Commit | Line | Data |
---|---|---|
1da177e4 LT |
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 | ||
ddda8e0a | 19 | #include <linux/module.h> |
1da177e4 LT |
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> | |
1da177e4 | 28 | #include <linux/pagemap.h> |
5a0e3ad6 | 29 | #include <linux/gfp.h> |
b608b283 | 30 | #include <linux/swap.h> |
1da177e4 LT |
31 | |
32 | #include <asm/uaccess.h> | |
1da177e4 LT |
33 | |
34 | #include "delegation.h" | |
94387fb1 | 35 | #include "internal.h" |
91d5b470 | 36 | #include "iostat.h" |
545db45f | 37 | #include "fscache.h" |
612aa983 | 38 | #include "pnfs.h" |
1da177e4 | 39 | |
f4ce1299 TM |
40 | #include "nfstrace.h" |
41 | ||
1da177e4 LT |
42 | #define NFSDBG_FACILITY NFSDBG_FILE |
43 | ||
f0f37e2f | 44 | static const struct vm_operations_struct nfs_file_vm_ops; |
94387fb1 | 45 | |
1da177e4 LT |
46 | /* Hack for future NFS swap support */ |
47 | #ifndef IS_SWAPFILE | |
48 | # define IS_SWAPFILE(inode) (0) | |
49 | #endif | |
50 | ||
ce4ef7c0 | 51 | int nfs_check_flags(int flags) |
1da177e4 LT |
52 | { |
53 | if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT)) | |
54 | return -EINVAL; | |
55 | ||
56 | return 0; | |
57 | } | |
89d77c8f | 58 | EXPORT_SYMBOL_GPL(nfs_check_flags); |
1da177e4 LT |
59 | |
60 | /* | |
61 | * Open file | |
62 | */ | |
63 | static int | |
64 | nfs_file_open(struct inode *inode, struct file *filp) | |
65 | { | |
1da177e4 LT |
66 | int res; |
67 | ||
6de1472f | 68 | dprintk("NFS: open file(%pD2)\n", filp); |
cc0dd2d1 | 69 | |
c2459dc4 | 70 | nfs_inc_stats(inode, NFSIOS_VFSOPEN); |
1da177e4 LT |
71 | res = nfs_check_flags(filp->f_flags); |
72 | if (res) | |
73 | return res; | |
74 | ||
46cb650c | 75 | res = nfs_open(inode, filp); |
1da177e4 LT |
76 | return res; |
77 | } | |
78 | ||
ce4ef7c0 | 79 | int |
1da177e4 LT |
80 | nfs_file_release(struct inode *inode, struct file *filp) |
81 | { | |
6de1472f | 82 | dprintk("NFS: release(%pD2)\n", filp); |
6da24bc9 | 83 | |
91d5b470 | 84 | nfs_inc_stats(inode, NFSIOS_VFSRELEASE); |
46cb650c | 85 | return nfs_release(inode, filp); |
1da177e4 | 86 | } |
89d77c8f | 87 | EXPORT_SYMBOL_GPL(nfs_file_release); |
1da177e4 | 88 | |
980802e3 TM |
89 | /** |
90 | * nfs_revalidate_size - Revalidate the file size | |
91 | * @inode - pointer to inode struct | |
92 | * @file - pointer to struct file | |
93 | * | |
94 | * Revalidates the file length. This is basically a wrapper around | |
95 | * nfs_revalidate_inode() that takes into account the fact that we may | |
96 | * have cached writes (in which case we don't care about the server's | |
97 | * idea of what the file length is), or O_DIRECT (in which case we | |
98 | * shouldn't trust the cache). | |
99 | */ | |
100 | static int nfs_revalidate_file_size(struct inode *inode, struct file *filp) | |
101 | { | |
102 | struct nfs_server *server = NFS_SERVER(inode); | |
103 | struct nfs_inode *nfsi = NFS_I(inode); | |
104 | ||
d7cf8dd0 TM |
105 | if (nfs_have_delegated_attributes(inode)) |
106 | goto out_noreval; | |
107 | ||
980802e3 TM |
108 | if (filp->f_flags & O_DIRECT) |
109 | goto force_reval; | |
d7cf8dd0 TM |
110 | if (nfsi->cache_validity & NFS_INO_REVAL_PAGECACHE) |
111 | goto force_reval; | |
112 | if (nfs_attribute_timeout(inode)) | |
113 | goto force_reval; | |
114 | out_noreval: | |
115 | return 0; | |
980802e3 TM |
116 | force_reval: |
117 | return __nfs_revalidate_inode(server, inode); | |
118 | } | |
119 | ||
965c8e59 | 120 | loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence) |
980802e3 | 121 | { |
6de1472f AV |
122 | dprintk("NFS: llseek file(%pD2, %lld, %d)\n", |
123 | filp, offset, whence); | |
b84e06c5 | 124 | |
06222e49 | 125 | /* |
965c8e59 | 126 | * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate |
06222e49 JB |
127 | * the cached file length |
128 | */ | |
965c8e59 | 129 | if (whence != SEEK_SET && whence != SEEK_CUR) { |
980802e3 | 130 | struct inode *inode = filp->f_mapping->host; |
d5e66348 | 131 | |
980802e3 TM |
132 | int retval = nfs_revalidate_file_size(inode, filp); |
133 | if (retval < 0) | |
134 | return (loff_t)retval; | |
79835a71 | 135 | } |
d5e66348 | 136 | |
965c8e59 | 137 | return generic_file_llseek(filp, offset, whence); |
980802e3 | 138 | } |
89d77c8f | 139 | EXPORT_SYMBOL_GPL(nfs_file_llseek); |
980802e3 | 140 | |
1da177e4 LT |
141 | /* |
142 | * Flush all dirty pages, and check for write errors. | |
1da177e4 | 143 | */ |
ce4ef7c0 | 144 | int |
75e1fcc0 | 145 | nfs_file_flush(struct file *file, fl_owner_t id) |
1da177e4 | 146 | { |
6de1472f | 147 | struct inode *inode = file_inode(file); |
