]>
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> |
e8edc6e0 | 29 | #include <linux/aio.h> |
5a0e3ad6 | 30 | #include <linux/gfp.h> |
b608b283 | 31 | #include <linux/swap.h> |
1da177e4 LT |
32 | |
33 | #include <asm/uaccess.h> | |
1da177e4 LT |
34 | |
35 | #include "delegation.h" | |
94387fb1 | 36 | #include "internal.h" |
91d5b470 | 37 | #include "iostat.h" |
545db45f | 38 | #include "fscache.h" |
612aa983 | 39 | #include "pnfs.h" |
1da177e4 | 40 | |
f4ce1299 TM |
41 | #include "nfstrace.h" |
42 | ||
1da177e4 LT |
43 | #define NFSDBG_FACILITY NFSDBG_FILE |
44 | ||
f0f37e2f | 45 | static const struct vm_operations_struct nfs_file_vm_ops; |
94387fb1 | 46 | |
1da177e4 LT |
47 | /* Hack for future NFS swap support */ |
48 | #ifndef IS_SWAPFILE | |
49 | # define IS_SWAPFILE(inode) (0) | |
50 | #endif | |
51 | ||
ce4ef7c0 | 52 | int nfs_check_flags(int flags) |
1da177e4 LT |
53 | { |
54 | if ((flags & (O_APPEND | O_DIRECT)) == (O_APPEND | O_DIRECT)) | |
55 | return -EINVAL; | |
56 | ||
57 | return 0; | |
58 | } | |
89d77c8f | 59 | EXPORT_SYMBOL_GPL(nfs_check_flags); |
1da177e4 LT |
60 | |
61 | /* | |
62 | * Open file | |
63 | */ | |
64 | static int | |
65 | nfs_file_open(struct inode *inode, struct file *filp) | |
66 | { | |
1da177e4 LT |
67 | int res; |
68 | ||
6de1472f | 69 | dprintk("NFS: open file(%pD2)\n", filp); |
cc0dd2d1 | 70 | |
c2459dc4 | 71 | nfs_inc_stats(inode, NFSIOS_VFSOPEN); |
1da177e4 LT |
72 | res = nfs_check_flags(filp->f_flags); |
73 | if (res) | |
74 | return res; | |
75 | ||
46cb650c | 76 | res = nfs_open(inode, filp); |
1da177e4 LT |
77 | return res; |
78 | } | |
79 | ||
ce4ef7c0 | 80 | int |
1da177e4 LT |
81 | nfs_file_release(struct inode *inode, struct file *filp) |
82 | { | |
6de1472f | 83 | dprintk("NFS: release(%pD2)\n", filp); |
6da24bc9 | 84 | |
91d5b470 | 85 | nfs_inc_stats(inode, NFSIOS_VFSRELEASE); |
46cb650c | 86 | return nfs_release(inode, filp); |
1da177e4 | 87 | } |
89d77c8f | 88 | EXPORT_SYMBOL_GPL(nfs_file_release); |
1da177e4 | 89 | |
980802e3 TM |
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 | ||
d7cf8dd0 TM |
106 | if (nfs_have_delegated_attributes(inode)) |
107 | goto out_noreval; | |
108 | ||
980802e3 TM |
109 | if (filp->f_flags & O_DIRECT) |
110 | goto force_reval; | |
d7cf8dd0 TM |
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; | |
980802e3 TM |
117 | force_reval: |
118 | return __nfs_revalidate_inode(server, inode); | |
119 | } | |
120 | ||
965c8e59 | 121 | loff_t nfs_file_llseek(struct file *filp, loff_t offset, int whence) |
980802e3 | 122 | { |
6de1472f AV |
123 | dprintk("NFS: llseek file(%pD2, %lld, %d)\n", |
124 | filp, offset, whence); | |
b84e06c5 | 125 | |
06222e49 | 126 | /* |
965c8e59 | 127 | * whence == SEEK_END || SEEK_DATA || SEEK_HOLE => we must revalidate |
06222e49 JB |
128 | * the cached file length |
129 | */ | |
965c8e59 | 130 | if (whence != SEEK_SET && whence != SEEK_CUR) { |
980802e3 | 131 | struct inode *inode = filp->f_mapping->host; |
d5e66348 | 132 | |
980802e3 TM |
133 | int retval = nfs_revalidate_file_size(inode, filp); |
134 | if (retval < 0) | |
135 | return (loff_t)retval; | |
79835a71 | 136 | } |
d5e66348 | 137 | |
965c8e59 | 138 | return generic_file_llseek(filp, offset, whence); |
980802e3 | 139 | } |
89d77c8f | 140 | EXPORT_SYMBOL_GPL(nfs_file_llseek); |
980802e3 | 141 | |
1da177e4 LT |
142 | /* |
143 | * Flush all dirty pages, and check for write errors. | |
1da177e4 | 144 | */ |
ce4ef7c0 | 145 | int |
75e1fcc0 | 146 | nfs_file_flush(struct file *file, fl_owner_t id) |
1da177e4 | 147 | { |
6de1472f | 148 | struct inode *inode = file_inode(file); |
1da177e4 | 149 | |
6de1472f | 150 | dprintk("NFS: flush(%pD2)\n", file); |
1da177e4 | 151 | |
c2459dc4 | 152 | nfs_inc_stats(inode, NFSIOS_VFSFLUSH); |
1da177e4 LT |
153 | if ((file->f_mode & FMODE_WRITE) == 0) |
154 | return 0; | |
7b159fc1 | 155 | |
14546c33 TM |
156 | /* |
157 | * If we're holding a write delegation, then just start the i/o | |
158 | * but don't wait for completion (or send a commit). | |
159 | */ | |
011e2a7f | 160 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_WRITE)) |
14546c33 TM |
161 | return filemap_fdatawrite(file->f_mapping); |
162 | ||
7fe5c398 | 163 | /* Flush writes to the server and return any errors */ |
af7fa165 | 164 | return vfs_fsync(file, 0); |
1da177e4 | 165 | } |
89d77c8f | 166 | EXPORT_SYMBOL_GPL(nfs_file_flush); |
