]>
Commit | Line | Data |
---|---|---|
b6aeaded MS |
1 | /* |
2 | FUSE: Filesystem in Userspace | |
1729a16c | 3 | Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu> |
b6aeaded MS |
4 | |
5 | This program can be distributed under the terms of the GNU GPL. | |
6 | See the file COPYING. | |
7 | */ | |
8 | ||
9 | #include "fuse_i.h" | |
10 | ||
11 | #include <linux/pagemap.h> | |
12 | #include <linux/slab.h> | |
13 | #include <linux/kernel.h> | |
e8edc6e0 | 14 | #include <linux/sched.h> |
7a36094d | 15 | #include <linux/sched/signal.h> |
08cbf542 | 16 | #include <linux/module.h> |
d9d318d3 | 17 | #include <linux/compat.h> |
478e0841 | 18 | #include <linux/swap.h> |
3634a632 | 19 | #include <linux/falloc.h> |
e2e40f2c | 20 | #include <linux/uio.h> |
b6aeaded | 21 | |
4b6f5d20 | 22 | static const struct file_operations fuse_direct_io_file_operations; |
45323fb7 | 23 | |
91fe96b4 MS |
24 | static int fuse_send_open(struct fuse_conn *fc, u64 nodeid, struct file *file, |
25 | int opcode, struct fuse_open_out *outargp) | |
b6aeaded | 26 | { |
b6aeaded | 27 | struct fuse_open_in inarg; |
7078187a | 28 | FUSE_ARGS(args); |
fd72faac MS |
29 | |
30 | memset(&inarg, 0, sizeof(inarg)); | |
6ff958ed MS |
31 | inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY); |
32 | if (!fc->atomic_o_trunc) | |
33 | inarg.flags &= ~O_TRUNC; | |
7078187a MS |
34 | args.in.h.opcode = opcode; |
35 | args.in.h.nodeid = nodeid; | |
36 | args.in.numargs = 1; | |
37 | args.in.args[0].size = sizeof(inarg); | |
38 | args.in.args[0].value = &inarg; | |
39 | args.out.numargs = 1; | |
40 | args.out.args[0].size = sizeof(*outargp); | |
41 | args.out.args[0].value = outargp; | |
fd72faac | 42 | |
7078187a | 43 | return fuse_simple_request(fc, &args); |
fd72faac MS |
44 | } |
45 | ||
acf99433 | 46 | struct fuse_file *fuse_file_alloc(struct fuse_conn *fc) |
fd72faac MS |
47 | { |
48 | struct fuse_file *ff; | |
6b2db28a | 49 | |
68227c03 | 50 | ff = kzalloc(sizeof(struct fuse_file), GFP_KERNEL); |
6b2db28a TH |
51 | if (unlikely(!ff)) |
52 | return NULL; | |
53 | ||
da5e4714 | 54 | ff->fc = fc; |
4250c066 | 55 | ff->reserved_req = fuse_request_alloc(0); |
6b2db28a TH |
56 | if (unlikely(!ff->reserved_req)) { |
57 | kfree(ff); | |
58 | return NULL; | |
fd72faac | 59 | } |
6b2db28a TH |
60 | |
61 | INIT_LIST_HEAD(&ff->write_entry); | |
4e8c2eb5 | 62 | refcount_set(&ff->count, 1); |
6b2db28a TH |
63 | RB_CLEAR_NODE(&ff->polled_node); |
64 | init_waitqueue_head(&ff->poll_wait); | |
65 | ||
66 | spin_lock(&fc->lock); | |
67 | ff->kh = ++fc->khctr; | |
68 | spin_unlock(&fc->lock); | |
69 | ||
fd72faac MS |
70 | return ff; |
71 | } | |
72 | ||
73 | void fuse_file_free(struct fuse_file *ff) | |
74 | { | |
33649c91 | 75 | fuse_request_free(ff->reserved_req); |
fd72faac MS |
76 | kfree(ff); |
77 | } | |
78 | ||
267d8444 | 79 | static struct fuse_file *fuse_file_get(struct fuse_file *ff) |
c756e0a4 | 80 | { |
4e8c2eb5 | 81 | refcount_inc(&ff->count); |
c756e0a4 MS |
82 | return ff; |
83 | } | |
84 | ||
819c4b3b MS |
85 | static void fuse_release_end(struct fuse_conn *fc, struct fuse_req *req) |
86 | { | |
baebccbe | 87 | iput(req->misc.release.inode); |
819c4b3b MS |
88 | } |
89 | ||
5a18ec17 | 90 | static void fuse_file_put(struct fuse_file *ff, bool sync) |
c756e0a4 | 91 | { |
4e8c2eb5 | 92 | if (refcount_dec_and_test(&ff->count)) { |
c756e0a4 | 93 | struct fuse_req *req = ff->reserved_req; |
8b0797a4 | 94 | |
7678ac50 AG |
95 | if (ff->fc->no_open) { |
96 | /* | |
97 | * Drop the release request when client does not | |
98 | * implement 'open' | |
99 | */ | |
825d6d33 | 100 | __clear_bit(FR_BACKGROUND, &req->flags); |
baebccbe | 101 | iput(req->misc.release.inode); |
7678ac50 AG |
102 | fuse_put_request(ff->fc, req); |
103 | } else if (sync) { | |
2e38bea9 | 104 | __set_bit(FR_FORCE, &req->flags); |
825d6d33 | 105 | __clear_bit(FR_BACKGROUND, &req->flags); |
5a18ec17 | 106 | fuse_request_send(ff->fc, req); |
baebccbe | 107 | iput(req->misc.release.inode); |
5a18ec17 MS |
108 | fuse_put_request(ff->fc, req); |
109 | } else { | |
110 | req->end = fuse_release_end; | |
825d6d33 | 111 | __set_bit(FR_BACKGROUND, &req->flags); |
5a18ec17 MS |
112 | fuse_request_send_background(ff->fc, req); |
113 | } | |
c756e0a4 MS |
114 | kfree(ff); |
115 | } | |
116 | } | |
117 | ||
08cbf542 TH |
118 | int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file, |
119 | bool isdir) | |
91fe96b4 | 120 | { |
91fe96b4 | 121 | struct fuse_file *ff; |
91fe96b4 MS |
122 | int opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN; |
123 | ||
124 | ff = fuse_file_alloc(fc); | |
125 | if (!ff) | |
126 | return -ENOMEM; | |
127 | ||
7678ac50 AG |
128 | ff->fh = 0; |
129 | ff->open_flags = FOPEN_KEEP_CACHE; /* Default for no-open */ | |
130 | if (!fc->no_open || isdir) { | |
131 | struct fuse_open_out outarg; | |
132 | int err; | |
133 | ||
134 | err = fuse_send_open(fc, nodeid, file, opcode, &outarg); | |
135 | if (!err) { | |
136 | ff->fh = outarg.fh; | |
137 | ff->open_flags = outarg.open_flags; | |
138 | ||
139 | } else if (err != -ENOSYS || isdir) { | |
140 | fuse_file_free(ff); | |
141 | return err; | |
142 | } else { | |
143 | fc->no_open = 1; | |
144 | } | |
91fe96b4 MS |
145 | } |
146 | ||
147 | if (isdir) | |
7678ac50 | 148 | ff->open_flags &= ~FOPEN_DIRECT_IO; |
91fe96b4 | 149 | |
91fe96b4 | 150 | ff->nodeid = nodeid; |
267d8444 | 151 | file->private_data = ff; |
91fe96b4 MS |
152 | |
153 | return 0; | |
154 | } | |
08cbf542 | 155 | EXPORT_SYMBOL_GPL(fuse_do_open); |
91fe96b4 | 156 | |
650b22b9 PE |
157 | static void fuse_link_write_file(struct file *file) |
158 | { | |
159 | struct inode *inode = file_inode(file); | |
160 | struct fuse_conn *fc = get_fuse_conn(inode); | |
161 | struct fuse_inode *fi = get_fuse_inode(inode); | |
162 | struct fuse_file *ff = file->private_data; | |
163 | /* | |
164 | * file may be written through mmap, so chain it onto the | |
165 | * inodes's write_file list | |
166 | */ | |
167 | spin_lock(&fc->lock); | |
168 | if (list_empty(&ff->write_entry)) | |
169 | list_add(&ff->write_entry, &fi->write_files); | |
170 | spin_unlock(&fc->lock); | |
171 | } | |
172 | ||
c7b7143c | 173 | void fuse_finish_open(struct inode *inode, struct file *file) |
fd72faac | 174 | { |
c7b7143c | 175 | struct fuse_file *ff = file->private_data; |
a0822c55 | 176 | struct fuse_conn *fc = get_fuse_conn(inode); |
c7b7143c MS |
177 | |
178 | if (ff->open_flags & FOPEN_DIRECT_IO) | |
fd72faac | 179 | file->f_op = &fuse_direct_io_file_operations; |
c7b7143c | 180 | if (!(ff->open_flags & FOPEN_KEEP_CACHE)) |
b1009979 | 181 | invalidate_inode_pages2(inode->i_mapping); |
c7b7143c | 182 | if (ff->open_flags & FOPEN_NONSEEKABLE) |
a7c1b990 | 183 | nonseekable_open(inode, file); |
a0822c55 KS |
