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

TOMOYO Linux Cross Reference
Linux/fs/nfs/write.c

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

  1 /*
  2  * linux/fs/nfs/write.c
  3  *
  4  * Write file data over NFS.
  5  *
  6  * Copyright (C) 1996, 1997, Olaf Kirch <okir@monad.swb.de>
  7  */
  8 
  9 #include <linux/types.h>
 10 #include <linux/slab.h>
 11 #include <linux/mm.h>
 12 #include <linux/pagemap.h>
 13 #include <linux/file.h>
 14 #include <linux/writeback.h>
 15 #include <linux/swap.h>
 16 #include <linux/migrate.h>
 17 
 18 #include <linux/sunrpc/clnt.h>
 19 #include <linux/nfs_fs.h>
 20 #include <linux/nfs_mount.h>
 21 #include <linux/nfs_page.h>
 22 #include <linux/backing-dev.h>
 23 #include <linux/export.h>
 24 #include <linux/freezer.h>
 25 #include <linux/wait.h>
 26 
 27 #include <asm/uaccess.h>
 28 
 29 #include "delegation.h"
 30 #include "internal.h"
 31 #include "iostat.h"
 32 #include "nfs4_fs.h"
 33 #include "fscache.h"
 34 #include "pnfs.h"
 35 
 36 #include "nfstrace.h"
 37 
 38 #define NFSDBG_FACILITY         NFSDBG_PAGECACHE
 39 
 40 #define MIN_POOL_WRITE          (32)
 41 #define MIN_POOL_COMMIT         (4)
 42 
 43 /*
 44  * Local function declarations
 45  */
 46 static void nfs_redirty_request(struct nfs_page *req);
 47 static const struct rpc_call_ops nfs_commit_ops;
 48 static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops;
 49 static const struct nfs_commit_completion_ops nfs_commit_completion_ops;
 50 static const struct nfs_rw_ops nfs_rw_write_ops;
 51 static void nfs_clear_request_commit(struct nfs_page *req);
 52 static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
 53                                       struct inode *inode);
 54 static struct nfs_page *
 55 nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi,
 56                                                 struct page *page);
 57 
 58 static struct kmem_cache *nfs_wdata_cachep;
 59 static mempool_t *nfs_wdata_mempool;
 60 static struct kmem_cache *nfs_cdata_cachep;
 61 static mempool_t *nfs_commit_mempool;
 62 
 63 struct nfs_commit_data *nfs_commitdata_alloc(void)
 64 {
 65         struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO);
 66 
 67         if (p) {
 68                 memset(p, 0, sizeof(*p));
 69                 INIT_LIST_HEAD(&p->pages);
 70         }
 71         return p;
 72 }
 73 EXPORT_SYMBOL_GPL(nfs_commitdata_alloc);
 74 
 75 void nfs_commit_free(struct nfs_commit_data *p)
 76 {
 77         mempool_free(p, nfs_commit_mempool);
 78 }
 79 EXPORT_SYMBOL_GPL(nfs_commit_free);
 80 
 81 static struct nfs_pgio_header *nfs_writehdr_alloc(void)
 82 {
 83         struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO);
 84 
 85         if (p)
 86                 memset(p, 0, sizeof(*p));
 87         return p;
 88 }
 89 
 90 static void nfs_writehdr_free(struct nfs_pgio_header *hdr)
 91 {
 92         mempool_free(hdr, nfs_wdata_mempool);
 93 }
 94 
 95 static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error)
 96 {
 97         ctx->error = error;
 98         smp_wmb();
 99         set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
100 }
101 
102 /*
103  * nfs_page_find_head_request_locked - find head request associated with @page
104  *
105  * must be called while holding the inode lock.
106  *
107  * returns matching head request with reference held, or NULL if not found.
108  */
109 static struct nfs_page *
110 nfs_page_find_head_request_locked(struct nfs_inode *nfsi, struct page *page)
111 {
112         struct nfs_page *req = NULL;
113 
114         if (PagePrivate(page))
115                 req = (struct nfs_page *)page_private(page);
116         else if (unlikely(PageSwapCache(page)))
117                 req = nfs_page_search_commits_for_head_request_locked(nfsi,
118                         page);
119 
120         if (req) {
121                 WARN_ON_ONCE(req->wb_head != req);
122                 kref_get(&req->wb_kref);
123         }
124 
125         return req;
126 }
127 
128 /*
129  * nfs_page_find_head_request - find head request associated with @page
130  *
131  * returns matching head request with reference held, or NULL if not found.
132  */
133 static struct nfs_page *nfs_page_find_head_request(struct page *page)
134 {
135         struct inode *inode = page_file_mapping(page)->host;
136         struct nfs_page *req = NULL;
137 
138         spin_lock(&inode->i_lock);
139         req = nfs_page_find_head_request_locked(NFS_I(inode), page);
140         spin_unlock(&inode->i_lock);
141         return req;
142 }
143 
144 /* Adjust the file length if we're writing beyond the end */
145 static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count)
146 {
147         struct inode *inode = page_file_mapping(page)->host;
148         loff_t end, i_size;
149         pgoff_t end_index;
150 
151         spin_lock(&inode->i_lock);
152         i_size = i_size_read(inode);
153         end_index = (i_size - 1) >> PAGE_SHIFT;
154         if (i_size > 0 && page_file_index(page) < end_index)
155                 goto out;
156         end = page_file_offset(page) + ((loff_t)offset+count);
157         if (i_size >= end)
158                 goto out;
159         i_size_write(inode, end);
160         nfs_inc_stats(inode, NFSIOS_EXTENDWRITE);
161 out:
162         spin_unlock(&inode->i_lock);
163 }
164 
165 /* A writeback failed: mark the page as bad, and invalidate the page cache */
166 static void nfs_set_pageerror(struct page *page)
167 {
168         nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page));
169 }
170 
171 /*
172  * nfs_page_group_search_locked
173  * @head - head request of page group
174  * @page_offset - offset into page
175  *
176  * Search page group with head @head to find a request that contains the
177  * page offset @page_offset.
178  *
179  * Returns a pointer to the first matching nfs request, or NULL if no
180  * match is found.
181  *
182  * Must be called with the page group lock held
183  */
184 static struct nfs_page *
185 nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset)
186 {
187         struct nfs_page *req;
188 
189         WARN_ON_ONCE(head != head->wb_head);
190         WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_head->wb_flags));
191 
192         req = head;
193         do {
194                 if (page_offset >= req->wb_pgbase &&
195                     page_offset < (req->wb_pgbase + req->wb_bytes))
196                         return req;
197 
198                 req = req->wb_this_page;
199         } while (req != head);
200 
201         return NULL;
202 }
203 
204 /*
205  * nfs_page_group_covers_page
206  * @head - head request of page group
207  *
208  * Return true if the page group with head @head covers the whole page,
209  * returns false otherwise
210  */
211 static bool nfs_page_group_covers_page(struct nfs_page *req)
212 {
213         struct nfs_page *tmp;
214         unsigned int pos = 0;
215         unsigned int len = nfs_page_length(req->wb_page);
216 
217         nfs_page_group_lock(req, false);
218 
219         do {
220                 tmp = nfs_page_group_search_locked(req->wb_head, pos);
221                 if (tmp) {
222                         /* no way this should happen */
223                         WARN_ON_ONCE(tmp->wb_pgbase != pos);
224                         pos += tmp->wb_bytes - (pos - tmp->wb_pgbase);
225                 }
226         } while (tmp && pos < len);
227 
228         nfs_page_group_unlock(req);
229         WARN_ON_ONCE(pos > len);
230         return pos == len;
231 }
232 
233 /* We can set the PG_uptodate flag if we see that a write request
234  * covers the full page.
235  */
236 static void nfs_mark_uptodate(struct nfs_page *req)
237 {
238         if (PageUptodate(req->wb_page))
239                 return;
240         if (!nfs_page_group_covers_page(req))
241                 return;
242         SetPageUptodate(req->wb_page);
243 }
244 
245 static int wb_priority(struct writeback_control *wbc)
246 {
247         int ret = 0;
248 
249         if (wbc->sync_mode == WB_SYNC_ALL)
250                 ret = FLUSH_COND_STABLE;
251         return ret;
252 }
253 
254 /*
255  * NFS congestion control
256  */
257 
258 int nfs_congestion_kb;
259 
260 #define NFS_CONGESTION_ON_THRESH        (nfs_congestion_kb >> (PAGE_SHIFT-10))
261 #define NFS_CONGESTION_OFF_THRESH       \
262         (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
263 
264 static void nfs_set_page_writeback(struct page *page)
265 {
266         struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host);
267         int ret = test_set_page_writeback(page);
268 
269         WARN_ON_ONCE(ret != 0);
270 
271         if (atomic_long_inc_return(&nfss->writeback) >
272                         NFS_CONGESTION_ON_THRESH) {
273                 set_bdi_congested(&nfss->backing_dev_info,
274                                         BLK_RW_ASYNC);
275         }
276 }
277 
278 static void nfs_end_page_writeback(struct nfs_page *req)
279 {
280         struct inode *inode = page_file_mapping(req->wb_page)->host;
281         struct nfs_server *nfss = NFS_SERVER(inode);
282 
283         if (!nfs_page_group_sync_on_bit(req, PG_WB_END))
284                 return;
285 
286         end_page_writeback(req->wb_page);
287         if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
288                 clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
289 }
290 
291 
292 /* nfs_page_group_clear_bits
293  *   @req - an nfs request
294  * clears all page group related bits from @req
295  */
296 static void
297 nfs_page_group_clear_bits(struct nfs_page *req)
298 {
299         clear_bit(PG_TEARDOWN, &req->wb_flags);
300         clear_bit(PG_UNLOCKPAGE, &req->wb_flags);
301         clear_bit(PG_UPTODATE, &req->wb_flags);
302         clear_bit(PG_WB_END, &req->wb_flags);
303         clear_bit(PG_REMOVE, &req->wb_flags);
304 }
305 
306 
307 /*
308  * nfs_unroll_locks_and_wait -  unlock all newly locked reqs and wait on @req
309  *
310  * this is a helper function for nfs_lock_and_join_requests
311  *
312  * @inode - inode associated with request page group, must be holding inode lock
313  * @head  - head request of page group, must be holding head lock
314  * @req   - request that couldn't lock and needs to wait on the req bit lock
315  * @nonblock - if true, don't actually wait
316  *
317  * NOTE: this must be called holding page_group bit lock and inode spin lock
318  *       and BOTH will be released before returning.