1da177e4 | 148 | |
6de1472f | 149 | dprintk("NFS: flush(%pD2)\n", file); |
1da177e4 | 150 | |
c2459dc4 | 151 | nfs_inc_stats(inode, NFSIOS_VFSFLUSH); |
1da177e4 LT |
152 | if ((file->f_mode & FMODE_WRITE) == 0) |
153 | return 0; | |
7b159fc1 | 154 | |
14546c33 TM |
155 | /* |
156 | * If we're holding a write delegation, then just start the i/o | |
157 | * but don't wait for completion (or send a commit). | |
158 | */ | |
011e2a7f | 159 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) |
14546c33 TM |
160 | return filemap_fdatawrite(file->f_mapping); |
161 | ||
7fe5c398 | 162 | /* Flush writes to the server and return any errors */ |
af7fa165 | 163 | return vfs_fsync(file, 0); |
1da177e4 | 164 | } |
89d77c8f | 165 | EXPORT_SYMBOL_GPL(nfs_file_flush); |
1da177e4 | 166 | |
ce4ef7c0 | 167 | ssize_t |
3aa2d199 | 168 | nfs_file_read(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 169 | { |
6de1472f | 170 | struct inode *inode = file_inode(iocb->ki_filp); |
1da177e4 LT |
171 | ssize_t result; |
172 | ||
2ba48ce5 | 173 | if (iocb->ki_flags & IOCB_DIRECT) |
e19a8a0a | 174 | return nfs_file_direct_read(iocb, to, iocb->ki_pos); |
1da177e4 | 175 | |
619d30b4 | 176 | dprintk("NFS: read(%pD2, %zu@%lu)\n", |
6de1472f | 177 | iocb->ki_filp, |
3aa2d199 | 178 | iov_iter_count(to), (unsigned long) iocb->ki_pos); |
1da177e4 | 179 | |
874f9463 | 180 | result = nfs_revalidate_mapping_protected(inode, iocb->ki_filp->f_mapping); |
4184dcf2 | 181 | if (!result) { |
3aa2d199 | 182 | result = generic_file_read_iter(iocb, to); |
4184dcf2 CL |
183 | if (result > 0) |
184 | nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result); | |
185 | } | |
1da177e4 LT |
186 | return result; |
187 | } | |
89d77c8f | 188 | EXPORT_SYMBOL_GPL(nfs_file_read); |
1da177e4 | 189 | |
ce4ef7c0 | 190 | ssize_t |
f0930fff JA |
191 | nfs_file_splice_read(struct file *filp, loff_t *ppos, |
192 | struct pipe_inode_info *pipe, size_t count, | |
193 | unsigned int flags) | |
1da177e4 | 194 | { |
6de1472f | 195 | struct inode *inode = file_inode(filp); |
1da177e4 LT |
196 | ssize_t res; |
197 | ||
6de1472f AV |
198 | dprintk("NFS: splice_read(%pD2, %lu@%Lu)\n", |
199 | filp, (unsigned long) count, (unsigned long long) *ppos); | |
1da177e4 | 200 | |
874f9463 | 201 | res = nfs_revalidate_mapping_protected(inode, filp->f_mapping); |
aa2f1ef1 | 202 | if (!res) { |
f0930fff | 203 | res = generic_file_splice_read(filp, ppos, pipe, count, flags); |
aa2f1ef1 CL |
204 | if (res > 0) |
205 | nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res); | |
206 | } | |
1da177e4 LT |
207 | return res; |
208 | } | |
89d77c8f | 209 | EXPORT_SYMBOL_GPL(nfs_file_splice_read); |
1da177e4 | 210 | |
ce4ef7c0 | 211 | int |
1da177e4 LT |
212 | nfs_file_mmap(struct file * file, struct vm_area_struct * vma) |
213 | { | |
6de1472f | 214 | struct inode *inode = file_inode(file); |
1da177e4 LT |
215 | int status; |
216 | ||
6de1472f | 217 | dprintk("NFS: mmap(%pD2)\n", file); |
1da177e4 | 218 | |
e1ebfd33 TM |
219 | /* Note: generic_file_mmap() returns ENOSYS on nommu systems |
220 | * so we call that before revalidating the mapping | |
221 | */ | |
222 | status = generic_file_mmap(file, vma); | |
94387fb1 TM |
223 | if (!status) { |
224 | vma->vm_ops = &nfs_file_vm_ops; | |
e1ebfd33 | 225 | status = nfs_revalidate_mapping(inode, file->f_mapping); |
94387fb1 | 226 | } |
1da177e4 LT |
227 | return status; |
228 | } | |
89d77c8f | 229 | EXPORT_SYMBOL_GPL(nfs_file_mmap); |
1da177e4 LT |
230 | |
231 | /* | |
232 | * Flush any dirty pages for this process, and check for write errors. | |
233 | * The return status from this call provides a reliable indication of | |
234 | * whether any write errors occurred for this process. | |
af7fa165 TM |
235 | * |
236 | * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to | |
237 | * disk, but it retrieves and clears ctx->error after synching, despite | |
238 | * the two being set at the same time in nfs_context_set_write_error(). | |
239 | * This is because the former is used to notify the _next_ call to | |
25985edc | 240 | * nfs_file_write() that a write error occurred, and hence cause it to |
af7fa165 | 241 | * fall back to doing a synchronous write. |
1da177e4 | 242 | */ |
ce4ef7c0 | 243 | int |
a5c58892 | 244 | nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync) |
1da177e4 | 245 | { |
cd3758e3 | 246 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
6de1472f | 247 | struct inode *inode = file_inode(file); |
05990d1b | 248 | int have_error, do_resend, status; |
af7fa165 TM |
