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> |
08cbf542 | 15 | #include <linux/module.h> |
d9d318d3 | 16 | #include <linux/compat.h> |
478e0841 | 17 | #include <linux/swap.h> |
3634a632 | 18 | #include <linux/falloc.h> |
e2e40f2c | 19 | #include <linux/uio.h> |
b6aeaded | 20 | |
4b6f5d20 | 21 | static const struct file_operations fuse_direct_io_file_operations; |
45323fb7 | 22 | |
91fe96b4 MS |
23 | static int fuse_send_open(struct fuse_conn *fc, u64 nodeid, struct file *file, |
24 | int opcode, struct fuse_open_out *outargp) | |
b6aeaded | 25 | { |
b6aeaded | 26 | struct fuse_open_in inarg; |
7078187a | 27 | FUSE_ARGS(args); |
fd72faac MS |
28 | |
29 | memset(&inarg, 0, sizeof(inarg)); | |
6ff958ed MS |
30 | inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY); |
31 | if (!fc->atomic_o_trunc) | |
32 | inarg.flags &= ~O_TRUNC; | |
7078187a MS |
33 | args.in.h.opcode = opcode; |
34 | args.in.h.nodeid = nodeid; | |
35 | args.in.numargs = 1; | |
36 | args.in.args[0].size = sizeof(inarg); | |
37 | args.in.args[0].value = &inarg; | |
38 | args.out.numargs = 1; | |
39 | args.out.args[0].size = sizeof(*outargp); | |
40 | args.out.args[0].value = outargp; | |
fd72faac | 41 | |
7078187a | 42 | return fuse_simple_request(fc, &args); |
fd72faac MS |
43 | } |
44 | ||
acf99433 | 45 | struct fuse_file *fuse_file_alloc(struct fuse_conn *fc) |
fd72faac MS |
46 | { |
47 | struct fuse_file *ff; | |
6b2db28a | 48 | |
fd72faac | 49 | ff = kmalloc(sizeof(struct fuse_file), GFP_KERNEL); |
6b2db28a TH |
50 | if (unlikely(!ff)) |
51 | return NULL; | |
52 | ||
da5e4714 | 53 | ff->fc = fc; |
4250c066 | 54 | ff->reserved_req = fuse_request_alloc(0); |
6b2db28a TH |
55 | if (unlikely(!ff->reserved_req)) { |
56 | kfree(ff); | |
57 | return NULL; | |
fd72faac | 58 | } |
6b2db28a TH |
59 | |
60 | INIT_LIST_HEAD(&ff->write_entry); | |
61 | atomic_set(&ff->count, 0); | |
62 | RB_CLEAR_NODE(&ff->polled_node); | |
63 | init_waitqueue_head(&ff->poll_wait); | |
64 | ||
65 | spin_lock(&fc->lock); | |
66 | ff->kh = ++fc->khctr; | |
67 | spin_unlock(&fc->lock); | |
68 | ||
fd72faac MS |
69 | return ff; |
70 | } | |
71 | ||
72 | void fuse_file_free(struct fuse_file *ff) | |
73 | { | |
33649c91 | 74 | fuse_request_free(ff->reserved_req); |
fd72faac MS |
75 | kfree(ff); |
76 | } | |
77 | ||
c7b7143c | 78 | struct fuse_file *fuse_file_get(struct fuse_file *ff) |
c756e0a4 MS |
79 | { |
80 | atomic_inc(&ff->count); | |
81 | return ff; | |
82 | } | |
83 | ||
819c4b3b MS |
84 | static void fuse_release_end(struct fuse_conn *fc, struct fuse_req *req) |
85 | { | |
baebccbe | 86 | iput(req->misc.release.inode); |
819c4b3b MS |
87 | } |
88 | ||
5a18ec17 | 89 | static void fuse_file_put(struct fuse_file *ff, bool sync) |
c756e0a4 MS |
90 | { |
91 | if (atomic_dec_and_test(&ff->count)) { | |
92 | struct fuse_req *req = ff->reserved_req; | |
8b0797a4 | 93 | |
7678ac50 AG |
94 | if (ff->fc->no_open) { |
95 | /* | |
96 | * Drop the release request when client does not | |
97 | * implement 'open' | |
98 | */ | |
825d6d33 | 99 | __clear_bit(FR_BACKGROUND, &req->flags); |
baebccbe | 100 | iput(req->misc.release.inode); |
7678ac50 AG |
101 | fuse_put_request(ff->fc, req); |
102 | } else if (sync) { | |
825d6d33 | 103 | __clear_bit(FR_BACKGROUND, &req->flags); |
5a18ec17 | 104 | fuse_request_send(ff->fc, req); |
baebccbe | 105 | iput(req->misc.release.inode); |
5a18ec17 MS |
106 | fuse_put_request(ff->fc, req); |
107 | } else { | |
108 | req->end = fuse_release_end; | |
825d6d33 | 109 | __set_bit(FR_BACKGROUND, &req->flags); |
5a18ec17 MS |
110 | fuse_request_send_background(ff->fc, req); |
111 | } | |
c756e0a4 MS |
112 | kfree(ff); |
113 | } | |
114 | } | |
115 | ||
08cbf542 TH |
116 | int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file, |
117 | bool isdir) | |
91fe96b4 | 118 | { |
91fe96b4 | 119 | struct fuse_file *ff; |
91fe96b4 MS |
120 | int opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN; |
121 | ||
122 | ff = fuse_file_alloc(fc); | |
123 | if (!ff) | |
124 | return -ENOMEM; | |
125 | ||
7678ac50 AG |
126 | ff->fh = 0; |
127 | ff->open_flags = FOPEN_KEEP_CACHE; /* Default for no-open */ | |
128 | if (!fc->no_open || isdir) { | |
129 | struct fuse_open_out outarg; | |
130 | int err; | |
131 | ||
132 | err = fuse_send_open(fc, nodeid, file, opcode, &outarg); | |
133 | if (!err) { | |
134 | ff->fh = outarg.fh; | |
135 | ff->open_flags = outarg.open_flags; | |
136 | ||
137 | } else if (err != -ENOSYS || isdir) { | |
138 | fuse_file_free(ff); | |
139 | return err; | |
140 | } else { | |
141 | fc->no_open = 1; | |
142 | } | |
91fe96b4 MS |
143 | } |
144 | ||
145 | if (isdir) | |
7678ac50 | 146 | ff->open_flags &= ~FOPEN_DIRECT_IO; |
91fe96b4 | 147 | |
91fe96b4 | 148 | ff->nodeid = nodeid; |
91fe96b4 MS |
149 | file->private_data = fuse_file_get(ff); |
150 | ||
151 | return 0; | |
152 | } | |
08cbf542 | 153 | EXPORT_SYMBOL_GPL(fuse_do_open); |
91fe96b4 | 154 | |
650b22b9 PE |
155 | static void fuse_link_write_file(struct file *file) |
156 | { | |
157 | struct inode *inode = file_inode(file); | |
158 | struct fuse_conn *fc = get_fuse_conn(inode); | |
159 | struct fuse_inode *fi = get_fuse_inode(inode); | |
160 | struct fuse_file *ff = file->private_data; | |
161 | /* | |
162 | * file may be written through mmap, so chain it onto the | |
163 | * inodes's write_file list | |
164 | */ | |
165 | spin_lock(&fc->lock); | |
166 | if (list_empty(&ff->write_entry)) | |
167 | list_add(&ff->write_entry, &fi->write_files); | |
168 | spin_unlock(&fc->lock); | |
169 | } | |
170 | ||
c7b7143c | 171 | void fuse_finish_open(struct inode *inode, struct file *file) |
fd72faac | 172 | { |
c7b7143c | 173 | struct fuse_file *ff = file->private_data; |
a0822c55 | 174 | struct fuse_conn *fc = get_fuse_conn(inode); |
c7b7143c MS |
175 | |
176 | if (ff->open_flags & FOPEN_DIRECT_IO) | |
fd72faac | 177 | file->f_op = &fuse_direct_io_file_operations; |
c7b7143c | 178 | if (!(ff->open_flags & FOPEN_KEEP_CACHE)) |
b1009979 | 179 | invalidate_inode_pages2(inode->i_mapping); |
c7b7143c | 180 | if (ff->open_flags & FOPEN_NONSEEKABLE) |
a7c1b990 | 181 | nonseekable_open(inode, file); |
a0822c55 KS |
182 | if (fc->atomic_o_trunc && (file->f_flags & O_TRUNC)) { |
183 | struct fuse_inode *fi = get_fuse_inode(inode); | |
184 | ||
185 | spin_lock(&fc->lock); | |
186 | fi->attr_version = ++fc->attr_version; | |
187 | i_size_write(inode, 0); | |
188 | spin_unlock(&fc->lock); | |
189 | fuse_invalidate_attr(inode); | |
75caeecd MP |
190 | if (fc->writeback_cache) |
191 | file_update_time(file); | |
a0822c55 | 192 | } |
4d99ff8f PE |
193 | if ((file->f_mode & FMODE_WRITE) && fc->writeback_cache) |
194 | fuse_link_write_file(file); | |
fd72faac MS |
195 | } |
196 | ||
91fe96b4 | 197 | int fuse_open_common(struct inode *inode, struct file *file, bool isdir) |
fd72faac | 198 | { |
acf99433 | 199 | struct fuse_conn *fc = get_fuse_conn(inode); |
b6aeaded | 200 | int err; |
75caeecd MP |
201 | bool lock_inode = (file->f_flags & O_TRUNC) && |
202 | fc->atomic_o_trunc && | |
203 | fc->writeback_cache; | |
b6aeaded MS |
204 | |
205 | err = generic_file_open(inode, file); | |
206 | if (err) | |
207 | return err; | |
208 | ||
75caeecd | 209 | if (lock_inode) |
5955102c | 210 | inode_lock(inode); |
75caeecd | 211 | |
91fe96b4 | 212 | err = fuse_do_open(fc, get_node_id(inode), file, isdir); |
b6aeaded | 213 | |
75caeecd MP |
214 | if (!err) |
215 | fuse_finish_open(inode, file); | |
91fe96b4 | 216 | |
75caeecd | 217 | if (lock_inode) |
5955102c | 218 | inode_unlock(inode); |
75caeecd MP |
219 | |
220 | return err; | |
b6aeaded MS |
221 | } |
222 | ||
8b0797a4 | 223 | static void fuse_prepare_release(struct fuse_file *ff, int flags, int opcode) |
64c6d8ed | 224 | { |
8b0797a4 | 225 | struct fuse_conn *fc = ff->fc; |
33649c91 | 226 | struct fuse_req *req = ff->reserved_req; |
b57d4264 | 227 | struct fuse_release_in *inarg = &req->misc.release.in; |
b6aeaded | 228 | |
8b0797a4 MS |
229 | spin_lock(&fc->lock); |
230 | list_del(&ff->write_entry); | |
231 | if (!RB_EMPTY_NODE(&ff->polled_node)) | |
232 | rb_erase(&ff->polled_node, &fc->polled_files); | |
233 | spin_unlock(&fc->lock); | |
234 | ||
357ccf2b | 235 | wake_up_interruptible_all(&ff->poll_wait); |
8b0797a4 | 236 | |
b6aeaded | 237 | inarg->fh = ff->fh; |
fd72faac | 238 | inarg->flags = flags; |
51eb01e7 | 239 | req->in.h.opcode = opcode; |
c7b7143c | 240 | req->in.h.nodeid = ff->nodeid; |
b6aeaded MS |
241 | req->in.numargs = 1; |
242 | req->in.args[0].size = sizeof(struct fuse_release_in); | |
243 | req->in.args[0].value = inarg; | |
fd72faac MS |
244 | } |
245 | ||
8b0797a4 | 246 | void fuse_release_common(struct file *file, int opcode) |
fd72faac | 247 | { |
6b2db28a TH |
248 | struct fuse_file *ff; |
249 | struct fuse_req *req; | |
b6aeaded | 250 | |
6b2db28a TH |
251 | ff = file->private_data; |
252 | if (unlikely(!ff)) | |
8b0797a4 | 253 | return; |
6b2db28a | 254 | |
6b2db28a | 255 | req = ff->reserved_req; |
8b0797a4 | 256 | fuse_prepare_release(ff, file->f_flags, opcode); |
6b2db28a | 257 | |
37fb3a30 MS |
258 | if (ff->flock) { |
259 | struct fuse_release_in *inarg = &req->misc.release.in; | |
260 | inarg->release_flags |= FUSE_RELEASE_FLOCK_UNLOCK; | |
261 | inarg->lock_owner = fuse_lock_owner_id(ff->fc, | |
262 | (fl_owner_t) file); | |
263 | } | |
baebccbe MS |
264 | /* Hold inode until release is finished */ |
265 | req->misc.release.inode = igrab(file_inode(file)); | |
6b2db28a | 266 | |
6b2db28a TH |
267 | /* |
268 | * Normally this will send the RELEASE request, however if | |
269 | * some asynchronous READ or WRITE requests are outstanding, | |
270 | * the sending will be delayed. | |
5a18ec17 MS |
271 | * |
272 | * Make the release synchronous if this is a fuseblk mount, | |
273 | * synchronous RELEASE is allowed (and desirable) in this case | |
274 | * because the server can be trusted not to screw up. | |
6b2db28a | 275 | */ |
5a18ec17 | 276 | fuse_file_put(ff, ff->fc->destroy_req != NULL); |
b6aeaded MS |
277 | } |
278 | ||
04730fef MS |
279 | static int fuse_open(struct inode *inode, struct file *file) |
280 | { | |
91fe96b4 | 281 | return fuse_open_common(inode, file, false); |
04730fef MS |
282 | } |
283 | ||
284 | static int fuse_release(struct inode *inode, struct file *file) | |
285 | { | |
e7cc133c PE |
286 | struct fuse_conn *fc = get_fuse_conn(inode); |
287 | ||
288 | /* see fuse_vma_close() for !writeback_cache case */ | |
289 | if (fc->writeback_cache) | |
1e18bda8 | 290 | write_inode_now(inode, 1); |
b0aa7606 | 291 | |
8b0797a4 MS |
292 | fuse_release_common(file, FUSE_RELEASE); |
293 | ||
294 | /* return value is ignored by VFS */ | |
295 | return 0; | |
296 | } | |
297 | ||
298 | void fuse_sync_release(struct fuse_file *ff, int flags) | |
299 | { | |
300 | WARN_ON(atomic_read(&ff->count) > 1); | |
301 | fuse_prepare_release(ff, flags, FUSE_RELEASE); | |
825d6d33 MS |
302 | __set_bit(FR_FORCE, &ff->reserved_req->flags); |
303 | __clear_bit(FR_BACKGROUND, &ff->reserved_req->flags); | |
8b0797a4 MS |
304 | fuse_request_send(ff->fc, ff->reserved_req); |
305 | fuse_put_request(ff->fc, ff->reserved_req); | |
306 | kfree(ff); | |
04730fef | 307 | } |
08cbf542 | 308 | EXPORT_SYMBOL_GPL(fuse_sync_release); |
04730fef | 309 | |
71421259 | 310 | /* |
9c8ef561 MS |
311 | * Scramble the ID space with XTEA, so that the value of the files_struct |
312 | * pointer is not exposed to userspace. | |
71421259 | 313 | */ |
f3332114 | 314 | u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id) |
71421259 | 315 | { |
9c8ef561 MS |
316 | u32 *k = fc->scramble_key; |
317 | u64 v = (unsigned long) id; | |
318 | u32 v0 = v; | |
319 | u32 v1 = v >> 32; | |
320 | u32 sum = 0; | |
321 | int i; | |
322 | ||
323 | for (i = 0; i < 32; i++) { | |
324 | v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]); | |
325 | sum += 0x9E3779B9; | |
326 | v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]); | |
327 | } | |
328 | ||
329 | return (u64) v0 + ((u64) v1 << 32); | |
71421259 MS |
330 | } |
331 | ||
3be5a52b | 332 | /* |
ea8cd333 | 333 | * Check if any page in a range is under writeback |
3be5a52b MS |
334 | * |
335 | * This is currently done by walking the list of writepage requests | |
336 | * for the inode, which can be pretty inefficient. | |
337 | */ | |
ea8cd333 PE |
338 | static bool fuse_range_is_writeback(struct inode *inode, pgoff_t idx_from, |
339 | pgoff_t idx_to) | |
3be5a52b MS |
340 | { |
341 | struct fuse_conn *fc = get_fuse_conn(inode); | |
342 | struct fuse_inode *fi = get_fuse_inode(inode); | |
343 | struct fuse_req *req; | |
344 | bool found = false; | |
345 | ||
346 | spin_lock(&fc->lock); | |
347 | list_for_each_entry(req, &fi->writepages, writepages_entry) { | |
348 | pgoff_t curr_index; | |
349 | ||
350 | BUG_ON(req->inode != inode); | |
09cbfeaf | 351 | curr_index = req->misc.write.in.offset >> PAGE_SHIFT; |
ea8cd333 PE |
352 | if (idx_from < curr_index + req->num_pages && |
353 | curr_index <= idx_to) { | |
3be5a52b MS |
354 | found = true; |
355 | break; | |
356 | } | |
357 | } | |
358 | spin_unlock(&fc->lock); | |
359 | ||
360 | return found; | |
361 | } | |
362 | ||
ea8cd333 PE |
363 | static inline bool fuse_page_is_writeback(struct inode *inode, pgoff_t index) |
364 | { | |
365 | return fuse_range_is_writeback(inode, index, index); | |
366 | } | |
367 | ||
3be5a52b MS |
368 | /* |
369 | * Wait for page writeback to be completed. | |
370 | * | |
371 | * Since fuse doesn't rely on the VM writeback tracking, this has to | |
372 | * use some other means. | |
373 | */ | |
374 | static int fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index) | |
375 | { | |
376 | struct fuse_inode *fi = get_fuse_inode(inode); | |
377 | ||
378 | wait_event(fi->page_waitq, !fuse_page_is_writeback(inode, index)); | |
379 | return 0; | |
380 | } | |
381 | ||
fe38d7df MP |
382 | /* |
383 | * Wait for all pending writepages on the inode to finish. | |
384 | * | |
385 | * This is currently done by blocking further writes with FUSE_NOWRITE | |
386 | * and waiting for all sent writes to complete. | |
387 | * | |
388 | * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage | |
389 | * could conflict with truncation. | |
390 | */ | |
391 | static void fuse_sync_writes(struct inode *inode) | |
392 | { | |
393 | fuse_set_nowrite(inode); | |
394 | fuse_release_nowrite(inode); | |
395 | } | |
396 | ||
75e1fcc0 | 397 | static int fuse_flush(struct file *file, fl_owner_t id) |
b6aeaded | 398 | { |
6131ffaa | 399 | struct inode *inode = file_inode(file); |
b6aeaded MS |
