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

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

Version: ~ [ linux-5.19-rc3 ] ~ [ linux-5.18.5 ] ~ [ linux-5.17.15 ] ~ [ linux-5.16.20 ] ~ [ linux-5.15.48 ] ~ [ linux-5.14.21 ] ~ [ linux-5.13.19 ] ~ [ linux-5.12.19 ] ~ [ linux-5.11.22 ] ~ [ linux-5.10.123 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.199 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.248 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.284 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.319 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.302 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.9 ] ~ [ 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 #include <linux/ceph/striper.h>
 19 
 20 /*
 21  * Ceph address space ops.
 22  *
 23  * There are a few funny things going on here.
 24  *
 25  * The page->private field is used to reference a struct
 26  * ceph_snap_context for _every_ dirty page.  This indicates which
 27  * snapshot the page was logically dirtied in, and thus which snap
 28  * context needs to be associated with the osd write during writeback.
 29  *
 30  * Similarly, struct ceph_inode_info maintains a set of counters to
 31  * count dirty pages on the inode.  In the absence of snapshots,
 32  * i_wrbuffer_ref == i_wrbuffer_ref_head == the dirty page count.
 33  *
 34  * When a snapshot is taken (that is, when the client receives
 35  * notification that a snapshot was taken), each inode with caps and
 36  * with dirty pages (dirty pages implies there is a cap) gets a new
 37  * ceph_cap_snap in the i_cap_snaps list (which is sorted in ascending
 38  * order, new snaps go to the tail).  The i_wrbuffer_ref_head count is
 39  * moved to capsnap->dirty. (Unless a sync write is currently in
 40  * progress.  In that case, the capsnap is said to be "pending", new
 41  * writes cannot start, and the capsnap isn't "finalized" until the
 42  * write completes (or fails) and a final size/mtime for the inode for
 43  * that snap can be settled upon.)  i_wrbuffer_ref_head is reset to 0.
 44  *
 45  * On writeback, we must submit writes to the osd IN SNAP ORDER.  So,
 46  * we look for the first capsnap in i_cap_snaps and write out pages in
 47  * that snap context _only_.  Then we move on to the next capsnap,
 48  * eventually reaching the "live" or "head" context (i.e., pages that
 49  * are not yet snapped) and are writing the most recently dirtied
 50  * pages.
 51  *
 52  * Invalidate and so forth must take care to ensure the dirty page
 53  * accounting is preserved.
 54  */
 55 
 56 #define CONGESTION_ON_THRESH(congestion_kb) (congestion_kb >> (PAGE_SHIFT-10))
 57 #define CONGESTION_OFF_THRESH(congestion_kb)                            \
 58         (CONGESTION_ON_THRESH(congestion_kb) -                          \
 59          (CONGESTION_ON_THRESH(congestion_kb) >> 2))
 60 
 61 static inline struct ceph_snap_context *page_snap_context(struct page *page)
 62 {
 63         if (PagePrivate(page))
 64                 return (void *)page->private;
 65         return NULL;
 66 }
 67 
 68 /*
 69  * Dirty a page.  Optimistically adjust accounting, on the assumption
 70  * that we won't race with invalidate.  If we do, readjust.
 71  */
 72 static int ceph_set_page_dirty(struct page *page)
 73 {
 74         struct address_space *mapping = page->mapping;
 75         struct inode *inode;
 76         struct ceph_inode_info *ci;
 77         struct ceph_snap_context *snapc;
 78         int ret;
 79 
 80         if (unlikely(!mapping))
 81                 return !TestSetPageDirty(page);
 82 
 83         if (PageDirty(page)) {
 84                 dout("%p set_page_dirty %p idx %lu -- already dirty\n",
 85                      mapping->host, page, page->index);
 86                 BUG_ON(!PagePrivate(page));
 87                 return 0;
 88         }
 89 
 90         inode = mapping->host;
 91         ci = ceph_inode(inode);
 92 
 93         /* dirty the head */
 94         spin_lock(&ci->i_ceph_lock);
 95         BUG_ON(ci->i_wr_ref == 0); // caller should hold Fw reference
 96         if (__ceph_have_pending_cap_snap(ci)) {
 97                 struct ceph_cap_snap *capsnap =
 98                                 list_last_entry(&ci->i_cap_snaps,
 99                                                 struct ceph_cap_snap,
100                                                 ci_item);
101                 snapc = ceph_get_snap_context(capsnap->context);
102                 capsnap->dirty_pages++;
103         } else {
104                 BUG_ON(!ci->i_head_snapc);
105                 snapc = ceph_get_snap_context(ci->i_head_snapc);
106                 ++ci->i_wrbuffer_ref_head;
107         }
108         if (ci->i_wrbuffer_ref == 0)
109                 ihold(inode);
110         ++ci->i_wrbuffer_ref;
111         dout("%p set_page_dirty %p idx %lu head %d/%d -> %d/%d "
112              "snapc %p seq %lld (%d snaps)\n",
113              mapping->host, page, page->index,
114              ci->i_wrbuffer_ref-1, ci->i_wrbuffer_ref_head-1,
115              ci->i_wrbuffer_ref, ci->i_wrbuffer_ref_head,
116              snapc, snapc->seq, snapc->num_snaps);
117         spin_unlock(&ci->i_ceph_lock);
118 
119         /*
120          * Reference snap context in page->private.  Also set
121          * PagePrivate so that we get invalidatepage callback.
122          */
123         BUG_ON(PagePrivate(page));
124         page->private = (unsigned long)snapc;
125         SetPagePrivate(page);
126 
127         ret = __set_page_dirty_nobuffers(page);
128         WARN_ON(!PageLocked(page));
129         WARN_ON(!page->mapping);
130 
131         return ret;
132 }
133 
134 /*
135  * If we are truncating the full page (i.e. offset == 0), adjust the
136  * dirty page counters appropriately.  Only called if there is private
137  * data on the page.
138  */
139 static void ceph_invalidatepage(struct page *page, unsigned int offset,
140                                 unsigned int length)
141 {
142         struct inode *inode;
143         struct ceph_inode_info *ci;
144         struct ceph_snap_context *snapc = page_snap_context(page);
145 
146         inode = page->mapping->host;
147         ci = ceph_inode(inode);
148 
149         if (offset != 0 || length != PAGE_SIZE) {
150                 dout("%p invalidatepage %p idx %lu partial dirty page %u~%u\n",
151                      inode, page, page->index, offset, length);
152                 return;
153         }
154 
155         ceph_invalidate_fscache_page(inode, page);
156 
157         WARN_ON(!PageLocked(page));
158         if (!PagePrivate(page))
159                 return;
160 
161         ClearPageChecked(page);
162 
163         dout("%p invalidatepage %p idx %lu full dirty page\n",
164              inode, page, page->index);
165 
166         ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
167         ceph_put_snap_context(snapc);
168         page->private = 0;
169         ClearPagePrivate(page);
170 }
171 
172 static int ceph_releasepage(struct page *page, gfp_t g)
173 {
174         dout("%p releasepage %p idx %lu (%sdirty)\n", page->mapping->host,
175              page, page->index, PageDirty(page) ? "" : "not ");
176 
177         /* Can we release the page from the cache? */
178         if (!ceph_release_fscache_page(page, g))
179                 return 0;
180 
181         return !PagePrivate(page);
182 }
183 
184 /*
185  * read a single page, without unlocking it.
186  */
187 static int ceph_do_readpage(struct file *filp, struct page *page)
188 {
189         struct inode *inode = file_inode(filp);
190         struct ceph_inode_info *ci = ceph_inode(inode);
191         struct ceph_osd_client *osdc =
192                 &ceph_inode_to_client(inode)->client->osdc;
193         int err = 0;
194         u64 off = page_offset(page);
195         u64 len = PAGE_SIZE;
196 
197         if (off >= i_size_read(inode)) {
198                 zero_user_segment(page, 0, PAGE_SIZE);
199                 SetPageUptodate(page);
200                 return 0;
201         }
202 
203         if (ci->i_inline_version != CEPH_INLINE_NONE) {
204                 /*
205                  * Uptodate inline data should have been added
206                  * into page cache while getting Fcr caps.
207                  */
208                 if (off == 0)
209                         return -EINVAL;
210                 zero_user_segment(page, 0, PAGE_SIZE);
211                 SetPageUptodate(page);
212                 return 0;
213         }
214 
215         err = ceph_readpage_from_fscache(inode, page);
216         if (err == 0)
217                 return -EINPROGRESS;
218 
219         dout("readpage inode %p file %p page %p index %lu\n",
220              inode, filp, page, page->index);
221         err = ceph_osdc_readpages(osdc, ceph_vino(inode), &ci->i_layout,
222                                   off, &len,
223                                   ci->i_truncate_seq, ci->i_truncate_size,
224                                   &page, 1, 0);
225         if (err == -ENOENT)
226                 err = 0;
227         if (err < 0) {
228                 SetPageError(page);
229                 ceph_fscache_readpage_cancel(inode, page);
230                 goto out;
231         }
232         if (err < PAGE_SIZE)
233                 /* zero fill remainder of page */
234                 zero_user_segment(page, err, PAGE_SIZE);
235         else
236                 flush_dcache_page(page);
237 
238         SetPageUptodate(page);
239         ceph_readpage_to_fscache(inode, page);
240 
241 out:
242         return err < 0 ? err : 0;
243 }
244 
245 static int ceph_readpage(struct file *filp, struct page *page)
246 {
247         int r = ceph_do_readpage(filp, page);
248         if (r != -EINPROGRESS)
249                 unlock_page(page);
250         else
251                 r = 0;
252         return r;
253 }
254 
255 /*
256  * Finish an async read(ahead) op.