1da177e4 | 167 | |
ce4ef7c0 | 168 | ssize_t |
3aa2d199 | 169 | nfs_file_read(struct kiocb *iocb, struct iov_iter *to) |
1da177e4 | 170 | { |
6de1472f | 171 | struct inode *inode = file_inode(iocb->ki_filp); |
1da177e4 LT |
172 | ssize_t result; |
173 | ||
1da177e4 | 174 | if (iocb->ki_filp->f_flags & O_DIRECT) |
e19a8a0a | 175 | return nfs_file_direct_read(iocb, to, iocb->ki_pos); |
1da177e4 | 176 | |
619d30b4 | 177 | dprintk("NFS: read(%pD2, %zu@%lu)\n", |
6de1472f | 178 | iocb->ki_filp, |
3aa2d199 | 179 | iov_iter_count(to), (unsigned long) iocb->ki_pos); |
1da177e4 | 180 | |
874f9463 | 181 | result = nfs_revalidate_mapping_protected(inode, iocb->ki_filp->f_mapping); |
4184dcf2 | 182 | if (!result) { |
3aa2d199 | 183 | result = generic_file_read_iter(iocb, to); |
4184dcf2 CL |
184 | if (result > 0) |
185 | nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, result); | |
186 | } | |
1da177e4 LT |
187 | return result; |
188 | } | |
89d77c8f | 189 | EXPORT_SYMBOL_GPL(nfs_file_read); |
1da177e4 | 190 | |
ce4ef7c0 | 191 | ssize_t |
f0930fff JA |
192 | nfs_file_splice_read(struct file *filp, loff_t *ppos, |
193 | struct pipe_inode_info *pipe, size_t count, | |
194 | unsigned int flags) | |
1da177e4 | 195 | { |
6de1472f | 196 | struct inode *inode = file_inode(filp); |
1da177e4 LT |
197 | ssize_t res; |
198 | ||
6de1472f AV |
199 | dprintk("NFS: splice_read(%pD2, %lu@%Lu)\n", |
200 | filp, (unsigned long) count, (unsigned long long) *ppos); | |
1da177e4 | 201 | |
874f9463 | 202 | res = nfs_revalidate_mapping_protected(inode, filp->f_mapping); |
aa2f1ef1 | 203 | if (!res) { |
f0930fff | 204 | res = generic_file_splice_read(filp, ppos, pipe, count, flags); |
aa2f1ef1 CL |
205 | if (res > 0) |
206 | nfs_add_stats(inode, NFSIOS_NORMALREADBYTES, res); | |
207 | } | |
1da177e4 LT |
208 | return res; |
209 | } | |
89d77c8f | 210 | EXPORT_SYMBOL_GPL(nfs_file_splice_read); |
1da177e4 | 211 | |
ce4ef7c0 | 212 | int |
1da177e4 LT |
213 | nfs_file_mmap(struct file * file, struct vm_area_struct * vma) |
214 | { | |
6de1472f | 215 | struct inode *inode = file_inode(file); |
1da177e4 LT |
216 | int status; |
217 | ||
6de1472f | 218 | dprintk("NFS: mmap(%pD2)\n", file); |
1da177e4 | 219 | |
e1ebfd33 TM |
220 | /* Note: generic_file_mmap() returns ENOSYS on nommu systems |
221 | * so we call that before revalidating the mapping | |
222 | */ | |
223 | status = generic_file_mmap(file, vma); | |
94387fb1 TM |
224 | if (!status) { |
225 | vma->vm_ops = &nfs_file_vm_ops; | |
e1ebfd33 | 226 | status = nfs_revalidate_mapping(inode, file->f_mapping); |
94387fb1 | 227 | } |
1da177e4 LT |
228 | return status; |
229 | } | |
89d77c8f | 230 | EXPORT_SYMBOL_GPL(nfs_file_mmap); |
1da177e4 LT |
231 | |
232 | /* | |
233 | * Flush any dirty pages for this process, and check for write errors. | |
234 | * The return status from this call provides a reliable indication of | |
235 | * whether any write errors occurred for this process. | |
af7fa165 TM |
236 | * |
237 | * Notice that it clears the NFS_CONTEXT_ERROR_WRITE before synching to | |
238 | * disk, but it retrieves and clears ctx->error after synching, despite | |
239 | * the two being set at the same time in nfs_context_set_write_error(). | |
240 | * This is because the former is used to notify the _next_ call to | |
25985edc | 241 | * nfs_file_write() that a write error occurred, and hence cause it to |
af7fa165 | 242 | * fall back to doing a synchronous write. |
1da177e4 | 243 | */ |
ce4ef7c0 | 244 | int |
a5c58892 | 245 | nfs_file_fsync_commit(struct file *file, loff_t start, loff_t end, int datasync) |
1da177e4 | 246 | { |
cd3758e3 | 247 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
6de1472f | 248 | struct inode *inode = file_inode(file); |
05990d1b | 249 | int have_error, do_resend, status; |
af7fa165 TM |
250 | int ret = 0; |
251 | ||
6de1472f | 252 | dprintk("NFS: fsync file(%pD2) datasync %d\n", file, datasync); |
1da177e4 | 253 | |
91d5b470 | 254 | nfs_inc_stats(inode, NFSIOS_VFSFSYNC); |
05990d1b | 255 | do_resend = test_and_clear_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags); |
af7fa165 TM |
256 | have_error = test_and_clear_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); |
257 | status = nfs_commit_inode(inode, FLUSH_SYNC); | |
258 | have_error |= test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags); | |
05990d1b | 259 | if (have_error) { |