400 | struct fuse_conn *fc = get_fuse_conn(inode); |
401 | struct fuse_file *ff = file->private_data; | |
402 | struct fuse_req *req; | |
403 | struct fuse_flush_in inarg; | |
404 | int err; | |
405 | ||
248d86e8 MS |
406 | if (is_bad_inode(inode)) |
407 | return -EIO; | |
408 | ||
b6aeaded MS |
409 | if (fc->no_flush) |
410 | return 0; | |
411 | ||
1e18bda8 | 412 | err = write_inode_now(inode, 1); |
fe38d7df MP |
413 | if (err) |
414 | return err; | |
415 | ||
5955102c | 416 | inode_lock(inode); |
fe38d7df | 417 | fuse_sync_writes(inode); |
5955102c | 418 | inode_unlock(inode); |
fe38d7df | 419 | |
9ebce595 MP |
420 | if (test_bit(AS_ENOSPC, &file->f_mapping->flags) && |
421 | test_and_clear_bit(AS_ENOSPC, &file->f_mapping->flags)) | |
422 | err = -ENOSPC; | |
423 | if (test_bit(AS_EIO, &file->f_mapping->flags) && | |
424 | test_and_clear_bit(AS_EIO, &file->f_mapping->flags)) | |
425 | err = -EIO; | |
426 | if (err) | |
427 | return err; | |
428 | ||
b111c8c0 | 429 | req = fuse_get_req_nofail_nopages(fc, file); |
b6aeaded MS |
430 | memset(&inarg, 0, sizeof(inarg)); |
431 | inarg.fh = ff->fh; | |
9c8ef561 | 432 | inarg.lock_owner = fuse_lock_owner_id(fc, id); |
b6aeaded MS |
433 | req->in.h.opcode = FUSE_FLUSH; |
434 | req->in.h.nodeid = get_node_id(inode); | |
b6aeaded MS |
435 | req->in.numargs = 1; |
436 | req->in.args[0].size = sizeof(inarg); | |
437 | req->in.args[0].value = &inarg; | |
825d6d33 | 438 | __set_bit(FR_FORCE, &req->flags); |
b93f858a | 439 | fuse_request_send(fc, req); |
b6aeaded MS |
440 | err = req->out.h.error; |
441 | fuse_put_request(fc, req); | |
442 | if (err == -ENOSYS) { | |
443 | fc->no_flush = 1; | |
444 | err = 0; | |
445 | } | |
446 | return err; | |
447 | } | |
448 | ||
02c24a82 JB |
449 | int fuse_fsync_common(struct file *file, loff_t start, loff_t end, |
450 | int datasync, int isdir) | |
b6aeaded | 451 | { |
7ea80859 | 452 | struct inode *inode = file->f_mapping->host; |
b6aeaded MS |
453 | struct fuse_conn *fc = get_fuse_conn(inode); |
454 | struct fuse_file *ff = file->private_data; | |
7078187a | 455 | FUSE_ARGS(args); |
b6aeaded MS |
456 | struct fuse_fsync_in inarg; |
457 | int err; | |
458 | ||
248d86e8 MS |
459 | if (is_bad_inode(inode)) |
460 | return -EIO; | |
461 | ||
5955102c | 462 | inode_lock(inode); |
02c24a82 | 463 | |
3be5a52b MS |
464 | /* |
465 | * Start writeback against all dirty pages of the inode, then | |
466 | * wait for all outstanding writes, before sending the FSYNC | |
467 | * request. | |
468 | */ | |
22401e7b | 469 | err = filemap_write_and_wait_range(inode->i_mapping, start, end); |
3be5a52b | 470 | if (err) |
02c24a82 | 471 | goto out; |
3be5a52b MS |
472 | |
473 | fuse_sync_writes(inode); | |
ac7f052b AK |
474 | |
475 | /* | |
476 | * Due to implementation of fuse writeback | |
477 | * filemap_write_and_wait_range() does not catch errors. | |
478 | * We have to do this directly after fuse_sync_writes() | |
479 | */ | |
480 | if (test_bit(AS_ENOSPC, &file->f_mapping->flags) && | |
481 | test_and_clear_bit(AS_ENOSPC, &file->f_mapping->flags)) | |
482 | err = -ENOSPC; | |
483 | if (test_bit(AS_EIO, &file->f_mapping->flags) && | |
484 | test_and_clear_bit(AS_EIO, &file->f_mapping->flags)) | |
485 | err = -EIO; | |
486 | if (err) | |
487 | goto out; | |
488 | ||
1e18bda8 MS |
489 | err = sync_inode_metadata(inode, 1); |
490 | if (err) | |
491 | goto out; | |
3be5a52b | 492 | |
22401e7b MS |
493 | if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir)) |
494 | goto out; | |
b0aa7606 | 495 | |
b6aeaded MS |
496 | memset(&inarg, 0, sizeof(inarg)); |
497 | inarg.fh = ff->fh; | |
498 | inarg.fsync_flags = datasync ? 1 : 0; | |
7078187a MS |
499 | args.in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC; |
500 | args.in.h.nodeid = get_node_id(inode); | |
501 | args.in.numargs = 1; | |
502 | args.in.args[0].size = sizeof(inarg); | |
503 | args.in.args[0].value = &inarg; | |
504 | err = fuse_simple_request(fc, &args); | |
b6aeaded | 505 | if (err == -ENOSYS) { |
82547981 MS |
506 | if (isdir) |
507 | fc->no_fsyncdir = 1; | |
508 | else | |
509 | fc->no_fsync = 1; | |
b6aeaded MS |
510 | err = 0; |
511 | } | |
02c24a82 | 512 | out: |
5955102c | 513 | inode_unlock(inode); |
b6aeaded MS |
514 | return err; |
515 | } | |
516 | ||
02c24a82 JB |
517 | static int fuse_fsync(struct file *file, loff_t start, loff_t end, |
518 | int datasync) | |
82547981 | 519 | { |
02c24a82 | 520 | return fuse_fsync_common(file, start, end, datasync, 0); |
82547981 MS |
521 | } |
522 | ||
2106cb18 MS |
523 | void fuse_read_fill(struct fuse_req *req, struct file *file, loff_t pos, |
524 | size_t count, int opcode) | |
b6aeaded | 525 | { |
5c5c5e51 | 526 | struct fuse_read_in *inarg = &req->misc.read.in; |
a6643094 | 527 | struct fuse_file *ff = file->private_data; |
b6aeaded | 528 | |
361b1eb5 MS |
529 | inarg->fh = ff->fh; |
530 | inarg->offset = pos; | |
531 | inarg->size = count; | |
a6643094 | 532 | inarg->flags = file->f_flags; |
361b1eb5 | 533 | req->in.h.opcode = opcode; |
2106cb18 | 534 | req->in.h.nodeid = ff->nodeid; |
b6aeaded MS |
535 | req->in.numargs = 1; |
536 | req->in.args[0].size = sizeof(struct fuse_read_in); | |
c1aa96a5 | 537 | req->in.args[0].value = inarg; |
b6aeaded MS |
538 | req->out.argvar = 1; |
539 | req->out.numargs = 1; | |
540 | req->out.args[0].size = count; | |
b6aeaded MS |
541 | } |
542 | ||
187c5c36 MP |
543 | static void fuse_release_user_pages(struct fuse_req *req, int write) |
544 | { | |
545 | unsigned i; | |
546 | ||
547 | for (i = 0; i < req->num_pages; i++) { | |
548 | struct page *page = req->pages[i]; | |
549 | if (write) | |
550 | set_page_dirty_lock(page); | |
551 | put_page(page); | |
552 | } | |
553 | } | |
554 | ||
744742d6 SF |
555 | static void fuse_io_release(struct kref *kref) |
556 | { | |
557 | kfree(container_of(kref, struct fuse_io_priv, refcnt)); | |
558 | } | |
559 | ||
9d5722b7 CH |
560 | static ssize_t fuse_get_res_by_io(struct fuse_io_priv *io) |
561 | { | |
562 | if (io->err) | |
563 | return io->err; | |
564 | ||
565 | if (io->bytes >= 0 && io->write) | |
566 | return -EIO; | |
567 | ||
568 | return io->bytes < 0 ? io->size : io->bytes; | |
569 | } | |
570 | ||
01e9d11a MP |
571 | /** |
572 | * In case of short read, the caller sets 'pos' to the position of | |
573 | * actual end of fuse request in IO request. Otherwise, if bytes_requested | |
574 | * == bytes_transferred or rw == WRITE, the caller sets 'pos' to -1. | |
575 | * | |
576 | * An example: | |
577 | * User requested DIO read of 64K. It was splitted into two 32K fuse requests, | |
578 | * both submitted asynchronously. The first of them was ACKed by userspace as | |
579 | * fully completed (req->out.args[0].size == 32K) resulting in pos == -1. The | |
580 | * second request was ACKed as short, e.g. only 1K was read, resulting in | |
581 | * pos == 33K. | |
582 | * | |
583 | * Thus, when all fuse requests are completed, the minimal non-negative 'pos' | |
584 | * will be equal to the length of the longest contiguous fragment of | |
585 | * transferred data starting from the beginning of IO request. | |
586 | */ | |
587 | static void fuse_aio_complete(struct fuse_io_priv *io, int err, ssize_t pos) | |
588 | { | |
589 | int left; | |
590 | ||
591 | spin_lock(&io->lock); | |
592 | if (err) | |
593 | io->err = io->err ? : err; | |
594 | else if (pos >= 0 && (io->bytes < 0 || pos < io->bytes)) | |
595 | io->bytes = pos; | |
596 | ||
597 | left = --io->reqs; | |
7879c4e5 | 598 | if (!left && io->blocking) |
9d5722b7 | 599 | complete(io->done); |
01e9d11a MP |
600 | spin_unlock(&io->lock); |
601 | ||
7879c4e5 | 602 | if (!left && !io->blocking) { |
9d5722b7 | 603 | ssize_t res = fuse_get_res_by_io(io); |
01e9d11a | 604 | |
9d5722b7 CH |
605 | if (res >= 0) { |
606 | struct inode *inode = file_inode(io->iocb->ki_filp); | |
607 | struct fuse_conn *fc = get_fuse_conn(inode); | |
608 | struct fuse_inode *fi = get_fuse_inode(inode); | |
01e9d11a | 609 | |
9d5722b7 CH |
610 | spin_lock(&fc->lock); |
611 | fi->attr_version = ++fc->attr_version; | |
612 | spin_unlock(&fc->lock); | |
01e9d11a MP |
613 | } |
614 | ||
04b2fa9f | 615 | io->iocb->ki_complete(io->iocb, res, 0); |
01e9d11a | 616 | } |
744742d6 SF |
617 | |
618 | kref_put(&io->refcnt, fuse_io_release); | |
01e9d11a MP |
619 | } |
620 | ||
621 | static void fuse_aio_complete_req(struct fuse_conn *fc, struct fuse_req *req) | |
622 | { | |
623 | struct fuse_io_priv *io = req->io; | |
624 | ssize_t pos = -1; | |
625 | ||
626 | fuse_release_user_pages(req, !io->write); | |
627 | ||
628 | if (io->write) { | |
629 | if (req->misc.write.in.size != req->misc.write.out.size) | |
630 | pos = req->misc.write.in.offset - io->offset + | |
631 | req->misc.write.out.size; | |
632 | } else { | |
633 | if (req->misc.read.in.size != req->out.args[0].size) | |
634 | pos = req->misc.read.in.offset - io->offset + | |
635 | req->out.args[0].size; | |
636 | } | |
637 | ||
638 | fuse_aio_complete(io, req->out.h.error, pos); | |
639 | } | |
640 | ||
641 | static size_t fuse_async_req_send(struct fuse_conn *fc, struct fuse_req *req, | |
642 | size_t num_bytes, struct fuse_io_priv *io) | |
643 | { | |
644 | spin_lock(&io->lock); | |
744742d6 | 645 | kref_get(&io->refcnt); |
01e9d11a MP |
646 | io->size += num_bytes; |
647 | io->reqs++; | |
648 | spin_unlock(&io->lock); | |
649 | ||
650 | req->io = io; | |
651 | req->end = fuse_aio_complete_req; | |
652 | ||
36cf66ed | 653 | __fuse_get_request(req); |
01e9d11a MP |
654 | fuse_request_send_background(fc, req); |
655 | ||
656 | return num_bytes; | |
657 | } | |
658 | ||
36cf66ed | 659 | static size_t fuse_send_read(struct fuse_req *req, struct fuse_io_priv *io, |
2106cb18 | 660 | loff_t pos, size_t count, fl_owner_t owner) |
04730fef | 661 | { |
36cf66ed | 662 | struct file *file = io->file; |
2106cb18 MS |
663 | struct fuse_file *ff = file->private_data; |
664 | struct fuse_conn *fc = ff->fc; | |
f3332114 | 665 | |
2106cb18 | 666 | fuse_read_fill(req, file, pos, count, FUSE_READ); |
f3332114 | 667 | if (owner != NULL) { |
5c5c5e51 | 668 | struct fuse_read_in *inarg = &req->misc.read.in; |
f3332114 MS |
669 | |
670 | inarg->read_flags |= FUSE_READ_LOCKOWNER; | |
671 | inarg->lock_owner = fuse_lock_owner_id(fc, owner); | |
672 | } | |
36cf66ed MP |
673 | |
674 | if (io->async) | |
675 | return fuse_async_req_send(fc, req, count, io); | |
676 | ||
b93f858a | 677 | fuse_request_send(fc, req); |
361b1eb5 | 678 | return req->out.args[0].size; |
04730fef MS |
679 | } |
680 | ||
5c5c5e51 MS |
681 | static void fuse_read_update_size(struct inode *inode, loff_t size, |
682 | u64 attr_ver) | |
683 | { | |
684 | struct fuse_conn *fc = get_fuse_conn(inode); | |
685 | struct fuse_inode *fi = get_fuse_inode(inode); | |
686 | ||
687 | spin_lock(&fc->lock); | |
06a7c3c2 MP |
688 | if (attr_ver == fi->attr_version && size < inode->i_size && |
689 | !test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) { | |
5c5c5e51 MS |
690 | fi->attr_version = ++fc->attr_version; |
691 | i_size_write(inode, size); | |
692 | } | |
693 | spin_unlock(&fc->lock); | |
694 | } | |
695 | ||
a92adc82 PE |
696 | static void fuse_short_read(struct fuse_req *req, struct inode *inode, |
697 | u64 attr_ver) | |
698 | { | |
699 | size_t num_read = req->out.args[0].size; | |
8373200b PE |
700 | struct fuse_conn *fc = get_fuse_conn(inode); |
701 | ||
702 | if (fc->writeback_cache) { | |
703 | /* | |
704 | * A hole in a file. Some data after the hole are in page cache, | |
705 | * but have not reached the client fs yet. So, the hole is not | |
706 | * present there. | |
707 | */ | |
708 | int i; | |
09cbfeaf KS |
709 | int start_idx = num_read >> PAGE_SHIFT; |
710 | size_t off = num_read & (PAGE_SIZE - 1); | |
a92adc82 | 711 | |
8373200b | 712 | for (i = start_idx; i < req->num_pages; i++) { |
09cbfeaf | 713 | zero_user_segment(req->pages[i], off, PAGE_SIZE); |
8373200b PE |
714 | off = 0; |
715 | } | |
716 | } else { | |
717 | loff_t pos = page_offset(req->pages[0]) + num_read; | |
718 | fuse_read_update_size(inode, pos, attr_ver); | |
719 | } | |
a92adc82 PE |
720 | } |
721 | ||
482fce55 | 722 | static int fuse_do_readpage(struct file *file, struct page *page) |
b6aeaded | 723 | { |
744742d6 | 724 | struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(file); |
b6aeaded MS |
725 | struct inode *inode = page->mapping->host; |
726 | struct fuse_conn *fc = get_fuse_conn(inode); | |
248d86e8 | 727 | struct fuse_req *req; |
5c5c5e51 MS |
728 | size_t num_read; |
729 | loff_t pos = page_offset(page); | |
09cbfeaf | 730 | size_t count = PAGE_SIZE; |
5c5c5e51 | 731 | u64 attr_ver; |
248d86e8 MS |
732 | int err; |
733 | ||
3be5a52b | 734 | /* |
25985edc | 735 | * Page writeback can extend beyond the lifetime of the |
3be5a52b MS |
736 | * page-cache page, so make sure we read a properly synced |
737 | * page. | |
738 | */ | |
739 | fuse_wait_on_page_writeback(inode, page->index); | |
740 | ||
b111c8c0 | 741 | req = fuse_get_req(fc, 1); |
ce1d5a49 | 742 | if (IS_ERR(req)) |
482fce55 | 743 | return PTR_ERR(req); |
b6aeaded | 744 | |
5c5c5e51 MS |
745 | attr_ver = fuse_get_attr_version(fc); |
746 | ||
b6aeaded | 747 | req->out.page_zeroing = 1; |
f4975c67 | 748 | req->out.argpages = 1; |
b6aeaded MS |
749 | req->num_pages = 1; |
750 | req->pages[0] = page; | |
85f40aec | 751 | req->page_descs[0].length = count; |
36cf66ed | 752 | num_read = fuse_send_read(req, &io, pos, count, NULL); |
b6aeaded | 753 | err = req->out.h.error; |
5c5c5e51 MS |
754 | |
755 | if (!err) { | |
756 | /* | |
757 | * Short read means EOF. If file size is larger, truncate it | |
758 | */ | |
759 | if (num_read < count) | |
a92adc82 | 760 | fuse_short_read(req, inode, attr_ver); |
5c5c5e51 | 761 | |
b6aeaded | 762 | SetPageUptodate(page); |
5c5c5e51 MS |
763 | } |
764 | ||
a92adc82 | 765 | fuse_put_request(fc, req); |
482fce55 MP |
766 | |
767 | return err; | |
768 | } | |
769 | ||
770 | static int fuse_readpage(struct file *file, struct page *page) | |
771 | { | |
772 | struct inode *inode = page->mapping->host; | |
773 | int err; | |
774 | ||
775 | err = -EIO; | |
776 | if (is_bad_inode(inode)) | |
777 | goto out; | |
778 | ||
779 | err = fuse_do_readpage(file, page); | |
451418fc | 780 | fuse_invalidate_atime(inode); |
b6aeaded MS |
781 | out: |
782 | unlock_page(page); | |
783 | return err; | |
784 | } | |
785 | ||
c1aa96a5 | 786 | static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req) |
db50b96c | 787 | { |
c1aa96a5 | 788 | int i; |
5c5c5e51 MS |
789 | size_t count = req->misc.read.in.size; |
790 | size_t num_read = req->out.args[0].size; | |