319  *
320  * returns 0 on success, < 0 on error.
321  */
322 static int
323 nfs_unroll_locks_and_wait(struct inode *inode, struct nfs_page *head,
324                           struct nfs_page *req, bool nonblock)
325         __releases(&inode->i_lock)
326 {
327         struct nfs_page *tmp;
328         int ret;
329 
330         /* relinquish all the locks successfully grabbed this run */
331         for (tmp = head ; tmp != req; tmp = tmp->wb_this_page)
332                 nfs_unlock_request(tmp);
333 
334         WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
335 
336         /* grab a ref on the request that will be waited on */
337         kref_get(&req->wb_kref);
338 
339         nfs_page_group_unlock(head);
340         spin_unlock(&inode->i_lock);
341 
342         /* release ref from nfs_page_find_head_request_locked */
343         nfs_release_request(head);
344 
345         if (!nonblock)
346                 ret = nfs_wait_on_request(req);
347         else
348                 ret = -EAGAIN;
349         nfs_release_request(req);
350 
351         return ret;
352 }
353 
354 /*
355  * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests
356  *
357  * @destroy_list - request list (using wb_this_page) terminated by @old_head
358  * @old_head - the old head of the list
359  *
360  * All subrequests must be locked and removed from all lists, so at this point
361  * they are only "active" in this function, and possibly in nfs_wait_on_request
362  * with a reference held by some other context.
363  */
364 static void
365 nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list,
366                                  struct nfs_page *old_head)
367 {
368         while (destroy_list) {
369                 struct nfs_page *subreq = destroy_list;
370 
371                 destroy_list = (subreq->wb_this_page == old_head) ?
372                                    NULL : subreq->wb_this_page;
373 
374                 WARN_ON_ONCE(old_head != subreq->wb_head);
375 
376                 /* make sure old group is not used */
377                 subreq->wb_head = subreq;
378                 subreq->wb_this_page = subreq;
379 
380                 /* subreq is now totally disconnected from page group or any
381                  * write / commit lists. last chance to wake any waiters */
382                 nfs_unlock_request(subreq);
383 
384                 if (!test_bit(PG_TEARDOWN, &subreq->wb_flags)) {
385                         /* release ref on old head request */
386                         nfs_release_request(old_head);
387 
388                         nfs_page_group_clear_bits(subreq);
389 
390                         /* release the PG_INODE_REF reference */
391                         if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags))
392                                 nfs_release_request(subreq);
393                         else
394                                 WARN_ON_ONCE(1);
395                 } else {
396                         WARN_ON_ONCE(test_bit(PG_CLEAN, &subreq->wb_flags));
397                         /* zombie requests have already released the last
398                          * reference and were waiting on the rest of the
399                          * group to complete. Since it's no longer part of a
400                          * group, simply free the request */
401                         nfs_page_group_clear_bits(subreq);
402                         nfs_free_request(subreq);
403                 }
404         }
405 }
406 
407 /*
408  * nfs_lock_and_join_requests - join all subreqs to the head req and return
409  *                              a locked reference, cancelling any pending
410  *                              operations for this page.
411  *
412  * @page - the page used to lookup the "page group" of nfs_page structures
413  * @nonblock - if true, don't block waiting for request locks
414  *
415  * This function joins all sub requests to the head request by first
416  * locking all requests in the group, cancelling any pending operations
417  * and finally updating the head request to cover the whole range covered by
418  * the (former) group.  All subrequests are removed from any write or commit
419  * lists, unlinked from the group and destroyed.
420  *
421  * Returns a locked, referenced pointer to the head request - which after
422  * this call is guaranteed to be the only request associated with the page.
423  * Returns NULL if no requests are found for @page, or a ERR_PTR if an
424  * error was encountered.
425  */
426 static struct nfs_page *
427 nfs_lock_and_join_requests(struct page *page, bool nonblock)
428 {
429         struct inode *inode = page_file_mapping(page)->host;
430         struct nfs_page *head, *subreq;
431         struct nfs_page *destroy_list = NULL;
432         unsigned int total_bytes;
433         int ret;
434 
435 try_again:
436         total_bytes = 0;
437 
438         WARN_ON_ONCE(destroy_list);
439 
440         spin_lock(&inode->i_lock);
441 
442         /*
443          * A reference is taken only on the head request which acts as a
444          * reference to the whole page group - the group will not be destroyed
445          * until the head reference is released.
446          */
447         head = nfs_page_find_head_request_locked(NFS_I(inode), page);
448 
449         if (!head) {
450                 spin_unlock(&inode->i_lock);
451                 return NULL;
452         }
453 
454         /* holding inode lock, so always make a non-blocking call to try the
455          * page group lock */
456         ret = nfs_page_group_lock(head, true);
457         if (ret < 0) {
458                 spin_unlock(&inode->i_lock);
459 
460                 if (!nonblock && ret == -EAGAIN) {
461                         nfs_page_group_lock_wait(head);
462                         nfs_release_request(head);
463                         goto try_again;
464                 }
465 
466                 nfs_release_request(head);
467                 return ERR_PTR(ret);
468         }
469 
470         /* lock each request in the page group */
471         subreq = head;
472         do {
473                 /*
474                  * Subrequests are always contiguous, non overlapping
475                  * and in order - but may be repeated (mirrored writes).
476                  */
477                 if (subreq->wb_offset == (head->wb_offset + total_bytes)) {
478                         /* keep track of how many bytes this group covers */
479                         total_bytes += subreq->wb_bytes;
480                 } else if (WARN_ON_ONCE(subreq->wb_offset < head->wb_offset ||
481                             ((subreq->wb_offset + subreq->wb_bytes) >
482                              (head->wb_offset + total_bytes)))) {
483                         nfs_page_group_unlock(head);
484                         spin_unlock(&inode->i_lock);
485                         return ERR_PTR(-EIO);
486                 }
487 
488                 if (!nfs_lock_request(subreq)) {
489                         /* releases page group bit lock and
490                          * inode spin lock and all references */
491                         ret = nfs_unroll_locks_and_wait(inode, head,
492                                 subreq, nonblock);
493 
494                         if (ret == 0)
495                                 goto try_again;
496 
497                         return ERR_PTR(ret);
498                 }
499 
500                 subreq = subreq->wb_this_page;
501         } while (subreq != head);
502 
503         /* Now that all requests are locked, make sure they aren't on any list.
504          * Commit list removal accounting is done after locks are dropped */
505         subreq = head;
506         do {
507                 nfs_clear_request_commit(subreq);
508                 subreq = subreq->wb_this_page;
509         } while (subreq != head);
510 
511         /* unlink subrequests from head, destroy them later */
512         if (head->wb_this_page != head) {
513                 /* destroy list will be terminated by head */
514                 destroy_list = head->wb_this_page;
515                 head->wb_this_page = head;
516 
517                 /* change head request to cover whole range that
518                  * the former page group covered */
519                 head->wb_bytes = total_bytes;
520         }
521 
522         /*
523          * prepare head request to be added to new pgio descriptor
524          */
525         nfs_page_group_clear_bits(head);
526 
527         /*
528          * some part of the group was still on the inode list - otherwise
529          * the group wouldn't be involved in async write.
530          * grab a reference for the head request, iff it needs one.
531          */
532         if (!test_and_set_bit(PG_INODE_REF, &head->wb_flags))
533                 kref_get(&head->wb_kref);
534 
535         nfs_page_group_unlock(head);
536 
537         /* drop lock to clean uprequests on destroy list */
538         spin_unlock(&inode->i_lock);
539 
540         nfs_destroy_unlinked_subrequests(destroy_list, head);
541 
542         /* still holds ref on head from nfs_page_find_head_request_locked
543          * and still has lock on head from lock loop */
544         return head;
545 }
546 
547 static void nfs_write_error_remove_page(struct nfs_page *req)
548 {
549         nfs_unlock_request(req);
550         nfs_end_page_writeback(req);
551         nfs_release_request(req);
552         generic_error_remove_page(page_file_mapping(req->wb_page),
553                                   req->wb_page);
554 }
555 
556 /*
557  * Find an associated nfs write request, and prepare to flush it out
558  * May return an error if the user signalled nfs_wait_on_request().
