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

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

Version: ~ [ linux-5.3-rc1 ] ~ [ linux-5.2.2 ] ~ [ linux-5.1.19 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.60 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.134 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.186 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.186 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.70 ] ~ [ 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 
 25 #include <asm/uaccess.h>
 26 
 27 #include "delegation.h"
 28 #include "internal.h"
 29 #include "iostat.h"
 30 #include "nfs4_fs.h"
 31 #include "fscache.h"
 32 #include "pnfs.h"
 33 
 34 #include "nfstrace.h"
 35 
 36 #define NFSDBG_FACILITY         NFSDBG_PAGECACHE
 37 
 38 #define MIN_POOL_WRITE          (32)
 39 #define MIN_POOL_COMMIT         (4)
 40 
 41 /*
 42  * Local function declarations
 43  */
 44 static void nfs_redirty_request(struct nfs_page *req);
 45 static const struct rpc_call_ops nfs_commit_ops;
 46 static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops;
 47 static const struct nfs_commit_completion_ops nfs_commit_completion_ops;
 48 static const struct nfs_rw_ops nfs_rw_write_ops;
 49 static void nfs_clear_request_commit(struct nfs_page *req);
 50 static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
 51                                       struct inode *inode);
 52 static struct nfs_page *
 53 nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi,
 54                                                 struct page *page);
 55 
 56 static struct kmem_cache *nfs_wdata_cachep;
 57 static mempool_t *nfs_wdata_mempool;
 58 static struct kmem_cache *nfs_cdata_cachep;
 59 static mempool_t *nfs_commit_mempool;
 60 
 61 struct nfs_commit_data *nfs_commitdata_alloc(void)
 62 {
 63         struct nfs_commit_data *p = mempool_alloc(nfs_commit_mempool, GFP_NOIO);
 64 
 65         if (p) {
 66                 memset(p, 0, sizeof(*p));
 67                 INIT_LIST_HEAD(&p->pages);
 68         }
 69         return p;
 70 }
 71 EXPORT_SYMBOL_GPL(nfs_commitdata_alloc);
 72 
 73 void nfs_commit_free(struct nfs_commit_data *p)
 74 {
 75         mempool_free(p, nfs_commit_mempool);
 76 }
 77 EXPORT_SYMBOL_GPL(nfs_commit_free);
 78 
 79 static struct nfs_pgio_header *nfs_writehdr_alloc(void)
 80 {
 81         struct nfs_pgio_header *p = mempool_alloc(nfs_wdata_mempool, GFP_NOIO);
 82 
 83         if (p)
 84                 memset(p, 0, sizeof(*p));
 85         return p;
 86 }
 87 
 88 static void nfs_writehdr_free(struct nfs_pgio_header *hdr)
 89 {
 90         mempool_free(hdr, nfs_wdata_mempool);
 91 }
 92 
 93 static void nfs_context_set_write_error(struct nfs_open_context *ctx, int error)
 94 {
 95         ctx->error = error;
 96         smp_wmb();
 97         set_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
 98 }
 99 
100 /*
101  * nfs_page_find_head_request_locked - find head request associated with @page
102  *
103  * must be called while holding the inode lock.
104  *
105  * returns matching head request with reference held, or NULL if not found.
106  */
107 static struct nfs_page *
108 nfs_page_find_head_request_locked(struct nfs_inode *nfsi, struct page *page)
109 {
110         struct nfs_page *req = NULL;
111 
112         if (PagePrivate(page))
113                 req = (struct nfs_page *)page_private(page);
114         else if (unlikely(PageSwapCache(page)))
115                 req = nfs_page_search_commits_for_head_request_locked(nfsi,
116                         page);
117 
118         if (req) {
119                 WARN_ON_ONCE(req->wb_head != req);
120                 kref_get(&req->wb_kref);
121         }
122 
123         return req;
124 }
125 
126 /*
127  * nfs_page_find_head_request - find head request associated with @page
128  *
129  * returns matching head request with reference held, or NULL if not found.
130  */
131 static struct nfs_page *nfs_page_find_head_request(struct page *page)
132 {
133         struct inode *inode = page_file_mapping(page)->host;
134         struct nfs_page *req = NULL;
135 
136         spin_lock(&inode->i_lock);
137         req = nfs_page_find_head_request_locked(NFS_I(inode), page);
138         spin_unlock(&inode->i_lock);
139         return req;
140 }
141 
142 /* Adjust the file length if we're writing beyond the end */
143 static void nfs_grow_file(struct page *page, unsigned int offset, unsigned int count)
144 {
145         struct inode *inode = page_file_mapping(page)->host;
146         loff_t end, i_size;
147         pgoff_t end_index;
148 
149         spin_lock(&inode->i_lock);
150         i_size = i_size_read(inode);
151         end_index = (i_size - 1) >> PAGE_CACHE_SHIFT;
152         if (i_size > 0 && page_file_index(page) < end_index)
153                 goto out;
154         end = page_file_offset(page) + ((loff_t)offset+count);
155         if (i_size >= end)
156                 goto out;
157         i_size_write(inode, end);
158         nfs_inc_stats(inode, NFSIOS_EXTENDWRITE);
159 out:
160         spin_unlock(&inode->i_lock);
161 }
162 
163 /* A writeback failed: mark the page as bad, and invalidate the page cache */
164 static void nfs_set_pageerror(struct page *page)
165 {
166         nfs_zap_mapping(page_file_mapping(page)->host, page_file_mapping(page));
167 }
168 
169 /*
170  * nfs_page_group_search_locked
171  * @head - head request of page group
172  * @page_offset - offset into page
173  *
174  * Search page group with head @head to find a request that contains the
175  * page offset @page_offset.
176  *
177  * Returns a pointer to the first matching nfs request, or NULL if no
178  * match is found.
179  *
180  * Must be called with the page group lock held
181  */
182 static struct nfs_page *
183 nfs_page_group_search_locked(struct nfs_page *head, unsigned int page_offset)
184 {
185         struct nfs_page *req;
186 
187         WARN_ON_ONCE(head != head->wb_head);
188         WARN_ON_ONCE(!test_bit(PG_HEADLOCK, &head->wb_head->wb_flags));
189 
190         req = head;
191         do {
192                 if (page_offset >= req->wb_pgbase &&
193                     page_offset < (req->wb_pgbase + req->wb_bytes))
194                         return req;
195 
196                 req = req->wb_this_page;
197         } while (req != head);
198 
199         return NULL;
200 }
201 
202 /*
203  * nfs_page_group_covers_page
204  * @head - head request of page group
205  *
206  * Return true if the page group with head @head covers the whole page,
207  * returns false otherwise
208  */
209 static bool nfs_page_group_covers_page(struct nfs_page *req)
210 {
211         struct nfs_page *tmp;
212         unsigned int pos = 0;
213         unsigned int len = nfs_page_length(req->wb_page);
214 
215         nfs_page_group_lock(req, false);
216 
217         do {
218                 tmp = nfs_page_group_search_locked(req->wb_head, pos);
219                 if (tmp) {
220                         /* no way this should happen */
221                         WARN_ON_ONCE(tmp->wb_pgbase != pos);
222                         pos += tmp->wb_bytes - (pos - tmp->wb_pgbase);
223                 }
224         } while (tmp && pos < len);
225 
226         nfs_page_group_unlock(req);
227         WARN_ON_ONCE(pos > len);
228         return pos == len;
229 }
230 
231 /* We can set the PG_uptodate flag if we see that a write request
232  * covers the full page.
233  */
234 static void nfs_mark_uptodate(struct nfs_page *req)
235 {
236         if (PageUptodate(req->wb_page))
237                 return;
238         if (!nfs_page_group_covers_page(req))
239                 return;
240         SetPageUptodate(req->wb_page);
241 }
242 
243 static int wb_priority(struct writeback_control *wbc)
244 {
245         int ret = 0;
246         if (wbc->for_reclaim)
247                 return FLUSH_HIGHPRI | FLUSH_STABLE;
248         if (wbc->sync_mode == WB_SYNC_ALL)
249                 ret = FLUSH_COND_STABLE;
250         if (wbc->for_kupdate || wbc->for_background)
251                 ret |= FLUSH_LOWPRI;
252         return ret;
253 }
254 
255 /*
256  * NFS congestion control
257  */
258 
259 int nfs_congestion_kb;
260 
261 #define NFS_CONGESTION_ON_THRESH        (nfs_congestion_kb >> (PAGE_SHIFT-10))
262 #define NFS_CONGESTION_OFF_THRESH       \
263         (NFS_CONGESTION_ON_THRESH - (NFS_CONGESTION_ON_THRESH >> 2))
264 
265 static void nfs_set_page_writeback(struct page *page)
266 {
267         struct nfs_server *nfss = NFS_SERVER(page_file_mapping(page)->host);
268         int ret = test_set_page_writeback(page);
269 
270         WARN_ON_ONCE(ret != 0);
271 
272         if (atomic_long_inc_return(&nfss->writeback) >
273                         NFS_CONGESTION_ON_THRESH) {
274                 set_bdi_congested(&nfss->backing_dev_info,
275                                         BLK_RW_ASYNC);
276         }
277 }
278 
279 static void nfs_end_page_writeback(struct nfs_page *req)
280 {
281         struct inode *inode = page_file_mapping(req->wb_page)->host;
282         struct nfs_server *nfss = NFS_SERVER(inode);
283 
284         if (!nfs_page_group_sync_on_bit(req, PG_WB_END))
285                 return;
286 
287         end_page_writeback(req->wb_page);
288         if (atomic_long_dec_return(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
289                 clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
290 }
291 
292 
293 /* nfs_page_group_clear_bits
294  *   @req - an nfs request
295  * clears all page group related bits from @req
296  */
297 static void
298 nfs_page_group_clear_bits(struct nfs_page *req)
299 {
300         clear_bit(PG_TEARDOWN, &req->wb_flags);
301         clear_bit(PG_UNLOCKPAGE, &req->wb_flags);
302         clear_bit(PG_UPTODATE, &req->wb_flags);
303         clear_bit(PG_WB_END, &req->wb_flags);
304         clear_bit(PG_REMOVE, &req->wb_flags);
305 }
306 
307 
308 /*
309  * nfs_unroll_locks_and_wait -  unlock all newly locked reqs and wait on @req
310  *
311  * this is a helper function for nfs_lock_and_join_requests
312  *
313  * @inode - inode associated with request page group, must be holding inode lock
314  * @head  - head request of page group, must be holding head lock
315  * @req   - request that couldn't lock and needs to wait on the req bit lock
316  * @nonblock - if true, don't actually wait
317  *
318  * NOTE: this must be called holding page_group bit lock and inode spin lock
319  *       and BOTH will be released before returning.
