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

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

Version: ~ [ linux-5.13-rc1 ] ~ [ linux-5.12.2 ] ~ [ linux-5.11.19 ] ~ [ linux-5.10.35 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.117 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.190 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.232 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.268 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.268 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  1 /*
  2  *  linux/fs/nfs/file.c
  3  *
  4  *  Copyright (C) 1992  Rick Sladkey
  5  *
  6  *  Changes Copyright (C) 1994 by Florian La Roche
  7  *   - Do not copy data too often around in the kernel.
  8  *   - In nfs_file_read the return value of kmalloc wasn't checked.
  9  *   - Put in a better version of read look-ahead buffering. Original idea
 10  *     and implementation by Wai S Kok elekokws@ee.nus.sg.
 11  *
 12  *  Expire cache on write to a file by Wai S Kok (Oct 1994).
 13  *
 14  *  Total rewrite of read side for new NFS buffer cache.. Linus.
 15  *
 16  *  nfs regular file handling functions
 17  */
 18 
 19 #include <linux/module.h>
 20 #include <linux/time.h>
 21 #include <linux/kernel.h>
 22 #include <linux/errno.h>
 23 #include <linux/fcntl.h>
 24 #include <linux/stat.h>
 25 #include <linux/nfs_fs.h>
 26 #include <linux/nfs_mount.h>
 27 #include <linux/mm.h>
 28 #include <linux/pagemap.h>
 29 #include <linux/gfp.h>
 30 #include <linux/swap.h>
 31 
 32 #include <asm/uaccess.h>
 33 
 34 #include "delegation.h"
 35 #include "internal.h"
 36 #include "iostat.h"
 37 #include "fscache.h"
 38 #include "pnfs.h"
 39 
 40 #include "nfstrace.h"
 41 
 42 #define NFSDBG_FACILITY         NFSDBG_FILE
 43 
 44 static const struct vm_operations_struct nfs_file_vm_ops;
 45 
 46 /* Hack for future NFS swap support */
 47 #ifndef IS_SWAPFILE
 48 # define IS_SWAPFILE(inode)     (0)
 49 #endif
 50 
 51 int nfs_check_flags(int flags)
 52 {
 53         if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT))
 54                 return -EINVAL;
 55 
 56         return 0;
 57 }
 58 EXPORT_SYMBOL_GPL(nfs_check_flags);
 59 
 60 /*
 61  * Open file
 62  */
 63 static int
 64 nfs_file_open(struct inode *inode, struct file *filp)
 65 {
 66         int res;
 67 
 68         dprintk("NFS: open file(%pD2)\n", filp);
 69 
 70         nfs_inc_stats(inode, NFSIOS_VFSOPEN);
 71         res = nfs_check_flags(filp->f_flags);
 72         if (res)
 73                 return res;
 74 
 75         res = nfs_open(inode, filp);
 76         return res;
 77 }
 78 
 79 int
 80 nfs_file_release(struct inode *inode, struct file *filp)
 81 {
 82         dprintk("NFS: release(%pD2)\n", filp);
 83 
 84         nfs_inc_stats(inode, NFSIOS_VFSRELEASE);
 85         nfs_file_clear_open_context(filp);
 86         return 0;
 87 }
 88 EXPORT_SYMBOL_GPL(nfs_file_release);
 89 
 90 /**
 91  * nfs_revalidate_size - Revalidate the file size
 92  * @inode - pointer to inode struct
 93  * @file - pointer to struct file
 94  *
 95  * Revalidates the file length. This is basically a wrapper around
 96  * nfs_revalidate_inode() that takes into account the fact that we may
 97  * have cached writes (in which case we don't care about the server's
 98  * idea of what the file length is), or O_DIRECT (in which case we
 99  * shouldn't trust the cache).