559  */
560 static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio,
561                                 struct page *page, bool nonblock,
562                                 bool launder)
563 {
564         struct nfs_page *req;
565         int ret = 0;
566 
567         req = nfs_lock_and_join_requests(page, nonblock);
568         if (!req)
569                 goto out;
570         ret = PTR_ERR(req);
571         if (IS_ERR(req))
572                 goto out;
573 
574         nfs_set_page_writeback(page);
575         WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags));
576 
577         ret = 0;
578         if (!nfs_pageio_add_request(pgio, req)) {
579                 ret = pgio->pg_error;
580                 /*
581                  * Remove the problematic req upon fatal errors
582                  * in launder case, while other dirty pages can
583                  * still be around until they get flushed.
584                  */
585                 if (nfs_error_is_fatal(ret)) {
586                         nfs_context_set_write_error(req->wb_context, ret);
587                         if (launder) {
588                                 nfs_write_error_remove_page(req);
589                                 goto out;
590                         }
591                 }
592                 nfs_redirty_request(req);
593                 ret = -EAGAIN;
594         } else
595                 nfs_add_stats(page_file_mapping(page)->host,
596                                 NFSIOS_WRITEPAGES, 1);
597 out:
598         return ret;
599 }
600 
601 static int nfs_do_writepage(struct page *page, struct writeback_control *wbc,
602                             struct nfs_pageio_descriptor *pgio, bool launder)
603 {
604         int ret;
605 
606         nfs_pageio_cond_complete(pgio, page_file_index(page));
607         ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE,
608                                    launder);
609         if (ret == -EAGAIN) {
610                 redirty_page_for_writepage(wbc, page);
611                 ret = 0;
612         }
613         return ret;
614 }
615 
616 /*
617  * Write an mmapped page to the server.
618  */
619 static int nfs_writepage_locked(struct page *page,
620                                 struct writeback_control *wbc,
621                                 bool launder)
622 {
623         struct nfs_pageio_descriptor pgio;
624         struct inode *inode = page_file_mapping(page)->host;
625         int err;
626 
627         nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
628         nfs_pageio_init_write(&pgio, inode, wb_priority(wbc),
629                                 false, &nfs_async_write_completion_ops);
630         err = nfs_do_writepage(page, wbc, &pgio, launder);
631         nfs_pageio_complete(&pgio);
632         if (err < 0)
633                 return err;
634         if (pgio.pg_error < 0)
635                 return pgio.pg_error;
636         return 0;
637 }
638 
639 int nfs_writepage(struct page *page, struct writeback_control *wbc)
640 {
641         int ret;
642 
643         ret = nfs_writepage_locked(page, wbc, false);
644         unlock_page(page);
645         return ret;
646 }
647 
648 static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data)
649 {
650         int ret;
651 
652         ret = nfs_do_writepage(page, wbc, data, false);
653         unlock_page(page);
654         return ret;
655 }
656 
657 int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
658 {
659         struct inode *inode = mapping->host;
660         unsigned long *bitlock = &NFS_I(inode)->flags;
661         struct nfs_pageio_descriptor pgio;
662         int err;
663 
664         /* Stop dirtying of new pages while we sync */
665         err = wait_on_bit_lock_action(bitlock, NFS_INO_FLUSHING,
666                         nfs_wait_bit_killable, TASK_KILLABLE);
667         if (err)
668                 goto out_err;
669 
670         nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);
671 
672         nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false,
673                                 &nfs_async_write_completion_ops);
674         err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio);
675         nfs_pageio_complete(&pgio);
676 
677         clear_bit_unlock(NFS_INO_FLUSHING, bitlock);
678         smp_mb__after_atomic();
679         wake_up_bit(bitlock, NFS_INO_FLUSHING);
680 
681         if (err < 0)
682                 goto out_err;
683         err = pgio.pg_error;
684         if (err < 0)
685                 goto out_err;
686         return 0;
687 out_err:
688         return err;
689 }
690 
691 /*
692  * Insert a write request into an inode
693  */
694 static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
695 {
696         struct nfs_inode *nfsi = NFS_I(inode);
697 
698         WARN_ON_ONCE(req->wb_this_page != req);
699 
700         /* Lock the request! */
701         nfs_lock_request(req);
702 
703         spin_lock(&inode->i_lock);
704         if (!nfsi->nrequests &&
705             NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
706                 inode->i_version++;
707         /*
708          * Swap-space should not get truncated. Hence no need to plug the race
709          * with invalidate/truncate.
710          */
711         if (likely(!PageSwapCache(req->wb_page))) {
712                 set_bit(PG_MAPPED, &req->wb_flags);
713                 SetPagePrivate(req->wb_page);
714                 set_page_private(req->wb_page, (unsigned long)req);
715         }
716         nfsi->nrequests++;
717         /* this a head request for a page group - mark it as having an
718          * extra reference so sub groups can follow suit.
719          * This flag also informs pgio layer when to bump nrequests when
720          * adding subrequests. */
721         WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags));
722         kref_get(&req->wb_kref);
723         spin_unlock(&inode->i_lock);
724 }
725 
726 /*
727  * Remove a write request from an inode
728  */
729 static void nfs_inode_remove_request(struct nfs_page *req)
730 {
731         struct inode *inode = d_inode(req->wb_context->dentry);
732         struct nfs_inode *nfsi = NFS_I(inode);
733         struct nfs_page *head;
734 
735         if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) {
736                 head = req->wb_head;
737 
738                 spin_lock(&inode->i_lock);
739                 if (likely(head->wb_page && !PageSwapCache(head->wb_page))) {
740                         set_page_private(head->wb_page, 0);
741                         ClearPagePrivate(head->wb_page);
742                         smp_mb__after_atomic();
743                         wake_up_page(head->wb_page, PG_private);
744                         clear_bit(PG_MAPPED, &head->wb_flags);
745                 }
746                 nfsi->nrequests--;
747                 spin_unlock(&inode->i_lock);
748         } else {
749                 spin_lock(&inode->i_lock);
750                 nfsi->nrequests--;
751                 spin_unlock(&inode->i_lock);
752         }
753 
754         if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags))
755                 nfs_release_request(req);
756 }
757 
758 static void
759 nfs_mark_request_dirty(struct nfs_page *req)
760 {
761         if (req->wb_page)
762                 __set_page_dirty_nobuffers(req->wb_page);
763 }
764 
765 /*
766  * nfs_page_search_commits_for_head_request_locked
767  *
768  * Search through commit lists on @inode for the head request for @page.
769  * Must be called while holding the inode (which is cinfo) lock.
770  *
771  * Returns the head request if found, or NULL if not found.
772  */
773 static struct nfs_page *
774 nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi,
775                                                 struct page *page)
776 {
777         struct nfs_page *freq, *t;
778         struct nfs_commit_info cinfo;
779         struct inode *inode = &nfsi->vfs_inode;
780 
781         nfs_init_cinfo_from_inode(&cinfo, inode);
782 
783         /* search through pnfs commit lists */
784         freq = pnfs_search_commit_reqs(inode, &cinfo, page);
785         if (freq)
786                 return freq->wb_head;
787 
788         /* Linearly search the commit list for the correct request */
789         list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) {
790                 if (freq->wb_page == page)
791                         return freq->wb_head;
792         }
793 
794         return NULL;
795 }
796 
797 /**
798  * nfs_request_add_commit_list_locked - add request to a commit list
799  * @req: pointer to a struct nfs_page
800  * @dst: commit list head
801  * @cinfo: holds list lock and accounting info
802  *
803  * This sets the PG_CLEAN bit, updates the cinfo count of
804  * number of outstanding requests requiring a commit as well as
805  * the MM page stats.
806  *
807  * The caller must hold cinfo->inode->i_lock, and the nfs_page lock.
808  */
809 void
810 nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst,
811                             struct nfs_commit_info *cinfo)
812 {
813         set_bit(PG_CLEAN, &req->wb_flags);
814         nfs_list_add_request(req, dst);
815         cinfo->mds->ncommit++;
816 }
817 EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked);
818 
819 /**
820  * nfs_request_add_commit_list - add request to a commit list
821  * @req: pointer to a struct nfs_page
822  * @dst: commit list head
823  * @cinfo: holds list lock and accounting info
824  *
825  * This sets the PG_CLEAN bit, updates the cinfo count of
826  * number of outstanding requests requiring a commit as well as
827  * the MM page stats.
828  *
829  * The caller must _not_ hold the cinfo->lock, but must be
830  * holding the nfs_page lock.