320  *
321  * returns 0 on success, < 0 on error.
322  */
323 static int
324 nfs_unroll_locks_and_wait(struct inode *inode, struct nfs_page *head,
325                           struct nfs_page *req, bool nonblock)
326         __releases(&inode->i_lock)
327 {
328         struct nfs_page *tmp;
329         int ret;
330 
331         /* relinquish all the locks successfully grabbed this run */
332         for (tmp = head ; tmp != req; tmp = tmp->wb_this_page)
333                 nfs_unlock_request(tmp);
334 
335         WARN_ON_ONCE(test_bit(PG_TEARDOWN, &req->wb_flags));
336 
337         /* grab a ref on the request that will be waited on */
338         kref_get(&req->wb_kref);
339 
340         nfs_page_group_unlock(head);
341         spin_unlock(&inode->i_lock);
342 
343         /* release ref from nfs_page_find_head_request_locked */
344         nfs_release_request(head);
345 
346         if (!nonblock)
347                 ret = nfs_wait_on_request(req);
348         else
349                 ret = -EAGAIN;
350         nfs_release_request(req);
351 
352         return ret;
353 }
354 
355 /*
356  * nfs_destroy_unlinked_subrequests - destroy recently unlinked subrequests
357  *
358  * @destroy_list - request list (using wb_this_page) terminated by @old_head
359  * @old_head - the old head of the list
360  *
361  * All subrequests must be locked and removed from all lists, so at this point
362  * they are only "active" in this function, and possibly in nfs_wait_on_request
363  * with a reference held by some other context.
364  */
365 static void
366 nfs_destroy_unlinked_subrequests(struct nfs_page *destroy_list,
367                                  struct nfs_page *old_head)
368 {
369         while (destroy_list) {
370                 struct nfs_page *subreq = destroy_list;
371 
372                 destroy_list = (subreq->wb_this_page == old_head) ?
373                                    NULL : subreq->wb_this_page;
374 
375                 WARN_ON_ONCE(old_head != subreq->wb_head);
376 
377                 /* make sure old group is not used */
378                 subreq->wb_head = subreq;
379                 subreq->wb_this_page = subreq;
380 
381                 /* subreq is now totally disconnected from page group or any
382                  * write / commit lists. last chance to wake any waiters */
383                 nfs_unlock_request(subreq);
384 
385                 if (!test_bit(PG_TEARDOWN, &subreq->wb_flags)) {
386                         /* release ref on old head request */
387                         nfs_release_request(old_head);
388 
389                         nfs_page_group_clear_bits(subreq);
390 
391                         /* release the PG_INODE_REF reference */
392                         if (test_and_clear_bit(PG_INODE_REF, &subreq->wb_flags))
393                                 nfs_release_request(subreq);
394                         else
395                                 WARN_ON_ONCE(1);
396                 } else {
397                         WARN_ON_ONCE(test_bit(PG_CLEAN, &subreq->wb_flags));
398                         /* zombie requests have already released the last
399                          * reference and were waiting on the rest of the
400                          * group to complete. Since it's no longer part of a
401                          * group, simply free the request */
402                         nfs_page_group_clear_bits(subreq);
403                         nfs_free_request(subreq);
404                 }
405         }
406 }
407 
408 /*
409  * nfs_lock_and_join_requests - join all subreqs to the head req and return
410  *                              a locked reference, cancelling any pending
411  *                              operations for this page.
412  *
413  * @page - the page used to lookup the "page group" of nfs_page structures
414  * @nonblock - if true, don't block waiting for request locks
415  *
416  * This function joins all sub requests to the head request by first
417  * locking all requests in the group, cancelling any pending operations
418  * and finally updating the head request to cover the whole range covered by
419  * the (former) group.  All subrequests are removed from any write or commit
420  * lists, unlinked from the group and destroyed.
421  *
422  * Returns a locked, referenced pointer to the head request - which after
423  * this call is guaranteed to be the only request associated with the page.
424  * Returns NULL if no requests are found for @page, or a ERR_PTR if an
425  * error was encountered.
426  */
427 static struct nfs_page *
428 nfs_lock_and_join_requests(struct page *page, bool nonblock)
429 {
430         struct inode *inode = page_file_mapping(page)->host;
431         struct nfs_page *head, *subreq;
432         struct nfs_page *destroy_list = NULL;
433         unsigned int total_bytes;
434         int ret;
435 
436 try_again:
437         total_bytes = 0;
438 
439         WARN_ON_ONCE(destroy_list);
440 
441         spin_lock(&inode->i_lock);
442 
443         /*
444          * A reference is taken only on the head request which acts as a
445          * reference to the whole page group - the group will not be destroyed
446          * until the head reference is released.
447          */
448         head = nfs_page_find_head_request_locked(NFS_I(inode), page);
449 
450         if (!head) {
451                 spin_unlock(&inode->i_lock);
452                 return NULL;
453         }
454 
455         /* holding inode lock, so always make a non-blocking call to try the
456          * page group lock */
457         ret = nfs_page_group_lock(head, true);
458         if (ret < 0) {
459                 spin_unlock(&inode->i_lock);
460 
461                 if (!nonblock && ret == -EAGAIN) {
462                         nfs_page_group_lock_wait(head);
463                         nfs_release_request(head);
464                         goto try_again;
465                 }
466 
467                 nfs_release_request(head);
468                 return ERR_PTR(ret);
469         }
470 
471         /* lock each request in the page group */
472         subreq = head;
473         do {
474                 /*
475                  * Subrequests are always contiguous, non overlapping
476                  * and in order - but may be repeated (mirrored writes).
477                  */
478                 if (subreq->wb_offset == (head->wb_offset + total_bytes)) {
479                         /* keep track of how many bytes this group covers */
480                         total_bytes += subreq->wb_bytes;
481                 } else if (WARN_ON_ONCE(subreq->wb_offset < head->wb_offset ||
482                             ((subreq->wb_offset + subreq->wb_bytes) >
483                              (head->wb_offset + total_bytes)))) {
484                         nfs_page_group_unlock(head);
485                         spin_unlock(&inode->i_lock);
486                         return ERR_PTR(-EIO);
487                 }
488 
489                 if (!nfs_lock_request(subreq)) {
490                         /* releases page group bit lock and
491                          * inode spin lock and all references */
492                         ret = nfs_unroll_locks_and_wait(inode, head,
493                                 subreq, nonblock);
494 
495                         if (ret == 0)
496                                 goto try_again;
497 
498                         return ERR_PTR(ret);
499                 }
500 
501                 subreq = subreq->wb_this_page;
502         } while (subreq != head);
503 
504         /* Now that all requests are locked, make sure they aren't on any list.
505          * Commit list removal accounting is done after locks are dropped */
506         subreq = head;
507         do {
508                 nfs_clear_request_commit(subreq);
509                 subreq = subreq->wb_this_page;
510         } while (subreq != head);
511 
512         /* unlink subrequests from head, destroy them later */
513         if (head->wb_this_page != head) {
514                 /* destroy list will be terminated by head */
515                 destroy_list = head->wb_this_page;
516                 head->wb_this_page = head;
517 
518                 /* change head request to cover whole range that
519                  * the former page group covered */
520                 head->wb_bytes = total_bytes;
521         }
522 
523         /*
524          * prepare head request to be added to new pgio descriptor
525          */
526         nfs_page_group_clear_bits(head);
527 
528         /*
529          * some part of the group was still on the inode list - otherwise
530          * the group wouldn't be involved in async write.
531          * grab a reference for the head request, iff it needs one.
532          */
533         if (!test_and_set_bit(PG_INODE_REF, &head->wb_flags))
534                 kref_get(&head->wb_kref);
535 
536         nfs_page_group_unlock(head);
537 
538         /* drop lock to clean uprequests on destroy list */
539         spin_unlock(&inode->i_lock);
540 
541         nfs_destroy_unlinked_subrequests(destroy_list, head);
542 
543         /* still holds ref on head from nfs_page_find_head_request_locked
544          * and still has lock on head from lock loop */
545         return head;
546 }
547 
548 /*
549  * Find an associated nfs write request, and prepare to flush it out
550  * May return an error if the user signalled nfs_wait_on_request().