257  */
258 static void finish_read(struct ceph_osd_request *req)
259 {
260         struct inode *inode = req->r_inode;
261         struct ceph_osd_data *osd_data;
262         int rc = req->r_result <= 0 ? req->r_result : 0;
263         int bytes = req->r_result >= 0 ? req->r_result : 0;
264         int num_pages;
265         int i;
266 
267         dout("finish_read %p req %p rc %d bytes %d\n", inode, req, rc, bytes);
268 
269         /* unlock all pages, zeroing any data we didn't read */
270         osd_data = osd_req_op_extent_osd_data(req, 0);
271         BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
272         num_pages = calc_pages_for((u64)osd_data->alignment,
273                                         (u64)osd_data->length);
274         for (i = 0; i < num_pages; i++) {
275                 struct page *page = osd_data->pages[i];
276 
277                 if (rc < 0 && rc != -ENOENT) {
278                         ceph_fscache_readpage_cancel(inode, page);
279                         goto unlock;
280                 }
281                 if (bytes < (int)PAGE_SIZE) {
282                         /* zero (remainder of) page */
283                         int s = bytes < 0 ? 0 : bytes;
284                         zero_user_segment(page, s, PAGE_SIZE);
285                 }
286                 dout("finish_read %p uptodate %p idx %lu\n", inode, page,
287                      page->index);
288                 flush_dcache_page(page);
289                 SetPageUptodate(page);
290                 ceph_readpage_to_fscache(inode, page);
291 unlock:
292                 unlock_page(page);
293                 put_page(page);
294                 bytes -= PAGE_SIZE;
295         }
296         kfree(osd_data->pages);
297 }
298 
299 /*
300  * start an async read(ahead) operation.  return nr_pages we submitted
301  * a read for on success, or negative error code.
302  */
303 static int start_read(struct inode *inode, struct ceph_rw_context *rw_ctx,
304                       struct list_head *page_list, int max)
305 {
306         struct ceph_osd_client *osdc =
307                 &ceph_inode_to_client(inode)->client->osdc;
308         struct ceph_inode_info *ci = ceph_inode(inode);
309         struct page *page = list_entry(page_list->prev, struct page, lru);
310         struct ceph_vino vino;
311         struct ceph_osd_request *req;
312         u64 off;
313         u64 len;
314         int i;
315         struct page **pages;
316         pgoff_t next_index;
317         int nr_pages = 0;
318         int got = 0;
319         int ret = 0;
320 
321         if (!rw_ctx) {
322                 /* caller of readpages does not hold buffer and read caps
323                  * (fadvise, madvise and readahead cases) */
324                 int want = CEPH_CAP_FILE_CACHE;
325                 ret = ceph_try_get_caps(ci, CEPH_CAP_FILE_RD, want, true, &got);
326                 if (ret < 0) {
327                         dout("start_read %p, error getting cap\n", inode);
328                 } else if (!(got & want)) {
329                         dout("start_read %p, no cache cap\n", inode);
330                         ret = 0;
331                 }
332                 if (ret <= 0) {
333                         if (got)
334                                 ceph_put_cap_refs(ci, got);
335                         while (!list_empty(page_list)) {
336                                 page = list_entry(page_list->prev,
337                                                   struct page, lru);
338                                 list_del(&page->lru);
339                                 put_page(page);
340                         }
341                         return ret;
342                 }
343         }
344 
345         off = (u64) page_offset(page);
346 
347         /* count pages */
348         next_index = page->index;
349         list_for_each_entry_reverse(page, page_list, lru) {
350                 if (page->index != next_index)
351                         break;
352                 nr_pages++;
353                 next_index++;
354                 if (max && nr_pages == max)
355                         break;
356         }
357         len = nr_pages << PAGE_SHIFT;
358         dout("start_read %p nr_pages %d is %lld~%lld\n", inode, nr_pages,
359              off, len);
360         vino = ceph_vino(inode);
361         req = ceph_osdc_new_request(osdc, &ci->i_layout, vino, off, &len,
362                                     0, 1, CEPH_OSD_OP_READ,
363                                     CEPH_OSD_FLAG_READ, NULL,
364                                     ci->i_truncate_seq, ci->i_truncate_size,
365                                     false);
366         if (IS_ERR(req)) {
367                 ret = PTR_ERR(req);
368                 goto out;
369         }
370 
371         /* build page vector */
372         nr_pages = calc_pages_for(0, len);
373         pages = kmalloc_array(nr_pages, sizeof(*pages), GFP_KERNEL);
374         if (!pages) {
375                 ret = -ENOMEM;
376                 goto out_put;
377         }
378         for (i = 0; i < nr_pages; ++i) {
379                 page = list_entry(page_list->prev, struct page, lru);
380                 BUG_ON(PageLocked(page));
381                 list_del(&page->lru);
382 
383                 dout("start_read %p adding %p idx %lu\n", inode, page,
384                      page->index);
385                 if (add_to_page_cache_lru(page, &inode->i_data, page->index,
386                                           GFP_KERNEL)) {
387                         ceph_fscache_uncache_page(inode, page);
388                         put_page(page);
389                         dout("start_read %p add_to_page_cache failed %p\n",
390                              inode, page);
391                         nr_pages = i;
392                         if (nr_pages > 0) {
393                                 len = nr_pages << PAGE_SHIFT;
394                                 osd_req_op_extent_update(req, 0, len);
395                                 break;
396                         }
397                         goto out_pages;
398                 }
399                 pages[i] = page;
400         }
401         osd_req_op_extent_osd_data_pages(req, 0, pages, len, 0, false, false);
402         req->r_callback = finish_read;
403         req->r_inode = inode;
404 
405         dout("start_read %p starting %p %lld~%lld\n", inode, req, off, len);
406         ret = ceph_osdc_start_request(osdc, req, false);
407         if (ret < 0)
408                 goto out_pages;
409         ceph_osdc_put_request(req);
410 
411         /* After adding locked pages to page cache, the inode holds cache cap.
412          * So we can drop our cap refs. */
413         if (got)
414                 ceph_put_cap_refs(ci, got);
415 
416         return nr_pages;
417 
418 out_pages:
419         for (i = 0; i < nr_pages; ++i) {
420                 ceph_fscache_readpage_cancel(inode, pages[i]);
421                 unlock_page(pages[i]);
422         }
423         ceph_put_page_vector(pages, nr_pages, false);
424 out_put:
425         ceph_osdc_put_request(req);
426 out:
427         if (got)
428                 ceph_put_cap_refs(ci, got);
429         return ret;
430 }
431 
432 
433 /*
434  * Read multiple pages.  Leave pages we don't read + unlock in page_list;
435  * the caller (VM) cleans them up.
436  */
437 static int ceph_readpages(struct file *file, struct address_space *mapping,
438                           struct list_head *page_list, unsigned nr_pages)
439 {
440         struct inode *inode = file_inode(file);
441         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
442         struct ceph_file_info *fi = file->private_data;
443         struct ceph_rw_context *rw_ctx;
444         int rc = 0;
445         int max = 0;
446 
447         if (ceph_inode(inode)->i_inline_version != CEPH_INLINE_NONE)
448                 return -EINVAL;
449 
450         rc = ceph_readpages_from_fscache(mapping->host, mapping, page_list,
451                                          &nr_pages);
452 
453         if (rc == 0)
454                 goto out;
455 
456         rw_ctx = ceph_find_rw_context(fi);
457         max = fsc->mount_options->rsize >> PAGE_SHIFT;
458         dout("readpages %p file %p ctx %p nr_pages %d max %d\n",
459              inode, file, rw_ctx, nr_pages, max);
460         while (!list_empty(page_list)) {
461                 rc = start_read(inode, rw_ctx, page_list, max);
462                 if (rc < 0)
463                         goto out;
464         }
465 out:
466         ceph_fscache_readpages_cancel(inode, page_list);
467 
468         dout("readpages %p file %p ret %d\n", inode, file, rc);
469         return rc;
470 }
471 
472 struct ceph_writeback_ctl
473 {
474         loff_t i_size;
475         u64 truncate_size;
476         u32 truncate_seq;
477         bool size_stable;
478         bool head_snapc;
479 };
480 
481 /*
482  * Get ref for the oldest snapc for an inode with dirty data... that is, the
483  * only snap context we are allowed to write back.
484  */
485 static struct ceph_snap_context *
486 get_oldest_context(struct inode *inode, struct ceph_writeback_ctl *ctl,
487                    struct ceph_snap_context *page_snapc)
488 {
489         struct ceph_inode_info *ci = ceph_inode(inode);
490         struct ceph_snap_context *snapc = NULL;
491         struct ceph_cap_snap *capsnap = NULL;
492 
493         spin_lock(&ci->i_ceph_lock);
494         list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
495                 dout(" cap_snap %p snapc %p has %d dirty pages\n", capsnap,
496                      capsnap->context, capsnap->dirty_pages);
497                 if (!capsnap->dirty_pages)
498                         continue;
499 
500                 /* get i_size, truncate_{seq,size} for page_snapc? */
501                 if (snapc && capsnap->context != page_snapc)
502                         continue;
503 
504                 if (ctl) {
505                         if (capsnap->writing) {
506                                 ctl->i_size = i_size_read(inode);
507                                 ctl->size_stable = false;
508                         } else {
509                                 ctl->i_size = capsnap->size;
510                                 ctl->size_stable = true;
511                         }
512                         ctl->truncate_size = capsnap->truncate_size;
513                         ctl->truncate_seq = capsnap->truncate_seq;
514                         ctl->head_snapc = false;
515                 }
516 
517                 if (snapc)
518                         break;
519 
520                 snapc = ceph_get_snap_context(capsnap->context);
521                 if (!page_snapc ||
522                     page_snapc == snapc ||
523                     page_snapc->seq > snapc->seq)
524                         break;
525         }
526         if (!snapc && ci->i_wrbuffer_ref_head) {
527                 snapc = ceph_get_snap_context(ci->i_head_snapc);
528                 dout(" head snapc %p has %d dirty pages\n",
529                      snapc, ci->i_wrbuffer_ref_head);
530                 if (ctl) {
531                         ctl->i_size = i_size_read(inode);
532                         ctl->truncate_size = ci->i_truncate_size;
533                         ctl->truncate_seq = ci->i_truncate_seq;
534                         ctl->size_stable = false;
535                         ctl->head_snapc = true;
536                 }
537         }
538         spin_unlock(&ci->i_ceph_lock);
539         return snapc;
540 }
541 
542 static u64 get_writepages_data_length(struct inode *inode,
543                                       struct page *page, u64 start)
544 {
545         struct ceph_inode_info *ci = ceph_inode(inode);
546         struct ceph_snap_context *snapc = page_snap_context(page);
547         struct ceph_cap_snap *capsnap = NULL;
548         u64 end = i_size_read(inode);
549 
550         if (snapc != ci->i_head_snapc) {
551                 bool found = false;
552                 spin_lock(&ci->i_ceph_lock);
553                 list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
554                         if (capsnap->context == snapc) {
555                                 if (!capsnap->writing)
556                                         end = capsnap->size;
557                                 found = true;
558                                 break;
559                         }
560                 }
561                 spin_unlock(&ci->i_ceph_lock);
562                 WARN_ON(!found);
563         }
564         if (end > page_offset(page) + PAGE_SIZE)
565                 end = page_offset(page) + PAGE_SIZE;
566         return end > start ? end - start : 0;
567 }
568 
569 /*
570  * Write a single page, but leave the page locked.