ce534fb0 | 791 | struct address_space *mapping = NULL; |
c1aa96a5 | 792 | |
ce534fb0 MS |
793 | for (i = 0; mapping == NULL && i < req->num_pages; i++) |
794 | mapping = req->pages[i]->mapping; | |
5c5c5e51 | 795 | |
ce534fb0 MS |
796 | if (mapping) { |
797 | struct inode *inode = mapping->host; | |
798 | ||
799 | /* | |
800 | * Short read means EOF. If file size is larger, truncate it | |
801 | */ | |
a92adc82 PE |
802 | if (!req->out.h.error && num_read < count) |
803 | fuse_short_read(req, inode, req->misc.read.attr_ver); | |
ce534fb0 | 804 | |
451418fc | 805 | fuse_invalidate_atime(inode); |
ce534fb0 | 806 | } |
c1aa96a5 | 807 | |
db50b96c MS |
808 | for (i = 0; i < req->num_pages; i++) { |
809 | struct page *page = req->pages[i]; | |
810 | if (!req->out.h.error) | |
811 | SetPageUptodate(page); | |
c1aa96a5 MS |
812 | else |
813 | SetPageError(page); | |
db50b96c | 814 | unlock_page(page); |
09cbfeaf | 815 | put_page(page); |
db50b96c | 816 | } |
c756e0a4 | 817 | if (req->ff) |
5a18ec17 | 818 | fuse_file_put(req->ff, false); |
c1aa96a5 MS |
819 | } |
820 | ||
2106cb18 | 821 | static void fuse_send_readpages(struct fuse_req *req, struct file *file) |
c1aa96a5 | 822 | { |
2106cb18 MS |
823 | struct fuse_file *ff = file->private_data; |
824 | struct fuse_conn *fc = ff->fc; | |
c1aa96a5 | 825 | loff_t pos = page_offset(req->pages[0]); |
09cbfeaf | 826 | size_t count = req->num_pages << PAGE_SHIFT; |
f4975c67 MS |
827 | |
828 | req->out.argpages = 1; | |
c1aa96a5 | 829 | req->out.page_zeroing = 1; |
ce534fb0 | 830 | req->out.page_replace = 1; |
2106cb18 | 831 | fuse_read_fill(req, file, pos, count, FUSE_READ); |
5c5c5e51 | 832 | req->misc.read.attr_ver = fuse_get_attr_version(fc); |
9cd68455 | 833 | if (fc->async_read) { |
c756e0a4 | 834 | req->ff = fuse_file_get(ff); |
9cd68455 | 835 | req->end = fuse_readpages_end; |
b93f858a | 836 | fuse_request_send_background(fc, req); |
9cd68455 | 837 | } else { |
b93f858a | 838 | fuse_request_send(fc, req); |
9cd68455 | 839 | fuse_readpages_end(fc, req); |
e9bb09dd | 840 | fuse_put_request(fc, req); |
9cd68455 | 841 | } |
db50b96c MS |
842 | } |
843 | ||
c756e0a4 | 844 | struct fuse_fill_data { |
db50b96c | 845 | struct fuse_req *req; |
a6643094 | 846 | struct file *file; |
db50b96c | 847 | struct inode *inode; |
f8dbdf81 | 848 | unsigned nr_pages; |
db50b96c MS |
849 | }; |
850 | ||
851 | static int fuse_readpages_fill(void *_data, struct page *page) | |
852 | { | |
c756e0a4 | 853 | struct fuse_fill_data *data = _data; |
db50b96c MS |
854 | struct fuse_req *req = data->req; |
855 | struct inode *inode = data->inode; | |
856 | struct fuse_conn *fc = get_fuse_conn(inode); | |
857 | ||
3be5a52b MS |
858 | fuse_wait_on_page_writeback(inode, page->index); |
859 | ||
db50b96c MS |
860 | if (req->num_pages && |
861 | (req->num_pages == FUSE_MAX_PAGES_PER_REQ || | |
09cbfeaf | 862 | (req->num_pages + 1) * PAGE_SIZE > fc->max_read || |
db50b96c | 863 | req->pages[req->num_pages - 1]->index + 1 != page->index)) { |
f8dbdf81 MP |
864 | int nr_alloc = min_t(unsigned, data->nr_pages, |
865 | FUSE_MAX_PAGES_PER_REQ); | |
2106cb18 | 866 | fuse_send_readpages(req, data->file); |
8b41e671 MP |
867 | if (fc->async_read) |
868 | req = fuse_get_req_for_background(fc, nr_alloc); | |
869 | else | |
870 | req = fuse_get_req(fc, nr_alloc); | |
871 | ||
872 | data->req = req; | |
ce1d5a49 | 873 | if (IS_ERR(req)) { |
db50b96c | 874 | unlock_page(page); |
ce1d5a49 | 875 | return PTR_ERR(req); |
db50b96c | 876 | } |
db50b96c | 877 | } |
f8dbdf81 MP |
878 | |
879 | if (WARN_ON(req->num_pages >= req->max_pages)) { | |
880 | fuse_put_request(fc, req); | |
881 | return -EIO; | |
882 | } | |
883 | ||
09cbfeaf | 884 | get_page(page); |
db50b96c | 885 | req->pages[req->num_pages] = page; |
85f40aec | 886 | req->page_descs[req->num_pages].length = PAGE_SIZE; |
1729a16c | 887 | req->num_pages++; |
f8dbdf81 | 888 | data->nr_pages--; |
db50b96c MS |
889 | return 0; |
890 | } | |
891 | ||
892 | static int fuse_readpages(struct file *file, struct address_space *mapping, | |
893 | struct list_head *pages, unsigned nr_pages) | |
894 | { | |
895 | struct inode *inode = mapping->host; | |
896 | struct fuse_conn *fc = get_fuse_conn(inode); | |
c756e0a4 | 897 | struct fuse_fill_data data; |
db50b96c | 898 | int err; |
f8dbdf81 | 899 | int nr_alloc = min_t(unsigned, nr_pages, FUSE_MAX_PAGES_PER_REQ); |
248d86e8 | 900 | |
1d7ea732 | 901 | err = -EIO; |
248d86e8 | 902 | if (is_bad_inode(inode)) |
2e990021 | 903 | goto out; |
248d86e8 | 904 | |
a6643094 | 905 | data.file = file; |
db50b96c | 906 | data.inode = inode; |
8b41e671 MP |
907 | if (fc->async_read) |
908 | data.req = fuse_get_req_for_background(fc, nr_alloc); | |
909 | else | |
910 | data.req = fuse_get_req(fc, nr_alloc); | |
f8dbdf81 | 911 | data.nr_pages = nr_pages; |
1d7ea732 | 912 | err = PTR_ERR(data.req); |
ce1d5a49 | 913 | if (IS_ERR(data.req)) |
2e990021 | 914 | goto out; |
db50b96c MS |
915 | |
916 | err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data); | |
d3406ffa MS |
917 | if (!err) { |
918 | if (data.req->num_pages) | |
2106cb18 | 919 | fuse_send_readpages(data.req, file); |
d3406ffa MS |
920 | else |
921 | fuse_put_request(fc, data.req); | |
922 | } | |
2e990021 | 923 | out: |
1d7ea732 | 924 | return err; |
db50b96c MS |
925 | } |
926 | ||
37c20f16 | 927 | static ssize_t fuse_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
bcb4be80 MS |
928 | { |
929 | struct inode *inode = iocb->ki_filp->f_mapping->host; | |
a8894274 | 930 | struct fuse_conn *fc = get_fuse_conn(inode); |
bcb4be80 | 931 | |
a8894274 BF |
932 | /* |
933 | * In auto invalidate mode, always update attributes on read. | |
934 | * Otherwise, only update if we attempt to read past EOF (to ensure | |
935 | * i_size is up to date). | |
936 | */ | |
937 | if (fc->auto_inval_data || | |
37c20f16 | 938 | (iocb->ki_pos + iov_iter_count(to) > i_size_read(inode))) { |
bcb4be80 | 939 | int err; |
bcb4be80 MS |
940 | err = fuse_update_attributes(inode, NULL, iocb->ki_filp, NULL); |
941 | if (err) | |
942 | return err; | |
943 | } | |
944 | ||
37c20f16 | 945 | return generic_file_read_iter(iocb, to); |
bcb4be80 MS |
946 | } |
947 | ||
2d698b07 | 948 | static void fuse_write_fill(struct fuse_req *req, struct fuse_file *ff, |
2106cb18 | 949 | loff_t pos, size_t count) |
b6aeaded | 950 | { |
b25e82e5 MS |
951 | struct fuse_write_in *inarg = &req->misc.write.in; |
952 | struct fuse_write_out *outarg = &req->misc.write.out; | |
b6aeaded | 953 | |
b25e82e5 MS |
954 | inarg->fh = ff->fh; |
955 | inarg->offset = pos; | |
956 | inarg->size = count; | |
b6aeaded | 957 | req->in.h.opcode = FUSE_WRITE; |
2106cb18 | 958 | req->in.h.nodeid = ff->nodeid; |
b6aeaded | 959 | req->in.numargs = 2; |
2106cb18 | 960 | if (ff->fc->minor < 9) |
f3332114 MS |
961 | req->in.args[0].size = FUSE_COMPAT_WRITE_IN_SIZE; |
962 | else | |
963 | req->in.args[0].size = sizeof(struct fuse_write_in); | |
b25e82e5 | 964 | req->in.args[0].value = inarg; |
b6aeaded MS |
965 | req->in.args[1].size = count; |
966 | req->out.numargs = 1; | |
967 | req->out.args[0].size = sizeof(struct fuse_write_out); | |
b25e82e5 MS |
968 | req->out.args[0].value = outarg; |
969 | } | |
970 | ||
36cf66ed | 971 | static size_t fuse_send_write(struct fuse_req *req, struct fuse_io_priv *io, |
2106cb18 | 972 | loff_t pos, size_t count, fl_owner_t owner) |
b25e82e5 | 973 | { |
36cf66ed | 974 | struct file *file = io->file; |
2106cb18 MS |
975 | struct fuse_file *ff = file->private_data; |
976 | struct fuse_conn *fc = ff->fc; | |
2d698b07 MS |
977 | struct fuse_write_in *inarg = &req->misc.write.in; |
978 | ||
2106cb18 | 979 | fuse_write_fill(req, ff, pos, count); |
2d698b07 | 980 | inarg->flags = file->f_flags; |
f3332114 | 981 | if (owner != NULL) { |
f3332114 MS |
982 | inarg->write_flags |= FUSE_WRITE_LOCKOWNER; |
983 | inarg->lock_owner = fuse_lock_owner_id(fc, owner); | |
984 | } | |
36cf66ed MP |
985 | |
986 | if (io->async) | |
987 | return fuse_async_req_send(fc, req, count, io); | |
988 | ||
b93f858a | 989 | fuse_request_send(fc, req); |
b25e82e5 | 990 | return req->misc.write.out.size; |
b6aeaded MS |
991 | } |
992 | ||
b0aa7606 | 993 | bool fuse_write_update_size(struct inode *inode, loff_t pos) |
854512ec MS |
994 | { |
995 | struct fuse_conn *fc = get_fuse_conn(inode); | |
996 | struct fuse_inode *fi = get_fuse_inode(inode); | |
b0aa7606 | 997 | bool ret = false; |
854512ec MS |
998 | |
999 | spin_lock(&fc->lock); | |
1000 | fi->attr_version = ++fc->attr_version; | |
b0aa7606 | 1001 | if (pos > inode->i_size) { |
854512ec | 1002 | i_size_write(inode, pos); |
b0aa7606 MP |
1003 | ret = true; |
1004 | } | |
854512ec | 1005 | spin_unlock(&fc->lock); |
b0aa7606 MP |
1006 | |
1007 | return ret; | |
854512ec MS |
1008 | } |
1009 | ||
ea9b9907 NP |
1010 | static size_t fuse_send_write_pages(struct fuse_req *req, struct file *file, |
1011 | struct inode *inode, loff_t pos, | |
1012 | size_t count) | |
1013 | { | |
1014 | size_t res; | |
1015 | unsigned offset; | |
1016 | unsigned i; | |
744742d6 | 1017 | struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(file); |
ea9b9907 NP |
1018 | |
1019 | for (i = 0; i < req->num_pages; i++) | |
1020 | fuse_wait_on_page_writeback(inode, req->pages[i]->index); | |
1021 | ||
36cf66ed | 1022 | res = fuse_send_write(req, &io, pos, count, NULL); |
ea9b9907 | 1023 | |
b2430d75 | 1024 | offset = req->page_descs[0].offset; |
ea9b9907 NP |
1025 | count = res; |
1026 | for (i = 0; i < req->num_pages; i++) { | |
1027 | struct page *page = req->pages[i]; | |
1028 | ||
09cbfeaf | 1029 | if (!req->out.h.error && !offset && count >= PAGE_SIZE) |
ea9b9907 NP |
1030 | SetPageUptodate(page); |
1031 | ||
09cbfeaf KS |
1032 | if (count > PAGE_SIZE - offset) |
1033 | count -= PAGE_SIZE - offset; | |
ea9b9907 NP |
1034 | else |
1035 | count = 0; | |
1036 | offset = 0; | |
1037 | ||
1038 | unlock_page(page); | |
09cbfeaf | 1039 | put_page(page); |
ea9b9907 NP |
1040 | } |
1041 | ||
1042 | return res; | |
1043 | } | |
1044 | ||
1045 | static ssize_t fuse_fill_write_pages(struct fuse_req *req, | |
1046 | struct address_space *mapping, | |
1047 | struct iov_iter *ii, loff_t pos) | |
1048 | { | |
1049 | struct fuse_conn *fc = get_fuse_conn(mapping->host); | |
09cbfeaf | 1050 | unsigned offset = pos & (PAGE_SIZE - 1); |
ea9b9907 NP |
1051 | size_t count = 0; |
1052 | int err; | |
1053 | ||
f4975c67 | 1054 | req->in.argpages = 1; |
b2430d75 | 1055 | req->page_descs[0].offset = offset; |
ea9b9907 NP |
1056 | |
1057 | do { | |
1058 | size_t tmp; | |
1059 | struct page *page; | |
09cbfeaf KS |
1060 | pgoff_t index = pos >> PAGE_SHIFT; |
1061 | size_t bytes = min_t(size_t, PAGE_SIZE - offset, | |
ea9b9907 NP |
1062 | iov_iter_count(ii)); |
1063 | ||
1064 | bytes = min_t(size_t, bytes, fc->max_write - count); | |
1065 | ||
1066 | again: | |
1067 | err = -EFAULT; | |
1068 | if (iov_iter_fault_in_readable(ii, bytes)) | |
1069 | break; | |
1070 | ||
1071 | err = -ENOMEM; | |
54566b2c | 1072 | page = grab_cache_page_write_begin(mapping, index, 0); |
ea9b9907 NP |
1073 | if (!page) |
1074 | break; | |
1075 | ||
931e80e4 | 1076 | if (mapping_writably_mapped(mapping)) |
1077 | flush_dcache_page(page); | |
1078 | ||
ea9b9907 | 1079 | tmp = iov_iter_copy_from_user_atomic(page, ii, offset, bytes); |
ea9b9907 NP |
1080 | flush_dcache_page(page); |
1081 | ||
3ca8138f | 1082 | iov_iter_advance(ii, tmp); |
ea9b9907 NP |
1083 | if (!tmp) { |
1084 | unlock_page(page); | |
09cbfeaf | 1085 | put_page(page); |
ea9b9907 NP |
1086 | bytes = min(bytes, iov_iter_single_seg_count(ii)); |
1087 | goto again; | |
1088 | } | |
1089 | ||
1090 | err = 0; | |
1091 | req->pages[req->num_pages] = page; | |
85f40aec | 1092 | req->page_descs[req->num_pages].length = tmp; |
ea9b9907 NP |
1093 | req->num_pages++; |
1094 | ||
ea9b9907 NP |
1095 | count += tmp; |
1096 | pos += tmp; | |
1097 | offset += tmp; | |
09cbfeaf | 1098 | if (offset == PAGE_SIZE) |
ea9b9907 NP |
1099 | offset = 0; |
1100 | ||
78bb6cb9 MS |
1101 | if (!fc->big_writes) |
1102 | break; | |
ea9b9907 | 1103 | } while (iov_iter_count(ii) && count < fc->max_write && |
d07f09f5 | 1104 | req->num_pages < req->max_pages && offset == 0); |
ea9b9907 NP |
1105 | |
1106 | return count > 0 ? count : err; | |
1107 | } | |
1108 | ||
d07f09f5 MP |
1109 | static inline unsigned fuse_wr_pages(loff_t pos, size_t len) |
1110 | { | |
1111 | return min_t(unsigned, | |
09cbfeaf KS |
1112 | ((pos + len - 1) >> PAGE_SHIFT) - |
1113 | (pos >> PAGE_SHIFT) + 1, | |
d07f09f5 MP |
1114 | FUSE_MAX_PAGES_PER_REQ); |
1115 | } | |
1116 | ||
ea9b9907 NP |
1117 | static ssize_t fuse_perform_write(struct file *file, |
1118 | struct address_space *mapping, | |
1119 | struct iov_iter *ii, loff_t pos) | |
1120 | { | |
1121 | struct inode *inode = mapping->host; | |
1122 | struct fuse_conn *fc = get_fuse_conn(inode); | |
06a7c3c2 | 1123 | struct fuse_inode *fi = get_fuse_inode(inode); |
ea9b9907 NP |
1124 | int err = 0; |
1125 | ssize_t res = 0; | |
1126 | ||
1127 | if (is_bad_inode(inode)) | |
1128 | return -EIO; | |
1129 | ||
06a7c3c2 MP |
1130 | if (inode->i_size < pos + iov_iter_count(ii)) |
1131 | set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); | |
1132 | ||
ea9b9907 NP |
1133 | do { |
1134 | struct fuse_req *req; | |
1135 | ssize_t count; | |
d07f09f5 | 1136 | unsigned nr_pages = fuse_wr_pages(pos, iov_iter_count(ii)); |
ea9b9907 | 1137 | |
d07f09f5 | 1138 | req = fuse_get_req(fc, nr_pages); |
ea9b9907 NP |
1139 | if (IS_ERR(req)) { |
1140 | err = PTR_ERR(req); | |
1141 | break; | |
1142 | } | |
1143 | ||
1144 | count = fuse_fill_write_pages(req, mapping, ii, pos); | |
1145 | if (count <= 0) { | |
1146 | err = count; | |
1147 | } else { | |
1148 | size_t num_written; | |
1149 | ||
1150 | num_written = fuse_send_write_pages(req, file, inode, | |
1151 | pos, count); | |
1152 | err = req->out.h.error; | |
1153 | if (!err) { | |
1154 | res += num_written; | |
1155 | pos += num_written; | |
1156 | ||
1157 | /* break out of the loop on short write */ | |
1158 | if (num_written != count) | |
1159 | err = -EIO; | |
1160 | } | |
1161 | } | |
1162 | fuse_put_request(fc, req); | |
1163 | } while (!err && iov_iter_count(ii)); | |
1164 | ||
1165 | if (res > 0) | |
1166 | fuse_write_update_size(inode, pos); | |
1167 | ||
06a7c3c2 | 1168 | clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); |
ea9b9907 NP |
1169 | fuse_invalidate_attr(inode); |
1170 | ||
1171 | return res > 0 ? res : err; | |
1172 | } | |
1173 | ||