af7fa165 | 260 | ret = xchg(&ctx->error, 0); |
05990d1b TM |
261 | if (ret) |
262 | goto out; | |
263 | } | |
264 | if (status < 0) { | |
af7fa165 | 265 | ret = status; |
05990d1b TM |
266 | goto out; |
267 | } | |
268 | do_resend |= test_bit(NFS_CONTEXT_RESEND_WRITES, &ctx->flags); | |
269 | if (do_resend) | |
270 | ret = -EAGAIN; | |
271 | out: | |
a5c58892 BS |
272 | return ret; |
273 | } | |
89d77c8f | 274 | EXPORT_SYMBOL_GPL(nfs_file_fsync_commit); |
a5c58892 BS |
275 | |
276 | static int | |
277 | nfs_file_fsync(struct file *file, loff_t start, loff_t end, int datasync) | |
278 | { | |
279 | int ret; | |
496ad9aa | 280 | struct inode *inode = file_inode(file); |
a5c58892 | 281 | |
f4ce1299 TM |
282 | trace_nfs_fsync_enter(inode); |
283 | ||
05990d1b TM |
284 | do { |
285 | ret = filemap_write_and_wait_range(inode->i_mapping, start, end); | |
286 | if (ret != 0) | |
287 | break; | |
288 | mutex_lock(&inode->i_mutex); | |
289 | ret = nfs_file_fsync_commit(file, start, end, datasync); | |
290 | mutex_unlock(&inode->i_mutex); | |
dcfc4f25 TM |
291 | /* |
292 | * If nfs_file_fsync_commit detected a server reboot, then | |
293 | * resend all dirty pages that might have been covered by | |
294 | * the NFS_CONTEXT_RESEND_WRITES flag | |
295 | */ | |
296 | start = 0; | |
297 | end = LLONG_MAX; | |
05990d1b TM |
298 | } while (ret == -EAGAIN); |
299 | ||
f4ce1299 | 300 | trace_nfs_fsync_exit(inode, ret); |
af7fa165 | 301 | return ret; |
1da177e4 LT |
302 | } |
303 | ||
38c73044 PS |
304 | /* |
305 | * Decide whether a read/modify/write cycle may be more efficient | |
306 | * then a modify/write/read cycle when writing to a page in the | |
307 | * page cache. | |
308 | * | |
309 | * The modify/write/read cycle may occur if a page is read before | |
310 | * being completely filled by the writer. In this situation, the | |
311 | * page must be completely written to stable storage on the server | |
312 | * before it can be refilled by reading in the page from the server. | |
313 | * This can lead to expensive, small, FILE_SYNC mode writes being | |
314 | * done. | |
315 | * | |
316 | * It may be more efficient to read the page first if the file is | |
317 | * open for reading in addition to writing, the page is not marked | |
318 | * as Uptodate, it is not dirty or waiting to be committed, | |
319 | * indicating that it was previously allocated and then modified, | |
320 | * that there were valid bytes of data in that range of the file, | |
321 | * and that the new data won't completely replace the old data in | |
322 | * that range of the file. | |
323 | */ | |
324 | static int nfs_want_read_modify_write(struct file *file, struct page *page, | |
325 | loff_t pos, unsigned len) | |
326 | { | |
327 | unsigned int pglen = nfs_page_length(page); | |
328 | unsigned int offset = pos & (PAGE_CACHE_SIZE - 1); | |
329 | unsigned int end = offset + len; | |
330 | ||
612aa983 CH |
331 | if (pnfs_ld_read_whole_page(file->f_mapping->host)) { |
332 | if (!PageUptodate(page)) | |
333 | return 1; | |
334 | return 0; | |
335 | } | |
336 | ||
38c73044 PS |
337 | if ((file->f_mode & FMODE_READ) && /* open for read? */ |
338 | !PageUptodate(page) && /* Uptodate? */ | |
339 | !PagePrivate(page) && /* i/o request already? */ | |
340 | pglen && /* valid bytes of file? */ | |
341 | (end < pglen || offset)) /* replace all valid bytes? */ | |
342 | return 1; | |
343 | return 0; | |
344 | } | |
345 | ||
1da177e4 | 346 | /* |
4899f9c8 NP |
347 | * This does the "real" work of the write. We must allocate and lock the |
348 | * page to be sent back to the generic routine, which then copies the | |
349 | * data from user space. | |
1da177e4 LT |
350 | * |
351 | * If the writer ends up delaying the write, the writer needs to | |
352 | * increment the page use counts until he is done with the page. | |
353 | */ | |
4899f9c8 NP |
354 | static int nfs_write_begin(struct file *file, struct address_space *mapping, |
355 | loff_t pos, unsigned len, unsigned flags, | |
356 | struct page **pagep, void **fsdata) | |
1da177e4 | 357 | { |
4899f9c8 | 358 | int ret; |
38c73044 | 359 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
4899f9c8 | 360 | struct page *page; |
38c73044 | 361 | int once_thru = 0; |
4899f9c8 | 362 | |
1e8968c5 | 363 | dfprintk(PAGECACHE, "NFS: write_begin(%pD2(%lu), %u@%lld)\n", |
6de1472f | 364 | file, mapping->host->i_ino, len, (long long) pos); |