100  */
101 static int nfs_revalidate_file_size(struct inode *inode, struct file *filp)
102 {
103         struct nfs_server *server = NFS_SERVER(inode);
104         struct nfs_inode *nfsi = NFS_I(inode);
105 
106         if (nfs_have_delegated_attributes(inode))
107                 goto out_noreval;
108 
109         if (filp->f_flags & O_DIRECT)
110                 goto force_reval;
111         if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE)
112                 goto force_reval;
113         if (nfs_attribute_timeout(inode))
114                 goto force_reval;
115 out_noreval:
116         return 0;
117 force_reval:
118         return __nfs_revalidate_inode(server, inode);
119 }
120 
121 loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence)
122 {
123         dprintk("NFS: llseek file(%pD2, %lld, %d)\n",
124                         filp, offset, whence);
125 
126         /*
127          * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate
128          * the cached file length
129          */
130         if (whence != SEEK_SET && whence != SEEK_CUR) {
131                 struct inode *inode = filp->f_mapping->host;
132 
133                 int retval = nfs_revalidate_file_size(inode, filp);
134                 if (retval < 0)
135                         return (loff_t)retval;
136         }
137 
138         return generic_file_llseek(filp, offset, whence);
139 }
140 EXPORT_SYMBOL_GPL(nfs_file_llseek);
141 
142 /*
143  * Flush all dirty pages, and check for write errors.
144  */
145 static int
146 nfs_file_flush(struct file *file, fl_owner_t id)
147 {
148         struct inode    *inode = file_inode(file);
149 
150         dprintk("NFS: flush(%pD2)\n", file);
151 
152         nfs_inc_stats(inode, NFSIOS_VFSFLUSH);
153         if ((file->f_mode & FMODE_WRITE) == 0)
154                 return 0;
155 
156         /* Flush writes to the server and return any errors */
157         return vfs_fsync(file, 0);
158 }
159 
160 ssize_t
161 nfs_file_read(struct kiocb *iocb, struct iov_iter *to)
162 {
163         struct inode *inode = file_inode(iocb->ki_filp);
164         ssize_t result;
165 
166         if (iocb->ki_flags & IOCB_DIRECT)
167                 return nfs_file_direct_read(iocb, to);
168 
169         dprintk("NFS: read(%pD2, %zu@%lu)\n",
170                 iocb->ki_filp,
171                 iov_iter_count(to), (unsigned long) iocb->ki_pos);
172 
173         nfs_start_io_read(inode);
174         result = nfs_revalidate_mapping(inode, iocb->ki_filp->f_mapping);
175         if (!result) {
176                 result = generic_file_read_iter(iocb, to);
177                 if (result > 0)
178                         nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result);
179         }
180         nfs_end_io_read(inode);
181         return result;
182 }
183 EXPORT_SYMBOL_GPL(nfs_file_read);
184 
185 ssize_t
186 nfs_file_splice_read(struct file *filp, loff_t *ppos,
187                      struct pipe_inode_info *pipe, size_t count,
188                      unsigned int flags)
189 {
190         struct inode *inode = file_inode(filp);
191         ssize_t res;
192 
193         dprintk("NFS: splice_read(%pD2, %lu@%Lu)\n",
194                 filp, (unsigned long) count, (unsigned long long) *ppos);
195 
196         nfs_start_io_read(inode);
197         res = nfs_revalidate_mapping(inode, filp->f_mapping);
198         if (!res) {
199                 res = generic_file_splice_read(filp, ppos, pipe, count, flags);
200                 if (res > 0)
201                         nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res);
202         }
203         nfs_end_io_read(inode);
204         return res;
205 }
206 EXPORT_SYMBOL_GPL(nfs_file_splice_read);
207 
208 int
209 nfs_file_mmap(struct file * file, struct vm_area_struct * vma)
210 {
211         struct inode *inode = file_inode(file);
212         int     status;
213 
214         dprintk("NFS: mmap(%pD2)\n", file);
215 
216         /* Note: generic_file_mmap() returns ENOSYS on nommu systems
217          *       so we call that before revalidating the mapping
218          */
219         status = generic_file_mmap(file, vma);
220         if (!status) {
221                 vma->vm_ops = &nfs_file_vm_ops;
222                 status = nfs_revalidate_mapping(inode, file->f_mapping);
223         }
224         return status;
225 }
226 EXPORT_SYMBOL_GPL(nfs_file_mmap);
227 
228 /*
229  * Flush any dirty pages for this process, and check for write errors.
230  * The return status from this call provides a reliable indication of
231  * whether any write errors occurred for this process.