571  *
572  * If we get a write error, set the page error bit, but still adjust the
573  * dirty page accounting (i.e., page is no longer dirty).
574  */
575 static int writepage_nounlock(struct page *page, struct writeback_control *wbc)
576 {
577         struct inode *inode;
578         struct ceph_inode_info *ci;
579         struct ceph_fs_client *fsc;
580         struct ceph_snap_context *snapc, *oldest;
581         loff_t page_off = page_offset(page);
582         int err, len = PAGE_SIZE;
583         struct ceph_writeback_ctl ceph_wbc;
584 
585         dout("writepage %p idx %lu\n", page, page->index);
586 
587         inode = page->mapping->host;
588         ci = ceph_inode(inode);
589         fsc = ceph_inode_to_client(inode);
590 
591         /* verify this is a writeable snap context */
592         snapc = page_snap_context(page);
593         if (!snapc) {
594                 dout("writepage %p page %p not dirty?\n", inode, page);
595                 return 0;
596         }
597         oldest = get_oldest_context(inode, &ceph_wbc, snapc);
598         if (snapc->seq > oldest->seq) {
599                 dout("writepage %p page %p snapc %p not writeable - noop\n",
600                      inode, page, snapc);
601                 /* we should only noop if called by kswapd */
602                 WARN_ON(!(current->flags & PF_MEMALLOC));
603                 ceph_put_snap_context(oldest);
604                 redirty_page_for_writepage(wbc, page);
605                 return 0;
606         }
607         ceph_put_snap_context(oldest);
608 
609         /* is this a partial page at end of file? */
610         if (page_off >= ceph_wbc.i_size) {
611                 dout("%p page eof %llu\n", page, ceph_wbc.i_size);
612                 page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE);
613                 return 0;
614         }
615 
616         if (ceph_wbc.i_size < page_off + len)
617                 len = ceph_wbc.i_size - page_off;
618 
619         dout("writepage %p page %p index %lu on %llu~%u snapc %p seq %lld\n",
620              inode, page, page->index, page_off, len, snapc, snapc->seq);
621 
622         if (atomic_long_inc_return(&fsc->writeback_count) >
623             CONGESTION_ON_THRESH(fsc->mount_options->congestion_kb))
624                 set_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC);
625 
626         set_page_writeback(page);
627         err = ceph_osdc_writepages(&fsc->client->osdc, ceph_vino(inode),
628                                    &ci->i_layout, snapc, page_off, len,
629                                    ceph_wbc.truncate_seq,
630                                    ceph_wbc.truncate_size,
631                                    &inode->i_mtime, &page, 1);
632         if (err < 0) {
633                 struct writeback_control tmp_wbc;
634                 if (!wbc)
635                         wbc = &tmp_wbc;
636                 if (err == -ERESTARTSYS) {
637                         /* killed by SIGKILL */
638                         dout("writepage interrupted page %p\n", page);
639                         redirty_page_for_writepage(wbc, page);
640                         end_page_writeback(page);
641                         return err;
642                 }
643                 dout("writepage setting page/mapping error %d %p\n",
644                      err, page);
645                 SetPageError(page);
646                 mapping_set_error(&inode->i_data, err);
647                 wbc->pages_skipped++;
648         } else {
649                 dout("writepage cleaned page %p\n", page);
650                 err = 0;  /* vfs expects us to return 0 */
651         }
652         page->private = 0;
653         ClearPagePrivate(page);
654         end_page_writeback(page);
655         ceph_put_wrbuffer_cap_refs(ci, 1, snapc);
656         ceph_put_snap_context(snapc);  /* page's reference */
657 
658         if (atomic_long_dec_return(&fsc->writeback_count) <
659             CONGESTION_OFF_THRESH(fsc->mount_options->congestion_kb))
660                 clear_bdi_congested(inode_to_bdi(inode), BLK_RW_ASYNC);
661 
662         return err;
663 }
664 
665 static int ceph_writepage(struct page *page, struct writeback_control *wbc)
666 {
667         int err;
668         struct inode *inode = page->mapping->host;
669         BUG_ON(!inode);
670         ihold(inode);
671         err = writepage_nounlock(page, wbc);
672         if (err == -ERESTARTSYS) {
673                 /* direct memory reclaimer was killed by SIGKILL. return 0
674                  * to prevent caller from setting mapping/page error */
675                 err = 0;
676         }
677         unlock_page(page);
678         iput(inode);
679         return err;
680 }
681 
682 /*
683  * lame release_pages helper.  release_pages() isn't exported to
684  * modules.
685  */
686 static void ceph_release_pages(struct page **pages, int num)
687 {
688         struct pagevec pvec;
689         int i;
690 
691         pagevec_init(&pvec);
692         for (i = 0; i < num; i++) {
693                 if (pagevec_add(&pvec, pages[i]) == 0)
694                         pagevec_release(&pvec);
695         }
696         pagevec_release(&pvec);
697 }
698 
699 /*
700  * async writeback completion handler.
701  *
702  * If we get an error, set the mapping error bit, but not the individual
703  * page error bits.
704  */
705 static void writepages_finish(struct ceph_osd_request *req)
706 {
707         struct inode *inode = req->r_inode;
708         struct ceph_inode_info *ci = ceph_inode(inode);
709         struct ceph_osd_data *osd_data;
710         struct page *page;
711         int num_pages, total_pages = 0;
712         int i, j;
713         int rc = req->r_result;
714         struct ceph_snap_context *snapc = req->r_snapc;
715         struct address_space *mapping = inode->i_mapping;
716         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
717         bool remove_page;
718 
719         dout("writepages_finish %p rc %d\n", inode, rc);
720         if (rc < 0) {
721                 mapping_set_error(mapping, rc);
722                 ceph_set_error_write(ci);
723         } else {
724                 ceph_clear_error_write(ci);
725         }
726 
727         /*
728          * We lost the cache cap, need to truncate the page before
729          * it is unlocked, otherwise we'd truncate it later in the
730          * page truncation thread, possibly losing some data that
731          * raced its way in
732          */
733         remove_page = !(ceph_caps_issued(ci) &
734                         (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO));
735 
736         /* clean all pages */
737         for (i = 0; i < req->r_num_ops; i++) {
738                 if (req->r_ops[i].op != CEPH_OSD_OP_WRITE)
739                         break;
740 
741                 osd_data = osd_req_op_extent_osd_data(req, i);
742                 BUG_ON(osd_data->type != CEPH_OSD_DATA_TYPE_PAGES);
743                 num_pages = calc_pages_for((u64)osd_data->alignment,
744                                            (u64)osd_data->length);
745                 total_pages += num_pages;
746                 for (j = 0; j < num_pages; j++) {
747                         page = osd_data->pages[j];
748                         BUG_ON(!page);
749                         WARN_ON(!PageUptodate(page));
750 
751                         if (atomic_long_dec_return(&fsc->writeback_count) <
752                              CONGESTION_OFF_THRESH(
753                                         fsc->mount_options->congestion_kb))
754                                 clear_bdi_congested(inode_to_bdi(inode),
755                                                     BLK_RW_ASYNC);
756 
757                         ceph_put_snap_context(page_snap_context(page));
758                         page->private = 0;
759                         ClearPagePrivate(page);
760                         dout("unlocking %p\n", page);
761                         end_page_writeback(page);
762 
763                         if (remove_page)
764                                 generic_error_remove_page(inode->i_mapping,
765                                                           page);
766 
767                         unlock_page(page);
768                 }
769                 dout("writepages_finish %p wrote %llu bytes cleaned %d pages\n",
770                      inode, osd_data->length, rc >= 0 ? num_pages : 0);
771 
772                 ceph_release_pages(osd_data->pages, num_pages);
773         }
774 
775         ceph_put_wrbuffer_cap_refs(ci, total_pages, snapc);
776 
777         osd_data = osd_req_op_extent_osd_data(req, 0);
778         if (osd_data->pages_from_pool)
779                 mempool_free(osd_data->pages,
780                              ceph_sb_to_client(inode->i_sb)->wb_pagevec_pool);
781         else
782                 kfree(osd_data->pages);
783         ceph_osdc_put_request(req);
784 }
785 
786 /*
787  * initiate async writeback
788  */
789 static int ceph_writepages_start(struct address_space *mapping,
790                                  struct writeback_control *wbc)
791 {
792         struct inode *inode = mapping->host;
793         struct ceph_inode_info *ci = ceph_inode(inode);
794         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
795         struct ceph_vino vino = ceph_vino(inode);
796         pgoff_t index, start_index, end = -1;
797         struct ceph_snap_context *snapc = NULL, *last_snapc = NULL, *pgsnapc;
798         struct pagevec pvec;
799         int rc = 0;
800         unsigned int wsize = i_blocksize(inode);
801         struct ceph_osd_request *req = NULL;
802         struct ceph_writeback_ctl ceph_wbc;
803         bool should_loop, range_whole = false;
804         bool done = false;
805 
806         dout("writepages_start %p (mode=%s)\n", inode,
807              wbc->sync_mode == WB_SYNC_NONE ? "NONE" :
808              (wbc->sync_mode == WB_SYNC_ALL ? "ALL" : "HOLD"));
809 
810         if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
811                 if (ci->i_wrbuffer_ref > 0) {
812                         pr_warn_ratelimited(
813                                 "writepage_start %p %lld forced umount\n",
814                                 inode, ceph_ino(inode));
815                 }
816                 mapping_set_error(mapping, -EIO);
817                 return -EIO; /* we're in a forced umount, don't write! */
818         }
819         if (fsc->mount_options->wsize < wsize)
820                 wsize = fsc->mount_options->wsize;
821 
822         pagevec_init(&pvec);
823 
824         start_index = wbc->range_cyclic ? mapping->writeback_index : 0;
825         index = start_index;
826 
827 retry:
828         /* find oldest snap context with dirty data */
829         snapc = get_oldest_context(inode, &ceph_wbc, NULL);
830         if (!snapc) {
831                 /* hmm, why does writepages get called when there
832                    is no dirty data? */
833                 dout(" no snap context with dirty data?\n");
834                 goto out;
835         }
836         dout(" oldest snapc is %p seq %lld (%d snaps)\n",
837              snapc, snapc->seq, snapc->num_snaps);
838 
839         should_loop = false;
840         if (ceph_wbc.head_snapc && snapc != last_snapc) {
841                 /* where to start/end? */
842                 if (wbc->range_cyclic) {
843                         index = start_index;
844                         end = -1;
845                         if (index > 0)
846                                 should_loop = true;
847                         dout(" cyclic, start at %lu\n", index);
848                 } else {
849                         index = wbc->range_start >> PAGE_SHIFT;
850                         end = wbc->range_end >> PAGE_SHIFT;
851                         if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX)
852                                 range_whole = true;
853                         dout(" not cyclic, %lu to %lu\n", index, end);
854                 }
855         } else if (!ceph_wbc.head_snapc) {
856                 /* Do not respect wbc->range_{start,end}. Dirty pages
857                  * in that range can be associated with newer snapc.