184 | if (fc->atomic_o_trunc && (file->f_flags & O_TRUNC)) { |
185 | struct fuse_inode *fi = get_fuse_inode(inode); | |
186 | ||
187 | spin_lock(&fc->lock); | |
188 | fi->attr_version = ++fc->attr_version; | |
189 | i_size_write(inode, 0); | |
190 | spin_unlock(&fc->lock); | |
191 | fuse_invalidate_attr(inode); | |
75caeecd MP |
192 | if (fc->writeback_cache) |
193 | file_update_time(file); | |
a0822c55 | 194 | } |
4d99ff8f PE |
195 | if ((file->f_mode & FMODE_WRITE) && fc->writeback_cache) |
196 | fuse_link_write_file(file); | |
fd72faac MS |
197 | } |
198 | ||
91fe96b4 | 199 | int fuse_open_common(struct inode *inode, struct file *file, bool isdir) |
fd72faac | 200 | { |
acf99433 | 201 | struct fuse_conn *fc = get_fuse_conn(inode); |
b6aeaded | 202 | int err; |
75caeecd MP |
203 | bool lock_inode = (file->f_flags & O_TRUNC) && |
204 | fc->atomic_o_trunc && | |
205 | fc->writeback_cache; | |
b6aeaded MS |
206 | |
207 | err = generic_file_open(inode, file); | |
208 | if (err) | |
209 | return err; | |
210 | ||
75caeecd | 211 | if (lock_inode) |
5955102c | 212 | inode_lock(inode); |
75caeecd | 213 | |
91fe96b4 | 214 | err = fuse_do_open(fc, get_node_id(inode), file, isdir); |
b6aeaded | 215 | |
75caeecd MP |
216 | if (!err) |
217 | fuse_finish_open(inode, file); | |
91fe96b4 | 218 | |
75caeecd | 219 | if (lock_inode) |
5955102c | 220 | inode_unlock(inode); |
75caeecd MP |
221 | |
222 | return err; | |
b6aeaded MS |
223 | } |
224 | ||
8b0797a4 | 225 | static void fuse_prepare_release(struct fuse_file *ff, int flags, int opcode) |
64c6d8ed | 226 | { |
8b0797a4 | 227 | struct fuse_conn *fc = ff->fc; |
33649c91 | 228 | struct fuse_req *req = ff->reserved_req; |
b57d4264 | 229 | struct fuse_release_in *inarg = &req->misc.release.in; |
b6aeaded | 230 | |
8b0797a4 MS |
231 | spin_lock(&fc->lock); |
232 | list_del(&ff->write_entry); | |
233 | if (!RB_EMPTY_NODE(&ff->polled_node)) | |
234 | rb_erase(&ff->polled_node, &fc->polled_files); | |
235 | spin_unlock(&fc->lock); | |
236 | ||
357ccf2b | 237 | wake_up_interruptible_all(&ff->poll_wait); |
8b0797a4 | 238 | |
b6aeaded | 239 | inarg->fh = ff->fh; |
fd72faac | 240 | inarg->flags = flags; |
51eb01e7 | 241 | req->in.h.opcode = opcode; |
c7b7143c | 242 | req->in.h.nodeid = ff->nodeid; |
b6aeaded MS |
243 | req->in.numargs = 1; |
244 | req->in.args[0].size = sizeof(struct fuse_release_in); | |
245 | req->in.args[0].value = inarg; | |
fd72faac MS |
246 | } |
247 | ||
8b0797a4 | 248 | void fuse_release_common(struct file *file, int opcode) |
fd72faac | 249 | { |
9a87ad3d MS |
250 | struct fuse_file *ff = file->private_data; |
251 | struct fuse_req *req = ff->reserved_req; | |
6b2db28a | 252 | |
8b0797a4 | 253 | fuse_prepare_release(ff, file->f_flags, opcode); |
6b2db28a | 254 | |
37fb3a30 MS |
255 | if (ff->flock) { |
256 | struct fuse_release_in *inarg = &req->misc.release.in; | |
257 | inarg->release_flags |= FUSE_RELEASE_FLOCK_UNLOCK; | |
258 | inarg->lock_owner = fuse_lock_owner_id(ff->fc, | |
259 | (fl_owner_t) file); | |
260 | } | |
baebccbe MS |
261 | /* Hold inode until release is finished */ |
262 | req->misc.release.inode = igrab(file_inode(file)); | |
6b2db28a | 263 | |
6b2db28a TH |
264 | /* |
265 | * Normally this will send the RELEASE request, however if | |
266 | * some asynchronous READ or WRITE requests are outstanding, | |
267 | * the sending will be delayed. | |
5a18ec17 MS |
268 | * |
269 | * Make the release synchronous if this is a fuseblk mount, | |
270 | * synchronous RELEASE is allowed (and desirable) in this case | |
271 | * because the server can be trusted not to screw up. | |
6b2db28a | 272 | */ |
5a18ec17 | 273 | fuse_file_put(ff, ff->fc->destroy_req != NULL); |
b6aeaded MS |
274 | } |
275 | ||
04730fef MS |
276 | static int fuse_open(struct inode *inode, struct file *file) |
277 | { | |
91fe96b4 | 278 | return fuse_open_common(inode, file, false); |
04730fef MS |
279 | } |
280 | ||
281 | static int fuse_release(struct inode *inode, struct file *file) | |
282 | { | |
e7cc133c PE |
283 | struct fuse_conn *fc = get_fuse_conn(inode); |
284 | ||
285 | /* see fuse_vma_close() for !writeback_cache case */ | |
286 | if (fc->writeback_cache) | |
1e18bda8 | 287 | write_inode_now(inode, 1); |
b0aa7606 | 288 | |
8b0797a4 MS |
289 | fuse_release_common(file, FUSE_RELEASE); |
290 | ||
291 | /* return value is ignored by VFS */ | |
292 | return 0; | |
293 | } | |
294 | ||
295 | void fuse_sync_release(struct fuse_file *ff, int flags) | |
296 | { | |
4e8c2eb5 | 297 | WARN_ON(refcount_read(&ff->count) > 1); |
8b0797a4 | 298 | fuse_prepare_release(ff, flags, FUSE_RELEASE); |
267d8444 MS |
299 | /* |
300 | * iput(NULL) is a no-op and since the refcount is 1 and everything's | |
301 | * synchronous, we are fine with not doing igrab() here" | |
302 | */ | |
303 | fuse_file_put(ff, true); | |
04730fef | 304 | } |
08cbf542 | 305 | EXPORT_SYMBOL_GPL(fuse_sync_release); |
04730fef | 306 | |
71421259 | 307 | /* |
9c8ef561 MS |
308 | * Scramble the ID space with XTEA, so that the value of the files_struct |
309 | * pointer is not exposed to userspace. | |
71421259 | 310 | */ |
f3332114 | 311 | u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id) |
71421259 | 312 | { |
9c8ef561 MS |
313 | u32 *k = fc->scramble_key; |
314 | u64 v = (unsigned long) id; | |
315 | u32 v0 = v; | |
316 | u32 v1 = v >> 32; | |
317 | u32 sum = 0; | |
318 | int i; | |
319 | ||
320 | for (i = 0; i < 32; i++) { | |
321 | v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]); | |
322 | sum += 0x9E3779B9; | |
323 | v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]); | |
324 | } | |
325 | ||
326 | return (u64) v0 + ((u64) v1 << 32); | |
71421259 MS |
327 | } |
328 | ||
3be5a52b | 329 | /* |
ea8cd333 | 330 | * Check if any page in a range is under writeback |
3be5a52b MS |
331 | * |
332 | * This is currently done by walking the list of writepage requests | |
333 | * for the inode, which can be pretty inefficient. | |
334 | */ | |
ea8cd333 PE |
335 | static bool fuse_range_is_writeback(struct inode *inode, pgoff_t idx_from, |
336 | pgoff_t idx_to) | |
3be5a52b MS |
337 | { |
338 | struct fuse_conn *fc = get_fuse_conn(inode); | |
339 | struct fuse_inode *fi = get_fuse_inode(inode); | |
340 | struct fuse_req *req; | |
341 | bool found = false; | |
342 | ||
343 | spin_lock(&fc->lock); | |
344 | list_for_each_entry(req, &fi->writepages, writepages_entry) { | |
345 | pgoff_t curr_index; | |
346 | ||
347 | BUG_ON(req->inode != inode); | |
09cbfeaf | 348 | curr_index = req->misc.write.in.offset >> PAGE_SHIFT; |
ea8cd333 PE |
349 | if (idx_from < curr_index + req->num_pages && |
350 | curr_index <= idx_to) { | |
3be5a52b MS |
351 | found = true; |
352 | break; | |
353 | } | |
354 | } | |
355 | spin_unlock(&fc->lock); | |
356 | ||
357 | return found; | |
358 | } | |
359 | ||