249 | int ret = 0; |
250 | ||
6de1472f | 251 | dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync); |
1da177e4 | 252 | |
91d5b470 | 253 | nfs_inc_stats(inode, NFSIOS_VFSFSYNC); |
05990d1b | 254 | do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags); |
af7fa165 TM |
255 | have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); |
256 | status = nfs_commit_inode(inode, FLUSH_SYNC); | |
257 | have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); | |
05990d1b | 258 | if (have_error) { |
af7fa165 | 259 | ret = xchg(&ctx->error, 0); |
05990d1b TM |
260 | if (ret) |
261 | goto out; | |
262 | } | |
263 | if (status < 0) { | |
af7fa165 | 264 | ret = status; |
05990d1b TM |
265 | goto out; |
266 | } | |
267 | do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags); | |
268 | if (do_resend) | |
269 | ret = -EAGAIN; | |
270 | out: | |
a5c58892 BS |
271 | return ret; |
272 | } | |
89d77c8f | 273 | EXPORT_SYMBOL_GPL(nfs_file_fsync_commit); |
a5c58892 BS |
274 | |
275 | static int | |
276 | nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync) | |
277 | { | |
278 | int ret; | |
496ad9aa | 279 | struct inode *inode = file_inode(file); |
a5c58892 | 280 | |
f4ce1299 TM |
281 | trace_nfs_fsync_enter(inode); |
282 | ||
05990d1b TM |
283 | do { |
284 | ret = filemap_write_and_wait_range(inode->i_mapping, start, end); | |
285 | if (ret != 0) | |
286 | break; | |
287 | mutex_lock(&inode->i_mutex); | |
288 | ret = nfs_file_fsync_commit(file, start, end, datasync); | |
289 | mutex_unlock(&inode->i_mutex); | |
dcfc4f25 TM |
290 | /* |
291 | * If nfs_file_fsync_commit detected a server reboot, then | |
292 | * resend all dirty pages that might have been covered by | |
293 | * the NFS_CONTEXT_RESEND_WRITES flag | |
294 | */ | |
295 | start = 0; | |
296 | end = LLONG_MAX; | |
05990d1b TM |
297 | } while (ret == -EAGAIN); |
298 | ||
f4ce1299 | 299 | trace_nfs_fsync_exit(inode, ret); |
af7fa165 | 300 | return ret; |
1da177e4 LT |
301 | } |
302 | ||
38c73044 PS |
303 | /* |
304 | * Decide whether a read/modify/write cycle may be more efficient | |
305 | * then a modify/write/read cycle when writing to a page in the | |
306 | * page cache. | |
307 | * | |
308 | * The modify/write/read cycle may occur if a page is read before | |
309 | * being completely filled by the writer. In this situation, the | |
310 | * page must be completely written to stable storage on the server | |
311 | * before it can be refilled by reading in the page from the server. | |
312 | * This can lead to expensive, small, FILE_SYNC mode writes being | |
313 | * done. | |
314 | * | |
315 | * It may be more efficient to read the page first if the file is | |
316 | * open for reading in addition to writing, the page is not marked | |
317 | * as Uptodate, it is not dirty or waiting to be committed, | |
318 | * indicating that it was previously allocated and then modified, | |
319 | * that there were valid bytes of data in that range of the file, | |
320 | * and that the new data won't completely replace the old data in | |
321 | * that range of the file. | |
322 | */ | |
323 | static int nfs_want_read_modify_write(struct file *file, struct page *page, | |
324 | loff_t pos, unsigned len) | |
325 | { | |
326 | unsigned int pglen = nfs_page_length(page); | |
327 | unsigned int offset = pos & (PAGE_CACHE_SIZE - 1); | |
328 | unsigned int end = offset + len; | |
329 | ||
612aa983 CH |
330 | if (pnfs_ld_read_whole_page(file->f_mapping->host)) { |
331 | if (!PageUptodate(page)) | |
332 | return 1; | |
333 | return 0; | |
334 | } | |
335 | ||
38c73044 PS |
336 | if ((file->f_mode & FMODE_READ) && /* open for read? */ |
337 | !PageUptodate(page) && /* Uptodate? */ | |
338 | !PagePrivate(page) && /* i/o request already? */ | |
339 | pglen && /* valid bytes of file? */ | |
340 | (end < pglen || offset)) /* replace all valid bytes? */ | |
341 | return 1; | |
342 | return 0; | |
343 | } | |
344 | ||
1da177e4 | 345 | /* |
4899f9c8 NP |
346 | * This does the "real" work of the write. We must allocate and lock the |
347 | * page to be sent back to the generic routine, which then copies the | |
348 | * data from user space. | |
1da177e4 LT |
349 | * |
350 | * If the writer ends up delaying the write, the writer needs to | |
351 | * increment the page use counts until he is done with the page. | |
352 | */ | |
4899f9c8 NP |
353 | static int nfs_write_begin(struct file *file, struct address_space *mapping, |
354 | loff_t pos, unsigned len, unsigned flags, | |
355 | struct page **pagep, void **fsdata) | |
1da177e4 | 356 | { |