84c3d55c | 1174 | static ssize_t fuse_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
ea9b9907 NP |
1175 | { |
1176 | struct file *file = iocb->ki_filp; | |
1177 | struct address_space *mapping = file->f_mapping; | |
ea9b9907 | 1178 | ssize_t written = 0; |
4273b793 | 1179 | ssize_t written_buffered = 0; |
ea9b9907 NP |
1180 | struct inode *inode = mapping->host; |
1181 | ssize_t err; | |
4273b793 | 1182 | loff_t endbyte = 0; |
ea9b9907 | 1183 | |
4d99ff8f PE |
1184 | if (get_fuse_conn(inode)->writeback_cache) { |
1185 | /* Update size (EOF optimization) and mode (SUID clearing) */ | |
1186 | err = fuse_update_attributes(mapping->host, NULL, file, NULL); | |
1187 | if (err) | |
1188 | return err; | |
1189 | ||
84c3d55c | 1190 | return generic_file_write_iter(iocb, from); |
4d99ff8f PE |
1191 | } |
1192 | ||
5955102c | 1193 | inode_lock(inode); |
ea9b9907 NP |
1194 | |
1195 | /* We can write back this queue in page reclaim */ | |
de1414a6 | 1196 | current->backing_dev_info = inode_to_bdi(inode); |
ea9b9907 | 1197 | |
3309dd04 AV |
1198 | err = generic_write_checks(iocb, from); |
1199 | if (err <= 0) | |
ea9b9907 NP |
1200 | goto out; |
1201 | ||
5fa8e0a1 | 1202 | err = file_remove_privs(file); |
ea9b9907 NP |
1203 | if (err) |
1204 | goto out; | |
1205 | ||
c3b2da31 JB |
1206 | err = file_update_time(file); |
1207 | if (err) | |
1208 | goto out; | |
ea9b9907 | 1209 | |
2ba48ce5 | 1210 | if (iocb->ki_flags & IOCB_DIRECT) { |
3309dd04 | 1211 | loff_t pos = iocb->ki_pos; |
1af5bb49 | 1212 | written = generic_file_direct_write(iocb, from); |
84c3d55c | 1213 | if (written < 0 || !iov_iter_count(from)) |
4273b793 AA |
1214 | goto out; |
1215 | ||
1216 | pos += written; | |
ea9b9907 | 1217 | |
84c3d55c | 1218 | written_buffered = fuse_perform_write(file, mapping, from, pos); |
4273b793 AA |
1219 | if (written_buffered < 0) { |
1220 | err = written_buffered; | |
1221 | goto out; | |
1222 | } | |
1223 | endbyte = pos + written_buffered - 1; | |
1224 | ||
1225 | err = filemap_write_and_wait_range(file->f_mapping, pos, | |
1226 | endbyte); | |
1227 | if (err) | |
1228 | goto out; | |
1229 | ||
1230 | invalidate_mapping_pages(file->f_mapping, | |
09cbfeaf KS |
1231 | pos >> PAGE_SHIFT, |
1232 | endbyte >> PAGE_SHIFT); | |
4273b793 AA |
1233 | |
1234 | written += written_buffered; | |
1235 | iocb->ki_pos = pos + written_buffered; | |
1236 | } else { | |
3309dd04 | 1237 | written = fuse_perform_write(file, mapping, from, iocb->ki_pos); |
4273b793 | 1238 | if (written >= 0) |
3309dd04 | 1239 | iocb->ki_pos += written; |
4273b793 | 1240 | } |
ea9b9907 NP |
1241 | out: |
1242 | current->backing_dev_info = NULL; | |
5955102c | 1243 | inode_unlock(inode); |
ea9b9907 NP |
1244 | |
1245 | return written ? written : err; | |
1246 | } | |
1247 | ||
7c190c8b MP |
1248 | static inline void fuse_page_descs_length_init(struct fuse_req *req, |
1249 | unsigned index, unsigned nr_pages) | |
85f40aec MP |
1250 | { |
1251 | int i; | |
1252 | ||
7c190c8b | 1253 | for (i = index; i < index + nr_pages; i++) |
85f40aec MP |
1254 | req->page_descs[i].length = PAGE_SIZE - |
1255 | req->page_descs[i].offset; | |
1256 | } | |
1257 | ||
7c190c8b MP |
1258 | static inline unsigned long fuse_get_user_addr(const struct iov_iter *ii) |
1259 | { | |
1260 | return (unsigned long)ii->iov->iov_base + ii->iov_offset; | |
1261 | } | |
1262 | ||
1263 | static inline size_t fuse_get_frag_size(const struct iov_iter *ii, | |
1264 | size_t max_size) | |
1265 | { | |
1266 | return min(iov_iter_single_seg_count(ii), max_size); | |
1267 | } | |
1268 | ||
b98d023a | 1269 | static int fuse_get_user_pages(struct fuse_req *req, struct iov_iter *ii, |
ce60a2f1 | 1270 | size_t *nbytesp, int write) |
413ef8cb | 1271 | { |
7c190c8b | 1272 | size_t nbytes = 0; /* # bytes already packed in req */ |
742f9927 | 1273 | ssize_t ret = 0; |
b98d023a | 1274 | |
f4975c67 | 1275 | /* Special case for kernel I/O: can copy directly into the buffer */ |
62a8067a | 1276 | if (ii->type & ITER_KVEC) { |
7c190c8b MP |
1277 | unsigned long user_addr = fuse_get_user_addr(ii); |
1278 | size_t frag_size = fuse_get_frag_size(ii, *nbytesp); | |
1279 | ||
f4975c67 MS |
1280 | if (write) |
1281 | req->in.args[1].value = (void *) user_addr; | |
1282 | else | |
1283 | req->out.args[0].value = (void *) user_addr; | |
1284 | ||
b98d023a MP |
1285 | iov_iter_advance(ii, frag_size); |
1286 | *nbytesp = frag_size; | |
f4975c67 MS |
1287 | return 0; |
1288 | } | |
413ef8cb | 1289 | |
5565a9d8 | 1290 | while (nbytes < *nbytesp && req->num_pages < req->max_pages) { |
7c190c8b | 1291 | unsigned npages; |
f67da30c | 1292 | size_t start; |
742f9927 | 1293 | ret = iov_iter_get_pages(ii, &req->pages[req->num_pages], |
2c80929c | 1294 | *nbytesp - nbytes, |
c7f3888a AV |
1295 | req->max_pages - req->num_pages, |
1296 | &start); | |
7c190c8b | 1297 | if (ret < 0) |
742f9927 | 1298 | break; |
7c190c8b | 1299 | |
c9c37e2e AV |
1300 | iov_iter_advance(ii, ret); |
1301 | nbytes += ret; | |
7c190c8b | 1302 | |
c9c37e2e AV |
1303 | ret += start; |
1304 | npages = (ret + PAGE_SIZE - 1) / PAGE_SIZE; | |
7c190c8b | 1305 | |
c9c37e2e | 1306 | req->page_descs[req->num_pages].offset = start; |
7c190c8b MP |
1307 | fuse_page_descs_length_init(req, req->num_pages, npages); |
1308 | ||
1309 | req->num_pages += npages; | |
1310 | req->page_descs[req->num_pages - 1].length -= | |
c9c37e2e | 1311 | (PAGE_SIZE - ret) & (PAGE_SIZE - 1); |
7c190c8b | 1312 | } |
f4975c67 MS |
1313 | |
1314 | if (write) | |
1315 | req->in.argpages = 1; | |
1316 | else | |
1317 | req->out.argpages = 1; | |
1318 | ||
7c190c8b | 1319 | *nbytesp = nbytes; |
f4975c67 | 1320 | |
2c932d4c | 1321 | return ret < 0 ? ret : 0; |
413ef8cb MS |
1322 | } |
1323 | ||
5565a9d8 MP |
1324 | static inline int fuse_iter_npages(const struct iov_iter *ii_p) |
1325 | { | |
f67da30c | 1326 | return iov_iter_npages(ii_p, FUSE_MAX_PAGES_PER_REQ); |
5565a9d8 MP |
1327 | } |
1328 | ||
d22a943f AV |
1329 | ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter, |
1330 | loff_t *ppos, int flags) | |
413ef8cb | 1331 | { |
ea8cd333 PE |
1332 | int write = flags & FUSE_DIO_WRITE; |
1333 | int cuse = flags & FUSE_DIO_CUSE; | |
36cf66ed | 1334 | struct file *file = io->file; |
ea8cd333 | 1335 | struct inode *inode = file->f_mapping->host; |
2106cb18 MS |
1336 | struct fuse_file *ff = file->private_data; |
1337 | struct fuse_conn *fc = ff->fc; | |
413ef8cb MS |
1338 | size_t nmax = write ? fc->max_write : fc->max_read; |
1339 | loff_t pos = *ppos; | |
d22a943f | 1340 | size_t count = iov_iter_count(iter); |
09cbfeaf KS |
1341 | pgoff_t idx_from = pos >> PAGE_SHIFT; |
1342 | pgoff_t idx_to = (pos + count - 1) >> PAGE_SHIFT; | |
413ef8cb | 1343 | ssize_t res = 0; |
248d86e8 | 1344 | struct fuse_req *req; |
742f9927 | 1345 | int err = 0; |
248d86e8 | 1346 | |
de82b923 | 1347 | if (io->async) |
d22a943f | 1348 | req = fuse_get_req_for_background(fc, fuse_iter_npages(iter)); |
de82b923 | 1349 | else |
d22a943f | 1350 | req = fuse_get_req(fc, fuse_iter_npages(iter)); |
ce1d5a49 MS |
1351 | if (IS_ERR(req)) |
1352 | return PTR_ERR(req); | |
413ef8cb | 1353 | |
ea8cd333 PE |
1354 | if (!cuse && fuse_range_is_writeback(inode, idx_from, idx_to)) { |
1355 | if (!write) | |
5955102c | 1356 | inode_lock(inode); |
ea8cd333 PE |
1357 | fuse_sync_writes(inode); |
1358 | if (!write) | |
5955102c | 1359 | inode_unlock(inode); |
ea8cd333 PE |
1360 | } |
1361 | ||
413ef8cb | 1362 | while (count) { |
413ef8cb | 1363 | size_t nres; |
2106cb18 | 1364 | fl_owner_t owner = current->files; |
f4975c67 | 1365 | size_t nbytes = min(count, nmax); |
742f9927 AS |
1366 | err = fuse_get_user_pages(req, iter, &nbytes, write); |
1367 | if (err && !nbytes) | |
413ef8cb | 1368 | break; |
f4975c67 | 1369 | |
413ef8cb | 1370 | if (write) |
36cf66ed | 1371 | nres = fuse_send_write(req, io, pos, nbytes, owner); |
413ef8cb | 1372 | else |
36cf66ed | 1373 | nres = fuse_send_read(req, io, pos, nbytes, owner); |
2106cb18 | 1374 | |
36cf66ed MP |
1375 | if (!io->async) |
1376 | fuse_release_user_pages(req, !write); | |
413ef8cb | 1377 | if (req->out.h.error) { |
742f9927 | 1378 | err = req->out.h.error; |
413ef8cb MS |
1379 | break; |
1380 | } else if (nres > nbytes) { | |
742f9927 AS |
1381 | res = 0; |
1382 | err = -EIO; | |
413ef8cb MS |
1383 | break; |
1384 | } | |
1385 | count -= nres; | |
1386 | res += nres; | |
1387 | pos += nres; | |
413ef8cb MS |
1388 | if (nres != nbytes) |
1389 | break; | |
56cf34ff MS |
1390 | if (count) { |
1391 | fuse_put_request(fc, req); | |
de82b923 BF |
1392 | if (io->async) |
1393 | req = fuse_get_req_for_background(fc, | |
d22a943f | 1394 | fuse_iter_npages(iter)); |
de82b923 | 1395 | else |
d22a943f | 1396 | req = fuse_get_req(fc, fuse_iter_npages(iter)); |
56cf34ff MS |
1397 | if (IS_ERR(req)) |
1398 | break; | |
1399 | } | |
413ef8cb | 1400 | } |
f60311d5 AA |
1401 | if (!IS_ERR(req)) |
1402 | fuse_put_request(fc, req); | |
d09cb9d7 | 1403 | if (res > 0) |
413ef8cb | 1404 | *ppos = pos; |
413ef8cb | 1405 | |
742f9927 | 1406 | return res > 0 ? res : err; |
413ef8cb | 1407 | } |
08cbf542 | 1408 | EXPORT_SYMBOL_GPL(fuse_direct_io); |
413ef8cb | 1409 | |
36cf66ed | 1410 | static ssize_t __fuse_direct_read(struct fuse_io_priv *io, |
d22a943f AV |
1411 | struct iov_iter *iter, |
1412 | loff_t *ppos) | |
413ef8cb | 1413 | { |
d09cb9d7 | 1414 | ssize_t res; |
36cf66ed | 1415 | struct file *file = io->file; |
6131ffaa | 1416 | struct inode *inode = file_inode(file); |
d09cb9d7 MS |
1417 | |
1418 | if (is_bad_inode(inode)) | |
1419 | return -EIO; | |
1420 | ||
d22a943f | 1421 | res = fuse_direct_io(io, iter, ppos, 0); |
d09cb9d7 MS |
1422 | |
1423 | fuse_invalidate_attr(inode); | |
1424 | ||
1425 | return res; | |
413ef8cb MS |
1426 | } |
1427 | ||
15316263 | 1428 | static ssize_t fuse_direct_read_iter(struct kiocb *iocb, struct iov_iter *to) |
b98d023a | 1429 | { |
744742d6 | 1430 | struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb->ki_filp); |
15316263 | 1431 | return __fuse_direct_read(&io, to, &iocb->ki_pos); |
b98d023a MP |
1432 | } |
1433 | ||
15316263 | 1434 | static ssize_t fuse_direct_write_iter(struct kiocb *iocb, struct iov_iter *from) |
4273b793 | 1435 | { |
15316263 | 1436 | struct file *file = iocb->ki_filp; |
6131ffaa | 1437 | struct inode *inode = file_inode(file); |
744742d6 | 1438 | struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(file); |
15316263 | 1439 | ssize_t res; |
4273b793 AA |
1440 | |
1441 | if (is_bad_inode(inode)) | |
1442 | return -EIO; | |
1443 | ||
1444 | /* Don't allow parallel writes to the same file */ | |
5955102c | 1445 | inode_lock(inode); |
3309dd04 AV |
1446 | res = generic_write_checks(iocb, from); |
1447 | if (res > 0) | |
812408fb | 1448 | res = fuse_direct_io(&io, from, &iocb->ki_pos, FUSE_DIO_WRITE); |
812408fb | 1449 | fuse_invalidate_attr(inode); |
bcba24cc | 1450 | if (res > 0) |
15316263 | 1451 | fuse_write_update_size(inode, iocb->ki_pos); |
5955102c | 1452 | inode_unlock(inode); |
4273b793 AA |
1453 | |
1454 | return res; | |
1455 | } | |
1456 | ||
3be5a52b | 1457 | static void fuse_writepage_free(struct fuse_conn *fc, struct fuse_req *req) |
b6aeaded | 1458 | { |
385b1268 PE |
1459 | int i; |
1460 | ||
1461 | for (i = 0; i < req->num_pages; i++) | |
1462 | __free_page(req->pages[i]); | |
8b284dc4 MS |
1463 | |
1464 | if (req->ff) | |
1465 | fuse_file_put(req->ff, false); | |
3be5a52b MS |
1466 | } |
1467 | ||
1468 | static void fuse_writepage_finish(struct fuse_conn *fc, struct fuse_req *req) | |
1469 | { | |
1470 | struct inode *inode = req->inode; | |
1471 | struct fuse_inode *fi = get_fuse_inode(inode); | |
de1414a6 | 1472 | struct backing_dev_info *bdi = inode_to_bdi(inode); |
385b1268 | 1473 | int i; |
3be5a52b MS |
1474 | |
1475 | list_del(&req->writepages_entry); | |
385b1268 | 1476 | for (i = 0; i < req->num_pages; i++) { |
93f78d88 | 1477 | dec_wb_stat(&bdi->wb, WB_WRITEBACK); |
385b1268 | 1478 | dec_zone_page_state(req->pages[i], NR_WRITEBACK_TEMP); |
93f78d88 | 1479 | wb_writeout_inc(&bdi->wb); |
385b1268 | 1480 | } |
3be5a52b MS |
1481 | wake_up(&fi->page_waitq); |
1482 | } | |
1483 | ||
1484 | /* Called under fc->lock, may release and reacquire it */ | |
6eaf4782 MP |
1485 | static void fuse_send_writepage(struct fuse_conn *fc, struct fuse_req *req, |
1486 | loff_t size) | |
b9ca67b2 MS |
1487 | __releases(fc->lock) |
1488 | __acquires(fc->lock) | |
3be5a52b MS |
1489 | { |
1490 | struct fuse_inode *fi = get_fuse_inode(req->inode); | |
3be5a52b | 1491 | struct fuse_write_in *inarg = &req->misc.write.in; |
09cbfeaf | 1492 | __u64 data_size = req->num_pages * PAGE_SIZE; |
3be5a52b MS |
1493 | |
1494 | if (!fc->connected) | |
1495 | goto out_free; | |
1496 | ||
385b1268 PE |
1497 | if (inarg->offset + data_size <= size) { |
1498 | inarg->size = data_size; | |
3be5a52b | 1499 | } else if (inarg->offset < size) { |
385b1268 | 1500 | inarg->size = size - inarg->offset; |
3be5a52b MS |
1501 | } else { |
1502 | /* Got truncated off completely */ | |
1503 | goto out_free; | |
b6aeaded | 1504 | } |
3be5a52b MS |
1505 | |
1506 | req->in.args[1].size = inarg->size; | |
1507 | fi->writectr++; | |
b93f858a | 1508 | fuse_request_send_background_locked(fc, req); |
3be5a52b MS |
1509 | return; |
1510 | ||
1511 | out_free: | |
1512 | fuse_writepage_finish(fc, req); | |
1513 | spin_unlock(&fc->lock); | |
1514 | fuse_writepage_free(fc, req); | |
e9bb09dd | 1515 | fuse_put_request(fc, req); |
3be5a52b | 1516 | spin_lock(&fc->lock); |
b6aeaded MS |
1517 | } |
1518 | ||
3be5a52b MS |
1519 | /* |
1520 | * If fi->writectr is positive (no truncate or fsync going on) send | |
1521 | * all queued writepage requests. | |
1522 | * | |
1523 | * Called with fc->lock | |
1524 | */ | |
1525 | void fuse_flush_writepages(struct inode *inode) | |
b9ca67b2 MS |
1526 | __releases(fc->lock) |
1527 | __acquires(fc->lock) | |
b6aeaded | 1528 | { |
3be5a52b MS |
1529 | struct fuse_conn *fc = get_fuse_conn(inode); |
1530 | struct fuse_inode *fi = get_fuse_inode(inode); | |
6eaf4782 | 1531 | size_t crop = i_size_read(inode); |
3be5a52b MS |
1532 | struct fuse_req *req; |
1533 | ||
1534 | while (fi->writectr >= 0 && !list_empty(&fi->queued_writes)) { | |
1535 | req = list_entry(fi->queued_writes.next, struct fuse_req, list); | |
1536 | list_del_init(&req->list); | |
6eaf4782 | 1537 | fuse_send_writepage(fc, req, crop); |
3be5a52b MS |
1538 | } |
1539 | } | |
1540 | ||
1541 | static void fuse_writepage_end(struct fuse_conn *fc, struct fuse_req *req) | |
1542 | { | |
1543 | struct inode *inode = req->inode; | |
1544 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1545 | ||