551  */
552 static int nfs_page_async_flush(struct nfs_pageio_descriptor *pgio,
553                                 struct page *page, bool nonblock)
554 {
555         struct nfs_page *req;
556         int ret = 0;
557 
558         req = nfs_lock_and_join_requests(page, nonblock);
559         if (!req)
560                 goto out;
561         ret = PTR_ERR(req);
562         if (IS_ERR(req))
563                 goto out;
564 
565         nfs_set_page_writeback(page);
566         WARN_ON_ONCE(test_bit(PG_CLEAN, &req->wb_flags));
567 
568         ret = 0;
569         if (!nfs_pageio_add_request(pgio, req)) {
570                 nfs_redirty_request(req);
571                 ret = pgio->pg_error;
572         } else
573                 nfs_add_stats(page_file_mapping(page)->host,
574                                 NFSIOS_WRITEPAGES, 1);
575 out:
576         return ret;
577 }
578 
579 static int nfs_do_writepage(struct page *page, struct writeback_control *wbc, struct nfs_pageio_descriptor *pgio)
580 {
581         int ret;
582 
583         nfs_pageio_cond_complete(pgio, page_file_index(page));
584         ret = nfs_page_async_flush(pgio, page, wbc->sync_mode == WB_SYNC_NONE);
585         if (ret == -EAGAIN) {
586                 redirty_page_for_writepage(wbc, page);
587                 ret = 0;
588         }
589         return ret;
590 }
591 
592 /*
593  * Write an mmapped page to the server.
594  */
595 static int nfs_writepage_locked(struct page *page, struct writeback_control *wbc)
596 {
597         struct nfs_pageio_descriptor pgio;
598         struct inode *inode = page_file_mapping(page)->host;
599         int err;
600 
601         nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGE);
602         nfs_pageio_init_write(&pgio, inode, wb_priority(wbc),
603                                 false, &nfs_async_write_completion_ops);
604         err = nfs_do_writepage(page, wbc, &pgio);
605         nfs_pageio_complete(&pgio);
606         if (err < 0)
607                 return err;
608         if (pgio.pg_error < 0)
609                 return pgio.pg_error;
610         return 0;
611 }
612 
613 int nfs_writepage(struct page *page, struct writeback_control *wbc)
614 {
615         int ret;
616 
617         ret = nfs_writepage_locked(page, wbc);
618         unlock_page(page);
619         return ret;
620 }
621 
622 static int nfs_writepages_callback(struct page *page, struct writeback_control *wbc, void *data)
623 {
624         int ret;
625 
626         ret = nfs_do_writepage(page, wbc, data);
627         unlock_page(page);
628         return ret;
629 }
630 
631 int nfs_writepages(struct address_space *mapping, struct writeback_control *wbc)
632 {
633         struct inode *inode = mapping->host;
634         unsigned long *bitlock = &NFS_I(inode)->flags;
635         struct nfs_pageio_descriptor pgio;
636         int err;
637 
638         /* Stop dirtying of new pages while we sync */
639         err = wait_on_bit_lock_action(bitlock, NFS_INO_FLUSHING,
640                         nfs_wait_bit_killable, TASK_KILLABLE);
641         if (err)
642                 goto out_err;
643 
644         nfs_inc_stats(inode, NFSIOS_VFSWRITEPAGES);
645 
646         nfs_pageio_init_write(&pgio, inode, wb_priority(wbc), false,
647                                 &nfs_async_write_completion_ops);
648         err = write_cache_pages(mapping, wbc, nfs_writepages_callback, &pgio);
649         nfs_pageio_complete(&pgio);
650 
651         clear_bit_unlock(NFS_INO_FLUSHING, bitlock);
652         smp_mb__after_atomic();
653         wake_up_bit(bitlock, NFS_INO_FLUSHING);
654 
655         if (err < 0)
656                 goto out_err;
657         err = pgio.pg_error;
658         if (err < 0)
659                 goto out_err;
660         return 0;
661 out_err:
662         return err;
663 }
664 
665 /*
666  * Insert a write request into an inode
667  */
668 static void nfs_inode_add_request(struct inode *inode, struct nfs_page *req)
669 {
670         struct nfs_inode *nfsi = NFS_I(inode);
671 
672         WARN_ON_ONCE(req->wb_this_page != req);
673 
674         /* Lock the request! */
675         nfs_lock_request(req);
676 
677         spin_lock(&inode->i_lock);
678         if (!nfsi->nrequests &&
679             NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
680                 inode->i_version++;
681         /*
682          * Swap-space should not get truncated. Hence no need to plug the race
683          * with invalidate/truncate.
684          */
685         if (likely(!PageSwapCache(req->wb_page))) {
686                 set_bit(PG_MAPPED, &req->wb_flags);
687                 SetPagePrivate(req->wb_page);
688                 set_page_private(req->wb_page, (unsigned long)req);
689         }
690         nfsi->nrequests++;
691         /* this a head request for a page group - mark it as having an
692          * extra reference so sub groups can follow suit.
693          * This flag also informs pgio layer when to bump nrequests when
694          * adding subrequests. */
695         WARN_ON(test_and_set_bit(PG_INODE_REF, &req->wb_flags));
696         kref_get(&req->wb_kref);
697         spin_unlock(&inode->i_lock);
698 }
699 
700 /*
701  * Remove a write request from an inode
702  */
703 static void nfs_inode_remove_request(struct nfs_page *req)
704 {
705         struct inode *inode = d_inode(req->wb_context->dentry);
706         struct nfs_inode *nfsi = NFS_I(inode);
707         struct nfs_page *head;
708 
709         if (nfs_page_group_sync_on_bit(req, PG_REMOVE)) {
710                 head = req->wb_head;
711 
712                 spin_lock(&inode->i_lock);
713                 if (likely(!PageSwapCache(head->wb_page))) {
714                         set_page_private(head->wb_page, 0);
715                         ClearPagePrivate(head->wb_page);
716                         smp_mb__after_atomic();
717                         wake_up_page(head->wb_page, PG_private);
718                         clear_bit(PG_MAPPED, &head->wb_flags);
719                 }
720                 nfsi->nrequests--;
721                 spin_unlock(&inode->i_lock);
722         } else {
723                 spin_lock(&inode->i_lock);
724                 nfsi->nrequests--;
725                 spin_unlock(&inode->i_lock);
726         }
727 
728         if (test_and_clear_bit(PG_INODE_REF, &req->wb_flags))
729                 nfs_release_request(req);
730 }
731 
732 static void
733 nfs_mark_request_dirty(struct nfs_page *req)
734 {
735         __set_page_dirty_nobuffers(req->wb_page);
736 }
737 
738 /*
739  * nfs_page_search_commits_for_head_request_locked
740  *
741  * Search through commit lists on @inode for the head request for @page.
742  * Must be called while holding the inode (which is cinfo) lock.
743  *
744  * Returns the head request if found, or NULL if not found.
745  */
746 static struct nfs_page *
747 nfs_page_search_commits_for_head_request_locked(struct nfs_inode *nfsi,
748                                                 struct page *page)
749 {
750         struct nfs_page *freq, *t;
751         struct nfs_commit_info cinfo;
752         struct inode *inode = &nfsi->vfs_inode;
753 
754         nfs_init_cinfo_from_inode(&cinfo, inode);
755 
756         /* search through pnfs commit lists */
757         freq = pnfs_search_commit_reqs(inode, &cinfo, page);
758         if (freq)
759                 return freq->wb_head;
760 
761         /* Linearly search the commit list for the correct request */
762         list_for_each_entry_safe(freq, t, &cinfo.mds->list, wb_list) {
763                 if (freq->wb_page == page)
764                         return freq->wb_head;
765         }
766 
767         return NULL;
768 }
769 
770 /**
771  * nfs_request_add_commit_list_locked - add request to a commit list
772  * @req: pointer to a struct nfs_page
773  * @dst: commit list head
774  * @cinfo: holds list lock and accounting info
775  *
776  * This sets the PG_CLEAN bit, updates the cinfo count of
777  * number of outstanding requests requiring a commit as well as
778  * the MM page stats.
779  *
780  * The caller must hold the cinfo->lock, and the nfs_page lock.
781  */
782 void
783 nfs_request_add_commit_list_locked(struct nfs_page *req, struct list_head *dst,
784                             struct nfs_commit_info *cinfo)
785 {
786         set_bit(PG_CLEAN, &req->wb_flags);
787         nfs_list_add_request(req, dst);
788         cinfo->mds->ncommit++;
789 }
790 EXPORT_SYMBOL_GPL(nfs_request_add_commit_list_locked);
791 
792 /**
793  * nfs_request_add_commit_list - add request to a commit list
794  * @req: pointer to a struct nfs_page
795  * @dst: commit list head
796  * @cinfo: holds list lock and accounting info
797  *
798  * This sets the PG_CLEAN bit, updates the cinfo count of
799  * number of outstanding requests requiring a commit as well as
800  * the MM page stats.
801  *
802  * The caller must _not_ hold the cinfo->lock, but must be
803  * holding the nfs_page lock.
804  */
805 void
806 nfs_request_add_commit_list(struct nfs_page *req, struct list_head *dst,
807                             struct nfs_commit_info *cinfo)
808 {
809         spin_lock(cinfo->lock);
810         nfs_request_add_commit_list_locked(req, dst, cinfo);
811         spin_unlock(cinfo->lock);
812         nfs_mark_page_unstable(req->wb_page, cinfo);
813 }
814 EXPORT_SYMBOL_GPL(nfs_request_add_commit_list);
815 
816 /**
817  * nfs_request_remove_commit_list - Remove request from a commit list
818  * @req: pointer to a nfs_page
819  * @cinfo: holds list lock and accounting info
820  *
821  * This clears the PG_CLEAN bit, and updates the cinfo's count of
822  * number of outstanding requests requiring a commit
823  * It does not update the MM page stats.