232  *
233  * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to
234  * disk, but it retrieves and clears ctx->error after synching, despite
235  * the two being set at the same time in nfs_context_set_write_error().
236  * This is because the former is used to notify the _next_ call to
237  * nfs_file_write() that a write error occurred, and hence cause it to
238  * fall back to doing a synchronous write.
239  */
240 static int
241 nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync)
242 {
243         struct nfs_open_context *ctx = nfs_file_open_context(file);
244         struct inode *inode = file_inode(file);
245         int have_error, do_resend, status;
246         int ret = 0;
247 
248         dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync);
249 
250         nfs_inc_stats(inode, NFSIOS_VFSFSYNC);
251         do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
252         have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
253         status = nfs_commit_inode(inode, FLUSH_SYNC);
254         have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags);
255         if (have_error) {
256                 ret = xchg(&ctx->error, 0);
257                 if (ret)
258                         goto out;
259         }
260         if (status < 0) {
261                 ret = status;
262                 goto out;
263         }
264         do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags);
265         if (do_resend)
266                 ret = -EAGAIN;
267 out:
268         return ret;
269 }
270 
271 int
272 nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync)
273 {
274         int ret;
275         struct inode *inode = file_inode(file);
276 
277         trace_nfs_fsync_enter(inode);
278 
279         do {
280                 ret = filemap_write_and_wait_range(inode->i_mapping, start, end);
281                 if (ret != 0)
282                         break;
283                 ret = nfs_file_fsync_commit(file, start, end, datasync);
284                 if (!ret)
285                         ret = pnfs_sync_inode(inode, !!datasync);
286                 /*
287                  * If nfs_file_fsync_commit detected a server reboot, then
288                  * resend all dirty pages that might have been covered by
289                  * the NFS_CONTEXT_RESEND_WRITES flag
290                  */
291                 start = 0;
292                 end = LLONG_MAX;
293         } while (ret == -EAGAIN);
294 
295         trace_nfs_fsync_exit(inode, ret);
296         return ret;
297 }
298 EXPORT_SYMBOL_GPL(nfs_file_fsync);
299 
300 /*
301  * Decide whether a read/modify/write cycle may be more efficient
302  * then a modify/write/read cycle when writing to a page in the
303  * page cache.
304  *
305  * The modify/write/read cycle may occur if a page is read before
306  * being completely filled by the writer.  In this situation, the
307  * page must be completely written to stable storage on the server
308  * before it can be refilled by reading in the page from the server.
309  * This can lead to expensive, small, FILE_SYNC mode writes being
310  * done.
311  *
312  * It may be more efficient to read the page first if the file is
313  * open for reading in addition to writing, the page is not marked
314  * as Uptodate, it is not dirty or waiting to be committed,
315  * indicating that it was previously allocated and then modified,
316  * that there were valid bytes of data in that range of the file,
317  * and that the new data won't completely replace the old data in
318  * that range of the file.
319  */
320 static int nfs_want_read_modify_write(struct file *file, struct page *page,
321                         loff_t pos, unsigned len)
322 {
323         unsigned int pglen = nfs_page_length(page);
324         unsigned int offset = pos & (PAGE_SIZE - 1);
325         unsigned int end = offset + len;
326 
327         if (pnfs_ld_read_whole_page(file->f_mapping->host)) {
328                 if (!PageUptodate(page))
329                         return 1;
330                 return 0;
331         }
332 
333         if ((file->f_mode & FMODE_READ) &&      /* open for read? */
334             !PageUptodate(page) &&              /* Uptodate? */
335             !PagePrivate(page) &&               /* i/o request already? */
336             pglen &&                            /* valid bytes of file? */
337             (end < pglen || offset))            /* replace all valid bytes? */
338                 return 1;
339         return 0;
340 }
341 
342 /*
343  * This does the "real" work of the write. We must allocate and lock the
344  * page to be sent back to the generic routine, which then copies the
345  * data from user space.
346  *
347  * If the writer ends up delaying the write, the writer needs to
348  * increment the page use counts until he is done with the page.