ea8cd333 PE |
360 | static inline bool fuse_page_is_writeback(struct inode *inode, pgoff_t index) |
361 | { | |
362 | return fuse_range_is_writeback(inode, index, index); | |
363 | } | |
364 | ||
3be5a52b MS |
365 | /* |
366 | * Wait for page writeback to be completed. | |
367 | * | |
368 | * Since fuse doesn't rely on the VM writeback tracking, this has to | |
369 | * use some other means. | |
370 | */ | |
371 | static int fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index) | |
372 | { | |
373 | struct fuse_inode *fi = get_fuse_inode(inode); | |
374 | ||
375 | wait_event(fi->page_waitq, !fuse_page_is_writeback(inode, index)); | |
376 | return 0; | |
377 | } | |
378 | ||
fe38d7df MP |
379 | /* |
380 | * Wait for all pending writepages on the inode to finish. | |
381 | * | |
382 | * This is currently done by blocking further writes with FUSE_NOWRITE | |
383 | * and waiting for all sent writes to complete. | |
384 | * | |
385 | * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage | |
386 | * could conflict with truncation. | |
387 | */ | |
388 | static void fuse_sync_writes(struct inode *inode) | |
389 | { | |
390 | fuse_set_nowrite(inode); | |
391 | fuse_release_nowrite(inode); | |
392 | } | |
393 | ||
75e1fcc0 | 394 | static int fuse_flush(struct file *file, fl_owner_t id) |
b6aeaded | 395 | { |
6131ffaa | 396 | struct inode *inode = file_inode(file); |
b6aeaded MS |
397 | struct fuse_conn *fc = get_fuse_conn(inode); |
398 | struct fuse_file *ff = file->private_data; | |
399 | struct fuse_req *req; | |
400 | struct fuse_flush_in inarg; | |
401 | int err; | |
402 | ||
248d86e8 MS |
403 | if (is_bad_inode(inode)) |
404 | return -EIO; | |
405 | ||
b6aeaded MS |
406 | if (fc->no_flush) |
407 | return 0; | |
408 | ||
1e18bda8 | 409 | err = write_inode_now(inode, 1); |
fe38d7df MP |
410 | if (err) |
411 | return err; | |
412 | ||
5955102c | 413 | inode_lock(inode); |
fe38d7df | 414 | fuse_sync_writes(inode); |
5955102c | 415 | inode_unlock(inode); |
fe38d7df | 416 | |
4a7f4e88 | 417 | err = filemap_check_errors(file->f_mapping); |
9ebce595 MP |
418 | if (err) |
419 | return err; | |
420 | ||
b111c8c0 | 421 | req = fuse_get_req_nofail_nopages(fc, file); |
b6aeaded MS |
422 | memset(&inarg, 0, sizeof(inarg)); |
423 | inarg.fh = ff->fh; | |
9c8ef561 | 424 | inarg.lock_owner = fuse_lock_owner_id(fc, id); |
b6aeaded MS |
425 | req->in.h.opcode = FUSE_FLUSH; |
426 | req->in.h.nodeid = get_node_id(inode); | |
b6aeaded MS |
427 | req->in.numargs = 1; |
428 | req->in.args[0].size = sizeof(inarg); | |
429 | req->in.args[0].value = &inarg; | |
825d6d33 | 430 | __set_bit(FR_FORCE, &req->flags); |
b93f858a | 431 | fuse_request_send(fc, req); |
b6aeaded MS |
432 | err = req->out.h.error; |
433 | fuse_put_request(fc, req); | |
434 | if (err == -ENOSYS) { | |
435 | fc->no_flush = 1; | |
436 | err = 0; | |
437 | } | |
438 | return err; | |
439 | } | |
440 | ||
02c24a82 JB |
441 | int fuse_fsync_common(struct file *file, loff_t start, loff_t end, |
442 | int datasync, int isdir) | |
b6aeaded | 443 | { |
7ea80859 | 444 | struct inode *inode = file->f_mapping->host; |
b6aeaded MS |
445 | struct fuse_conn *fc = get_fuse_conn(inode); |
446 | struct fuse_file *ff = file->private_data; | |
7078187a | 447 | FUSE_ARGS(args); |
b6aeaded MS |
448 | struct fuse_fsync_in inarg; |
449 | int err; | |
450 | ||
248d86e8 MS |
451 | if (is_bad_inode(inode)) |
452 | return -EIO; | |
453 | ||
5955102c | 454 | inode_lock(inode); |
02c24a82 | 455 | |
3be5a52b MS |
456 | /* |
457 | * Start writeback against all dirty pages of the inode, then | |
458 | * wait for all outstanding writes, before sending the FSYNC | |
459 | * request. | |
460 | */ | |
7e51fe1d | 461 | err = file_write_and_wait_range(file, start, end); |
3be5a52b | 462 | if (err) |
02c24a82 | 463 | goto out; |
3be5a52b MS |
464 | |
465 | fuse_sync_writes(inode); | |
ac7f052b AK |
466 | |
467 | /* | |
468 | * Due to implementation of fuse writeback | |
7e51fe1d | 469 | * file_write_and_wait_range() does not catch errors. |
ac7f052b AK |
470 | * We have to do this directly after fuse_sync_writes() |
471 | */ | |
7e51fe1d | 472 | err = file_check_and_advance_wb_err(file); |
ac7f052b AK |
473 | if (err) |
474 | goto out; | |
475 | ||
1e18bda8 MS |
476 | err = sync_inode_metadata(inode, 1); |
477 | if (err) | |
478 | goto out; | |
3be5a52b | 479 | |
22401e7b MS |
480 | if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir)) |
481 | goto out; | |
b0aa7606 | 482 | |
b6aeaded MS |
483 | memset(&inarg, 0, sizeof(inarg)); |
484 | inarg.fh = ff->fh; | |
485 | inarg.fsync_flags = datasync ? 1 : 0; | |
7078187a MS |
486 | args.in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC; |
487 | args.in.h.nodeid = get_node_id(inode); | |
488 | args.in.numargs = 1; | |
489 | args.in.args[0].size = sizeof(inarg); | |
490 | args.in.args[0].value = &inarg; | |
491 | err = fuse_simple_request(fc, &args); | |
b6aeaded | 492 | if (err == -ENOSYS) { |
82547981 MS |
493 | if (isdir) |
494 | fc->no_fsyncdir = 1; | |
495 | else | |
496 | fc->no_fsync = 1; | |
b6aeaded MS |
497 | err = 0; |
498 | } | |
02c24a82 | 499 | out: |
5955102c | 500 | inode_unlock(inode); |
b6aeaded MS |
501 | return err; |
502 | } | |
503 | ||
02c24a82 JB |
504 | static int fuse_fsync(struct file *file, loff_t start, loff_t end, |
505 | int datasync) | |
82547981 | 506 | { |
02c24a82 | 507 | return fuse_fsync_common(file, start, end, datasync, 0); |
82547981 MS |
508 | } |
509 | ||
2106cb18 MS |
510 | void fuse_read_fill(struct fuse_req *req, struct file *file, loff_t pos, |
511 | size_t count, int opcode) | |
b6aeaded | 512 | { |
5c5c5e51 | 513 | struct fuse_read_in *inarg = &req->misc.read.in; |
a6643094 | 514 | struct fuse_file *ff = file->private_data; |
b6aeaded | 515 | |
361b1eb5 MS |
516 | inarg->fh = ff->fh; |
517 | inarg->offset = pos; | |
518 | inarg->size = count; | |
a6643094 | 519 | inarg->flags = file->f_flags; |
361b1eb5 | 520 | req->in.h.opcode = opcode; |
2106cb18 | 521 | req->in.h.nodeid = ff->nodeid; |
b6aeaded MS |
522 | req->in.numargs = 1; |
523 | req->in.args[0].size = sizeof(struct fuse_read_in); | |
c1aa96a5 | 524 | req->in.args[0].value = inarg; |
b6aeaded MS |
525 | req->out.argvar = 1; |
526 | req->out.numargs = 1; | |
527 | req->out.args[0].size = count; | |
b6aeaded MS |
528 | } |
529 | ||
8fba54ae | 530 | static void fuse_release_user_pages(struct fuse_req *req, bool should_dirty) |
187c5c36 MP |
531 | { |
532 | unsigned i; | |
533 | ||
534 | for (i = 0; i < req->num_pages; i++) { | |
535 | struct page *page = req->pages[i]; | |
8fba54ae | 536 | if (should_dirty) |
187c5c36 MP |
537 | set_page_dirty_lock(page); |
538 | put_page(page); | |
539 | } | |
540 | } | |
541 | ||
744742d6 SF |
542 | static void fuse_io_release(struct kref *kref) |
543 | { | |