824  *
825  * The caller _must_ hold the cinfo->lock and the nfs_page lock.
826  */
827 void
828 nfs_request_remove_commit_list(struct nfs_page *req,
829                                struct nfs_commit_info *cinfo)
830 {
831         if (!test_and_clear_bit(PG_CLEAN, &(req)->wb_flags))
832                 return;
833         nfs_list_remove_request(req);
834         cinfo->mds->ncommit--;
835 }
836 EXPORT_SYMBOL_GPL(nfs_request_remove_commit_list);
837 
838 static void nfs_init_cinfo_from_inode(struct nfs_commit_info *cinfo,
839                                       struct inode *inode)
840 {
841         cinfo->lock = &inode->i_lock;
842         cinfo->mds = &NFS_I(inode)->commit_info;
843         cinfo->ds = pnfs_get_ds_info(inode);
844         cinfo->dreq = NULL;
845         cinfo->completion_ops = &nfs_commit_completion_ops;
846 }
847 
848 void nfs_init_cinfo(struct nfs_commit_info *cinfo,
849                     struct inode *inode,
850                     struct nfs_direct_req *dreq)
851 {
852         if (dreq)
853                 nfs_init_cinfo_from_dreq(cinfo, dreq);
854         else
855                 nfs_init_cinfo_from_inode(cinfo, inode);
856 }
857 EXPORT_SYMBOL_GPL(nfs_init_cinfo);
858 
859 /*
860  * Add a request to the inode's commit list.
861  */
862 void
863 nfs_mark_request_commit(struct nfs_page *req, struct pnfs_layout_segment *lseg,
864                         struct nfs_commit_info *cinfo, u32 ds_commit_idx)
865 {
866         if (pnfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx))
867                 return;
868         nfs_request_add_commit_list(req, &cinfo->mds->list, cinfo);
869 }
870 
871 static void
872 nfs_clear_page_commit(struct page *page)
873 {
874         dec_zone_page_state(page, NR_UNSTABLE_NFS);
875         dec_wb_stat(&inode_to_bdi(page_file_mapping(page)->host)->wb,
876                     WB_RECLAIMABLE);
877 }
878 
879 /* Called holding inode (/cinfo) lock */
880 static void
881 nfs_clear_request_commit(struct nfs_page *req)
882 {
883         if (test_bit(PG_CLEAN, &req->wb_flags)) {
884                 struct inode *inode = d_inode(req->wb_context->dentry);
885                 struct nfs_commit_info cinfo;
886 
887                 nfs_init_cinfo_from_inode(&cinfo, inode);
888                 if (!pnfs_clear_request_commit(req, &cinfo)) {
889                         nfs_request_remove_commit_list(req, &cinfo);
890                 }
891                 nfs_clear_page_commit(req->wb_page);
892         }
893 }
894 
895 int nfs_write_need_commit(struct nfs_pgio_header *hdr)
896 {
897         if (hdr->verf.committed == NFS_DATA_SYNC)
898                 return hdr->lseg == NULL;
899         return hdr->verf.committed != NFS_FILE_SYNC;
900 }
901 
902 static void nfs_write_completion(struct nfs_pgio_header *hdr)
903 {
904         struct nfs_commit_info cinfo;
905         unsigned long bytes = 0;
906 
907         if (test_bit(NFS_IOHDR_REDO, &hdr->flags))
908                 goto out;
909         nfs_init_cinfo_from_inode(&cinfo, hdr->inode);
910         while (!list_empty(&hdr->pages)) {
911                 struct nfs_page *req = nfs_list_entry(hdr->pages.next);
912 
913                 bytes += req->wb_bytes;
914                 nfs_list_remove_request(req);
915                 if (test_bit(NFS_IOHDR_ERROR, &hdr->flags) &&
916                     (hdr->good_bytes < bytes)) {
917                         nfs_set_pageerror(req->wb_page);
918                         nfs_context_set_write_error(req->wb_context, hdr->error);
919                         goto remove_req;
920                 }
921                 if (nfs_write_need_commit(hdr)) {
922                         memcpy(&req->wb_verf, &hdr->verf.verifier, sizeof(req->wb_verf));
923                         nfs_mark_request_commit(req, hdr->lseg, &cinfo,
924                                 hdr->pgio_mirror_idx);
925                         goto next;
926                 }
927 remove_req:
928                 nfs_inode_remove_request(req);
929 next:
930                 nfs_unlock_request(req);
931                 nfs_end_page_writeback(req);
932                 nfs_release_request(req);
933         }
934 out:
935         hdr->release(hdr);
936 }
937 
938 unsigned long
939 nfs_reqs_to_commit(struct nfs_commit_info *cinfo)
940 {
941         return cinfo->mds->ncommit;
942 }
943 
944 /* cinfo->lock held by caller */
945 int
946 nfs_scan_commit_list(struct list_head *src, struct list_head *dst,
947                      struct nfs_commit_info *cinfo, int max)
948 {
949         struct nfs_page *req, *tmp;
950         int ret = 0;
951 
952         list_for_each_entry_safe(req, tmp, src, wb_list) {
953                 if (!nfs_lock_request(req))
954                         continue;
955                 kref_get(&req->wb_kref);
956                 if (cond_resched_lock(cinfo->lock))
957                         list_safe_reset_next(req, tmp, wb_list);
958                 nfs_request_remove_commit_list(req, cinfo);
959                 nfs_list_add_request(req, dst);
960                 ret++;
961                 if ((ret == max) && !cinfo->dreq)
962                         break;
963         }
964         return ret;
965 }
966 
967 /*
968  * nfs_scan_commit - Scan an inode for commit requests
969  * @inode: NFS inode to scan
970  * @dst: mds destination list
971  * @cinfo: mds and ds lists of reqs ready to commit
972  *
973  * Moves requests from the inode's 'commit' request list.
974  * The requests are *not* checked to ensure that they form a contiguous set.
975  */
976 int
977 nfs_scan_commit(struct inode *inode, struct list_head *dst,
978                 struct nfs_commit_info *cinfo)
979 {
980         int ret = 0;
981 
982         spin_lock(cinfo->lock);
983         if (cinfo->mds->ncommit > 0) {
984                 const int max = INT_MAX;
985 
986                 ret = nfs_scan_commit_list(&cinfo->mds->list, dst,
987                                            cinfo, max);
988                 ret += pnfs_scan_commit_lists(inode, cinfo, max - ret);
989         }
990         spin_unlock(cinfo->lock);
991         return ret;
992 }
993 
994 /*
995  * Search for an existing write request, and attempt to update
996  * it to reflect a new dirty region on a given page.
997  *
998  * If the attempt fails, then the existing request is flushed out
999  * to disk.
1000  */
1001 static struct nfs_page *nfs_try_to_update_request(struct inode *inode,
1002                 struct page *page,
1003                 unsigned int offset,
1004                 unsigned int bytes)
1005 {
1006         struct nfs_page *req;
1007         unsigned int rqend;
1008         unsigned int end;
1009         int error;
1010 
1011         if (!PagePrivate(page))
1012                 return NULL;
1013 
1014         end = offset + bytes;
1015         spin_lock(&inode->i_lock);
1016 
1017         for (;;) {
1018                 req = nfs_page_find_head_request_locked(NFS_I(inode), page);
1019                 if (req == NULL)
1020                         goto out_unlock;
1021 
1022                 /* should be handled by nfs_flush_incompatible */
1023                 WARN_ON_ONCE(req->wb_head != req);
1024                 WARN_ON_ONCE(req->wb_this_page != req);
1025 
1026                 rqend = req->wb_offset + req->wb_bytes;
1027                 /*
1028                  * Tell the caller to flush out the request if
1029                  * the offsets are non-contiguous.
1030                  * Note: nfs_flush_incompatible() will already
1031                  * have flushed out requests having wrong owners.
1032                  */
1033                 if (offset > rqend
1034                     || end < req->wb_offset)
1035                         goto out_flushme;
1036 
1037                 if (nfs_lock_request(req))
1038                         break;
1039 
1040                 /* The request is locked, so wait and then retry */
1041                 spin_unlock(&inode->i_lock);
1042                 error = nfs_wait_on_request(req);
1043                 nfs_release_request(req);
1044                 if (error != 0)
1045                         goto out_err;
1046                 spin_lock(&inode->i_lock);
1047         }
1048 
1049         /* Okay, the request matches. Update the region */
1050         if (offset < req->wb_offset) {
1051                 req->wb_offset = offset;
1052                 req->wb_pgbase = offset;
1053         }
1054         if (end > rqend)
1055                 req->wb_bytes = end - req->wb_offset;
1056         else
1057                 req->wb_bytes = rqend - req->wb_offset;
1058 out_unlock:
1059         if (req)
1060                 nfs_clear_request_commit(req);
1061         spin_unlock(&inode->i_lock);
1062         return req;
1063 out_flushme:
1064         spin_unlock(&inode->i_lock);
1065         nfs_release_request(req);
1066         error = nfs_wb_page(inode, page);
1067 out_err:
1068         return ERR_PTR(error);
1069 }
1070 
1071 /*
1072  * Try to update an existing write request, or create one if there is none.
1073  *
1074  * Note: Should always be called with the Page Lock held to prevent races
1075  * if we have to add a new request. Also assumes that the caller has
1076  * already called nfs_flush_incompatible() if necessary.