349  */
350 static int nfs_write_begin(struct file *file, struct address_space *mapping,
351                         loff_t pos, unsigned len, unsigned flags,
352                         struct page **pagep, void **fsdata)
353 {
354         int ret;
355         pgoff_t index = pos >> PAGE_SHIFT;
356         struct page *page;
357         int once_thru = 0;
358 
359         dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n",
360                 file, mapping->host->i_ino, len, (long long) pos);
361 
362 start:
363         page = grab_cache_page_write_begin(mapping, index, flags);
364         if (!page)
365                 return -ENOMEM;
366         *pagep = page;
367 
368         ret = nfs_flush_incompatible(file, page);
369         if (ret) {
370                 unlock_page(page);
371                 put_page(page);
372         } else if (!once_thru &&
373                    nfs_want_read_modify_write(file, page, pos, len)) {
374                 once_thru = 1;
375                 ret = nfs_readpage(file, page);
376                 put_page(page);
377                 if (!ret)
378                         goto start;
379         }
380         return ret;
381 }
382 
383 static int nfs_write_end(struct file *file, struct address_space *mapping,
384                         loff_t pos, unsigned len, unsigned copied,
385                         struct page *page, void *fsdata)
386 {
387         unsigned offset = pos & (PAGE_SIZE - 1);
388         struct nfs_open_context *ctx = nfs_file_open_context(file);
389         int status;
390 
391         dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n",
392                 file, mapping->host->i_ino, len, (long long) pos);
393 
394         /*
395          * Zero any uninitialised parts of the page, and then mark the page
396          * as up to date if it turns out that we're extending the file.
397          */
398         if (!PageUptodate(page)) {
399                 unsigned pglen = nfs_page_length(page);
400                 unsigned end = offset + copied;
401 
402                 if (pglen == 0) {
403                         zero_user_segments(page, 0, offset,
404                                         end, PAGE_SIZE);
405                         SetPageUptodate(page);
406                 } else if (end >= pglen) {
407                         zero_user_segment(page, end, PAGE_SIZE);
408                         if (offset == 0)
409                                 SetPageUptodate(page);
410                 } else
411                         zero_user_segment(page, pglen, PAGE_SIZE);
412         }
413 
414         status = nfs_updatepage(file, page, offset, copied);
415 
416         unlock_page(page);
417         put_page(page);
418 
419         if (status < 0)
420                 return status;
421         NFS_I(mapping->host)->write_io += copied;
422 
423         if (nfs_ctx_key_to_expire(ctx, mapping->host)) {
424                 status = nfs_wb_all(mapping->host);
425                 if (status < 0)
426                         return status;
427         }
428 
429         return copied;
430 }
431 
432 /*
433  * Partially or wholly invalidate a page
434  * - Release the private state associated with a page if undergoing complete
435  *   page invalidation
436  * - Called if either PG_private or PG_fscache is set on the page
437  * - Caller holds page lock
438  */
439 static void nfs_invalidate_page(struct page *page, unsigned int offset,
440                                 unsigned int length)
441 {
442         dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n",
443                  page, offset, length);
444 
445         if (offset != 0 || length < PAGE_SIZE)
446                 return;
447         /* Cancel any unstarted writes on this page */
448         nfs_wb_page_cancel(page_file_mapping(page)->host, page);
449 
450         nfs_fscache_invalidate_page(page, page->mapping->host);
451 }
452 
453 /*
454  * Attempt to release the private state associated with a page
455  * - Called if either PG_private or PG_fscache is set on the page
456  * - Caller holds page lock
457  * - Return true (may release page) or false (may not)
458  */
459 static int nfs_release_page(struct page *page, gfp_t gfp)
460 {
461         dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
462 
463         /* If PagePrivate() is set, then the page is not freeable */
464         if (PagePrivate(page))
465                 return 0;
466         return nfs_fscache_release_page(page, gfp);
467 }
468 
469 static void nfs_check_dirty_writeback(struct page *page,
470                                 bool *dirty, bool *writeback)
471 {
472         struct nfs_inode *nfsi;
473         struct address_space *mapping = page_file_mapping(page);
474 
475         if (!mapping || PageSwapCache(page))
476                 return;
477 
478         /*
479          * Check if an unstable page is currently being committed and
480          * if so, have the VM treat it as if the page is under writeback
481          * so it will not block due to pages that will shortly be freeable.