1546 | mapping_set_error(inode->i_mapping, req->out.h.error); | |
1547 | spin_lock(&fc->lock); | |
8b284dc4 | 1548 | while (req->misc.write.next) { |
6eaf4782 MP |
1549 | struct fuse_conn *fc = get_fuse_conn(inode); |
1550 | struct fuse_write_in *inarg = &req->misc.write.in; | |
8b284dc4 MS |
1551 | struct fuse_req *next = req->misc.write.next; |
1552 | req->misc.write.next = next->misc.write.next; | |
1553 | next->misc.write.next = NULL; | |
ce128de6 | 1554 | next->ff = fuse_file_get(req->ff); |
8b284dc4 | 1555 | list_add(&next->writepages_entry, &fi->writepages); |
6eaf4782 MP |
1556 | |
1557 | /* | |
1558 | * Skip fuse_flush_writepages() to make it easy to crop requests | |
1559 | * based on primary request size. | |
1560 | * | |
1561 | * 1st case (trivial): there are no concurrent activities using | |
1562 | * fuse_set/release_nowrite. Then we're on safe side because | |
1563 | * fuse_flush_writepages() would call fuse_send_writepage() | |
1564 | * anyway. | |
1565 | * | |
1566 | * 2nd case: someone called fuse_set_nowrite and it is waiting | |
1567 | * now for completion of all in-flight requests. This happens | |
1568 | * rarely and no more than once per page, so this should be | |
1569 | * okay. | |
1570 | * | |
1571 | * 3rd case: someone (e.g. fuse_do_setattr()) is in the middle | |
1572 | * of fuse_set_nowrite..fuse_release_nowrite section. The fact | |
1573 | * that fuse_set_nowrite returned implies that all in-flight | |
1574 | * requests were completed along with all of their secondary | |
1575 | * requests. Further primary requests are blocked by negative | |
1576 | * writectr. Hence there cannot be any in-flight requests and | |
1577 | * no invocations of fuse_writepage_end() while we're in | |
1578 | * fuse_set_nowrite..fuse_release_nowrite section. | |
1579 | */ | |
1580 | fuse_send_writepage(fc, next, inarg->offset + inarg->size); | |
8b284dc4 | 1581 | } |
3be5a52b MS |
1582 | fi->writectr--; |
1583 | fuse_writepage_finish(fc, req); | |
1584 | spin_unlock(&fc->lock); | |
1585 | fuse_writepage_free(fc, req); | |
1586 | } | |
1587 | ||
1e18bda8 MS |
1588 | static struct fuse_file *__fuse_write_file_get(struct fuse_conn *fc, |
1589 | struct fuse_inode *fi) | |
adcadfa8 | 1590 | { |
72523425 | 1591 | struct fuse_file *ff = NULL; |
adcadfa8 PE |
1592 | |
1593 | spin_lock(&fc->lock); | |
1e18bda8 | 1594 | if (!list_empty(&fi->write_files)) { |
72523425 MS |
1595 | ff = list_entry(fi->write_files.next, struct fuse_file, |
1596 | write_entry); | |
1597 | fuse_file_get(ff); | |
1598 | } | |
adcadfa8 PE |
1599 | spin_unlock(&fc->lock); |
1600 | ||
1601 | return ff; | |
1602 | } | |
1603 | ||
1e18bda8 MS |
1604 | static struct fuse_file *fuse_write_file_get(struct fuse_conn *fc, |
1605 | struct fuse_inode *fi) | |
1606 | { | |
1607 | struct fuse_file *ff = __fuse_write_file_get(fc, fi); | |
1608 | WARN_ON(!ff); | |
1609 | return ff; | |
1610 | } | |
1611 | ||
1612 | int fuse_write_inode(struct inode *inode, struct writeback_control *wbc) | |
1613 | { | |
1614 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1615 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1616 | struct fuse_file *ff; | |
1617 | int err; | |
1618 | ||
1619 | ff = __fuse_write_file_get(fc, fi); | |
ab9e13f7 | 1620 | err = fuse_flush_times(inode, ff); |
1e18bda8 MS |
1621 | if (ff) |
1622 | fuse_file_put(ff, 0); | |
1623 | ||
1624 | return err; | |
1625 | } | |
1626 | ||
3be5a52b MS |
1627 | static int fuse_writepage_locked(struct page *page) |
1628 | { | |
1629 | struct address_space *mapping = page->mapping; | |
1630 | struct inode *inode = mapping->host; | |
1631 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1632 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1633 | struct fuse_req *req; | |
3be5a52b | 1634 | struct page *tmp_page; |
72523425 | 1635 | int error = -ENOMEM; |
3be5a52b MS |
1636 | |
1637 | set_page_writeback(page); | |
1638 | ||
4250c066 | 1639 | req = fuse_request_alloc_nofs(1); |
3be5a52b MS |
1640 | if (!req) |
1641 | goto err; | |
1642 | ||
825d6d33 MS |
1643 | /* writeback always goes to bg_queue */ |
1644 | __set_bit(FR_BACKGROUND, &req->flags); | |
3be5a52b MS |
1645 | tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); |
1646 | if (!tmp_page) | |
1647 | goto err_free; | |
1648 | ||
72523425 | 1649 | error = -EIO; |
26d614df | 1650 | req->ff = fuse_write_file_get(fc, fi); |
72523425 | 1651 | if (!req->ff) |
27f1b363 | 1652 | goto err_nofile; |
72523425 | 1653 | |
adcadfa8 | 1654 | fuse_write_fill(req, req->ff, page_offset(page), 0); |
3be5a52b MS |
1655 | |
1656 | copy_highpage(tmp_page, page); | |
2d698b07 | 1657 | req->misc.write.in.write_flags |= FUSE_WRITE_CACHE; |
8b284dc4 | 1658 | req->misc.write.next = NULL; |
f4975c67 | 1659 | req->in.argpages = 1; |
3be5a52b MS |
1660 | req->num_pages = 1; |
1661 | req->pages[0] = tmp_page; | |
b2430d75 | 1662 | req->page_descs[0].offset = 0; |
85f40aec | 1663 | req->page_descs[0].length = PAGE_SIZE; |
3be5a52b MS |
1664 | req->end = fuse_writepage_end; |
1665 | req->inode = inode; | |
1666 | ||
93f78d88 | 1667 | inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK); |
3be5a52b | 1668 | inc_zone_page_state(tmp_page, NR_WRITEBACK_TEMP); |
3be5a52b MS |
1669 | |
1670 | spin_lock(&fc->lock); | |
1671 | list_add(&req->writepages_entry, &fi->writepages); | |
1672 | list_add_tail(&req->list, &fi->queued_writes); | |
1673 | fuse_flush_writepages(inode); | |
1674 | spin_unlock(&fc->lock); | |
1675 | ||
4a4ac4eb MP |
1676 | end_page_writeback(page); |
1677 | ||
3be5a52b MS |
1678 | return 0; |
1679 | ||
27f1b363 MP |
1680 | err_nofile: |
1681 | __free_page(tmp_page); | |
3be5a52b MS |
1682 | err_free: |
1683 | fuse_request_free(req); | |
1684 | err: | |
1685 | end_page_writeback(page); | |
72523425 | 1686 | return error; |
3be5a52b MS |
1687 | } |
1688 | ||
1689 | static int fuse_writepage(struct page *page, struct writeback_control *wbc) | |
1690 | { | |
1691 | int err; | |
1692 | ||
ff17be08 MS |
1693 | if (fuse_page_is_writeback(page->mapping->host, page->index)) { |
1694 | /* | |
1695 | * ->writepages() should be called for sync() and friends. We | |
1696 | * should only get here on direct reclaim and then we are | |
1697 | * allowed to skip a page which is already in flight | |
1698 | */ | |
1699 | WARN_ON(wbc->sync_mode == WB_SYNC_ALL); | |
1700 | ||
1701 | redirty_page_for_writepage(wbc, page); | |
1702 | return 0; | |
1703 | } | |
1704 | ||
3be5a52b MS |
1705 | err = fuse_writepage_locked(page); |
1706 | unlock_page(page); | |
1707 | ||
1708 | return err; | |
1709 | } | |
1710 | ||
26d614df PE |
1711 | struct fuse_fill_wb_data { |
1712 | struct fuse_req *req; | |
1713 | struct fuse_file *ff; | |
1714 | struct inode *inode; | |
2d033eaa | 1715 | struct page **orig_pages; |
26d614df PE |
1716 | }; |
1717 | ||
1718 | static void fuse_writepages_send(struct fuse_fill_wb_data *data) | |
1719 | { | |
1720 | struct fuse_req *req = data->req; | |
1721 | struct inode *inode = data->inode; | |
1722 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1723 | struct fuse_inode *fi = get_fuse_inode(inode); | |
2d033eaa MP |
1724 | int num_pages = req->num_pages; |
1725 | int i; | |
26d614df PE |
1726 | |
1727 | req->ff = fuse_file_get(data->ff); | |
1728 | spin_lock(&fc->lock); | |
1729 | list_add_tail(&req->list, &fi->queued_writes); | |
1730 | fuse_flush_writepages(inode); | |
1731 | spin_unlock(&fc->lock); | |
2d033eaa MP |
1732 | |
1733 | for (i = 0; i < num_pages; i++) | |
1734 | end_page_writeback(data->orig_pages[i]); | |
26d614df PE |
1735 | } |
1736 | ||
8b284dc4 MS |
1737 | static bool fuse_writepage_in_flight(struct fuse_req *new_req, |
1738 | struct page *page) | |
1739 | { | |
1740 | struct fuse_conn *fc = get_fuse_conn(new_req->inode); | |
1741 | struct fuse_inode *fi = get_fuse_inode(new_req->inode); | |
1742 | struct fuse_req *tmp; | |
1743 | struct fuse_req *old_req; | |
1744 | bool found = false; | |
1745 | pgoff_t curr_index; | |
1746 | ||
1747 | BUG_ON(new_req->num_pages != 0); | |
1748 | ||
1749 | spin_lock(&fc->lock); | |
1750 | list_del(&new_req->writepages_entry); | |
8b284dc4 MS |
1751 | list_for_each_entry(old_req, &fi->writepages, writepages_entry) { |
1752 | BUG_ON(old_req->inode != new_req->inode); | |
09cbfeaf | 1753 | curr_index = old_req->misc.write.in.offset >> PAGE_SHIFT; |
8b284dc4 MS |
1754 | if (curr_index <= page->index && |
1755 | page->index < curr_index + old_req->num_pages) { | |
1756 | found = true; | |
1757 | break; | |
1758 | } | |
1759 | } | |
f6011081 MP |
1760 | if (!found) { |
1761 | list_add(&new_req->writepages_entry, &fi->writepages); | |
8b284dc4 | 1762 | goto out_unlock; |
f6011081 | 1763 | } |
8b284dc4 | 1764 | |
f6011081 | 1765 | new_req->num_pages = 1; |
8b284dc4 MS |
1766 | for (tmp = old_req; tmp != NULL; tmp = tmp->misc.write.next) { |
1767 | BUG_ON(tmp->inode != new_req->inode); | |
09cbfeaf | 1768 | curr_index = tmp->misc.write.in.offset >> PAGE_SHIFT; |
8b284dc4 MS |
1769 | if (tmp->num_pages == 1 && |
1770 | curr_index == page->index) { | |
1771 | old_req = tmp; | |
1772 | } | |
1773 | } | |
1774 | ||
33e14b4d | 1775 | if (old_req->num_pages == 1 && test_bit(FR_PENDING, &old_req->flags)) { |
de1414a6 | 1776 | struct backing_dev_info *bdi = inode_to_bdi(page->mapping->host); |
41b6e41f | 1777 | |
8b284dc4 MS |
1778 | copy_highpage(old_req->pages[0], page); |
1779 | spin_unlock(&fc->lock); | |
1780 | ||
93f78d88 | 1781 | dec_wb_stat(&bdi->wb, WB_WRITEBACK); |
8b284dc4 | 1782 | dec_zone_page_state(page, NR_WRITEBACK_TEMP); |
93f78d88 | 1783 | wb_writeout_inc(&bdi->wb); |
8b284dc4 MS |
1784 | fuse_writepage_free(fc, new_req); |
1785 | fuse_request_free(new_req); | |
1786 | goto out; | |
1787 | } else { | |
1788 | new_req->misc.write.next = old_req->misc.write.next; | |
1789 | old_req->misc.write.next = new_req; | |
1790 | } | |
1791 | out_unlock: | |
1792 | spin_unlock(&fc->lock); | |
1793 | out: | |
1794 | return found; | |
1795 | } | |
1796 | ||
26d614df PE |
1797 | static int fuse_writepages_fill(struct page *page, |
1798 | struct writeback_control *wbc, void *_data) | |
1799 | { | |
1800 | struct fuse_fill_wb_data *data = _data; | |
1801 | struct fuse_req *req = data->req; | |
1802 | struct inode *inode = data->inode; | |
1803 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1804 | struct page *tmp_page; | |
8b284dc4 | 1805 | bool is_writeback; |
26d614df PE |
1806 | int err; |
1807 | ||
1808 | if (!data->ff) { | |
1809 | err = -EIO; | |
1810 | data->ff = fuse_write_file_get(fc, get_fuse_inode(inode)); | |
1811 | if (!data->ff) | |
1812 | goto out_unlock; | |
1813 | } | |
1814 | ||
8b284dc4 MS |
1815 | /* |
1816 | * Being under writeback is unlikely but possible. For example direct | |
1817 | * read to an mmaped fuse file will set the page dirty twice; once when | |
1818 | * the pages are faulted with get_user_pages(), and then after the read | |
1819 | * completed. | |
1820 | */ | |
1821 | is_writeback = fuse_page_is_writeback(inode, page->index); | |
1822 | ||
1823 | if (req && req->num_pages && | |
1824 | (is_writeback || req->num_pages == FUSE_MAX_PAGES_PER_REQ || | |
09cbfeaf | 1825 | (req->num_pages + 1) * PAGE_SIZE > fc->max_write || |
8b284dc4 MS |
1826 | data->orig_pages[req->num_pages - 1]->index + 1 != page->index)) { |
1827 | fuse_writepages_send(data); | |
1828 | data->req = NULL; | |
26d614df PE |
1829 | } |
1830 | err = -ENOMEM; | |
1831 | tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
1832 | if (!tmp_page) | |
1833 | goto out_unlock; | |
1834 | ||
1835 | /* | |
1836 | * The page must not be redirtied until the writeout is completed | |
1837 | * (i.e. userspace has sent a reply to the write request). Otherwise | |
1838 | * there could be more than one temporary page instance for each real | |
1839 | * page. | |
1840 | * | |
1841 | * This is ensured by holding the page lock in page_mkwrite() while | |
1842 | * checking fuse_page_is_writeback(). We already hold the page lock | |
1843 | * since clear_page_dirty_for_io() and keep it held until we add the | |
1844 | * request to the fi->writepages list and increment req->num_pages. | |
1845 | * After this fuse_page_is_writeback() will indicate that the page is | |
1846 | * under writeback, so we can release the page lock. | |
1847 | */ | |
1848 | if (data->req == NULL) { | |
1849 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1850 | ||
1851 | err = -ENOMEM; | |
1852 | req = fuse_request_alloc_nofs(FUSE_MAX_PAGES_PER_REQ); | |
1853 | if (!req) { | |
1854 | __free_page(tmp_page); | |
1855 | goto out_unlock; | |
1856 | } | |
1857 | ||
1858 | fuse_write_fill(req, data->ff, page_offset(page), 0); | |
1859 | req->misc.write.in.write_flags |= FUSE_WRITE_CACHE; | |
8b284dc4 | 1860 | req->misc.write.next = NULL; |
26d614df | 1861 | req->in.argpages = 1; |
825d6d33 | 1862 | __set_bit(FR_BACKGROUND, &req->flags); |
26d614df PE |
1863 | req->num_pages = 0; |
1864 | req->end = fuse_writepage_end; | |
1865 | req->inode = inode; | |
1866 | ||
1867 | spin_lock(&fc->lock); | |
1868 | list_add(&req->writepages_entry, &fi->writepages); | |
1869 | spin_unlock(&fc->lock); | |
1870 | ||
1871 | data->req = req; | |
1872 | } | |
1873 | set_page_writeback(page); | |
1874 | ||
1875 | copy_highpage(tmp_page, page); | |
1876 | req->pages[req->num_pages] = tmp_page; | |
1877 | req->page_descs[req->num_pages].offset = 0; | |
1878 | req->page_descs[req->num_pages].length = PAGE_SIZE; | |
1879 | ||
93f78d88 | 1880 | inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK); |
26d614df | 1881 | inc_zone_page_state(tmp_page, NR_WRITEBACK_TEMP); |
8b284dc4 MS |
1882 | |
1883 | err = 0; | |
1884 | if (is_writeback && fuse_writepage_in_flight(req, page)) { | |
1885 | end_page_writeback(page); | |
1886 | data->req = NULL; | |
1887 | goto out_unlock; | |
1888 | } | |
2d033eaa | 1889 | data->orig_pages[req->num_pages] = page; |
26d614df PE |
1890 | |
1891 | /* | |
1892 | * Protected by fc->lock against concurrent access by | |
1893 | * fuse_page_is_writeback(). | |
1894 | */ | |
1895 | spin_lock(&fc->lock); | |
1896 | req->num_pages++; | |
1897 | spin_unlock(&fc->lock); | |
1898 | ||
26d614df PE |
1899 | out_unlock: |
1900 | unlock_page(page); | |
1901 | ||
1902 | return err; | |
1903 | } | |
1904 | ||
1905 | static int fuse_writepages(struct address_space *mapping, | |
1906 | struct writeback_control *wbc) | |
1907 | { | |
1908 | struct inode *inode = mapping->host; | |
1909 | struct fuse_fill_wb_data data; | |
1910 | int err; | |
1911 | ||
1912 | err = -EIO; | |
1913 | if (is_bad_inode(inode)) | |