1077  */
1078 static struct nfs_page * nfs_setup_write_request(struct nfs_open_context* ctx,
1079                 struct page *page, unsigned int offset, unsigned int bytes)
1080 {
1081         struct inode *inode = page_file_mapping(page)->host;
1082         struct nfs_page *req;
1083 
1084         req = nfs_try_to_update_request(inode, page, offset, bytes);
1085         if (req != NULL)
1086                 goto out;
1087         req = nfs_create_request(ctx, page, NULL, offset, bytes);
1088         if (IS_ERR(req))
1089                 goto out;
1090         nfs_inode_add_request(inode, req);
1091 out:
1092         return req;
1093 }
1094 
1095 static int nfs_writepage_setup(struct nfs_open_context *ctx, struct page *page,
1096                 unsigned int offset, unsigned int count)
1097 {
1098         struct nfs_page *req;
1099 
1100         req = nfs_setup_write_request(ctx, page, offset, count);
1101         if (IS_ERR(req))
1102                 return PTR_ERR(req);
1103         /* Update file length */
1104         nfs_grow_file(page, offset, count);
1105         nfs_mark_uptodate(req);
1106         nfs_mark_request_dirty(req);
1107         nfs_unlock_and_release_request(req);
1108         return 0;
1109 }
1110 
1111 int nfs_flush_incompatible(struct file *file, struct page *page)
1112 {
1113         struct nfs_open_context *ctx = nfs_file_open_context(file);
1114         struct nfs_lock_context *l_ctx;
1115         struct file_lock_context *flctx = file_inode(file)->i_flctx;
1116         struct nfs_page *req;
1117         int do_flush, status;
1118         /*
1119          * Look for a request corresponding to this page. If there
1120          * is one, and it belongs to another file, we flush it out
1121          * before we try to copy anything into the page. Do this
1122          * due to the lack of an ACCESS-type call in NFSv2.
1123          * Also do the same if we find a request from an existing
1124          * dropped page.
1125          */
1126         do {
1127                 req = nfs_page_find_head_request(page);
1128                 if (req == NULL)
1129                         return 0;
1130                 l_ctx = req->wb_lock_context;
1131                 do_flush = req->wb_page != page || req->wb_context != ctx;
1132                 /* for now, flush if more than 1 request in page_group */
1133                 do_flush |= req->wb_this_page != req;
1134                 if (l_ctx && flctx &&
1135                     !(list_empty_careful(&flctx->flc_posix) &&
1136                       list_empty_careful(&flctx->flc_flock))) {
1137                         do_flush |= l_ctx->lockowner.l_owner != current->files
1138                                 || l_ctx->lockowner.l_pid != current->tgid;
1139                 }
1140                 nfs_release_request(req);
1141                 if (!do_flush)
1142                         return 0;
1143                 status = nfs_wb_page(page_file_mapping(page)->host, page);
1144         } while (status == 0);
1145         return status;
1146 }
1147 
1148 /*
1149  * Avoid buffered writes when a open context credential's key would
1150  * expire soon.
1151  *
1152  * Returns -EACCES if the key will expire within RPC_KEY_EXPIRE_FAIL.
1153  *
1154  * Return 0 and set a credential flag which triggers the inode to flush
1155  * and performs  NFS_FILE_SYNC writes if the key will expired within
1156  * RPC_KEY_EXPIRE_TIMEO.
1157  */
1158 int
1159 nfs_key_timeout_notify(struct file *filp, struct inode *inode)
1160 {
1161         struct nfs_open_context *ctx = nfs_file_open_context(filp);
1162         struct rpc_auth *auth = NFS_SERVER(inode)->client->cl_auth;
1163 
1164         return rpcauth_key_timeout_notify(auth, ctx->cred);
1165 }
1166 
1167 /*
1168  * Test if the open context credential key is marked to expire soon.
1169  */
1170 bool nfs_ctx_key_to_expire(struct nfs_open_context *ctx)
1171 {
1172         return rpcauth_cred_key_to_expire(ctx->cred);
1173 }
1174 
1175 /*
1176  * If the page cache is marked as unsafe or invalid, then we can't rely on
1177  * the PageUptodate() flag. In this case, we will need to turn off
1178  * write optimisations that depend on the page contents being correct.
1179  */
1180 static bool nfs_write_pageuptodate(struct page *page, struct inode *inode)
1181 {
1182         struct nfs_inode *nfsi = NFS_I(inode);
1183 
1184         if (nfs_have_delegated_attributes(inode))
1185                 goto out;
1186         if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
1187                 return false;
1188         smp_rmb();
1189         if (test_bit(NFS_INO_INVALIDATING, &nfsi->flags))
1190                 return false;
1191 out:
1192         if (nfsi->cache_validity & NFS_INO_INVALID_DATA)
1193                 return false;
1194         return PageUptodate(page) != 0;
1195 }
1196 
1197 static bool
1198 is_whole_file_wrlock(struct file_lock *fl)
1199 {
1200         return fl->fl_start == 0 && fl->fl_end == OFFSET_MAX &&
1201                         fl->fl_type == F_WRLCK;
1202 }
1203 
1204 /* If we know the page is up to date, and we're not using byte range locks (or
1205  * if we have the whole file locked for writing), it may be more efficient to
1206  * extend the write to cover the entire page in order to avoid fragmentation
1207  * inefficiencies.
1208  *
1209  * If the file is opened for synchronous writes then we can just skip the rest
1210  * of the checks.
1211  */
1212 static int nfs_can_extend_write(struct file *file, struct page *page, struct inode *inode)
1213 {
1214         int ret;
1215         struct file_lock_context *flctx = inode->i_flctx;
1216         struct file_lock *fl;
1217 
1218         if (file->f_flags & O_DSYNC)
1219                 return 0;
1220         if (!nfs_write_pageuptodate(page, inode))
1221                 return 0;
1222         if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE))
1223                 return 1;
1224         if (!flctx || (list_empty_careful(&flctx->flc_flock) &&
1225                        list_empty_careful(&flctx->flc_posix)))
1226                 return 1;
1227 
1228         /* Check to see if there are whole file write locks */
1229         ret = 0;
1230         spin_lock(&flctx->flc_lock);
1231         if (!list_empty(&flctx->flc_posix)) {
1232                 fl = list_first_entry(&flctx->flc_posix, struct file_lock,
1233                                         fl_list);
1234                 if (is_whole_file_wrlock(fl))
1235                         ret = 1;
1236         } else if (!list_empty(&flctx->flc_flock)) {
1237                 fl = list_first_entry(&flctx->flc_flock, struct file_lock,
1238                                         fl_list);
1239                 if (fl->fl_type == F_WRLCK)
1240                         ret = 1;
1241         }
1242         spin_unlock(&flctx->flc_lock);
1243         return ret;
1244 }
1245 
1246 /*
1247  * Update and possibly write a cached page of an NFS file.
1248  *
1249  * XXX: Keep an eye on generic_file_read to make sure it doesn't do bad
1250  * things with a page scheduled for an RPC call (e.g. invalidate it).
1251  */
1252 int nfs_updatepage(struct file *file, struct page *page,
1253                 unsigned int offset, unsigned int count)
1254 {
1255         struct nfs_open_context *ctx = nfs_file_open_context(file);
1256         struct inode    *inode = page_file_mapping(page)->host;
1257         int             status = 0;
1258 
1259         nfs_inc_stats(inode, NFSIOS_VFSUPDATEPAGE);
1260 
1261         dprintk("NFS:       nfs_updatepage(%pD2 %d@%lld)\n",
1262                 file, count, (long long)(page_file_offset(page) + offset));
1263 
1264         if (nfs_can_extend_write(file, page, inode)) {
1265                 count = max(count + offset, nfs_page_length(page));
1266                 offset = 0;
1267         }
1268 
1269         status = nfs_writepage_setup(ctx, page, offset, count);
1270         if (status < 0)
1271                 nfs_set_pageerror(page);
1272         else
1273                 __set_page_dirty_nobuffers(page);
1274 
1275         dprintk("NFS:       nfs_updatepage returns %d (isize %lld)\n",
1276                         status, (long long)i_size_read(inode));
1277         return status;
1278 }
1279 
1280 static int flush_task_priority(int how)
1281 {
1282         switch (how & (FLUSH_HIGHPRI|FLUSH_LOWPRI)) {
1283                 case FLUSH_HIGHPRI:
1284                         return RPC_PRIORITY_HIGH;
1285                 case FLUSH_LOWPRI:
1286                         return RPC_PRIORITY_LOW;
1287         }
1288         return RPC_PRIORITY_NORMAL;
1289 }
1290 
1291 static void nfs_initiate_write(struct nfs_pgio_header *hdr,
1292                                struct rpc_message *msg,
1293                                const struct nfs_rpc_ops *rpc_ops,
1294                                struct rpc_task_setup *task_setup_data, int how)
1295 {
1296         int priority = flush_task_priority(how);
1297 
1298         task_setup_data->priority = priority;
1299         rpc_ops->write_setup(hdr, msg);
1300 
1301         nfs4_state_protect_write(NFS_SERVER(hdr->inode)->nfs_client,
1302                                  &task_setup_data->rpc_client, msg, hdr);
1303 }
1304 
1305 /* If a nfs_flush_* function fails, it should remove reqs from @head and
1306  * call this on each, which will prepare them to be retried on next
1307  * writeback using standard nfs.
