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