831  */
832 void
833 nfs_request_add_commit_list(struct nfs_page *req, struct nfs_commit_info *cinfo)
834 {
835         spin_lock(&cinfo->inode->i_lock);
836         nfs_request_add_commit_list_locked(req, &cinfo->mds->list, cinfo);
837         spin_unlock(&cinfo->inode->i_lock);
838         if (req->wb_page)
839                 nfs_mark_page_unstable(req->wb_page, cinfo);
840 }
841 EXPORT_SYMBOL_GPL(nfs_request_add_commit_list);
842 
843 /**
844  * nfs_request_remove_commit_list - Remove request from a commit list
845  * @req: pointer to a nfs_page
846  * @cinfo: holds list lock and accounting info
847  *
848  * This clears the PG_CLEAN bit, and updates the cinfo's count of
849  * number of outstanding requests requiring a commit
850  * It does not update the MM page stats.
851  *
852  * The caller _must_ hold the cinfo->lock and the nfs_page lock.
853  */
854 void
855 nfs_request_remove_commit_list(struct nfs_page *req,
856                                struct nfs_commit_info *cinfo)
857 {
858         if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags))
859                 return;
860         nfs_list_remove_request(req);
861         cinfo->mds->ncommit--;
862 }
863 EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list);
864 
865 static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
866                                       struct inode *inode)
867 {
868         cinfo->inode = inode;
869         cinfo->mds = &NFS_I(inode)->commit_info;
870         cinfo->ds = pnfs_get_ds_info(inode);
871         cinfo->dreq = NULL;
872         cinfo->completion_ops = &nfs_commit_completion_ops;
873 }
874 
875 void nfs_init_cinfo(struct nfs_commit_info *cinfo,
876                     struct inode *inode,
877                     struct nfs_direct_req *dreq)
878 {
879         if (dreq)
880                 nfs_init_cinfo_from_dreq(cinfo, dreq);
881         else
882                 nfs_init_cinfo_from_inode(cinfo, inode);
883 }
884 EXPORT_SYMBOL_GPL(nfs_init_cinfo);
885 
886 /*
887  * Add a request to the inode's commit list.
888  */
889 void
890 nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg,
891                         struct nfs_commit_info *cinfo, u32 ds_commit_idx)
892 {
893         if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx))
894                 return;
895         nfs_request_add_commit_list(req, cinfo);
896 }
897 
898 static void
899 nfs_clear_page_commit(struct page *page)
900 {
901         dec_zone_page_state(page, NR_UNSTABLE_NFS);
902         dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb,
903                     WB_RECLAIMABLE);
904 }
905 
906 /* Called holding inode (/cinfo) lock */
907 static void
908 nfs_clear_request_commit(struct nfs_page *req)
909 {
910         if (test_bit(PG_CLEAN, &req->wb_flags)) {
911                 struct inode *inode = d_inode(req->wb_context->dentry);
912                 struct nfs_commit_info cinfo;
913 
914                 nfs_init_cinfo_from_inode(&cinfo, inode);
915                 if (!pnfs_clear_request_commit(req, &cinfo)) {
916                         nfs_request_remove_commit_list(req, &cinfo);
917                 }
918                 nfs_clear_page_commit(req->wb_page);
919         }
920 }
921 
922 int nfs_write_need_commit(struct nfs_pgio_header *hdr)
923 {
924         if (hdr->verf.committed == NFS_DATA_SYNC)
925                 return hdr->lseg == NULL;
926         return hdr->verf.committed != NFS_FILE_SYNC;
927 }
928 
929 static void nfs_write_completion(struct nfs_pgio_header *hdr)
930 {
931         struct nfs_commit_info cinfo;
932         unsigned long bytes = 0;
933 
934         if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
935                 goto out;
936         nfs_init_cinfo_from_inode(&cinfo, hdr->inode);
937         while (!list_empty(&hdr->pages)) {
938                 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
939 
940                 bytes += req->wb_bytes;
941                 nfs_list_remove_request(req);
942                 if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) &&
943                     (hdr->good_bytes < bytes)) {
944                         nfs_set_pageerror(req->wb_page);
945                         nfs_context_set_write_error(req->wb_context, hdr->error);
946                         goto remove_req;
947                 }
948                 if (nfs_write_need_commit(hdr)) {
949                         memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf));
950                         nfs_mark_request_commit(req, hdr->lseg, &cinfo,
951                                 hdr->pgio_mirror_idx);
952                         goto next;
953                 }
954 remove_req:
955                 nfs_inode_remove_request(req);
956 next:
957                 nfs_unlock_request(req);
958                 nfs_end_page_writeback(req);
959                 nfs_release_request(req);
960         }
961 out:
962         hdr->release(hdr);
963 }
964 
965 unsigned long
966 nfs_reqs_to_commit(struct nfs_commit_info *cinfo)
967 {
968         return cinfo->mds->ncommit;
969 }
970 
971 /* cinfo->inode->i_lock held by caller */
972 int
973 nfs_scan_commit_list(struct list_head *src, struct list_head *dst,
974                      struct nfs_commit_info *cinfo, int max)
975 {
976         struct nfs_page *req, *tmp;
977         int ret = 0;
978 
979         list_for_each_entry_safe(req, tmp, src, wb_list) {
980                 if (!nfs_lock_request(req))
981                         continue;
982                 kref_get(&req->wb_kref);
983                 if (cond_resched_lock(&cinfo->inode->i_lock))
984                         list_safe_reset_next(req, tmp, wb_list);
985                 nfs_request_remove_commit_list(req, cinfo);
986                 nfs_list_add_request(req, dst);
987                 ret++;
988                 if ((ret == max) && !cinfo->dreq)
989                         break;
990         }
991         return ret;
992 }
993 
994 /*
995  * nfs_scan_commit - Scan an inode for commit requests
996  * @inode: NFS inode to scan
997  * @dst: mds destination list
998  * @cinfo: mds and ds lists of reqs ready to commit
999  *
1000  * Moves requests from the inode's 'commit' request list.
1001  * The requests are *not* checked to ensure that they form a contiguous set.
1002  */
1003 int
1004 nfs_scan_commit(struct inode *inode, struct list_head *dst,
1005                 struct nfs_commit_info *cinfo)
1006 {
1007         int ret = 0;
1008 
1009         spin_lock(&cinfo->inode->i_lock);
1010         if (cinfo->mds->ncommit > 0) {
1011                 const int max = INT_MAX;
1012 
1013                 ret = nfs_scan_commit_list(&cinfo->mds->list, dst,
1014                                            cinfo, max);
1015                 ret += pnfs_scan_commit_lists(inode, cinfo, max - ret);
1016         }
1017         spin_unlock(&cinfo->inode->i_lock);
1018         return ret;
1019 }
1020 
1021 /*
1022  * Search for an existing write request, and attempt to update
1023  * it to reflect a new dirty region on a given page.
1024  *
1025  * If the attempt fails, then the existing request is flushed out
1026  * to disk.
1027  */
1028 static struct nfs_page *nfs_try_to_update_request(struct inode *inode,
1029                 struct page *page,
1030                 unsigned int offset,
1031                 unsigned int bytes)
1032 {
1033         struct nfs_page *req;
1034         unsigned int rqend;
1035         unsigned int end;
1036         int error;
1037 
1038         if (!PagePrivate(page))
1039                 return NULL;
1040 
1041         end = offset + bytes;
1042         spin_lock(&inode->i_lock);
1043 
1044         for (;;) {
1045                 req = nfs_page_find_head_request_locked(NFS_I(inode), page);
1046                 if (req == NULL)
1047                         goto out_unlock;
1048 
1049                 /* should be handled by nfs_flush_incompatible */
1050                 WARN_ON_ONCE(req->wb_head != req);
1051                 WARN_ON_ONCE(req->wb_this_page != req);
1052 
1053                 rqend = req->wb_offset + req->wb_bytes;
1054                 /*
1055                  * Tell the caller to flush out the request if
1056                  * the offsets are non-contiguous.
1057                  * Note: nfs_flush_incompatible() will already
1058                  * have flushed out requests having wrong owners.
1059                  */
1060                 if (offset > rqend
1061                     || end < req->wb_offset)
1062                         goto out_flushme;
1063 
1064                 if (nfs_lock_request(req))
1065                         break;
1066 
1067                 /* The request is locked, so wait and then retry */
1068                 spin_unlock(&inode->i_lock);
1069                 error = nfs_wait_on_request(req);
1070                 nfs_release_request(req);
1071                 if (error != 0)
1072                         goto out_err;
1073                 spin_lock(&inode->i_lock);
1074         }
1075 
1076         /* Okay, the request matches. Update the region */
1077         if (offset < req->wb_offset) {
1078                 req->wb_offset = offset;
1079                 req->wb_pgbase = offset;
1080         }
1081         if (end > rqend)
1082                 req->wb_bytes = end - req->wb_offset;
1083         else
1084                 req->wb_bytes = rqend - req->wb_offset;
1085 out_unlock:
1086         if (req)
1087                 nfs_clear_request_commit(req);
1088         spin_unlock(&inode->i_lock);
1089         return req;
1090 out_flushme:
1091         spin_unlock(&inode->i_lock);
1092         nfs_release_request(req);
1093         error = nfs_wb_page(inode, page);
1094 out_err:
1095         return ERR_PTR(error);
1096 }
1097 
1098 /*
1099  * Try to update an existing write request, or create one if there is none.
1100  *
1101  * Note: Should always be called with the Page Lock held to prevent races
1102  * if we have to add a new request. Also assumes that the caller has
1103  * already called nfs_flush_incompatible() if necessary.