1308  */
1309 static void nfs_redirty_request(struct nfs_page *req)
1310 {
1311         nfs_mark_request_dirty(req);
1312         set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags);
1313         nfs_unlock_request(req);
1314         nfs_end_page_writeback(req);
1315         nfs_release_request(req);
1316 }
1317 
1318 static void nfs_async_write_error(struct list_head *head)
1319 {
1320         struct nfs_page *req;
1321 
1322         while (!list_empty(head)) {
1323                 req = nfs_list_entry(head->next);
1324                 nfs_list_remove_request(req);
1325                 nfs_redirty_request(req);
1326         }
1327 }
1328 
1329 static const struct nfs_pgio_completion_ops nfs_async_write_completion_ops = {
1330         .error_cleanup = nfs_async_write_error,
1331         .completion = nfs_write_completion,
1332 };
1333 
1334 void nfs_pageio_init_write(struct nfs_pageio_descriptor *pgio,
1335                                struct inode *inode, int ioflags, bool force_mds,
1336                                const struct nfs_pgio_completion_ops *compl_ops)
1337 {
1338         struct nfs_server *server = NFS_SERVER(inode);
1339         const struct nfs_pageio_ops *pg_ops = &nfs_pgio_rw_ops;
1340 
1341 #ifdef CONFIG_NFS_V4_1
1342         if (server->pnfs_curr_ld && !force_mds)
1343                 pg_ops = server->pnfs_curr_ld->pg_write_ops;
1344 #endif
1345         nfs_pageio_init(pgio, inode, pg_ops, compl_ops, &nfs_rw_write_ops,
1346                         server->wsize, ioflags);
1347 }
1348 EXPORT_SYMBOL_GPL(nfs_pageio_init_write);
1349 
1350 void nfs_pageio_reset_write_mds(struct nfs_pageio_descriptor *pgio)
1351 {
1352         struct nfs_pgio_mirror *mirror;
1353 
1354         if (pgio->pg_ops && pgio->pg_ops->pg_cleanup)
1355                 pgio->pg_ops->pg_cleanup(pgio);
1356 
1357         pgio->pg_ops = &nfs_pgio_rw_ops;
1358 
1359         nfs_pageio_stop_mirroring(pgio);
1360 
1361         mirror = &pgio->pg_mirrors[0];
1362         mirror->pg_bsize = NFS_SERVER(pgio->pg_inode)->wsize;
1363 }
1364 EXPORT_SYMBOL_GPL(nfs_pageio_reset_write_mds);
1365 
1366 
1367 void nfs_commit_prepare(struct rpc_task *task, void *calldata)
1368 {
1369         struct nfs_commit_data *data = calldata;
1370 
1371         NFS_PROTO(data->inode)->commit_rpc_prepare(task, data);
1372 }
1373 
1374 /*
1375  * Special version of should_remove_suid() that ignores capabilities.
1376  */
1377 static int nfs_should_remove_suid(const struct inode *inode)
1378 {
1379         umode_t mode = inode->i_mode;
1380         int kill = 0;
1381 
1382         /* suid always must be killed */
1383         if (unlikely(mode & S_ISUID))
1384                 kill = ATTR_KILL_SUID;
1385 
1386         /*
1387          * sgid without any exec bits is just a mandatory locking mark; leave
1388          * it alone.  If some exec bits are set, it's a real sgid; kill it.
1389          */
1390         if (unlikely((mode & S_ISGID) && (mode & S_IXGRP)))
1391                 kill |= ATTR_KILL_SGID;
1392 
1393         if (unlikely(kill && S_ISREG(mode)))
1394                 return kill;
1395 
1396         return 0;
1397 }
1398 
1399 static void nfs_writeback_check_extend(struct nfs_pgio_header *hdr,
1400                 struct nfs_fattr *fattr)
1401 {
1402         struct nfs_pgio_args *argp = &hdr->args;
1403         struct nfs_pgio_res *resp = &hdr->res;
1404         u64 size = argp->offset + resp->count;
1405 
1406         if (!(fattr->valid & NFS_ATTR_FATTR_SIZE))
1407                 fattr->size = size;
1408         if (nfs_size_to_loff_t(fattr->size) < i_size_read(hdr->inode)) {
1409                 fattr->valid &= ~NFS_ATTR_FATTR_SIZE;
1410                 return;
1411         }
1412         if (size != fattr->size)
1413                 return;
1414         /* Set attribute barrier */
1415         nfs_fattr_set_barrier(fattr);
1416         /* ...and update size */
1417         fattr->valid |= NFS_ATTR_FATTR_SIZE;
1418 }
1419 
1420 void nfs_writeback_update_inode(struct nfs_pgio_header *hdr)
1421 {
1422         struct nfs_fattr *fattr = &hdr->fattr;
1423         struct inode *inode = hdr->inode;
1424 
1425         spin_lock(&inode->i_lock);
1426         nfs_writeback_check_extend(hdr, fattr);
1427         nfs_post_op_update_inode_force_wcc_locked(inode, fattr);
1428         spin_unlock(&inode->i_lock);
1429 }
1430 EXPORT_SYMBOL_GPL(nfs_writeback_update_inode);
1431 
1432 /*
1433  * This function is called when the WRITE call is complete.
1434  */
1435 static int nfs_writeback_done(struct rpc_task *task,
1436                               struct nfs_pgio_header *hdr,
1437                               struct inode *inode)
1438 {
1439         int status;
1440 
1441         /*
1442          * ->write_done will attempt to use post-op attributes to detect
1443          * conflicting writes by other clients.  A strict interpretation
1444          * of close-to-open would allow us to continue caching even if
1445          * another writer had changed the file, but some applications
1446          * depend on tighter cache coherency when writing.
1447          */
1448         status = NFS_PROTO(inode)->write_done(task, hdr);
1449         if (status != 0)
1450                 return status;
1451         nfs_add_stats(inode, NFSIOS_SERVERWRITTENBYTES, hdr->res.count);
1452 
1453         if (hdr->res.verf->committed < hdr->args.stable &&
1454             task->tk_status >= 0) {
1455                 /* We tried a write call, but the server did not
1456                  * commit data to stable storage even though we
1457                  * requested it.
1458                  * Note: There is a known bug in Tru64 < 5.0 in which
1459                  *       the server reports NFS_DATA_SYNC, but performs
1460                  *       NFS_FILE_SYNC. We therefore implement this checking
1461                  *       as a dprintk() in order to avoid filling syslog.
1462                  */
1463                 static unsigned long    complain;
1464 
1465                 /* Note this will print the MDS for a DS write */
1466                 if (time_before(complain, jiffies)) {
1467                         dprintk("NFS:       faulty NFS server %s:"
1468                                 " (committed = %d) != (stable = %d)\n",
1469                                 NFS_SERVER(inode)->nfs_client->cl_hostname,
1470                                 hdr->res.verf->committed, hdr->args.stable);
1471                         complain = jiffies + 300 * HZ;
1472                 }
1473         }
1474 
1475         /* Deal with the suid/sgid bit corner case */
1476         if (nfs_should_remove_suid(inode))
1477                 nfs_mark_for_revalidate(inode);
1478         return 0;
1479 }
1480 
1481 /*
1482  * This function is called when the WRITE call is complete.
1483  */
1484 static void nfs_writeback_result(struct rpc_task *task,
1485                                  struct nfs_pgio_header *hdr)
1486 {
1487         struct nfs_pgio_args    *argp = &hdr->args;
1488         struct nfs_pgio_res     *resp = &hdr->res;
1489 
1490         if (resp->count < argp->count) {
1491                 static unsigned long    complain;
1492 
1493                 /* This a short write! */
1494                 nfs_inc_stats(hdr->inode, NFSIOS_SHORTWRITE);
1495 
1496                 /* Has the server at least made some progress? */
1497                 if (resp->count == 0) {
1498                         if (time_before(complain, jiffies)) {
1499                                 printk(KERN_WARNING
1500                                        "NFS: Server wrote zero bytes, expected %u.\n",
1501                                        argp->count);
1502                                 complain = jiffies + 300 * HZ;
1503                         }
1504                         nfs_set_pgio_error(hdr, -EIO, argp->offset);
1505                         task->tk_status = -EIO;
1506                         return;
1507                 }
1508                 /* Was this an NFSv2 write or an NFSv3 stable write? */
1509                 if (resp->verf->committed != NFS_UNSTABLE) {
1510                         /* Resend from where the server left off */
1511                         hdr->mds_offset += resp->count;
1512                         argp->offset += resp->count;
1513                         argp->pgbase += resp->count;
1514                         argp->count -= resp->count;
1515                 } else {
1516                         /* Resend as a stable write in order to avoid
1517                          * headaches in the case of a server crash.
