]>
Commit | Line | Data |
---|---|---|
b6aeaded MS |
1 | /* |
2 | FUSE: Filesystem in Userspace | |
1729a16c | 3 | Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu> |
b6aeaded MS |
4 | |
5 | This program can be distributed under the terms of the GNU GPL. | |
6 | See the file COPYING. | |
7 | */ | |
8 | ||
9 | #include "fuse_i.h" | |
10 | ||
11 | #include <linux/pagemap.h> | |
12 | #include <linux/slab.h> | |
13 | #include <linux/kernel.h> | |
e8edc6e0 | 14 | #include <linux/sched.h> |
7a36094d | 15 | #include <linux/sched/signal.h> |
08cbf542 | 16 | #include <linux/module.h> |
d9d318d3 | 17 | #include <linux/compat.h> |
478e0841 | 18 | #include <linux/swap.h> |
3634a632 | 19 | #include <linux/falloc.h> |
e2e40f2c | 20 | #include <linux/uio.h> |
b6aeaded | 21 | |
7213394c MS |
22 | static struct page **fuse_pages_alloc(unsigned int npages, gfp_t flags, |
23 | struct fuse_page_desc **desc) | |
4c4f03f7 MS |
24 | { |
25 | struct page **pages; | |
26 | ||
27 | pages = kzalloc(npages * (sizeof(struct page *) + | |
28 | sizeof(struct fuse_page_desc)), flags); | |
29 | *desc = (void *) (pages + npages); | |
30 | ||
31 | return pages; | |
32 | } | |
33 | ||
91fe96b4 MS |
34 | static int fuse_send_open(struct fuse_conn *fc, u64 nodeid, struct file *file, |
35 | int opcode, struct fuse_open_out *outargp) | |
b6aeaded | 36 | { |
b6aeaded | 37 | struct fuse_open_in inarg; |
7078187a | 38 | FUSE_ARGS(args); |
fd72faac MS |
39 | |
40 | memset(&inarg, 0, sizeof(inarg)); | |
6ff958ed MS |
41 | inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY); |
42 | if (!fc->atomic_o_trunc) | |
43 | inarg.flags &= ~O_TRUNC; | |
d5b48543 MS |
44 | args.opcode = opcode; |
45 | args.nodeid = nodeid; | |
46 | args.in_numargs = 1; | |
47 | args.in_args[0].size = sizeof(inarg); | |
48 | args.in_args[0].value = &inarg; | |
49 | args.out_numargs = 1; | |
50 | args.out_args[0].size = sizeof(*outargp); | |
51 | args.out_args[0].value = outargp; | |
fd72faac | 52 | |
7078187a | 53 | return fuse_simple_request(fc, &args); |
fd72faac MS |
54 | } |
55 | ||
4cb54866 MS |
56 | struct fuse_release_args { |
57 | struct fuse_args args; | |
58 | struct fuse_release_in inarg; | |
59 | struct inode *inode; | |
60 | }; | |
61 | ||
acf99433 | 62 | struct fuse_file *fuse_file_alloc(struct fuse_conn *fc) |
fd72faac MS |
63 | { |
64 | struct fuse_file *ff; | |
6b2db28a | 65 | |
dc69e98c | 66 | ff = kzalloc(sizeof(struct fuse_file), GFP_KERNEL_ACCOUNT); |
6b2db28a TH |
67 | if (unlikely(!ff)) |
68 | return NULL; | |
69 | ||
da5e4714 | 70 | ff->fc = fc; |
dc69e98c KK |
71 | ff->release_args = kzalloc(sizeof(*ff->release_args), |
72 | GFP_KERNEL_ACCOUNT); | |
4cb54866 | 73 | if (!ff->release_args) { |
6b2db28a TH |
74 | kfree(ff); |
75 | return NULL; | |
fd72faac | 76 | } |
6b2db28a TH |
77 | |
78 | INIT_LIST_HEAD(&ff->write_entry); | |
5d7bc7e8 | 79 | mutex_init(&ff->readdir.lock); |
4e8c2eb5 | 80 | refcount_set(&ff->count, 1); |
6b2db28a TH |
81 | RB_CLEAR_NODE(&ff->polled_node); |
82 | init_waitqueue_head(&ff->poll_wait); | |
83 | ||
75126f55 | 84 | ff->kh = atomic64_inc_return(&fc->khctr); |
6b2db28a | 85 | |
fd72faac MS |
86 | return ff; |
87 | } | |
88 | ||
89 | void fuse_file_free(struct fuse_file *ff) | |
90 | { | |
4cb54866 | 91 | kfree(ff->release_args); |
5d7bc7e8 | 92 | mutex_destroy(&ff->readdir.lock); |
fd72faac MS |
93 | kfree(ff); |
94 | } | |
95 | ||
267d8444 | 96 | static struct fuse_file *fuse_file_get(struct fuse_file *ff) |
c756e0a4 | 97 | { |
4e8c2eb5 | 98 | refcount_inc(&ff->count); |
c756e0a4 MS |
99 | return ff; |
100 | } | |
101 | ||
4cb54866 MS |
102 | static void fuse_release_end(struct fuse_conn *fc, struct fuse_args *args, |
103 | int error) | |
819c4b3b | 104 | { |
4cb54866 MS |
105 | struct fuse_release_args *ra = container_of(args, typeof(*ra), args); |
106 | ||
107 | iput(ra->inode); | |
108 | kfree(ra); | |
819c4b3b MS |
109 | } |
110 | ||
2e64ff15 | 111 | static void fuse_file_put(struct fuse_file *ff, bool sync, bool isdir) |
c756e0a4 | 112 | { |
4e8c2eb5 | 113 | if (refcount_dec_and_test(&ff->count)) { |
4cb54866 | 114 | struct fuse_args *args = &ff->release_args->args; |
8b0797a4 | 115 | |
d9a9ea94 | 116 | if (isdir ? ff->fc->no_opendir : ff->fc->no_open) { |
4cb54866 MS |
117 | /* Do nothing when client does not implement 'open' */ |
118 | fuse_release_end(ff->fc, args, 0); | |
7678ac50 | 119 | } else if (sync) { |
4cb54866 MS |
120 | fuse_simple_request(ff->fc, args); |
121 | fuse_release_end(ff->fc, args, 0); | |
5a18ec17 | 122 | } else { |
4cb54866 MS |
123 | args->end = fuse_release_end; |
124 | if (fuse_simple_background(ff->fc, args, | |
125 | GFP_KERNEL | __GFP_NOFAIL)) | |
126 | fuse_release_end(ff->fc, args, -ENOTCONN); | |
5a18ec17 | 127 | } |
c756e0a4 MS |
128 | kfree(ff); |
129 | } | |
130 | } | |
131 | ||
08cbf542 TH |
132 | int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file, |
133 | bool isdir) | |
91fe96b4 | 134 | { |
91fe96b4 | 135 | struct fuse_file *ff; |
91fe96b4 MS |
136 | int opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN; |
137 | ||
138 | ff = fuse_file_alloc(fc); | |
139 | if (!ff) | |
140 | return -ENOMEM; | |
141 | ||
7678ac50 | 142 | ff->fh = 0; |
fabf7e02 CA |
143 | /* Default for no-open */ |
144 | ff->open_flags = FOPEN_KEEP_CACHE | (isdir ? FOPEN_CACHE_DIR : 0); | |
d9a9ea94 | 145 | if (isdir ? !fc->no_opendir : !fc->no_open) { |
7678ac50 AG |
146 | struct fuse_open_out outarg; |
147 | int err; | |
148 | ||
149 | err = fuse_send_open(fc, nodeid, file, opcode, &outarg); | |
150 | if (!err) { | |
151 | ff->fh = outarg.fh; | |
152 | ff->open_flags = outarg.open_flags; | |
153 | ||
d9a9ea94 | 154 | } else if (err != -ENOSYS) { |
7678ac50 AG |
155 | fuse_file_free(ff); |
156 | return err; | |
157 | } else { | |
d9a9ea94 CA |
158 | if (isdir) |
159 | fc->no_opendir = 1; | |
160 | else | |
161 | fc->no_open = 1; | |
7678ac50 | 162 | } |
91fe96b4 MS |
163 | } |
164 | ||
165 | if (isdir) | |
7678ac50 | 166 | ff->open_flags &= ~FOPEN_DIRECT_IO; |
91fe96b4 | 167 | |
91fe96b4 | 168 | ff->nodeid = nodeid; |
267d8444 | 169 | file->private_data = ff; |
91fe96b4 MS |
170 | |
171 | return 0; | |
172 | } | |
08cbf542 | 173 | EXPORT_SYMBOL_GPL(fuse_do_open); |
91fe96b4 | 174 | |
650b22b9 PE |
175 | static void fuse_link_write_file(struct file *file) |
176 | { | |
177 | struct inode *inode = file_inode(file); | |
650b22b9 PE |
178 | struct fuse_inode *fi = get_fuse_inode(inode); |
179 | struct fuse_file *ff = file->private_data; | |
180 | /* | |
181 | * file may be written through mmap, so chain it onto the | |
182 | * inodes's write_file list | |
183 | */ | |
f15ecfef | 184 | spin_lock(&fi->lock); |
650b22b9 PE |
185 | if (list_empty(&ff->write_entry)) |
186 | list_add(&ff->write_entry, &fi->write_files); | |
f15ecfef | 187 | spin_unlock(&fi->lock); |
650b22b9 PE |
188 | } |
189 | ||
c7b7143c | 190 | void fuse_finish_open(struct inode *inode, struct file *file) |
fd72faac | 191 | { |
c7b7143c | 192 | struct fuse_file *ff = file->private_data; |
a0822c55 | 193 | struct fuse_conn *fc = get_fuse_conn(inode); |
c7b7143c | 194 | |
c7b7143c | 195 | if (!(ff->open_flags & FOPEN_KEEP_CACHE)) |
b1009979 | 196 | invalidate_inode_pages2(inode->i_mapping); |
bbd84f33 KS |
197 | if (ff->open_flags & FOPEN_STREAM) |
198 | stream_open(inode, file); | |
199 | else if (ff->open_flags & FOPEN_NONSEEKABLE) | |
a7c1b990 | 200 | nonseekable_open(inode, file); |
a0822c55 KS |
201 | if (fc->atomic_o_trunc && (file->f_flags & O_TRUNC)) { |
202 | struct fuse_inode *fi = get_fuse_inode(inode); | |
203 | ||
f15ecfef | 204 | spin_lock(&fi->lock); |
4510d86f | 205 | fi->attr_version = atomic64_inc_return(&fc->attr_version); |
a0822c55 | 206 | i_size_write(inode, 0); |
f15ecfef | 207 | spin_unlock(&fi->lock); |
a0822c55 | 208 | fuse_invalidate_attr(inode); |
75caeecd MP |
209 | if (fc->writeback_cache) |
210 | file_update_time(file); | |
a0822c55 | 211 | } |
4d99ff8f PE |
212 | if ((file->f_mode & FMODE_WRITE) && fc->writeback_cache) |
213 | fuse_link_write_file(file); | |
fd72faac MS |
214 | } |
215 | ||
91fe96b4 | 216 | int fuse_open_common(struct inode *inode, struct file *file, bool isdir) |
fd72faac | 217 | { |
acf99433 | 218 | struct fuse_conn *fc = get_fuse_conn(inode); |
b6aeaded | 219 | int err; |
e4648309 | 220 | bool is_wb_truncate = (file->f_flags & O_TRUNC) && |
75caeecd MP |
221 | fc->atomic_o_trunc && |
222 | fc->writeback_cache; | |
b6aeaded MS |
223 | |
224 | err = generic_file_open(inode, file); | |
225 | if (err) | |
226 | return err; | |
227 | ||
e4648309 | 228 | if (is_wb_truncate) { |
5955102c | 229 | inode_lock(inode); |
e4648309 MS |
230 | fuse_set_nowrite(inode); |
231 | } | |
75caeecd | 232 | |
91fe96b4 | 233 | err = fuse_do_open(fc, get_node_id(inode), file, isdir); |
b6aeaded | 234 | |
75caeecd MP |
235 | if (!err) |
236 | fuse_finish_open(inode, file); | |
91fe96b4 | 237 | |
e4648309 MS |
238 | if (is_wb_truncate) { |
239 | fuse_release_nowrite(inode); | |
5955102c | 240 | inode_unlock(inode); |
e4648309 | 241 | } |
75caeecd MP |
242 | |
243 | return err; | |
b6aeaded MS |
244 | } |
245 | ||
ebf84d0c KT |
246 | static void fuse_prepare_release(struct fuse_inode *fi, struct fuse_file *ff, |
247 | int flags, int opcode) | |
64c6d8ed | 248 | { |
8b0797a4 | 249 | struct fuse_conn *fc = ff->fc; |
4cb54866 | 250 | struct fuse_release_args *ra = ff->release_args; |
b6aeaded | 251 | |
f15ecfef KT |
252 | /* Inode is NULL on error path of fuse_create_open() */ |
253 | if (likely(fi)) { | |
254 | spin_lock(&fi->lock); | |
255 | list_del(&ff->write_entry); | |
256 | spin_unlock(&fi->lock); | |
257 | } | |
8b0797a4 | 258 | spin_lock(&fc->lock); |
8b0797a4 MS |
259 | if (!RB_EMPTY_NODE(&ff->polled_node)) |
260 | rb_erase(&ff->polled_node, &fc->polled_files); | |
261 | spin_unlock(&fc->lock); | |
262 | ||
357ccf2b | 263 | wake_up_interruptible_all(&ff->poll_wait); |
8b0797a4 | 264 | |
4cb54866 MS |
265 | ra->inarg.fh = ff->fh; |
266 | ra->inarg.flags = flags; | |
267 | ra->args.in_numargs = 1; | |
268 | ra->args.in_args[0].size = sizeof(struct fuse_release_in); | |
269 | ra->args.in_args[0].value = &ra->inarg; | |
270 | ra->args.opcode = opcode; | |
271 | ra->args.nodeid = ff->nodeid; | |
272 | ra->args.force = true; | |
273 | ra->args.nocreds = true; | |
fd72faac MS |
274 | } |
275 | ||
2e64ff15 | 276 | void fuse_release_common(struct file *file, bool isdir) |
fd72faac | 277 | { |
ebf84d0c | 278 | struct fuse_inode *fi = get_fuse_inode(file_inode(file)); |
9a87ad3d | 279 | struct fuse_file *ff = file->private_data; |
4cb54866 | 280 | struct fuse_release_args *ra = ff->release_args; |
2e64ff15 | 281 | int opcode = isdir ? FUSE_RELEASEDIR : FUSE_RELEASE; |
6b2db28a | 282 | |
ebf84d0c | 283 | fuse_prepare_release(fi, ff, file->f_flags, opcode); |
6b2db28a | 284 | |
37fb3a30 | 285 | if (ff->flock) { |
4cb54866 MS |
286 | ra->inarg.release_flags |= FUSE_RELEASE_FLOCK_UNLOCK; |
287 | ra->inarg.lock_owner = fuse_lock_owner_id(ff->fc, | |
288 | (fl_owner_t) file); | |
37fb3a30 | 289 | } |
baebccbe | 290 | /* Hold inode until release is finished */ |
4cb54866 | 291 | ra->inode = igrab(file_inode(file)); |
6b2db28a | 292 | |
6b2db28a TH |
293 | /* |
294 | * Normally this will send the RELEASE request, however if | |
295 | * some asynchronous READ or WRITE requests are outstanding, | |
296 | * the sending will be delayed. | |
5a18ec17 MS |
297 | * |
298 | * Make the release synchronous if this is a fuseblk mount, | |
299 | * synchronous RELEASE is allowed (and desirable) in this case | |
300 | * because the server can be trusted not to screw up. | |
6b2db28a | 301 | */ |
1ccd1ea2 | 302 | fuse_file_put(ff, ff->fc->destroy, isdir); |
b6aeaded MS |
303 | } |
304 | ||
04730fef MS |
305 | static int fuse_open(struct inode *inode, struct file *file) |
306 | { | |
91fe96b4 | 307 | return fuse_open_common(inode, file, false); |
04730fef MS |
308 | } |
309 | ||
310 | static int fuse_release(struct inode *inode, struct file *file) | |
311 | { | |
e7cc133c PE |
312 | struct fuse_conn *fc = get_fuse_conn(inode); |
313 | ||
314 | /* see fuse_vma_close() for !writeback_cache case */ | |
315 | if (fc->writeback_cache) | |
1e18bda8 | 316 | write_inode_now(inode, 1); |
b0aa7606 | 317 | |
2e64ff15 | 318 | fuse_release_common(file, false); |
8b0797a4 MS |
319 | |
320 | /* return value is ignored by VFS */ | |
321 | return 0; | |
322 | } | |
323 | ||
ebf84d0c | 324 | void fuse_sync_release(struct fuse_inode *fi, struct fuse_file *ff, int flags) |
8b0797a4 | 325 | { |
4e8c2eb5 | 326 | WARN_ON(refcount_read(&ff->count) > 1); |
ebf84d0c | 327 | fuse_prepare_release(fi, ff, flags, FUSE_RELEASE); |
267d8444 MS |
328 | /* |
329 | * iput(NULL) is a no-op and since the refcount is 1 and everything's | |
330 | * synchronous, we are fine with not doing igrab() here" | |
331 | */ | |
2e64ff15 | 332 | fuse_file_put(ff, true, false); |
04730fef | 333 | } |
08cbf542 | 334 | EXPORT_SYMBOL_GPL(fuse_sync_release); |
04730fef | 335 | |
71421259 | 336 | /* |
9c8ef561 MS |
337 | * Scramble the ID space with XTEA, so that the value of the files_struct |
338 | * pointer is not exposed to userspace. | |
71421259 | 339 | */ |
f3332114 | 340 | u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id) |
71421259 | 341 | { |
9c8ef561 MS |
342 | u32 *k = fc->scramble_key; |
343 | u64 v = (unsigned long) id; | |
344 | u32 v0 = v; | |
345 | u32 v1 = v >> 32; | |
346 | u32 sum = 0; | |
347 | int i; | |
348 | ||
349 | for (i = 0; i < 32; i++) { | |
350 | v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]); | |
351 | sum += 0x9E3779B9; | |
352 | v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]); | |
353 | } | |
354 | ||
355 | return (u64) v0 + ((u64) v1 << 32); | |
71421259 MS |
356 | } |
357 | ||
33826ebb MS |
358 | struct fuse_writepage_args { |
359 | struct fuse_io_args ia; | |
6b2fb799 | 360 | struct rb_node writepages_entry; |
33826ebb MS |
361 | struct list_head queue_entry; |
362 | struct fuse_writepage_args *next; | |
363 | struct inode *inode; | |
364 | }; | |
365 | ||
366 | static struct fuse_writepage_args *fuse_find_writeback(struct fuse_inode *fi, | |
2fe93bd4 MS |
367 | pgoff_t idx_from, pgoff_t idx_to) |
368 | { | |
6b2fb799 MP |
369 | struct rb_node *n; |
370 | ||
371 | n = fi->writepages.rb_node; | |
2fe93bd4 | 372 | |
6b2fb799 MP |
373 | while (n) { |
374 | struct fuse_writepage_args *wpa; | |
2fe93bd4 MS |
375 | pgoff_t curr_index; |
376 | ||
6b2fb799 | 377 | wpa = rb_entry(n, struct fuse_writepage_args, writepages_entry); |
33826ebb MS |
378 | WARN_ON(get_fuse_inode(wpa->inode) != fi); |
379 | curr_index = wpa->ia.write.in.offset >> PAGE_SHIFT; | |
6b2fb799 MP |
380 | if (idx_from >= curr_index + wpa->ia.ap.num_pages) |
381 | n = n->rb_right; | |
382 | else if (idx_to < curr_index) | |
383 | n = n->rb_left; | |
384 | else | |
33826ebb | 385 | return wpa; |
2fe93bd4 MS |
386 | } |
387 | return NULL; | |
388 | } | |
389 | ||
3be5a52b | 390 | /* |
ea8cd333 | 391 | * Check if any page in a range is under writeback |
3be5a52b MS |
392 | * |
393 | * This is currently done by walking the list of writepage requests | |
394 | * for the inode, which can be pretty inefficient. | |
395 | */ | |
ea8cd333 PE |
396 | static bool fuse_range_is_writeback(struct inode *inode, pgoff_t idx_from, |
397 | pgoff_t idx_to) | |
3be5a52b | 398 | { |
3be5a52b | 399 | struct fuse_inode *fi = get_fuse_inode(inode); |
2fe93bd4 | 400 | bool found; |
3be5a52b | 401 | |
f15ecfef | 402 | spin_lock(&fi->lock); |
2fe93bd4 | 403 | found = fuse_find_writeback(fi, idx_from, idx_to); |
f15ecfef | 404 | spin_unlock(&fi->lock); |
3be5a52b MS |
405 | |
406 | return found; | |
407 | } | |
408 | ||
ea8cd333 PE |
409 | static inline bool fuse_page_is_writeback(struct inode *inode, pgoff_t index) |
410 | { | |
411 | return fuse_range_is_writeback(inode, index, index); | |
412 | } | |
413 | ||
3be5a52b MS |
414 | /* |
415 | * Wait for page writeback to be completed. | |
416 | * | |
417 | * Since fuse doesn't rely on the VM writeback tracking, this has to | |
418 | * use some other means. | |
419 | */ | |
17b2cbe2 | 420 | static void fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index) |
3be5a52b MS |
421 | { |
422 | struct fuse_inode *fi = get_fuse_inode(inode); | |
423 | ||
424 | wait_event(fi->page_waitq, !fuse_page_is_writeback(inode, index)); | |
3be5a52b MS |
425 | } |
426 | ||
fe38d7df MP |
427 | /* |
428 | * Wait for all pending writepages on the inode to finish. | |
429 | * | |
430 | * This is currently done by blocking further writes with FUSE_NOWRITE | |
