,
1706                                         CEPH_STAT_CAP_INLINE_DATA, true);
1707                 if (err < 0) {
1708                         /* no inline data */
1709                         if (err == -ENODATA)
1710                                 err = 0;
1711                         goto out;
1712                 }
1713                 len = err;
1714         }
1715 
1716         req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1717                                     ceph_vino(inode), 0, &len, 0, 1,
1718                                     CEPH_OSD_OP_CREATE, CEPH_OSD_FLAG_WRITE,
1719                                     NULL, 0, 0, false);
1720         if (IS_ERR(req)) {
1721                 err = PTR_ERR(req);
1722                 goto out;
1723         }
1724 
1725         req->r_mtime = inode->i_mtime;
1726         err = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1727         if (!err)
1728                 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1729         ceph_osdc_put_request(req);
1730         if (err < 0)
1731                 goto out;
1732 
1733         req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1734                                     ceph_vino(inode), 0, &len, 1, 3,
1735                                     CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
1736                                     NULL, ci->i_truncate_seq,
1737                                     ci->i_truncate_size, false);
1738         if (IS_ERR(req)) {
1739                 err = PTR_ERR(req);
1740                 goto out;
1741         }
1742 
1743         osd_req_op_extent_osd_data_pages(req, 1, &page, len, 0, false, false);
1744 
1745         {
1746                 __le64 xattr_buf = cpu_to_le64(inline_version);
1747                 err = osd_req_op_xattr_init(req, 0, CEPH_OSD_OP_CMPXATTR,
1748                                             "inline_version", &xattr_buf,
1749                                             sizeof(xattr_buf),
1750                                             CEPH_OSD_CMPXATTR_OP_GT,
1751                                             CEPH_OSD_CMPXATTR_MODE_U64);
1752                 if (err)
1753                         goto out_put;
1754         }
1755 
1756         {
1757                 char xattr_buf[32];
1758                 int xattr_len = snprintf(xattr_buf, sizeof(xattr_buf),
1759                                          "%llu", inline_version);
1760                 err = osd_req_op_xattr_init(req, 2, CEPH_OSD_OP_SETXATTR,
1761                                             "inline_version",
1762                                             xattr_buf, xattr_len, 0, 0);
1763                 if (err)
1764                         goto out_put;
1765         }
1766 
1767         req->r_mtime = inode->i_mtime;
1768         err = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1769         if (!err)
1770                 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1771 out_put:
1772         ceph_osdc_put_request(req);
1773         if (err == -ECANCELED)
1774                 err = 0;
1775 out:
1776         if (page && page != locked_page) {
1777                 if (from_pagecache) {
1778                         unlock_page(page);
1779                         put_page(page);
1780                 } else
1781                         __free_pages(page, 0);
1782         }
1783 
1784         dout("uninline_data %p %llx.%llx inline_version %llu = %d\n",
1785              inode, ceph_vinop(inode), inline_version, err);
1786         return err;
1787 }
1788 
1789 static const struct vm_operations_struct ceph_vmops = {
1790         .fault          = ceph_filemap_fault,
1791         .page_mkwrite   = ceph_page_mkwrite,
1792 };
1793 
1794 int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1795 {
1796         struct address_space *mapping = file->f_mapping;
1797 
1798         if (!mapping->a_ops->readpage)
1799                 return -ENOEXEC;
1800         file_accessed(file);
1801         vma->vm_ops = &ceph_vmops;
1802         return 0;
1803 }
1804 
1805 enum {
1806         POOL_READ       = 1,
1807         POOL_WRITE      = 2,
1808 };
1809 
1810 static int __ceph_pool_perm_get(struct ceph_inode_info *ci,
1811                                 s64 pool, struct ceph_string *pool_ns)
1812 {
1813         struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode);
1814         struct ceph_mds_client *mdsc = fsc->mdsc;
1815         struct ceph_osd_request *rd_req = NULL, *wr_req = NULL;
1816         struct rb_node **p, *parent;
1817         struct ceph_pool_perm *perm;
1818         struct page **pages;
1819         size_t pool_ns_len;
1820         int err = 0, err2 = 0, have = 0;
1821 