482          */
483         nfsi = NFS_I(mapping->host);
484         if (atomic_read(&nfsi->commit_info.rpcs_out)) {
485                 *writeback = true;
486                 return;
487         }
488 
489         /*
490          * If PagePrivate() is set, then the page is not freeable and as the
491          * inode is not being committed, it's not going to be cleaned in the
492          * near future so treat it as dirty
493          */
494         if (PagePrivate(page))
495                 *dirty = true;
496 }
497 
498 /*
499  * Attempt to clear the private state associated with a page when an error
500  * occurs that requires the cached contents of an inode to be written back or
501  * destroyed
502  * - Called if either PG_private or fscache is set on the page
503  * - Caller holds page lock
504  * - Return 0 if successful, -error otherwise
505  */
506 static int nfs_launder_page(struct page *page)
507 {
508         struct inode *inode = page_file_mapping(page)->host;
509         struct nfs_inode *nfsi = NFS_I(inode);
510 
511         dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
512                 inode->i_ino, (long long)page_offset(page));
513 
514         nfs_fscache_wait_on_page_write(nfsi, page);
515         return nfs_wb_launder_page(inode, page);
516 }
517 
518 static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file,
519                                                 sector_t *span)
520 {
521         struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
522 
523         *span = sis->pages;
524 
525         return rpc_clnt_swap_activate(clnt);
526 }
527 
528 static void nfs_swap_deactivate(struct file *file)
529 {
530         struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host);
531 
532         rpc_clnt_swap_deactivate(clnt);
533 }
534 
535 const struct address_space_operations nfs_file_aops = {
536         .readpage = nfs_readpage,
537         .readpages = nfs_readpages,
538         .set_page_dirty = __set_page_dirty_nobuffers,
539         .writepage = nfs_writepage,
540         .writepages = nfs_writepages,
541         .write_begin = nfs_write_begin,
542         .write_end = nfs_write_end,
543         .invalidatepage = nfs_invalidate_page,
544         .releasepage = nfs_release_page,
545         .direct_IO = nfs_direct_IO,
546         .migratepage = nfs_migrate_page,
547         .launder_page = nfs_launder_page,
548         .is_dirty_writeback = nfs_check_dirty_writeback,
549         .error_remove_page = generic_error_remove_page,
550         .swap_activate = nfs_swap_activate,
551         .swap_deactivate = nfs_swap_deactivate,
552 };
553 
554 /*
555  * Notification that a PTE pointing to an NFS page is about to be made
556  * writable, implying that someone is about to modify the page through a
557  * shared-writable mapping
558  */
559 static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
560 {
561         struct page *page = vmf->page;
562         struct file *filp = vma->vm_file;
563         struct inode *inode = file_inode(filp);
564         unsigned pagelen;
565         int ret = VM_FAULT_NOPAGE;
566         struct address_space *mapping;
567 
568         dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n",
569                 filp, filp->f_mapping->host->i_ino,
570                 (long long)page_offset(page));
571 
572         sb_start_pagefault(inode->i_sb);
573 
574         /* make sure the cache has finished storing the page */
575         nfs_fscache_wait_on_page_write(NFS_I(inode), page);
576 
577         wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING,
578                         nfs_wait_bit_killable, TASK_KILLABLE);
579 
580         lock_page(page);
581         mapping = page_file_mapping(page);
582         if (mapping != inode->i_mapping)
583                 goto out_unlock;
584 
585         wait_on_page_writeback(page);
586 
587         pagelen = nfs_page_length(page);
588         if (pagelen == 0)
589                 goto out_unlock;
590 
591         ret = VM_FAULT_LOCKED;
592         if (nfs_flush_incompatible(filp, page) == 0 &&
593             nfs_updatepage(filp, page, 0, pagelen) == 0)
594                 goto out;
595 
596         ret = VM_FAULT_SIGBUS;
597 out_unlock:
598         unlock_page(page);
599 out:
600         sb_end_pagefault(inode->i_sb);
601         return ret;
602 }
603 
604 static const struct vm_operations_struct nfs_file_vm_ops = {
605         .fault = filemap_fault,
606         .map_pages = filemap_map_pages,
607         .page_mkwrite = nfs_vm_page_mkwrite,
608 };
609 
610 static int nfs_need_check_write(struct file *filp, struct inode *inode)
611 {
612         struct nfs_open_context *ctx;
613 
614         ctx = nfs_file_open_context(filp);
615         if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) ||
616             nfs_ctx_key_to_expire(ctx, inode))
617                 return 1;
618         return 0;
619 }
620 
621 ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from)
622 {
623         struct file *file = iocb->ki_filp;
624         struct inode *inode = file_inode(file);
625         unsigned long written = 0;
626         ssize_t result;
627 
628         result = nfs_key_timeout_notify(file, inode);
629         if (result)
630                 return result;
631 
632         if (iocb->ki_flags & IOCB_DIRECT)
633                 return nfs_file_direct_write(iocb, from);
634 
635         dprintk("NFS: write(%pD2, %zu@%Ld)\n",
636                 file, iov_iter_count(from), (long long) iocb->ki_pos);
637 
638         if (IS_SWAPFILE(inode))
639                 goto out_swapfile;
640         /*
641          * O_APPEND implies that we must revalidate the file length.