4899f9c8 | 357 | int ret; |
38c73044 | 358 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
4899f9c8 | 359 | struct page *page; |
38c73044 | 360 | int once_thru = 0; |
4899f9c8 | 361 | |
1e8968c5 | 362 | dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n", |
6de1472f | 363 | file, mapping->host->i_ino, len, (long long) pos); |
b7eaefaa | 364 | |
38c73044 | 365 | start: |
72cb77f4 TM |
366 | /* |
367 | * Prevent starvation issues if someone is doing a consistency | |
368 | * sync-to-disk | |
369 | */ | |
74316201 N |
370 | ret = wait_on_bit_action(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING, |
371 | nfs_wait_bit_killable, TASK_KILLABLE); | |
72cb77f4 TM |
372 | if (ret) |
373 | return ret; | |
aa5accea TM |
374 | /* |
375 | * Wait for O_DIRECT to complete | |
376 | */ | |
377 | nfs_inode_dio_wait(mapping->host); | |
72cb77f4 | 378 | |
54566b2c | 379 | page = grab_cache_page_write_begin(mapping, index, flags); |
4899f9c8 NP |
380 | if (!page) |
381 | return -ENOMEM; | |
382 | *pagep = page; | |
383 | ||
384 | ret = nfs_flush_incompatible(file, page); | |
385 | if (ret) { | |
386 | unlock_page(page); | |
387 | page_cache_release(page); | |
38c73044 PS |
388 | } else if (!once_thru && |
389 | nfs_want_read_modify_write(file, page, pos, len)) { | |
390 | once_thru = 1; | |
391 | ret = nfs_readpage(file, page); | |
392 | page_cache_release(page); | |
393 | if (!ret) | |
394 | goto start; | |
4899f9c8 NP |
395 | } |
396 | return ret; | |
1da177e4 LT |
397 | } |
398 | ||
4899f9c8 NP |
399 | static int nfs_write_end(struct file *file, struct address_space *mapping, |
400 | loff_t pos, unsigned len, unsigned copied, | |
401 | struct page *page, void *fsdata) | |
1da177e4 | 402 | { |
4899f9c8 | 403 | unsigned offset = pos & (PAGE_CACHE_SIZE - 1); |
dc24826b | 404 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
4899f9c8 | 405 | int status; |
1da177e4 | 406 | |
1e8968c5 | 407 | dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n", |
6de1472f | 408 | file, mapping->host->i_ino, len, (long long) pos); |
b7eaefaa | 409 | |
efc91ed0 TM |
410 | /* |
411 | * Zero any uninitialised parts of the page, and then mark the page | |
412 | * as up to date if it turns out that we're extending the file. | |
413 | */ | |
414 | if (!PageUptodate(page)) { | |
415 | unsigned pglen = nfs_page_length(page); | |
416 | unsigned end = offset + len; | |
417 | ||
418 | if (pglen == 0) { | |
419 | zero_user_segments(page, 0, offset, | |
420 | end, PAGE_CACHE_SIZE); | |
421 | SetPageUptodate(page); | |
422 | } else if (end >= pglen) { | |
423 | zero_user_segment(page, end, PAGE_CACHE_SIZE); | |
424 | if (offset == 0) | |
425 | SetPageUptodate(page); | |
426 | } else | |
427 | zero_user_segment(page, pglen, PAGE_CACHE_SIZE); | |
428 | } | |
429 | ||
4899f9c8 | 430 | status = nfs_updatepage(file, page, offset, copied); |
4899f9c8 NP |
431 | |
432 | unlock_page(page); | |
433 | page_cache_release(page); | |
434 | ||
3d509e54 CL |
435 | if (status < 0) |
436 | return status; | |
2701d086 | 437 | NFS_I(mapping->host)->write_io += copied; |
dc24826b AA |
438 | |
439 | if (nfs_ctx_key_to_expire(ctx)) { | |
440 | status = nfs_wb_all(mapping->host); | |
441 | if (status < 0) | |
442 | return status; | |
443 | } | |
444 | ||
3d509e54 | 445 | return copied; |
1da177e4 LT |
446 | } |
447 | ||
6b9b3514 DH |
448 | /* |
449 | * Partially or wholly invalidate a page | |
450 | * - Release the private state associated with a page if undergoing complete | |
451 | * page invalidation | |
545db45f | 452 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
453 | * - Caller holds page lock |
454 | */ | |
d47992f8 LC |
455 | static void nfs_invalidate_page(struct page *page, unsigned int offset, |
456 | unsigned int length) | |
cd52ed35 | 457 | { |
d47992f8 LC |
458 | dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n", |
459 | page, offset, length); | |
b7eaefaa | 460 | |
d47992f8 | 461 | if (offset != 0 || length < PAGE_CACHE_SIZE) |
1c75950b | 462 | return; |
d2ccddf0 | 463 | /* Cancel any unstarted writes on this page */ |
d56b4ddf | 464 | nfs_wb_page_cancel(page_file_mapping(page)->host, page); |
545db45f DH |
465 | |
466 | nfs_fscache_invalidate_page(page, page->mapping->host); | |
cd52ed35 TM |
467 | } |
468 | ||
6b9b3514 DH |
469 | /* |
470 | * Attempt to release the private state associated with a page | |
545db45f | 471 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
472 | * - Caller holds page lock |
473 | * - Return true (may release page) or false (may not) | |
474 | */ | |
cd52ed35 TM |