1914 | goto out; | |
1915 | ||
1916 | data.inode = inode; | |
1917 | data.req = NULL; | |
1918 | data.ff = NULL; | |
1919 | ||
2d033eaa | 1920 | err = -ENOMEM; |
f2b3455e FF |
1921 | data.orig_pages = kcalloc(FUSE_MAX_PAGES_PER_REQ, |
1922 | sizeof(struct page *), | |
2d033eaa MP |
1923 | GFP_NOFS); |
1924 | if (!data.orig_pages) | |
1925 | goto out; | |
1926 | ||
26d614df PE |
1927 | err = write_cache_pages(mapping, wbc, fuse_writepages_fill, &data); |
1928 | if (data.req) { | |
1929 | /* Ignore errors if we can write at least one page */ | |
1930 | BUG_ON(!data.req->num_pages); | |
1931 | fuse_writepages_send(&data); | |
1932 | err = 0; | |
1933 | } | |
1934 | if (data.ff) | |
1935 | fuse_file_put(data.ff, false); | |
2d033eaa MP |
1936 | |
1937 | kfree(data.orig_pages); | |
26d614df PE |
1938 | out: |
1939 | return err; | |
1940 | } | |
1941 | ||
6b12c1b3 PE |
1942 | /* |
1943 | * It's worthy to make sure that space is reserved on disk for the write, | |
1944 | * but how to implement it without killing performance need more thinking. | |
1945 | */ | |
1946 | static int fuse_write_begin(struct file *file, struct address_space *mapping, | |
1947 | loff_t pos, unsigned len, unsigned flags, | |
1948 | struct page **pagep, void **fsdata) | |
1949 | { | |
09cbfeaf | 1950 | pgoff_t index = pos >> PAGE_SHIFT; |
a455589f | 1951 | struct fuse_conn *fc = get_fuse_conn(file_inode(file)); |
6b12c1b3 PE |
1952 | struct page *page; |
1953 | loff_t fsize; | |
1954 | int err = -ENOMEM; | |
1955 | ||
1956 | WARN_ON(!fc->writeback_cache); | |
1957 | ||
1958 | page = grab_cache_page_write_begin(mapping, index, flags); | |
1959 | if (!page) | |
1960 | goto error; | |
1961 | ||
1962 | fuse_wait_on_page_writeback(mapping->host, page->index); | |
1963 | ||
09cbfeaf | 1964 | if (PageUptodate(page) || len == PAGE_SIZE) |
6b12c1b3 PE |
1965 | goto success; |
1966 | /* | |
1967 | * Check if the start this page comes after the end of file, in which | |
1968 | * case the readpage can be optimized away. | |
1969 | */ | |
1970 | fsize = i_size_read(mapping->host); | |
09cbfeaf KS |
1971 | if (fsize <= (pos & PAGE_MASK)) { |
1972 | size_t off = pos & ~PAGE_MASK; | |
6b12c1b3 PE |
1973 | if (off) |
1974 | zero_user_segment(page, 0, off); | |
1975 | goto success; | |
1976 | } | |
1977 | err = fuse_do_readpage(file, page); | |
1978 | if (err) | |
1979 | goto cleanup; | |
1980 | success: | |
1981 | *pagep = page; | |
1982 | return 0; | |
1983 | ||
1984 | cleanup: | |
1985 | unlock_page(page); | |
09cbfeaf | 1986 | put_page(page); |
6b12c1b3 PE |
1987 | error: |
1988 | return err; | |
1989 | } | |
1990 | ||
1991 | static int fuse_write_end(struct file *file, struct address_space *mapping, | |
1992 | loff_t pos, unsigned len, unsigned copied, | |
1993 | struct page *page, void *fsdata) | |
1994 | { | |
1995 | struct inode *inode = page->mapping->host; | |
1996 | ||
1997 | if (!PageUptodate(page)) { | |
1998 | /* Zero any unwritten bytes at the end of the page */ | |
09cbfeaf | 1999 | size_t endoff = (pos + copied) & ~PAGE_MASK; |
6b12c1b3 | 2000 | if (endoff) |
09cbfeaf | 2001 | zero_user_segment(page, endoff, PAGE_SIZE); |
6b12c1b3 PE |
2002 | SetPageUptodate(page); |
2003 | } | |
2004 | ||
2005 | fuse_write_update_size(inode, pos + copied); | |
2006 | set_page_dirty(page); | |
2007 | unlock_page(page); | |
09cbfeaf | 2008 | put_page(page); |
6b12c1b3 PE |
2009 | |
2010 | return copied; | |
2011 | } | |
2012 | ||
3be5a52b MS |
2013 | static int fuse_launder_page(struct page *page) |
2014 | { | |
2015 | int err = 0; | |
2016 | if (clear_page_dirty_for_io(page)) { | |
2017 | struct inode *inode = page->mapping->host; | |
2018 | err = fuse_writepage_locked(page); | |
2019 | if (!err) | |
2020 | fuse_wait_on_page_writeback(inode, page->index); | |
2021 | } | |
2022 | return err; | |
2023 | } | |
2024 | ||
2025 | /* | |
2026 | * Write back dirty pages now, because there may not be any suitable | |
2027 | * open files later | |
2028 | */ | |
2029 | static void fuse_vma_close(struct vm_area_struct *vma) | |
2030 | { | |
2031 | filemap_write_and_wait(vma->vm_file->f_mapping); | |
2032 | } | |
2033 | ||
2034 | /* | |
2035 | * Wait for writeback against this page to complete before allowing it | |
2036 | * to be marked dirty again, and hence written back again, possibly | |
2037 | * before the previous writepage completed. | |
2038 | * | |
2039 | * Block here, instead of in ->writepage(), so that the userspace fs | |
2040 | * can only block processes actually operating on the filesystem. | |
2041 | * | |
2042 | * Otherwise unprivileged userspace fs would be able to block | |
2043 | * unrelated: | |
2044 | * | |
2045 | * - page migration | |
2046 | * - sync(2) | |
2047 | * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER | |
2048 | */ | |
c2ec175c | 2049 | static int fuse_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
3be5a52b | 2050 | { |
c2ec175c | 2051 | struct page *page = vmf->page; |
cca24370 MS |
2052 | struct inode *inode = file_inode(vma->vm_file); |
2053 | ||
2054 | file_update_time(vma->vm_file); | |
2055 | lock_page(page); | |
2056 | if (page->mapping != inode->i_mapping) { | |
2057 | unlock_page(page); | |
2058 | return VM_FAULT_NOPAGE; | |
2059 | } | |
3be5a52b MS |
2060 | |
2061 | fuse_wait_on_page_writeback(inode, page->index); | |
cca24370 | 2062 | return VM_FAULT_LOCKED; |
3be5a52b MS |
2063 | } |
2064 | ||
f0f37e2f | 2065 | static const struct vm_operations_struct fuse_file_vm_ops = { |
3be5a52b MS |
2066 | .close = fuse_vma_close, |
2067 | .fault = filemap_fault, | |
f1820361 | 2068 | .map_pages = filemap_map_pages, |
3be5a52b MS |
2069 | .page_mkwrite = fuse_page_mkwrite, |
2070 | }; | |
2071 | ||
2072 | static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma) | |
2073 | { | |
650b22b9 PE |
2074 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) |
2075 | fuse_link_write_file(file); | |
2076 | ||
3be5a52b MS |
2077 | file_accessed(file); |
2078 | vma->vm_ops = &fuse_file_vm_ops; | |
b6aeaded MS |
2079 | return 0; |
2080 | } | |
2081 | ||
fc280c96 MS |
2082 | static int fuse_direct_mmap(struct file *file, struct vm_area_struct *vma) |
2083 | { | |
2084 | /* Can't provide the coherency needed for MAP_SHARED */ | |
2085 | if (vma->vm_flags & VM_MAYSHARE) | |
2086 | return -ENODEV; | |
2087 | ||
3121bfe7 MS |
2088 | invalidate_inode_pages2(file->f_mapping); |
2089 | ||
fc280c96 MS |
2090 | return generic_file_mmap(file, vma); |
2091 | } | |
2092 | ||
71421259 MS |
2093 | static int convert_fuse_file_lock(const struct fuse_file_lock *ffl, |
2094 | struct file_lock *fl) | |
2095 | { | |
2096 | switch (ffl->type) { | |
2097 | case F_UNLCK: | |
2098 | break; | |
2099 | ||
2100 | case F_RDLCK: | |
2101 | case F_WRLCK: | |
2102 | if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX || | |
2103 | ffl->end < ffl->start) | |
2104 | return -EIO; | |
2105 | ||
2106 | fl->fl_start = ffl->start; | |
2107 | fl->fl_end = ffl->end; | |
2108 | fl->fl_pid = ffl->pid; | |
2109 | break; | |
2110 | ||
2111 | default: | |
2112 | return -EIO; | |
2113 | } | |
2114 | fl->fl_type = ffl->type; | |
2115 | return 0; | |
2116 | } | |
2117 | ||
7078187a | 2118 | static void fuse_lk_fill(struct fuse_args *args, struct file *file, |
a9ff4f87 | 2119 | const struct file_lock *fl, int opcode, pid_t pid, |
7078187a | 2120 | int flock, struct fuse_lk_in *inarg) |
71421259 | 2121 | { |
6131ffaa | 2122 | struct inode *inode = file_inode(file); |
9c8ef561 | 2123 | struct fuse_conn *fc = get_fuse_conn(inode); |
71421259 | 2124 | struct fuse_file *ff = file->private_data; |
7078187a MS |
2125 | |
2126 | memset(inarg, 0, sizeof(*inarg)); | |
2127 | inarg->fh = ff->fh; | |
2128 | inarg->owner = fuse_lock_owner_id(fc, fl->fl_owner); | |
2129 | inarg->lk.start = fl->fl_start; | |
2130 | inarg->lk.end = fl->fl_end; | |
2131 | inarg->lk.type = fl->fl_type; | |
2132 | inarg->lk.pid = pid; | |
a9ff4f87 | 2133 | if (flock) |
7078187a MS |
2134 | inarg->lk_flags |= FUSE_LK_FLOCK; |
2135 | args->in.h.opcode = opcode; | |
2136 | args->in.h.nodeid = get_node_id(inode); | |
2137 | args->in.numargs = 1; | |
2138 | args->in.args[0].size = sizeof(*inarg); | |
2139 | args->in.args[0].value = inarg; | |
71421259 MS |
2140 | } |
2141 | ||
2142 | static int fuse_getlk(struct file *file, struct file_lock *fl) | |
2143 | { | |
6131ffaa | 2144 | struct inode *inode = file_inode(file); |
71421259 | 2145 | struct fuse_conn *fc = get_fuse_conn(inode); |
7078187a MS |
2146 | FUSE_ARGS(args); |
2147 | struct fuse_lk_in inarg; | |
71421259 MS |
2148 | struct fuse_lk_out outarg; |
2149 | int err; | |
2150 | ||
7078187a MS |
2151 | fuse_lk_fill(&args, file, fl, FUSE_GETLK, 0, 0, &inarg); |
2152 | args.out.numargs = 1; | |
2153 | args.out.args[0].size = sizeof(outarg); | |
2154 | args.out.args[0].value = &outarg; | |
2155 | err = fuse_simple_request(fc, &args); | |
71421259 MS |
2156 | if (!err) |
2157 | err = convert_fuse_file_lock(&outarg.lk, fl); | |
2158 | ||
2159 | return err; | |
2160 | } | |
2161 | ||
a9ff4f87 | 2162 | static int fuse_setlk(struct file *file, struct file_lock *fl, int flock) |
71421259 | 2163 | { |
6131ffaa | 2164 | struct inode *inode = file_inode(file); |
71421259 | 2165 | struct fuse_conn *fc = get_fuse_conn(inode); |
7078187a MS |
2166 | FUSE_ARGS(args); |
2167 | struct fuse_lk_in inarg; | |
71421259 MS |
2168 | int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK; |
2169 | pid_t pid = fl->fl_type != F_UNLCK ? current->tgid : 0; | |
2170 | int err; | |
2171 | ||
8fb47a4f | 2172 | if (fl->fl_lmops && fl->fl_lmops->lm_grant) { |
48e90761 MS |
2173 | /* NLM needs asynchronous locks, which we don't support yet */ |
2174 | return -ENOLCK; | |
2175 | } | |
2176 | ||
71421259 MS |
2177 | /* Unlock on close is handled by the flush method */ |
2178 | if (fl->fl_flags & FL_CLOSE) | |
2179 | return 0; | |
2180 | ||
7078187a MS |
2181 | fuse_lk_fill(&args, file, fl, opcode, pid, flock, &inarg); |
2182 | err = fuse_simple_request(fc, &args); | |
71421259 | 2183 | |
a4d27e75 MS |
2184 | /* locking is restartable */ |
2185 | if (err == -EINTR) | |
2186 | err = -ERESTARTSYS; | |
7078187a | 2187 | |
71421259 MS |
2188 | return err; |
2189 | } | |
2190 | ||
2191 | static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl) | |
2192 | { | |
6131ffaa | 2193 | struct inode *inode = file_inode(file); |
71421259 MS |
2194 | struct fuse_conn *fc = get_fuse_conn(inode); |
2195 | int err; | |
2196 | ||
48e90761 MS |
2197 | if (cmd == F_CANCELLK) { |
2198 | err = 0; | |
2199 | } else if (cmd == F_GETLK) { | |
71421259 | 2200 | if (fc->no_lock) { |
9d6a8c5c | 2201 | posix_test_lock(file, fl); |
71421259 MS |
2202 | err = 0; |
2203 | } else | |
2204 | err = fuse_getlk(file, fl); | |
2205 | } else { | |
2206 | if (fc->no_lock) | |
48e90761 | 2207 | err = posix_lock_file(file, fl, NULL); |
71421259 | 2208 | else |
a9ff4f87 | 2209 | err = fuse_setlk(file, fl, 0); |
71421259 MS |
2210 | } |
2211 | return err; | |
2212 | } | |
2213 | ||
a9ff4f87 MS |
2214 | static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl) |
2215 | { | |
6131ffaa | 2216 | struct inode *inode = file_inode(file); |
a9ff4f87 MS |
2217 | struct fuse_conn *fc = get_fuse_conn(inode); |
2218 | int err; | |
2219 | ||
37fb3a30 | 2220 | if (fc->no_flock) { |
4f656367 | 2221 | err = locks_lock_file_wait(file, fl); |
a9ff4f87 | 2222 | } else { |
37fb3a30 MS |
2223 | struct fuse_file *ff = file->private_data; |
2224 | ||
a9ff4f87 | 2225 | /* emulate flock with POSIX locks */ |
37fb3a30 | 2226 | ff->flock = true; |
a9ff4f87 MS |
2227 | err = fuse_setlk(file, fl, 1); |
2228 | } | |
2229 | ||
2230 | return err; | |
2231 | } | |
2232 | ||
b2d2272f MS |
2233 | static sector_t fuse_bmap(struct address_space *mapping, sector_t block) |
2234 | { | |
2235 | struct inode *inode = mapping->host; | |
2236 | struct fuse_conn *fc = get_fuse_conn(inode); | |
7078187a | 2237 | FUSE_ARGS(args); |
b2d2272f MS |
2238 | struct fuse_bmap_in inarg; |
2239 | struct fuse_bmap_out outarg; | |
2240 | int err; | |
2241 | ||
2242 | if (!inode->i_sb->s_bdev || fc->no_bmap) | |
2243 | return 0; | |
2244 | ||
b2d2272f MS |
2245 | memset(&inarg, 0, sizeof(inarg)); |
2246 | inarg.block = block; | |
2247 | inarg.blocksize = inode->i_sb->s_blocksize; | |
7078187a MS |
2248 | args.in.h.opcode = FUSE_BMAP; |
2249 | args.in.h.nodeid = get_node_id(inode); | |
2250 | args.in.numargs = 1; | |
2251 | args.in.args[0].size = sizeof(inarg); | |
2252 | args.in.args[0].value = &inarg; | |
2253 | args.out.numargs = 1; | |
2254 | args.out.args[0].size = sizeof(outarg); | |
2255 | args.out.args[0].value = &outarg; | |
2256 | err = fuse_simple_request(fc, &args); | |
b2d2272f MS |
2257 | if (err == -ENOSYS) |
2258 | fc->no_bmap = 1; | |
2259 | ||
2260 | return err ? 0 : outarg.block; | |
2261 | } | |
2262 | ||
0b5da8db R |
2263 | static loff_t fuse_lseek(struct file *file, loff_t offset, int whence) |
2264 | { | |
2265 | struct inode *inode = file->f_mapping->host; | |
2266 | struct fuse_conn *fc = get_fuse_conn(inode); | |
2267 | struct fuse_file *ff = file->private_data; | |
2268 | FUSE_ARGS(args); | |
2269 | struct fuse_lseek_in inarg = { | |
2270 | .fh = ff->fh, | |
2271 | .offset = offset, | |
2272 | .whence = whence | |
2273 | }; | |
2274 | struct fuse_lseek_out outarg; | |
2275 | int err; | |
2276 | ||
2277 | if (fc->no_lseek) | |
2278 | goto fallback; | |
2279 | ||
2280 | args.in.h.opcode = FUSE_LSEEK; | |
2281 | args.in.h.nodeid = ff->nodeid; | |
2282 | args.in.numargs = 1; | |
2283 | args.in.args[0].size = sizeof(inarg); | |
2284 | args.in.args[0].value = &inarg; | |
2285 | args.out.numargs = 1; | |
2286 | args.out.args[0].size = sizeof(outarg); | |
2287 | args.out.args[0].value = &outarg; | |
2288 | err = fuse_simple_request(fc, &args); | |
2289 | if (err) { | |
2290 | if (err == -ENOSYS) { | |
2291 | fc->no_lseek = 1; | |
2292 | goto fallback; | |
2293 | } | |
2294 | return err; | |
2295 | } | |
2296 | ||
2297 | return vfs_setpos(file, outarg.offset, inode->i_sb->s_maxbytes); | |
2298 | ||
2299 | fallback: | |
2300 | err = fuse_update_attributes(inode, NULL, file, NULL); | |
2301 | if (!err) | |
2302 | return generic_file_llseek(file, offset, whence); | |
2303 | else | |
2304 | return err; | |
2305 | } | |
2306 | ||
965c8e59 | 2307 | static loff_t fuse_file_llseek(struct file *file, loff_t offset, int whence) |
5559b8f4 MS |
2308 | { |
2309 | loff_t retval; | |
6131ffaa | 2310 | struct inode *inode = file_inode(file); |
5559b8f4 | 2311 | |
0b5da8db R |
2312 | switch (whence) { |
2313 | case SEEK_SET: | |
2314 | case SEEK_CUR: | |