858                  * They are not writeable until we write all dirty pages
859                  * associated with 'snapc' get written */
860                 if (index > 0)
861                         should_loop = true;
862                 dout(" non-head snapc, range whole\n");
863         }
864 
865         ceph_put_snap_context(last_snapc);
866         last_snapc = snapc;
867 
868         while (!done && index <= end) {
869                 int num_ops = 0, op_idx;
870                 unsigned i, pvec_pages, max_pages, locked_pages = 0;
871                 struct page **pages = NULL, **data_pages;
872                 mempool_t *pool = NULL; /* Becomes non-null if mempool used */
873                 struct page *page;
874                 pgoff_t strip_unit_end = 0;
875                 u64 offset = 0, len = 0;
876 
877                 max_pages = wsize >> PAGE_SHIFT;
878 
879 get_more_pages:
880                 pvec_pages = pagevec_lookup_range_nr_tag(&pvec, mapping, &index,
881                                                 end, PAGECACHE_TAG_DIRTY,
882                                                 max_pages - locked_pages);
883                 dout("pagevec_lookup_range_tag got %d\n", pvec_pages);
884                 if (!pvec_pages && !locked_pages)
885                         break;
886                 for (i = 0; i < pvec_pages && locked_pages < max_pages; i++) {
887                         page = pvec.pages[i];
888                         dout("? %p idx %lu\n", page, page->index);
889                         if (locked_pages == 0)
890                                 lock_page(page);  /* first page */
891                         else if (!trylock_page(page))
892                                 break;
893 
894                         /* only dirty pages, or our accounting breaks */
895                         if (unlikely(!PageDirty(page)) ||
896                             unlikely(page->mapping != mapping)) {
897                                 dout("!dirty or !mapping %p\n", page);
898                                 unlock_page(page);
899                                 continue;
900                         }
901                         /* only if matching snap context */
902                         pgsnapc = page_snap_context(page);
903                         if (pgsnapc != snapc) {
904                                 dout("page snapc %p %lld != oldest %p %lld\n",
905                                      pgsnapc, pgsnapc->seq, snapc, snapc->seq);
906                                 if (!should_loop &&
907                                     !ceph_wbc.head_snapc &&
908                                     wbc->sync_mode != WB_SYNC_NONE)
909                                         should_loop = true;
910                                 unlock_page(page);
911                                 continue;
912                         }
913                         if (page_offset(page) >= ceph_wbc.i_size) {
914                                 dout("%p page eof %llu\n",
915                                      page, ceph_wbc.i_size);
916                                 if (ceph_wbc.size_stable ||
917                                     page_offset(page) >= i_size_read(inode))
918                                         mapping->a_ops->invalidatepage(page,
919                                                                 0, PAGE_SIZE);
920                                 unlock_page(page);
921                                 continue;
922                         }
923                         if (strip_unit_end && (page->index > strip_unit_end)) {
924                                 dout("end of strip unit %p\n", page);
925                                 unlock_page(page);
926                                 break;
927                         }
928                         if (PageWriteback(page)) {
929                                 if (wbc->sync_mode == WB_SYNC_NONE) {
930                                         dout("%p under writeback\n", page);
931                                         unlock_page(page);
932                                         continue;
933                                 }
934                                 dout("waiting on writeback %p\n", page);
935                                 wait_on_page_writeback(page);
936                         }
937 
938                         if (!clear_page_dirty_for_io(page)) {
939                                 dout("%p !clear_page_dirty_for_io\n", page);
940                                 unlock_page(page);
941                                 continue;
942                         }
943 
944                         /*
945                          * We have something to write.  If this is
946                          * the first locked page this time through,
947                          * calculate max possinle write size and
948                          * allocate a page array
949                          */
950                         if (locked_pages == 0) {
951                                 u64 objnum;
952                                 u64 objoff;
953                                 u32 xlen;
954 
955                                 /* prepare async write request */
956                                 offset = (u64)page_offset(page);
957                                 ceph_calc_file_object_mapping(&ci->i_layout,
958                                                               offset, wsize,
959                                                               &objnum, &objoff,
960                                                               &xlen);
961                                 len = xlen;
962 
963                                 num_ops = 1;
964                                 strip_unit_end = page->index +
965                                         ((len - 1) >> PAGE_SHIFT);
966 
967                                 BUG_ON(pages);
968                                 max_pages = calc_pages_for(0, (u64)len);
969                                 pages = kmalloc_array(max_pages,
970                                                       sizeof(*pages),
971                                                       GFP_NOFS);
972                                 if (!pages) {
973                                         pool = fsc->wb_pagevec_pool;
974                                         pages = mempool_alloc(pool, GFP_NOFS);
975                                         BUG_ON(!pages);
976                                 }
977 
978                                 len = 0;
979                         } else if (page->index !=
980                                    (offset + len) >> PAGE_SHIFT) {
981                                 if (num_ops >= (pool ?  CEPH_OSD_SLAB_OPS :
982                                                         CEPH_OSD_MAX_OPS)) {
983                                         redirty_page_for_writepage(wbc, page);
984                                         unlock_page(page);
985                                         break;
986                                 }
987 
988                                 num_ops++;
989                                 offset = (u64)page_offset(page);
990                                 len = 0;
991                         }
992 
993                         /* note position of first page in pvec */
994                         dout("%p will write page %p idx %lu\n",
995                              inode, page, page->index);
996 
997                         if (atomic_long_inc_return(&fsc->writeback_count) >
998                             CONGESTION_ON_THRESH(
999                                     fsc->mount_options->congestion_kb)) {
1000                                 set_bdi_congested(inode_to_bdi(inode),
1001                                                   BLK_RW_ASYNC);
1002                         }
1003 
1004 
1005                         pages[locked_pages++] = page;
1006                         pvec.pages[i] = NULL;
1007 
1008                         len += PAGE_SIZE;
1009                 }
1010 
1011                 /* did we get anything? */
1012                 if (!locked_pages)
1013                         goto release_pvec_pages;
1014                 if (i) {
1015                         unsigned j, n = 0;
1016                         /* shift unused page to beginning of pvec */
1017                         for (j = 0; j < pvec_pages; j++) {
1018                                 if (!pvec.pages[j])
1019                                         continue;
1020                                 if (n < j)
1021                                         pvec.pages[n] = pvec.pages[j];
1022                                 n++;
1023                         }
1024                         pvec.nr = n;
1025 
1026                         if (pvec_pages && i == pvec_pages &&
1027                             locked_pages < max_pages) {
1028                                 dout("reached end pvec, trying for more\n");
1029                                 pagevec_release(&pvec);
1030                                 goto get_more_pages;
1031                         }
1032                 }
1033 
1034 new_request:
1035                 offset = page_offset(pages[0]);
1036                 len = wsize;
1037 
1038                 req = ceph_osdc_new_request(&fsc->client->osdc,
1039                                         &ci->i_layout, vino,
1040                                         offset, &len, 0, num_ops,
1041                                         CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
1042                                         snapc, ceph_wbc.truncate_seq,
1043                                         ceph_wbc.truncate_size, false);
1044                 if (IS_ERR(req)) {
1045                         req = ceph_osdc_new_request(&fsc->client->osdc,
1046                                                 &ci->i_layout, vino,
1047                                                 offset, &len, 0,
1048                                                 min(num_ops,
1049                                                     CEPH_OSD_SLAB_OPS),
1050                                                 CEPH_OSD_OP_WRITE,
1051                                                 CEPH_OSD_FLAG_WRITE,
1052                                                 snapc, ceph_wbc.truncate_seq,
1053                                                 ceph_wbc.truncate_size, true);
1054                         BUG_ON(IS_ERR(req));
1055                 }
1056                 BUG_ON(len < page_offset(pages[locked_pages - 1]) +
1057                              PAGE_SIZE - offset);
1058 
1059                 req->r_callback = writepages_finish;
1060                 req->r_inode = inode;
1061 
1062                 /* Format the osd request message and submit the write */
1063                 len = 0;
1064                 data_pages = pages;
1065                 op_idx = 0;
1066                 for (i = 0; i < locked_pages; i++) {
1067                         u64 cur_offset = page_offset(pages[i]);
1068                         if (offset + len != cur_offset) {
1069                                 if (op_idx + 1 == req->r_num_ops)
1070                                         break;
1071                                 osd_req_op_extent_dup_last(req, op_idx,