431 | * and waiting for all sent writes to complete. | |
432 | * | |
433 | * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage | |
434 | * could conflict with truncation. | |
435 | */ | |
436 | static void fuse_sync_writes(struct inode *inode) | |
437 | { | |
438 | fuse_set_nowrite(inode); | |
439 | fuse_release_nowrite(inode); | |
440 | } | |
441 | ||
75e1fcc0 | 442 | static int fuse_flush(struct file *file, fl_owner_t id) |
b6aeaded | 443 | { |
6131ffaa | 444 | struct inode *inode = file_inode(file); |
b6aeaded MS |
445 | struct fuse_conn *fc = get_fuse_conn(inode); |
446 | struct fuse_file *ff = file->private_data; | |
b6aeaded | 447 | struct fuse_flush_in inarg; |
c500ebaa | 448 | FUSE_ARGS(args); |
b6aeaded MS |
449 | int err; |
450 | ||
248d86e8 MS |
451 | if (is_bad_inode(inode)) |
452 | return -EIO; | |
453 | ||
1e18bda8 | 454 | err = write_inode_now(inode, 1); |
fe38d7df MP |
455 | if (err) |
456 | return err; | |
457 | ||
5955102c | 458 | inode_lock(inode); |
fe38d7df | 459 | fuse_sync_writes(inode); |
5955102c | 460 | inode_unlock(inode); |
fe38d7df | 461 | |
4a7f4e88 | 462 | err = filemap_check_errors(file->f_mapping); |
9ebce595 MP |
463 | if (err) |
464 | return err; | |
465 | ||
614c026e MS |
466 | err = 0; |
467 | if (fc->no_flush) | |
468 | goto inval_attr_out; | |
469 | ||
b6aeaded MS |
470 | memset(&inarg, 0, sizeof(inarg)); |
471 | inarg.fh = ff->fh; | |
9c8ef561 | 472 | inarg.lock_owner = fuse_lock_owner_id(fc, id); |
c500ebaa MS |
473 | args.opcode = FUSE_FLUSH; |
474 | args.nodeid = get_node_id(inode); | |
475 | args.in_numargs = 1; | |
476 | args.in_args[0].size = sizeof(inarg); | |
477 | args.in_args[0].value = &inarg; | |
478 | args.force = true; | |
479 | ||
480 | err = fuse_simple_request(fc, &args); | |
b6aeaded MS |
481 | if (err == -ENOSYS) { |
482 | fc->no_flush = 1; | |
483 | err = 0; | |
484 | } | |
cf576c58 EG |
485 | |
486 | inval_attr_out: | |
487 | /* | |
488 | * In memory i_blocks is not maintained by fuse, if writeback cache is | |
489 | * enabled, i_blocks from cached attr may not be accurate. | |
490 | */ | |
491 | if (!err && fc->writeback_cache) | |
492 | fuse_invalidate_attr(inode); | |
b6aeaded MS |
493 | return err; |
494 | } | |
495 | ||
02c24a82 | 496 | int fuse_fsync_common(struct file *file, loff_t start, loff_t end, |
a9c2d1e8 | 497 | int datasync, int opcode) |
b6aeaded | 498 | { |
7ea80859 | 499 | struct inode *inode = file->f_mapping->host; |
b6aeaded MS |
500 | struct fuse_conn *fc = get_fuse_conn(inode); |
501 | struct fuse_file *ff = file->private_data; | |
7078187a | 502 | FUSE_ARGS(args); |
b6aeaded | 503 | struct fuse_fsync_in inarg; |
a9c2d1e8 MS |
504 | |
505 | memset(&inarg, 0, sizeof(inarg)); | |
506 | inarg.fh = ff->fh; | |
154603fe | 507 | inarg.fsync_flags = datasync ? FUSE_FSYNC_FDATASYNC : 0; |
d5b48543 MS |
508 | args.opcode = opcode; |
509 | args.nodeid = get_node_id(inode); | |
510 | args.in_numargs = 1; | |
511 | args.in_args[0].size = sizeof(inarg); | |
512 | args.in_args[0].value = &inarg; | |
a9c2d1e8 MS |
513 | return fuse_simple_request(fc, &args); |
514 | } | |
515 | ||
516 | static int fuse_fsync(struct file *file, loff_t start, loff_t end, | |
517 | int datasync) | |
518 | { | |
519 | struct inode *inode = file->f_mapping->host; | |
520 | struct fuse_conn *fc = get_fuse_conn(inode); | |
b6aeaded MS |
521 | int err; |
522 | ||
248d86e8 MS |
523 | if (is_bad_inode(inode)) |
524 | return -EIO; | |
525 | ||
5955102c | 526 | inode_lock(inode); |
02c24a82 | 527 | |
3be5a52b MS |
528 | /* |
529 | * Start writeback against all dirty pages of the inode, then | |
530 | * wait for all outstanding writes, before sending the FSYNC | |
531 | * request. | |
532 | */ | |
7e51fe1d | 533 | err = file_write_and_wait_range(file, start, end); |
3be5a52b | 534 | if (err) |
02c24a82 | 535 | goto out; |
3be5a52b MS |
536 | |
537 | fuse_sync_writes(inode); | |
ac7f052b AK |
538 | |
539 | /* | |
540 | * Due to implementation of fuse writeback | |
7e51fe1d | 541 | * file_write_and_wait_range() does not catch errors. |
ac7f052b AK |
542 | * We have to do this directly after fuse_sync_writes() |
543 | */ | |
7e51fe1d | 544 | err = file_check_and_advance_wb_err(file); |
ac7f052b AK |
545 | if (err) |
546 | goto out; | |
547 | ||
1e18bda8 MS |
548 | err = sync_inode_metadata(inode, 1); |
549 | if (err) | |
550 | goto out; | |
3be5a52b | 551 | |
a9c2d1e8 | 552 | if (fc->no_fsync) |
22401e7b | 553 | goto out; |
b0aa7606 | 554 | |
a9c2d1e8 | 555 | err = fuse_fsync_common(file, start, end, datasync, FUSE_FSYNC); |
b6aeaded | 556 | if (err == -ENOSYS) { |
a9c2d1e8 | 557 | fc->no_fsync = 1; |
b6aeaded MS |
558 | err = 0; |
559 | } | |
02c24a82 | 560 | out: |
5955102c | 561 | inode_unlock(inode); |
b6aeaded | 562 | |
a9c2d1e8 | 563 | return err; |
82547981 MS |
564 | } |
565 | ||
00793ca5 MS |
566 | void fuse_read_args_fill(struct fuse_io_args *ia, struct file *file, loff_t pos, |
567 | size_t count, int opcode) | |
568 | { | |
569 | struct fuse_file *ff = file->private_data; | |
570 | struct fuse_args *args = &ia->ap.args; | |
571 | ||
572 | ia->read.in.fh = ff->fh; | |
573 | ia->read.in.offset = pos; | |
574 | ia->read.in.size = count; | |
575 | ia->read.in.flags = file->f_flags; | |
576 | args->opcode = opcode; | |
577 | args->nodeid = ff->nodeid; | |
578 | args->in_numargs = 1; | |
579 | args->in_args[0].size = sizeof(ia->read.in); | |
580 | args->in_args[0].value = &ia->read.in; | |
581 | args->out_argvar = true; | |
582 | args->out_numargs = 1; | |
583 | args->out_args[0].size = count; | |
584 | } | |
585 | ||
45ac96ed MS |
586 | static void fuse_release_user_pages(struct fuse_args_pages *ap, |
587 | bool should_dirty) | |
187c5c36 | 588 | { |
45ac96ed | 589 | unsigned int i; |
187c5c36 | 590 | |
45ac96ed | 591 | for (i = 0; i < ap->num_pages; i++) { |
8fba54ae | 592 | if (should_dirty) |
45ac96ed MS |
593 | set_page_dirty_lock(ap->pages[i]); |
594 | put_page(ap->pages[i]); | |
187c5c36 MP |
595 | } |
596 | } | |
597 | ||
744742d6 SF |
598 | static void fuse_io_release(struct kref *kref) |
599 | { | |
600 | kfree(container_of(kref, struct fuse_io_priv, refcnt)); | |
601 | } | |
602 | ||
9d5722b7 CH |
603 | static ssize_t fuse_get_res_by_io(struct fuse_io_priv *io) |
604 | { | |
605 | if (io->err) | |
606 | return io->err; | |
607 | ||
608 | if (io->bytes >= 0 && io->write) | |
609 | return -EIO; | |
610 | ||
611 | return io->bytes < 0 ? io->size : io->bytes; | |
612 | } | |
613 | ||
01e9d11a MP |
614 | /** |
615 | * In case of short read, the caller sets 'pos' to the position of | |
616 | * actual end of fuse request in IO request. Otherwise, if bytes_requested | |
617 | * == bytes_transferred or rw == WRITE, the caller sets 'pos' to -1. | |
618 | * | |
619 | * An example: | |
620 | * User requested DIO read of 64K. It was splitted into two 32K fuse requests, | |
621 | * both submitted asynchronously. The first of them was ACKed by userspace as | |
622 | * fully completed (req->out.args[0].size == 32K) resulting in pos == -1. The | |
623 | * second request was ACKed as short, e.g. only 1K was read, resulting in | |
624 | * pos == 33K. | |
625 | * | |
626 | * Thus, when all fuse requests are completed, the minimal non-negative 'pos' | |
627 | * will be equal to the length of the longest contiguous fragment of | |
628 | * transferred data starting from the beginning of IO request. | |
629 | */ | |
630 | static void fuse_aio_complete(struct fuse_io_priv *io, int err, ssize_t pos) | |
631 | { | |
632 | int left; | |
633 | ||
634 | spin_lock(&io->lock); | |
635 | if (err) | |
636 | io->err = io->err ? : err; | |
637 | else if (pos >= 0 && (io->bytes < 0 || pos < io->bytes)) | |
638 | io->bytes = pos; | |
639 | ||
640 | left = --io->reqs; | |
7879c4e5 | 641 | if (!left && io->blocking) |
9d5722b7 | 642 | complete(io->done); |
01e9d11a MP |
643 | spin_unlock(&io->lock); |
644 | ||
7879c4e5 | 645 | if (!left && !io->blocking) { |
9d5722b7 | 646 | ssize_t res = fuse_get_res_by_io(io); |
01e9d11a | 647 | |
9d5722b7 CH |
648 | if (res >= 0) { |
649 | struct inode *inode = file_inode(io->iocb->ki_filp); | |
650 | struct fuse_conn *fc = get_fuse_conn(inode); | |
651 | struct fuse_inode *fi = get_fuse_inode(inode); | |
01e9d11a | 652 | |
f15ecfef | 653 | spin_lock(&fi->lock); |
4510d86f | 654 | fi->attr_version = atomic64_inc_return(&fc->attr_version); |
f15ecfef | 655 | spin_unlock(&fi->lock); |
01e9d11a MP |
656 | } |
657 | ||
04b2fa9f | 658 | io->iocb->ki_complete(io->iocb, res, 0); |
01e9d11a | 659 | } |
744742d6 SF |
660 | |
661 | kref_put(&io->refcnt, fuse_io_release); | |
01e9d11a MP |
662 | } |
663 | ||
45ac96ed MS |
664 | static struct fuse_io_args *fuse_io_alloc(struct fuse_io_priv *io, |
665 | unsigned int npages) | |
666 | { | |
667 | struct fuse_io_args *ia; | |
668 | ||
669 | ia = kzalloc(sizeof(*ia), GFP_KERNEL); | |
670 | if (ia) { | |
671 | ia->io = io; | |
672 | ia->ap.pages = fuse_pages_alloc(npages, GFP_KERNEL, | |
673 | &ia->ap.descs); | |
674 | if (!ia->ap.pages) { | |
675 | kfree(ia); | |
676 | ia = NULL; | |
677 | } | |
678 | } | |
679 | return ia; | |
680 | } | |
681 | ||
682 | static void fuse_io_free(struct fuse_io_args *ia) | |
01e9d11a | 683 | { |
45ac96ed MS |
684 | kfree(ia->ap.pages); |
685 | kfree(ia); | |
686 | } | |
687 | ||
688 | static void fuse_aio_complete_req(struct fuse_conn *fc, struct fuse_args *args, | |
689 | int err) | |
690 | { | |
691 | struct fuse_io_args *ia = container_of(args, typeof(*ia), ap.args); | |
692 | struct fuse_io_priv *io = ia->io; | |
01e9d11a MP |
693 | ssize_t pos = -1; |
694 | ||
45ac96ed | 695 | fuse_release_user_pages(&ia->ap, io->should_dirty); |
01e9d11a | 696 | |
45ac96ed MS |
697 | if (err) { |
698 | /* Nothing */ | |
699 | } else if (io->write) { | |
700 | if (ia->write.out.size > ia->write.in.size) { | |
701 | err = -EIO; | |
702 | } else if (ia->write.in.size != ia->write.out.size) { | |
703 | pos = ia->write.in.offset - io->offset + | |
704 | ia->write.out.size; | |
705 | } | |
01e9d11a | 706 | } else { |
45ac96ed MS |
707 | u32 outsize = args->out_args[0].size; |
708 | ||
709 | if (ia->read.in.size != outsize) | |
710 | pos = ia->read.in.offset - io->offset + outsize; | |
01e9d11a MP |
711 | } |
712 | ||
45ac96ed MS |
713 | fuse_aio_complete(io, err, pos); |
714 | fuse_io_free(ia); | |
01e9d11a MP |
715 | } |
716 | ||
45ac96ed MS |
717 | static ssize_t fuse_async_req_send(struct fuse_conn *fc, |
718 | struct fuse_io_args *ia, size_t num_bytes) | |
01e9d11a | 719 | { |
45ac96ed MS |
720 | ssize_t err; |
721 | struct fuse_io_priv *io = ia->io; | |
722 | ||
01e9d11a | 723 | spin_lock(&io->lock); |
744742d6 | 724 | kref_get(&io->refcnt); |
01e9d11a MP |
725 | io->size += num_bytes; |
726 | io->reqs++; | |
727 | spin_unlock(&io->lock); | |
728 | ||
45ac96ed | 729 | ia->ap.args.end = fuse_aio_complete_req; |
bb737bbe | 730 | ia->ap.args.may_block = io->should_dirty; |
45ac96ed | 731 | err = fuse_simple_background(fc, &ia->ap.args, GFP_KERNEL); |
f1ebdeff MS |
732 | if (err) |
733 | fuse_aio_complete_req(fc, &ia->ap.args, err); | |
01e9d11a | 734 | |
f1ebdeff | 735 | return num_bytes; |
01e9d11a MP |
736 | } |
737 | ||
45ac96ed MS |
738 | static ssize_t fuse_send_read(struct fuse_io_args *ia, loff_t pos, size_t count, |
739 | fl_owner_t owner) | |
04730fef | 740 | { |
45ac96ed | 741 | struct file *file = ia->io->iocb->ki_filp; |
2106cb18 MS |
742 | struct fuse_file *ff = file->private_data; |
743 | struct fuse_conn *fc = ff->fc; | |
f3332114 | 744 | |
45ac96ed | 745 | fuse_read_args_fill(ia, file, pos, count, FUSE_READ); |
f3332114 | 746 | if (owner != NULL) { |
45ac96ed MS |
747 | ia->read.in.read_flags |= FUSE_READ_LOCKOWNER; |
748 | ia->read.in.lock_owner = fuse_lock_owner_id(fc, owner); | |
f3332114 | 749 | } |
36cf66ed | 750 | |
45ac96ed MS |
751 | if (ia->io->async) |
752 | return fuse_async_req_send(fc, ia, count); | |
36cf66ed | 753 | |
45ac96ed | 754 | return fuse_simple_request(fc, &ia->ap.args); |
04730fef MS |
755 | } |
756 | ||
5c5c5e51 MS |
757 | static void fuse_read_update_size(struct inode *inode, loff_t size, |
758 | u64 attr_ver) | |
759 | { | |
760 | struct fuse_conn *fc = get_fuse_conn(inode); | |
761 | struct fuse_inode *fi = get_fuse_inode(inode); | |
762 | ||
f15ecfef | 763 | spin_lock(&fi->lock); |
06a7c3c2 MP |
764 | if (attr_ver == fi->attr_version && size < inode->i_size && |
765 | !test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) { | |
4510d86f | 766 | fi->attr_version = atomic64_inc_return(&fc->attr_version); |
5c5c5e51 MS |
767 | i_size_write(inode, size); |
768 | } | |
f15ecfef | 769 | spin_unlock(&fi->lock); |
5c5c5e51 MS |
770 | } |
771 | ||
a0d45d84 | 772 | static void fuse_short_read(struct inode *inode, u64 attr_ver, size_t num_read, |
134831e3 | 773 | struct fuse_args_pages *ap) |
a92adc82 | 774 | { |
8373200b PE |
775 | struct fuse_conn *fc = get_fuse_conn(inode); |
776 | ||
777 | if (fc->writeback_cache) { | |
778 | /* | |
779 | * A hole in a file. Some data after the hole are in page cache, | |
780 | * but have not reached the client fs yet. So, the hole is not | |
781 | * present there. | |
782 | */ | |
783 | int i; | |
09cbfeaf KS |
784 | int start_idx = num_read >> PAGE_SHIFT; |
785 | size_t off = num_read & (PAGE_SIZE - 1); | |
a92adc82 | 786 | |
134831e3 MS |
787 | for (i = start_idx; i < ap->num_pages; i++) { |
788 | zero_user_segment(ap->pages[i], off, PAGE_SIZE); | |
8373200b PE |
789 | off = 0; |
790 | } | |
791 | } else { | |
134831e3 | 792 | loff_t pos = page_offset(ap->pages[0]) + num_read; |
8373200b PE |
793 | fuse_read_update_size(inode, pos, attr_ver); |
794 | } | |
a92adc82 PE |
795 | } |
796 | ||
482fce55 | 797 | static int fuse_do_readpage(struct file *file, struct page *page) |
b6aeaded MS |
798 | { |
799 | struct inode *inode = page->mapping->host; | |
800 | struct fuse_conn *fc = get_fuse_conn(inode); | |
5c5c5e51 | 801 | loff_t pos = page_offset(page); |
00793ca5 MS |
802 | struct fuse_page_desc desc = { .length = PAGE_SIZE }; |
803 | struct fuse_io_args ia = { | |
804 | .ap.args.page_zeroing = true, | |
805 | .ap.args.out_pages = true, | |
806 | .ap.num_pages = 1, | |
807 | .ap.pages = &page, | |
808 | .ap.descs = &desc, | |
809 | }; | |
810 | ssize_t res; | |
5c5c5e51 | 811 | u64 attr_ver; |
248d86e8 | 812 | |
3be5a52b | 813 | /* |
25985edc | 814 | * Page writeback can extend beyond the lifetime of the |
3be5a52b MS |
815 | * page-cache page, so make sure we read a properly synced |
816 | * page. | |
817 | */ | |
818 | fuse_wait_on_page_writeback(inode, page->index); | |
819 | ||
5c5c5e51 MS |
820 | attr_ver = fuse_get_attr_version(fc); |
821 | ||
2f139829 MS |
822 | /* Don't overflow end offset */ |
823 | if (pos + (desc.length - 1) == LLONG_MAX) | |
824 | desc.length--; | |
825 | ||
00793ca5 MS |
826 | fuse_read_args_fill(&ia, file, pos, desc.length, FUSE_READ); |
827 | res = fuse_simple_request(fc, &ia.ap.args); | |
828 | if (res < 0) | |
829 | return res; | |
830 | /* | |
831 | * Short read means EOF. If file size is larger, truncate it | |
832 | */ | |
833 | if (res < desc.length) | |
134831e3 | 834 | fuse_short_read(inode, attr_ver, res, &ia.ap); |
5c5c5e51 | 835 | |
00793ca5 | 836 | SetPageUptodate(page); |
482fce55 | 837 | |
00793ca5 | 838 | return 0; |
482fce55 MP |
839 | } |
840 | ||
841 | static int fuse_readpage(struct file *file, struct page *page) | |
842 | { | |
843 | struct inode *inode = page->mapping->host; | |
844 | int err; | |
845 | ||
846 | err = -EIO; | |
847 | if (is_bad_inode(inode)) | |
848 | goto out; | |
849 | ||
850 | err = fuse_do_readpage(file, page); | |
451418fc | 851 | fuse_invalidate_atime(inode); |
b6aeaded MS |
852 | out: |
853 | unlock_page(page); | |
854 | return err; | |
855 | } | |
856 | ||
134831e3 MS |
857 | static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_args *args, |
858 | int err) | |
db50b96c | 859 | { |
c1aa96a5 | 860 | int i; |
134831e3 MS |
861 | struct fuse_io_args *ia = container_of(args, typeof(*ia), ap.args); |
862 | struct fuse_args_pages *ap = &ia->ap; | |
863 | size_t count = ia->read.in.size; | |
864 | size_t num_read = args->out_args[0].size; | |
ce534fb0 | 865 | struct address_space *mapping = NULL; |
c1aa96a5 | 866 | |
134831e3 MS |
867 | for (i = 0; mapping == NULL && i < ap->num_pages; i++) |
868 | mapping = ap->pages[i]->mapping; | |
5c5c5e51 | 869 | |
ce534fb0 MS |
870 | if (mapping) { |
871 | struct inode *inode = mapping->host; | |
872 | ||
873 | /* | |
874 | * Short read means EOF. If file size is larger, truncate it | |
875 | */ | |