2315 | /* No i_mutex protection necessary for SEEK_CUR and SEEK_SET */ | |
965c8e59 | 2316 | retval = generic_file_llseek(file, offset, whence); |
0b5da8db R |
2317 | break; |
2318 | case SEEK_END: | |
5955102c | 2319 | inode_lock(inode); |
0b5da8db R |
2320 | retval = fuse_update_attributes(inode, NULL, file, NULL); |
2321 | if (!retval) | |
2322 | retval = generic_file_llseek(file, offset, whence); | |
5955102c | 2323 | inode_unlock(inode); |
0b5da8db R |
2324 | break; |
2325 | case SEEK_HOLE: | |
2326 | case SEEK_DATA: | |
5955102c | 2327 | inode_lock(inode); |
0b5da8db | 2328 | retval = fuse_lseek(file, offset, whence); |
5955102c | 2329 | inode_unlock(inode); |
0b5da8db R |
2330 | break; |
2331 | default: | |
2332 | retval = -EINVAL; | |
2333 | } | |
c07c3d19 | 2334 | |
5559b8f4 MS |
2335 | return retval; |
2336 | } | |
2337 | ||
59efec7b TH |
2338 | static int fuse_ioctl_copy_user(struct page **pages, struct iovec *iov, |
2339 | unsigned int nr_segs, size_t bytes, bool to_user) | |
2340 | { | |
2341 | struct iov_iter ii; | |
2342 | int page_idx = 0; | |
2343 | ||
2344 | if (!bytes) | |
2345 | return 0; | |
2346 | ||
71d8e532 | 2347 | iov_iter_init(&ii, to_user ? READ : WRITE, iov, nr_segs, bytes); |
59efec7b TH |
2348 | |
2349 | while (iov_iter_count(&ii)) { | |
2350 | struct page *page = pages[page_idx++]; | |
2351 | size_t todo = min_t(size_t, PAGE_SIZE, iov_iter_count(&ii)); | |
4aa0edd2 | 2352 | void *kaddr; |
59efec7b | 2353 | |
4aa0edd2 | 2354 | kaddr = kmap(page); |
59efec7b TH |
2355 | |
2356 | while (todo) { | |
2357 | char __user *uaddr = ii.iov->iov_base + ii.iov_offset; | |
2358 | size_t iov_len = ii.iov->iov_len - ii.iov_offset; | |
2359 | size_t copy = min(todo, iov_len); | |
2360 | size_t left; | |
2361 | ||
2362 | if (!to_user) | |
2363 | left = copy_from_user(kaddr, uaddr, copy); | |
2364 | else | |
2365 | left = copy_to_user(uaddr, kaddr, copy); | |
2366 | ||
2367 | if (unlikely(left)) | |
2368 | return -EFAULT; | |
2369 | ||
2370 | iov_iter_advance(&ii, copy); | |
2371 | todo -= copy; | |
2372 | kaddr += copy; | |
2373 | } | |
2374 | ||
0bd87182 | 2375 | kunmap(page); |
59efec7b TH |
2376 | } |
2377 | ||
2378 | return 0; | |
2379 | } | |
2380 | ||
d9d318d3 MS |
2381 | /* |
2382 | * CUSE servers compiled on 32bit broke on 64bit kernels because the | |
2383 | * ABI was defined to be 'struct iovec' which is different on 32bit | |
2384 | * and 64bit. Fortunately we can determine which structure the server | |
2385 | * used from the size of the reply. | |
2386 | */ | |
1baa26b2 MS |
2387 | static int fuse_copy_ioctl_iovec_old(struct iovec *dst, void *src, |
2388 | size_t transferred, unsigned count, | |
2389 | bool is_compat) | |
d9d318d3 MS |
2390 | { |
2391 | #ifdef CONFIG_COMPAT | |
2392 | if (count * sizeof(struct compat_iovec) == transferred) { | |
2393 | struct compat_iovec *ciov = src; | |
2394 | unsigned i; | |
2395 | ||
2396 | /* | |
2397 | * With this interface a 32bit server cannot support | |
2398 | * non-compat (i.e. ones coming from 64bit apps) ioctl | |
2399 | * requests | |
2400 | */ | |
2401 | if (!is_compat) | |
2402 | return -EINVAL; | |
2403 | ||
2404 | for (i = 0; i < count; i++) { | |
2405 | dst[i].iov_base = compat_ptr(ciov[i].iov_base); | |
2406 | dst[i].iov_len = ciov[i].iov_len; | |
2407 | } | |
2408 | return 0; | |
2409 | } | |
2410 | #endif | |
2411 | ||
2412 | if (count * sizeof(struct iovec) != transferred) | |
2413 | return -EIO; | |
2414 | ||
2415 | memcpy(dst, src, transferred); | |
2416 | return 0; | |
2417 | } | |
2418 | ||
7572777e MS |
2419 | /* Make sure iov_length() won't overflow */ |
2420 | static int fuse_verify_ioctl_iov(struct iovec *iov, size_t count) | |
2421 | { | |
2422 | size_t n; | |
2423 | u32 max = FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT; | |
2424 | ||
fb6ccff6 | 2425 | for (n = 0; n < count; n++, iov++) { |
7572777e MS |
2426 | if (iov->iov_len > (size_t) max) |
2427 | return -ENOMEM; | |
2428 | max -= iov->iov_len; | |
2429 | } | |
2430 | return 0; | |
2431 | } | |
2432 | ||
1baa26b2 MS |
2433 | static int fuse_copy_ioctl_iovec(struct fuse_conn *fc, struct iovec *dst, |
2434 | void *src, size_t transferred, unsigned count, | |
2435 | bool is_compat) | |
2436 | { | |
2437 | unsigned i; | |
2438 | struct fuse_ioctl_iovec *fiov = src; | |
2439 | ||
2440 | if (fc->minor < 16) { | |
2441 | return fuse_copy_ioctl_iovec_old(dst, src, transferred, | |
2442 | count, is_compat); | |
2443 | } | |
2444 | ||
2445 | if (count * sizeof(struct fuse_ioctl_iovec) != transferred) | |
2446 | return -EIO; | |
2447 | ||
2448 | for (i = 0; i < count; i++) { | |
2449 | /* Did the server supply an inappropriate value? */ | |
2450 | if (fiov[i].base != (unsigned long) fiov[i].base || | |
2451 | fiov[i].len != (unsigned long) fiov[i].len) | |
2452 | return -EIO; | |
2453 | ||
2454 | dst[i].iov_base = (void __user *) (unsigned long) fiov[i].base; | |
2455 | dst[i].iov_len = (size_t) fiov[i].len; | |
2456 | ||
2457 | #ifdef CONFIG_COMPAT | |
2458 | if (is_compat && | |
2459 | (ptr_to_compat(dst[i].iov_base) != fiov[i].base || | |
2460 | (compat_size_t) dst[i].iov_len != fiov[i].len)) | |
2461 | return -EIO; | |
2462 | #endif | |
2463 | } | |
2464 | ||
2465 | return 0; | |
2466 | } | |
2467 | ||
2468 | ||
59efec7b TH |
2469 | /* |
2470 | * For ioctls, there is no generic way to determine how much memory | |
2471 | * needs to be read and/or written. Furthermore, ioctls are allowed | |
2472 | * to dereference the passed pointer, so the parameter requires deep | |
2473 | * copying but FUSE has no idea whatsoever about what to copy in or | |
2474 | * out. | |
2475 | * | |
2476 | * This is solved by allowing FUSE server to retry ioctl with | |
2477 | * necessary in/out iovecs. Let's assume the ioctl implementation | |
2478 | * needs to read in the following structure. | |
2479 | * | |
2480 | * struct a { | |
2481 | * char *buf; | |
2482 | * size_t buflen; | |
2483 | * } | |
2484 | * | |
2485 | * On the first callout to FUSE server, inarg->in_size and | |
2486 | * inarg->out_size will be NULL; then, the server completes the ioctl | |
2487 | * with FUSE_IOCTL_RETRY set in out->flags, out->in_iovs set to 1 and | |
2488 | * the actual iov array to | |
2489 | * | |
2490 | * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) } } | |
2491 | * | |
2492 | * which tells FUSE to copy in the requested area and retry the ioctl. | |
2493 | * On the second round, the server has access to the structure and | |
2494 | * from that it can tell what to look for next, so on the invocation, | |
2495 | * it sets FUSE_IOCTL_RETRY, out->in_iovs to 2 and iov array to | |
2496 | * | |
2497 | * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) }, | |
2498 | * { .iov_base = a.buf, .iov_len = a.buflen } } | |
2499 | * | |
2500 | * FUSE will copy both struct a and the pointed buffer from the | |
2501 | * process doing the ioctl and retry ioctl with both struct a and the | |
2502 | * buffer. | |
2503 | * | |
2504 | * This time, FUSE server has everything it needs and completes ioctl | |
2505 | * without FUSE_IOCTL_RETRY which finishes the ioctl call. | |
2506 | * | |
2507 | * Copying data out works the same way. | |
2508 | * | |
2509 | * Note that if FUSE_IOCTL_UNRESTRICTED is clear, the kernel | |
2510 | * automatically initializes in and out iovs by decoding @cmd with | |
2511 | * _IOC_* macros and the server is not allowed to request RETRY. This | |
2512 | * limits ioctl data transfers to well-formed ioctls and is the forced | |
2513 | * behavior for all FUSE servers. | |
2514 | */ | |
08cbf542 TH |
2515 | long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg, |
2516 | unsigned int flags) | |
59efec7b | 2517 | { |
59efec7b | 2518 | struct fuse_file *ff = file->private_data; |
d36f2487 | 2519 | struct fuse_conn *fc = ff->fc; |
59efec7b TH |
2520 | struct fuse_ioctl_in inarg = { |
2521 | .fh = ff->fh, | |
2522 | .cmd = cmd, | |
2523 | .arg = arg, | |
2524 | .flags = flags | |
2525 | }; | |
2526 | struct fuse_ioctl_out outarg; | |
2527 | struct fuse_req *req = NULL; | |
2528 | struct page **pages = NULL; | |
8ac83505 | 2529 | struct iovec *iov_page = NULL; |
59efec7b TH |
2530 | struct iovec *in_iov = NULL, *out_iov = NULL; |
2531 | unsigned int in_iovs = 0, out_iovs = 0, num_pages = 0, max_pages; | |
2532 | size_t in_size, out_size, transferred; | |
2533 | int err; | |
2534 | ||
1baa26b2 MS |
2535 | #if BITS_PER_LONG == 32 |
2536 | inarg.flags |= FUSE_IOCTL_32BIT; | |
2537 | #else | |
2538 | if (flags & FUSE_IOCTL_COMPAT) | |
2539 | inarg.flags |= FUSE_IOCTL_32BIT; | |
2540 | #endif | |
2541 | ||
59efec7b | 2542 | /* assume all the iovs returned by client always fits in a page */ |
1baa26b2 | 2543 | BUILD_BUG_ON(sizeof(struct fuse_ioctl_iovec) * FUSE_IOCTL_MAX_IOV > PAGE_SIZE); |
59efec7b | 2544 | |
59efec7b | 2545 | err = -ENOMEM; |
c411cc88 | 2546 | pages = kcalloc(FUSE_MAX_PAGES_PER_REQ, sizeof(pages[0]), GFP_KERNEL); |
8ac83505 | 2547 | iov_page = (struct iovec *) __get_free_page(GFP_KERNEL); |
59efec7b TH |
2548 | if (!pages || !iov_page) |
2549 | goto out; | |
2550 | ||
2551 | /* | |
2552 | * If restricted, initialize IO parameters as encoded in @cmd. | |
2553 | * RETRY from server is not allowed. | |
2554 | */ | |
2555 | if (!(flags & FUSE_IOCTL_UNRESTRICTED)) { | |
8ac83505 | 2556 | struct iovec *iov = iov_page; |
59efec7b | 2557 | |
c9f0523d | 2558 | iov->iov_base = (void __user *)arg; |
59efec7b TH |
2559 | iov->iov_len = _IOC_SIZE(cmd); |
2560 | ||
2561 | if (_IOC_DIR(cmd) & _IOC_WRITE) { | |
2562 | in_iov = iov; | |
2563 | in_iovs = 1; | |
2564 | } | |
2565 | ||
2566 | if (_IOC_DIR(cmd) & _IOC_READ) { | |
2567 | out_iov = iov; | |
2568 | out_iovs = 1; | |
2569 | } | |
2570 | } | |
2571 | ||
2572 | retry: | |
2573 | inarg.in_size = in_size = iov_length(in_iov, in_iovs); | |
2574 | inarg.out_size = out_size = iov_length(out_iov, out_iovs); | |
2575 | ||
2576 | /* | |
2577 | * Out data can be used either for actual out data or iovs, | |
2578 | * make sure there always is at least one page. | |
2579 | */ | |
2580 | out_size = max_t(size_t, out_size, PAGE_SIZE); | |
2581 | max_pages = DIV_ROUND_UP(max(in_size, out_size), PAGE_SIZE); | |
2582 | ||
2583 | /* make sure there are enough buffer pages and init request with them */ | |
2584 | err = -ENOMEM; | |
2585 | if (max_pages > FUSE_MAX_PAGES_PER_REQ) | |
2586 | goto out; | |
2587 | while (num_pages < max_pages) { | |
2588 | pages[num_pages] = alloc_page(GFP_KERNEL | __GFP_HIGHMEM); | |
2589 | if (!pages[num_pages]) | |
2590 | goto out; | |
2591 | num_pages++; | |
2592 | } | |
2593 | ||
54b96670 | 2594 | req = fuse_get_req(fc, num_pages); |
59efec7b TH |
2595 | if (IS_ERR(req)) { |
2596 | err = PTR_ERR(req); | |
2597 | req = NULL; | |
2598 | goto out; | |
2599 | } | |
2600 | memcpy(req->pages, pages, sizeof(req->pages[0]) * num_pages); | |
2601 | req->num_pages = num_pages; | |
7c190c8b | 2602 | fuse_page_descs_length_init(req, 0, req->num_pages); |
59efec7b TH |
2603 | |
2604 | /* okay, let's send it to the client */ | |
2605 | req->in.h.opcode = FUSE_IOCTL; | |
d36f2487 | 2606 | req->in.h.nodeid = ff->nodeid; |
59efec7b TH |
2607 | req->in.numargs = 1; |
2608 | req->in.args[0].size = sizeof(inarg); | |
2609 | req->in.args[0].value = &inarg; | |
2610 | if (in_size) { | |
2611 | req->in.numargs++; | |
2612 | req->in.args[1].size = in_size; | |
2613 | req->in.argpages = 1; | |
2614 | ||
2615 | err = fuse_ioctl_copy_user(pages, in_iov, in_iovs, in_size, | |
2616 | false); | |
2617 | if (err) | |
2618 | goto out; | |
2619 | } | |
2620 | ||
2621 | req->out.numargs = 2; | |
2622 | req->out.args[0].size = sizeof(outarg); | |
2623 | req->out.args[0].value = &outarg; | |
2624 | req->out.args[1].size = out_size; | |
2625 | req->out.argpages = 1; | |
2626 | req->out.argvar = 1; | |
2627 | ||
b93f858a | 2628 | fuse_request_send(fc, req); |
59efec7b TH |
2629 | err = req->out.h.error; |
2630 | transferred = req->out.args[1].size; | |
2631 | fuse_put_request(fc, req); | |
2632 | req = NULL; | |
2633 | if (err) | |
2634 | goto out; | |
2635 | ||
2636 | /* did it ask for retry? */ | |
2637 | if (outarg.flags & FUSE_IOCTL_RETRY) { | |
8ac83505 | 2638 | void *vaddr; |
59efec7b TH |
2639 | |
2640 | /* no retry if in restricted mode */ | |
2641 | err = -EIO; | |
2642 | if (!(flags & FUSE_IOCTL_UNRESTRICTED)) | |
2643 | goto out; | |
2644 | ||
2645 | in_iovs = outarg.in_iovs; | |
2646 | out_iovs = outarg.out_iovs; | |
2647 | ||
2648 | /* | |
2649 | * Make sure things are in boundary, separate checks | |
2650 | * are to protect against overflow. | |
2651 | */ | |
2652 | err = -ENOMEM; | |
2653 | if (in_iovs > FUSE_IOCTL_MAX_IOV || | |
2654 | out_iovs > FUSE_IOCTL_MAX_IOV || | |
2655 | in_iovs + out_iovs > FUSE_IOCTL_MAX_IOV) | |
2656 | goto out; | |
2657 | ||
2408f6ef | 2658 | vaddr = kmap_atomic(pages[0]); |
1baa26b2 | 2659 | err = fuse_copy_ioctl_iovec(fc, iov_page, vaddr, |
d9d318d3 MS |
2660 | transferred, in_iovs + out_iovs, |
2661 | (flags & FUSE_IOCTL_COMPAT) != 0); | |
2408f6ef | 2662 | kunmap_atomic(vaddr); |
d9d318d3 MS |
2663 | if (err) |
2664 | goto out; | |
59efec7b | 2665 | |
8ac83505 | 2666 | in_iov = iov_page; |
59efec7b TH |
2667 | out_iov = in_iov + in_iovs; |
2668 | ||
7572777e MS |
2669 | err = fuse_verify_ioctl_iov(in_iov, in_iovs); |
2670 | if (err) | |
2671 | goto out; | |
2672 | ||
2673 | err = fuse_verify_ioctl_iov(out_iov, out_iovs); | |
2674 | if (err) | |
2675 | goto out; | |
2676 | ||
59efec7b TH |
2677 | goto retry; |
2678 | } | |
2679 | ||
2680 | err = -EIO; | |
2681 | if (transferred > inarg.out_size) | |
2682 | goto out; | |
2683 | ||
2684 | err = fuse_ioctl_copy_user(pages, out_iov, out_iovs, transferred, true); | |
2685 | out: | |
2686 | if (req) | |
2687 | fuse_put_request(fc, req); | |
8ac83505 | 2688 | free_page((unsigned long) iov_page); |
59efec7b TH |
2689 | while (num_pages) |
2690 | __free_page(pages[--num_pages]); | |
2691 | kfree(pages); | |
2692 | ||
2693 | return err ? err : outarg.result; | |
2694 | } | |
08cbf542 | 2695 | EXPORT_SYMBOL_GPL(fuse_do_ioctl); |
59efec7b | 2696 | |
b18da0c5 MS |
2697 | long fuse_ioctl_common(struct file *file, unsigned int cmd, |
2698 | unsigned long arg, unsigned int flags) | |
d36f2487 | 2699 | { |
6131ffaa | 2700 | struct inode *inode = file_inode(file); |
d36f2487 MS |
2701 | struct fuse_conn *fc = get_fuse_conn(inode); |
2702 | ||
c2132c1b | 2703 | if (!fuse_allow_current_process(fc)) |
d36f2487 MS |
2704 | return -EACCES; |
2705 | ||
2706 | if (is_bad_inode(inode)) | |
2707 | return -EIO; | |
2708 | ||