1072                                                            cur_offset - offset);
1073                                 dout("writepages got pages at %llu~%llu\n",
1074                                      offset, len);
1075                                 osd_req_op_extent_osd_data_pages(req, op_idx,
1076                                                         data_pages, len, 0,
1077                                                         !!pool, false);
1078                                 osd_req_op_extent_update(req, op_idx, len);
1079 
1080                                 len = 0;
1081                                 offset = cur_offset; 
1082                                 data_pages = pages + i;
1083                                 op_idx++;
1084                         }
1085 
1086                         set_page_writeback(pages[i]);
1087                         len += PAGE_SIZE;
1088                 }
1089 
1090                 if (ceph_wbc.size_stable) {
1091                         len = min(len, ceph_wbc.i_size - offset);
1092                 } else if (i == locked_pages) {
1093                         /* writepages_finish() clears writeback pages
1094                          * according to the data length, so make sure
1095                          * data length covers all locked pages */
1096                         u64 min_len = len + 1 - PAGE_SIZE;
1097                         len = get_writepages_data_length(inode, pages[i - 1],
1098                                                          offset);
1099                         len = max(len, min_len);
1100                 }
1101                 dout("writepages got pages at %llu~%llu\n", offset, len);
1102 
1103                 osd_req_op_extent_osd_data_pages(req, op_idx, data_pages, len,
1104                                                  0, !!pool, false);
1105                 osd_req_op_extent_update(req, op_idx, len);
1106 
1107                 BUG_ON(op_idx + 1 != req->r_num_ops);
1108 
1109                 pool = NULL;
1110                 if (i < locked_pages) {
1111                         BUG_ON(num_ops <= req->r_num_ops);
1112                         num_ops -= req->r_num_ops;
1113                         locked_pages -= i;
1114 
1115                         /* allocate new pages array for next request */
1116                         data_pages = pages;
1117                         pages = kmalloc_array(locked_pages, sizeof(*pages),
1118                                               GFP_NOFS);
1119                         if (!pages) {
1120                                 pool = fsc->wb_pagevec_pool;
1121                                 pages = mempool_alloc(pool, GFP_NOFS);
1122                                 BUG_ON(!pages);
1123                         }
1124                         memcpy(pages, data_pages + i,
1125                                locked_pages * sizeof(*pages));
1126                         memset(data_pages + i, 0,
1127                                locked_pages * sizeof(*pages));
1128                 } else {
1129                         BUG_ON(num_ops != req->r_num_ops);
1130                         index = pages[i - 1]->index + 1;
1131                         /* request message now owns the pages array */
1132                         pages = NULL;
1133                 }
1134 
1135                 req->r_mtime = inode->i_mtime;
1136                 rc = ceph_osdc_start_request(&fsc->client->osdc, req, true);
1137                 BUG_ON(rc);
1138                 req = NULL;
1139 
1140                 wbc->nr_to_write -= i;
1141                 if (pages)
1142                         goto new_request;
1143 
1144                 /*
1145                  * We stop writing back only if we are not doing
1146                  * integrity sync. In case of integrity sync we have to
1147                  * keep going until we have written all the pages
1148                  * we tagged for writeback prior to entering this loop.
1149                  */
1150                 if (wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE)
1151                         done = true;
1152 
1153 release_pvec_pages:
1154                 dout("pagevec_release on %d pages (%p)\n", (int)pvec.nr,
1155                      pvec.nr ? pvec.pages[0] : NULL);
1156                 pagevec_release(&pvec);
1157         }
1158 
1159         if (should_loop && !done) {
1160                 /* more to do; loop back to beginning of file */
1161                 dout("writepages looping back to beginning of file\n");
1162                 end = start_index - 1; /* OK even when start_index == 0 */
1163 
1164                 /* to write dirty pages associated with next snapc,
1165                  * we need to wait until current writes complete */
1166                 if (wbc->sync_mode != WB_SYNC_NONE &&
1167                     start_index == 0 && /* all dirty pages were checked */
1168                     !ceph_wbc.head_snapc) {
1169                         struct page *page;
1170                         unsigned i, nr;
1171                         index = 0;
1172                         while ((index <= end) &&
1173                                (nr = pagevec_lookup_tag(&pvec, mapping, &index,
1174                                                 PAGECACHE_TAG_WRITEBACK))) {
1175                                 for (i = 0; i < nr; i++) {
1176                                         page = pvec.pages[i];
1177                                         if (page_snap_context(page) != snapc)
1178                                                 continue;
1179                                         wait_on_page_writeback(page);
1180                                 }
1181                                 pagevec_release(&pvec);
1182                                 cond_resched();
1183                         }
1184                 }
1185 
1186                 start_index = 0;
1187                 index = 0;
1188                 goto retry;
1189         }
1190 
1191         if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0))
1192                 mapping->writeback_index = index;
1193 
1194 out:
1195         ceph_osdc_put_request(req);
1196         ceph_put_snap_context(last_snapc);
1197         dout("writepages dend - startone, rc = %d\n", rc);
1198         return rc;
1199 }
1200 
1201 
1202 
1203 /*
1204  * See if a given @snapc is either writeable, or already written.
1205  */
1206 static int context_is_writeable_or_written(struct inode *inode,
1207                                            struct ceph_snap_context *snapc)
1208 {
1209         struct ceph_snap_context *oldest = get_oldest_context(inode, NULL, NULL);
1210         int ret = !oldest || snapc->seq <= oldest->seq;
1211 
1212         ceph_put_snap_context(oldest);
1213         return ret;
1214 }
1215 
1216 /*
1217  * We are only allowed to write into/dirty the page if the page is
1218  * clean, or already dirty within the same snap context.
1219  *
1220  * called with page locked.
1221  * return success with page locked,
1222  * or any failure (incl -EAGAIN) with page unlocked.
1223  */
1224 static int ceph_update_writeable_page(struct file *file,
1225                             loff_t pos, unsigned len,
1226                             struct page *page)
1227 {
1228         struct inode *inode = file_inode(file);
1229         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1230         struct ceph_inode_info *ci = ceph_inode(inode);
1231         loff_t page_off = pos & PAGE_MASK;
1232         int pos_in_page = pos & ~PAGE_MASK;
1233         int end_in_page = pos_in_page + len;
1234         loff_t i_size;
1235         int r;
1236         struct ceph_snap_context *snapc, *oldest;
1237 
1238         if (READ_ONCE(fsc->mount_state) == CEPH_MOUNT_SHUTDOWN) {
1239                 dout(" page %p forced umount\n", page);
1240                 unlock_page(page);
1241                 return -EIO;
1242         }
1243 
1244 retry_locked:
1245         /* writepages currently holds page lock, but if we change that later, */
1246         wait_on_page_writeback(page);
1247 
1248         snapc = page_snap_context(page);
1249         if (snapc && snapc != ci->i_head_snapc) {
1250                 /*
1251                  * this page is already dirty in another (older) snap
1252                  * context!  is it writeable now?
1253                  */
1254                 oldest = get_oldest_context(inode, NULL, NULL);
1255                 if (snapc->seq > oldest->seq) {
1256                         ceph_put_snap_context(oldest);
1257                         dout(" page %p snapc %p not current or oldest\n",
1258                              page, snapc);
1259                         /*
1260                          * queue for writeback, and wait for snapc to
1261                          * be writeable or written
1262                          */
1263                         snapc = ceph_get_snap_context(snapc);
1264                         unlock_page(page);
1265                         ceph_queue_writeback(inode);
1266                         r = wait_event_killable(ci->i_cap_wq,
1267                                context_is_writeable_or_written(inode, snapc));
1268                         ceph_put_snap_context(snapc);
1269                         if (r == -ERESTARTSYS)
1270                                 return r;
1271                         return -EAGAIN;
1272                 }
1273                 ceph_put_snap_context(oldest);
1274 
1275                 /* yay, writeable, do it now (without dropping page lock) */
1276                 dout(" page %p snapc %p not current, but oldest\n",
1277                      page, snapc);
1278                 if (!clear_page_dirty_for_io(page))
1279                         goto retry_locked;
1280                 r = writepage_nounlock(page, NULL);
1281                 if (r < 0)
1282                         goto fail_unlock;
1283                 goto retry_locked;
1284         }
1285 
1286         if (PageUptodate(page)) {
1287                 dout(" page %p already uptodate\n", page);
1288                 return 0;
1289         }
1290 
1291         /* full page? */
1292         if (pos_in_page == 0 && len == PAGE_SIZE)
1293                 return 0;
1294 
1295         /* past end of file? */
1296         i_size = i_size_read(inode);
1297 
1298         if (page_off >= i_size ||
1299             (pos_in_page == 0 && (pos+len) >= i_size &&
1300              end_in_page - pos_in_page != PAGE_SIZE)) {
1301                 dout(" zeroing %p 0 - %d and %d - %d\n",
1302                      page, pos_in_page, end_in_page, (int)PAGE_SIZE);
1303                 zero_user_segments(page,
1304                                    0, pos_in_page,
1305                                    end_in_page, PAGE_SIZE);
1306                 return 0;
1307         }
1308 
1309         /* we need to read it. */
1310         r = ceph_do_readpage(file, page);
1311         if (r < 0) {
1312                 if (r == -EINPROGRESS)
1313                         return -EAGAIN;
1314                 goto fail_unlock;
1315         }
1316         goto retry_locked;
1317 fail_unlock:
1318         unlock_page(page);
1319         return r;
1320 }
1321 
1322 /*
1323  * We are only allowed to write into/dirty the page if the page is
1324  * clean, or already dirty within the same snap context.