544 | kfree(container_of(kref, struct fuse_io_priv, refcnt)); | |
545 | } | |
546 | ||
9d5722b7 CH |
547 | static ssize_t fuse_get_res_by_io(struct fuse_io_priv *io) |
548 | { | |
549 | if (io->err) | |
550 | return io->err; | |
551 | ||
552 | if (io->bytes >= 0 && io->write) | |
553 | return -EIO; | |
554 | ||
555 | return io->bytes < 0 ? io->size : io->bytes; | |
556 | } | |
557 | ||
01e9d11a MP |
558 | /** |
559 | * In case of short read, the caller sets 'pos' to the position of | |
560 | * actual end of fuse request in IO request. Otherwise, if bytes_requested | |
561 | * == bytes_transferred or rw == WRITE, the caller sets 'pos' to -1. | |
562 | * | |
563 | * An example: | |
564 | * User requested DIO read of 64K. It was splitted into two 32K fuse requests, | |
565 | * both submitted asynchronously. The first of them was ACKed by userspace as | |
566 | * fully completed (req->out.args[0].size == 32K) resulting in pos == -1. The | |
567 | * second request was ACKed as short, e.g. only 1K was read, resulting in | |
568 | * pos == 33K. | |
569 | * | |
570 | * Thus, when all fuse requests are completed, the minimal non-negative 'pos' | |
571 | * will be equal to the length of the longest contiguous fragment of | |
572 | * transferred data starting from the beginning of IO request. | |
573 | */ | |
574 | static void fuse_aio_complete(struct fuse_io_priv *io, int err, ssize_t pos) | |
575 | { | |
576 | int left; | |
577 | ||
578 | spin_lock(&io->lock); | |
579 | if (err) | |
580 | io->err = io->err ? : err; | |
581 | else if (pos >= 0 && (io->bytes < 0 || pos < io->bytes)) | |
582 | io->bytes = pos; | |
583 | ||
584 | left = --io->reqs; | |
7879c4e5 | 585 | if (!left && io->blocking) |
9d5722b7 | 586 | complete(io->done); |
01e9d11a MP |
587 | spin_unlock(&io->lock); |
588 | ||
7879c4e5 | 589 | if (!left && !io->blocking) { |
9d5722b7 | 590 | ssize_t res = fuse_get_res_by_io(io); |
01e9d11a | 591 | |
9d5722b7 CH |
592 | if (res >= 0) { |
593 | struct inode *inode = file_inode(io->iocb->ki_filp); | |
594 | struct fuse_conn *fc = get_fuse_conn(inode); | |
595 | struct fuse_inode *fi = get_fuse_inode(inode); | |
01e9d11a | 596 | |
9d5722b7 CH |
597 | spin_lock(&fc->lock); |
598 | fi->attr_version = ++fc->attr_version; | |
599 | spin_unlock(&fc->lock); | |
01e9d11a MP |
600 | } |
601 | ||
04b2fa9f | 602 | io->iocb->ki_complete(io->iocb, res, 0); |
01e9d11a | 603 | } |
744742d6 SF |
604 | |
605 | kref_put(&io->refcnt, fuse_io_release); | |
01e9d11a MP |
606 | } |
607 | ||
608 | static void fuse_aio_complete_req(struct fuse_conn *fc, struct fuse_req *req) | |
609 | { | |
610 | struct fuse_io_priv *io = req->io; | |
611 | ssize_t pos = -1; | |
612 | ||
61c12b49 | 613 | fuse_release_user_pages(req, io->should_dirty); |
01e9d11a MP |
614 | |
615 | if (io->write) { | |
616 | if (req->misc.write.in.size != req->misc.write.out.size) | |
617 | pos = req->misc.write.in.offset - io->offset + | |
618 | req->misc.write.out.size; | |
619 | } else { | |
620 | if (req->misc.read.in.size != req->out.args[0].size) | |
621 | pos = req->misc.read.in.offset - io->offset + | |
622 | req->out.args[0].size; | |
623 | } | |
624 | ||
625 | fuse_aio_complete(io, req->out.h.error, pos); | |
626 | } | |
627 | ||
628 | static size_t fuse_async_req_send(struct fuse_conn *fc, struct fuse_req *req, | |
629 | size_t num_bytes, struct fuse_io_priv *io) | |
630 | { | |
631 | spin_lock(&io->lock); | |
744742d6 | 632 | kref_get(&io->refcnt); |
01e9d11a MP |
633 | io->size += num_bytes; |
634 | io->reqs++; | |
635 | spin_unlock(&io->lock); | |
636 | ||
637 | req->io = io; | |
638 | req->end = fuse_aio_complete_req; | |
639 | ||
36cf66ed | 640 | __fuse_get_request(req); |
01e9d11a MP |
641 | fuse_request_send_background(fc, req); |
642 | ||
643 | return num_bytes; | |
644 | } | |
645 | ||
36cf66ed | 646 | static size_t fuse_send_read(struct fuse_req *req, struct fuse_io_priv *io, |
2106cb18 | 647 | loff_t pos, size_t count, fl_owner_t owner) |
04730fef | 648 | { |
e1c0eecb | 649 | struct file *file = io->iocb->ki_filp; |
2106cb18 MS |
650 | struct fuse_file *ff = file->private_data; |
651 | struct fuse_conn *fc = ff->fc; | |
f3332114 | 652 | |
2106cb18 | 653 | fuse_read_fill(req, file, pos, count, FUSE_READ); |
f3332114 | 654 | if (owner != NULL) { |
5c5c5e51 | 655 | struct fuse_read_in *inarg = &req->misc.read.in; |
f3332114 MS |
656 | |
657 | inarg->read_flags |= FUSE_READ_LOCKOWNER; | |
658 | inarg->lock_owner = fuse_lock_owner_id(fc, owner); | |
659 | } | |
36cf66ed MP |
660 | |
661 | if (io->async) | |
662 | return fuse_async_req_send(fc, req, count, io); | |
663 | ||
b93f858a | 664 | fuse_request_send(fc, req); |
361b1eb5 | 665 | return req->out.args[0].size; |
04730fef MS |
666 | } |
667 | ||
5c5c5e51 MS |
668 | static void fuse_read_update_size(struct inode *inode, loff_t size, |
669 | u64 attr_ver) | |
670 | { | |
671 | struct fuse_conn *fc = get_fuse_conn(inode); | |
672 | struct fuse_inode *fi = get_fuse_inode(inode); | |
673 | ||
674 | spin_lock(&fc->lock); | |
06a7c3c2 MP |
675 | if (attr_ver == fi->attr_version && size < inode->i_size && |
676 | !test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) { | |
5c5c5e51 MS |
677 | fi->attr_version = ++fc->attr_version; |
678 | i_size_write(inode, size); | |
679 | } | |
680 | spin_unlock(&fc->lock); | |
681 | } | |
682 | ||
a92adc82 PE |
683 | static void fuse_short_read(struct fuse_req *req, struct inode *inode, |
684 | u64 attr_ver) | |
685 | { | |
686 | size_t num_read = req->out.args[0].size; | |
8373200b PE |
687 | struct fuse_conn *fc = get_fuse_conn(inode); |
688 | ||
689 | if (fc->writeback_cache) { | |
690 | /* | |
691 | * A hole in a file. Some data after the hole are in page cache, | |
692 | * but have not reached the client fs yet. So, the hole is not | |
693 | * present there. | |
694 | */ | |
695 | int i; | |
09cbfeaf KS |
696 | int start_idx = num_read >> PAGE_SHIFT; |
697 | size_t off = num_read & (PAGE_SIZE - 1); | |
a92adc82 | 698 | |
8373200b | 699 | for (i = start_idx; i < req->num_pages; i++) { |
09cbfeaf | 700 | zero_user_segment(req->pages[i], off, PAGE_SIZE); |
8373200b PE |
701 | off = 0; |
702 | } | |
703 | } else { | |
704 | loff_t pos = page_offset(req->pages[0]) + num_read; | |
705 | fuse_read_update_size(inode, pos, attr_ver); | |
706 | } | |
a92adc82 PE |
707 | } |
708 | ||
482fce55 | 709 | static int fuse_do_readpage(struct file *file, struct page *page) |
b6aeaded | 710 | { |
e1c0eecb MS |
711 | struct kiocb iocb; |
712 | struct fuse_io_priv io; | |
b6aeaded MS |
713 | struct inode *inode = page->mapping->host; |
714 | struct fuse_conn *fc = get_fuse_conn(inode); | |
248d86e8 | 715 | struct fuse_req *req; |
5c5c5e51 MS |
716 | size_t num_read; |
717 | loff_t pos = page_offset(page); | |