134831e3 MS |
876 | if (!err && num_read < count) |
877 | fuse_short_read(inode, ia->read.attr_ver, num_read, ap); | |
ce534fb0 | 878 | |
451418fc | 879 | fuse_invalidate_atime(inode); |
ce534fb0 | 880 | } |
c1aa96a5 | 881 | |
134831e3 MS |
882 | for (i = 0; i < ap->num_pages; i++) { |
883 | struct page *page = ap->pages[i]; | |
884 | ||
885 | if (!err) | |
db50b96c | 886 | SetPageUptodate(page); |
c1aa96a5 MS |
887 | else |
888 | SetPageError(page); | |
db50b96c | 889 | unlock_page(page); |
09cbfeaf | 890 | put_page(page); |
db50b96c | 891 | } |
134831e3 MS |
892 | if (ia->ff) |
893 | fuse_file_put(ia->ff, false, false); | |
894 | ||
895 | fuse_io_free(ia); | |
c1aa96a5 MS |
896 | } |
897 | ||
134831e3 | 898 | static void fuse_send_readpages(struct fuse_io_args *ia, struct file *file) |
c1aa96a5 | 899 | { |
2106cb18 MS |
900 | struct fuse_file *ff = file->private_data; |
901 | struct fuse_conn *fc = ff->fc; | |
134831e3 MS |
902 | struct fuse_args_pages *ap = &ia->ap; |
903 | loff_t pos = page_offset(ap->pages[0]); | |
904 | size_t count = ap->num_pages << PAGE_SHIFT; | |
7df1e988 | 905 | ssize_t res; |
134831e3 MS |
906 | int err; |
907 | ||
908 | ap->args.out_pages = true; | |
909 | ap->args.page_zeroing = true; | |
910 | ap->args.page_replace = true; | |
2f139829 MS |
911 | |
912 | /* Don't overflow end offset */ | |
913 | if (pos + (count - 1) == LLONG_MAX) { | |
914 | count--; | |
915 | ap->descs[ap->num_pages - 1].length--; | |
916 | } | |
917 | WARN_ON((loff_t) (pos + count) < 0); | |
918 | ||
134831e3 MS |
919 | fuse_read_args_fill(ia, file, pos, count, FUSE_READ); |
920 | ia->read.attr_ver = fuse_get_attr_version(fc); | |
9cd68455 | 921 | if (fc->async_read) { |
134831e3 MS |
922 | ia->ff = fuse_file_get(ff); |
923 | ap->args.end = fuse_readpages_end; | |
924 | err = fuse_simple_background(fc, &ap->args, GFP_KERNEL); | |
925 | if (!err) | |
926 | return; | |
9cd68455 | 927 | } else { |
7df1e988 MS |
928 | res = fuse_simple_request(fc, &ap->args); |
929 | err = res < 0 ? res : 0; | |
9cd68455 | 930 | } |
134831e3 | 931 | fuse_readpages_end(fc, &ap->args, err); |
db50b96c MS |
932 | } |
933 | ||
c756e0a4 | 934 | struct fuse_fill_data { |
134831e3 | 935 | struct fuse_io_args *ia; |
a6643094 | 936 | struct file *file; |
db50b96c | 937 | struct inode *inode; |
134831e3 MS |
938 | unsigned int nr_pages; |
939 | unsigned int max_pages; | |
db50b96c MS |
940 | }; |
941 | ||
942 | static int fuse_readpages_fill(void *_data, struct page *page) | |
943 | { | |
c756e0a4 | 944 | struct fuse_fill_data *data = _data; |
134831e3 MS |
945 | struct fuse_io_args *ia = data->ia; |
946 | struct fuse_args_pages *ap = &ia->ap; | |
db50b96c MS |
947 | struct inode *inode = data->inode; |
948 | struct fuse_conn *fc = get_fuse_conn(inode); | |
949 | ||
3be5a52b MS |
950 | fuse_wait_on_page_writeback(inode, page->index); |
951 | ||
134831e3 MS |
952 | if (ap->num_pages && |
953 | (ap->num_pages == fc->max_pages || | |
954 | (ap->num_pages + 1) * PAGE_SIZE > fc->max_read || | |
955 | ap->pages[ap->num_pages - 1]->index + 1 != page->index)) { | |
956 | data->max_pages = min_t(unsigned int, data->nr_pages, | |
957 | fc->max_pages); | |
958 | fuse_send_readpages(ia, data->file); | |
959 | data->ia = ia = fuse_io_alloc(NULL, data->max_pages); | |
960 | if (!ia) { | |
db50b96c | 961 | unlock_page(page); |
134831e3 | 962 | return -ENOMEM; |
db50b96c | 963 | } |
134831e3 | 964 | ap = &ia->ap; |
db50b96c | 965 | } |
f8dbdf81 | 966 | |
134831e3 | 967 | if (WARN_ON(ap->num_pages >= data->max_pages)) { |
109728cc | 968 | unlock_page(page); |
134831e3 | 969 | fuse_io_free(ia); |
f8dbdf81 MP |
970 | return -EIO; |
971 | } | |
972 | ||
09cbfeaf | 973 | get_page(page); |
134831e3 MS |
974 | ap->pages[ap->num_pages] = page; |
975 | ap->descs[ap->num_pages].length = PAGE_SIZE; | |
976 | ap->num_pages++; | |
f8dbdf81 | 977 | data->nr_pages--; |
db50b96c MS |
978 | return 0; |
979 | } | |
980 | ||
981 | static int fuse_readpages(struct file *file, struct address_space *mapping, | |
982 | struct list_head *pages, unsigned nr_pages) | |
983 | { | |
984 | struct inode *inode = mapping->host; | |
985 | struct fuse_conn *fc = get_fuse_conn(inode); | |
c756e0a4 | 986 | struct fuse_fill_data data; |
db50b96c | 987 | int err; |
248d86e8 | 988 | |
1d7ea732 | 989 | err = -EIO; |
248d86e8 | 990 | if (is_bad_inode(inode)) |
2e990021 | 991 | goto out; |
248d86e8 | 992 | |
a6643094 | 993 | data.file = file; |
db50b96c | 994 | data.inode = inode; |
f8dbdf81 | 995 | data.nr_pages = nr_pages; |
134831e3 MS |
996 | data.max_pages = min_t(unsigned int, nr_pages, fc->max_pages); |
997 | ; | |
998 | data.ia = fuse_io_alloc(NULL, data.max_pages); | |
999 | err = -ENOMEM; | |
1000 | if (!data.ia) | |
2e990021 | 1001 | goto out; |
db50b96c MS |
1002 | |
1003 | err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data); | |
d3406ffa | 1004 | if (!err) { |
134831e3 MS |
1005 | if (data.ia->ap.num_pages) |
1006 | fuse_send_readpages(data.ia, file); | |
d3406ffa | 1007 | else |
134831e3 | 1008 | fuse_io_free(data.ia); |
d3406ffa | 1009 | } |
2e990021 | 1010 | out: |
1d7ea732 | 1011 | return err; |
db50b96c MS |
1012 | } |
1013 | ||
55752a3a | 1014 | static ssize_t fuse_cache_read_iter(struct kiocb *iocb, struct iov_iter *to) |
bcb4be80 MS |
1015 | { |
1016 | struct inode *inode = iocb->ki_filp->f_mapping->host; | |
a8894274 | 1017 | struct fuse_conn *fc = get_fuse_conn(inode); |
bcb4be80 | 1018 | |
a8894274 BF |
1019 | /* |
1020 | * In auto invalidate mode, always update attributes on read. | |
1021 | * Otherwise, only update if we attempt to read past EOF (to ensure | |
1022 | * i_size is up to date). | |
1023 | */ | |
1024 | if (fc->auto_inval_data || | |
37c20f16 | 1025 | (iocb->ki_pos + iov_iter_count(to) > i_size_read(inode))) { |
bcb4be80 | 1026 | int err; |
5b97eeac | 1027 | err = fuse_update_attributes(inode, iocb->ki_filp); |
bcb4be80 MS |
1028 | if (err) |
1029 | return err; | |
1030 | } | |
1031 | ||
37c20f16 | 1032 | return generic_file_read_iter(iocb, to); |
bcb4be80 MS |
1033 | } |
1034 | ||
338f2e3f MS |
1035 | static void fuse_write_args_fill(struct fuse_io_args *ia, struct fuse_file *ff, |
1036 | loff_t pos, size_t count) | |
1037 | { | |
1038 | struct fuse_args *args = &ia->ap.args; | |
1039 | ||
1040 | ia->write.in.fh = ff->fh; | |
1041 | ia->write.in.offset = pos; | |
1042 | ia->write.in.size = count; | |
1043 | args->opcode = FUSE_WRITE; | |
1044 | args->nodeid = ff->nodeid; | |
1045 | args->in_numargs = 2; | |
1046 | if (ff->fc->minor < 9) | |
1047 | args->in_args[0].size = FUSE_COMPAT_WRITE_IN_SIZE; | |
1048 | else | |
1049 | args->in_args[0].size = sizeof(ia->write.in); | |
1050 | args->in_args[0].value = &ia->write.in; | |
1051 | args->in_args[1].size = count; | |
1052 | args->out_numargs = 1; | |
1053 | args->out_args[0].size = sizeof(ia->write.out); | |
1054 | args->out_args[0].value = &ia->write.out; | |
1055 | } | |
1056 | ||
1057 | static unsigned int fuse_write_flags(struct kiocb *iocb) | |
1058 | { | |
1059 | unsigned int flags = iocb->ki_filp->f_flags; | |
1060 | ||
1061 | if (iocb->ki_flags & IOCB_DSYNC) | |
1062 | flags |= O_DSYNC; | |
1063 | if (iocb->ki_flags & IOCB_SYNC) | |
1064 | flags |= O_SYNC; | |
1065 | ||
1066 | return flags; | |
1067 | } | |
1068 | ||
45ac96ed MS |
1069 | static ssize_t fuse_send_write(struct fuse_io_args *ia, loff_t pos, |
1070 | size_t count, fl_owner_t owner) | |
b25e82e5 | 1071 | { |
45ac96ed | 1072 | struct kiocb *iocb = ia->io->iocb; |
e1c0eecb | 1073 | struct file *file = iocb->ki_filp; |
2106cb18 MS |
1074 | struct fuse_file *ff = file->private_data; |
1075 | struct fuse_conn *fc = ff->fc; | |
45ac96ed MS |
1076 | struct fuse_write_in *inarg = &ia->write.in; |
1077 | ssize_t err; | |
2d698b07 | 1078 | |
45ac96ed | 1079 | fuse_write_args_fill(ia, ff, pos, count); |
338f2e3f | 1080 | inarg->flags = fuse_write_flags(iocb); |
f3332114 | 1081 | if (owner != NULL) { |
f3332114 MS |
1082 | inarg->write_flags |= FUSE_WRITE_LOCKOWNER; |
1083 | inarg->lock_owner = fuse_lock_owner_id(fc, owner); | |
1084 | } | |
36cf66ed | 1085 | |
45ac96ed MS |
1086 | if (ia->io->async) |
1087 | return fuse_async_req_send(fc, ia, count); | |
1088 | ||
1089 | err = fuse_simple_request(fc, &ia->ap.args); | |
1090 | if (!err && ia->write.out.size > count) | |
1091 | err = -EIO; | |
36cf66ed | 1092 | |
45ac96ed | 1093 | return err ?: ia->write.out.size; |
b6aeaded MS |
1094 | } |
1095 | ||
b0aa7606 | 1096 | bool fuse_write_update_size(struct inode *inode, loff_t pos) |
854512ec MS |
1097 | { |
1098 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1099 | struct fuse_inode *fi = get_fuse_inode(inode); | |
b0aa7606 | 1100 | bool ret = false; |
854512ec | 1101 | |
f15ecfef | 1102 | spin_lock(&fi->lock); |
4510d86f | 1103 | fi->attr_version = atomic64_inc_return(&fc->attr_version); |
b0aa7606 | 1104 | if (pos > inode->i_size) { |
854512ec | 1105 | i_size_write(inode, pos); |
b0aa7606 MP |
1106 | ret = true; |
1107 | } | |
f15ecfef | 1108 | spin_unlock(&fi->lock); |
b0aa7606 MP |
1109 | |
1110 | return ret; | |
854512ec MS |
1111 | } |
1112 | ||
338f2e3f MS |
1113 | static ssize_t fuse_send_write_pages(struct fuse_io_args *ia, |
1114 | struct kiocb *iocb, struct inode *inode, | |
1115 | loff_t pos, size_t count) | |
ea9b9907 | 1116 | { |
338f2e3f MS |
1117 | struct fuse_args_pages *ap = &ia->ap; |
1118 | struct file *file = iocb->ki_filp; | |
1119 | struct fuse_file *ff = file->private_data; | |
1120 | struct fuse_conn *fc = ff->fc; | |
1121 | unsigned int offset, i; | |
1122 | int err; | |
ea9b9907 | 1123 | |
338f2e3f MS |
1124 | for (i = 0; i < ap->num_pages; i++) |
1125 | fuse_wait_on_page_writeback(inode, ap->pages[i]->index); | |
ea9b9907 | 1126 | |
338f2e3f MS |
1127 | fuse_write_args_fill(ia, ff, pos, count); |
1128 | ia->write.in.flags = fuse_write_flags(iocb); | |
ea9b9907 | 1129 | |
338f2e3f | 1130 | err = fuse_simple_request(fc, &ap->args); |
8aab336b MS |
1131 | if (!err && ia->write.out.size > count) |
1132 | err = -EIO; | |
338f2e3f MS |
1133 | |
1134 | offset = ap->descs[0].offset; | |
1135 | count = ia->write.out.size; | |
1136 | for (i = 0; i < ap->num_pages; i++) { | |
1137 | struct page *page = ap->pages[i]; | |
ea9b9907 | 1138 | |
338f2e3f | 1139 | if (!err && !offset && count >= PAGE_SIZE) |
ea9b9907 NP |
1140 | SetPageUptodate(page); |
1141 | ||
09cbfeaf KS |
1142 | if (count > PAGE_SIZE - offset) |
1143 | count -= PAGE_SIZE - offset; | |
ea9b9907 NP |
1144 | else |
1145 | count = 0; | |
1146 | offset = 0; | |
1147 | ||
1148 | unlock_page(page); | |
09cbfeaf | 1149 | put_page(page); |
ea9b9907 NP |
1150 | } |
1151 | ||
338f2e3f | 1152 | return err; |
ea9b9907 NP |
1153 | } |
1154 | ||
338f2e3f MS |
1155 | static ssize_t fuse_fill_write_pages(struct fuse_args_pages *ap, |
1156 | struct address_space *mapping, | |
1157 | struct iov_iter *ii, loff_t pos, | |
1158 | unsigned int max_pages) | |
ea9b9907 NP |
1159 | { |
1160 | struct fuse_conn *fc = get_fuse_conn(mapping->host); | |
09cbfeaf | 1161 | unsigned offset = pos & (PAGE_SIZE - 1); |
ea9b9907 NP |
1162 | size_t count = 0; |
1163 | int err; | |
1164 | ||
338f2e3f MS |
1165 | ap->args.in_pages = true; |
1166 | ap->descs[0].offset = offset; | |
ea9b9907 NP |
1167 | |
1168 | do { | |
1169 | size_t tmp; | |
1170 | struct page *page; | |
09cbfeaf KS |
1171 | pgoff_t index = pos >> PAGE_SHIFT; |
1172 | size_t bytes = min_t(size_t, PAGE_SIZE - offset, | |
ea9b9907 NP |
1173 | iov_iter_count(ii)); |
1174 | ||
1175 | bytes = min_t(size_t, bytes, fc->max_write - count); | |
1176 | ||
1177 | again: | |
1178 | err = -EFAULT; | |
1179 | if (iov_iter_fault_in_readable(ii, bytes)) | |
1180 | break; | |
1181 | ||
1182 | err = -ENOMEM; | |
54566b2c | 1183 | page = grab_cache_page_write_begin(mapping, index, 0); |
ea9b9907 NP |
1184 | if (!page) |
1185 | break; | |
1186 | ||
931e80e4 | 1187 | if (mapping_writably_mapped(mapping)) |
1188 | flush_dcache_page(page); | |
1189 | ||
ea9b9907 | 1190 | tmp = iov_iter_copy_from_user_atomic(page, ii, offset, bytes); |
ea9b9907 NP |
1191 | flush_dcache_page(page); |
1192 | ||
3ca8138f | 1193 | iov_iter_advance(ii, tmp); |
ea9b9907 NP |
1194 | if (!tmp) { |
1195 | unlock_page(page); | |
09cbfeaf | 1196 | put_page(page); |
ea9b9907 NP |
1197 | bytes = min(bytes, iov_iter_single_seg_count(ii)); |
1198 | goto again; | |
1199 | } | |
1200 | ||
1201 | err = 0; | |
338f2e3f MS |
1202 | ap->pages[ap->num_pages] = page; |
1203 | ap->descs[ap->num_pages].length = tmp; | |
1204 | ap->num_pages++; | |
ea9b9907 | 1205 | |
ea9b9907 NP |
1206 | count += tmp; |
1207 | pos += tmp; | |
1208 | offset += tmp; | |
09cbfeaf | 1209 | if (offset == PAGE_SIZE) |
ea9b9907 NP |
1210 | offset = 0; |
1211 | ||
78bb6cb9 MS |
1212 | if (!fc->big_writes) |
1213 | break; | |
ea9b9907 | 1214 | } while (iov_iter_count(ii) && count < fc->max_write && |
338f2e3f | 1215 | ap->num_pages < max_pages && offset == 0); |
ea9b9907 NP |
1216 | |
1217 | return count > 0 ? count : err; | |
1218 | } | |
1219 | ||
5da784cc CS |
1220 | static inline unsigned int fuse_wr_pages(loff_t pos, size_t len, |
1221 | unsigned int max_pages) | |
d07f09f5 | 1222 | { |
5da784cc | 1223 | return min_t(unsigned int, |
09cbfeaf KS |
1224 | ((pos + len - 1) >> PAGE_SHIFT) - |
1225 | (pos >> PAGE_SHIFT) + 1, | |
5da784cc | 1226 | max_pages); |
d07f09f5 MP |
1227 | } |
1228 | ||
e1c0eecb | 1229 | static ssize_t fuse_perform_write(struct kiocb *iocb, |
ea9b9907 NP |
1230 | struct address_space *mapping, |
1231 | struct iov_iter *ii, loff_t pos) | |
1232 | { | |
1233 | struct inode *inode = mapping->host; | |
1234 | struct fuse_conn *fc = get_fuse_conn(inode); | |
06a7c3c2 | 1235 | struct fuse_inode *fi = get_fuse_inode(inode); |
ea9b9907 NP |
1236 | int err = 0; |
1237 | ssize_t res = 0; | |
1238 | ||
06a7c3c2 MP |
1239 | if (inode->i_size < pos + iov_iter_count(ii)) |
1240 | set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); | |
1241 | ||
ea9b9907 | 1242 | do { |
ea9b9907 | 1243 | ssize_t count; |
338f2e3f MS |
1244 | struct fuse_io_args ia = {}; |
1245 | struct fuse_args_pages *ap = &ia.ap; | |
5da784cc CS |
1246 | unsigned int nr_pages = fuse_wr_pages(pos, iov_iter_count(ii), |
1247 | fc->max_pages); | |
ea9b9907 | 1248 | |
338f2e3f MS |
1249 | ap->pages = fuse_pages_alloc(nr_pages, GFP_KERNEL, &ap->descs); |
1250 | if (!ap->pages) { | |
1251 | err = -ENOMEM; | |
ea9b9907 NP |
1252 | break; |
1253 | } | |
1254 | ||
338f2e3f | 1255 | count = fuse_fill_write_pages(ap, mapping, ii, pos, nr_pages); |
ea9b9907 NP |
1256 | if (count <= 0) { |
1257 | err = count; | |
1258 | } else { | |
338f2e3f MS |
1259 | err = fuse_send_write_pages(&ia, iocb, inode, |
1260 | pos, count); | |
ea9b9907 | 1261 | if (!err) { |
338f2e3f MS |
1262 | size_t num_written = ia.write.out.size; |
1263 | ||
ea9b9907 NP |
1264 | res += num_written; |
1265 | pos += num_written; | |
1266 | ||
1267 | /* break out of the loop on short write */ | |
1268 | if (num_written != count) | |
1269 | err = -EIO; | |
1270 | } | |
1271 | } | |
338f2e3f | 1272 | kfree(ap->pages); |
ea9b9907 NP |
1273 | } while (!err && iov_iter_count(ii)); |
1274 | ||
1275 | if (res > 0) | |
1276 | fuse_write_update_size(inode, pos); | |
1277 | ||
06a7c3c2 | 1278 | clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); |
ea9b9907 NP |
1279 | fuse_invalidate_attr(inode); |
1280 | ||
1281 | return res > 0 ? res : err; | |
1282 | } | |
1283 | ||
55752a3a | 1284 | static ssize_t fuse_cache_write_iter(struct kiocb *iocb, struct iov_iter *from) |
ea9b9907 NP |
1285 | { |
1286 | struct file *file = iocb->ki_filp; | |
1287 | struct address_space *mapping = file->f_mapping; | |
ea9b9907 | 1288 | ssize_t written = 0; |
4273b793 | 1289 | ssize_t written_buffered = 0; |
ea9b9907 NP |
1290 | struct inode *inode = mapping->host; |
1291 | ssize_t err; | |
4273b793 | 1292 | loff_t endbyte = 0; |
ea9b9907 | 1293 | |
4d99ff8f PE |
1294 | if (get_fuse_conn(inode)->writeback_cache) { |
1295 | /* Update size (EOF optimization) and mode (SUID clearing) */ | |
5b97eeac | 1296 | err = fuse_update_attributes(mapping->host, file); |
4d99ff8f PE |
1297 | if (err) |
1298 | return err; | |
1299 | ||
84c3d55c | 1300 | return generic_file_write_iter(iocb, from); |
4d99ff8f PE |
1301 | } |
1302 | ||
5955102c | 1303 | inode_lock(inode); |
ea9b9907 NP |
1304 | |
1305 | /* We can write back this queue in page reclaim */ | |
de1414a6 | 1306 | current->backing_dev_info = inode_to_bdi(inode); |
ea9b9907 | 1307 | |
3309dd04 AV |
1308 | err = generic_write_checks(iocb, from); |
1309 | if (err <= 0) | |