1822         down_read(&mdsc->pool_perm_rwsem);
1823         p = &mdsc->pool_perm_tree.rb_node;
1824         while (*p) {
1825                 perm = rb_entry(*p, struct ceph_pool_perm, node);
1826                 if (pool < perm->pool)
1827                         p = &(*p)->rb_left;
1828                 else if (pool > perm->pool)
1829                         p = &(*p)->rb_right;
1830                 else {
1831                         int ret = ceph_compare_string(pool_ns,
1832                                                 perm->pool_ns,
1833                                                 perm->pool_ns_len);
1834                         if (ret < 0)
1835                                 p = &(*p)->rb_left;
1836                         else if (ret > 0)
1837                                 p = &(*p)->rb_right;
1838                         else {
1839                                 have = perm->perm;
1840                                 break;
1841                         }
1842                 }
1843         }
1844         up_read(&mdsc->pool_perm_rwsem);
1845         if (*p)
1846                 goto out;
1847 
1848         if (pool_ns)
1849                 dout("__ceph_pool_perm_get pool %lld ns %.*s no perm cached\n",
1850                      pool, (int)pool_ns->len, pool_ns->str);
1851         else
1852                 dout("__ceph_pool_perm_get pool %lld no perm cached\n", pool);
1853 
1854         down_write(&mdsc->pool_perm_rwsem);
1855         p = &mdsc->pool_perm_tree.rb_node;
1856         parent = NULL;
1857         while (*p) {
1858                 parent = *p;
1859                 perm = rb_entry(parent, struct ceph_pool_perm, node);
1860                 if (pool < perm->pool)
1861                         p = &(*p)->rb_left;
1862                 else if (pool > perm->pool)
1863                         p = &(*p)->rb_right;
1864                 else {
1865                         int ret = ceph_compare_string(pool_ns,
1866                                                 perm->pool_ns,
1867                                                 perm->pool_ns_len);
1868                         if (ret < 0)
1869                                 p = &(*p)->rb_left;
1870                         else if (ret > 0)
1871                                 p = &(*p)->rb_right;
1872                         else {
1873                                 have = perm->perm;
1874                                 break;
1875                         }
1876                 }
1877         }
1878         if (*p) {
1879                 up_write(&mdsc->pool_perm_rwsem);
1880                 goto out;
1881         }
1882 
1883         rd_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1884                                          1, false, GFP_NOFS);
1885         if (!rd_req) {
1886                 err = -ENOMEM;
1887                 goto out_unlock;
1888         }
1889 
1890         rd_req->r_flags = CEPH_OSD_FLAG_READ;
1891         osd_req_op_init(rd_req, 0, CEPH_OSD_OP_STAT, 0);
1892         rd_req->r_base_oloc.pool = pool;
1893         if (pool_ns)
1894                 rd_req->r_base_oloc.pool_ns = ceph_get_string(pool_ns);
1895         ceph_oid_printf(&rd_req->r_base_oid, "%llx.00000000", ci->i_vino.ino);
1896 
1897         err = ceph_osdc_alloc_messages(rd_req, GFP_NOFS);
1898         if (err)
1899                 goto out_unlock;
1900 
1901         wr_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1902                                          1, false, GFP_NOFS);
1903         if (!wr_req) {
1904                 err = -ENOMEM;
1905                 goto out_unlock;
1906         }
1907 
1908         wr_req->r_flags = CEPH_OSD_FLAG_WRITE;
1909         osd_req_op_init(wr_req, 0, CEPH_OSD_OP_CREATE, CEPH_OSD_OP_FLAG_EXCL);
1910         ceph_oloc_copy(&wr_req->r_base_oloc, &rd_req->r_base_oloc);
1911         ceph_oid_copy(&wr_req->r_base_oid, &rd_req->r_base_oid);
1912 
1913         err = ceph_osdc_alloc_messages(wr_req, GFP_NOFS);
1914         if (err)
1915                 goto out_unlock;
1916 
1917         /* one page should be large enough for STAT data */
1918         pages = ceph_alloc_page_vector(1, GFP_KERNEL);
1919         if (IS_ERR(pages)) {
1920                 err = PTR_ERR(pages);
1921                 goto out_unlock;
1922         }
1923 
1924         osd_req_op_raw_data_in_pages(rd_req, 0, pages, PAGE_SIZE,
1925                                      0, false, true);
1926         err = ceph_osdc_start_request(&fsc->client->osdc, rd_req, false);
1927 
1928         wr_req->r_mtime = ci->vfs_inode.i_mtime;
1929         wr_req->r_abort_on_full = true;
1930         err2 = ceph_osdc_start_request(&fsc->client->osdc, wr_req, false);
1931 