b7eaefaa | 365 | |
38c73044 | 366 | start: |
72cb77f4 TM |
367 | /* |
368 | * Prevent starvation issues if someone is doing a consistency | |
369 | * sync-to-disk | |
370 | */ | |
74316201 N |
371 | ret = wait_on_bit_action(&NFS_I(mapping->host)->flags, NFS_INO_FLUSHING, |
372 | nfs_wait_bit_killable, TASK_KILLABLE); | |
72cb77f4 TM |
373 | if (ret) |
374 | return ret; | |
aa5accea TM |
375 | /* |
376 | * Wait for O_DIRECT to complete | |
377 | */ | |
378 | nfs_inode_dio_wait(mapping->host); | |
72cb77f4 | 379 | |
54566b2c | 380 | page = grab_cache_page_write_begin(mapping, index, flags); |
4899f9c8 NP |
381 | if (!page) |
382 | return -ENOMEM; | |
383 | *pagep = page; | |
384 | ||
385 | ret = nfs_flush_incompatible(file, page); | |
386 | if (ret) { | |
387 | unlock_page(page); | |
388 | page_cache_release(page); | |
38c73044 PS |
389 | } else if (!once_thru && |
390 | nfs_want_read_modify_write(file, page, pos, len)) { | |
391 | once_thru = 1; | |
392 | ret = nfs_readpage(file, page); | |
393 | page_cache_release(page); | |
394 | if (!ret) | |
395 | goto start; | |
4899f9c8 NP |
396 | } |
397 | return ret; | |
1da177e4 LT |
398 | } |
399 | ||
4899f9c8 NP |
400 | static int nfs_write_end(struct file *file, struct address_space *mapping, |
401 | loff_t pos, unsigned len, unsigned copied, | |
402 | struct page *page, void *fsdata) | |
1da177e4 | 403 | { |
4899f9c8 | 404 | unsigned offset = pos & (PAGE_CACHE_SIZE - 1); |
dc24826b | 405 | struct nfs_open_context *ctx = nfs_file_open_context(file); |
4899f9c8 | 406 | int status; |
1da177e4 | 407 | |
1e8968c5 | 408 | dfprintk(PAGECACHE, "NFS: write_end(%pD2(%lu), %u@%lld)\n", |
6de1472f | 409 | file, mapping->host->i_ino, len, (long long) pos); |
b7eaefaa | 410 | |
efc91ed0 TM |
411 | /* |
412 | * Zero any uninitialised parts of the page, and then mark the page | |
413 | * as up to date if it turns out that we're extending the file. | |
414 | */ | |
415 | if (!PageUptodate(page)) { | |
416 | unsigned pglen = nfs_page_length(page); | |
417 | unsigned end = offset + len; | |
418 | ||
419 | if (pglen == 0) { | |
420 | zero_user_segments(page, 0, offset, | |
421 | end, PAGE_CACHE_SIZE); | |
422 | SetPageUptodate(page); | |
423 | } else if (end >= pglen) { | |
424 | zero_user_segment(page, end, PAGE_CACHE_SIZE); | |
425 | if (offset == 0) | |
426 | SetPageUptodate(page); | |
427 | } else | |
428 | zero_user_segment(page, pglen, PAGE_CACHE_SIZE); | |
429 | } | |
430 | ||
4899f9c8 | 431 | status = nfs_updatepage(file, page, offset, copied); |
4899f9c8 NP |
432 | |
433 | unlock_page(page); | |
434 | page_cache_release(page); | |
435 | ||
3d509e54 CL |
436 | if (status < 0) |
437 | return status; | |
2701d086 | 438 | NFS_I(mapping->host)->write_io += copied; |
dc24826b AA |
439 | |
440 | if (nfs_ctx_key_to_expire(ctx)) { | |
441 | status = nfs_wb_all(mapping->host); | |
442 | if (status < 0) | |
443 | return status; | |
444 | } | |
445 | ||
3d509e54 | 446 | return copied; |
1da177e4 LT |
447 | } |
448 | ||
6b9b3514 DH |
449 | /* |
450 | * Partially or wholly invalidate a page | |
451 | * - Release the private state associated with a page if undergoing complete | |
452 | * page invalidation | |
545db45f | 453 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
454 | * - Caller holds page lock |
455 | */ | |
d47992f8 LC |
456 | static void nfs_invalidate_page(struct page *page, unsigned int offset, |
457 | unsigned int length) | |
cd52ed35 | 458 | { |
d47992f8 LC |
459 | dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %u, %u)\n", |
460 | page, offset, length); | |
b7eaefaa | 461 | |
d47992f8 | 462 | if (offset != 0 || length < PAGE_CACHE_SIZE) |
1c75950b | 463 | return; |
d2ccddf0 | 464 | /* Cancel any unstarted writes on this page */ |
d56b4ddf | 465 | nfs_wb_page_cancel(page_file_mapping(page)->host, page); |
545db45f DH |
466 | |
467 | nfs_fscache_invalidate_page(page, page->mapping->host); | |
cd52ed35 TM |
468 | } |
469 | ||
6b9b3514 DH |
470 | /* |
471 | * Attempt to release the private state associated with a page | |
545db45f | 472 | * - Called if either PG_private or PG_fscache is set on the page |
6b9b3514 DH |
473 | * - Caller holds page lock |
474 | * - Return true (may release page) or false (may not) | |
475 | */ | |
cd52ed35 TM |
476 | static int nfs_release_page(struct page *page, gfp_t gfp) |
477 | { | |
b608b283 TM |
478 | struct address_space *mapping = page->mapping; |