ea9b9907 NP |
1310 | goto out; |
1311 | ||
5fa8e0a1 | 1312 | err = file_remove_privs(file); |
ea9b9907 NP |
1313 | if (err) |
1314 | goto out; | |
1315 | ||
c3b2da31 JB |
1316 | err = file_update_time(file); |
1317 | if (err) | |
1318 | goto out; | |
ea9b9907 | 1319 | |
2ba48ce5 | 1320 | if (iocb->ki_flags & IOCB_DIRECT) { |
3309dd04 | 1321 | loff_t pos = iocb->ki_pos; |
1af5bb49 | 1322 | written = generic_file_direct_write(iocb, from); |
84c3d55c | 1323 | if (written < 0 || !iov_iter_count(from)) |
4273b793 AA |
1324 | goto out; |
1325 | ||
1326 | pos += written; | |
ea9b9907 | 1327 | |
e1c0eecb | 1328 | written_buffered = fuse_perform_write(iocb, mapping, from, pos); |
4273b793 AA |
1329 | if (written_buffered < 0) { |
1330 | err = written_buffered; | |
1331 | goto out; | |
1332 | } | |
1333 | endbyte = pos + written_buffered - 1; | |
1334 | ||
1335 | err = filemap_write_and_wait_range(file->f_mapping, pos, | |
1336 | endbyte); | |
1337 | if (err) | |
1338 | goto out; | |
1339 | ||
1340 | invalidate_mapping_pages(file->f_mapping, | |
09cbfeaf KS |
1341 | pos >> PAGE_SHIFT, |
1342 | endbyte >> PAGE_SHIFT); | |
4273b793 AA |
1343 | |
1344 | written += written_buffered; | |
1345 | iocb->ki_pos = pos + written_buffered; | |
1346 | } else { | |
e1c0eecb | 1347 | written = fuse_perform_write(iocb, mapping, from, iocb->ki_pos); |
4273b793 | 1348 | if (written >= 0) |
3309dd04 | 1349 | iocb->ki_pos += written; |
4273b793 | 1350 | } |
ea9b9907 NP |
1351 | out: |
1352 | current->backing_dev_info = NULL; | |
5955102c | 1353 | inode_unlock(inode); |
e1c0eecb MS |
1354 | if (written > 0) |
1355 | written = generic_write_sync(iocb, written); | |
ea9b9907 NP |
1356 | |
1357 | return written ? written : err; | |
1358 | } | |
1359 | ||
093f38a2 MS |
1360 | static inline void fuse_page_descs_length_init(struct fuse_page_desc *descs, |
1361 | unsigned int index, | |
1362 | unsigned int nr_pages) | |
85f40aec MP |
1363 | { |
1364 | int i; | |
1365 | ||
7c190c8b | 1366 | for (i = index; i < index + nr_pages; i++) |
093f38a2 | 1367 | descs[i].length = PAGE_SIZE - descs[i].offset; |
85f40aec MP |
1368 | } |
1369 | ||
7c190c8b MP |
1370 | static inline unsigned long fuse_get_user_addr(const struct iov_iter *ii) |
1371 | { | |
1372 | return (unsigned long)ii->iov->iov_base + ii->iov_offset; | |
1373 | } | |
1374 | ||
1375 | static inline size_t fuse_get_frag_size(const struct iov_iter *ii, | |
1376 | size_t max_size) | |
1377 | { | |
1378 | return min(iov_iter_single_seg_count(ii), max_size); | |
1379 | } | |
1380 | ||
45ac96ed MS |
1381 | static int fuse_get_user_pages(struct fuse_args_pages *ap, struct iov_iter *ii, |
1382 | size_t *nbytesp, int write, | |
1383 | unsigned int max_pages) | |
413ef8cb | 1384 | { |
7c190c8b | 1385 | size_t nbytes = 0; /* # bytes already packed in req */ |
742f9927 | 1386 | ssize_t ret = 0; |
b98d023a | 1387 | |
f4975c67 | 1388 | /* Special case for kernel I/O: can copy directly into the buffer */ |
00e23707 | 1389 | if (iov_iter_is_kvec(ii)) { |
7c190c8b MP |
1390 | unsigned long user_addr = fuse_get_user_addr(ii); |
1391 | size_t frag_size = fuse_get_frag_size(ii, *nbytesp); | |
1392 | ||
f4975c67 | 1393 | if (write) |
45ac96ed | 1394 | ap->args.in_args[1].value = (void *) user_addr; |
f4975c67 | 1395 | else |
45ac96ed | 1396 | ap->args.out_args[0].value = (void *) user_addr; |
f4975c67 | 1397 | |
b98d023a MP |
1398 | iov_iter_advance(ii, frag_size); |
1399 | *nbytesp = frag_size; | |
f4975c67 MS |
1400 | return 0; |
1401 | } | |
413ef8cb | 1402 | |
45ac96ed | 1403 | while (nbytes < *nbytesp && ap->num_pages < max_pages) { |
7c190c8b | 1404 | unsigned npages; |
f67da30c | 1405 | size_t start; |
45ac96ed | 1406 | ret = iov_iter_get_pages(ii, &ap->pages[ap->num_pages], |
2c80929c | 1407 | *nbytesp - nbytes, |
45ac96ed | 1408 | max_pages - ap->num_pages, |
c7f3888a | 1409 | &start); |
7c190c8b | 1410 | if (ret < 0) |
742f9927 | 1411 | break; |
7c190c8b | 1412 | |
c9c37e2e AV |
1413 | iov_iter_advance(ii, ret); |
1414 | nbytes += ret; | |
7c190c8b | 1415 | |
c9c37e2e AV |
1416 | ret += start; |
1417 | npages = (ret + PAGE_SIZE - 1) / PAGE_SIZE; | |
7c190c8b | 1418 | |
45ac96ed MS |
1419 | ap->descs[ap->num_pages].offset = start; |
1420 | fuse_page_descs_length_init(ap->descs, ap->num_pages, npages); | |
7c190c8b | 1421 | |
45ac96ed MS |
1422 | ap->num_pages += npages; |
1423 | ap->descs[ap->num_pages - 1].length -= | |
c9c37e2e | 1424 | (PAGE_SIZE - ret) & (PAGE_SIZE - 1); |
7c190c8b | 1425 | } |
f4975c67 MS |
1426 | |
1427 | if (write) | |
cabdb4fa | 1428 | ap->args.in_pages = true; |
f4975c67 | 1429 | else |
cabdb4fa | 1430 | ap->args.out_pages = true; |
f4975c67 | 1431 | |
7c190c8b | 1432 | *nbytesp = nbytes; |
f4975c67 | 1433 | |
2c932d4c | 1434 | return ret < 0 ? ret : 0; |
413ef8cb MS |
1435 | } |
1436 | ||
d22a943f AV |
1437 | ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter, |
1438 | loff_t *ppos, int flags) | |
413ef8cb | 1439 | { |
ea8cd333 PE |
1440 | int write = flags & FUSE_DIO_WRITE; |
1441 | int cuse = flags & FUSE_DIO_CUSE; | |
e1c0eecb | 1442 | struct file *file = io->iocb->ki_filp; |
ea8cd333 | 1443 | struct inode *inode = file->f_mapping->host; |
2106cb18 MS |
1444 | struct fuse_file *ff = file->private_data; |
1445 | struct fuse_conn *fc = ff->fc; | |
413ef8cb MS |
1446 | size_t nmax = write ? fc->max_write : fc->max_read; |
1447 | loff_t pos = *ppos; | |
d22a943f | 1448 | size_t count = iov_iter_count(iter); |
09cbfeaf KS |
1449 | pgoff_t idx_from = pos >> PAGE_SHIFT; |
1450 | pgoff_t idx_to = (pos + count - 1) >> PAGE_SHIFT; | |
413ef8cb | 1451 | ssize_t res = 0; |
742f9927 | 1452 | int err = 0; |
45ac96ed MS |
1453 | struct fuse_io_args *ia; |
1454 | unsigned int max_pages; | |
248d86e8 | 1455 | |
45ac96ed MS |
1456 | max_pages = iov_iter_npages(iter, fc->max_pages); |
1457 | ia = fuse_io_alloc(io, max_pages); | |
1458 | if (!ia) | |
1459 | return -ENOMEM; | |
413ef8cb | 1460 | |
45ac96ed | 1461 | ia->io = io; |
ea8cd333 PE |
1462 | if (!cuse && fuse_range_is_writeback(inode, idx_from, idx_to)) { |
1463 | if (!write) | |
5955102c | 1464 | inode_lock(inode); |
ea8cd333 PE |
1465 | fuse_sync_writes(inode); |
1466 | if (!write) | |
5955102c | 1467 | inode_unlock(inode); |
ea8cd333 PE |
1468 | } |
1469 | ||
61c12b49 | 1470 | io->should_dirty = !write && iter_is_iovec(iter); |
413ef8cb | 1471 | while (count) { |
45ac96ed | 1472 | ssize_t nres; |
2106cb18 | 1473 | fl_owner_t owner = current->files; |
f4975c67 | 1474 | size_t nbytes = min(count, nmax); |
45ac96ed MS |
1475 | |
1476 | err = fuse_get_user_pages(&ia->ap, iter, &nbytes, write, | |
1477 | max_pages); | |
742f9927 | 1478 | if (err && !nbytes) |
413ef8cb | 1479 | break; |
f4975c67 | 1480 | |
4a2abf99 | 1481 | if (write) { |
45ac96ed MS |
1482 | if (!capable(CAP_FSETID)) |
1483 | ia->write.in.write_flags |= FUSE_WRITE_KILL_PRIV; | |
4a2abf99 | 1484 | |
45ac96ed | 1485 | nres = fuse_send_write(ia, pos, nbytes, owner); |
4a2abf99 | 1486 | } else { |
45ac96ed | 1487 | nres = fuse_send_read(ia, pos, nbytes, owner); |
4a2abf99 | 1488 | } |
2106cb18 | 1489 | |
45ac96ed MS |
1490 | if (!io->async || nres < 0) { |
1491 | fuse_release_user_pages(&ia->ap, io->should_dirty); | |
1492 | fuse_io_free(ia); | |
1493 | } | |
1494 | ia = NULL; | |
1495 | if (nres < 0) { | |
f658adee | 1496 | iov_iter_revert(iter, nbytes); |
45ac96ed | 1497 | err = nres; |
413ef8cb MS |
1498 | break; |
1499 | } | |
45ac96ed MS |
1500 | WARN_ON(nres > nbytes); |
1501 | ||
413ef8cb MS |
1502 | count -= nres; |
1503 | res += nres; | |
1504 | pos += nres; | |
f658adee MS |
1505 | if (nres != nbytes) { |
1506 | iov_iter_revert(iter, nbytes - nres); | |
413ef8cb | 1507 | break; |
f658adee | 1508 | } |
56cf34ff | 1509 | if (count) { |
45ac96ed MS |
1510 | max_pages = iov_iter_npages(iter, fc->max_pages); |
1511 | ia = fuse_io_alloc(io, max_pages); | |
1512 | if (!ia) | |
56cf34ff MS |
1513 | break; |
1514 | } | |
413ef8cb | 1515 | } |
45ac96ed MS |
1516 | if (ia) |
1517 | fuse_io_free(ia); | |
d09cb9d7 | 1518 | if (res > 0) |
413ef8cb | 1519 | *ppos = pos; |
413ef8cb | 1520 | |
742f9927 | 1521 | return res > 0 ? res : err; |
413ef8cb | 1522 | } |
08cbf542 | 1523 | EXPORT_SYMBOL_GPL(fuse_direct_io); |
413ef8cb | 1524 | |
36cf66ed | 1525 | static ssize_t __fuse_direct_read(struct fuse_io_priv *io, |
d22a943f AV |
1526 | struct iov_iter *iter, |
1527 | loff_t *ppos) | |
413ef8cb | 1528 | { |
d09cb9d7 | 1529 | ssize_t res; |
e1c0eecb | 1530 | struct inode *inode = file_inode(io->iocb->ki_filp); |
d09cb9d7 | 1531 | |
d22a943f | 1532 | res = fuse_direct_io(io, iter, ppos, 0); |
d09cb9d7 | 1533 | |
9a2eb24d | 1534 | fuse_invalidate_atime(inode); |
d09cb9d7 MS |
1535 | |
1536 | return res; | |
413ef8cb MS |
1537 | } |
1538 | ||
23c94e1c MR |
1539 | static ssize_t fuse_direct_IO(struct kiocb *iocb, struct iov_iter *iter); |
1540 | ||
15316263 | 1541 | static ssize_t fuse_direct_read_iter(struct kiocb *iocb, struct iov_iter *to) |
b98d023a | 1542 | { |
23c94e1c MR |
1543 | ssize_t res; |
1544 | ||
1545 | if (!is_sync_kiocb(iocb) && iocb->ki_flags & IOCB_DIRECT) { | |
23c94e1c MR |
1546 | res = fuse_direct_IO(iocb, to); |
1547 | } else { | |
1548 | struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb); | |
1549 | ||
1550 | res = __fuse_direct_read(&io, to, &iocb->ki_pos); | |
1551 | } | |
1552 | ||
1553 | return res; | |
b98d023a MP |
1554 | } |
1555 | ||
15316263 | 1556 | static ssize_t fuse_direct_write_iter(struct kiocb *iocb, struct iov_iter *from) |
4273b793 | 1557 | { |
e1c0eecb MS |
1558 | struct inode *inode = file_inode(iocb->ki_filp); |
1559 | struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb); | |
15316263 | 1560 | ssize_t res; |
4273b793 | 1561 | |
4273b793 | 1562 | /* Don't allow parallel writes to the same file */ |
5955102c | 1563 | inode_lock(inode); |
3309dd04 | 1564 | res = generic_write_checks(iocb, from); |
23c94e1c MR |
1565 | if (res > 0) { |
1566 | if (!is_sync_kiocb(iocb) && iocb->ki_flags & IOCB_DIRECT) { | |
1567 | res = fuse_direct_IO(iocb, from); | |
1568 | } else { | |
1569 | res = fuse_direct_io(&io, from, &iocb->ki_pos, | |
1570 | FUSE_DIO_WRITE); | |
1571 | } | |
1572 | } | |
812408fb | 1573 | fuse_invalidate_attr(inode); |
bcba24cc | 1574 | if (res > 0) |
15316263 | 1575 | fuse_write_update_size(inode, iocb->ki_pos); |
5955102c | 1576 | inode_unlock(inode); |
4273b793 AA |
1577 | |
1578 | return res; | |
1579 | } | |
1580 | ||
55752a3a MS |
1581 | static ssize_t fuse_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
1582 | { | |
2f7b6f5b MS |
1583 | struct file *file = iocb->ki_filp; |
1584 | struct fuse_file *ff = file->private_data; | |
1585 | ||
1586 | if (is_bad_inode(file_inode(file))) | |
1587 | return -EIO; | |
55752a3a MS |
1588 | |
1589 | if (!(ff->open_flags & FOPEN_DIRECT_IO)) | |
1590 | return fuse_cache_read_iter(iocb, to); | |
1591 | else | |
1592 | return fuse_direct_read_iter(iocb, to); | |
1593 | } | |
1594 | ||
1595 | static ssize_t fuse_file_write_iter(struct kiocb *iocb, struct iov_iter *from) | |
1596 | { | |
2f7b6f5b MS |
1597 | struct file *file = iocb->ki_filp; |
1598 | struct fuse_file *ff = file->private_data; | |
1599 | ||
1600 | if (is_bad_inode(file_inode(file))) | |
1601 | return -EIO; | |
55752a3a MS |
1602 | |
1603 | if (!(ff->open_flags & FOPEN_DIRECT_IO)) | |
1604 | return fuse_cache_write_iter(iocb, from); | |
1605 | else | |
1606 | return fuse_direct_write_iter(iocb, from); | |
1607 | } | |
1608 | ||
33826ebb | 1609 | static void fuse_writepage_free(struct fuse_writepage_args *wpa) |
b6aeaded | 1610 | { |
33826ebb | 1611 | struct fuse_args_pages *ap = &wpa->ia.ap; |
385b1268 PE |
1612 | int i; |
1613 | ||
33826ebb MS |
1614 | for (i = 0; i < ap->num_pages; i++) |
1615 | __free_page(ap->pages[i]); | |
1616 | ||
1617 | if (wpa->ia.ff) | |
1618 | fuse_file_put(wpa->ia.ff, false, false); | |
8b284dc4 | 1619 | |
33826ebb MS |
1620 | kfree(ap->pages); |
1621 | kfree(wpa); | |
3be5a52b MS |
1622 | } |
1623 | ||
33826ebb MS |
1624 | static void fuse_writepage_finish(struct fuse_conn *fc, |
1625 | struct fuse_writepage_args *wpa) | |
3be5a52b | 1626 | { |
33826ebb MS |
1627 | struct fuse_args_pages *ap = &wpa->ia.ap; |
1628 | struct inode *inode = wpa->inode; | |
3be5a52b | 1629 | struct fuse_inode *fi = get_fuse_inode(inode); |
de1414a6 | 1630 | struct backing_dev_info *bdi = inode_to_bdi(inode); |
385b1268 | 1631 | int i; |
3be5a52b | 1632 | |
6b2fb799 | 1633 | rb_erase(&wpa->writepages_entry, &fi->writepages); |
33826ebb | 1634 | for (i = 0; i < ap->num_pages; i++) { |
93f78d88 | 1635 | dec_wb_stat(&bdi->wb, WB_WRITEBACK); |
33826ebb | 1636 | dec_node_page_state(ap->pages[i], NR_WRITEBACK_TEMP); |
93f78d88 | 1637 | wb_writeout_inc(&bdi->wb); |
385b1268 | 1638 | } |
3be5a52b MS |
1639 | wake_up(&fi->page_waitq); |
1640 | } | |
1641 | ||
f15ecfef | 1642 | /* Called under fi->lock, may release and reacquire it */ |
33826ebb MS |
1643 | static void fuse_send_writepage(struct fuse_conn *fc, |
1644 | struct fuse_writepage_args *wpa, loff_t size) | |
f15ecfef KT |
1645 | __releases(fi->lock) |
1646 | __acquires(fi->lock) | |
3be5a52b | 1647 | { |
33826ebb MS |
1648 | struct fuse_writepage_args *aux, *next; |
1649 | struct fuse_inode *fi = get_fuse_inode(wpa->inode); | |
1650 | struct fuse_write_in *inarg = &wpa->ia.write.in; | |
1651 | struct fuse_args *args = &wpa->ia.ap.args; | |
1652 | __u64 data_size = wpa->ia.ap.num_pages * PAGE_SIZE; | |
1653 | int err; | |
3be5a52b | 1654 | |
33826ebb | 1655 | fi->writectr++; |
385b1268 PE |
1656 | if (inarg->offset + data_size <= size) { |
1657 | inarg->size = data_size; | |
3be5a52b | 1658 | } else if (inarg->offset < size) { |
385b1268 | 1659 | inarg->size = size - inarg->offset; |
3be5a52b MS |
1660 | } else { |
1661 | /* Got truncated off completely */ | |
1662 | goto out_free; | |
b6aeaded | 1663 | } |
3be5a52b | 1664 | |
33826ebb MS |
1665 | args->in_args[1].size = inarg->size; |
1666 | args->force = true; | |
1667 | args->nocreds = true; | |
1668 | ||
1669 | err = fuse_simple_background(fc, args, GFP_ATOMIC); | |
1670 | if (err == -ENOMEM) { | |
1671 | spin_unlock(&fi->lock); | |
1672 | err = fuse_simple_background(fc, args, GFP_NOFS | __GFP_NOFAIL); | |
1673 | spin_lock(&fi->lock); | |
1674 | } | |
1675 | ||
f15ecfef | 1676 | /* Fails on broken connection only */ |
33826ebb | 1677 | if (unlikely(err)) |
f15ecfef KT |
1678 | goto out_free; |
1679 | ||
3be5a52b MS |
1680 | return; |
1681 | ||
1682 | out_free: | |
33826ebb MS |
1683 | fi->writectr--; |
1684 | fuse_writepage_finish(fc, wpa); | |
f15ecfef | 1685 | spin_unlock(&fi->lock); |
e2653bd5 MS |
1686 | |
1687 | /* After fuse_writepage_finish() aux request list is private */ | |
33826ebb MS |
1688 | for (aux = wpa->next; aux; aux = next) { |
1689 | next = aux->next; | |
1690 | aux->next = NULL; | |
1691 | fuse_writepage_free(aux); | |
e2653bd5 MS |
1692 | } |
1693 | ||
33826ebb | 1694 | fuse_writepage_free(wpa); |
f15ecfef | 1695 | spin_lock(&fi->lock); |
b6aeaded MS |
1696 | } |
1697 | ||
3be5a52b MS |
1698 | /* |
1699 | * If fi->writectr is positive (no truncate or fsync going on) send | |
1700 | * all queued writepage requests. | |
1701 | * | |
f15ecfef | 1702 | * Called with fi->lock |
3be5a52b MS |
1703 | */ |
1704 | void fuse_flush_writepages(struct inode *inode) | |
f15ecfef KT |
1705 | __releases(fi->lock) |
1706 | __acquires(fi->lock) | |
b6aeaded | 1707 | { |
3be5a52b MS |
1708 | struct fuse_conn *fc = get_fuse_conn(inode); |
1709 | struct fuse_inode *fi = get_fuse_inode(inode); | |
9de5be06 | 1710 | loff_t crop = i_size_read(inode); |
33826ebb | 1711 | struct fuse_writepage_args *wpa; |
3be5a52b MS |
1712 | |
1713 | while (fi->writectr >= 0 && !list_empty(&fi->queued_writes)) { | |
33826ebb MS |
1714 | wpa = list_entry(fi->queued_writes.next, |
1715 | struct fuse_writepage_args, queue_entry); | |
1716 | list_del_init(&wpa->queue_entry); | |
1717 | fuse_send_writepage(fc, wpa, crop); | |
3be5a52b MS |
1718 | } |
1719 | } | |
1720 | ||
6b2fb799 MP |
1721 | static void tree_insert(struct rb_root *root, struct fuse_writepage_args *wpa) |
1722 | { | |
1723 | pgoff_t idx_from = wpa->ia.write.in.offset >> PAGE_SHIFT; | |
1724 | pgoff_t idx_to = idx_from + wpa->ia.ap.num_pages - 1; | |
1725 | struct rb_node **p = &root->rb_node; | |
1726 | struct rb_node *parent = NULL; | |
1727 | ||
1728 | WARN_ON(!wpa->ia.ap.num_pages); | |
1729 | while (*p) { | |
1730 | struct fuse_writepage_args *curr; | |
1731 | pgoff_t curr_index; | |
1732 | ||
1733 | parent = *p; | |