1325  */
1326 static int ceph_write_begin(struct file *file, struct address_space *mapping,
1327                             loff_t pos, unsigned len, unsigned flags,
1328                             struct page **pagep, void **fsdata)
1329 {
1330         struct inode *inode = file_inode(file);
1331         struct page *page;
1332         pgoff_t index = pos >> PAGE_SHIFT;
1333         int r;
1334 
1335         do {
1336                 /* get a page */
1337                 page = grab_cache_page_write_begin(mapping, index, 0);
1338                 if (!page)
1339                         return -ENOMEM;
1340 
1341                 dout("write_begin file %p inode %p page %p %d~%d\n", file,
1342                      inode, page, (int)pos, (int)len);
1343 
1344                 r = ceph_update_writeable_page(file, pos, len, page);
1345                 if (r < 0)
1346                         put_page(page);
1347                 else
1348                         *pagep = page;
1349         } while (r == -EAGAIN);
1350 
1351         return r;
1352 }
1353 
1354 /*
1355  * we don't do anything in here that simple_write_end doesn't do
1356  * except adjust dirty page accounting
1357  */
1358 static int ceph_write_end(struct file *file, struct address_space *mapping,
1359                           loff_t pos, unsigned len, unsigned copied,
1360                           struct page *page, void *fsdata)
1361 {
1362         struct inode *inode = file_inode(file);
1363         bool check_cap = false;
1364 
1365         dout("write_end file %p inode %p page %p %d~%d (%d)\n", file,
1366              inode, page, (int)pos, (int)copied, (int)len);
1367 
1368         /* zero the stale part of the page if we did a short copy */
1369         if (!PageUptodate(page)) {
1370                 if (copied < len) {
1371                         copied = 0;
1372                         goto out;
1373                 }
1374                 SetPageUptodate(page);
1375         }
1376 
1377         /* did file size increase? */
1378         if (pos+copied > i_size_read(inode))
1379                 check_cap = ceph_inode_set_size(inode, pos+copied);
1380 
1381         set_page_dirty(page);
1382 
1383 out:
1384         unlock_page(page);
1385         put_page(page);
1386 
1387         if (check_cap)
1388                 ceph_check_caps(ceph_inode(inode), CHECK_CAPS_AUTHONLY, NULL);
1389 
1390         return copied;
1391 }
1392 
1393 /*
1394  * we set .direct_IO to indicate direct io is supported, but since we
1395  * intercept O_DIRECT reads and writes early, this function should
1396  * never get called.
1397  */
1398 static ssize_t ceph_direct_io(struct kiocb *iocb, struct iov_iter *iter)
1399 {
1400         WARN_ON(1);
1401         return -EINVAL;
1402 }
1403 
1404 const struct address_space_operations ceph_aops = {
1405         .readpage = ceph_readpage,
1406         .readpages = ceph_readpages,
1407         .writepage = ceph_writepage,
1408         .writepages = ceph_writepages_start,
1409         .write_begin = ceph_write_begin,
1410         .write_end = ceph_write_end,
1411         .set_page_dirty = ceph_set_page_dirty,
1412         .invalidatepage = ceph_invalidatepage,
1413         .releasepage = ceph_releasepage,
1414         .direct_IO = ceph_direct_io,
1415 };
1416 
1417 static void ceph_block_sigs(sigset_t *oldset)
1418 {
1419         sigset_t mask;
1420         siginitsetinv(&mask, sigmask(SIGKILL));
1421         sigprocmask(SIG_BLOCK, &mask, oldset);
1422 }
1423 
1424 static void ceph_restore_sigs(sigset_t *oldset)
1425 {
1426         sigprocmask(SIG_SETMASK, oldset, NULL);
1427 }
1428 
1429 /*
1430  * vm ops
1431  */
1432 static vm_fault_t ceph_filemap_fault(struct vm_fault *vmf)
1433 {
1434         struct vm_area_struct *vma = vmf->vma;
1435         struct inode *inode = file_inode(vma->vm_file);
1436         struct ceph_inode_info *ci = ceph_inode(inode);
1437         struct ceph_file_info *fi = vma->vm_file->private_data;
1438         struct page *pinned_page = NULL;
1439         loff_t off = vmf->pgoff << PAGE_SHIFT;
1440         int want, got, err;
1441         sigset_t oldset;
1442         vm_fault_t ret = VM_FAULT_SIGBUS;
1443 
1444         ceph_block_sigs(&oldset);
1445 
1446         dout("filemap_fault %p %llx.%llx %llu~%zd trying to get caps\n",
1447              inode, ceph_vinop(inode), off, (size_t)PAGE_SIZE);
1448         if (fi->fmode & CEPH_FILE_MODE_LAZY)
1449                 want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO;
1450         else
1451                 want = CEPH_CAP_FILE_CACHE;
1452 
1453         got = 0;
1454         err = ceph_get_caps(ci, CEPH_CAP_FILE_RD, want, -1, &got, &pinned_page);
1455         if (err < 0)
1456                 goto out_restore;
1457 
1458         dout("filemap_fault %p %llu~%zd got cap refs on %s\n",
1459              inode, off, (size_t)PAGE_SIZE, ceph_cap_string(got));
1460 
1461         if ((got & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) ||
1462             ci->i_inline_version == CEPH_INLINE_NONE) {
1463                 CEPH_DEFINE_RW_CONTEXT(rw_ctx, got);
1464                 ceph_add_rw_context(fi, &rw_ctx);
1465                 ret = filemap_fault(vmf);
1466                 ceph_del_rw_context(fi, &rw_ctx);
1467                 dout("filemap_fault %p %llu~%zd drop cap refs %s ret %x\n",
1468                         inode, off, (size_t)PAGE_SIZE,
1469                                 ceph_cap_string(got), ret);
1470         } else
1471                 err = -EAGAIN;
1472 
1473         if (pinned_page)
1474                 put_page(pinned_page);
1475         ceph_put_cap_refs(ci, got);
1476 
1477         if (err != -EAGAIN)
1478                 goto out_restore;
1479 
1480         /* read inline data */
1481         if (off >= PAGE_SIZE) {
1482                 /* does not support inline data > PAGE_SIZE */
1483                 ret = VM_FAULT_SIGBUS;
1484         } else {
1485                 struct address_space *mapping = inode->i_mapping;
1486                 struct page *page = find_or_create_page(mapping, 0,
1487                                                 mapping_gfp_constraint(mapping,
1488                                                 ~__GFP_FS));
1489                 if (!page) {
1490                         ret = VM_FAULT_OOM;
1491                         goto out_inline;
1492                 }
1493                 err = __ceph_do_getattr(inode, page,
1494                                          CEPH_STAT_CAP_INLINE_DATA, true);
1495                 if (err < 0 || off >= i_size_read(inode)) {
1496                         unlock_page(page);
1497                         put_page(page);
1498                         if (err == -ENOMEM)
1499                                 ret = VM_FAULT_OOM;
1500                         else
1501                                 ret = VM_FAULT_SIGBUS;
1502                         goto out_inline;
1503                 }
1504                 if (err < PAGE_SIZE)
1505                         zero_user_segment(page, err, PAGE_SIZE);
1506                 else
1507                         flush_dcache_page(page);
1508                 SetPageUptodate(page);
1509                 vmf->page = page;
1510                 ret = VM_FAULT_MAJOR | VM_FAULT_LOCKED;
1511 out_inline:
1512                 dout("filemap_fault %p %llu~%zd read inline data ret %x\n",
1513                      inode, off, (size_t)PAGE_SIZE, ret);
1514         }
1515 out_restore:
1516         ceph_restore_sigs(&oldset);
1517         if (err < 0)
1518                 ret = vmf_error(err);
1519 
1520         return ret;
1521 }
1522 
1523 /*
1524  * Reuse write_begin here for simplicity.