09cbfeaf | 718 | size_t count = PAGE_SIZE; |
5c5c5e51 | 719 | u64 attr_ver; |
248d86e8 MS |
720 | int err; |
721 | ||
3be5a52b | 722 | /* |
25985edc | 723 | * Page writeback can extend beyond the lifetime of the |
3be5a52b MS |
724 | * page-cache page, so make sure we read a properly synced |
725 | * page. | |
726 | */ | |
727 | fuse_wait_on_page_writeback(inode, page->index); | |
728 | ||
b111c8c0 | 729 | req = fuse_get_req(fc, 1); |
ce1d5a49 | 730 | if (IS_ERR(req)) |
482fce55 | 731 | return PTR_ERR(req); |
b6aeaded | 732 | |
5c5c5e51 MS |
733 | attr_ver = fuse_get_attr_version(fc); |
734 | ||
b6aeaded | 735 | req->out.page_zeroing = 1; |
f4975c67 | 736 | req->out.argpages = 1; |
b6aeaded MS |
737 | req->num_pages = 1; |
738 | req->pages[0] = page; | |
85f40aec | 739 | req->page_descs[0].length = count; |
e1c0eecb MS |
740 | init_sync_kiocb(&iocb, file); |
741 | io = (struct fuse_io_priv) FUSE_IO_PRIV_SYNC(&iocb); | |
36cf66ed | 742 | num_read = fuse_send_read(req, &io, pos, count, NULL); |
b6aeaded | 743 | err = req->out.h.error; |
5c5c5e51 MS |
744 | |
745 | if (!err) { | |
746 | /* | |
747 | * Short read means EOF. If file size is larger, truncate it | |
748 | */ | |
749 | if (num_read < count) | |
a92adc82 | 750 | fuse_short_read(req, inode, attr_ver); |
5c5c5e51 | 751 | |
b6aeaded | 752 | SetPageUptodate(page); |
5c5c5e51 MS |
753 | } |
754 | ||
a92adc82 | 755 | fuse_put_request(fc, req); |
482fce55 MP |
756 | |
757 | return err; | |
758 | } | |
759 | ||
760 | static int fuse_readpage(struct file *file, struct page *page) | |
761 | { | |
762 | struct inode *inode = page->mapping->host; | |
763 | int err; | |
764 | ||
765 | err = -EIO; | |
766 | if (is_bad_inode(inode)) | |
767 | goto out; | |
768 | ||
769 | err = fuse_do_readpage(file, page); | |
451418fc | 770 | fuse_invalidate_atime(inode); |
b6aeaded MS |
771 | out: |
772 | unlock_page(page); | |
773 | return err; | |
774 | } | |
775 | ||
c1aa96a5 | 776 | static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req) |
db50b96c | 777 | { |
c1aa96a5 | 778 | int i; |
5c5c5e51 MS |
779 | size_t count = req->misc.read.in.size; |
780 | size_t num_read = req->out.args[0].size; | |
ce534fb0 | 781 | struct address_space *mapping = NULL; |
c1aa96a5 | 782 | |
ce534fb0 MS |
783 | for (i = 0; mapping == NULL && i < req->num_pages; i++) |
784 | mapping = req->pages[i]->mapping; | |
5c5c5e51 | 785 | |
ce534fb0 MS |
786 | if (mapping) { |
787 | struct inode *inode = mapping->host; | |
788 | ||
789 | /* | |
790 | * Short read means EOF. If file size is larger, truncate it | |
791 | */ | |
a92adc82 PE |
792 | if (!req->out.h.error && num_read < count) |
793 | fuse_short_read(req, inode, req->misc.read.attr_ver); | |
ce534fb0 | 794 | |
451418fc | 795 | fuse_invalidate_atime(inode); |
ce534fb0 | 796 | } |
c1aa96a5 | 797 | |
db50b96c MS |
798 | for (i = 0; i < req->num_pages; i++) { |
799 | struct page *page = req->pages[i]; | |
800 | if (!req->out.h.error) | |
801 | SetPageUptodate(page); | |
c1aa96a5 MS |
802 | else |
803 | SetPageError(page); | |
db50b96c | 804 | unlock_page(page); |
09cbfeaf | 805 | put_page(page); |
db50b96c | 806 | } |
c756e0a4 | 807 | if (req->ff) |
5a18ec17 | 808 | fuse_file_put(req->ff, false); |
c1aa96a5 MS |
809 | } |
810 | ||
2106cb18 | 811 | static void fuse_send_readpages(struct fuse_req *req, struct file *file) |
c1aa96a5 | 812 | { |
2106cb18 MS |
813 | struct fuse_file *ff = file->private_data; |
814 | struct fuse_conn *fc = ff->fc; | |
c1aa96a5 | 815 | loff_t pos = page_offset(req->pages[0]); |
09cbfeaf | 816 | size_t count = req->num_pages << PAGE_SHIFT; |
f4975c67 MS |
817 | |
818 | req->out.argpages = 1; | |
c1aa96a5 | 819 | req->out.page_zeroing = 1; |
ce534fb0 | 820 | req->out.page_replace = 1; |
2106cb18 | 821 | fuse_read_fill(req, file, pos, count, FUSE_READ); |
5c5c5e51 | 822 | req->misc.read.attr_ver = fuse_get_attr_version(fc); |
9cd68455 | 823 | if (fc->async_read) { |
c756e0a4 | 824 | req->ff = fuse_file_get(ff); |
9cd68455 | 825 | req->end = fuse_readpages_end; |
b93f858a | 826 | fuse_request_send_background(fc, req); |
9cd68455 | 827 | } else { |
b93f858a | 828 | fuse_request_send(fc, req); |
9cd68455 | 829 | fuse_readpages_end(fc, req); |
e9bb09dd | 830 | fuse_put_request(fc, req); |
9cd68455 | 831 | } |
db50b96c MS |
832 | } |
833 | ||
c756e0a4 | 834 | struct fuse_fill_data { |
db50b96c | 835 | struct fuse_req *req; |
a6643094 | 836 | struct file *file; |
db50b96c | 837 | struct inode *inode; |
f8dbdf81 | 838 | unsigned nr_pages; |
db50b96c MS |
839 | }; |
840 | ||
841 | static int fuse_readpages_fill(void *_data, struct page *page) | |
842 | { | |
c756e0a4 | 843 | struct fuse_fill_data *data = _data; |
db50b96c MS |
844 | struct fuse_req *req = data->req; |
845 | struct inode *inode = data->inode; | |
846 | struct fuse_conn *fc = get_fuse_conn(inode); | |
847 | ||
3be5a52b MS |
848 | fuse_wait_on_page_writeback(inode, page->index); |
849 | ||
db50b96c MS |
850 | if (req->num_pages && |
851 | (req->num_pages == FUSE_MAX_PAGES_PER_REQ || | |
09cbfeaf | 852 | (req->num_pages + 1) * PAGE_SIZE > fc->max_read || |
db50b96c | 853 | req->pages[req->num_pages - 1]->index + 1 != page->index)) { |
f8dbdf81 MP |
854 | int nr_alloc = min_t(unsigned, data->nr_pages, |
855 | FUSE_MAX_PAGES_PER_REQ); | |
2106cb18 | 856 | fuse_send_readpages(req, data->file); |
8b41e671 MP |
857 | if (fc->async_read) |
858 | req = fuse_get_req_for_background(fc, nr_alloc); | |
859 | else | |
860 | req = fuse_get_req(fc, nr_alloc); | |
861 | ||
862 | data->req = req; | |
ce1d5a49 | 863 | if (IS_ERR(req)) { |
db50b96c | 864 | unlock_page(page); |
ce1d5a49 | 865 | return PTR_ERR(req); |
db50b96c | 866 | } |
db50b96c | 867 | } |
f8dbdf81 MP |
868 | |
869 | if (WARN_ON(req->num_pages >= req->max_pages)) { | |
109728cc | 870 | unlock_page(page); |
f8dbdf81 MP |
871 | fuse_put_request(fc, req); |
872 | return -EIO; | |
873 | } | |
874 | ||
09cbfeaf | 875 | get_page(page); |
db50b96c | 876 | req->pages[req->num_pages] = page; |
85f40aec | 877 | req->page_descs[req->num_pages].length = PAGE_SIZE; |
1729a16c | 878 | req->num_pages++; |
f8dbdf81 | 879 | data->nr_pages--; |
db50b96c MS |
880 | return 0; |
881 | } | |
882 | ||
883 | static int fuse_readpages(struct file *file, struct address_space *mapping, | |
884 | struct list_head *pages, unsigned nr_pages) | |
885 | { | |
886 | struct inode *inode = mapping->host; | |
887 | struct fuse_conn *fc = get_fuse_conn(inode); | |
c756e0a4 | 888 | struct fuse_fill_data data; |
db50b96c | 889 | int err; |
f8dbdf81 | 890 | int nr_alloc = min_t(unsigned, nr_pages, FUSE_MAX_PAGES_PER_REQ); |
248d86e8 | 891 | |
1d7ea732 | 892 | err = -EIO; |
248d86e8 | 893 | if (is_bad_inode(inode)) |
2e990021 | 894 | goto out; |
248d86e8 | 895 | |
a6643094 | 896 | data.file = file; |