1518                          */
1519                         argp->stable = NFS_FILE_SYNC;
1520                 }
1521                 rpc_restart_call_prepare(task);
1522         }
1523 }
1524 
1525 
1526 static int nfs_commit_set_lock(struct nfs_inode *nfsi, int may_wait)
1527 {
1528         int ret;
1529 
1530         if (!test_and_set_bit(NFS_INO_COMMIT, &nfsi->flags))
1531                 return 1;
1532         if (!may_wait)
1533                 return 0;
1534         ret = out_of_line_wait_on_bit_lock(&nfsi->flags,
1535                                 NFS_INO_COMMIT,
1536                                 nfs_wait_bit_killable,
1537                                 TASK_KILLABLE);
1538         return (ret < 0) ? ret : 1;
1539 }
1540 
1541 static void nfs_commit_clear_lock(struct nfs_inode *nfsi)
1542 {
1543         clear_bit(NFS_INO_COMMIT, &nfsi->flags);
1544         smp_mb__after_atomic();
1545         wake_up_bit(&nfsi->flags, NFS_INO_COMMIT);
1546 }
1547 
1548 void nfs_commitdata_release(struct nfs_commit_data *data)
1549 {
1550         put_nfs_open_context(data->context);
1551         nfs_commit_free(data);
1552 }
1553 EXPORT_SYMBOL_GPL(nfs_commitdata_release);
1554 
1555 int nfs_initiate_commit(struct rpc_clnt *clnt, struct nfs_commit_data *data,
1556                         const struct nfs_rpc_ops *nfs_ops,
1557                         const struct rpc_call_ops *call_ops,
1558                         int how, int flags)
1559 {
1560         struct rpc_task *task;
1561         int priority = flush_task_priority(how);
1562         struct rpc_message msg = {
1563                 .rpc_argp = &data->args,
1564                 .rpc_resp = &data->res,
1565                 .rpc_cred = data->cred,
1566         };
1567         struct rpc_task_setup task_setup_data = {
1568                 .task = &data->task,
1569                 .rpc_client = clnt,
1570                 .rpc_message = &msg,
1571                 .callback_ops = call_ops,
1572                 .callback_data = data,
1573                 .workqueue = nfsiod_workqueue,
1574                 .flags = RPC_TASK_ASYNC | flags,
1575                 .priority = priority,
1576         };
1577         /* Set up the initial task struct.  */
1578         nfs_ops->commit_setup(data, &msg);
1579 
1580         dprintk("NFS: initiated commit call\n");
1581 
1582         nfs4_state_protect(NFS_SERVER(data->inode)->nfs_client,
1583                 NFS_SP4_MACH_CRED_COMMIT, &task_setup_data.rpc_client, &msg);
1584 
1585         task = rpc_run_task(&task_setup_data);
1586         if (IS_ERR(task))
1587                 return PTR_ERR(task);
1588         if (how & FLUSH_SYNC)
1589                 rpc_wait_for_completion_task(task);
1590         rpc_put_task(task);
1591         return 0;
1592 }
1593 EXPORT_SYMBOL_GPL(nfs_initiate_commit);
1594 
1595 static loff_t nfs_get_lwb(struct list_head *head)
1596 {
1597         loff_t lwb = 0;
1598         struct nfs_page *req;
1599 
1600         list_for_each_entry(req, head, wb_list)
1601                 if (lwb < (req_offset(req) + req->wb_bytes))
1602                         lwb = req_offset(req) + req->wb_bytes;
1603 
1604         return lwb;
1605 }
1606 
1607 /*
1608  * Set up the argument/result storage required for the RPC call.
1609  */
1610 void nfs_init_commit(struct nfs_commit_data *data,
1611                      struct list_head *head,
1612                      struct pnfs_layout_segment *lseg,
1613                      struct nfs_commit_info *cinfo)
1614 {
1615         struct nfs_page *first = nfs_list_entry(head->next);
1616         struct inode *inode = d_inode(first->wb_context->dentry);
1617 
1618         /* Set up the RPC argument and reply structs
1619          * NB: take care not to mess about with data->commit et al. */
1620 
1621         list_splice_init(head, &data->pages);
1622 
1623         data->inode       = inode;
1624         data->cred        = first->wb_context->cred;
1625         data->lseg        = lseg; /* reference transferred */
1626         /* only set lwb for pnfs commit */
1627         if (lseg)
1628                 data->lwb = nfs_get_lwb(&data->pages);
1629         data->mds_ops     = &nfs_commit_ops;
1630         data->completion_ops = cinfo->completion_ops;
1631         data->dreq        = cinfo->dreq;
1632 
1633         data->args.fh     = NFS_FH(data->inode);
1634         /* Note: we always request a commit of the entire inode */
1635         data->args.offset = 0;
1636         data->args.count  = 0;
1637         data->context     = get_nfs_open_context(first->wb_context);
1638         data->res.fattr   = &data->fattr;
1639         data->res.verf    = &data->verf;
1640         nfs_fattr_init(&data->fattr);
1641 }
1642 EXPORT_SYMBOL_GPL(nfs_init_commit);
1643 
1644 void nfs_retry_commit(struct list_head *page_list,
1645                       struct pnfs_layout_segment *lseg,
1646                       struct nfs_commit_info *cinfo,
1647                       u32 ds_commit_idx)
1648 {
1649         struct nfs_page *req;
1650 
1651         while (!list_empty(page_list)) {
1652                 req = nfs_list_entry(page_list->next);
1653                 nfs_list_remove_request(req);
1654                 nfs_mark_request_commit(req, lseg, cinfo, ds_commit_idx);
1655                 if (!cinfo->dreq)
1656                         nfs_clear_page_commit(req->wb_page);
1657                 nfs_unlock_and_release_request(req);
1658         }
1659 }
1660 EXPORT_SYMBOL_GPL(nfs_retry_commit);
1661 
1662 /*
1663  * Commit dirty pages
1664  */
1665 static int
1666 nfs_commit_list(struct inode *inode, struct list_head *head, int how,
1667                 struct nfs_commit_info *cinfo)
1668 {
1669         struct nfs_commit_data  *data;
1670 
1671         data = nfs_commitdata_alloc();
1672 
1673         if (!data)
1674                 goto out_bad;
1675 
1676         /* Set up the argument struct */
1677         nfs_init_commit(data, head, NULL, cinfo);
1678         atomic_inc(&cinfo->mds->rpcs_out);
1679         return nfs_initiate_commit(NFS_CLIENT(inode), data, NFS_PROTO(inode),
1680                                    data->mds_ops, how, 0);
1681  out_bad:
1682         nfs_retry_commit(head, NULL, cinfo, 0);
1683         cinfo->completion_ops->error_cleanup(NFS_I(inode));
1684         return -ENOMEM;
1685 }
1686 
1687 /*
1688  * COMMIT call returned
1689  */
1690 static void nfs_commit_done(struct rpc_task *task, void *calldata)
1691 {
1692         struct nfs_commit_data  *data = calldata;
1693 
1694         dprintk("NFS: %5u nfs_commit_done (status %d)\n",
1695                                 task->tk_pid, task->tk_status);
1696 
1697         /* Call the NFS version-specific code */
1698         NFS_PROTO(data->inode)->commit_done(task, data);
1699 }
1700 
1701 static void nfs_commit_release_pages(struct nfs_commit_data *data)
1702 {
1703         struct nfs_page *req;
1704         int status = data->task.tk_status;
1705         struct nfs_commit_info cinfo;
1706         struct nfs_server *nfss;
1707 
1708         while (!list_empty(&data->pages)) {
1709                 req = nfs_list_entry(data->pages.next);
1710                 nfs_list_remove_request(req);
1711                 nfs_clear_page_commit(req->wb_page);
1712 
1713                 dprintk("NFS:       commit (%s/%llu %d@%lld)",
1714                         req->wb_context->dentry->d_sb->s_id,
1715                         (unsigned long long)NFS_FILEID(d_inode(req->wb_context->dentry)),
1716                         req->wb_bytes,
1717                         (long long)req_offset(req));
1718                 if (status < 0) {
1719                         nfs_context_set_write_error(req->wb_context, status);
1720                         nfs_inode_remove_request(req);
1721                         dprintk(", error = %d\n", status);
1722                         goto next;
1723                 }
1724 
1725                 /* Okay, COMMIT succeeded, apparently. Check the verifier
1726                  * returned by the server against all stored verfs. */
1727                 if (!memcmp(&req->wb_verf, &data->verf.verifier, sizeof(req->wb_verf))) {
1728                         /* We have a match */
1729                         nfs_inode_remove_request(req);
1730                         dprintk(" OK\n");
1731                         goto next;
1732                 }
1733                 /* We have a mismatch. Write the page again */
1734                 dprintk(" mismatch\n");
1735                 nfs_mark_request_dirty(req);
1736                 set_bit(NFS_CONTEXT_RESEND_WRITES, &req->wb_context->flags);
1737         next:
1738                 nfs_unlock_and_release_request(req);
1739         }
1740         nfss = NFS_SERVER(data->inode);
1741         if (atomic_long_read(&nfss->writeback) < NFS_CONGESTION_OFF_THRESH)
1742                 clear_bdi_congested(&nfss->backing_dev_info, BLK_RW_ASYNC);
1743 
1744         nfs_init_cinfo(&cinfo, data->inode, data->dreq);
1745         if (atomic_dec_and_test(&cinfo.mds->rpcs_out))
1746                 nfs_commit_clear_lock(NFS_I(data->inode));
1747 }
1748 
1749 static void nfs_commit_release(void *calldata)
1750 {
1751         struct nfs_commit_data *data = calldata;
1752 
1753         data->completion_ops->completion(data);
1754         nfs_commitdata_release(calldata);
1755 }
1756 
1757 static const struct rpc_call_ops nfs_commit_ops = {
1758         .rpc_call_prepare = nfs_commit_prepare,
1759         .rpc_call_done = nfs_commit_done,
1760         .rpc_release = nfs_commit_release,
1761 };
1762 
1763 static const struct nfs_commit_completion_ops nfs_commit_completion_ops = {
1764         .completion = nfs_commit_release_pages,
1765         .error_cleanup = nfs_commit_clear_lock,
1766 };
1767 
1768 int nfs_generic_commit_list(struct inode *inode, struct list_head *head,
1769                             int how, struct nfs_commit_info *cinfo)
1770 {
1771         int status;
1772 
1773         status = pnfs_commit_list(inode, head, how, cinfo);
1774         if (status == PNFS_NOT_ATTEMPTED)
1775                 status = nfs_commit_list(inode, head, how, cinfo);
1776         return status;
1777 }
1778 
1779 int nfs_commit_inode(struct inode *inode, int how)
1780 {
1781         LIST_HEAD(head);
1782         struct nfs_commit_info cinfo;
1783         int may_wait = how & FLUSH_SYNC;
1784         int res;
1785 
1786         res = nfs_commit_set_lock(NFS_I(inode), may_wait);
1787         if (res <= 0)
1788                 goto out_mark_dirty;
1789         nfs_init_cinfo_from_inode(&cinfo, inode);
1790         res = nfs_scan_commit(inode, &head, &cinfo);
1791         if (res) {
1792                 int error;
1793 
1794                 error = nfs_generic_commit_list(inode, &head, how, &cinfo);
1795                 if (error < 0)
1796                         return error;
1797                 if (!may_wait)
1798                         goto out_mark_dirty;
1799                 error = wait_on_bit_action(&NFS_I(inode)->flags,
1800                                 NFS_INO_COMMIT,
1801                                 nfs_wait_bit_killable,
1802                                 TASK_KILLABLE);
1803                 if (error < 0)
1804                         return error;
1805         } else
1806                 nfs_commit_clear_lock(NFS_I(inode));
1807         return res;
1808         /* Note: If we exit without ensuring that the commit is complete,
1809          * we must mark the inode as dirty. Otherwise, future calls to
1810          * sync_inode() with the WB_SYNC_ALL flag set will fail to ensure
1811          * that the data is on the disk.