1525  */
1526 static vm_fault_t ceph_page_mkwrite(struct vm_fault *vmf)
1527 {
1528         struct vm_area_struct *vma = vmf->vma;
1529         struct inode *inode = file_inode(vma->vm_file);
1530         struct ceph_inode_info *ci = ceph_inode(inode);
1531         struct ceph_file_info *fi = vma->vm_file->private_data;
1532         struct ceph_cap_flush *prealloc_cf;
1533         struct page *page = vmf->page;
1534         loff_t off = page_offset(page);
1535         loff_t size = i_size_read(inode);
1536         size_t len;
1537         int want, got, err;
1538         sigset_t oldset;
1539         vm_fault_t ret = VM_FAULT_SIGBUS;
1540 
1541         prealloc_cf = ceph_alloc_cap_flush();
1542         if (!prealloc_cf)
1543                 return VM_FAULT_OOM;
1544 
1545         ceph_block_sigs(&oldset);
1546 
1547         if (ci->i_inline_version != CEPH_INLINE_NONE) {
1548                 struct page *locked_page = NULL;
1549                 if (off == 0) {
1550                         lock_page(page);
1551                         locked_page = page;
1552                 }
1553                 err = ceph_uninline_data(vma->vm_file, locked_page);
1554                 if (locked_page)
1555                         unlock_page(locked_page);
1556                 if (err < 0)
1557                         goto out_free;
1558         }
1559 
1560         if (off + PAGE_SIZE <= size)
1561                 len = PAGE_SIZE;
1562         else
1563                 len = size & ~PAGE_MASK;
1564 
1565         dout("page_mkwrite %p %llx.%llx %llu~%zd getting caps i_size %llu\n",
1566              inode, ceph_vinop(inode), off, len, size);
1567         if (fi->fmode & CEPH_FILE_MODE_LAZY)
1568                 want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO;
1569         else
1570                 want = CEPH_CAP_FILE_BUFFER;
1571 
1572         got = 0;
1573         err = ceph_get_caps(ci, CEPH_CAP_FILE_WR, want, off + len,
1574                             &got, NULL);
1575         if (err < 0)
1576                 goto out_free;
1577 
1578         dout("page_mkwrite %p %llu~%zd got cap refs on %s\n",
1579              inode, off, len, ceph_cap_string(got));
1580 
1581         /* Update time before taking page lock */
1582         file_update_time(vma->vm_file);
1583 
1584         do {
1585                 lock_page(page);
1586 
1587                 if ((off > size) || (page->mapping != inode->i_mapping)) {
1588                         unlock_page(page);
1589                         ret = VM_FAULT_NOPAGE;
1590                         break;
1591                 }
1592 
1593                 err = ceph_update_writeable_page(vma->vm_file, off, len, page);
1594                 if (err >= 0) {
1595                         /* success.  we'll keep the page locked. */
1596                         set_page_dirty(page);
1597                         ret = VM_FAULT_LOCKED;
1598                 }
1599         } while (err == -EAGAIN);
1600 
1601         if (ret == VM_FAULT_LOCKED ||
1602             ci->i_inline_version != CEPH_INLINE_NONE) {
1603                 int dirty;
1604                 spin_lock(&ci->i_ceph_lock);
1605                 ci->i_inline_version = CEPH_INLINE_NONE;
1606                 dirty = __ceph_mark_dirty_caps(ci, CEPH_CAP_FILE_WR,
1607                                                &prealloc_cf);
1608                 spin_unlock(&ci->i_ceph_lock);
1609                 if (dirty)
1610                         __mark_inode_dirty(inode, dirty);
1611         }
1612 
1613         dout("page_mkwrite %p %llu~%zd dropping cap refs on %s ret %x\n",
1614              inode, off, len, ceph_cap_string(got), ret);
1615         ceph_put_cap_refs(ci, got);
1616 out_free:
1617         ceph_restore_sigs(&oldset);
1618         ceph_free_cap_flush(prealloc_cf);
1619         if (err < 0)
1620                 ret = vmf_error(err);
1621         return ret;
1622 }
1623 
1624 void ceph_fill_inline_data(struct inode *inode, struct page *locked_page,
1625                            char *data, size_t len)
1626 {
1627         struct address_space *mapping = inode->i_mapping;
1628         struct page *page;
1629 
1630         if (locked_page) {
1631                 page = locked_page;
1632         } else {
1633                 if (i_size_read(inode) == 0)
1634                         return;
1635                 page = find_or_create_page(mapping, 0,
1636                                            mapping_gfp_constraint(mapping,
1637                                            ~__GFP_FS));
1638                 if (!page)
1639                         return;
1640                 if (PageUptodate(page)) {
1641                         unlock_page(page);
1642                         put_page(page);
1643                         return;
1644                 }
1645         }
1646 
1647         dout("fill_inline_data %p %llx.%llx len %zu locked_page %p\n",
1648              inode, ceph_vinop(inode), len, locked_page);
1649 
1650         if (len > 0) {
1651                 void *kaddr = kmap_atomic(page);
1652                 memcpy(kaddr, data, len);
1653                 kunmap_atomic(kaddr);
1654         }
1655 
1656         if (page != locked_page) {
1657                 if (len < PAGE_SIZE)
1658                         zero_user_segment(page, len, PAGE_SIZE);
1659                 else
1660                         flush_dcache_page(page);
1661 
1662                 SetPageUptodate(page);
1663                 unlock_page(page);
1664                 put_page(page);
1665         }
1666 }
1667 
1668 int ceph_uninline_data(struct file *filp, struct page *locked_page)
1669 {
1670         struct inode *inode = file_inode(filp);
1671         struct ceph_inode_info *ci = ceph_inode(inode);
1672         struct ceph_fs_client *fsc = ceph_inode_to_client(inode);
1673         struct ceph_osd_request *req;
1674         struct page *page = NULL;
1675         u64 len, inline_version;
1676         int err = 0;
1677         bool from_pagecache = false;
1678 
1679         spin_lock(&ci->i_ceph_lock);
1680         inline_version = ci->i_inline_version;
1681         spin_unlock(&ci->i_ceph_lock);
1682 
1683         dout("uninline_data %p %llx.%llx inline_version %llu\n",
1684              inode, ceph_vinop(inode), inline_version);
1685 
1686         if (inline_version == 1 || /* initial version, no data */
1687             inline_version == CEPH_INLINE_NONE)
1688                 goto out;
1689 
1690         if (locked_page) {
1691                 page = locked_page;
1692                 WARN_ON(!PageUptodate(page));
1693         } else if (ceph_caps_issued(ci) &
1694                    (CEPH_CAP_FILE_CACHE|CEPH_CAP_FILE_LAZYIO)) {
1695                 page = find_get_page(inode->i_mapping, 0);
1696                 if (page) {
1697                         if (PageUptodate(page)) {
1698                                 from_pagecache = true;
1699                                 lock_page(page);
1700                         } else {
1701                                 put_page(page);
1702                                 page = NULL;
1703                         }
1704                 }
1705         }
1706 
1707         if (page) {
1708                 len = i_size_read(inode);
1709                 if (len > PAGE_SIZE)
1710                         len = PAGE_SIZE;
1711         } else {
1712                 page = __page_cache_alloc(GFP_NOFS);
1713                 if (!page) {
1714                         err = -ENOMEM;
1715                         goto out;
1716                 }
1717                 err = __ceph_do_getattr(inode, page,
1718                                         CEPH_STAT_CAP_INLINE_DATA, true);
1719                 if (err < 0) {
1720                         /* no inline data */
1721                         if (err == -ENODATA)
1722                                 err = 0;
1723                         goto out;
1724                 }
1725                 len = err;
1726         }
1727 
1728         req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1729                                     ceph_vino(inode), 0, &len, 0, 1,
1730                                     CEPH_OSD_OP_CREATE, CEPH_OSD_FLAG_WRITE,
1731                                     NULL, 0, 0, false);
1732         if (IS_ERR(req)) {
1733                 err = PTR_ERR(req);
1734                 goto out;
1735         }
1736 
1737         req->r_mtime = inode->i_mtime;
1738         err = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1739         if (!err)
1740                 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1741         ceph_osdc_put_request(req);
1742         if (err < 0)
1743                 goto out;
1744 
1745         req = ceph_osdc_new_request(&fsc->client->osdc, &ci->i_layout,
1746                                     ceph_vino(inode), 0, &len, 1, 3,
1747                                     CEPH_OSD_OP_WRITE, CEPH_OSD_FLAG_WRITE,
1748                                     NULL, ci->i_truncate_seq,
1749                                     ci->i_truncate_size, false);
1750         if (IS_ERR(req)) {
1751                 err = PTR_ERR(req);
1752                 goto out;
1753         }
1754 
1755         osd_req_op_extent_osd_data_pages(req, 1, &page, len, 0, false, false);
1756 
1757         {
1758                 __le64 xattr_buf = cpu_to_le64(inline_version);
1759                 err = osd_req_op_xattr_init(req, 0, CEPH_OSD_OP_CMPXATTR,
1760                                             "inline_version", &xattr_buf,
1761                                             sizeof(xattr_buf),
1762                                             CEPH_OSD_CMPXATTR_OP_GT,
1763                                             CEPH_OSD_CMPXATTR_MODE_U64);
1764                 if (err)
1765                         goto out_put;
1766         }
1767 
1768         {
1769                 char xattr_buf[32];
1770                 int xattr_len = snprintf(xattr_buf, sizeof(xattr_buf),
1771                                          "%llu", inline_version);
1772                 err = osd_req_op_xattr_init(req, 2, CEPH_OSD_OP_SETXATTR,
1773                                             "inline_version",
1774                                             xattr_buf, xattr_len, 0, 0);
1775                 if (err)
1776                         goto out_put;
1777         }
1778 
1779         req->r_mtime = inode->i_mtime;
1780         err = ceph_osdc_start_request(&fsc->client->osdc, req, false);
1781         if (!err)
1782                 err = ceph_osdc_wait_request(&fsc->client->osdc, req);
1783 out_put:
1784         ceph_osdc_put_request(req);
1785         if (err == -ECANCELED)
1786                 err = 0;
1787 out:
1788         if (page && page != locked_page) {
1789                 if (from_pagecache) {
1790                         unlock_page(page);