479 | ||
b7eaefaa CL |
480 | dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page); |
481 | ||
95905446 | 482 | /* Always try to initiate a 'commit' if relevant, but only |
1aff5256 N |
483 | * wait for it if __GFP_WAIT is set. Even then, only wait 1 |
484 | * second and only if the 'bdi' is not congested. | |
95905446 | 485 | * Waiting indefinitely can cause deadlocks when the NFS |
1aff5256 N |
486 | * server is on this machine, when a new TCP connection is |
487 | * needed and in other rare cases. There is no particular | |
488 | * need to wait extensively here. A short wait has the | |
489 | * benefit that someone else can worry about the freezer. | |
5cf02d09 | 490 | */ |
95905446 N |
491 | if (mapping) { |
492 | struct nfs_server *nfss = NFS_SERVER(mapping->host); | |
493 | nfs_commit_inode(mapping->host, 0); | |
494 | if ((gfp & __GFP_WAIT) && | |
353db796 | 495 | !bdi_write_congested(&nfss->backing_dev_info)) { |
95905446 N |
496 | wait_on_page_bit_killable_timeout(page, PG_private, |
497 | HZ); | |
353db796 N |
498 | if (PagePrivate(page)) |
499 | set_bdi_congested(&nfss->backing_dev_info, | |
500 | BLK_RW_ASYNC); | |
95905446 | 501 | } |
b608b283 | 502 | } |
e3db7691 | 503 | /* If PagePrivate() is set, then the page is not freeable */ |
545db45f DH |
504 | if (PagePrivate(page)) |
505 | return 0; | |
506 | return nfs_fscache_release_page(page, gfp); | |
e3db7691 TM |
507 | } |
508 | ||
f919b196 MG |
509 | static void nfs_check_dirty_writeback(struct page *page, |
510 | bool *dirty, bool *writeback) | |
511 | { | |
512 | struct nfs_inode *nfsi; | |
513 | struct address_space *mapping = page_file_mapping(page); | |
514 | ||
515 | if (!mapping || PageSwapCache(page)) | |
516 | return; | |
517 | ||
518 | /* | |
519 | * Check if an unstable page is currently being committed and | |
520 | * if so, have the VM treat it as if the page is under writeback | |
521 | * so it will not block due to pages that will shortly be freeable. | |
522 | */ | |
523 | nfsi = NFS_I(mapping->host); | |
524 | if (test_bit(NFS_INO_COMMIT, &nfsi->flags)) { | |
525 | *writeback = true; | |
526 | return; | |
527 | } | |
528 | ||
529 | /* | |
530 | * If PagePrivate() is set, then the page is not freeable and as the | |
531 | * inode is not being committed, it's not going to be cleaned in the | |
532 | * near future so treat it as dirty | |
533 | */ | |
534 | if (PagePrivate(page)) | |
535 | *dirty = true; | |
536 | } | |
537 | ||
6b9b3514 DH |
538 | /* |
539 | * Attempt to clear the private state associated with a page when an error | |
540 | * occurs that requires the cached contents of an inode to be written back or | |
541 | * destroyed | |
545db45f | 542 | * - Called if either PG_private or fscache is set on the page |
6b9b3514 DH |
543 | * - Caller holds page lock |
544 | * - Return 0 if successful, -error otherwise | |
545 | */ | |
e3db7691 TM |
546 | static int nfs_launder_page(struct page *page) |
547 | { | |
d56b4ddf | 548 | struct inode *inode = page_file_mapping(page)->host; |
545db45f | 549 | struct nfs_inode *nfsi = NFS_I(inode); |
b7eaefaa CL |
550 | |
551 | dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n", | |
552 | inode->i_ino, (long long)page_offset(page)); | |
553 | ||
545db45f | 554 | nfs_fscache_wait_on_page_write(nfsi, page); |
b7eaefaa | 555 | return nfs_wb_page(inode, page); |
cd52ed35 TM |
556 | } |
557 | ||
a564b8f0 MG |
558 | #ifdef CONFIG_NFS_SWAP |
559 | static int nfs_swap_activate(struct swap_info_struct *sis, struct file *file, | |
560 | sector_t *span) | |
561 | { | |
dad2b015 JL |
562 | int ret; |
563 | struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host); | |
564 | ||
a564b8f0 | 565 | *span = sis->pages; |
dad2b015 JL |
566 | |
567 | rcu_read_lock(); | |
568 | ret = xs_swapper(rcu_dereference(clnt->cl_xprt), 1); | |
569 | rcu_read_unlock(); | |
570 | ||
571 | return ret; | |
a564b8f0 MG |
572 | } |
573 | ||
574 | static void nfs_swap_deactivate(struct file *file) | |
575 | { | |
dad2b015 JL |
576 | struct rpc_clnt *clnt = NFS_CLIENT(file->f_mapping->host); |
577 | ||
578 | rcu_read_lock(); | |
579 | xs_swapper(rcu_dereference(clnt->cl_xprt), 0); | |
580 | rcu_read_unlock(); | |
a564b8f0 MG |
581 | } |
582 | #endif | |
583 | ||
f5e54d6e | 584 | const struct address_space_operations nfs_file_aops = { |
1da177e4 LT |
585 | .readpage = nfs_readpage, |
586 | .readpages = nfs_readpages, | |