1734 | curr = rb_entry(parent, struct fuse_writepage_args, | |
1735 | writepages_entry); | |
1736 | WARN_ON(curr->inode != wpa->inode); | |
1737 | curr_index = curr->ia.write.in.offset >> PAGE_SHIFT; | |
1738 | ||
1739 | if (idx_from >= curr_index + curr->ia.ap.num_pages) | |
1740 | p = &(*p)->rb_right; | |
1741 | else if (idx_to < curr_index) | |
1742 | p = &(*p)->rb_left; | |
1743 | else | |
1744 | return (void) WARN_ON(true); | |
1745 | } | |
1746 | ||
1747 | rb_link_node(&wpa->writepages_entry, parent, p); | |
1748 | rb_insert_color(&wpa->writepages_entry, root); | |
1749 | } | |
1750 | ||
33826ebb MS |
1751 | static void fuse_writepage_end(struct fuse_conn *fc, struct fuse_args *args, |
1752 | int error) | |
3be5a52b | 1753 | { |
33826ebb MS |
1754 | struct fuse_writepage_args *wpa = |
1755 | container_of(args, typeof(*wpa), ia.ap.args); | |
1756 | struct inode *inode = wpa->inode; | |
3be5a52b MS |
1757 | struct fuse_inode *fi = get_fuse_inode(inode); |
1758 | ||
33826ebb | 1759 | mapping_set_error(inode->i_mapping, error); |
f15ecfef | 1760 | spin_lock(&fi->lock); |
33826ebb | 1761 | while (wpa->next) { |
6eaf4782 | 1762 | struct fuse_conn *fc = get_fuse_conn(inode); |
33826ebb MS |
1763 | struct fuse_write_in *inarg = &wpa->ia.write.in; |
1764 | struct fuse_writepage_args *next = wpa->next; | |
1765 | ||
1766 | wpa->next = next->next; | |
1767 | next->next = NULL; | |
1768 | next->ia.ff = fuse_file_get(wpa->ia.ff); | |
6b2fb799 | 1769 | tree_insert(&fi->writepages, next); |
6eaf4782 MP |
1770 | |
1771 | /* | |
1772 | * Skip fuse_flush_writepages() to make it easy to crop requests | |
1773 | * based on primary request size. | |
1774 | * | |
1775 | * 1st case (trivial): there are no concurrent activities using | |
1776 | * fuse_set/release_nowrite. Then we're on safe side because | |
1777 | * fuse_flush_writepages() would call fuse_send_writepage() | |
1778 | * anyway. | |
1779 | * | |
1780 | * 2nd case: someone called fuse_set_nowrite and it is waiting | |
1781 | * now for completion of all in-flight requests. This happens | |
1782 | * rarely and no more than once per page, so this should be | |
1783 | * okay. | |
1784 | * | |
1785 | * 3rd case: someone (e.g. fuse_do_setattr()) is in the middle | |
1786 | * of fuse_set_nowrite..fuse_release_nowrite section. The fact | |
1787 | * that fuse_set_nowrite returned implies that all in-flight | |
1788 | * requests were completed along with all of their secondary | |
1789 | * requests. Further primary requests are blocked by negative | |
1790 | * writectr. Hence there cannot be any in-flight requests and | |
1791 | * no invocations of fuse_writepage_end() while we're in | |
1792 | * fuse_set_nowrite..fuse_release_nowrite section. | |
1793 | */ | |
1794 | fuse_send_writepage(fc, next, inarg->offset + inarg->size); | |
8b284dc4 | 1795 | } |
3be5a52b | 1796 | fi->writectr--; |
33826ebb | 1797 | fuse_writepage_finish(fc, wpa); |
f15ecfef | 1798 | spin_unlock(&fi->lock); |
33826ebb | 1799 | fuse_writepage_free(wpa); |
3be5a52b MS |
1800 | } |
1801 | ||
1e18bda8 MS |
1802 | static struct fuse_file *__fuse_write_file_get(struct fuse_conn *fc, |
1803 | struct fuse_inode *fi) | |
adcadfa8 | 1804 | { |
72523425 | 1805 | struct fuse_file *ff = NULL; |
adcadfa8 | 1806 | |
f15ecfef | 1807 | spin_lock(&fi->lock); |
1e18bda8 | 1808 | if (!list_empty(&fi->write_files)) { |
72523425 MS |
1809 | ff = list_entry(fi->write_files.next, struct fuse_file, |
1810 | write_entry); | |
1811 | fuse_file_get(ff); | |
1812 | } | |
f15ecfef | 1813 | spin_unlock(&fi->lock); |
adcadfa8 PE |
1814 | |
1815 | return ff; | |
1816 | } | |
1817 | ||
1e18bda8 MS |
1818 | static struct fuse_file *fuse_write_file_get(struct fuse_conn *fc, |
1819 | struct fuse_inode *fi) | |
1820 | { | |
1821 | struct fuse_file *ff = __fuse_write_file_get(fc, fi); | |
1822 | WARN_ON(!ff); | |
1823 | return ff; | |
1824 | } | |
1825 | ||
1826 | int fuse_write_inode(struct inode *inode, struct writeback_control *wbc) | |
1827 | { | |
1828 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1829 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1830 | struct fuse_file *ff; | |
1831 | int err; | |
1832 | ||
1833 | ff = __fuse_write_file_get(fc, fi); | |
ab9e13f7 | 1834 | err = fuse_flush_times(inode, ff); |
1e18bda8 | 1835 | if (ff) |
2e64ff15 | 1836 | fuse_file_put(ff, false, false); |
1e18bda8 MS |
1837 | |
1838 | return err; | |
1839 | } | |
1840 | ||
33826ebb MS |
1841 | static struct fuse_writepage_args *fuse_writepage_args_alloc(void) |
1842 | { | |
1843 | struct fuse_writepage_args *wpa; | |
1844 | struct fuse_args_pages *ap; | |
1845 | ||
1846 | wpa = kzalloc(sizeof(*wpa), GFP_NOFS); | |
1847 | if (wpa) { | |
1848 | ap = &wpa->ia.ap; | |
1849 | ap->num_pages = 0; | |
1850 | ap->pages = fuse_pages_alloc(1, GFP_NOFS, &ap->descs); | |
1851 | if (!ap->pages) { | |
1852 | kfree(wpa); | |
1853 | wpa = NULL; | |
1854 | } | |
1855 | } | |
1856 | return wpa; | |
1857 | ||
1858 | } | |
1859 | ||
3be5a52b MS |
1860 | static int fuse_writepage_locked(struct page *page) |
1861 | { | |
1862 | struct address_space *mapping = page->mapping; | |
1863 | struct inode *inode = mapping->host; | |
1864 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1865 | struct fuse_inode *fi = get_fuse_inode(inode); | |
33826ebb MS |
1866 | struct fuse_writepage_args *wpa; |
1867 | struct fuse_args_pages *ap; | |
3be5a52b | 1868 | struct page *tmp_page; |
72523425 | 1869 | int error = -ENOMEM; |
3be5a52b MS |
1870 | |
1871 | set_page_writeback(page); | |
1872 | ||
33826ebb MS |
1873 | wpa = fuse_writepage_args_alloc(); |
1874 | if (!wpa) | |
3be5a52b | 1875 | goto err; |
33826ebb | 1876 | ap = &wpa->ia.ap; |
3be5a52b MS |
1877 | |
1878 | tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
1879 | if (!tmp_page) | |
1880 | goto err_free; | |
1881 | ||
72523425 | 1882 | error = -EIO; |
33826ebb MS |
1883 | wpa->ia.ff = fuse_write_file_get(fc, fi); |
1884 | if (!wpa->ia.ff) | |
27f1b363 | 1885 | goto err_nofile; |
72523425 | 1886 | |
33826ebb | 1887 | fuse_write_args_fill(&wpa->ia, wpa->ia.ff, page_offset(page), 0); |
3be5a52b MS |
1888 | |
1889 | copy_highpage(tmp_page, page); | |
33826ebb MS |
1890 | wpa->ia.write.in.write_flags |= FUSE_WRITE_CACHE; |
1891 | wpa->next = NULL; | |
1892 | ap->args.in_pages = true; | |
1893 | ap->num_pages = 1; | |
1894 | ap->pages[0] = tmp_page; | |
1895 | ap->descs[0].offset = 0; | |
1896 | ap->descs[0].length = PAGE_SIZE; | |
1897 | ap->args.end = fuse_writepage_end; | |
1898 | wpa->inode = inode; | |
3be5a52b | 1899 | |
93f78d88 | 1900 | inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK); |
11fb9989 | 1901 | inc_node_page_state(tmp_page, NR_WRITEBACK_TEMP); |
3be5a52b | 1902 | |
f15ecfef | 1903 | spin_lock(&fi->lock); |
6b2fb799 | 1904 | tree_insert(&fi->writepages, wpa); |
33826ebb | 1905 | list_add_tail(&wpa->queue_entry, &fi->queued_writes); |
3be5a52b | 1906 | fuse_flush_writepages(inode); |
f15ecfef | 1907 | spin_unlock(&fi->lock); |
3be5a52b | 1908 | |
4a4ac4eb MP |
1909 | end_page_writeback(page); |
1910 | ||
3be5a52b MS |
1911 | return 0; |
1912 | ||
27f1b363 MP |
1913 | err_nofile: |
1914 | __free_page(tmp_page); | |
3be5a52b | 1915 | err_free: |
33826ebb | 1916 | kfree(wpa); |
3be5a52b | 1917 | err: |
9183976e | 1918 | mapping_set_error(page->mapping, error); |
3be5a52b | 1919 | end_page_writeback(page); |
72523425 | 1920 | return error; |
3be5a52b MS |
1921 | } |
1922 | ||
1923 | static int fuse_writepage(struct page *page, struct writeback_control *wbc) | |
1924 | { | |
1925 | int err; | |
1926 | ||
ff17be08 MS |
1927 | if (fuse_page_is_writeback(page->mapping->host, page->index)) { |
1928 | /* | |
1929 | * ->writepages() should be called for sync() and friends. We | |
1930 | * should only get here on direct reclaim and then we are | |
1931 | * allowed to skip a page which is already in flight | |
1932 | */ | |
1933 | WARN_ON(wbc->sync_mode == WB_SYNC_ALL); | |
1934 | ||
1935 | redirty_page_for_writepage(wbc, page); | |
d5880c7a | 1936 | unlock_page(page); |
ff17be08 MS |
1937 | return 0; |
1938 | } | |
1939 | ||
3be5a52b MS |
1940 | err = fuse_writepage_locked(page); |
1941 | unlock_page(page); | |
1942 | ||
1943 | return err; | |
1944 | } | |
1945 | ||
26d614df | 1946 | struct fuse_fill_wb_data { |
33826ebb | 1947 | struct fuse_writepage_args *wpa; |
26d614df PE |
1948 | struct fuse_file *ff; |
1949 | struct inode *inode; | |
2d033eaa | 1950 | struct page **orig_pages; |
33826ebb | 1951 | unsigned int max_pages; |
26d614df PE |
1952 | }; |
1953 | ||
33826ebb MS |
1954 | static bool fuse_pages_realloc(struct fuse_fill_wb_data *data) |
1955 | { | |
1956 | struct fuse_args_pages *ap = &data->wpa->ia.ap; | |
1957 | struct fuse_conn *fc = get_fuse_conn(data->inode); | |
1958 | struct page **pages; | |
1959 | struct fuse_page_desc *descs; | |
1960 | unsigned int npages = min_t(unsigned int, | |
1961 | max_t(unsigned int, data->max_pages * 2, | |
1962 | FUSE_DEFAULT_MAX_PAGES_PER_REQ), | |
1963 | fc->max_pages); | |
1964 | WARN_ON(npages <= data->max_pages); | |
1965 | ||
1966 | pages = fuse_pages_alloc(npages, GFP_NOFS, &descs); | |
1967 | if (!pages) | |
1968 | return false; | |
1969 | ||
1970 | memcpy(pages, ap->pages, sizeof(struct page *) * ap->num_pages); | |
1971 | memcpy(descs, ap->descs, sizeof(struct fuse_page_desc) * ap->num_pages); | |
1972 | kfree(ap->pages); | |
1973 | ap->pages = pages; | |
1974 | ap->descs = descs; | |
1975 | data->max_pages = npages; | |
1976 | ||
1977 | return true; | |
1978 | } | |
1979 | ||
26d614df PE |
1980 | static void fuse_writepages_send(struct fuse_fill_wb_data *data) |
1981 | { | |
33826ebb | 1982 | struct fuse_writepage_args *wpa = data->wpa; |
26d614df | 1983 | struct inode *inode = data->inode; |
26d614df | 1984 | struct fuse_inode *fi = get_fuse_inode(inode); |
33826ebb | 1985 | int num_pages = wpa->ia.ap.num_pages; |
2d033eaa | 1986 | int i; |
26d614df | 1987 | |
33826ebb | 1988 | wpa->ia.ff = fuse_file_get(data->ff); |
f15ecfef | 1989 | spin_lock(&fi->lock); |
33826ebb | 1990 | list_add_tail(&wpa->queue_entry, &fi->queued_writes); |
26d614df | 1991 | fuse_flush_writepages(inode); |
f15ecfef | 1992 | spin_unlock(&fi->lock); |
2d033eaa MP |
1993 | |
1994 | for (i = 0; i < num_pages; i++) | |
1995 | end_page_writeback(data->orig_pages[i]); | |
26d614df PE |
1996 | } |
1997 | ||
7f305ca1 MS |
1998 | /* |
1999 | * First recheck under fi->lock if the offending offset is still under | |
419234d5 MS |
2000 | * writeback. If yes, then iterate auxiliary write requests, to see if there's |
2001 | * one already added for a page at this offset. If there's none, then insert | |
2002 | * this new request onto the auxiliary list, otherwise reuse the existing one by | |
7f305ca1 MS |
2003 | * copying the new page contents over to the old temporary page. |
2004 | */ | |
33826ebb | 2005 | static bool fuse_writepage_in_flight(struct fuse_writepage_args *new_wpa, |
8b284dc4 MS |
2006 | struct page *page) |
2007 | { | |
33826ebb MS |
2008 | struct fuse_inode *fi = get_fuse_inode(new_wpa->inode); |
2009 | struct fuse_writepage_args *tmp; | |
2010 | struct fuse_writepage_args *old_wpa; | |
2011 | struct fuse_args_pages *new_ap = &new_wpa->ia.ap; | |
8b284dc4 | 2012 | |
33826ebb | 2013 | WARN_ON(new_ap->num_pages != 0); |
8b284dc4 | 2014 | |
f15ecfef | 2015 | spin_lock(&fi->lock); |
6b2fb799 | 2016 | rb_erase(&new_wpa->writepages_entry, &fi->writepages); |
33826ebb MS |
2017 | old_wpa = fuse_find_writeback(fi, page->index, page->index); |
2018 | if (!old_wpa) { | |
6b2fb799 | 2019 | tree_insert(&fi->writepages, new_wpa); |
f15ecfef | 2020 | spin_unlock(&fi->lock); |
2fe93bd4 | 2021 | return false; |
f6011081 | 2022 | } |
8b284dc4 | 2023 | |
33826ebb MS |
2024 | new_ap->num_pages = 1; |
2025 | for (tmp = old_wpa->next; tmp; tmp = tmp->next) { | |
7f305ca1 MS |
2026 | pgoff_t curr_index; |
2027 | ||
33826ebb MS |
2028 | WARN_ON(tmp->inode != new_wpa->inode); |
2029 | curr_index = tmp->ia.write.in.offset >> PAGE_SHIFT; | |
419234d5 | 2030 | if (curr_index == page->index) { |
33826ebb MS |
2031 | WARN_ON(tmp->ia.ap.num_pages != 1); |
2032 | swap(tmp->ia.ap.pages[0], new_ap->pages[0]); | |
7f305ca1 | 2033 | break; |
8b284dc4 MS |
2034 | } |
2035 | } | |
2036 | ||
7f305ca1 | 2037 | if (!tmp) { |
33826ebb MS |
2038 | new_wpa->next = old_wpa->next; |
2039 | old_wpa->next = new_wpa; | |
7f305ca1 | 2040 | } |
41b6e41f | 2041 | |
f15ecfef | 2042 | spin_unlock(&fi->lock); |
7f305ca1 MS |
2043 | |
2044 | if (tmp) { | |
33826ebb | 2045 | struct backing_dev_info *bdi = inode_to_bdi(new_wpa->inode); |
8b284dc4 | 2046 | |
93f78d88 | 2047 | dec_wb_stat(&bdi->wb, WB_WRITEBACK); |
33826ebb | 2048 | dec_node_page_state(new_ap->pages[0], NR_WRITEBACK_TEMP); |
93f78d88 | 2049 | wb_writeout_inc(&bdi->wb); |
33826ebb | 2050 | fuse_writepage_free(new_wpa); |
8b284dc4 | 2051 | } |
7f305ca1 | 2052 | |
2fe93bd4 | 2053 | return true; |
8b284dc4 MS |
2054 | } |
2055 | ||
26d614df PE |
2056 | static int fuse_writepages_fill(struct page *page, |
2057 | struct writeback_control *wbc, void *_data) | |
2058 | { | |
2059 | struct fuse_fill_wb_data *data = _data; | |
33826ebb MS |
2060 | struct fuse_writepage_args *wpa = data->wpa; |
2061 | struct fuse_args_pages *ap = &wpa->ia.ap; | |
26d614df | 2062 | struct inode *inode = data->inode; |
f15ecfef | 2063 | struct fuse_inode *fi = get_fuse_inode(inode); |
26d614df PE |
2064 | struct fuse_conn *fc = get_fuse_conn(inode); |
2065 | struct page *tmp_page; | |
8b284dc4 | 2066 | bool is_writeback; |
26d614df PE |
2067 | int err; |
2068 | ||
2069 | if (!data->ff) { | |
2070 | err = -EIO; | |
091d1a72 | 2071 | data->ff = fuse_write_file_get(fc, fi); |
26d614df PE |
2072 | if (!data->ff) |
2073 | goto out_unlock; | |
2074 | } | |
2075 | ||
8b284dc4 MS |
2076 | /* |
2077 | * Being under writeback is unlikely but possible. For example direct | |
2078 | * read to an mmaped fuse file will set the page dirty twice; once when | |
2079 | * the pages are faulted with get_user_pages(), and then after the read | |
2080 | * completed. | |
2081 | */ | |
2082 | is_writeback = fuse_page_is_writeback(inode, page->index); | |
2083 | ||
33826ebb MS |
2084 | if (wpa && ap->num_pages && |
2085 | (is_writeback || ap->num_pages == fc->max_pages || | |
2086 | (ap->num_pages + 1) * PAGE_SIZE > fc->max_write || | |
2087 | data->orig_pages[ap->num_pages - 1]->index + 1 != page->index)) { | |
8b284dc4 | 2088 | fuse_writepages_send(data); |
33826ebb MS |
2089 | data->wpa = NULL; |
2090 | } else if (wpa && ap->num_pages == data->max_pages) { | |
2091 | if (!fuse_pages_realloc(data)) { | |
e52a8250 | 2092 | fuse_writepages_send(data); |
33826ebb | 2093 | data->wpa = NULL; |
e52a8250 | 2094 | } |
26d614df | 2095 | } |
e52a8250 | 2096 | |
26d614df PE |
2097 | err = -ENOMEM; |
2098 | tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
2099 | if (!tmp_page) | |
2100 | goto out_unlock; | |
2101 | ||
2102 | /* | |
2103 | * The page must not be redirtied until the writeout is completed | |
2104 | * (i.e. userspace has sent a reply to the write request). Otherwise | |
2105 | * there could be more than one temporary page instance for each real | |
2106 | * page. | |
2107 | * | |
2108 | * This is ensured by holding the page lock in page_mkwrite() while | |
2109 | * checking fuse_page_is_writeback(). We already hold the page lock | |
2110 | * since clear_page_dirty_for_io() and keep it held until we add the | |
33826ebb | 2111 | * request to the fi->writepages list and increment ap->num_pages. |
26d614df PE |
2112 | * After this fuse_page_is_writeback() will indicate that the page is |
2113 | * under writeback, so we can release the page lock. | |
2114 | */ | |
33826ebb | 2115 | if (data->wpa == NULL) { |
26d614df | 2116 | err = -ENOMEM; |
33826ebb MS |
2117 | wpa = fuse_writepage_args_alloc(); |
2118 | if (!wpa) { | |
26d614df PE |
2119 | __free_page(tmp_page); |
2120 | goto out_unlock; | |
2121 | } | |
33826ebb | 2122 | data->max_pages = 1; |
26d614df | 2123 | |
33826ebb MS |
2124 | ap = &wpa->ia.ap; |
2125 | fuse_write_args_fill(&wpa->ia, data->ff, page_offset(page), 0); | |
2126 | wpa->ia.write.in.write_flags |= FUSE_WRITE_CACHE; | |
2127 | wpa->next = NULL; | |
2128 | ap->args.in_pages = true; | |
2129 | ap->args.end = fuse_writepage_end; | |
2130 | ap->num_pages = 0; | |
2131 | wpa->inode = inode; | |
26d614df | 2132 | |
f15ecfef | 2133 | spin_lock(&fi->lock); |
6b2fb799 | 2134 | tree_insert(&fi->writepages, wpa); |
f15ecfef | 2135 | spin_unlock(&fi->lock); |
26d614df | 2136 | |
33826ebb | 2137 | data->wpa = wpa; |