475 | static int nfs_release_page(struct page *page, gfp_t gfp) |
476 | { | |
b608b283 TM |
477 | struct address_space *mapping = page->mapping; |
478 | ||
b7eaefaa CL |
479 | dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page); |
480 | ||
95905446 | 481 | /* Always try to initiate a 'commit' if relevant, but only |
1aff5256 N |
482 | * wait for it if __GFP_WAIT is set. Even then, only wait 1 |
483 | * second and only if the 'bdi' is not congested. | |
95905446 | 484 | * Waiting indefinitely can cause deadlocks when the NFS |
1aff5256 N |
485 | * server is on this machine, when a new TCP connection is |
486 | * needed and in other rare cases. There is no particular | |
487 | * need to wait extensively here. A short wait has the | |
488 | * benefit that someone else can worry about the freezer. | |
5cf02d09 | 489 | */ |
95905446 N |
490 | if (mapping) { |
491 | struct nfs_server *nfss = NFS_SERVER(mapping->host); | |
492 | nfs_commit_inode(mapping->host, 0); | |
493 | if ((gfp & __GFP_WAIT) && | |
353db796 | 494 | !bdi_write_congested(&nfss->backing_dev_info)) { |
95905446 N |
495 | wait_on_page_bit_killable_timeout(page, PG_private, |
496 | HZ); | |
353db796 N |
497 | if (PagePrivate(page)) |
498 | set_bdi_congested(&nfss->backing_dev_info, | |
499 | BLK_RW_ASYNC); | |
95905446 | 500 | } |
b608b283 | 501 | } |
e3db7691 | 502 | /* If PagePrivate() is set, then the page is not freeable */ |
545db45f DH |
503 | if (PagePrivate(page)) |
504 | return 0; | |
505 | return nfs_fscache_release_page(page, gfp); | |
e3db7691 TM |
506 | } |
507 | ||
f919b196 MG |
508 | static void nfs_check_dirty_writeback(struct page *page, |
509 | bool *dirty, bool *writeback) | |
510 | { | |
511 | struct nfs_inode *nfsi; | |
512 | struct address_space *mapping = page_file_mapping(page); | |
513 | ||
514 | if (!mapping || PageSwapCache(page)) | |
515 | return; | |
516 | ||
517 | /* | |
518 | * Check if an unstable page is currently being committed and | |
519 | * if so, have the VM treat it as if the page is under writeback | |
520 | * so it will not block due to pages that will shortly be freeable. | |
521 | */ | |
522 | nfsi = NFS_I(mapping->host); | |
523 | if (test_bit(NFS_INO_COMMIT, &nfsi->flags)) { | |
524 | *writeback = true; | |
525 | return; | |
526 | } | |
527 | ||
528 | /* | |
529 | * If PagePrivate() is set, then the page is not freeable and as the | |
530 | * inode is not being committed, it's not going to be cleaned in the | |
531 | * near future so treat it as dirty | |
532 | */ | |
533 | if (PagePrivate(page)) | |
534 | *dirty = true; | |
535 | } | |
536 | ||
6b9b3514 DH |
537 | /* |
538 | * Attempt to clear the private state associated with a page when an error | |
539 | * occurs that requires the cached contents of an inode to be written back or | |
540 | * destroyed | |
545db45f | 541 | * - Called if either PG_private or fscache is set on the page |
6b9b3514 DH |
542 | * - Caller holds page lock |
543 | * - Return 0 if successful, -error otherwise | |
544 | */ | |
e3db7691 TM |
545 | static int nfs_launder_page(struct page *page) |
546 | { | |
d56b4ddf | 547 | struct inode *inode = page_file_mapping(page)->host; |
545db45f | 548 | struct nfs_inode *nfsi = NFS_I(inode); |
b7eaefaa CL |
549 | |
550 | dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n", | |
551 | inode->i_ino, (long long)page_offset(page)); | |
552 | ||
545db45f | 553 | nfs_fscache_wait_on_page_write(nfsi, page); |
b7eaefaa | 554 | return nfs_wb_page(inode, page); |
cd52ed35 TM |
555 | } |
556 | ||
a564b8f0 MG |
557 | #ifdef CONFIG_NFS_SWAP |
558 | static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file, | |
559 | sector_t *span) | |
560 | { | |
dad2b015 JL |
561 | int ret; |
562 | struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host); | |
563 | ||
a564b8f0 | 564 | *span = sis->pages; |
dad2b015 JL |
565 | |
566 | rcu_read_lock(); | |
567 | ret = xs_swapper(rcu_dereference(clnt->cl_xprt), 1); | |
568 | rcu_read_unlock(); | |
569 | ||
570 | return ret; | |
a564b8f0 MG |
571 | } |
572 | ||
573 | static void nfs_swap_deactivate(struct file *file) | |
574 | { | |
dad2b015 JL |
575 | struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host); |
576 | ||
577 | rcu_read_lock(); | |
578 | xs_swapper(rcu_dereference(clnt->cl_xprt), 0); | |
579 | rcu_read_unlock(); | |
a564b8f0 MG |
580 | } |
581 | #endif | |
582 | ||
f5e54d6e | 583 | const struct address_space_operations nfs_file_aops = { |
1da177e4 LT |
584 | .readpage = nfs_readpage, |
585 | .readpages = nfs_readpages, | |