9cccef95 | 587 | .set_page_dirty = __set_page_dirty_nobuffers, |
1da177e4 LT |
588 | .writepage = nfs_writepage, |
589 | .writepages = nfs_writepages, | |
4899f9c8 NP |
590 | .write_begin = nfs_write_begin, |
591 | .write_end = nfs_write_end, | |
cd52ed35 TM |
592 | .invalidatepage = nfs_invalidate_page, |
593 | .releasepage = nfs_release_page, | |
1da177e4 | 594 | .direct_IO = nfs_direct_IO, |
074cc1de | 595 | .migratepage = nfs_migrate_page, |
e3db7691 | 596 | .launder_page = nfs_launder_page, |
f919b196 | 597 | .is_dirty_writeback = nfs_check_dirty_writeback, |
f590f333 | 598 | .error_remove_page = generic_error_remove_page, |
a564b8f0 MG |
599 | #ifdef CONFIG_NFS_SWAP |
600 | .swap_activate = nfs_swap_activate, | |
601 | .swap_deactivate = nfs_swap_deactivate, | |
602 | #endif | |
1da177e4 LT |
603 | }; |
604 | ||
6b9b3514 DH |
605 | /* |
606 | * Notification that a PTE pointing to an NFS page is about to be made | |
607 | * writable, implying that someone is about to modify the page through a | |
608 | * shared-writable mapping | |
609 | */ | |
c2ec175c | 610 | static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
94387fb1 | 611 | { |
c2ec175c | 612 | struct page *page = vmf->page; |
94387fb1 | 613 | struct file *filp = vma->vm_file; |
6de1472f | 614 | struct inode *inode = file_inode(filp); |
94387fb1 | 615 | unsigned pagelen; |
bc4866b6 | 616 | int ret = VM_FAULT_NOPAGE; |
4899f9c8 | 617 | struct address_space *mapping; |
94387fb1 | 618 | |
1e8968c5 | 619 | dfprintk(PAGECACHE, "NFS: vm_page_mkwrite(%pD2(%lu), offset %lld)\n", |
6de1472f | 620 | filp, filp->f_mapping->host->i_ino, |
b7eaefaa CL |
621 | (long long)page_offset(page)); |
622 | ||
545db45f | 623 | /* make sure the cache has finished storing the page */ |
6de1472f | 624 | nfs_fscache_wait_on_page_write(NFS_I(inode), page); |
545db45f | 625 | |
ef070dcb TM |
626 | wait_on_bit_action(&NFS_I(inode)->flags, NFS_INO_INVALIDATING, |
627 | nfs_wait_bit_killable, TASK_KILLABLE); | |
628 | ||
94387fb1 | 629 | lock_page(page); |
d56b4ddf | 630 | mapping = page_file_mapping(page); |
6de1472f | 631 | if (mapping != inode->i_mapping) |
8b1f9ee5 TM |
632 | goto out_unlock; |
633 | ||
2aeb98f4 TM |
634 | wait_on_page_writeback(page); |
635 | ||
94387fb1 | 636 | pagelen = nfs_page_length(page); |
8b1f9ee5 TM |
637 | if (pagelen == 0) |
638 | goto out_unlock; | |
4899f9c8 | 639 | |
bc4866b6 TM |
640 | ret = VM_FAULT_LOCKED; |
641 | if (nfs_flush_incompatible(filp, page) == 0 && | |
642 | nfs_updatepage(filp, page, 0, pagelen) == 0) | |
643 | goto out; | |
8b1f9ee5 | 644 | |
bc4866b6 | 645 | ret = VM_FAULT_SIGBUS; |
8b1f9ee5 TM |
646 | out_unlock: |
647 | unlock_page(page); | |
bc4866b6 TM |
648 | out: |
649 | return ret; | |
94387fb1 TM |
650 | } |
651 | ||
f0f37e2f | 652 | static const struct vm_operations_struct nfs_file_vm_ops = { |
94387fb1 | 653 | .fault = filemap_fault, |
f1820361 | 654 | .map_pages = filemap_map_pages, |
94387fb1 TM |
655 | .page_mkwrite = nfs_vm_page_mkwrite, |
656 | }; | |
657 | ||
7b159fc1 TM |
658 | static int nfs_need_sync_write(struct file *filp, struct inode *inode) |
659 | { | |
660 | struct nfs_open_context *ctx; | |
661 | ||
6b2f3d1f | 662 | if (IS_SYNC(inode) || (filp->f_flags & O_DSYNC)) |
7b159fc1 | 663 | return 1; |
cd3758e3 | 664 | ctx = nfs_file_open_context(filp); |
dc24826b AA |
665 | if (test_bit(NFS_CONTEXT_ERROR_WRITE, &ctx->flags) || |
666 | nfs_ctx_key_to_expire(ctx)) | |
7b159fc1 TM |
667 | return 1; |
668 | return 0; | |
669 | } | |
670 | ||
edaf4369 | 671 | ssize_t nfs_file_write(struct kiocb *iocb, struct iov_iter *from) |
1da177e4 | 672 | { |
6de1472f AV |
673 | struct file *file = iocb->ki_filp; |
674 | struct inode *inode = file_inode(file); | |
7e381172 | 675 | unsigned long written = 0; |
1da177e4 | 676 | ssize_t result; |
edaf4369 AV |
677 | size_t count = iov_iter_count(from); |
678 | loff_t pos = iocb->ki_pos; | |
1da177e4 | 679 | |
6de1472f | 680 | result = nfs_key_timeout_notify(file, inode); |
dc24826b AA |
681 | if (result) |
682 | return result; | |
683 | ||
6de1472f | 684 | if (file->f_flags & O_DIRECT) |
e19a8a0a | 685 | return nfs_file_direct_write(iocb, from, pos); |
1da177e4 | 686 | |
619d30b4 AV |
687 | dprintk("NFS: write(%pD2, %zu@%Ld)\n", |
688 | file, count, (long long) pos); | |
1da177e4 LT |
689 | |
690 | result = -EBUSY; | |
691 | if (IS_SWAPFILE(inode)) | |