1791                         put_page(page);
1792                 } else
1793                         __free_pages(page, 0);
1794         }
1795 
1796         dout("uninline_data %p %llx.%llx inline_version %llu = %d\n",
1797              inode, ceph_vinop(inode), inline_version, err);
1798         return err;
1799 }
1800 
1801 static const struct vm_operations_struct ceph_vmops = {
1802         .fault          = ceph_filemap_fault,
1803         .page_mkwrite   = ceph_page_mkwrite,
1804 };
1805 
1806 int ceph_mmap(struct file *file, struct vm_area_struct *vma)
1807 {
1808         struct address_space *mapping = file->f_mapping;
1809 
1810         if (!mapping->a_ops->readpage)
1811                 return -ENOEXEC;
1812         file_accessed(file);
1813         vma->vm_ops = &ceph_vmops;
1814         return 0;
1815 }
1816 
1817 enum {
1818         POOL_READ       = 1,
1819         POOL_WRITE      = 2,
1820 };
1821 
1822 static int __ceph_pool_perm_get(struct ceph_inode_info *ci,
1823                                 s64 pool, struct ceph_string *pool_ns)
1824 {
1825         struct ceph_fs_client *fsc = ceph_inode_to_client(&ci->vfs_inode);
1826         struct ceph_mds_client *mdsc = fsc->mdsc;
1827         struct ceph_osd_request *rd_req = NULL, *wr_req = NULL;
1828         struct rb_node **p, *parent;
1829         struct ceph_pool_perm *perm;
1830         struct page **pages;
1831         size_t pool_ns_len;
1832         int err = 0, err2 = 0, have = 0;
1833 
1834         down_read(&mdsc->pool_perm_rwsem);
1835         p = &mdsc->pool_perm_tree.rb_node;
1836         while (*p) {
1837                 perm = rb_entry(*p, struct ceph_pool_perm, node);
1838                 if (pool < perm->pool)
1839                         p = &(*p)->rb_left;
1840                 else if (pool > perm->pool)
1841                         p = &(*p)->rb_right;
1842                 else {
1843                         int ret = ceph_compare_string(pool_ns,
1844                                                 perm->pool_ns,
1845                                                 perm->pool_ns_len);
1846                         if (ret < 0)
1847                                 p = &(*p)->rb_left;
1848                         else if (ret > 0)
1849                                 p = &(*p)->rb_right;
1850                         else {
1851                                 have = perm->perm;
1852                                 break;
1853                         }
1854                 }
1855         }
1856         up_read(&mdsc->pool_perm_rwsem);
1857         if (*p)
1858                 goto out;
1859 
1860         if (pool_ns)
1861                 dout("__ceph_pool_perm_get pool %lld ns %.*s no perm cached\n",
1862                      pool, (int)pool_ns->len, pool_ns->str);
1863         else
1864                 dout("__ceph_pool_perm_get pool %lld no perm cached\n", pool);
1865 
1866         down_write(&mdsc->pool_perm_rwsem);
1867         p = &mdsc->pool_perm_tree.rb_node;
1868         parent = NULL;
1869         while (*p) {
1870                 parent = *p;
1871                 perm = rb_entry(parent, struct ceph_pool_perm, node);
1872                 if (pool < perm->pool)
1873                         p = &(*p)->rb_left;
1874                 else if (pool > perm->pool)
1875                         p = &(*p)->rb_right;
1876                 else {
1877                         int ret = ceph_compare_string(pool_ns,
1878                                                 perm->pool_ns,
1879                                                 perm->pool_ns_len);
1880                         if (ret < 0)
1881                                 p = &(*p)->rb_left;
1882                         else if (ret > 0)
1883                                 p = &(*p)->rb_right;
1884                         else {
1885                                 have = perm->perm;
1886                                 break;
1887                         }
1888                 }
1889         }
1890         if (*p) {
1891                 up_write(&mdsc->pool_perm_rwsem);
1892                 goto out;
1893         }
1894 
1895         rd_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1896                                          1, false, GFP_NOFS);
1897         if (!rd_req) {
1898                 err = -ENOMEM;
1899                 goto out_unlock;
1900         }
1901 
1902         rd_req->r_flags = CEPH_OSD_FLAG_READ;
1903         osd_req_op_init(rd_req, 0, CEPH_OSD_OP_STAT, 0);
1904         rd_req->r_base_oloc.pool = pool;
1905         if (pool_ns)
1906                 rd_req->r_base_oloc.pool_ns = ceph_get_string(pool_ns);
1907         ceph_oid_printf(&rd_req->r_base_oid, "%llx.00000000", ci->i_vino.ino);
1908 
1909         err = ceph_osdc_alloc_messages(rd_req, GFP_NOFS);
1910         if (err)
1911                 goto out_unlock;
1912 
1913         wr_req = ceph_osdc_alloc_request(&fsc->client->osdc, NULL,
1914                                          1, false, GFP_NOFS);
1915         if (!wr_req) {
1916                 err = -ENOMEM;
1917                 goto out_unlock;
1918         }
1919 
1920         wr_req->r_flags = CEPH_OSD_FLAG_WRITE;
1921         osd_req_op_init(wr_req, 0, CEPH_OSD_OP_CREATE, CEPH_OSD_OP_FLAG_EXCL);
1922         ceph_oloc_copy(&wr_req->r_base_oloc, &rd_req->r_base_oloc);
1923         ceph_oid_copy(&wr_req->r_base_oid, &rd_req->r_base_oid);
1924 
1925         err = ceph_osdc_alloc_messages(wr_req, GFP_NOFS);
1926         if (err)
1927                 goto out_unlock;
1928 
1929         /* one page should be large enough for STAT data */
1930         pages = ceph_alloc_page_vector(1, GFP_KERNEL);
1931         if (IS_ERR(pages)) {
1932                 err = PTR_ERR(pages);
1933                 goto out_unlock;
1934         }
1935 
1936         osd_req_op_raw_data_in_pages(rd_req, 0, pages, PAGE_SIZE,
1937                                      0, false, true);
1938         err = ceph_osdc_start_request(&fsc->client->osdc, rd_req, false);
1939 
1940         wr_req->r_mtime = ci->vfs_inode.i_mtime;
1941         err2 = ceph_osdc_start_request(&fsc->client->osdc, wr_req, false);
1942 
1943         if (!err)
1944                 err = ceph_osdc_wait_request(&fsc->client->osdc, rd_req);
1945         if (!err2)
1946                 err2 = ceph_osdc_wait_request(&fsc->client->osdc, wr_req);
1947 
1948         if (err >= 0 || err == -ENOENT)
1949                 have |= POOL_READ;
1950         else if (err != -EPERM)
1951                 goto out_unlock;
1952 
1953         if (err2 == 0 || err2 == -EEXIST)
1954                 have |= POOL_WRITE;
1955         else if (err2 != -EPERM) {
1956                 err = err2;
1957                 goto out_unlock;
1958         }
1959 
1960         pool_ns_len = pool_ns ? pool_ns->len : 0;
1961         perm = kmalloc(sizeof(*perm) + pool_ns_len + 1, GFP_NOFS);
1962         if (!perm) {
1963                 err = -ENOMEM;
1964                 goto out_unlock;
1965         }
1966 
1967         perm->pool = pool;
1968         perm->perm = have;
1969         perm->pool_ns_len = pool_ns_len;
1970         if (pool_ns_len > 0)
1971                 memcpy(perm->pool_ns, pool_ns->str, pool_ns_len);
1972         perm->pool_ns[pool_ns_len] = 0;
1973 
1974         rb_link_node(&perm->node, parent, p);
1975         rb_insert_color(&perm->node, &mdsc->pool_perm_tree);
1976         err = 0;
1977 out_unlock:
1978         up_write(&mdsc->pool_perm_rwsem);
1979 
1980         ceph_osdc_put_request(rd_req);
1981         ceph_osdc_put_request(wr_req);
1982 out:
1983         if (!err)
1984                 err = have;
1985         if (pool_ns)
1986                 dout("__ceph_pool_perm_get pool %lld ns %.*s result = %d\n",
1987                      pool, (int)pool_ns->len, pool_ns->str, err);
1988         else
1989                 dout("__ceph_pool_perm_get pool %lld result = %d\n", pool, err);
1990         return err;
1991 }
1992 
1993 int ceph_pool_perm_check(struct ceph_inode_info *ci, int need)
1994 {
1995         s64 pool;
1996         struct ceph_string *pool_ns;
1997         int ret, flags;
1998 
1999         if (ci->i_vino.snap != CEPH_NOSNAP) {
2000                 /*
2001                  * Pool permission check needs to write to the first object.
2002                  * But for snapshot, head of the first object may have alread
2003                  * been deleted. Skip check to avoid creating orphan object.
2004                  */
2005                 return 0;
2006         }
2007 
2008         if (ceph_test_mount_opt(ceph_inode_to_client(&ci->vfs_inode),
2009                                 NOPOOLPERM))
2010                 return 0;
2011 
2012         spin_lock(&ci->i_ceph_lock);
2013         flags = ci->i_ceph_flags;
2014         pool = ci->i_layout.pool_id;
2015         spin_unlock(&ci->i_ceph_lock);
2016 check:
2017         if (flags & CEPH_I_POOL_PERM) {
2018                 if ((need & CEPH_CAP_FILE_RD) && !(flags & CEPH_I_POOL_RD)) {
2019                         dout("ceph_pool_perm_check pool %lld no read perm\n",
2020                              pool);
2021                         return -EPERM;
2022                 }
2023                 if ((need & CEPH_CAP_FILE_WR) && !(flags & CEPH_I_POOL_WR)) {
2024                         dout("ceph_pool_perm_check pool %lld no write perm\n",
2025                              pool);
2026                         return -EPERM;
2027                 }
2028                 return 0;
2029         }
2030 
2031         pool_ns = ceph_try_get_string(ci->i_layout.pool_ns);
2032         ret = __ceph_pool_perm_get(ci, pool, pool_ns);
2033         ceph_put_string(pool_ns);
2034         if (ret < 0)
2035                 return ret;
2036 
2037         flags = CEPH_I_POOL_PERM;
2038         if (ret & POOL_READ)
2039                 flags |= CEPH_I_POOL_RD;
2040         if (ret & POOL_WRITE)
2041                 flags |= CEPH_I_POOL_WR;
2042 
2043         spin_lock(&ci->i_ceph_lock);
2044         if (pool == ci->i_layout.pool_id &&
2045             pool_ns == rcu_dereference_raw(ci->i_layout.pool_ns)) {
2046                 ci->i_ceph_flags |= flags;
2047         } else {
2048                 pool = ci->i_layout.pool_id;
2049                 flags = ci->i_ceph_flags;
2050         }
2051         spin_unlock(&ci->i_ceph_lock);
2052         goto check;
2053 }
2054 
2055 void ceph_pool_perm_destroy(struct ceph_mds_client *mdsc)
2056 {
2057         struct ceph_pool_perm *perm;
2058         struct rb_node *n;
2059 
2060         while (!RB_EMPTY_ROOT(&mdsc->pool_perm_tree)) {
2061                 n = rb_first(&mdsc->pool_perm_tree);
2062                 perm = rb_entry(n, struct ceph_pool_perm, node);
2063                 rb_erase(n, &mdsc->pool_perm_tree);
2064                 kfree(perm);
2065         }
2066 }
2067 

~ [ 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