9cccef95 | 586 | .set_page_dirty = __set_page_dirty_nobuffers, |
1da177e4 LT |
587 | .writepage = nfs_writepage, |
588 | .writepages = nfs_writepages, | |
4899f9c8 NP |
589 | .write_begin = nfs_write_begin, |
590 | .write_end = nfs_write_end, | |
cd52ed35 TM |
591 | .invalidatepage = nfs_invalidate_page, |
592 | .releasepage = nfs_release_page, | |
1da177e4 | 593 | .direct_IO = nfs_direct_IO, |
074cc1de | 594 | .migratepage = nfs_migrate_page, |
e3db7691 | 595 | .launder_page = nfs_launder_page, |
f919b196 | 596 | .is_dirty_writeback = nfs_check_dirty_writeback, |
f590f333 | 597 | .error_remove_page = generic_error_remove_page, |
a564b8f0 MG |
598 | #ifdef CONFIG_NFS_SWAP |
599 | .swap_activate = nfs_swap_activate, | |
600 | .swap_deactivate = nfs_swap_deactivate, | |
601 | #endif | |
1da177e4 LT |
602 | }; |
603 | ||
6b9b3514 DH |
604 | /* |
605 | * Notification that a PTE pointing to an NFS page is about to be made | |
606 | * writable, implying that someone is about to modify the page through a | |
607 | * shared-writable mapping | |
608 | */ | |
c2ec175c | 609 | static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
94387fb1 | 610 | { |
c2ec175c | 611 | struct page *page = vmf->page; |
94387fb1 | 612 | struct file *filp = vma->vm_file; |
6de1472f | 613 | struct inode *inode = file_inode(filp); |
94387fb1 | 614 | unsigned pagelen; |
bc4866b6 | 615 | int ret = VM_FAULT_NOPAGE; |
4899f9c8 | 616 | struct address_space *mapping; |
94387fb1 | 617 | |
1e8968c5 | 618 | dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n", |
6de1472f | 619 | filp, filp->f_mapping->host->i_ino, |
b7eaefaa CL |
620 | (long long)page_offset(page)); |
621 | ||
545db45f | 622 | /* make sure the cache has finished storing the page */ |
6de1472f | 623 | nfs_fscache_wait_on_page_write(NFS_I(inode), page); |
545db45f | 624 | |
ef070dcb TM |
625 | wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING, |
626 | nfs_wait_bit_killable, TASK_KILLABLE); | |
627 | ||
94387fb1 | 628 | lock_page(page); |
d56b4ddf | 629 | mapping = page_file_mapping(page); |
6de1472f | 630 | if (mapping != inode->i_mapping) |
8b1f9ee5 TM |
631 | goto out_unlock; |
632 | ||
2aeb98f4 TM |
633 | wait_on_page_writeback(page); |
634 | ||
94387fb1 | 635 | pagelen = nfs_page_length(page); |
8b1f9ee5 TM |
636 | if (pagelen == 0) |
637 | goto out_unlock; | |
4899f9c8 | 638 | |
bc4866b6 TM |
639 | ret = VM_FAULT_LOCKED; |
640 | if (nfs_flush_incompatible(filp, page) == 0 && | |
641 | nfs_updatepage(filp, page, 0, pagelen) == 0) | |
642 | goto out; | |
8b1f9ee5 | 643 | |
bc4866b6 | 644 | ret = VM_FAULT_SIGBUS; |
8b1f9ee5 TM |
645 | out_unlock: |
646 | unlock_page(page); | |
bc4866b6 TM |
647 | out: |
648 | return ret; | |
94387fb1 TM |
649 | } |
650 | ||
f0f37e2f | 651 | static const struct vm_operations_struct nfs_file_vm_ops = { |
94387fb1 | 652 | .fault = filemap_fault, |
f1820361 | 653 | .map_pages = filemap_map_pages, |
94387fb1 TM |
654 | .page_mkwrite = nfs_vm_page_mkwrite, |
655 | }; | |
656 | ||
7b159fc1 TM |
657 | static int nfs_need_sync_write(struct file *filp, struct inode *inode) |
658 | { | |
659 | struct nfs_open_context *ctx; | |
660 | ||
6b2f3d1f | 661 | if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC)) |
7b159fc1 | 662 | return 1; |
cd3758e3 | 663 | ctx = nfs_file_open_context(filp); |
dc24826b AA |
664 | if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) || |
665 | nfs_ctx_key_to_expire(ctx)) | |
7b159fc1 TM |
666 | return 1; |
667 | return 0; | |
668 | } | |
669 | ||
edaf4369 | 670 | ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 671 | { |
6de1472f AV |
672 | struct file *file = iocb->ki_filp; |
673 | struct inode *inode = file_inode(file); | |
7e381172 | 674 | unsigned long written = 0; |
1da177e4 | 675 | ssize_t result; |
edaf4369 | 676 | size_t count = iov_iter_count(from); |
1da177e4 | 677 | |
6de1472f | 678 | result = nfs_key_timeout_notify(file, inode); |
dc24826b AA |
679 | if (result) |
680 | return result; | |
681 | ||
65a4a1ca AV |
682 | if (iocb->ki_flags & IOCB_DIRECT) { |
683 | result = generic_write_checks(iocb, from); | |
684 | if (result <= 0) | |
685 | return result; | |
686 | return nfs_file_direct_write(iocb, from); | |
687 | } | |
1da177e4 | 688 | |
619d30b4 | 689 | dprintk("NFS: write(%pD2, %zu@%Ld)\n", |
65a4a1ca | 690 | file, count, (long long) iocb->ki_pos); |
1da177e4 LT |
691 | |
692 | result = -EBUSY; | |
693 | if (IS_SWAPFILE(inode)) | |
694 | goto out_swapfile; | |
7d52e862 TM |
695 | /* |
696 | * O_APPEND implies that we must revalidate the file length. | |
697 | */ | |