26d614df PE |
2138 | } |
2139 | set_page_writeback(page); | |
2140 | ||
2141 | copy_highpage(tmp_page, page); | |
33826ebb MS |
2142 | ap->pages[ap->num_pages] = tmp_page; |
2143 | ap->descs[ap->num_pages].offset = 0; | |
2144 | ap->descs[ap->num_pages].length = PAGE_SIZE; | |
26d614df | 2145 | |
93f78d88 | 2146 | inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK); |
11fb9989 | 2147 | inc_node_page_state(tmp_page, NR_WRITEBACK_TEMP); |
8b284dc4 MS |
2148 | |
2149 | err = 0; | |
33826ebb | 2150 | if (is_writeback && fuse_writepage_in_flight(wpa, page)) { |
8b284dc4 | 2151 | end_page_writeback(page); |
33826ebb | 2152 | data->wpa = NULL; |
8b284dc4 MS |
2153 | goto out_unlock; |
2154 | } | |
33826ebb | 2155 | data->orig_pages[ap->num_pages] = page; |
26d614df PE |
2156 | |
2157 | /* | |
f15ecfef | 2158 | * Protected by fi->lock against concurrent access by |
26d614df PE |
2159 | * fuse_page_is_writeback(). |
2160 | */ | |
f15ecfef | 2161 | spin_lock(&fi->lock); |
33826ebb | 2162 | ap->num_pages++; |
f15ecfef | 2163 | spin_unlock(&fi->lock); |
26d614df | 2164 | |
26d614df PE |
2165 | out_unlock: |
2166 | unlock_page(page); | |
2167 | ||
2168 | return err; | |
2169 | } | |
2170 | ||
2171 | static int fuse_writepages(struct address_space *mapping, | |
2172 | struct writeback_control *wbc) | |
2173 | { | |
2174 | struct inode *inode = mapping->host; | |
5da784cc | 2175 | struct fuse_conn *fc = get_fuse_conn(inode); |
26d614df PE |
2176 | struct fuse_fill_wb_data data; |
2177 | int err; | |
2178 | ||
2179 | err = -EIO; | |
2180 | if (is_bad_inode(inode)) | |
2181 | goto out; | |
2182 | ||
2183 | data.inode = inode; | |
33826ebb | 2184 | data.wpa = NULL; |
26d614df PE |
2185 | data.ff = NULL; |
2186 | ||
2d033eaa | 2187 | err = -ENOMEM; |
5da784cc | 2188 | data.orig_pages = kcalloc(fc->max_pages, |
f2b3455e | 2189 | sizeof(struct page *), |
2d033eaa MP |
2190 | GFP_NOFS); |
2191 | if (!data.orig_pages) | |
2192 | goto out; | |
2193 | ||
26d614df | 2194 | err = write_cache_pages(mapping, wbc, fuse_writepages_fill, &data); |
33826ebb | 2195 | if (data.wpa) { |
26d614df | 2196 | /* Ignore errors if we can write at least one page */ |
33826ebb | 2197 | WARN_ON(!data.wpa->ia.ap.num_pages); |
26d614df PE |
2198 | fuse_writepages_send(&data); |
2199 | err = 0; | |
2200 | } | |
2201 | if (data.ff) | |
2e64ff15 | 2202 | fuse_file_put(data.ff, false, false); |
2d033eaa MP |
2203 | |
2204 | kfree(data.orig_pages); | |
26d614df PE |
2205 | out: |
2206 | return err; | |
2207 | } | |
2208 | ||
6b12c1b3 PE |
2209 | /* |
2210 | * It's worthy to make sure that space is reserved on disk for the write, | |
2211 | * but how to implement it without killing performance need more thinking. | |
2212 | */ | |
2213 | static int fuse_write_begin(struct file *file, struct address_space *mapping, | |
2214 | loff_t pos, unsigned len, unsigned flags, | |
2215 | struct page **pagep, void **fsdata) | |
2216 | { | |
09cbfeaf | 2217 | pgoff_t index = pos >> PAGE_SHIFT; |
a455589f | 2218 | struct fuse_conn *fc = get_fuse_conn(file_inode(file)); |
6b12c1b3 PE |
2219 | struct page *page; |
2220 | loff_t fsize; | |
2221 | int err = -ENOMEM; | |
2222 | ||
2223 | WARN_ON(!fc->writeback_cache); | |
2224 | ||
2225 | page = grab_cache_page_write_begin(mapping, index, flags); | |
2226 | if (!page) | |
2227 | goto error; | |
2228 | ||
2229 | fuse_wait_on_page_writeback(mapping->host, page->index); | |
2230 | ||
09cbfeaf | 2231 | if (PageUptodate(page) || len == PAGE_SIZE) |
6b12c1b3 PE |
2232 | goto success; |
2233 | /* | |
2234 | * Check if the start this page comes after the end of file, in which | |
2235 | * case the readpage can be optimized away. | |
2236 | */ | |
2237 | fsize = i_size_read(mapping->host); | |
09cbfeaf KS |
2238 | if (fsize <= (pos & PAGE_MASK)) { |
2239 | size_t off = pos & ~PAGE_MASK; | |
6b12c1b3 PE |
2240 | if (off) |
2241 | zero_user_segment(page, 0, off); | |
2242 | goto success; | |
2243 | } | |
2244 | err = fuse_do_readpage(file, page); | |
2245 | if (err) | |
2246 | goto cleanup; | |
2247 | success: | |
2248 | *pagep = page; | |
2249 | return 0; | |
2250 | ||
2251 | cleanup: | |
2252 | unlock_page(page); | |
09cbfeaf | 2253 | put_page(page); |
6b12c1b3 PE |
2254 | error: |
2255 | return err; | |
2256 | } | |
2257 | ||
2258 | static int fuse_write_end(struct file *file, struct address_space *mapping, | |
2259 | loff_t pos, unsigned len, unsigned copied, | |
2260 | struct page *page, void *fsdata) | |
2261 | { | |
2262 | struct inode *inode = page->mapping->host; | |
2263 | ||
59c3b76c MS |
2264 | /* Haven't copied anything? Skip zeroing, size extending, dirtying. */ |
2265 | if (!copied) | |
2266 | goto unlock; | |
2267 | ||
6b12c1b3 PE |
2268 | if (!PageUptodate(page)) { |
2269 | /* Zero any unwritten bytes at the end of the page */ | |
09cbfeaf | 2270 | size_t endoff = (pos + copied) & ~PAGE_MASK; |
6b12c1b3 | 2271 | if (endoff) |
09cbfeaf | 2272 | zero_user_segment(page, endoff, PAGE_SIZE); |
6b12c1b3 PE |
2273 | SetPageUptodate(page); |
2274 | } | |
2275 | ||
2276 | fuse_write_update_size(inode, pos + copied); | |
2277 | set_page_dirty(page); | |
59c3b76c MS |
2278 | |
2279 | unlock: | |
6b12c1b3 | 2280 | unlock_page(page); |
09cbfeaf | 2281 | put_page(page); |
6b12c1b3 PE |
2282 | |
2283 | return copied; | |
2284 | } | |
2285 | ||
3be5a52b MS |
2286 | static int fuse_launder_page(struct page *page) |
2287 | { | |
2288 | int err = 0; | |
2289 | if (clear_page_dirty_for_io(page)) { | |
2290 | struct inode *inode = page->mapping->host; | |
2291 | err = fuse_writepage_locked(page); | |
2292 | if (!err) | |
2293 | fuse_wait_on_page_writeback(inode, page->index); | |
2294 | } | |
2295 | return err; | |
2296 | } | |
2297 | ||
2298 | /* | |
2299 | * Write back dirty pages now, because there may not be any suitable | |
2300 | * open files later | |
2301 | */ | |
2302 | static void fuse_vma_close(struct vm_area_struct *vma) | |
2303 | { | |
2304 | filemap_write_and_wait(vma->vm_file->f_mapping); | |
2305 | } | |
2306 | ||
2307 | /* | |
2308 | * Wait for writeback against this page to complete before allowing it | |
2309 | * to be marked dirty again, and hence written back again, possibly | |
2310 | * before the previous writepage completed. | |
2311 | * | |
2312 | * Block here, instead of in ->writepage(), so that the userspace fs | |
2313 | * can only block processes actually operating on the filesystem. | |
2314 | * | |
2315 | * Otherwise unprivileged userspace fs would be able to block | |
2316 | * unrelated: | |
2317 | * | |
2318 | * - page migration | |
2319 | * - sync(2) | |
2320 | * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER | |
2321 | */ | |
46fb504a | 2322 | static vm_fault_t fuse_page_mkwrite(struct vm_fault *vmf) |
3be5a52b | 2323 | { |
c2ec175c | 2324 | struct page *page = vmf->page; |
11bac800 | 2325 | struct inode *inode = file_inode(vmf->vma->vm_file); |
cca24370 | 2326 | |
11bac800 | 2327 | file_update_time(vmf->vma->vm_file); |
cca24370 MS |
2328 | lock_page(page); |
2329 | if (page->mapping != inode->i_mapping) { | |
2330 | unlock_page(page); | |
2331 | return VM_FAULT_NOPAGE; | |
2332 | } | |
3be5a52b MS |
2333 | |
2334 | fuse_wait_on_page_writeback(inode, page->index); | |
cca24370 | 2335 | return VM_FAULT_LOCKED; |
3be5a52b MS |
2336 | } |
2337 | ||
f0f37e2f | 2338 | static const struct vm_operations_struct fuse_file_vm_ops = { |
3be5a52b MS |
2339 | .close = fuse_vma_close, |
2340 | .fault = filemap_fault, | |
f1820361 | 2341 | .map_pages = filemap_map_pages, |
3be5a52b MS |
2342 | .page_mkwrite = fuse_page_mkwrite, |
2343 | }; | |
2344 | ||
2345 | static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma) | |
2346 | { | |
55752a3a MS |
2347 | struct fuse_file *ff = file->private_data; |
2348 | ||
2349 | if (ff->open_flags & FOPEN_DIRECT_IO) { | |
2350 | /* Can't provide the coherency needed for MAP_SHARED */ | |
2351 | if (vma->vm_flags & VM_MAYSHARE) | |
2352 | return -ENODEV; | |
2353 | ||
2354 | invalidate_inode_pages2(file->f_mapping); | |
2355 | ||
2356 | return generic_file_mmap(file, vma); | |
2357 | } | |
2358 | ||
650b22b9 PE |
2359 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) |
2360 | fuse_link_write_file(file); | |
2361 | ||
3be5a52b MS |
2362 | file_accessed(file); |
2363 | vma->vm_ops = &fuse_file_vm_ops; | |
b6aeaded MS |
2364 | return 0; |
2365 | } | |
2366 | ||
0b6e9ea0 SF |
2367 | static int convert_fuse_file_lock(struct fuse_conn *fc, |
2368 | const struct fuse_file_lock *ffl, | |
71421259 MS |
2369 | struct file_lock *fl) |
2370 | { | |
2371 | switch (ffl->type) { | |
2372 | case F_UNLCK: | |
2373 | break; | |
2374 | ||
2375 | case F_RDLCK: | |
2376 | case F_WRLCK: | |
2377 | if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX || | |
2378 | ffl->end < ffl->start) | |
2379 | return -EIO; | |
2380 | ||
2381 | fl->fl_start = ffl->start; | |
2382 | fl->fl_end = ffl->end; | |
0b6e9ea0 SF |
2383 | |
2384 | /* | |
9d5b86ac BC |
2385 | * Convert pid into init's pid namespace. The locks API will |
2386 | * translate it into the caller's pid namespace. | |
0b6e9ea0 SF |
2387 | */ |
2388 | rcu_read_lock(); | |
9d5b86ac | 2389 | fl->fl_pid = pid_nr_ns(find_pid_ns(ffl->pid, fc->pid_ns), &init_pid_ns); |
0b6e9ea0 | 2390 | rcu_read_unlock(); |
71421259 MS |
2391 | break; |
2392 | ||
2393 | default: | |
2394 | return -EIO; | |
2395 | } | |
2396 | fl->fl_type = ffl->type; | |
2397 | return 0; | |
2398 | } | |
2399 | ||
7078187a | 2400 | static void fuse_lk_fill(struct fuse_args *args, struct file *file, |
a9ff4f87 | 2401 | const struct file_lock *fl, int opcode, pid_t pid, |
7078187a | 2402 | int flock, struct fuse_lk_in *inarg) |
71421259 | 2403 | { |
6131ffaa | 2404 | struct inode *inode = file_inode(file); |
9c8ef561 | 2405 | struct fuse_conn *fc = get_fuse_conn(inode); |
71421259 | 2406 | struct fuse_file *ff = file->private_data; |
7078187a MS |
2407 | |
2408 | memset(inarg, 0, sizeof(*inarg)); | |
2409 | inarg->fh = ff->fh; | |
2410 | inarg->owner = fuse_lock_owner_id(fc, fl->fl_owner); | |
2411 | inarg->lk.start = fl->fl_start; | |
2412 | inarg->lk.end = fl->fl_end; | |
2413 | inarg->lk.type = fl->fl_type; | |
2414 | inarg->lk.pid = pid; | |
a9ff4f87 | 2415 | if (flock) |
7078187a | 2416 | inarg->lk_flags |= FUSE_LK_FLOCK; |
d5b48543 MS |
2417 | args->opcode = opcode; |
2418 | args->nodeid = get_node_id(inode); | |
2419 | args->in_numargs = 1; | |
2420 | args->in_args[0].size = sizeof(*inarg); | |
2421 | args->in_args[0].value = inarg; | |
71421259 MS |
2422 | } |
2423 | ||
2424 | static int fuse_getlk(struct file *file, struct file_lock *fl) | |
2425 | { | |
6131ffaa | 2426 | struct inode *inode = file_inode(file); |
71421259 | 2427 | struct fuse_conn *fc = get_fuse_conn(inode); |
7078187a MS |
2428 | FUSE_ARGS(args); |
2429 | struct fuse_lk_in inarg; | |
71421259 MS |
2430 | struct fuse_lk_out outarg; |
2431 | int err; | |
2432 | ||
7078187a | 2433 | fuse_lk_fill(&args, file, fl, FUSE_GETLK, 0, 0, &inarg); |
d5b48543 MS |
2434 | args.out_numargs = 1; |
2435 | args.out_args[0].size = sizeof(outarg); | |
2436 | args.out_args[0].value = &outarg; | |
7078187a | 2437 | err = fuse_simple_request(fc, &args); |
71421259 | 2438 | if (!err) |
0b6e9ea0 | 2439 | err = convert_fuse_file_lock(fc, &outarg.lk, fl); |
71421259 MS |
2440 | |
2441 | return err; | |
2442 | } | |
2443 | ||
a9ff4f87 | 2444 | static int fuse_setlk(struct file *file, struct file_lock *fl, int flock) |
71421259 | 2445 | { |
6131ffaa | 2446 | struct inode *inode = file_inode(file); |
71421259 | 2447 | struct fuse_conn *fc = get_fuse_conn(inode); |
7078187a MS |
2448 | FUSE_ARGS(args); |
2449 | struct fuse_lk_in inarg; | |
71421259 | 2450 | int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK; |
0b6e9ea0 SF |
2451 | struct pid *pid = fl->fl_type != F_UNLCK ? task_tgid(current) : NULL; |
2452 | pid_t pid_nr = pid_nr_ns(pid, fc->pid_ns); | |
71421259 MS |
2453 | int err; |
2454 | ||
8fb47a4f | 2455 | if (fl->fl_lmops && fl->fl_lmops->lm_grant) { |
48e90761 MS |
2456 | /* NLM needs asynchronous locks, which we don't support yet */ |
2457 | return -ENOLCK; | |
2458 | } | |
2459 | ||
71421259 | 2460 | /* Unlock on close is handled by the flush method */ |
50f2112c | 2461 | if ((fl->fl_flags & FL_CLOSE_POSIX) == FL_CLOSE_POSIX) |
71421259 MS |
2462 | return 0; |
2463 | ||
0b6e9ea0 | 2464 | fuse_lk_fill(&args, file, fl, opcode, pid_nr, flock, &inarg); |
7078187a | 2465 | err = fuse_simple_request(fc, &args); |
71421259 | 2466 | |
a4d27e75 MS |
2467 | /* locking is restartable */ |
2468 | if (err == -EINTR) | |
2469 | err = -ERESTARTSYS; | |
7078187a | 2470 | |
71421259 MS |
2471 | return err; |
2472 | } | |
2473 | ||
2474 | static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl) | |
2475 | { | |
6131ffaa | 2476 | struct inode *inode = file_inode(file); |
71421259 MS |
2477 | struct fuse_conn *fc = get_fuse_conn(inode); |
2478 | int err; | |
2479 | ||
48e90761 MS |
2480 | if (cmd == F_CANCELLK) { |
2481 | err = 0; | |
2482 | } else if (cmd == F_GETLK) { | |
71421259 | 2483 | if (fc->no_lock) { |
9d6a8c5c | 2484 | posix_test_lock(file, fl); |
71421259 MS |
2485 | err = 0; |
2486 | } else | |
2487 | err = fuse_getlk(file, fl); | |
2488 | } else { | |
2489 | if (fc->no_lock) | |
48e90761 | 2490 | err = posix_lock_file(file, fl, NULL); |
71421259 | 2491 | else |
a9ff4f87 | 2492 | err = fuse_setlk(file, fl, 0); |
71421259 MS |
2493 | } |
2494 | return err; | |
2495 | } | |
2496 | ||
a9ff4f87 MS |
2497 | static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl) |
2498 | { | |
6131ffaa | 2499 | struct inode *inode = file_inode(file); |
a9ff4f87 MS |
2500 | struct fuse_conn *fc = get_fuse_conn(inode); |
2501 | int err; | |
2502 | ||
37fb3a30 | 2503 | if (fc->no_flock) { |
4f656367 | 2504 | err = locks_lock_file_wait(file, fl); |
a9ff4f87 | 2505 | } else { |
37fb3a30 MS |
2506 | struct fuse_file *ff = file->private_data; |
2507 | ||
a9ff4f87 | 2508 | /* emulate flock with POSIX locks */ |
37fb3a30 | 2509 | ff->flock = true; |
a9ff4f87 MS |
2510 | err = fuse_setlk(file, fl, 1); |
2511 | } | |
2512 | ||
2513 | return err; | |
2514 | } | |
2515 | ||
b2d2272f MS |
2516 | static sector_t fuse_bmap(struct address_space *mapping, sector_t block) |
2517 | { | |
2518 | struct inode *inode = mapping->host; | |
2519 | struct fuse_conn *fc = get_fuse_conn(inode); | |
7078187a | 2520 | FUSE_ARGS(args); |
b2d2272f MS |
2521 | struct fuse_bmap_in inarg; |
2522 | struct fuse_bmap_out outarg; | |
2523 | int err; | |
2524 | ||
2525 | if (!inode->i_sb->s_bdev || fc->no_bmap) | |
2526 | return 0; | |
2527 | ||
b2d2272f MS |
2528 | memset(&inarg, 0, sizeof(inarg)); |
2529 | inarg.block = block; | |
2530 | inarg.blocksize = inode->i_sb->s_blocksize; | |
d5b48543 MS |
2531 | args.opcode = FUSE_BMAP; |
2532 | args.nodeid = get_node_id(inode); | |
2533 | args.in_numargs = 1; | |
2534 | args.in_args[0].size = sizeof(inarg); | |
2535 | args.in_args[0].value = &inarg; | |
2536 | args.out_numargs = 1; | |
2537 | args.out_args[0].size = sizeof(outarg); | |
2538 | args.out_args[0].value = &outarg; | |
7078187a | 2539 | err = fuse_simple_request(fc, &args); |
b2d2272f MS |
2540 | if (err == -ENOSYS) |
2541 | fc->no_bmap = 1; | |
2542 | ||
2543 | return err ? 0 : outarg.block; | |
2544 | } | |
2545 | ||
0b5da8db R |
2546 | static loff_t fuse_lseek(struct file *file, loff_t offset, int whence) |
2547 | { | |
2548 | struct inode *inode = file->f_mapping->host; | |
2549 | struct fuse_conn *fc = get_fuse_conn(inode); | |
2550 | struct fuse_file *ff = file->private_data; | |
2551 | FUSE_ARGS(args); | |
2552 | struct fuse_lseek_in inarg = { | |
2553 | .fh = ff->fh, | |
2554 | .offset = offset, | |
2555 | .whence = whence | |
2556 | }; | |
2557 | struct fuse_lseek_out outarg; | |
2558 | int err; | |
2559 | ||
2560 | if (fc->no_lseek) | |
2561 | goto fallback; | |
2562 | ||
d5b48543 MS |
2563 | args.opcode = FUSE_LSEEK; |
2564 | args.nodeid = ff->nodeid; | |
2565 | args.in_numargs = 1; | |
2566 | args.in_args[0].size = sizeof(inarg); | |
2567 | args.in_args[0].value = &inarg; | |
2568 | args.out_numargs = 1; | |
2569 | args.out_args[0].size = sizeof(outarg); | |
2570 | args.out_args[0].value = &outarg; | |
0b5da8db R |
2571 | err = fuse_simple_request(fc, &args); |
2572 | if (err) { | |
2573 | if (err == -ENOSYS) { | |
2574 | fc->no_lseek = 1; | |
2575 | goto fallback; | |
2576 | } | |
2577 | return err; | |
2578 | } | |
2579 | ||
2580 | return vfs_setpos(file, outarg.offset, inode->i_sb->s_maxbytes); | |
2581 | ||
2582 | fallback: | |