1104  */
1105 static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx,
1106                 struct page *page, unsigned int offset, unsigned int bytes)
1107 {
1108         struct inode *inode = page_file_mapping(page)->host;
1109         struct nfs_page *req;
1110 
1111         req = nfs_try_to_update_request(inode, page, offset, bytes);
1112         if (req != NULL)
1113                 goto out;
1114         req = nfs_create_request(ctx, page, NULL, offset, bytes);
1115         if (IS_ERR(req))
1116                 goto out;
1117         nfs_inode_add_request(inode, req);
1118 out:
1119         return req;
1120 }
1121 
1122 static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
1123                 unsigned int offset, unsigned int count)
1124 {
1125         struct nfs_page *req;
1126 
1127         req = nfs_setup_write_request(ctx, page, offset, count);
1128         if (IS_ERR(req))
1129                 return PTR_ERR(req);
1130         /* Update file length */
1131         nfs_grow_file(page, offset, count);
1132         nfs_mark_uptodate(req);
1133         nfs_mark_request_dirty(req);
1134         nfs_unlock_and_release_request(req);
1135         return 0;
1136 }
1137 
1138 int nfs_flush_incompatible(struct file *file, struct page *page)
1139 {
1140         struct nfs_open_context *ctx = nfs_file_open_context(file);
1141         struct nfs_lock_context *l_ctx;
1142         struct file_lock_context *flctx = file_inode(file)->i_flctx;
1143         struct nfs_page *req;
1144         int do_flush, status;
1145         /*
1146          * Look for a request corresponding to this page. If there
1147          * is one, and it belongs to another file, we flush it out
1148          * before we try to copy anything into the page. Do this
1149          * due to the lack of an ACCESS-type call in NFSv2.
1150          * Also do the same if we find a request from an existing
1151          * dropped page.
1152          */
1153         do {
1154                 req = nfs_page_find_head_request(page);
1155                 if (req == NULL)
1156                         return 0;
1157                 l_ctx = req->wb_lock_context;
1158                 do_flush = req->wb_page != page ||
1159                         !nfs_match_open_context(req->wb_context, ctx);
1160                 /* for now, flush if more than 1 request in page_group */
1161                 do_flush |= req->wb_this_page != req;
1162                 if (l_ctx && flctx &&
1163                     !(list_empty_careful(&flctx->flc_posix) &&
1164                       list_empty_careful(&flctx->flc_flock))) {
1165                         do_flush |= l_ctx->lockowner.l_owner != current->files
1166                                 || l_ctx->lockowner.l_pid != current->tgid;
1167                 }
1168                 nfs_release_request(req);
1169                 if (!do_flush)
1170                         return 0;
1171                 status = nfs_wb_page(page_file_mapping(page)->host, page);
1172         } while (status == 0);
1173         return status;
1174 }
1175 
1176 /*
1177  * Avoid buffered writes when a open context credential's key would
1178  * expire soon.
1179  *
1180  * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL.
1181  *
1182  * Return 0 and set a credential flag which triggers the inode to flush
1183  * and performs  NFS_FILE_SYNC writes if the key will expired within
1184  * RPC_KEY_EXPIRE_TIMEO.
1185  */
1186 int
1187 nfs_key_timeout_notify(struct file *filp, struct inode *inode)
1188 {
1189         struct nfs_open_context *ctx = nfs_file_open_context(filp);
1190         struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth;
1191 
1192         return rpcauth_key_timeout_notify(auth, ctx->cred);
1193 }
1194 
1195 /*
1196  * Test if the open context credential key is marked to expire soon.
1197  */
1198 bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx)
1199 {
1200         return rpcauth_cred_key_to_expire(ctx->cred);
1201 }
1202 
1203 /*
1204  * If the page cache is marked as unsafe or invalid, then we can't rely on
1205  * the PageUptodate() flag. In this case, we will need to turn off
1206  * write optimisations that depend on the page contents being correct.
1207  */
1208 static bool nfs_write_pageuptodate(struct page *page, struct inode *inode)
1209 {
1210         struct nfs_inode *nfsi = NFS_I(inode);
1211 
1212         if (nfs_have_delegated_attributes(inode))
1213                 goto out;
1214         if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
1215                 return false;
1216         smp_rmb();
1217         if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags))
1218                 return false;
1219 out:
1220         if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1221                 return false;
1222         return PageUptodate(page) != 0;
1223 }
1224 
1225 static bool
1226 is_whole_file_wrlock(struct file_lock *fl)
1227 {
1228         return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX &&
1229                         fl->fl_type == F_WRLCK;
1230 }
1231 
1232 /* If we know the page is up to date, and we're not using byte range locks (or
1233  * if we have the whole file locked for writing), it may be more efficient to
1234  * extend the write to cover the entire page in order to avoid fragmentation
1235  * inefficiencies.
1236  *
1237  * If the file is opened for synchronous writes then we can just skip the rest
1238  * of the checks.
1239  */
1240 static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode)
1241 {
1242         int ret;
1243         struct file_lock_context *flctx = inode->i_flctx;
1244         struct file_lock *fl;
1245 
1246         if (file->f_flags & O_DSYNC)
1247                 return 0;
1248         if (!nfs_write_pageuptodate(page, inode))
1249                 return 0;
1250         if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
1251                 return 1;
1252         if (!flctx || (list_empty_careful(&flctx->flc_flock) &&
1253                        list_empty_careful(&flctx->flc_posix)))
1254                 return 1;
1255 
1256         /* Check to see if there are whole file write locks */
1257         ret = 0;
1258         spin_lock(&flctx->flc_lock);
1259         if (!list_empty(&flctx->flc_posix)) {
1260                 fl = list_first_entry(&flctx->flc_posix, struct file_lock,
1261                                         fl_list);
1262                 if (is_whole_file_wrlock(fl))
1263                         ret = 1;
1264         } else if (!list_empty(&flctx->flc_flock)) {
1265                 fl = list_first_entry(&flctx->flc_flock, struct file_lock,
1266                                         fl_list);
1267                 if (fl->fl_type == F_WRLCK)
1268                         ret = 1;
1269         }
1270         spin_unlock(&flctx->flc_lock);
1271         return ret;
1272 }
1273 
1274 /*
1275  * Update and possibly write a cached page of an NFS file.
1276  *
1277  * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
1278  * things with a page scheduled for an RPC call (e.g. invalidate it).
1279  */
1280 int nfs_updatepage(struct file *file, struct page *page,
1281                 unsigned int offset, unsigned int count)
1282 {
1283         struct nfs_open_context *ctx = nfs_file_open_context(file);
1284         struct inode    *inode = page_file_mapping(page)->host;
1285         int             status = 0;
1286 
1287         nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
1288 
1289         dprintk("NFS:       nfs_updatepage(%pD2 %d@%lld)\n",
1290                 file, count, (long long)(page_file_offset(page) + offset));
1291 
1292         if (!count)
1293                 goto out;
1294 
1295         if (nfs_can_extend_write(file, page, inode)) {
1296                 count = max(count + offset, nfs_page_length(page));
1297                 offset = 0;
1298         }
1299 
1300         status = nfs_writepage_setup(ctx, page, offset, count);
1301         if (status < 0)
1302                 nfs_set_pageerror(page);
1303         else
1304                 __set_page_dirty_nobuffers(page);
1305 out:
1306         dprintk("NFS:       nfs_updatepage returns %d (isize %lld)\n",
1307                         status, (long long)i_size_read(inode));
1308         return status;
1309 }
1310 
1311 static int flush_task_priority(int how)
1312 {
1313         switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
1314                 case FLUSH_HIGHPRI:
1315                         return RPC_PRIORITY_HIGH;
1316                 case FLUSH_LOWPRI:
1317                         return RPC_PRIORITY_LOW;
1318         }
1319         return RPC_PRIORITY_NORMAL;
1320 }
1321 
1322 static void nfs_initiate_write(struct nfs_pgio_header *hdr,
1323                                struct rpc_message *msg,
1324                                const struct nfs_rpc_ops *rpc_ops,
1325                                struct rpc_task_setup *task_setup_data, int how)
1326 {
1327         int priority = flush_task_priority(how);
1328 
1329         task_setup_data->priority = priority;
1330         rpc_ops->write_setup(hdr, msg);
1331 
1332         nfs4_state_protect_write(NFS_SERVER(hdr->inode)->nfs_client,
1333                                  &task_setup_data->rpc_client, msg, hdr);
1334 }
1335 
1336 /* If a nfs_flush_* function fails, it should remove reqs from @head and
1337  * call this on each, which will prepare them to be retried on next
1338  * writeback using standard nfs.