692 | goto out_swapfile; | |
7d52e862 TM |
693 | /* |
694 | * O_APPEND implies that we must revalidate the file length. | |
695 | */ | |
6de1472f AV |
696 | if (file->f_flags & O_APPEND) { |
697 | result = nfs_revalidate_file_size(inode, file); | |
7d52e862 TM |
698 | if (result) |
699 | goto out; | |
fe51beec | 700 | } |
1da177e4 LT |
701 | |
702 | result = count; | |
703 | if (!count) | |
704 | goto out; | |
705 | ||
edaf4369 | 706 | result = generic_file_write_iter(iocb, from); |
7e381172 CL |
707 | if (result > 0) |
708 | written = result; | |
709 | ||
6b2f3d1f | 710 | /* Return error values for O_DSYNC and IS_SYNC() */ |
6de1472f AV |
711 | if (result >= 0 && nfs_need_sync_write(file, inode)) { |
712 | int err = vfs_fsync(file, 0); | |
200baa21 TM |
713 | if (err < 0) |
714 | result = err; | |
715 | } | |
7e381172 CL |
716 | if (result > 0) |
717 | nfs_add_stats(inode, NFSIOS_NORMALWRITTENBYTES, written); | |
1da177e4 LT |
718 | out: |
719 | return result; | |
720 | ||
721 | out_swapfile: | |
722 | printk(KERN_INFO "NFS: attempt to write to active swap file!\n"); | |
723 | goto out; | |
724 | } | |
89d77c8f | 725 | EXPORT_SYMBOL_GPL(nfs_file_write); |
1da177e4 | 726 | |
5eebde23 SJ |
727 | static int |
728 | do_getlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
729 | { |
730 | struct inode *inode = filp->f_mapping->host; | |
731 | int status = 0; | |
21ac19d4 | 732 | unsigned int saved_type = fl->fl_type; |
1da177e4 | 733 | |
039c4d7a | 734 | /* Try local locking first */ |
6d34ac19 BF |
735 | posix_test_lock(filp, fl); |
736 | if (fl->fl_type != F_UNLCK) { | |
737 | /* found a conflict */ | |
039c4d7a | 738 | goto out; |
1da177e4 | 739 | } |
21ac19d4 | 740 | fl->fl_type = saved_type; |
039c4d7a | 741 | |
011e2a7f | 742 | if (NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) |
039c4d7a TM |
743 | goto out_noconflict; |
744 | ||
5eebde23 | 745 | if (is_local) |
039c4d7a TM |
746 | goto out_noconflict; |
747 | ||
748 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); | |
749 | out: | |
1da177e4 | 750 | return status; |
039c4d7a TM |
751 | out_noconflict: |
752 | fl->fl_type = F_UNLCK; | |
753 | goto out; | |
1da177e4 LT |
754 | } |
755 | ||
756 | static int do_vfs_lock(struct file *file, struct file_lock *fl) | |
757 | { | |
758 | int res = 0; | |
759 | switch (fl->fl_flags & (FL_POSIX|FL_FLOCK)) { | |
760 | case FL_POSIX: | |
761 | res = posix_lock_file_wait(file, fl); | |
762 | break; | |
763 | case FL_FLOCK: | |
764 | res = flock_lock_file_wait(file, fl); | |
765 | break; | |
766 | default: | |
767 | BUG(); | |
768 | } | |
1da177e4 LT |
769 | return res; |
770 | } | |
771 | ||
5eebde23 SJ |
772 | static int |
773 | do_unlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
774 | { |
775 | struct inode *inode = filp->f_mapping->host; | |
7a8203d8 | 776 | struct nfs_lock_context *l_ctx; |
1da177e4 LT |
777 | int status; |
778 | ||
1da177e4 LT |
779 | /* |
780 | * Flush all pending writes before doing anything | |
781 | * with locks.. | |
782 | */ | |
29884df0 | 783 | nfs_sync_mapping(filp->f_mapping); |
1da177e4 | 784 | |
7a8203d8 TM |
785 | l_ctx = nfs_get_lock_context(nfs_file_open_context(filp)); |
786 | if (!IS_ERR(l_ctx)) { | |
787 | status = nfs_iocounter_wait(&l_ctx->io_count); | |
788 | nfs_put_lock_context(l_ctx); | |
789 | if (status < 0) | |
790 | return status; | |
791 | } | |
792 | ||
1da177e4 LT |
793 | /* NOTE: special case |
794 | * If we're signalled while cleaning up locks on process exit, we | |
795 | * still need to complete the unlock. | |
796 | */ | |
5eebde23 SJ |
797 | /* |
798 | * Use local locking if mounted with "-onolock" or with appropriate | |
799 | * "-olocal_lock=" | |
800 | */ | |
801 | if (!is_local) | |
1da177e4 LT |
802 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
803 | else | |
804 | status = do_vfs_lock(filp, fl); | |
1da177e4 LT |
805 | return status; |
806 | } | |
807 | ||
6b96724e RL |
808 | static int |
809 | is_time_granular(struct timespec *ts) { | |
810 | return ((ts->tv_sec == 0) && (ts->tv_nsec <= 1000)); | |
811 | } | |
812 | ||
5eebde23 SJ |
813 | static int |
814 | do_setlk(struct file *filp, int cmd, struct file_lock *fl, int is_local) | |
1da177e4 LT |
815 | { |
816 | struct inode *inode = filp->f_mapping->host; | |
1da177e4 LT |
817 | int status; |
818 | ||
1da177e4 LT |
819 | /* |
820 | * Flush all pending writes before doing anything | |