642          */
643         if (iocb->ki_flags & IOCB_APPEND) {
644                 result = nfs_revalidate_file_size(inode, file);
645                 if (result)
646                         goto out;
647         }
648 
649         nfs_start_io_write(inode);
650         result = generic_write_checks(iocb, from);
651         if (result > 0) {
652                 current->backing_dev_info = inode_to_bdi(inode);
653                 result = generic_perform_write(file, from, iocb->ki_pos);
654                 current->backing_dev_info = NULL;
655         }
656         nfs_end_io_write(inode);
657         if (result <= 0)
658                 goto out;
659 
660         result = generic_write_sync(iocb, result);
661         if (result < 0)
662                 goto out;
663         written = result;
664         iocb->ki_pos += written;
665 
666         /* Return error values */
667         if (nfs_need_check_write(file, inode)) {
668                 int err = vfs_fsync(file, 0);
669                 if (err < 0)
670                         result = err;
671         }
672         nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written);
673 out:
674         return result;
675 
676 out_swapfile:
677         printk(KERN_INFO "NFS: attempt to write to active swap file!\n");
678         return -EBUSY;
679 }
680 EXPORT_SYMBOL_GPL(nfs_file_write);
681 
682 static int
683 do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
684 {
685         struct inode *inode = filp->f_mapping->host;
686         int status = 0;
687         unsigned int saved_type = fl->fl_type;
688 
689         /* Try local locking first */
690         posix_test_lock(filp, fl);
691         if (fl->fl_type != F_UNLCK) {
692                 /* found a conflict */
693                 goto out;
694         }
695         fl->fl_type = saved_type;
696 
697         if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
698                 goto out_noconflict;
699 
700         if (is_local)
701                 goto out_noconflict;
702 
703         status = NFS_PROTO(inode)->lock(filp, cmd, fl);
704 out:
705         return status;
706 out_noconflict:
707         fl->fl_type = F_UNLCK;
708         goto out;
709 }
710 
711 static int do_vfs_lock(struct file *file, struct file_lock *fl)
712 {
713         return locks_lock_file_wait(file, fl);
714 }
715 
716 static int
717 do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
718 {
719         struct inode *inode = filp->f_mapping->host;
720         struct nfs_lock_context *l_ctx;
721         int status;
722 
723         /*
724          * Flush all pending writes before doing anything
725          * with locks..
726          */
727         vfs_fsync(filp, 0);
728 
729         l_ctx = nfs_get_lock_context(nfs_file_open_context(filp));
730         if (!IS_ERR(l_ctx)) {
731                 status = nfs_iocounter_wait(l_ctx);
732                 nfs_put_lock_context(l_ctx);
733                 if (status < 0)
734                         return status;
735         }
736 
737         /* NOTE: special case
738          *      If we're signalled while cleaning up locks on process exit, we
739          *      still need to complete the unlock.
740          */
741         /*
742          * Use local locking if mounted with "-onolock" or with appropriate
743          * "-olocal_lock="
744          */
745         if (!is_local)
746                 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
747         else
748                 status = do_vfs_lock(filp, fl);
749         return status;
750 }
751 
752 static int
753 do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local)
754 {
755         struct inode *inode = filp->f_mapping->host;
756         int status;
757 
758         /*
759          * Flush all pending writes before doing anything
760          * with locks..