1812          */
1813 out_mark_dirty:
1814         __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1815         return res;
1816 }
1817 
1818 int nfs_write_inode(struct inode *inode, struct writeback_control *wbc)
1819 {
1820         struct nfs_inode *nfsi = NFS_I(inode);
1821         int flags = FLUSH_SYNC;
1822         int ret = 0;
1823 
1824         /* no commits means nothing needs to be done */
1825         if (!nfsi->commit_info.ncommit)
1826                 return ret;
1827 
1828         if (wbc->sync_mode == WB_SYNC_NONE) {
1829                 /* Don't commit yet if this is a non-blocking flush and there
1830                  * are a lot of outstanding writes for this mapping.
1831                  */
1832                 if (nfsi->commit_info.ncommit <= (nfsi->nrequests >> 1))
1833                         goto out_mark_dirty;
1834 
1835                 /* don't wait for the COMMIT response */
1836                 flags = 0;
1837         }
1838 
1839         ret = nfs_commit_inode(inode, flags);
1840         if (ret >= 0) {
1841                 if (wbc->sync_mode == WB_SYNC_NONE) {
1842                         if (ret < wbc->nr_to_write)
1843                                 wbc->nr_to_write -= ret;
1844                         else
1845                                 wbc->nr_to_write = 0;
1846                 }
1847                 return 0;
1848         }
1849 out_mark_dirty:
1850         __mark_inode_dirty(inode, I_DIRTY_DATASYNC);
1851         return ret;
1852 }
1853 EXPORT_SYMBOL_GPL(nfs_write_inode);
1854 
1855 /*
1856  * flush the inode to disk.
1857  */
1858 int nfs_wb_all(struct inode *inode)
1859 {
1860         int ret;
1861 
1862         trace_nfs_writeback_inode_enter(inode);
1863 
1864         ret = filemap_write_and_wait(inode->i_mapping);
1865         if (ret)
1866                 goto out;
1867         ret = nfs_commit_inode(inode, FLUSH_SYNC);
1868         if (ret < 0)
1869                 goto out;
1870         pnfs_sync_inode(inode, true);
1871         ret = 0;
1872 
1873 out:
1874         trace_nfs_writeback_inode_exit(inode, ret);
1875         return ret;
1876 }
1877 EXPORT_SYMBOL_GPL(nfs_wb_all);
1878 
1879 int nfs_wb_page_cancel(struct inode *inode, struct page *page)
1880 {
1881         struct nfs_page *req;
1882         int ret = 0;
1883 
1884         wait_on_page_writeback(page);
1885 
1886         /* blocking call to cancel all requests and join to a single (head)
1887          * request */
1888         req = nfs_lock_and_join_requests(page, false);
1889 
1890         if (IS_ERR(req)) {
1891                 ret = PTR_ERR(req);
1892         } else if (req) {
1893                 /* all requests from this page have been cancelled by
1894                  * nfs_lock_and_join_requests, so just remove the head
1895                  * request from the inode / page_private pointer and
1896                  * release it */
1897                 nfs_inode_remove_request(req);
1898                 nfs_unlock_and_release_request(req);
1899         }
1900 
1901         return ret;
1902 }
1903 
1904 /*
1905  * Write back all requests on one page - we do this before reading it.
1906  */
1907 int nfs_wb_page(struct inode *inode, struct page *page)
1908 {
1909         loff_t range_start = page_file_offset(page);
1910         loff_t range_end = range_start + (loff_t)(PAGE_CACHE_SIZE - 1);
1911         struct writeback_control wbc = {
1912                 .sync_mode = WB_SYNC_ALL,
1913                 .nr_to_write = 0,
1914                 .range_start = range_start,
1915                 .range_end = range_end,
1916         };
1917         int ret;
1918 
1919         trace_nfs_writeback_page_enter(inode);
1920 
1921         for (;;) {
1922                 wait_on_page_writeback(page);
1923                 if (clear_page_dirty_for_io(page)) {
1924                         ret = nfs_writepage_locked(page, &wbc);
1925                         if (ret < 0)
1926                                 goto out_error;
1927                         continue;
1928                 }
1929                 ret = 0;
1930                 if (!PagePrivate(page))
1931                         break;
1932                 ret = nfs_commit_inode(inode, FLUSH_SYNC);
1933                 if (ret < 0)
1934                         goto out_error;
1935         }
1936 out_error:
1937         trace_nfs_writeback_page_exit(inode, ret);
1938         return ret;
1939 }
1940 
1941 #ifdef CONFIG_MIGRATION
1942 int nfs_migrate_page(struct address_space *mapping, struct page *newpage,
1943                 struct page *page, enum migrate_mode mode)
1944 {
1945         /*
1946          * If PagePrivate is set, then the page is currently associated with
1947          * an in-progress read or write request. Don't try to migrate it.
1948          *
1949          * FIXME: we could do this in principle, but we'll need a way to ensure
1950          *        that we can safely release the inode reference while holding
1951          *        the page lock.
1952          */
1953         if (PagePrivate(page))
1954                 return -EBUSY;
1955 
1956         if (!nfs_fscache_release_page(page, GFP_KERNEL))
1957                 return -EBUSY;
1958 
1959         return migrate_page(mapping, newpage, page, mode);
1960 }
1961 #endif
1962 
1963 int __init nfs_init_writepagecache(void)
1964 {
1965         nfs_wdata_cachep = kmem_cache_create("nfs_write_data",
1966                                              sizeof(struct nfs_pgio_header),
1967                                              0, SLAB_HWCACHE_ALIGN,
1968                                              NULL);
1969         if (nfs_wdata_cachep == NULL)
1970                 return -ENOMEM;
1971 
1972         nfs_wdata_mempool = mempool_create_slab_pool(MIN_POOL_WRITE,
1973                                                      nfs_wdata_cachep);
1974         if (nfs_wdata_mempool == NULL)
1975                 goto out_destroy_write_cache;
1976 
1977         nfs_cdata_cachep = kmem_cache_create("nfs_commit_data",
1978                                              sizeof(struct nfs_commit_data),
1979                                              0, SLAB_HWCACHE_ALIGN,
1980                                              NULL);
1981         if (nfs_cdata_cachep == NULL)
1982                 goto out_destroy_write_mempool;
1983 
1984         nfs_commit_mempool = mempool_create_slab_pool(MIN_POOL_COMMIT,
1985                                                       nfs_cdata_cachep);
1986         if (nfs_commit_mempool == NULL)
1987                 goto out_destroy_commit_cache;
1988 
1989         /*
1990          * NFS congestion size, scale with available memory.
1991          *
1992          *  64MB:    8192k
1993          * 128MB:   11585k
1994          * 256MB:   16384k
1995          * 512MB:   23170k
1996          *   1GB:   32768k
1997          *   2GB:   46340k
1998          *   4GB:   65536k
1999          *   8GB:   92681k
2000          *  16GB:  131072k
2001          *
2002          * This allows larger machines to have larger/more transfers.
2003          * Limit the default to 256M
2004          */
2005         nfs_congestion_kb = (16*int_sqrt(totalram_pages)) << (PAGE_SHIFT-10);
2006         if (nfs_congestion_kb > 256*1024)
2007                 nfs_congestion_kb = 256*1024;
2008 
2009         return 0;
2010 
2011 out_destroy_commit_cache:
2012         kmem_cache_destroy(nfs_cdata_cachep);
2013 out_destroy_write_mempool:
2014         mempool_destroy(nfs_wdata_mempool);
2015 out_destroy_write_cache:
2016         kmem_cache_destroy(nfs_wdata_cachep);
2017         return -ENOMEM;
2018 }
2019 
2020 void nfs_destroy_writepagecache(void)
2021 {
2022         mempool_destroy(nfs_commit_mempool);
2023         kmem_cache_destroy(nfs_cdata_cachep);
2024         mempool_destroy(nfs_wdata_mempool);
2025         kmem_cache_destroy(nfs_wdata_cachep);
2026 }
2027 
2028 static const struct nfs_rw_ops nfs_rw_write_ops = {
2029         .rw_mode                = FMODE_WRITE,
2030         .rw_alloc_header        = nfs_writehdr_alloc,
2031         .rw_free_header         = nfs_writehdr_free,
2032         .rw_done                = nfs_writeback_done,
2033         .rw_result              = nfs_writeback_result,
2034         .rw_initiate            = nfs_initiate_write,
2035 };
2036 

~ [ 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