1339  */
1340 static void nfs_redirty_request(struct nfs_page *req)
1341 {
1342         nfs_mark_request_dirty(req);
1343         set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags);
1344         nfs_unlock_request(req);
1345         nfs_end_page_writeback(req);
1346         nfs_release_request(req);
1347 }
1348 
1349 static void nfs_async_write_error(struct list_head *head)
1350 {
1351         struct nfs_page *req;
1352 
1353         while (!list_empty(head)) {
1354                 req = nfs_list_entry(head->next);
1355                 nfs_list_remove_request(req);
1356                 nfs_redirty_request(req);
1357         }
1358 }
1359 
1360 static void nfs_async_write_reschedule_io(struct nfs_pgio_header *hdr)
1361 {
1362         nfs_async_write_error(&hdr->pages);
1363 }
1364 
1365 static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = {
1366         .error_cleanup = nfs_async_write_error,
1367         .completion = nfs_write_completion,
1368         .reschedule_io = nfs_async_write_reschedule_io,
1369 };
1370 
1371 void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
1372                                struct inode *inode, int ioflags, bool force_mds,
1373                                const struct nfs_pgio_completion_ops *compl_ops)
1374 {
1375         struct nfs_server *server = NFS_SERVER(inode);
1376         const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops;
1377 
1378 #ifdef CONFIG_NFS_V4_1
1379         if (server->pnfs_curr_ld && !force_mds)
1380                 pg_ops = server->pnfs_curr_ld->pg_write_ops;
1381 #endif
1382         nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops,
1383                         server->wsize, ioflags);
1384 }
1385 EXPORT_SYMBOL_GPL(nfs_pageio_init_write);
1386 
1387 void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio)
1388 {
1389         struct nfs_pgio_mirror *mirror;
1390 
1391         if (pgio->pg_ops && pgio->pg_ops->pg_cleanup)
1392                 pgio->pg_ops->pg_cleanup(pgio);
1393 
1394         pgio->pg_ops = &nfs_pgio_rw_ops;
1395 
1396         nfs_pageio_stop_mirroring(pgio);
1397 
1398         mirror = &pgio->pg_mirrors[0];
1399         mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize;
1400 }
1401 EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds);
1402 
1403 
1404 void nfs_commit_prepare(struct rpc_task *task, void *calldata)
1405 {
1406         struct nfs_commit_data *data = calldata;
1407 
1408         NFS_PROTO(data->inode)->commit_rpc_prepare(task, data);
1409 }
1410 
1411 /*
1412  * Special version of should_remove_suid() that ignores capabilities.
1413  */
1414 static int nfs_should_remove_suid(const struct inode *inode)
1415 {
1416         umode_t mode = inode->i_mode;
1417         int kill = 0;
1418 
1419         /* suid always must be killed */
1420         if (unlikely(mode & S_ISUID))
1421                 kill = ATTR_KILL_SUID;
1422 
1423         /*
1424          * sgid without any exec bits is just a mandatory locking mark; leave
1425          * it alone.  If some exec bits are set, it's a real sgid; kill it.
1426          */
1427         if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1428                 kill |= ATTR_KILL_SGID;
1429 
1430         if (unlikely(kill && S_ISREG(mode)))
1431                 return kill;
1432 
1433         return 0;
1434 }
1435 
1436 static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr,
1437                 struct nfs_fattr *fattr)
1438 {
1439         struct nfs_pgio_args *argp = &hdr->args;
1440         struct nfs_pgio_res *resp = &hdr->res;
1441         u64 size = argp->offset + resp->count;
1442 
1443         if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
1444                 fattr->size = size;
1445         if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) {
1446                 fattr->valid &= ~NFS_ATTR_FATTR_SIZE;
1447                 return;
1448         }
1449         if (size != fattr->size)
1450                 return;
1451         /* Set attribute barrier */
1452         nfs_fattr_set_barrier(fattr);
1453         /* ...and update size */
1454         fattr->valid |= NFS_ATTR_FATTR_SIZE;
1455 }
1456 
1457 void nfs_writeback_update_inode(struct nfs_pgio_header *hdr)
1458 {
1459         struct nfs_fattr *fattr = &hdr->fattr;
1460         struct inode *inode = hdr->inode;
1461 
1462         spin_lock(&inode->i_lock);
1463         nfs_writeback_check_extend(hdr, fattr);
1464         nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1465         spin_unlock(&inode->i_lock);
1466 }
1467 EXPORT_SYMBOL_GPL(nfs_writeback_update_inode);
1468 
1469 /*
1470  * This function is called when the WRITE call is complete.
1471  */
1472 static int nfs_writeback_done(struct rpc_task *task,
1473                               struct nfs_pgio_header *hdr,
1474                               struct inode *inode)
1475 {
1476         int status;
1477 
1478         /*
1479          * ->write_done will attempt to use post-op attributes to detect
1480          * conflicting writes by other clients.  A strict interpretation
1481          * of close-to-open would allow us to continue caching even if
1482          * another writer had changed the file, but some applications
1483          * depend on tighter cache coherency when writing.
1484          */
1485         status = NFS_PROTO(inode)->write_done(task, hdr);
1486         if (status != 0)
1487                 return status;
1488         nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count);
1489 
1490         if (hdr->res.verf->committed < hdr->args.stable &&
1491             task->tk_status >= 0) {
1492                 /* We tried a write call, but the server did not
1493                  * commit data to stable storage even though we
1494                  * requested it.
1495                  * Note: There is a known bug in Tru64 < 5.0 in which
1496                  *       the server reports NFS_DATA_SYNC, but performs
1497                  *       NFS_FILE_SYNC. We therefore implement this checking
1498                  *       as a dprintk() in order to avoid filling syslog.
1499                  */
1500                 static unsigned long    complain;
1501 
1502                 /* Note this will print the MDS for a DS write */
1503                 if (time_before(complain, jiffies)) {
1504                         dprintk("NFS:       faulty NFS server %s:"
1505                                 " (committed = %d) != (stable = %d)\n",
1506                                 NFS_SERVER(inode)->nfs_client->cl_hostname,
1507                                 hdr->res.verf->committed, hdr->args.stable);
1508                         complain = jiffies + 300 * HZ;
1509                 }
1510         }
1511 
1512         /* Deal with the suid/sgid bit corner case */
1513         if (nfs_should_remove_suid(inode))
1514                 nfs_mark_for_revalidate(inode);
1515         return 0;
1516 }
1517 
1518 /*
1519  * This function is called when the WRITE call is complete.
1520  */
1521 static void nfs_writeback_result(struct rpc_task *task,
1522                                  struct nfs_pgio_header *hdr)
1523 {
1524         struct nfs_pgio_args    *argp = &hdr->args;
1525         struct nfs_pgio_res     *resp = &hdr->res;
1526 
1527         if (resp->count < argp->count) {
1528                 static unsigned long    complain;
1529 
1530                 /* This a short write! */
1531                 nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE);
1532 
1533                 /* Has the server at least made some progress? */
1534                 if (resp->count == 0) {
1535                         if (time_before(complain, jiffies)) {
1536                                 printk(KERN_WARNING
1537                                        "NFS: Server wrote zero bytes, expected %u.\n",
1538                                        argp->count);
1539                                 complain = jiffies + 300 * HZ;
1540                         }
1541                         nfs_set_pgio_error(hdr, -EIO, argp->offset);
1542                         task->tk_status = -EIO;
1543                         return;
1544                 }
1545 
1546                 /* For non rpc-based layout drivers, retry-through-MDS */
1547                 if (!task->tk_ops) {
1548                         hdr->pnfs_error = -EAGAIN;
1549                         return;
1550                 }
1551 
1552                 /* Was this an NFSv2 write or an NFSv3 stable write? */
1553                 if (resp->verf->committed != NFS_UNSTABLE) {
1554                         /* Resend from where the server left off */
1555                         hdr->mds_offset += resp->count;
1556                         argp->offset += resp->count;
1557                         argp->pgbase += resp->count;
1558                         argp->count -= resp->count;
1559                 } else {
1560                         /* Resend as a stable write in order to avoid
1561                          * headaches in the case of a server crash.
1562                          */
1563                         argp->stable = NFS_FILE_SYNC;
1564                 }
1565                 rpc_restart_call_prepare(task);
1566         }
1567 }
1568 
1569 static int wait_on_commit(struct nfs_mds_commit_info *cinfo)
1570 {
1571         return wait_on_atomic_t(&cinfo->rpcs_out,
1572                         nfs_wait_atomic_killable, TASK_KILLABLE);
1573 }
1574 
1575 static void nfs_commit_begin(struct nfs_mds_commit_info *cinfo)
1576 {
1577         atomic_inc(&cinfo->rpcs_out);
1578 }
1579 
1580 static void nfs_commit_end(struct nfs_mds_commit_info *cinfo)
1581 {
1582         if (atomic_dec_and_test(&cinfo->rpcs_out))
1583                 wake_up_atomic_t(&cinfo->rpcs_out);
1584 }
1585 
1586 void nfs_commitdata_release(struct nfs_commit_data *data)
1587 {
1588         put_nfs_open_context(data->context);
1589         nfs_commit_free(data);
1590 }
1591 EXPORT_SYMBOL_GPL(nfs_commitdata_release);
1592 
1593 int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data,
1594                         const struct nfs_rpc_ops *nfs_ops,
1595                         const struct rpc_call_ops *call_ops,
1596                         int how, int flags)
1597 {
1598         struct rpc_task *task;
1599         int priority = flush_task_priority(how);
1600         struct rpc_message msg = {
1601                 .rpc_argp = &data->args,
1602                 .rpc_resp = &data->res,
1603                 .rpc_cred = data->cred,
1604         };
1605         struct rpc_task_setup task_setup_data = {
1606                 .task = &data->task,
1607                 .rpc_client = clnt,
1608                 .rpc_message = &msg,
1609                 .callback_ops = call_ops,
1610                 .callback_data = data,
1611                 .workqueue = nfsiod_workqueue,
1612                 .flags = RPC_TASK_ASYNC | flags,
1613                 .priority = priority,
1614         };
1615         /* Set up the initial task struct.  */
1616         nfs_ops->commit_setup(data, &msg);
1617 
1618         dprintk("NFS: initiated commit call\n");
1619 
1620         nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client,
1621                 NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg);
1622 
1623         task = rpc_run_task(&task_setup_data);
1624         if (IS_ERR(task))
1625                 return PTR_ERR(task);
1626         if (how & FLUSH_SYNC)
1627                 rpc_wait_for_completion_task(task);
1628         rpc_put_task(task);
1629         return 0;
1630 }
1631 EXPORT_SYMBOL_GPL(nfs_initiate_commit);
1632 
1633 static loff_t nfs_get_lwb(struct list_head *head)
1634 {
1635         loff_t lwb = 0;
1636         struct nfs_page *req;
1637 
1638         list_for_each_entry(req, head, wb_list)
1639                 if (lwb < (req_offset(req) + req->wb_bytes))
1640                         lwb = req_offset(req) + req->wb_bytes;
1641 
1642         return lwb;
1643 }
1644 
1645 /*
1646  * Set up the argument/result storage required for the RPC call.