db50b96c | 897 | data.inode = inode; |
8b41e671 MP |
898 | if (fc->async_read) |
899 | data.req = fuse_get_req_for_background(fc, nr_alloc); | |
900 | else | |
901 | data.req = fuse_get_req(fc, nr_alloc); | |
f8dbdf81 | 902 | data.nr_pages = nr_pages; |
1d7ea732 | 903 | err = PTR_ERR(data.req); |
ce1d5a49 | 904 | if (IS_ERR(data.req)) |
2e990021 | 905 | goto out; |
db50b96c MS |
906 | |
907 | err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data); | |
d3406ffa MS |
908 | if (!err) { |
909 | if (data.req->num_pages) | |
2106cb18 | 910 | fuse_send_readpages(data.req, file); |
d3406ffa MS |
911 | else |
912 | fuse_put_request(fc, data.req); | |
913 | } | |
2e990021 | 914 | out: |
1d7ea732 | 915 | return err; |
db50b96c MS |
916 | } |
917 | ||
37c20f16 | 918 | static ssize_t fuse_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
bcb4be80 MS |
919 | { |
920 | struct inode *inode = iocb->ki_filp->f_mapping->host; | |
a8894274 | 921 | struct fuse_conn *fc = get_fuse_conn(inode); |
bcb4be80 | 922 | |
a8894274 BF |
923 | /* |
924 | * In auto invalidate mode, always update attributes on read. | |
925 | * Otherwise, only update if we attempt to read past EOF (to ensure | |
926 | * i_size is up to date). | |
927 | */ | |
928 | if (fc->auto_inval_data || | |
37c20f16 | 929 | (iocb->ki_pos + iov_iter_count(to) > i_size_read(inode))) { |
bcb4be80 | 930 | int err; |
5b97eeac | 931 | err = fuse_update_attributes(inode, iocb->ki_filp); |
bcb4be80 MS |
932 | if (err) |
933 | return err; | |
934 | } | |
935 | ||
37c20f16 | 936 | return generic_file_read_iter(iocb, to); |
bcb4be80 MS |
937 | } |
938 | ||
2d698b07 | 939 | static void fuse_write_fill(struct fuse_req *req, struct fuse_file *ff, |
2106cb18 | 940 | loff_t pos, size_t count) |
b6aeaded | 941 | { |
b25e82e5 MS |
942 | struct fuse_write_in *inarg = &req->misc.write.in; |
943 | struct fuse_write_out *outarg = &req->misc.write.out; | |
b6aeaded | 944 | |
b25e82e5 MS |
945 | inarg->fh = ff->fh; |
946 | inarg->offset = pos; | |
947 | inarg->size = count; | |
b6aeaded | 948 | req->in.h.opcode = FUSE_WRITE; |
2106cb18 | 949 | req->in.h.nodeid = ff->nodeid; |
b6aeaded | 950 | req->in.numargs = 2; |
2106cb18 | 951 | if (ff->fc->minor < 9) |
f3332114 MS |
952 | req->in.args[0].size = FUSE_COMPAT_WRITE_IN_SIZE; |
953 | else | |
954 | req->in.args[0].size = sizeof(struct fuse_write_in); | |
b25e82e5 | 955 | req->in.args[0].value = inarg; |
b6aeaded MS |
956 | req->in.args[1].size = count; |
957 | req->out.numargs = 1; | |
958 | req->out.args[0].size = sizeof(struct fuse_write_out); | |
b25e82e5 MS |
959 | req->out.args[0].value = outarg; |
960 | } | |
961 | ||
36cf66ed | 962 | static size_t fuse_send_write(struct fuse_req *req, struct fuse_io_priv *io, |
2106cb18 | 963 | loff_t pos, size_t count, fl_owner_t owner) |
b25e82e5 | 964 | { |
e1c0eecb MS |
965 | struct kiocb *iocb = io->iocb; |
966 | struct file *file = iocb->ki_filp; | |
2106cb18 MS |
967 | struct fuse_file *ff = file->private_data; |
968 | struct fuse_conn *fc = ff->fc; | |
2d698b07 MS |
969 | struct fuse_write_in *inarg = &req->misc.write.in; |
970 | ||
2106cb18 | 971 | fuse_write_fill(req, ff, pos, count); |
2d698b07 | 972 | inarg->flags = file->f_flags; |
e1c0eecb MS |
973 | if (iocb->ki_flags & IOCB_DSYNC) |
974 | inarg->flags |= O_DSYNC; | |
975 | if (iocb->ki_flags & IOCB_SYNC) | |
976 | inarg->flags |= O_SYNC; | |
f3332114 | 977 | if (owner != NULL) { |
f3332114 MS |
978 | inarg->write_flags |= FUSE_WRITE_LOCKOWNER; |
979 | inarg->lock_owner = fuse_lock_owner_id(fc, owner); | |
980 | } | |
36cf66ed MP |
981 | |
982 | if (io->async) | |
983 | return fuse_async_req_send(fc, req, count, io); | |
984 | ||
b93f858a | 985 | fuse_request_send(fc, req); |
b25e82e5 | 986 | return req->misc.write.out.size; |
b6aeaded MS |
987 | } |
988 | ||
b0aa7606 | 989 | bool fuse_write_update_size(struct inode *inode, loff_t pos) |
854512ec MS |
990 | { |
991 | struct fuse_conn *fc = get_fuse_conn(inode); | |
992 | struct fuse_inode *fi = get_fuse_inode(inode); | |
b0aa7606 | 993 | bool ret = false; |
854512ec MS |
994 | |
995 | spin_lock(&fc->lock); | |
996 | fi->attr_version = ++fc->attr_version; | |
b0aa7606 | 997 | if (pos > inode->i_size) { |
854512ec | 998 | i_size_write(inode, pos); |
b0aa7606 MP |
999 | ret = true; |
1000 | } | |
854512ec | 1001 | spin_unlock(&fc->lock); |
b0aa7606 MP |
1002 | |
1003 | return ret; | |
854512ec MS |
1004 | } |
1005 | ||
e1c0eecb | 1006 | static size_t fuse_send_write_pages(struct fuse_req *req, struct kiocb *iocb, |
ea9b9907 NP |
1007 | struct inode *inode, loff_t pos, |
1008 | size_t count) | |
1009 | { | |
1010 | size_t res; | |
1011 | unsigned offset; | |
1012 | unsigned i; | |
e1c0eecb | 1013 | struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb); |
ea9b9907 NP |
1014 | |
1015 | for (i = 0; i < req->num_pages; i++) | |
1016 | fuse_wait_on_page_writeback(inode, req->pages[i]->index); | |
1017 | ||
36cf66ed | 1018 | res = fuse_send_write(req, &io, pos, count, NULL); |
ea9b9907 | 1019 | |
b2430d75 | 1020 | offset = req->page_descs[0].offset; |
ea9b9907 NP |
1021 | count = res; |
1022 | for (i = 0; i < req->num_pages; i++) { | |
1023 | struct page *page = req->pages[i]; | |
1024 | ||
09cbfeaf | 1025 | if (!req->out.h.error && !offset && count >= PAGE_SIZE) |
ea9b9907 NP |
1026 | SetPageUptodate(page); |
1027 | ||
09cbfeaf KS |
1028 | if (count > PAGE_SIZE - offset) |
1029 | count -= PAGE_SIZE - offset; | |
ea9b9907 NP |
1030 | else |
1031 | count = 0; | |
1032 | offset = 0; | |
1033 | ||
1034 | unlock_page(page); | |
09cbfeaf | 1035 | put_page(page); |
ea9b9907 NP |
1036 | } |
1037 | ||
1038 | return res; | |
1039 | } | |
1040 | ||
1041 | static ssize_t fuse_fill_write_pages(struct fuse_req *req, | |
1042 | struct address_space *mapping, | |
1043 | struct iov_iter *ii, loff_t pos) | |
1044 | { | |
1045 | struct fuse_conn *fc = get_fuse_conn(mapping->host); | |
09cbfeaf | 1046 | unsigned offset = pos & (PAGE_SIZE - 1); |
ea9b9907 NP |
1047 | size_t count = 0; |
1048 | int err; | |
1049 | ||
f4975c67 | 1050 | req->in.argpages = 1; |
b2430d75 | 1051 | req->page_descs[0].offset = offset; |
ea9b9907 NP |
1052 | |
1053 | do { | |
1054 | size_t tmp; | |
1055 | struct page *page; | |
09cbfeaf KS |
1056 | pgoff_t index = pos >> PAGE_SHIFT; |
1057 | size_t bytes = min_t(size_t, PAGE_SIZE - offset, | |
ea9b9907 NP |
1058 | iov_iter_count(ii)); |
1059 | ||
1060 | bytes = min_t(size_t, bytes, fc->max_write - count); | |
1061 | ||
1062 | again: | |
1063 | err = -EFAULT; | |
1064 | if (iov_iter_fault_in_readable(ii, bytes)) | |
1065 | break; | |
1066 | ||
1067 | err = -ENOMEM; | |
54566b2c | 1068 | page = grab_cache_page_write_begin(mapping, index, 0); |
ea9b9907 NP |
1069 | if (!page) |
1070 | break; | |
1071 | ||
931e80e4 | 1072 | if (mapping_writably_mapped(mapping)) |