821 | * with locks.. | |
822 | */ | |
29884df0 TM |
823 | status = nfs_sync_mapping(filp->f_mapping); |
824 | if (status != 0) | |
1da177e4 LT |
825 | goto out; |
826 | ||
5eebde23 SJ |
827 | /* |
828 | * Use local locking if mounted with "-onolock" or with appropriate | |
829 | * "-olocal_lock=" | |
830 | */ | |
831 | if (!is_local) | |
1da177e4 | 832 | status = NFS_PROTO(inode)->lock(filp, cmd, fl); |
c4d7c402 | 833 | else |
1da177e4 | 834 | status = do_vfs_lock(filp, fl); |
1da177e4 LT |
835 | if (status < 0) |
836 | goto out; | |
6b96724e | 837 | |
1da177e4 | 838 | /* |
6b96724e RL |
839 | * Revalidate the cache if the server has time stamps granular |
840 | * enough to detect subsecond changes. Otherwise, clear the | |
841 | * cache to prevent missing any changes. | |
842 | * | |
1da177e4 LT |
843 | * This makes locking act as a cache coherency point. |
844 | */ | |
29884df0 | 845 | nfs_sync_mapping(filp->f_mapping); |
011e2a7f | 846 | if (!NFS_PROTO(inode)->have_delegation(inode, FMODE_READ)) { |
6b96724e RL |
847 | if (is_time_granular(&NFS_SERVER(inode)->time_delta)) |
848 | __nfs_revalidate_inode(NFS_SERVER(inode), inode); | |
849 | else | |
850 | nfs_zap_caches(inode); | |
851 | } | |
1da177e4 | 852 | out: |
1da177e4 LT |
853 | return status; |
854 | } | |
855 | ||
856 | /* | |
857 | * Lock a (portion of) a file | |
858 | */ | |
ce4ef7c0 | 859 | int nfs_lock(struct file *filp, int cmd, struct file_lock *fl) |
1da177e4 | 860 | { |
6da24bc9 | 861 | struct inode *inode = filp->f_mapping->host; |
2116271a | 862 | int ret = -ENOLCK; |
5eebde23 | 863 | int is_local = 0; |
1da177e4 | 864 | |
6de1472f AV |
865 | dprintk("NFS: lock(%pD2, t=%x, fl=%x, r=%lld:%lld)\n", |
866 | filp, fl->fl_type, fl->fl_flags, | |
1da177e4 | 867 | (long long)fl->fl_start, (long long)fl->fl_end); |
6da24bc9 | 868 | |
91d5b470 | 869 | nfs_inc_stats(inode, NFSIOS_VFSLOCK); |
1da177e4 LT |
870 | |
871 | /* No mandatory locks over NFS */ | |
dfad9441 | 872 | if (__mandatory_lock(inode) && fl->fl_type != F_UNLCK) |
2116271a TM |
873 | goto out_err; |
874 | ||
5eebde23 SJ |
875 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FCNTL) |
876 | is_local = 1; | |
877 | ||
2116271a TM |
878 | if (NFS_PROTO(inode)->lock_check_bounds != NULL) { |
879 | ret = NFS_PROTO(inode)->lock_check_bounds(fl); | |
880 | if (ret < 0) | |
881 | goto out_err; | |
882 | } | |
1da177e4 LT |
883 | |
884 | if (IS_GETLK(cmd)) | |
5eebde23 | 885 | ret = do_getlk(filp, cmd, fl, is_local); |
2116271a | 886 | else if (fl->fl_type == F_UNLCK) |
5eebde23 | 887 | ret = do_unlk(filp, cmd, fl, is_local); |
2116271a | 888 | else |
5eebde23 | 889 | ret = do_setlk(filp, cmd, fl, is_local); |
2116271a TM |
890 | out_err: |
891 | return ret; | |
1da177e4 | 892 | } |
89d77c8f | 893 | EXPORT_SYMBOL_GPL(nfs_lock); |
1da177e4 LT |
894 | |
895 | /* | |
896 | * Lock a (portion of) a file | |
897 | */ | |
ce4ef7c0 | 898 | int nfs_flock(struct file *filp, int cmd, struct file_lock *fl) |
1da177e4 | 899 | { |
5eebde23 SJ |
900 | struct inode *inode = filp->f_mapping->host; |
901 | int is_local = 0; | |
902 | ||
6de1472f AV |
903 | dprintk("NFS: flock(%pD2, t=%x, fl=%x)\n", |
904 | filp, fl->fl_type, fl->fl_flags); | |
1da177e4 | 905 | |
1da177e4 LT |
906 | if (!(fl->fl_flags & FL_FLOCK)) |
907 | return -ENOLCK; | |
908 | ||
ad0fcd4e JL |
909 | /* |
910 | * The NFSv4 protocol doesn't support LOCK_MAND, which is not part of | |
911 | * any standard. In principle we might be able to support LOCK_MAND | |
912 | * on NFSv2/3 since NLMv3/4 support DOS share modes, but for now the | |
913 | * NFS code is not set up for it. | |
914 | */ | |
915 | if (fl->fl_type & LOCK_MAND) | |
916 | return -EINVAL; | |
917 | ||
5eebde23 SJ |
918 | if (NFS_SERVER(inode)->flags & NFS_MOUNT_LOCAL_FLOCK) |
919 | is_local = 1; | |
920 | ||
1da177e4 | 921 | /* We're simulating flock() locks using posix locks on the server */ |
1da177e4 | 922 | if (fl->fl_type == F_UNLCK) |
5eebde23 SJ |
923 | return do_unlk(filp, cmd, fl, is_local); |
924 | return do_setlk(filp, cmd, fl, is_local); | |
1da177e4 | 925 | } |
89d77c8f | 926 | EXPORT_SYMBOL_GPL(nfs_flock); |
370f6599 | 927 | |
0486958f JL |
928 | const struct file_operations nfs_file_operations = { |
929 | .llseek = nfs_file_llseek, | |
3aa2d199 | 930 | .read = new_sync_read, |
edaf4369 | 931 | .write = new_sync_write, |
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); |