2ba48ce5 | 698 | if (iocb->ki_flags & IOCB_APPEND) { |
6de1472f | 699 | result = nfs_revalidate_file_size(inode, file); |
7d52e862 TM |
700 | if (result) |
701 | goto out; | |
fe51beec | 702 | } |
1da177e4 LT |
703 | |
704 | result = count; | |
705 | if (!count) | |
706 | goto out; | |
707 | ||
edaf4369 | 708 | result = generic_file_write_iter(iocb, from); |
7e381172 CL |
709 | if (result > 0) |
710 | written = result; | |
711 | ||
6b2f3d1f | 712 | /* Return error values for O_DSYNC and IS_SYNC() */ |
6de1472f AV |
713 | if (result >= 0 && nfs_need_sync_write(file, inode)) { |
714 | int err = vfs_fsync(file, 0); | |
200baa21 TM |
715 | if (err < 0) |
716 | result = err; | |
717 | } | |
7e381172 CL |
718 | if (result > 0) |
719 | nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written); | |
1da177e4 LT |
720 | out: |
721 | return result; | |
722 | ||
723 | out_swapfile: | |
724 | printk(KERN_INFO "NFS: attempt to write to active swap file!\n"); | |
725 | goto out; | |
726 | } | |
89d77c8f | 727 | EXPORT_SYMBOL_GPL(nfs_file_write); |
1da177e4 | 728 | |
5eebde23 SJ |
729 | static int |
730 | do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
731 | { |
732 | struct inode *inode = filp->f_mapping->host; | |
733 | int status = 0; | |
21ac19d4 | 734 | unsigned int saved_type = fl->fl_type; |
1da177e4 | 735 | |
039c4d7a | 736 | /* Try local locking first */ |
6d34ac19 BF |
737 | posix_test_lock(filp, fl); |
738 | if (fl->fl_type != F_UNLCK) { | |
739 | /* found a conflict */ | |
039c4d7a | 740 | goto out; |
1da177e4 | 741 | } |
21ac19d4 | 742 | fl->fl_type = saved_type; |
039c4d7a | 743 | |
011e2a7f | 744 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) |
039c4d7a TM |
745 | goto out_noconflict; |
746 | ||
5eebde23 | 747 | if (is_local) |
039c4d7a TM |
748 | goto out_noconflict; |
749 | ||
750 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); | |
751 | out: | |
1da177e4 | 752 | return status; |
039c4d7a TM |
753 | out_noconflict: |
754 | fl->fl_type = F_UNLCK; | |
755 | goto out; | |
1da177e4 LT |
756 | } |
757 | ||
758 | static int do_vfs_lock(struct file *file, struct file_lock *fl) | |
759 | { | |
760 | int res = 0; | |
761 | switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) { | |
762 | case FL_POSIX: | |
763 | res = posix_lock_file_wait(file, fl); | |
764 | break; | |
765 | case FL_FLOCK: | |
766 | res = flock_lock_file_wait(file, fl); | |
767 | break; | |
768 | default: | |
769 | BUG(); | |
770 | } | |
1da177e4 LT |
771 | return res; |
772 | } | |
773 | ||
5eebde23 SJ |
774 | static int |
775 | do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
776 | { |
777 | struct inode *inode = filp->f_mapping->host; | |
7a8203d8 | 778 | struct nfs_lock_context *l_ctx; |
1da177e4 LT |
779 | int status; |
780 | ||
1da177e4 LT |
781 | /* |
782 | * Flush all pending writes before doing anything | |
783 | * with locks.. | |
784 | */ | |
29884df0 | 785 | nfs_sync_mapping(filp->f_mapping); |
1da177e4 | 786 | |
7a8203d8 TM |
787 | l_ctx = nfs_get_lock_context(nfs_file_open_context(filp)); |
788 | if (!IS_ERR(l_ctx)) { | |
789 | status = nfs_iocounter_wait(&l_ctx->io_count); | |
790 | nfs_put_lock_context(l_ctx); | |
791 | if (status < 0) | |
792 | return status; | |
793 | } | |
794 | ||
1da177e4 LT |
795 | /* NOTE: special case |
796 | * If we're signalled while cleaning up locks on process exit, we | |
797 | * still need to complete the unlock. | |
798 | */ | |
5eebde23 SJ |
799 | /* |
800 | * Use local locking if mounted with "-onolock" or with appropriate | |
801 | * "-olocal_lock=" | |
802 | */ | |
803 | if (!is_local) | |
1da177e4 LT |
804 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
805 | else | |
806 | status = do_vfs_lock(filp, fl); | |
1da177e4 LT |
807 | return status; |
808 | } | |
809 | ||
6b96724e RL |
810 | static int |
811 | is_time_granular(struct timespec *ts) { | |
812 | return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000)); | |
813 | } | |
814 | ||
5eebde23 SJ |
815 | static int |
816 | do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
817 | { |
818 | struct inode *inode = filp->f_mapping->host; | |
1da177e4 LT |
819 | int status; |
820 | ||
1da177e4 LT |
821 | /* |
822 | * Flush all pending writes before doing anything | |
823 | * with locks.. | |
824 | */ | |
29884df0 TM |
825 | status = nfs_sync_mapping(filp->f_mapping); |
826 | if (status != 0) | |
1da177e4 LT |
827 | goto out; |
828 | ||
5eebde23 SJ |
829 | /* |
830 | * Use local locking if mounted with "-onolock" or with appropriate | |
831 | * "-olocal_lock=" | |
832 | */ | |