761          */
762         status = nfs_sync_mapping(filp->f_mapping);
763         if (status != 0)
764                 goto out;
765 
766         /*
767          * Use local locking if mounted with "-onolock" or with appropriate
768          * "-olocal_lock="
769          */
770         if (!is_local)
771                 status = NFS_PROTO(inode)->lock(filp, cmd, fl);
772         else
773                 status = do_vfs_lock(filp, fl);
774         if (status < 0)
775                 goto out;
776 
777         /*
778          * Revalidate the cache if the server has time stamps granular
779          * enough to detect subsecond changes.  Otherwise, clear the
780          * cache to prevent missing any changes.
781          *
782          * This makes locking act as a cache coherency point.
783          */
784         nfs_sync_mapping(filp->f_mapping);
785         if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ))
786                 nfs_zap_mapping(inode, filp->f_mapping);
787 out:
788         return status;
789 }
790 
791 /*
792  * Lock a (portion of) a file
793  */
794 int nfs_lock(struct file *filp, int cmd, struct file_lock *fl)
795 {
796         struct inode *inode = filp->f_mapping->host;
797         int ret = -ENOLCK;
798         int is_local = 0;
799 
800         dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n",
801                         filp, fl->fl_type, fl->fl_flags,
802                         (long long)fl->fl_start, (long long)fl->fl_end);
803 
804         nfs_inc_stats(inode, NFSIOS_VFSLOCK);
805 
806         /* No mandatory locks over NFS */
807         if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK)
808                 goto out_err;
809 
810         if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL)
811                 is_local = 1;
812 
813         if (NFS_PROTO(inode)->lock_check_bounds != NULL) {
814                 ret = NFS_PROTO(inode)->lock_check_bounds(fl);
815                 if (ret < 0)
816                         goto out_err;
817         }
818 
819         if (IS_GETLK(cmd))
820                 ret = do_getlk(filp, cmd, fl, is_local);
821         else if (fl->fl_type == F_UNLCK)
822                 ret = do_unlk(filp, cmd, fl, is_local);
823         else
824                 ret = do_setlk(filp, cmd, fl, is_local);
825 out_err:
826         return ret;
827 }
828 EXPORT_SYMBOL_GPL(nfs_lock);
829 
830 /*
831  * Lock a (portion of) a file
832  */
833 int nfs_flock(struct file *filp, int cmd, struct file_lock *fl)
834 {
835         struct inode *inode = filp->f_mapping->host;
836         int is_local = 0;
837 
838         dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n",
839                         filp, fl->fl_type, fl->fl_flags);
840 
841         if (!(fl->fl_flags & FL_FLOCK))
842                 return -ENOLCK;
843 
844         /*
845          * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of
846          * any standard. In principle we might be able to support LOCK_MAND
847          * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the
848          * NFS code is not set up for it.
849          */
850         if (fl->fl_type & LOCK_MAND)
851                 return -EINVAL;
852 
853         if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK)
854                 is_local = 1;
855 
856         /* We're simulating flock() locks using posix locks on the server */
857         if (fl->fl_type == F_UNLCK)
858                 return do_unlk(filp, cmd, fl, is_local);
859         return do_setlk(filp, cmd, fl, is_local);
860 }
861 EXPORT_SYMBOL_GPL(nfs_flock);
862 
863 const struct file_operations nfs_file_operations = {
864         .llseek         = nfs_file_llseek,
865         .read_iter      = nfs_file_read,
866         .write_iter     = nfs_file_write,
867         .mmap           = nfs_file_mmap,
868         .open           = nfs_file_open,
869         .flush          = nfs_file_flush,
870         .release        = nfs_file_release,
871         .fsync          = nfs_file_fsync,
872         .lock           = nfs_lock,
873         .flock          = nfs_flock,
874         .splice_read    = nfs_file_splice_read,
875         .splice_write   = iter_file_splice_write,
876         .check_flags    = nfs_check_flags,
877         .setlease       = simple_nosetlease,
878 };
879 EXPORT_SYMBOL_GPL(nfs_file_operations);
880 

~ [ 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