5b97eeac | 2583 | err = fuse_update_attributes(inode, file); |
0b5da8db R |
2584 | if (!err) |
2585 | return generic_file_llseek(file, offset, whence); | |
2586 | else | |
2587 | return err; | |
2588 | } | |
2589 | ||
965c8e59 | 2590 | static loff_t fuse_file_llseek(struct file *file, loff_t offset, int whence) |
5559b8f4 MS |
2591 | { |
2592 | loff_t retval; | |
6131ffaa | 2593 | struct inode *inode = file_inode(file); |
5559b8f4 | 2594 | |
0b5da8db R |
2595 | switch (whence) { |
2596 | case SEEK_SET: | |
2597 | case SEEK_CUR: | |
2598 | /* No i_mutex protection necessary for SEEK_CUR and SEEK_SET */ | |
965c8e59 | 2599 | retval = generic_file_llseek(file, offset, whence); |
0b5da8db R |
2600 | break; |
2601 | case SEEK_END: | |
5955102c | 2602 | inode_lock(inode); |
5b97eeac | 2603 | retval = fuse_update_attributes(inode, file); |
0b5da8db R |
2604 | if (!retval) |
2605 | retval = generic_file_llseek(file, offset, whence); | |
5955102c | 2606 | inode_unlock(inode); |
0b5da8db R |
2607 | break; |
2608 | case SEEK_HOLE: | |
2609 | case SEEK_DATA: | |
5955102c | 2610 | inode_lock(inode); |
0b5da8db | 2611 | retval = fuse_lseek(file, offset, whence); |
5955102c | 2612 | inode_unlock(inode); |
0b5da8db R |
2613 | break; |
2614 | default: | |
2615 | retval = -EINVAL; | |
2616 | } | |
c07c3d19 | 2617 | |
5559b8f4 MS |
2618 | return retval; |
2619 | } | |
2620 | ||
d9d318d3 MS |
2621 | /* |
2622 | * CUSE servers compiled on 32bit broke on 64bit kernels because the | |
2623 | * ABI was defined to be 'struct iovec' which is different on 32bit | |
2624 | * and 64bit. Fortunately we can determine which structure the server | |
2625 | * used from the size of the reply. | |
2626 | */ | |
1baa26b2 MS |
2627 | static int fuse_copy_ioctl_iovec_old(struct iovec *dst, void *src, |
2628 | size_t transferred, unsigned count, | |
2629 | bool is_compat) | |
d9d318d3 MS |
2630 | { |
2631 | #ifdef CONFIG_COMPAT | |
2632 | if (count * sizeof(struct compat_iovec) == transferred) { | |
2633 | struct compat_iovec *ciov = src; | |
2634 | unsigned i; | |
2635 | ||
2636 | /* | |
2637 | * With this interface a 32bit server cannot support | |
2638 | * non-compat (i.e. ones coming from 64bit apps) ioctl | |
2639 | * requests | |
2640 | */ | |
2641 | if (!is_compat) | |
2642 | return -EINVAL; | |
2643 | ||
2644 | for (i = 0; i < count; i++) { | |
2645 | dst[i].iov_base = compat_ptr(ciov[i].iov_base); | |
2646 | dst[i].iov_len = ciov[i].iov_len; | |
2647 | } | |
2648 | return 0; | |
2649 | } | |
2650 | #endif | |
2651 | ||
2652 | if (count * sizeof(struct iovec) != transferred) | |
2653 | return -EIO; | |
2654 | ||
2655 | memcpy(dst, src, transferred); | |
2656 | return 0; | |
2657 | } | |
2658 | ||
7572777e | 2659 | /* Make sure iov_length() won't overflow */ |
5da784cc CS |
2660 | static int fuse_verify_ioctl_iov(struct fuse_conn *fc, struct iovec *iov, |
2661 | size_t count) | |
7572777e MS |
2662 | { |
2663 | size_t n; | |
5da784cc | 2664 | u32 max = fc->max_pages << PAGE_SHIFT; |
7572777e | 2665 | |
fb6ccff6 | 2666 | for (n = 0; n < count; n++, iov++) { |
7572777e MS |
2667 | if (iov->iov_len > (size_t) max) |
2668 | return -ENOMEM; | |
2669 | max -= iov->iov_len; | |
2670 | } | |
2671 | return 0; | |
2672 | } | |
2673 | ||
1baa26b2 MS |
2674 | static int fuse_copy_ioctl_iovec(struct fuse_conn *fc, struct iovec *dst, |
2675 | void *src, size_t transferred, unsigned count, | |
2676 | bool is_compat) | |
2677 | { | |
2678 | unsigned i; | |
2679 | struct fuse_ioctl_iovec *fiov = src; | |
2680 | ||
2681 | if (fc->minor < 16) { | |
2682 | return fuse_copy_ioctl_iovec_old(dst, src, transferred, | |
2683 | count, is_compat); | |
2684 | } | |
2685 | ||
2686 | if (count * sizeof(struct fuse_ioctl_iovec) != transferred) | |
2687 | return -EIO; | |
2688 | ||
2689 | for (i = 0; i < count; i++) { | |
2690 | /* Did the server supply an inappropriate value? */ | |
2691 | if (fiov[i].base != (unsigned long) fiov[i].base || | |
2692 | fiov[i].len != (unsigned long) fiov[i].len) | |
2693 | return -EIO; | |
2694 | ||
2695 | dst[i].iov_base = (void __user *) (unsigned long) fiov[i].base; | |
2696 | dst[i].iov_len = (size_t) fiov[i].len; | |
2697 | ||
2698 | #ifdef CONFIG_COMPAT | |
2699 | if (is_compat && | |
2700 | (ptr_to_compat(dst[i].iov_base) != fiov[i].base || | |
2701 | (compat_size_t) dst[i].iov_len != fiov[i].len)) | |
2702 | return -EIO; | |
2703 | #endif | |
2704 | } | |
2705 | ||
2706 | return 0; | |
2707 | } | |
2708 | ||
2709 | ||
59efec7b TH |
2710 | /* |
2711 | * For ioctls, there is no generic way to determine how much memory | |
2712 | * needs to be read and/or written. Furthermore, ioctls are allowed | |
2713 | * to dereference the passed pointer, so the parameter requires deep | |
2714 | * copying but FUSE has no idea whatsoever about what to copy in or | |
2715 | * out. | |
2716 | * | |
2717 | * This is solved by allowing FUSE server to retry ioctl with | |
2718 | * necessary in/out iovecs. Let's assume the ioctl implementation | |
2719 | * needs to read in the following structure. | |
2720 | * | |
2721 | * struct a { | |
2722 | * char *buf; | |
2723 | * size_t buflen; | |
2724 | * } | |
2725 | * | |
2726 | * On the first callout to FUSE server, inarg->in_size and | |
2727 | * inarg->out_size will be NULL; then, the server completes the ioctl | |
2728 | * with FUSE_IOCTL_RETRY set in out->flags, out->in_iovs set to 1 and | |
2729 | * the actual iov array to | |
2730 | * | |
2731 | * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) } } | |
2732 | * | |
2733 | * which tells FUSE to copy in the requested area and retry the ioctl. | |
2734 | * On the second round, the server has access to the structure and | |
2735 | * from that it can tell what to look for next, so on the invocation, | |
2736 | * it sets FUSE_IOCTL_RETRY, out->in_iovs to 2 and iov array to | |
2737 | * | |
2738 | * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) }, | |
2739 | * { .iov_base = a.buf, .iov_len = a.buflen } } | |
2740 | * | |
2741 | * FUSE will copy both struct a and the pointed buffer from the | |
2742 | * process doing the ioctl and retry ioctl with both struct a and the | |
2743 | * buffer. | |
2744 | * | |
2745 | * This time, FUSE server has everything it needs and completes ioctl | |
2746 | * without FUSE_IOCTL_RETRY which finishes the ioctl call. | |
2747 | * | |
2748 | * Copying data out works the same way. | |
2749 | * | |
2750 | * Note that if FUSE_IOCTL_UNRESTRICTED is clear, the kernel | |
2751 | * automatically initializes in and out iovs by decoding @cmd with | |
2752 | * _IOC_* macros and the server is not allowed to request RETRY. This | |
2753 | * limits ioctl data transfers to well-formed ioctls and is the forced | |
2754 | * behavior for all FUSE servers. | |
2755 | */ | |
08cbf542 TH |
2756 | long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg, |
2757 | unsigned int flags) | |
59efec7b | 2758 | { |
59efec7b | 2759 | struct fuse_file *ff = file->private_data; |
d36f2487 | 2760 | struct fuse_conn *fc = ff->fc; |
59efec7b TH |
2761 | struct fuse_ioctl_in inarg = { |
2762 | .fh = ff->fh, | |
2763 | .cmd = cmd, | |
2764 | .arg = arg, | |
2765 | .flags = flags | |
2766 | }; | |
2767 | struct fuse_ioctl_out outarg; | |
8ac83505 | 2768 | struct iovec *iov_page = NULL; |
59efec7b | 2769 | struct iovec *in_iov = NULL, *out_iov = NULL; |
093f38a2 MS |
2770 | unsigned int in_iovs = 0, out_iovs = 0, max_pages; |
2771 | size_t in_size, out_size, c; | |
2772 | ssize_t transferred; | |
acbe5fda MS |
2773 | int err, i; |
2774 | struct iov_iter ii; | |
093f38a2 | 2775 | struct fuse_args_pages ap = {}; |
59efec7b | 2776 | |
1baa26b2 MS |
2777 | #if BITS_PER_LONG == 32 |
2778 | inarg.flags |= FUSE_IOCTL_32BIT; | |
2779 | #else | |
6407f44a | 2780 | if (flags & FUSE_IOCTL_COMPAT) { |
1baa26b2 | 2781 | inarg.flags |= FUSE_IOCTL_32BIT; |
6407f44a IA |
2782 | #ifdef CONFIG_X86_X32 |
2783 | if (in_x32_syscall()) | |
2784 | inarg.flags |= FUSE_IOCTL_COMPAT_X32; | |
2785 | #endif | |
2786 | } | |
1baa26b2 MS |
2787 | #endif |
2788 | ||
59efec7b | 2789 | /* assume all the iovs returned by client always fits in a page */ |
1baa26b2 | 2790 | BUILD_BUG_ON(sizeof(struct fuse_ioctl_iovec) * FUSE_IOCTL_MAX_IOV > PAGE_SIZE); |
59efec7b | 2791 | |
59efec7b | 2792 | err = -ENOMEM; |
093f38a2 | 2793 | ap.pages = fuse_pages_alloc(fc->max_pages, GFP_KERNEL, &ap.descs); |
8ac83505 | 2794 | iov_page = (struct iovec *) __get_free_page(GFP_KERNEL); |
093f38a2 | 2795 | if (!ap.pages || !iov_page) |
59efec7b TH |
2796 | goto out; |
2797 | ||
093f38a2 MS |
2798 | fuse_page_descs_length_init(ap.descs, 0, fc->max_pages); |
2799 | ||
59efec7b TH |
2800 | /* |
2801 | * If restricted, initialize IO parameters as encoded in @cmd. | |
2802 | * RETRY from server is not allowed. | |
2803 | */ | |
2804 | if (!(flags & FUSE_IOCTL_UNRESTRICTED)) { | |
8ac83505 | 2805 | struct iovec *iov = iov_page; |
59efec7b | 2806 | |
c9f0523d | 2807 | iov->iov_base = (void __user *)arg; |
59efec7b TH |
2808 | iov->iov_len = _IOC_SIZE(cmd); |
2809 | ||
2810 | if (_IOC_DIR(cmd) & _IOC_WRITE) { | |
2811 | in_iov = iov; | |
2812 | in_iovs = 1; | |
2813 | } | |
2814 | ||
2815 | if (_IOC_DIR(cmd) & _IOC_READ) { | |
2816 | out_iov = iov; | |
2817 | out_iovs = 1; | |
2818 | } | |
2819 | } | |
2820 | ||
2821 | retry: | |
2822 | inarg.in_size = in_size = iov_length(in_iov, in_iovs); | |
2823 | inarg.out_size = out_size = iov_length(out_iov, out_iovs); | |
2824 | ||
2825 | /* | |
2826 | * Out data can be used either for actual out data or iovs, | |
2827 | * make sure there always is at least one page. | |
2828 | */ | |
2829 | out_size = max_t(size_t, out_size, PAGE_SIZE); | |
2830 | max_pages = DIV_ROUND_UP(max(in_size, out_size), PAGE_SIZE); | |
2831 | ||
2832 | /* make sure there are enough buffer pages and init request with them */ | |
2833 | err = -ENOMEM; | |
5da784cc | 2834 | if (max_pages > fc->max_pages) |
59efec7b | 2835 | goto out; |
093f38a2 MS |
2836 | while (ap.num_pages < max_pages) { |
2837 | ap.pages[ap.num_pages] = alloc_page(GFP_KERNEL | __GFP_HIGHMEM); | |
2838 | if (!ap.pages[ap.num_pages]) | |
59efec7b | 2839 | goto out; |
093f38a2 | 2840 | ap.num_pages++; |
59efec7b TH |
2841 | } |
2842 | ||
59efec7b TH |
2843 | |
2844 | /* okay, let's send it to the client */ | |
093f38a2 MS |
2845 | ap.args.opcode = FUSE_IOCTL; |
2846 | ap.args.nodeid = ff->nodeid; | |
2847 | ap.args.in_numargs = 1; | |
2848 | ap.args.in_args[0].size = sizeof(inarg); | |
2849 | ap.args.in_args[0].value = &inarg; | |
59efec7b | 2850 | if (in_size) { |
093f38a2 MS |
2851 | ap.args.in_numargs++; |
2852 | ap.args.in_args[1].size = in_size; | |
2853 | ap.args.in_pages = true; | |
59efec7b | 2854 | |
acbe5fda MS |
2855 | err = -EFAULT; |
2856 | iov_iter_init(&ii, WRITE, in_iov, in_iovs, in_size); | |
093f38a2 MS |
2857 | for (i = 0; iov_iter_count(&ii) && !WARN_ON(i >= ap.num_pages); i++) { |
2858 | c = copy_page_from_iter(ap.pages[i], 0, PAGE_SIZE, &ii); | |
acbe5fda MS |
2859 | if (c != PAGE_SIZE && iov_iter_count(&ii)) |
2860 | goto out; | |
2861 | } | |
59efec7b TH |
2862 | } |
2863 | ||
093f38a2 MS |
2864 | ap.args.out_numargs = 2; |
2865 | ap.args.out_args[0].size = sizeof(outarg); | |
2866 | ap.args.out_args[0].value = &outarg; | |
2867 | ap.args.out_args[1].size = out_size; | |
2868 | ap.args.out_pages = true; | |
2869 | ap.args.out_argvar = true; | |
59efec7b | 2870 | |
093f38a2 MS |
2871 | transferred = fuse_simple_request(fc, &ap.args); |
2872 | err = transferred; | |
2873 | if (transferred < 0) | |
59efec7b TH |
2874 | goto out; |
2875 | ||
2876 | /* did it ask for retry? */ | |
2877 | if (outarg.flags & FUSE_IOCTL_RETRY) { | |
8ac83505 | 2878 | void *vaddr; |
59efec7b TH |
2879 | |
2880 | /* no retry if in restricted mode */ | |
2881 | err = -EIO; | |
2882 | if (!(flags & FUSE_IOCTL_UNRESTRICTED)) | |
2883 | goto out; | |
2884 | ||
2885 | in_iovs = outarg.in_iovs; | |
2886 | out_iovs = outarg.out_iovs; | |
2887 | ||
2888 | /* | |
2889 | * Make sure things are in boundary, separate checks | |
2890 | * are to protect against overflow. | |
2891 | */ | |
2892 | err = -ENOMEM; | |
2893 | if (in_iovs > FUSE_IOCTL_MAX_IOV || | |
2894 | out_iovs > FUSE_IOCTL_MAX_IOV || | |
2895 | in_iovs + out_iovs > FUSE_IOCTL_MAX_IOV) | |
2896 | goto out; | |
2897 | ||
093f38a2 | 2898 | vaddr = kmap_atomic(ap.pages[0]); |
1baa26b2 | 2899 | err = fuse_copy_ioctl_iovec(fc, iov_page, vaddr, |
d9d318d3 MS |
2900 | transferred, in_iovs + out_iovs, |
2901 | (flags & FUSE_IOCTL_COMPAT) != 0); | |
2408f6ef | 2902 | kunmap_atomic(vaddr); |
d9d318d3 MS |
2903 | if (err) |
2904 | goto out; | |
59efec7b | 2905 | |
8ac83505 | 2906 | in_iov = iov_page; |
59efec7b TH |
2907 | out_iov = in_iov + in_iovs; |
2908 | ||
5da784cc | 2909 | err = fuse_verify_ioctl_iov(fc, in_iov, in_iovs); |
7572777e MS |
2910 | if (err) |
2911 | goto out; | |
2912 | ||
5da784cc | 2913 | err = fuse_verify_ioctl_iov(fc, out_iov, out_iovs); |
7572777e MS |
2914 | if (err) |
2915 | goto out; | |
2916 | ||
59efec7b TH |
2917 | goto retry; |
2918 | } | |
2919 | ||
2920 | err = -EIO; | |
2921 | if (transferred > inarg.out_size) | |
2922 | goto out; | |
2923 | ||
acbe5fda MS |
2924 | err = -EFAULT; |
2925 | iov_iter_init(&ii, READ, out_iov, out_iovs, transferred); | |
093f38a2 MS |
2926 | for (i = 0; iov_iter_count(&ii) && !WARN_ON(i >= ap.num_pages); i++) { |
2927 | c = copy_page_to_iter(ap.pages[i], 0, PAGE_SIZE, &ii); | |
acbe5fda MS |
2928 | if (c != PAGE_SIZE && iov_iter_count(&ii)) |
2929 | goto out; | |
2930 | } | |
2931 | err = 0; | |
59efec7b | 2932 | out: |
8ac83505 | 2933 | free_page((unsigned long) iov_page); |
093f38a2 MS |
2934 | while (ap.num_pages) |
2935 | __free_page(ap.pages[--ap.num_pages]); | |
2936 | kfree(ap.pages); | |
59efec7b TH |
2937 | |
2938 | return err ? err : outarg.result; | |
2939 | } | |
08cbf542 | 2940 | EXPORT_SYMBOL_GPL(fuse_do_ioctl); |
59efec7b | 2941 | |
b18da0c5 MS |
2942 | long fuse_ioctl_common(struct file *file, unsigned int cmd, |
2943 | unsigned long arg, unsigned int flags) | |
d36f2487 | 2944 | { |
6131ffaa | 2945 | struct inode *inode = file_inode(file); |
d36f2487 MS |
2946 | struct fuse_conn *fc = get_fuse_conn(inode); |
2947 | ||
c2132c1b | 2948 | if (!fuse_allow_current_process(fc)) |
d36f2487 MS |
2949 | return -EACCES; |
2950 | ||
2951 | if (is_bad_inode(inode)) | |
2952 | return -EIO; | |
2953 | ||
2954 | return fuse_do_ioctl(file, cmd, arg, flags); | |
2955 | } | |
2956 | ||
59efec7b TH |
2957 | static long fuse_file_ioctl(struct file *file, unsigned int cmd, |
2958 | unsigned long arg) | |
2959 | { | |
b18da0c5 | 2960 | return fuse_ioctl_common(file, cmd, arg, 0); |
59efec7b TH |
2961 | } |
2962 | ||
2963 | static long fuse_file_compat_ioctl(struct file *file, unsigned int cmd, | |
2964 | unsigned long arg) | |
2965 | { | |
b18da0c5 | 2966 | return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_COMPAT); |
59efec7b TH |
2967 | } |
2968 | ||
95668a69 TH |
2969 | /* |
2970 | * All files which have been polled are linked to RB tree | |
2971 | * fuse_conn->polled_files which is indexed by kh. Walk the tree and | |
2972 | * find the matching one. | |
2973 | */ | |
2974 | static struct rb_node **fuse_find_polled_node(struct fuse_conn *fc, u64 kh, | |
2975 | struct rb_node **parent_out) | |
2976 | { | |
2977 | struct rb_node **link = &fc->polled_files.rb_node; | |
2978 | struct rb_node *last = NULL; | |
2979 | ||
2980 | while (*link) { | |
2981 | struct fuse_file *ff; | |
2982 | ||
2983 | last = *link; | |
2984 | ff = rb_entry(last, struct fuse_file, polled_node); | |
2985 | ||
2986 | if (kh < ff->kh) | |
2987 | link = &last->rb_left; | |
2988 | else if (kh > ff->kh) | |
2989 | link = &last->rb_right; | |
2990 | else | |
2991 | return link; | |
2992 | } | |
2993 | ||
2994 | if (parent_out) | |
2995 | *parent_out = last; | |
2996 | return link; | |
2997 | } | |
2998 | ||
2999 | /* | |
3000 | * The file is about to be polled. Make sure it's on the polled_files | |
3001 | * RB tree. Note that files once added to the polled_files tree are | |
3002 | * not removed before the file is released. This is because a file | |
3003 | * polled once is likely to be polled again. | |
3004 | */ | |
3005 | static void fuse_register_polled_file(struct fuse_conn *fc, | |
3006 | struct fuse_file *ff) | |
3007 | { | |
3008 | spin_lock(&fc->lock); | |
3009 | if (RB_EMPTY_NODE(&ff->polled_node)) { | |
f3846266 | 3010 | struct rb_node **link, *uninitialized_var(parent); |
95668a69 TH |
3011 | |
3012 | link = fuse_find_polled_node(fc, ff->kh, &parent); | |