1647  */
1648 void nfs_init_commit(struct nfs_commit_data *data,
1649                      struct list_head *head,
1650                      struct pnfs_layout_segment *lseg,
1651                      struct nfs_commit_info *cinfo)
1652 {
1653         struct nfs_page *first = nfs_list_entry(head->next);
1654         struct inode *inode = d_inode(first->wb_context->dentry);
1655 
1656         /* Set up the RPC argument and reply structs
1657          * NB: take care not to mess about with data->commit et al. */
1658 
1659         list_splice_init(head, &data->pages);
1660 
1661         data->inode       = inode;
1662         data->cred        = first->wb_context->cred;
1663         data->lseg        = lseg; /* reference transferred */
1664         /* only set lwb for pnfs commit */
1665         if (lseg)
1666                 data->lwb = nfs_get_lwb(&data->pages);
1667         data->mds_ops     = &nfs_commit_ops;
1668         data->completion_ops = cinfo->completion_ops;
1669         data->dreq        = cinfo->dreq;
1670 
1671         data->args.fh     = NFS_FH(data->inode);
1672         /* Note: we always request a commit of the entire inode */
1673         data->args.offset = 0;
1674         data->args.count  = 0;
1675         data->context     = get_nfs_open_context(first->wb_context);
1676         data->res.fattr   = &data->fattr;
1677         data->res.verf    = &data->verf;
1678         nfs_fattr_init(&data->fattr);
1679 }
1680 EXPORT_SYMBOL_GPL(nfs_init_commit);
1681 
1682 void nfs_retry_commit(struct list_head *page_list,
1683                       struct pnfs_layout_segment *lseg,
1684                       struct nfs_commit_info *cinfo,
1685                       u32 ds_commit_idx)
1686 {
1687         struct nfs_page *req;
1688 
1689         while (!list_empty(page_list)) {
1690                 req = nfs_list_entry(page_list->next);
1691                 nfs_list_remove_request(req);
1692                 nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx);
1693                 if (!cinfo->dreq)
1694                         nfs_clear_page_commit(req->wb_page);
1695                 nfs_unlock_and_release_request(req);
1696         }
1697 }
1698 EXPORT_SYMBOL_GPL(nfs_retry_commit);
1699 
1700 static void
1701 nfs_commit_resched_write(struct nfs_commit_info *cinfo,
1702                 struct nfs_page *req)
1703 {
1704         __set_page_dirty_nobuffers(req->wb_page);
1705 }
1706 
1707 /*
1708  * Commit dirty pages
1709  */
1710 static int
1711 nfs_commit_list(struct inode *inode, struct list_head *head, int how,
1712                 struct nfs_commit_info *cinfo)
1713 {
1714         struct nfs_commit_data  *data;
1715 
1716         /* another commit raced with us */
1717         if (list_empty(head))
1718                 return 0;
1719 
1720         data = nfs_commitdata_alloc();
1721 
1722         if (!data)
1723                 goto out_bad;
1724 
1725         /* Set up the argument struct */
1726         nfs_init_commit(data, head, NULL, cinfo);
1727         atomic_inc(&cinfo->mds->rpcs_out);
1728         return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode),
1729                                    data->mds_ops, how, 0);
1730  out_bad:
1731         nfs_retry_commit(head, NULL, cinfo, 0);
1732         return -ENOMEM;
1733 }
1734 
1735 int nfs_commit_file(struct file *file, struct nfs_write_verifier *verf)
1736 {
1737         struct inode *inode = file_inode(file);
1738         struct nfs_open_context *open;
1739         struct nfs_commit_info cinfo;
1740         struct nfs_page *req;
1741         int ret;
1742 
1743         open = get_nfs_open_context(nfs_file_open_context(file));
1744         req  = nfs_create_request(open, NULL, NULL, 0, i_size_read(inode));
1745         if (IS_ERR(req)) {
1746                 ret = PTR_ERR(req);
1747                 goto out_put;
1748         }
1749 
1750         nfs_init_cinfo_from_inode(&cinfo, inode);
1751 
1752         memcpy(&req->wb_verf, verf, sizeof(struct nfs_write_verifier));
1753         nfs_request_add_commit_list(req, &cinfo);
1754         ret = nfs_commit_inode(inode, FLUSH_SYNC);
1755         if (ret > 0)
1756                 ret = 0;
1757 
1758         nfs_free_request(req);
1759 out_put:
1760         put_nfs_open_context(open);
1761         return ret;
1762 }
1763 EXPORT_SYMBOL_GPL(nfs_commit_file);
1764 
1765 /*
1766  * COMMIT call returned
1767  */
1768 static void nfs_commit_done(struct rpc_task *task, void *calldata)
1769 {
1770         struct nfs_commit_data  *data = calldata;
1771 
1772         dprintk("NFS: %5u nfs_commit_done (status %d)\n",
1773                                 task->tk_pid, task->tk_status);
1774 
1775         /* Call the NFS version-specific code */
1776         NFS_PROTO(data->inode)->commit_done(task, data);
1777 }
1778 
1779 static void nfs_commit_release_pages(struct nfs_commit_data *data)
1780 {
1781         struct nfs_page *req;
1782         int status = data->task.tk_status;
1783         struct nfs_commit_info cinfo;
1784         struct nfs_server *nfss;
1785 
1786         while (!list_empty(&data->pages)) {
1787                 req = nfs_list_entry(data->pages.next);
1788                 nfs_list_remove_request(req);
1789                 if (req->wb_page)
1790                         nfs_clear_page_commit(req->wb_page);
1791 
1792                 dprintk("NFS:       commit (%s/%llu %d@%lld)",
1793                         req->wb_context->dentry->d_sb->s_id,
1794                         (unsigned long long)NFS_FILEID(d_inode(req->wb_context->dentry)),
1795                         req->wb_bytes,
1796                         (long long)req_offset(req));
1797                 if (status < 0) {
1798                         nfs_context_set_write_error(req->wb_context, status);
1799                         nfs_inode_remove_request(req);
1800                         dprintk(", error = %d\n", status);
1801                         goto next;
1802                 }
1803 
1804                 /* Okay, COMMIT succeeded, apparently. Check the verifier
1805                  * returned by the server against all stored verfs. */
1806                 if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) {
1807                         /* We have a match */
1808                         nfs_inode_remove_request(req);
1809                         dprintk(" OK\n");
1810                         goto next;
1811                 }
1812                 /* We have a mismatch. Write the page again */
1813                 dprintk(" mismatch\n");
1814                 nfs_mark_request_dirty(req);
1815                 set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags);
1816         next:
1817                 nfs_unlock_and_release_request(req);
1818         }
1819         nfss = NFS_SERVER(data->inode);
1820         if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
1821                 clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
1822 
1823         nfs_init_cinfo(&cinfo, data->inode, data->dreq);
1824         nfs_commit_end(cinfo.mds);
1825 }
1826 
1827 static void nfs_commit_release(void *calldata)
1828 {
1829         struct nfs_commit_data *data = calldata;
1830 
1831         data->completion_ops->completion(data);
1832         nfs_commitdata_release(calldata);
1833 }
1834 
1835 static const struct rpc_call_ops nfs_commit_ops = {
1836         .rpc_call_prepare = nfs_commit_prepare,
1837         .rpc_call_done = nfs_commit_done,
1838         .rpc_release = nfs_commit_release,
1839 };
1840 
1841 static const struct nfs_commit_completion_ops nfs_commit_completion_ops = {
1842         .completion = nfs_commit_release_pages,
1843         .resched_write = nfs_commit_resched_write,
1844 };
1845 
1846 int nfs_generic_commit_list(struct inode *inode, struct list_head *head,
1847                             int how, struct nfs_commit_info *cinfo)
1848 {
1849         int status;
1850 
1851         status = pnfs_commit_list(inode, head, how, cinfo);
1852         if (status == PNFS_NOT_ATTEMPTED)
1853                 status = nfs_commit_list(inode, head, how, cinfo);
1854         return status;
1855 }
1856 
1857 int nfs_commit_inode(struct inode *inode, int how)
1858 {
1859         LIST_HEAD(head);
1860         struct nfs_commit_info cinfo;
1861         int may_wait = how & FLUSH_SYNC;
1862         int error = 0;
1863         int res;
1864 
1865         nfs_init_cinfo_from_inode(&cinfo, inode);
1866         nfs_commit_begin(cinfo.mds);
1867         res = nfs_scan_commit(inode, &head, &cinfo);
1868         if (res)
1869                 error = nfs_generic_commit_list(inode, &head, how, &cinfo);
1870         nfs_commit_end(cinfo.mds);
1871         if (error < 0)
1872                 goto out_error;
1873         if (!may_wait)
1874                 goto out_mark_dirty;
1875         error = wait_on_commit(cinfo.mds);
1876         if (error < 0)
1877                 return error;
1878         return res;
1879 out_error:
1880         res = error;
1881         /* Note: If we exit without ensuring that the commit is complete,
1882          * we must mark the inode as dirty. Otherwise, future calls to
1883          * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1884          * that the data is on the disk.