1932         if (!err)
1933                 err = ceph_osdc_wait_request(&fsc->client->osdc, rd_req);
1934         if (!err2)
1935                 err2 = ceph_osdc_wait_request(&fsc->client->osdc, wr_req);
1936 
1937         if (err >= 0 || err == -ENOENT)
1938                 have |= POOL_READ;
1939         else if (err != -EPERM)
1940                 goto out_unlock;
1941 
1942         if (err2 == 0 || err2 == -EEXIST)
1943                 have |= POOL_WRITE;
1944         else if (err2 != -EPERM) {
1945                 err = err2;
1946                 goto out_unlock;
1947         }
1948 
1949         pool_ns_len = pool_ns ? pool_ns->len : 0;
1950         perm = kmalloc(sizeof(*perm) + pool_ns_len + 1, GFP_NOFS);
1951         if (!perm) {
1952                 err = -ENOMEM;
1953                 goto out_unlock;
1954         }
1955 
1956         perm->pool = pool;
1957         perm->perm = have;
1958         perm->pool_ns_len = pool_ns_len;
1959         if (pool_ns_len > 0)
1960                 memcpy(perm->pool_ns, pool_ns->str, pool_ns_len);
1961         perm->pool_ns[pool_ns_len] = 0;
1962 
1963         rb_link_node(&perm->node, parent, p);
1964         rb_insert_color(&perm->node, &mdsc->pool_perm_tree);
1965         err = 0;
1966 out_unlock:
1967         up_write(&mdsc->pool_perm_rwsem);
1968 
1969         ceph_osdc_put_request(rd_req);
1970         ceph_osdc_put_request(wr_req);
1971 out:
1972         if (!err)
1973                 err = have;
1974         if (pool_ns)
1975                 dout("__ceph_pool_perm_get pool %lld ns %.*s result = %d\n",
1976                      pool, (int)pool_ns->len, pool_ns->str, err);
1977         else
1978                 dout("__ceph_pool_perm_get pool %lld result = %d\n", pool, err);
1979         return err;
1980 }
1981 
1982 int ceph_pool_perm_check(struct ceph_inode_info *ci, int need)
1983 {
1984         s64 pool;
1985         struct ceph_string *pool_ns;
1986         int ret, flags;
1987 
1988         if (ci->i_vino.snap != CEPH_NOSNAP) {
1989                 /*
1990                  * Pool permission check needs to write to the first object.
1991                  * But for snapshot, head of the first object may have alread
1992                  * been deleted. Skip check to avoid creating orphan object.
1993                  */
1994                 return 0;
1995         }
1996 
1997         if (ceph_test_mount_opt(ceph_inode_to_client(&ci->vfs_inode),
1998                                 NOPOOLPERM))
1999                 return 0;
2000 
2001         spin_lock(&ci->i_ceph_lock);
2002         flags = ci->i_ceph_flags;
2003         pool = ci->i_layout.pool_id;
2004         spin_unlock(&ci->i_ceph_lock);
2005 check:
2006         if (flags & CEPH_I_POOL_PERM) {
2007                 if ((need & CEPH_CAP_FILE_RD) && !(flags & CEPH_I_POOL_RD)) {
2008                         dout("ceph_pool_perm_check pool %lld no read perm\n",
2009                              pool);
2010                         return -EPERM;
2011                 }
2012                 if ((need & CEPH_CAP_FILE_WR) && !(flags & CEPH_I_POOL_WR)) {
2013                         dout("ceph_pool_perm_check pool %lld no write perm\n",
2014                              pool);
2015                         return -EPERM;
2016                 }
2017                 return 0;
2018         }
2019 
2020         pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
2021         ret = __ceph_pool_perm_get(ci, pool, pool_ns);
2022         ceph_put_string(pool_ns);
2023         if (ret < 0)
2024                 return ret;
2025 
2026         flags = CEPH_I_POOL_PERM;
2027         if (ret & POOL_READ)
2028                 flags |= CEPH_I_POOL_RD;
2029         if (ret & POOL_WRITE)
2030                 flags |= CEPH_I_POOL_WR;
2031 
2032         spin_lock(&ci->i_ceph_lock);
2033         if (pool == ci->i_layout.pool_id &&
2034             pool_ns == rcu_dereference_raw(ci->i_layout.pool_ns)) {
2035                 ci->i_ceph_flags |= flags;
2036         } else {
2037                 pool = ci->i_layout.pool_id;
2038                 flags = ci->i_ceph_flags;
2039         }
2040         spin_unlock(&ci->i_ceph_lock);
2041         goto check;
2042 }
2043 
2044 void ceph_pool_perm_destroy(struct ceph_mds_client *mdsc)
2045 {
2046         struct ceph_pool_perm *perm;
2047         struct rb_node *n;
2048 
2049         while (!RB_EMPTY_ROOT(&mdsc->pool_perm_tree)) {
2050                 n = rb_first(&mdsc->pool_perm_tree);
2051                 perm = rb_entry(n, struct ceph_pool_perm, node);
2052                 rb_erase(n, &mdsc->pool_perm_tree);
2053                 kfree(perm);
2054         }
2055 }
2056 

~ [ 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