3013 | BUG_ON(*link); | |
3014 | rb_link_node(&ff->polled_node, parent, link); | |
3015 | rb_insert_color(&ff->polled_node, &fc->polled_files); | |
3016 | } | |
3017 | spin_unlock(&fc->lock); | |
3018 | } | |
3019 | ||
076ccb76 | 3020 | __poll_t fuse_file_poll(struct file *file, poll_table *wait) |
95668a69 | 3021 | { |
95668a69 | 3022 | struct fuse_file *ff = file->private_data; |
797759aa | 3023 | struct fuse_conn *fc = ff->fc; |
95668a69 TH |
3024 | struct fuse_poll_in inarg = { .fh = ff->fh, .kh = ff->kh }; |
3025 | struct fuse_poll_out outarg; | |
7078187a | 3026 | FUSE_ARGS(args); |
95668a69 TH |
3027 | int err; |
3028 | ||
3029 | if (fc->no_poll) | |
3030 | return DEFAULT_POLLMASK; | |
3031 | ||
3032 | poll_wait(file, &ff->poll_wait, wait); | |
c71d227f | 3033 | inarg.events = mangle_poll(poll_requested_events(wait)); |
95668a69 TH |
3034 | |
3035 | /* | |
3036 | * Ask for notification iff there's someone waiting for it. | |
3037 | * The client may ignore the flag and always notify. | |
3038 | */ | |
3039 | if (waitqueue_active(&ff->poll_wait)) { | |
3040 | inarg.flags |= FUSE_POLL_SCHEDULE_NOTIFY; | |
3041 | fuse_register_polled_file(fc, ff); | |
3042 | } | |
3043 | ||
d5b48543 MS |
3044 | args.opcode = FUSE_POLL; |
3045 | args.nodeid = ff->nodeid; | |
3046 | args.in_numargs = 1; | |
3047 | args.in_args[0].size = sizeof(inarg); | |
3048 | args.in_args[0].value = &inarg; | |
3049 | args.out_numargs = 1; | |
3050 | args.out_args[0].size = sizeof(outarg); | |
3051 | args.out_args[0].value = &outarg; | |
7078187a | 3052 | err = fuse_simple_request(fc, &args); |
95668a69 TH |
3053 | |
3054 | if (!err) | |
c71d227f | 3055 | return demangle_poll(outarg.revents); |
95668a69 TH |
3056 | if (err == -ENOSYS) { |
3057 | fc->no_poll = 1; | |
3058 | return DEFAULT_POLLMASK; | |
3059 | } | |
a9a08845 | 3060 | return EPOLLERR; |
95668a69 | 3061 | } |
08cbf542 | 3062 | EXPORT_SYMBOL_GPL(fuse_file_poll); |
95668a69 TH |
3063 | |
3064 | /* | |
3065 | * This is called from fuse_handle_notify() on FUSE_NOTIFY_POLL and | |
3066 | * wakes up the poll waiters. | |
3067 | */ | |
3068 | int fuse_notify_poll_wakeup(struct fuse_conn *fc, | |
3069 | struct fuse_notify_poll_wakeup_out *outarg) | |
3070 | { | |
3071 | u64 kh = outarg->kh; | |
3072 | struct rb_node **link; | |
3073 | ||
3074 | spin_lock(&fc->lock); | |
3075 | ||
3076 | link = fuse_find_polled_node(fc, kh, NULL); | |
3077 | if (*link) { | |
3078 | struct fuse_file *ff; | |
3079 | ||
3080 | ff = rb_entry(*link, struct fuse_file, polled_node); | |
3081 | wake_up_interruptible_sync(&ff->poll_wait); | |
3082 | } | |
3083 | ||
3084 | spin_unlock(&fc->lock); | |
3085 | return 0; | |
3086 | } | |
3087 | ||
efb9fa9e MP |
3088 | static void fuse_do_truncate(struct file *file) |
3089 | { | |
3090 | struct inode *inode = file->f_mapping->host; | |
3091 | struct iattr attr; | |
3092 | ||
3093 | attr.ia_valid = ATTR_SIZE; | |
3094 | attr.ia_size = i_size_read(inode); | |
3095 | ||
3096 | attr.ia_file = file; | |
3097 | attr.ia_valid |= ATTR_FILE; | |
3098 | ||
62490330 | 3099 | fuse_do_setattr(file_dentry(file), &attr, file); |
efb9fa9e MP |
3100 | } |
3101 | ||
5da784cc | 3102 | static inline loff_t fuse_round_up(struct fuse_conn *fc, loff_t off) |
e5c5f05d | 3103 | { |
5da784cc | 3104 | return round_up(off, fc->max_pages << PAGE_SHIFT); |
e5c5f05d MP |
3105 | } |
3106 | ||
4273b793 | 3107 | static ssize_t |
c8b8e32d | 3108 | fuse_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
4273b793 | 3109 | { |
9d5722b7 | 3110 | DECLARE_COMPLETION_ONSTACK(wait); |
4273b793 | 3111 | ssize_t ret = 0; |
60b9df7a MS |
3112 | struct file *file = iocb->ki_filp; |
3113 | struct fuse_file *ff = file->private_data; | |
e5c5f05d | 3114 | bool async_dio = ff->fc->async_dio; |
4273b793 | 3115 | loff_t pos = 0; |
bcba24cc MP |
3116 | struct inode *inode; |
3117 | loff_t i_size; | |
a6cbcd4a | 3118 | size_t count = iov_iter_count(iter); |
c8b8e32d | 3119 | loff_t offset = iocb->ki_pos; |
36cf66ed | 3120 | struct fuse_io_priv *io; |
4273b793 | 3121 | |
4273b793 | 3122 | pos = offset; |
bcba24cc MP |
3123 | inode = file->f_mapping->host; |
3124 | i_size = i_size_read(inode); | |
4273b793 | 3125 | |
6f673763 | 3126 | if ((iov_iter_rw(iter) == READ) && (offset > i_size)) |
9fe55eea SW |
3127 | return 0; |
3128 | ||
439ee5f0 | 3129 | /* optimization for short read */ |
6f673763 | 3130 | if (async_dio && iov_iter_rw(iter) != WRITE && offset + count > i_size) { |
439ee5f0 MP |
3131 | if (offset >= i_size) |
3132 | return 0; | |
5da784cc | 3133 | iov_iter_truncate(iter, fuse_round_up(ff->fc, i_size - offset)); |
6b775b18 | 3134 | count = iov_iter_count(iter); |
439ee5f0 MP |
3135 | } |
3136 | ||
bcba24cc | 3137 | io = kmalloc(sizeof(struct fuse_io_priv), GFP_KERNEL); |
36cf66ed MP |
3138 | if (!io) |
3139 | return -ENOMEM; | |
bcba24cc | 3140 | spin_lock_init(&io->lock); |
744742d6 | 3141 | kref_init(&io->refcnt); |
bcba24cc MP |
3142 | io->reqs = 1; |
3143 | io->bytes = -1; | |
3144 | io->size = 0; | |
3145 | io->offset = offset; | |
6f673763 | 3146 | io->write = (iov_iter_rw(iter) == WRITE); |
bcba24cc | 3147 | io->err = 0; |
bcba24cc MP |
3148 | /* |
3149 | * By default, we want to optimize all I/Os with async request | |
60b9df7a | 3150 | * submission to the client filesystem if supported. |
bcba24cc | 3151 | */ |
e5c5f05d | 3152 | io->async = async_dio; |
bcba24cc | 3153 | io->iocb = iocb; |
7879c4e5 | 3154 | io->blocking = is_sync_kiocb(iocb); |
bcba24cc MP |
3155 | |
3156 | /* | |
7879c4e5 AS |
3157 | * We cannot asynchronously extend the size of a file. |
3158 | * In such case the aio will behave exactly like sync io. | |
bcba24cc | 3159 | */ |
7879c4e5 AS |
3160 | if ((offset + count > i_size) && iov_iter_rw(iter) == WRITE) |
3161 | io->blocking = true; | |
4273b793 | 3162 | |
7879c4e5 | 3163 | if (io->async && io->blocking) { |
744742d6 SF |
3164 | /* |
3165 | * Additional reference to keep io around after | |
3166 | * calling fuse_aio_complete() | |
3167 | */ | |
3168 | kref_get(&io->refcnt); | |
9d5722b7 | 3169 | io->done = &wait; |
744742d6 | 3170 | } |
9d5722b7 | 3171 | |
6f673763 | 3172 | if (iov_iter_rw(iter) == WRITE) { |
6b775b18 | 3173 | ret = fuse_direct_io(io, iter, &pos, FUSE_DIO_WRITE); |
812408fb AV |
3174 | fuse_invalidate_attr(inode); |
3175 | } else { | |
d22a943f | 3176 | ret = __fuse_direct_read(io, iter, &pos); |
812408fb | 3177 | } |
36cf66ed | 3178 | |
bcba24cc | 3179 | if (io->async) { |
ebacb812 LC |
3180 | bool blocking = io->blocking; |
3181 | ||
bcba24cc MP |
3182 | fuse_aio_complete(io, ret < 0 ? ret : 0, -1); |
3183 | ||
3184 | /* we have a non-extending, async request, so return */ | |
ebacb812 | 3185 | if (!blocking) |
bcba24cc MP |
3186 | return -EIOCBQUEUED; |
3187 | ||
9d5722b7 CH |
3188 | wait_for_completion(&wait); |
3189 | ret = fuse_get_res_by_io(io); | |
bcba24cc MP |
3190 | } |
3191 | ||
744742d6 | 3192 | kref_put(&io->refcnt, fuse_io_release); |
9d5722b7 | 3193 | |
6f673763 | 3194 | if (iov_iter_rw(iter) == WRITE) { |
efb9fa9e MP |
3195 | if (ret > 0) |
3196 | fuse_write_update_size(inode, pos); | |
3197 | else if (ret < 0 && offset + count > i_size) | |
3198 | fuse_do_truncate(file); | |
3199 | } | |
4273b793 AA |
3200 | |
3201 | return ret; | |
3202 | } | |
3203 | ||
26eb3bae MS |
3204 | static int fuse_writeback_range(struct inode *inode, loff_t start, loff_t end) |
3205 | { | |
3206 | int err = filemap_write_and_wait_range(inode->i_mapping, start, end); | |
3207 | ||
3208 | if (!err) | |
3209 | fuse_sync_writes(inode); | |
3210 | ||
3211 | return err; | |
3212 | } | |
3213 | ||
cdadb11c MS |
3214 | static long fuse_file_fallocate(struct file *file, int mode, loff_t offset, |
3215 | loff_t length) | |
05ba1f08 AP |
3216 | { |
3217 | struct fuse_file *ff = file->private_data; | |
1c68271c | 3218 | struct inode *inode = file_inode(file); |
0ab08f57 | 3219 | struct fuse_inode *fi = get_fuse_inode(inode); |
05ba1f08 | 3220 | struct fuse_conn *fc = ff->fc; |
7078187a | 3221 | FUSE_ARGS(args); |
05ba1f08 AP |
3222 | struct fuse_fallocate_in inarg = { |
3223 | .fh = ff->fh, | |
3224 | .offset = offset, | |
3225 | .length = length, | |
3226 | .mode = mode | |
3227 | }; | |
3228 | int err; | |
14c14414 MP |
3229 | bool lock_inode = !(mode & FALLOC_FL_KEEP_SIZE) || |
3230 | (mode & FALLOC_FL_PUNCH_HOLE); | |
05ba1f08 | 3231 | |
4adb8302 MS |
3232 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
3233 | return -EOPNOTSUPP; | |
3234 | ||
519c6040 MS |
3235 | if (fc->no_fallocate) |
3236 | return -EOPNOTSUPP; | |
3237 | ||
14c14414 | 3238 | if (lock_inode) { |
5955102c | 3239 | inode_lock(inode); |
bde52788 MP |
3240 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
3241 | loff_t endbyte = offset + length - 1; | |
26eb3bae MS |
3242 | |
3243 | err = fuse_writeback_range(inode, offset, endbyte); | |
bde52788 MP |
3244 | if (err) |
3245 | goto out; | |
bde52788 | 3246 | } |
3634a632 BF |
3247 | } |
3248 | ||
0cbade02 LB |
3249 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
3250 | offset + length > i_size_read(inode)) { | |
3251 | err = inode_newsize_ok(inode, offset + length); | |
3252 | if (err) | |
35d6fcbb | 3253 | goto out; |
0cbade02 LB |
3254 | } |
3255 | ||
0ab08f57 MP |
3256 | if (!(mode & FALLOC_FL_KEEP_SIZE)) |
3257 | set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); | |
3258 | ||
d5b48543 MS |
3259 | args.opcode = FUSE_FALLOCATE; |
3260 | args.nodeid = ff->nodeid; | |
3261 | args.in_numargs = 1; | |
3262 | args.in_args[0].size = sizeof(inarg); | |
3263 | args.in_args[0].value = &inarg; | |
7078187a | 3264 | err = fuse_simple_request(fc, &args); |
519c6040 MS |
3265 | if (err == -ENOSYS) { |
3266 | fc->no_fallocate = 1; | |
3267 | err = -EOPNOTSUPP; | |
3268 | } | |
bee6c307 BF |
3269 | if (err) |
3270 | goto out; | |
3271 | ||
3272 | /* we could have extended the file */ | |
b0aa7606 MP |
3273 | if (!(mode & FALLOC_FL_KEEP_SIZE)) { |
3274 | bool changed = fuse_write_update_size(inode, offset + length); | |
3275 | ||
93d2269d MS |
3276 | if (changed && fc->writeback_cache) |
3277 | file_update_time(file); | |
b0aa7606 | 3278 | } |
bee6c307 BF |
3279 | |
3280 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
3281 | truncate_pagecache_range(inode, offset, offset + length - 1); | |
3282 | ||
3283 | fuse_invalidate_attr(inode); | |
3284 | ||
3634a632 | 3285 | out: |
0ab08f57 MP |
3286 | if (!(mode & FALLOC_FL_KEEP_SIZE)) |
3287 | clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); | |
3288 | ||
bde52788 | 3289 | if (lock_inode) |
5955102c | 3290 | inode_unlock(inode); |
3634a632 | 3291 | |
05ba1f08 AP |
3292 | return err; |
3293 | } | |
05ba1f08 | 3294 | |
64bf5ff5 DC |
3295 | static ssize_t __fuse_copy_file_range(struct file *file_in, loff_t pos_in, |
3296 | struct file *file_out, loff_t pos_out, | |
3297 | size_t len, unsigned int flags) | |
88bc7d50 NV |
3298 | { |
3299 | struct fuse_file *ff_in = file_in->private_data; | |
3300 | struct fuse_file *ff_out = file_out->private_data; | |
a2bc9236 | 3301 | struct inode *inode_in = file_inode(file_in); |
88bc7d50 NV |
3302 | struct inode *inode_out = file_inode(file_out); |
3303 | struct fuse_inode *fi_out = get_fuse_inode(inode_out); | |
3304 | struct fuse_conn *fc = ff_in->fc; | |
3305 | FUSE_ARGS(args); | |
3306 | struct fuse_copy_file_range_in inarg = { | |
3307 | .fh_in = ff_in->fh, | |
3308 | .off_in = pos_in, | |
3309 | .nodeid_out = ff_out->nodeid, | |
3310 | .fh_out = ff_out->fh, | |
3311 | .off_out = pos_out, | |
3312 | .len = len, | |
3313 | .flags = flags | |
3314 | }; | |
3315 | struct fuse_write_out outarg; | |
3316 | ssize_t err; | |
3317 | /* mark unstable when write-back is not used, and file_out gets | |
3318 | * extended */ | |
3319 | bool is_unstable = (!fc->writeback_cache) && | |
3320 | ((pos_out + len) > inode_out->i_size); | |
3321 | ||
3322 | if (fc->no_copy_file_range) | |
3323 | return -EOPNOTSUPP; | |
3324 | ||
5dae222a AG |
3325 | if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb) |
3326 | return -EXDEV; | |
3327 | ||
a2bc9236 MS |
3328 | if (fc->writeback_cache) { |
3329 | inode_lock(inode_in); | |
26eb3bae | 3330 | err = fuse_writeback_range(inode_in, pos_in, pos_in + len); |
a2bc9236 MS |
3331 | inode_unlock(inode_in); |
3332 | if (err) | |
3333 | return err; | |
3334 | } | |
3335 | ||
88bc7d50 NV |
3336 | inode_lock(inode_out); |
3337 | ||
fe0da9c0 AG |
3338 | err = file_modified(file_out); |
3339 | if (err) | |
3340 | goto out; | |
3341 | ||
88bc7d50 | 3342 | if (fc->writeback_cache) { |
26eb3bae | 3343 | err = fuse_writeback_range(inode_out, pos_out, pos_out + len); |
88bc7d50 NV |
3344 | if (err) |
3345 | goto out; | |
88bc7d50 NV |
3346 | } |
3347 | ||
3348 | if (is_unstable) | |
3349 | set_bit(FUSE_I_SIZE_UNSTABLE, &fi_out->state); | |
3350 | ||
d5b48543 MS |
3351 | args.opcode = FUSE_COPY_FILE_RANGE; |
3352 | args.nodeid = ff_in->nodeid; | |
3353 | args.in_numargs = 1; | |
3354 | args.in_args[0].size = sizeof(inarg); | |
3355 | args.in_args[0].value = &inarg; | |
3356 | args.out_numargs = 1; | |
3357 | args.out_args[0].size = sizeof(outarg); | |
3358 | args.out_args[0].value = &outarg; | |
88bc7d50 NV |
3359 | err = fuse_simple_request(fc, &args); |
3360 | if (err == -ENOSYS) { | |
3361 | fc->no_copy_file_range = 1; | |
3362 | err = -EOPNOTSUPP; | |
3363 | } | |
3364 | if (err) | |
3365 | goto out; | |
3366 | ||
3367 | if (fc->writeback_cache) { | |
3368 | fuse_write_update_size(inode_out, pos_out + outarg.size); | |
3369 | file_update_time(file_out); | |
3370 | } | |
3371 | ||
3372 | fuse_invalidate_attr(inode_out); | |
3373 | ||
3374 | err = outarg.size; | |
3375 | out: | |
3376 | if (is_unstable) | |
3377 | clear_bit(FUSE_I_SIZE_UNSTABLE, &fi_out->state); | |
3378 | ||
3379 | inode_unlock(inode_out); | |
fe0da9c0 | 3380 | file_accessed(file_in); |
88bc7d50 NV |
3381 | |
3382 | return err; | |
3383 | } | |
3384 | ||
64bf5ff5 DC |
3385 | static ssize_t fuse_copy_file_range(struct file *src_file, loff_t src_off, |
3386 | struct file *dst_file, loff_t dst_off, | |
3387 | size_t len, unsigned int flags) | |
3388 | { | |
3389 | ssize_t ret; | |
3390 | ||
3391 | ret = __fuse_copy_file_range(src_file, src_off, dst_file, dst_off, | |
3392 | len, flags); | |
3393 | ||
5dae222a | 3394 | if (ret == -EOPNOTSUPP || ret == -EXDEV) |
64bf5ff5 DC |
3395 | ret = generic_copy_file_range(src_file, src_off, dst_file, |
3396 | dst_off, len, flags); | |
3397 | return ret; | |
3398 | } | |
3399 | ||
4b6f5d20 | 3400 | static const struct file_operations fuse_file_operations = { |
5559b8f4 | 3401 | .llseek = fuse_file_llseek, |
37c20f16 | 3402 | .read_iter = fuse_file_read_iter, |
84c3d55c | 3403 | .write_iter = fuse_file_write_iter, |
b6aeaded MS |
3404 | .mmap = fuse_file_mmap, |
3405 | .open = fuse_open, | |
3406 | .flush = fuse_flush, | |
3407 | .release = fuse_release, | |
3408 | .fsync = fuse_fsync, | |
71421259 | 3409 | .lock = fuse_file_lock, |
a9ff4f87 | 3410 | .flock = fuse_file_flock, |
3c3db095 MS |
3411 | .splice_read = generic_file_splice_read, |
3412 | .splice_write = iter_file_splice_write, | |
59efec7b TH |
3413 | .unlocked_ioctl = fuse_file_ioctl, |
3414 | .compat_ioctl = fuse_file_compat_ioctl, | |
95668a69 | 3415 | .poll = fuse_file_poll, |
05ba1f08 | 3416 | .fallocate = fuse_file_fallocate, |
d4136d60 | 3417 | .copy_file_range = fuse_copy_file_range, |
413ef8cb MS |
3418 | }; |
3419 | ||
f5e54d6e | 3420 | static const struct address_space_operations fuse_file_aops = { |
b6aeaded | 3421 | .readpage = fuse_readpage, |
3be5a52b | 3422 | .writepage = fuse_writepage, |
26d614df | 3423 | .writepages = fuse_writepages, |
3be5a52b | 3424 | .launder_page = fuse_launder_page, |
db50b96c | 3425 | .readpages = fuse_readpages, |
3be5a52b | 3426 | .set_page_dirty = __set_page_dirty_nobuffers, |
b2d2272f | 3427 | .bmap = fuse_bmap, |
4273b793 | 3428 | .direct_IO = fuse_direct_IO, |
6b12c1b3 PE |
3429 | .write_begin = fuse_write_begin, |
3430 | .write_end = fuse_write_end, | |
b6aeaded MS |
3431 | }; |
3432 | ||
3433 | void fuse_init_file_inode(struct inode *inode) | |
3434 | { | |
ab2257e9 MS |
3435 | struct fuse_inode *fi = get_fuse_inode(inode); |
3436 | ||
45323fb7 MS |
3437 | inode->i_fop = &fuse_file_operations; |
3438 | inode->i_data.a_ops = &fuse_file_aops; | |
ab2257e9 MS |
3439 | |
3440 | INIT_LIST_HEAD(&fi->write_files); | |
3441 | INIT_LIST_HEAD(&fi->queued_writes); | |
3442 | fi->writectr = 0; | |
3443 | init_waitqueue_head(&fi->page_waitq); | |
6b2fb799 | 3444 | fi->writepages = RB_ROOT; |
b6aeaded | 3445 | } |