1885          */
1886 out_mark_dirty:
1887         __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1888         return res;
1889 }
1890 EXPORT_SYMBOL_GPL(nfs_commit_inode);
1891 
1892 int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1893 {
1894         struct nfs_inode *nfsi = NFS_I(inode);
1895         int flags = FLUSH_SYNC;
1896         int ret = 0;
1897 
1898         /* no commits means nothing needs to be done */
1899         if (!nfsi->commit_info.ncommit)
1900                 return ret;
1901 
1902         if (wbc->sync_mode == WB_SYNC_NONE) {
1903                 /* Don't commit yet if this is a non-blocking flush and there
1904                  * are a lot of outstanding writes for this mapping.
1905                  */
1906                 if (nfsi->commit_info.ncommit <= (nfsi->nrequests >> 1))
1907                         goto out_mark_dirty;
1908 
1909                 /* don't wait for the COMMIT response */
1910                 flags = 0;
1911         }
1912 
1913         ret = nfs_commit_inode(inode, flags);
1914         if (ret >= 0) {
1915                 if (wbc->sync_mode == WB_SYNC_NONE) {
1916                         if (ret < wbc->nr_to_write)
1917                                 wbc->nr_to_write -= ret;
1918                         else
1919                                 wbc->nr_to_write = 0;
1920                 }
1921                 return 0;
1922         }
1923 out_mark_dirty:
1924         __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1925         return ret;
1926 }
1927 EXPORT_SYMBOL_GPL(nfs_write_inode);
1928 
1929 /*
1930  * flush the inode to disk.
1931  */
1932 int nfs_wb_all(struct inode *inode)
1933 {
1934         int ret;
1935 
1936         trace_nfs_writeback_inode_enter(inode);
1937 
1938         ret = filemap_write_and_wait(inode->i_mapping);
1939         if (ret)
1940                 goto out;
1941         ret = nfs_commit_inode(inode, FLUSH_SYNC);
1942         if (ret < 0)
1943                 goto out;
1944         pnfs_sync_inode(inode, true);
1945         ret = 0;
1946 
1947 out:
1948         trace_nfs_writeback_inode_exit(inode, ret);
1949         return ret;
1950 }
1951 EXPORT_SYMBOL_GPL(nfs_wb_all);
1952 
1953 int nfs_wb_page_cancel(struct inode *inode, struct page *page)
1954 {
1955         struct nfs_page *req;
1956         int ret = 0;
1957 
1958         wait_on_page_writeback(page);
1959 
1960         /* blocking call to cancel all requests and join to a single (head)
1961          * request */
1962         req = nfs_lock_and_join_requests(page, false);
1963 
1964         if (IS_ERR(req)) {
1965                 ret = PTR_ERR(req);
1966         } else if (req) {
1967                 /* all requests from this page have been cancelled by
1968                  * nfs_lock_and_join_requests, so just remove the head
1969                  * request from the inode / page_private pointer and
1970                  * release it */
1971                 nfs_inode_remove_request(req);
1972                 nfs_unlock_and_release_request(req);
1973         }
1974 
1975         return ret;
1976 }
1977 
1978 /*
1979  * Write back all requests on one page - we do this before reading it.
1980  */
1981 int nfs_wb_single_page(struct inode *inode, struct page *page, bool launder)
1982 {
1983         loff_t range_start = page_file_offset(page);
1984         loff_t range_end = range_start + (loff_t)(PAGE_SIZE - 1);
1985         struct writeback_control wbc = {
1986                 .sync_mode = WB_SYNC_ALL,
1987                 .nr_to_write = 0,
1988                 .range_start = range_start,
1989                 .range_end = range_end,
1990         };
1991         int ret;
1992 
1993         trace_nfs_writeback_page_enter(inode);
1994 
1995         for (;;) {
1996                 wait_on_page_writeback(page);
1997                 if (clear_page_dirty_for_io(page)) {
1998                         ret = nfs_writepage_locked(page, &wbc, launder);
1999                         if (ret < 0)
2000                                 goto out_error;
2001                         continue;
2002                 }
2003                 ret = 0;
2004                 if (!PagePrivate(page))
2005                         break;
2006                 ret = nfs_commit_inode(inode, FLUSH_SYNC);
2007                 if (ret < 0)
2008                         goto out_error;
2009         }
2010 out_error:
2011         trace_nfs_writeback_page_exit(inode, ret);
2012         return ret;
2013 }
2014 
2015 #ifdef CONFIG_MIGRATION
2016 int nfs_migrate_page(struct address_space *mapping, struct page *newpage,
2017                 struct page *page, enum migrate_mode mode)
2018 {
2019         /*
2020          * If PagePrivate is set, then the page is currently associated with
2021          * an in-progress read or write request. Don't try to migrate it.
2022          *
2023          * FIXME: we could do this in principle, but we'll need a way to ensure
2024          *        that we can safely release the inode reference while holding
2025          *        the page lock.
2026          */
2027         if (PagePrivate(page))
2028                 return -EBUSY;
2029 
2030         if (!nfs_fscache_release_page(page, GFP_KERNEL))
2031                 return -EBUSY;
2032 
2033         return migrate_page(mapping, newpage, page, mode);
2034 }
2035 #endif
2036 
2037 int __init nfs_init_writepagecache(void)
2038 {
2039         nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
2040                                              sizeof(struct nfs_pgio_header),
2041                                              0, SLAB_HWCACHE_ALIGN,
2042                                              NULL);
2043         if (nfs_wdata_cachep == NULL)
2044                 return -ENOMEM;
2045 
2046         nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
2047                                                      nfs_wdata_cachep);
2048         if (nfs_wdata_mempool == NULL)
2049                 goto out_destroy_write_cache;
2050 
2051         nfs_cdata_cachep = kmem_cache_create("nfs_commit_data",
2052                                              sizeof(struct nfs_commit_data),
2053                                              0, SLAB_HWCACHE_ALIGN,
2054                                              NULL);
2055         if (nfs_cdata_cachep == NULL)
2056                 goto out_destroy_write_mempool;
2057 
2058         nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
2059                                                       nfs_cdata_cachep);
2060         if (nfs_commit_mempool == NULL)
2061                 goto out_destroy_commit_cache;
2062 
2063         /*
2064          * NFS congestion size, scale with available memory.
2065          *
2066          *  64MB:    8192k
2067          * 128MB:   11585k
2068          * 256MB:   16384k
2069          * 512MB:   23170k
2070          *   1GB:   32768k
2071          *   2GB:   46340k
2072          *   4GB:   65536k
2073          *   8GB:   92681k
2074          *  16GB:  131072k
2075          *
2076          * This allows larger machines to have larger/more transfers.
2077          * Limit the default to 256M
2078          */
2079         nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
2080         if (nfs_congestion_kb > 256*1024)
2081                 nfs_congestion_kb = 256*1024;
2082 
2083         return 0;
2084 
2085 out_destroy_commit_cache:
2086         kmem_cache_destroy(nfs_cdata_cachep);
2087 out_destroy_write_mempool:
2088         mempool_destroy(nfs_wdata_mempool);
2089 out_destroy_write_cache:
2090         kmem_cache_destroy(nfs_wdata_cachep);
2091         return -ENOMEM;
2092 }
2093 
2094 void nfs_destroy_writepagecache(void)
2095 {
2096         mempool_destroy(nfs_commit_mempool);
2097         kmem_cache_destroy(nfs_cdata_cachep);
2098         mempool_destroy(nfs_wdata_mempool);
2099         kmem_cache_destroy(nfs_wdata_cachep);
2100 }
2101 
2102 static const struct nfs_rw_ops nfs_rw_write_ops = {
2103         .rw_mode                = FMODE_WRITE,
2104         .rw_alloc_header        = nfs_writehdr_alloc,
2105         .rw_free_header         = nfs_writehdr_free,
2106         .rw_done                = nfs_writeback_done,
2107         .rw_result              = nfs_writeback_result,
2108         .rw_initiate            = nfs_initiate_write,
2109 };
2110 

~ [ 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