1073 | flush_dcache_page(page); | |
1074 | ||
ea9b9907 | 1075 | tmp = iov_iter_copy_from_user_atomic(page, ii, offset, bytes); |
ea9b9907 NP |
1076 | flush_dcache_page(page); |
1077 | ||
3ca8138f | 1078 | iov_iter_advance(ii, tmp); |
ea9b9907 NP |
1079 | if (!tmp) { |
1080 | unlock_page(page); | |
09cbfeaf | 1081 | put_page(page); |
ea9b9907 NP |
1082 | bytes = min(bytes, iov_iter_single_seg_count(ii)); |
1083 | goto again; | |
1084 | } | |
1085 | ||
1086 | err = 0; | |
1087 | req->pages[req->num_pages] = page; | |
85f40aec | 1088 | req->page_descs[req->num_pages].length = tmp; |
ea9b9907 NP |
1089 | req->num_pages++; |
1090 | ||
ea9b9907 NP |
1091 | count += tmp; |
1092 | pos += tmp; | |
1093 | offset += tmp; | |
09cbfeaf | 1094 | if (offset == PAGE_SIZE) |
ea9b9907 NP |
1095 | offset = 0; |
1096 | ||
78bb6cb9 MS |
1097 | if (!fc->big_writes) |
1098 | break; | |
ea9b9907 | 1099 | } while (iov_iter_count(ii) && count < fc->max_write && |
d07f09f5 | 1100 | req->num_pages < req->max_pages && offset == 0); |
ea9b9907 NP |
1101 | |
1102 | return count > 0 ? count : err; | |
1103 | } | |
1104 | ||
d07f09f5 MP |
1105 | static inline unsigned fuse_wr_pages(loff_t pos, size_t len) |
1106 | { | |
1107 | return min_t(unsigned, | |
09cbfeaf KS |
1108 | ((pos + len - 1) >> PAGE_SHIFT) - |
1109 | (pos >> PAGE_SHIFT) + 1, | |
d07f09f5 MP |
1110 | FUSE_MAX_PAGES_PER_REQ); |
1111 | } | |
1112 | ||
e1c0eecb | 1113 | static ssize_t fuse_perform_write(struct kiocb *iocb, |
ea9b9907 NP |
1114 | struct address_space *mapping, |
1115 | struct iov_iter *ii, loff_t pos) | |
1116 | { | |
1117 | struct inode *inode = mapping->host; | |
1118 | struct fuse_conn *fc = get_fuse_conn(inode); | |
06a7c3c2 | 1119 | struct fuse_inode *fi = get_fuse_inode(inode); |
ea9b9907 NP |
1120 | int err = 0; |
1121 | ssize_t res = 0; | |
1122 | ||
1123 | if (is_bad_inode(inode)) | |
1124 | return -EIO; | |
1125 | ||
06a7c3c2 MP |
1126 | if (inode->i_size < pos + iov_iter_count(ii)) |
1127 | set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); | |
1128 | ||
ea9b9907 NP |
1129 | do { |
1130 | struct fuse_req *req; | |
1131 | ssize_t count; | |
d07f09f5 | 1132 | unsigned nr_pages = fuse_wr_pages(pos, iov_iter_count(ii)); |
ea9b9907 | 1133 | |
d07f09f5 | 1134 | req = fuse_get_req(fc, nr_pages); |
ea9b9907 NP |
1135 | if (IS_ERR(req)) { |
1136 | err = PTR_ERR(req); | |
1137 | break; | |
1138 | } | |
1139 | ||
1140 | count = fuse_fill_write_pages(req, mapping, ii, pos); | |
1141 | if (count <= 0) { | |
1142 | err = count; | |
1143 | } else { | |
1144 | size_t num_written; | |
1145 | ||
e1c0eecb | 1146 | num_written = fuse_send_write_pages(req, iocb, inode, |
ea9b9907 NP |
1147 | pos, count); |
1148 | err = req->out.h.error; | |
1149 | if (!err) { | |
1150 | res += num_written; | |
1151 | pos += num_written; | |
1152 | ||
1153 | /* break out of the loop on short write */ | |
1154 | if (num_written != count) | |
1155 | err = -EIO; | |
1156 | } | |
1157 | } | |
1158 | fuse_put_request(fc, req); | |
1159 | } while (!err && iov_iter_count(ii)); | |
1160 | ||
1161 | if (res > 0) | |
1162 | fuse_write_update_size(inode, pos); | |
1163 | ||
06a7c3c2 | 1164 | clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); |
ea9b9907 NP |
1165 | fuse_invalidate_attr(inode); |
1166 | ||
1167 | return res > 0 ? res : err; | |
1168 | } | |
1169 | ||
84c3d55c | 1170 | static ssize_t fuse_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
ea9b9907 NP |
1171 | { |
1172 | struct file *file = iocb->ki_filp; | |
1173 | struct address_space *mapping = file->f_mapping; | |
ea9b9907 | 1174 | ssize_t written = 0; |
4273b793 | 1175 | ssize_t written_buffered = 0; |
ea9b9907 NP |
1176 | struct inode *inode = mapping->host; |
1177 | ssize_t err; | |
4273b793 | 1178 | loff_t endbyte = 0; |
ea9b9907 | 1179 | |
4d99ff8f PE |
1180 | if (get_fuse_conn(inode)->writeback_cache) { |
1181 | /* Update size (EOF optimization) and mode (SUID clearing) */ | |
5b97eeac | 1182 | err = fuse_update_attributes(mapping->host, file); |
4d99ff8f PE |
1183 | if (err) |
1184 | return err; | |
1185 | ||
84c3d55c | 1186 | return generic_file_write_iter(iocb, from); |
4d99ff8f PE |
1187 | } |
1188 | ||
5955102c | 1189 | inode_lock(inode); |
ea9b9907 NP |
1190 | |
1191 | /* We can write back this queue in page reclaim */ | |
de1414a6 | 1192 | current->backing_dev_info = inode_to_bdi(inode); |
ea9b9907 | 1193 | |
3309dd04 AV |
1194 | err = generic_write_checks(iocb, from); |
1195 | if (err <= 0) | |
ea9b9907 NP |
1196 | goto out; |
1197 | ||
5fa8e0a1 | 1198 | err = file_remove_privs(file); |
ea9b9907 NP |
1199 | if (err) |
1200 | goto out; | |
1201 | ||
c3b2da31 JB |
1202 | err = file_update_time(file); |
1203 | if (err) | |
1204 | goto out; | |
ea9b9907 | 1205 | |
2ba48ce5 | 1206 | if (iocb->ki_flags & IOCB_DIRECT) { |
3309dd04 | 1207 | loff_t pos = iocb->ki_pos; |
1af5bb49 | 1208 | written = generic_file_direct_write(iocb, from); |
84c3d55c | 1209 | if (written < 0 || !iov_iter_count(from)) |
4273b793 AA |
1210 | goto out; |
1211 | ||
1212 | pos += written; | |
ea9b9907 | 1213 | |
e1c0eecb | 1214 | written_buffered = fuse_perform_write(iocb, mapping, from, pos); |
4273b793 AA |
1215 | if (written_buffered < 0) { |
1216 | err = written_buffered; | |
1217 | goto out; | |
1218 | } | |
1219 | endbyte = pos + written_buffered - 1; | |
1220 | ||
1221 | err = filemap_write_and_wait_range(file->f_mapping, pos, | |
1222 | endbyte); | |
1223 | if (err) | |
1224 | goto out; | |
1225 | ||
1226 | invalidate_mapping_pages(file->f_mapping, | |
09cbfeaf KS |
1227 | pos >> PAGE_SHIFT, |
1228 | endbyte >> PAGE_SHIFT); | |
4273b793 AA |
1229 | |
1230 | written += written_buffered; | |
1231 | iocb->ki_pos = pos + written_buffered; | |
1232 | } else { | |
e1c0eecb | 1233 | written = fuse_perform_write(iocb, mapping, from, iocb->ki_pos); |
4273b793 | 1234 | if (written >= 0) |
3309dd04 | 1235 | iocb->ki_pos += written; |
4273b793 | 1236 | } |
ea9b9907 NP |
1237 | out: |
1238 | current->backing_dev_info = NULL; | |
5955102c | 1239 | inode_unlock(inode); |
e1c0eecb MS |
1240 | if (written > 0) |
1241 | written = generic_write_sync(iocb, written); | |
ea9b9907 NP |
1242 | |
1243 | return written ? written : err; | |
1244 | } | |
1245 | ||
7c190c8b MP |
1246 | static inline void fuse_page_descs_length_init(struct fuse_req *req, |
1247 | unsigned index, unsigned nr_pages) | |
85f40aec MP |
1248 | { |
1249 | int i; | |
1250 | ||
7c190c8b | 1251 | for (i = index; i < index + nr_pages; i++) |
85f40aec MP |
1252 | req->page_descs[i].length = PAGE_SIZE - |
1253 | req->page_descs[i].offset; | |
1254 | } | |
1255 | ||
7c190c8b MP |
1256 | static inline unsigned long fuse_get_user_addr(const struct iov_iter *ii) |
1257 | { | |
1258 | return (unsigned long)ii->iov->iov_base + ii->iov_offset; | |