833 | if (!is_local) | |
1da177e4 | 834 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
c4d7c402 | 835 | else |
1da177e4 | 836 | status = do_vfs_lock(filp, fl); |
1da177e4 LT |
837 | if (status < 0) |
838 | goto out; | |
6b96724e | 839 | |
1da177e4 | 840 | /* |
6b96724e RL |
841 | * Revalidate the cache if the server has time stamps granular |
842 | * enough to detect subsecond changes. Otherwise, clear the | |
843 | * cache to prevent missing any changes. | |
844 | * | |
1da177e4 LT |
845 | * This makes locking act as a cache coherency point. |
846 | */ | |
29884df0 | 847 | nfs_sync_mapping(filp->f_mapping); |
011e2a7f | 848 | if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) { |
6b96724e RL |
849 | if (is_time_granular(&NFS_SERVER(inode)->time_delta)) |
850 | __nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
851 | else | |
852 | nfs_zap_caches(inode); | |
853 | } | |
1da177e4 | 854 | out: |
1da177e4 LT |
855 | return status; |
856 | } | |
857 | ||
858 | /* | |
859 | * Lock a (portion of) a file | |
860 | */ | |
ce4ef7c0 | 861 | int nfs_lock(struct file *filp, int cmd, struct file_lock *fl) |
1da177e4 | 862 | { |
6da24bc9 | 863 | struct inode *inode = filp->f_mapping->host; |
2116271a | 864 | int ret = -ENOLCK; |
5eebde23 | 865 | int is_local = 0; |
1da177e4 | 866 | |
6de1472f AV |
867 | dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n", |
868 | filp, fl->fl_type, fl->fl_flags, | |
1da177e4 | 869 | (long long)fl->fl_start, (long long)fl->fl_end); |
6da24bc9 | 870 | |
91d5b470 | 871 | nfs_inc_stats(inode, NFSIOS_VFSLOCK); |
1da177e4 LT |
872 | |
873 | /* No mandatory locks over NFS */ | |
dfad9441 | 874 | if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK) |
2116271a TM |
875 | goto out_err; |
876 | ||
5eebde23 SJ |
877 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL) |
878 | is_local = 1; | |
879 | ||
2116271a TM |
880 | if (NFS_PROTO(inode)->lock_check_bounds != NULL) { |
881 | ret = NFS_PROTO(inode)->lock_check_bounds(fl); | |
882 | if (ret < 0) | |
883 | goto out_err; | |
884 | } | |
1da177e4 LT |
885 | |
886 | if (IS_GETLK(cmd)) | |
5eebde23 | 887 | ret = do_getlk(filp, cmd, fl, is_local); |
2116271a | 888 | else if (fl->fl_type == F_UNLCK) |
5eebde23 | 889 | ret = do_unlk(filp, cmd, fl, is_local); |
2116271a | 890 | else |
5eebde23 | 891 | ret = do_setlk(filp, cmd, fl, is_local); |
2116271a TM |
892 | out_err: |
893 | return ret; | |
1da177e4 | 894 | } |
89d77c8f | 895 | EXPORT_SYMBOL_GPL(nfs_lock); |
1da177e4 LT |
896 | |
897 | /* | |
898 | * Lock a (portion of) a file | |
899 | */ | |
ce4ef7c0 | 900 | int nfs_flock(struct file *filp, int cmd, struct file_lock *fl) |
1da177e4 | 901 | { |
5eebde23 SJ |
902 | struct inode *inode = filp->f_mapping->host; |
903 | int is_local = 0; | |
904 | ||
6de1472f AV |
905 | dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n", |
906 | filp, fl->fl_type, fl->fl_flags); | |
1da177e4 | 907 | |
1da177e4 LT |
908 | if (!(fl->fl_flags & FL_FLOCK)) |
909 | return -ENOLCK; | |
910 | ||
ad0fcd4e JL |
911 | /* |
912 | * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of | |
913 | * any standard. In principle we might be able to support LOCK_MAND | |
914 | * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the | |
915 | * NFS code is not set up for it. | |
916 | */ | |
917 | if (fl->fl_type & LOCK_MAND) | |
918 | return -EINVAL; | |
919 | ||
5eebde23 SJ |
920 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK) |
921 | is_local = 1; | |
922 | ||
1da177e4 | 923 | /* We're simulating flock() locks using posix locks on the server */ |
1da177e4 | 924 | if (fl->fl_type == F_UNLCK) |
5eebde23 SJ |
925 | return do_unlk(filp, cmd, fl, is_local); |
926 | return do_setlk(filp, cmd, fl, is_local); | |
1da177e4 | 927 | } |
89d77c8f | 928 | EXPORT_SYMBOL_GPL(nfs_flock); |
370f6599 | 929 | |
0486958f JL |
930 | const struct file_operations nfs_file_operations = { |
931 | .llseek = nfs_file_llseek, | |
3aa2d199 | 932 | .read_iter = nfs_file_read, |
edaf4369 | 933 | .write_iter = nfs_file_write, |
0486958f JL |
934 | .mmap = nfs_file_mmap, |
935 | .open = nfs_file_open, | |
936 | .flush = nfs_file_flush, | |
937 | .release = nfs_file_release, | |
938 | .fsync = nfs_file_fsync, | |
939 | .lock = nfs_lock, | |
940 | .flock = nfs_flock, | |
941 | .splice_read = nfs_file_splice_read, | |
4da54c21 | 942 | .splice_write = iter_file_splice_write, |
0486958f | 943 | .check_flags = nfs_check_flags, |
1c994a09 | 944 | .setlease = simple_nosetlease, |
0486958f | 945 | }; |
ddda8e0a | 946 | EXPORT_SYMBOL_GPL(nfs_file_operations); |