2709 | return fuse_do_ioctl(file, cmd, arg, flags); | |
2710 | } | |
2711 | ||
59efec7b TH |
2712 | static long fuse_file_ioctl(struct file *file, unsigned int cmd, |
2713 | unsigned long arg) | |
2714 | { | |
b18da0c5 | 2715 | return fuse_ioctl_common(file, cmd, arg, 0); |
59efec7b TH |
2716 | } |
2717 | ||
2718 | static long fuse_file_compat_ioctl(struct file *file, unsigned int cmd, | |
2719 | unsigned long arg) | |
2720 | { | |
b18da0c5 | 2721 | return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_COMPAT); |
59efec7b TH |
2722 | } |
2723 | ||
95668a69 TH |
2724 | /* |
2725 | * All files which have been polled are linked to RB tree | |
2726 | * fuse_conn->polled_files which is indexed by kh. Walk the tree and | |
2727 | * find the matching one. | |
2728 | */ | |
2729 | static struct rb_node **fuse_find_polled_node(struct fuse_conn *fc, u64 kh, | |
2730 | struct rb_node **parent_out) | |
2731 | { | |
2732 | struct rb_node **link = &fc->polled_files.rb_node; | |
2733 | struct rb_node *last = NULL; | |
2734 | ||
2735 | while (*link) { | |
2736 | struct fuse_file *ff; | |
2737 | ||
2738 | last = *link; | |
2739 | ff = rb_entry(last, struct fuse_file, polled_node); | |
2740 | ||
2741 | if (kh < ff->kh) | |
2742 | link = &last->rb_left; | |
2743 | else if (kh > ff->kh) | |
2744 | link = &last->rb_right; | |
2745 | else | |
2746 | return link; | |
2747 | } | |
2748 | ||
2749 | if (parent_out) | |
2750 | *parent_out = last; | |
2751 | return link; | |
2752 | } | |
2753 | ||
2754 | /* | |
2755 | * The file is about to be polled. Make sure it's on the polled_files | |
2756 | * RB tree. Note that files once added to the polled_files tree are | |
2757 | * not removed before the file is released. This is because a file | |
2758 | * polled once is likely to be polled again. | |
2759 | */ | |
2760 | static void fuse_register_polled_file(struct fuse_conn *fc, | |
2761 | struct fuse_file *ff) | |
2762 | { | |
2763 | spin_lock(&fc->lock); | |
2764 | if (RB_EMPTY_NODE(&ff->polled_node)) { | |
f3846266 | 2765 | struct rb_node **link, *uninitialized_var(parent); |
95668a69 TH |
2766 | |
2767 | link = fuse_find_polled_node(fc, ff->kh, &parent); | |
2768 | BUG_ON(*link); | |
2769 | rb_link_node(&ff->polled_node, parent, link); | |
2770 | rb_insert_color(&ff->polled_node, &fc->polled_files); | |
2771 | } | |
2772 | spin_unlock(&fc->lock); | |
2773 | } | |
2774 | ||
08cbf542 | 2775 | unsigned fuse_file_poll(struct file *file, poll_table *wait) |
95668a69 | 2776 | { |
95668a69 | 2777 | struct fuse_file *ff = file->private_data; |
797759aa | 2778 | struct fuse_conn *fc = ff->fc; |
95668a69 TH |
2779 | struct fuse_poll_in inarg = { .fh = ff->fh, .kh = ff->kh }; |
2780 | struct fuse_poll_out outarg; | |
7078187a | 2781 | FUSE_ARGS(args); |
95668a69 TH |
2782 | int err; |
2783 | ||
2784 | if (fc->no_poll) | |
2785 | return DEFAULT_POLLMASK; | |
2786 | ||
2787 | poll_wait(file, &ff->poll_wait, wait); | |
0415d291 | 2788 | inarg.events = (__u32)poll_requested_events(wait); |
95668a69 TH |
2789 | |
2790 | /* | |
2791 | * Ask for notification iff there's someone waiting for it. | |
2792 | * The client may ignore the flag and always notify. | |
2793 | */ | |
2794 | if (waitqueue_active(&ff->poll_wait)) { | |
2795 | inarg.flags |= FUSE_POLL_SCHEDULE_NOTIFY; | |
2796 | fuse_register_polled_file(fc, ff); | |
2797 | } | |
2798 | ||
7078187a MS |
2799 | args.in.h.opcode = FUSE_POLL; |
2800 | args.in.h.nodeid = ff->nodeid; | |
2801 | args.in.numargs = 1; | |
2802 | args.in.args[0].size = sizeof(inarg); | |
2803 | args.in.args[0].value = &inarg; | |
2804 | args.out.numargs = 1; | |
2805 | args.out.args[0].size = sizeof(outarg); | |
2806 | args.out.args[0].value = &outarg; | |
2807 | err = fuse_simple_request(fc, &args); | |
95668a69 TH |
2808 | |
2809 | if (!err) | |
2810 | return outarg.revents; | |
2811 | if (err == -ENOSYS) { | |
2812 | fc->no_poll = 1; | |
2813 | return DEFAULT_POLLMASK; | |
2814 | } | |
2815 | return POLLERR; | |
2816 | } | |
08cbf542 | 2817 | EXPORT_SYMBOL_GPL(fuse_file_poll); |
95668a69 TH |
2818 | |
2819 | /* | |
2820 | * This is called from fuse_handle_notify() on FUSE_NOTIFY_POLL and | |
2821 | * wakes up the poll waiters. | |
2822 | */ | |
2823 | int fuse_notify_poll_wakeup(struct fuse_conn *fc, | |
2824 | struct fuse_notify_poll_wakeup_out *outarg) | |
2825 | { | |
2826 | u64 kh = outarg->kh; | |
2827 | struct rb_node **link; | |
2828 | ||
2829 | spin_lock(&fc->lock); | |
2830 | ||
2831 | link = fuse_find_polled_node(fc, kh, NULL); | |
2832 | if (*link) { | |
2833 | struct fuse_file *ff; | |
2834 | ||
2835 | ff = rb_entry(*link, struct fuse_file, polled_node); | |
2836 | wake_up_interruptible_sync(&ff->poll_wait); | |
2837 | } | |
2838 | ||
2839 | spin_unlock(&fc->lock); | |
2840 | return 0; | |
2841 | } | |
2842 | ||
efb9fa9e MP |
2843 | static void fuse_do_truncate(struct file *file) |
2844 | { | |
2845 | struct inode *inode = file->f_mapping->host; | |
2846 | struct iattr attr; | |
2847 | ||
2848 | attr.ia_valid = ATTR_SIZE; | |
2849 | attr.ia_size = i_size_read(inode); | |
2850 | ||
2851 | attr.ia_file = file; | |
2852 | attr.ia_valid |= ATTR_FILE; | |
2853 | ||
2854 | fuse_do_setattr(inode, &attr, file); | |
2855 | } | |
2856 | ||
e5c5f05d MP |
2857 | static inline loff_t fuse_round_up(loff_t off) |
2858 | { | |
2859 | return round_up(off, FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT); | |
2860 | } | |
2861 | ||
4273b793 | 2862 | static ssize_t |
c8b8e32d | 2863 | fuse_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
4273b793 | 2864 | { |
9d5722b7 | 2865 | DECLARE_COMPLETION_ONSTACK(wait); |
4273b793 | 2866 | ssize_t ret = 0; |
60b9df7a MS |
2867 | struct file *file = iocb->ki_filp; |
2868 | struct fuse_file *ff = file->private_data; | |
e5c5f05d | 2869 | bool async_dio = ff->fc->async_dio; |
4273b793 | 2870 | loff_t pos = 0; |
bcba24cc MP |
2871 | struct inode *inode; |
2872 | loff_t i_size; | |
a6cbcd4a | 2873 | size_t count = iov_iter_count(iter); |
c8b8e32d | 2874 | loff_t offset = iocb->ki_pos; |
36cf66ed | 2875 | struct fuse_io_priv *io; |
4273b793 | 2876 | |
4273b793 | 2877 | pos = offset; |
bcba24cc MP |
2878 | inode = file->f_mapping->host; |
2879 | i_size = i_size_read(inode); | |
4273b793 | 2880 | |
6f673763 | 2881 | if ((iov_iter_rw(iter) == READ) && (offset > i_size)) |
9fe55eea SW |
2882 | return 0; |
2883 | ||
439ee5f0 | 2884 | /* optimization for short read */ |
6f673763 | 2885 | if (async_dio && iov_iter_rw(iter) != WRITE && offset + count > i_size) { |
439ee5f0 MP |
2886 | if (offset >= i_size) |
2887 | return 0; | |
6b775b18 AV |
2888 | iov_iter_truncate(iter, fuse_round_up(i_size - offset)); |
2889 | count = iov_iter_count(iter); | |
439ee5f0 MP |
2890 | } |
2891 | ||
bcba24cc | 2892 | io = kmalloc(sizeof(struct fuse_io_priv), GFP_KERNEL); |
36cf66ed MP |
2893 | if (!io) |
2894 | return -ENOMEM; | |
bcba24cc | 2895 | spin_lock_init(&io->lock); |
744742d6 | 2896 | kref_init(&io->refcnt); |
bcba24cc MP |
2897 | io->reqs = 1; |
2898 | io->bytes = -1; | |
2899 | io->size = 0; | |
2900 | io->offset = offset; | |
6f673763 | 2901 | io->write = (iov_iter_rw(iter) == WRITE); |
bcba24cc | 2902 | io->err = 0; |
36cf66ed | 2903 | io->file = file; |
bcba24cc MP |
2904 | /* |
2905 | * By default, we want to optimize all I/Os with async request | |
60b9df7a | 2906 | * submission to the client filesystem if supported. |
bcba24cc | 2907 | */ |
e5c5f05d | 2908 | io->async = async_dio; |
bcba24cc | 2909 | io->iocb = iocb; |
7879c4e5 | 2910 | io->blocking = is_sync_kiocb(iocb); |
bcba24cc MP |
2911 | |
2912 | /* | |
7879c4e5 AS |
2913 | * We cannot asynchronously extend the size of a file. |
2914 | * In such case the aio will behave exactly like sync io. | |
bcba24cc | 2915 | */ |
7879c4e5 AS |
2916 | if ((offset + count > i_size) && iov_iter_rw(iter) == WRITE) |
2917 | io->blocking = true; | |
4273b793 | 2918 | |
7879c4e5 | 2919 | if (io->async && io->blocking) { |
744742d6 SF |
2920 | /* |
2921 | * Additional reference to keep io around after | |
2922 | * calling fuse_aio_complete() | |
2923 | */ | |
2924 | kref_get(&io->refcnt); | |
9d5722b7 | 2925 | io->done = &wait; |
744742d6 | 2926 | } |
9d5722b7 | 2927 | |
6f673763 | 2928 | if (iov_iter_rw(iter) == WRITE) { |
6b775b18 | 2929 | ret = fuse_direct_io(io, iter, &pos, FUSE_DIO_WRITE); |
812408fb AV |
2930 | fuse_invalidate_attr(inode); |
2931 | } else { | |
d22a943f | 2932 | ret = __fuse_direct_read(io, iter, &pos); |
812408fb | 2933 | } |
36cf66ed | 2934 | |
bcba24cc MP |
2935 | if (io->async) { |
2936 | fuse_aio_complete(io, ret < 0 ? ret : 0, -1); | |
2937 | ||
2938 | /* we have a non-extending, async request, so return */ | |
7879c4e5 | 2939 | if (!io->blocking) |
bcba24cc MP |
2940 | return -EIOCBQUEUED; |
2941 | ||
9d5722b7 CH |
2942 | wait_for_completion(&wait); |
2943 | ret = fuse_get_res_by_io(io); | |
bcba24cc MP |
2944 | } |
2945 | ||
744742d6 | 2946 | kref_put(&io->refcnt, fuse_io_release); |
9d5722b7 | 2947 | |
6f673763 | 2948 | if (iov_iter_rw(iter) == WRITE) { |
efb9fa9e MP |
2949 | if (ret > 0) |
2950 | fuse_write_update_size(inode, pos); | |
2951 | else if (ret < 0 && offset + count > i_size) | |
2952 | fuse_do_truncate(file); | |
2953 | } | |
4273b793 AA |
2954 | |
2955 | return ret; | |
2956 | } | |
2957 | ||
cdadb11c MS |
2958 | static long fuse_file_fallocate(struct file *file, int mode, loff_t offset, |
2959 | loff_t length) | |
05ba1f08 AP |
2960 | { |
2961 | struct fuse_file *ff = file->private_data; | |
1c68271c | 2962 | struct inode *inode = file_inode(file); |
0ab08f57 | 2963 | struct fuse_inode *fi = get_fuse_inode(inode); |
05ba1f08 | 2964 | struct fuse_conn *fc = ff->fc; |
7078187a | 2965 | FUSE_ARGS(args); |
05ba1f08 AP |
2966 | struct fuse_fallocate_in inarg = { |
2967 | .fh = ff->fh, | |
2968 | .offset = offset, | |
2969 | .length = length, | |
2970 | .mode = mode | |
2971 | }; | |
2972 | int err; | |
14c14414 MP |
2973 | bool lock_inode = !(mode & FALLOC_FL_KEEP_SIZE) || |
2974 | (mode & FALLOC_FL_PUNCH_HOLE); | |
05ba1f08 | 2975 | |
4adb8302 MS |
2976 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
2977 | return -EOPNOTSUPP; | |
2978 | ||
519c6040 MS |
2979 | if (fc->no_fallocate) |
2980 | return -EOPNOTSUPP; | |
2981 | ||
14c14414 | 2982 | if (lock_inode) { |
5955102c | 2983 | inode_lock(inode); |
bde52788 MP |
2984 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
2985 | loff_t endbyte = offset + length - 1; | |
2986 | err = filemap_write_and_wait_range(inode->i_mapping, | |
2987 | offset, endbyte); | |
2988 | if (err) | |
2989 | goto out; | |
2990 | ||
2991 | fuse_sync_writes(inode); | |
2992 | } | |
3634a632 BF |
2993 | } |
2994 | ||
0ab08f57 MP |
2995 | if (!(mode & FALLOC_FL_KEEP_SIZE)) |
2996 | set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); | |
2997 | ||
7078187a MS |
2998 | args.in.h.opcode = FUSE_FALLOCATE; |
2999 | args.in.h.nodeid = ff->nodeid; | |
3000 | args.in.numargs = 1; | |
3001 | args.in.args[0].size = sizeof(inarg); | |
3002 | args.in.args[0].value = &inarg; | |
3003 | err = fuse_simple_request(fc, &args); | |
519c6040 MS |
3004 | if (err == -ENOSYS) { |
3005 | fc->no_fallocate = 1; | |
3006 | err = -EOPNOTSUPP; | |
3007 | } | |
bee6c307 BF |
3008 | if (err) |
3009 | goto out; | |
3010 | ||
3011 | /* we could have extended the file */ | |
b0aa7606 MP |
3012 | if (!(mode & FALLOC_FL_KEEP_SIZE)) { |
3013 | bool changed = fuse_write_update_size(inode, offset + length); | |
3014 | ||
93d2269d MS |
3015 | if (changed && fc->writeback_cache) |
3016 | file_update_time(file); | |
b0aa7606 | 3017 | } |
bee6c307 BF |
3018 | |
3019 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
3020 | truncate_pagecache_range(inode, offset, offset + length - 1); | |
3021 | ||
3022 | fuse_invalidate_attr(inode); | |
3023 | ||
3634a632 | 3024 | out: |
0ab08f57 MP |
3025 | if (!(mode & FALLOC_FL_KEEP_SIZE)) |
3026 | clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); | |
3027 | ||
bde52788 | 3028 | if (lock_inode) |
5955102c | 3029 | inode_unlock(inode); |
3634a632 | 3030 | |
05ba1f08 AP |
3031 | return err; |
3032 | } | |
05ba1f08 | 3033 | |
4b6f5d20 | 3034 | static const struct file_operations fuse_file_operations = { |
5559b8f4 | 3035 | .llseek = fuse_file_llseek, |
37c20f16 | 3036 | .read_iter = fuse_file_read_iter, |
84c3d55c | 3037 | .write_iter = fuse_file_write_iter, |
b6aeaded MS |
3038 | .mmap = fuse_file_mmap, |
3039 | .open = fuse_open, | |
3040 | .flush = fuse_flush, | |
3041 | .release = fuse_release, | |
3042 | .fsync = fuse_fsync, | |
71421259 | 3043 | .lock = fuse_file_lock, |
a9ff4f87 | 3044 | .flock = fuse_file_flock, |
5ffc4ef4 | 3045 | .splice_read = generic_file_splice_read, |
59efec7b TH |
3046 | .unlocked_ioctl = fuse_file_ioctl, |
3047 | .compat_ioctl = fuse_file_compat_ioctl, | |
95668a69 | 3048 | .poll = fuse_file_poll, |
05ba1f08 | 3049 | .fallocate = fuse_file_fallocate, |
b6aeaded MS |
3050 | }; |
3051 | ||
4b6f5d20 | 3052 | static const struct file_operations fuse_direct_io_file_operations = { |
5559b8f4 | 3053 | .llseek = fuse_file_llseek, |
15316263 | 3054 | .read_iter = fuse_direct_read_iter, |
15316263 | 3055 | .write_iter = fuse_direct_write_iter, |
fc280c96 | 3056 | .mmap = fuse_direct_mmap, |
413ef8cb MS |
3057 | .open = fuse_open, |
3058 | .flush = fuse_flush, | |
3059 | .release = fuse_release, | |
3060 | .fsync = fuse_fsync, | |
71421259 | 3061 | .lock = fuse_file_lock, |
a9ff4f87 | 3062 | .flock = fuse_file_flock, |
59efec7b TH |
3063 | .unlocked_ioctl = fuse_file_ioctl, |
3064 | .compat_ioctl = fuse_file_compat_ioctl, | |
95668a69 | 3065 | .poll = fuse_file_poll, |
05ba1f08 | 3066 | .fallocate = fuse_file_fallocate, |
fc280c96 | 3067 | /* no splice_read */ |
413ef8cb MS |
3068 | }; |
3069 | ||
f5e54d6e | 3070 | static const struct address_space_operations fuse_file_aops = { |
b6aeaded | 3071 | .readpage = fuse_readpage, |
3be5a52b | 3072 | .writepage = fuse_writepage, |
26d614df | 3073 | .writepages = fuse_writepages, |
3be5a52b | 3074 | .launder_page = fuse_launder_page, |
db50b96c | 3075 | .readpages = fuse_readpages, |
3be5a52b | 3076 | .set_page_dirty = __set_page_dirty_nobuffers, |
b2d2272f | 3077 | .bmap = fuse_bmap, |
4273b793 | 3078 | .direct_IO = fuse_direct_IO, |
6b12c1b3 PE |
3079 | .write_begin = fuse_write_begin, |
3080 | .write_end = fuse_write_end, | |
b6aeaded MS |
3081 | }; |
3082 | ||
3083 | void fuse_init_file_inode(struct inode *inode) | |
3084 | { | |
45323fb7 MS |
3085 | inode->i_fop = &fuse_file_operations; |
3086 | inode->i_data.a_ops = &fuse_file_aops; | |
b6aeaded | 3087 | } |