1259 | } | |
1260 | ||
1261 | static inline size_t fuse_get_frag_size(const struct iov_iter *ii, | |
1262 | size_t max_size) | |
1263 | { | |
1264 | return min(iov_iter_single_seg_count(ii), max_size); | |
1265 | } | |
1266 | ||
b98d023a | 1267 | static int fuse_get_user_pages(struct fuse_req *req, struct iov_iter *ii, |
ce60a2f1 | 1268 | size_t *nbytesp, int write) |
413ef8cb | 1269 | { |
7c190c8b | 1270 | size_t nbytes = 0; /* # bytes already packed in req */ |
742f9927 | 1271 | ssize_t ret = 0; |
b98d023a | 1272 | |
f4975c67 | 1273 | /* Special case for kernel I/O: can copy directly into the buffer */ |
00e23707 | 1274 | if (iov_iter_is_kvec(ii)) { |
7c190c8b MP |
1275 | unsigned long user_addr = fuse_get_user_addr(ii); |
1276 | size_t frag_size = fuse_get_frag_size(ii, *nbytesp); | |
1277 | ||
f4975c67 MS |
1278 | if (write) |
1279 | req->in.args[1].value = (void *) user_addr; | |
1280 | else | |
1281 | req->out.args[0].value = (void *) user_addr; | |
1282 | ||
b98d023a MP |
1283 | iov_iter_advance(ii, frag_size); |
1284 | *nbytesp = frag_size; | |
f4975c67 MS |
1285 | return 0; |
1286 | } | |
413ef8cb | 1287 | |
5565a9d8 | 1288 | while (nbytes < *nbytesp && req->num_pages < req->max_pages) { |
7c190c8b | 1289 | unsigned npages; |
f67da30c | 1290 | size_t start; |
742f9927 | 1291 | ret = iov_iter_get_pages(ii, &req->pages[req->num_pages], |
2c80929c | 1292 | *nbytesp - nbytes, |
c7f3888a AV |
1293 | req->max_pages - req->num_pages, |
1294 | &start); | |
7c190c8b | 1295 | if (ret < 0) |
742f9927 | 1296 | break; |
7c190c8b | 1297 | |
c9c37e2e AV |
1298 | iov_iter_advance(ii, ret); |
1299 | nbytes += ret; | |
7c190c8b | 1300 | |
c9c37e2e AV |
1301 | ret += start; |
1302 | npages = (ret + PAGE_SIZE - 1) / PAGE_SIZE; | |
7c190c8b | 1303 | |
c9c37e2e | 1304 | req->page_descs[req->num_pages].offset = start; |
7c190c8b MP |
1305 | fuse_page_descs_length_init(req, req->num_pages, npages); |
1306 | ||
1307 | req->num_pages += npages; | |
1308 | req->page_descs[req->num_pages - 1].length -= | |
c9c37e2e | 1309 | (PAGE_SIZE - ret) & (PAGE_SIZE - 1); |
7c190c8b | 1310 | } |
f4975c67 MS |
1311 | |
1312 | if (write) | |
1313 | req->in.argpages = 1; | |
1314 | else | |
1315 | req->out.argpages = 1; | |
1316 | ||
7c190c8b | 1317 | *nbytesp = nbytes; |
f4975c67 | 1318 | |
2c932d4c | 1319 | return ret < 0 ? ret : 0; |
413ef8cb MS |
1320 | } |
1321 | ||
5565a9d8 MP |
1322 | static inline int fuse_iter_npages(const struct iov_iter *ii_p) |
1323 | { | |
f67da30c | 1324 | return iov_iter_npages(ii_p, FUSE_MAX_PAGES_PER_REQ); |
5565a9d8 MP |
1325 | } |
1326 | ||
d22a943f AV |
1327 | ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter, |
1328 | loff_t *ppos, int flags) | |
413ef8cb | 1329 | { |
ea8cd333 PE |
1330 | int write = flags & FUSE_DIO_WRITE; |
1331 | int cuse = flags & FUSE_DIO_CUSE; | |
e1c0eecb | 1332 | struct file *file = io->iocb->ki_filp; |
ea8cd333 | 1333 | struct inode *inode = file->f_mapping->host; |
2106cb18 MS |
1334 | struct fuse_file *ff = file->private_data; |
1335 | struct fuse_conn *fc = ff->fc; | |
413ef8cb MS |
1336 | size_t nmax = write ? fc->max_write : fc->max_read; |
1337 | loff_t pos = *ppos; | |
d22a943f | 1338 | size_t count = iov_iter_count(iter); |
09cbfeaf KS |
1339 | pgoff_t idx_from = pos >> PAGE_SHIFT; |
1340 | pgoff_t idx_to = (pos + count - 1) >> PAGE_SHIFT; | |
413ef8cb | 1341 | ssize_t res = 0; |
248d86e8 | 1342 | struct fuse_req *req; |
742f9927 | 1343 | int err = 0; |
248d86e8 | 1344 | |
de82b923 | 1345 | if (io->async) |
d22a943f | 1346 | req = fuse_get_req_for_background(fc, fuse_iter_npages(iter)); |
de82b923 | 1347 | else |
d22a943f | 1348 | req = fuse_get_req(fc, fuse_iter_npages(iter)); |
ce1d5a49 MS |
1349 | if (IS_ERR(req)) |
1350 | return PTR_ERR(req); | |
413ef8cb | 1351 | |
ea8cd333 PE |
1352 | if (!cuse && fuse_range_is_writeback(inode, idx_from, idx_to)) { |
1353 | if (!write) | |
5955102c | 1354 | inode_lock(inode); |
ea8cd333 PE |
1355 | fuse_sync_writes(inode); |
1356 | if (!write) | |
5955102c | 1357 | inode_unlock(inode); |
ea8cd333 PE |
1358 | } |
1359 | ||
61c12b49 | 1360 | io->should_dirty = !write && iter_is_iovec(iter); |
413ef8cb | 1361 | while (count) { |
413ef8cb | 1362 | size_t nres; |
2106cb18 | 1363 | fl_owner_t owner = current->files; |
f4975c67 | 1364 | size_t nbytes = min(count, nmax); |
742f9927 AS |
1365 | err = fuse_get_user_pages(req, iter, &nbytes, write); |
1366 | if (err && !nbytes) | |
413ef8cb | 1367 | break; |
f4975c67 | 1368 | |
413ef8cb | 1369 | if (write) |
36cf66ed | 1370 | nres = fuse_send_write(req, io, pos, nbytes, owner); |
413ef8cb | 1371 | else |
36cf66ed | 1372 | nres = fuse_send_read(req, io, pos, nbytes, owner); |
2106cb18 | 1373 | |
36cf66ed | 1374 | if (!io->async) |
61c12b49 | 1375 | fuse_release_user_pages(req, io->should_dirty); |
413ef8cb | 1376 | if (req->out.h.error) { |
742f9927 | 1377 | err = req->out.h.error; |
413ef8cb MS |
1378 | break; |
1379 | } else if (nres > nbytes) { | |
742f9927 AS |
1380 | res = 0; |
1381 | err = -EIO; | |
413ef8cb MS |
1382 | break; |
1383 | } | |
1384 | count -= nres; | |
1385 | res += nres; | |
1386 | pos += nres; | |
413ef8cb MS |
1387 | if (nres != nbytes) |
1388 | break; | |
56cf34ff MS |
1389 | if (count) { |
1390 | fuse_put_request(fc, req); | |
de82b923 BF |
1391 | if (io->async) |
1392 | req = fuse_get_req_for_background(fc, | |
d22a943f | 1393 | fuse_iter_npages(iter)); |
de82b923 | 1394 | else |
d22a943f | 1395 | req = fuse_get_req(fc, fuse_iter_npages(iter)); |
56cf34ff MS |
1396 | if (IS_ERR(req)) |
1397 | break; | |
1398 | } | |
413ef8cb | 1399 | } |
f60311d5 AA |
1400 | if (!IS_ERR(req)) |
1401 | fuse_put_request(fc, req); | |
d09cb9d7 | 1402 | if (res > 0) |
413ef8cb | 1403 | *ppos = pos; |
413ef8cb | 1404 | |
742f9927 | 1405 | return res > 0 ? res : err; |
413ef8cb | 1406 | } |
08cbf542 | 1407 | EXPORT_SYMBOL_GPL(fuse_direct_io); |
413ef8cb | 1408 | |
36cf66ed | 1409 | static ssize_t __fuse_direct_read(struct fuse_io_priv *io, |
d22a943f AV |
1410 | struct iov_iter *iter, |
1411 | loff_t *ppos) | |
413ef8cb | 1412 | { |
d09cb9d7 | 1413 | ssize_t res; |
e1c0eecb | 1414 | struct inode *inode = file_inode(io->iocb->ki_filp); |
d09cb9d7 MS |
1415 | |
1416 | if (is_bad_inode(inode)) | |
1417 | return -EIO; | |
1418 | ||
d22a943f | 1419 | res = fuse_direct_io(io, iter, ppos, 0); |
d09cb9d7 MS |
1420 | |
1421 | fuse_invalidate_attr(inode); | |
1422 | ||
1423 | return res; | |
413ef8cb MS |
1424 | } |
1425 | ||
15316263 | 1426 | static ssize_t fuse_direct_read_iter(struct kiocb *iocb, struct iov_iter *to) |
b98d023a | 1427 | { |
e1c0eecb | 1428 | struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb); |
15316263 | 1429 | return __fuse_direct_read(&io, to, &iocb->ki_pos); |
b98d023a MP |
1430 | } |
1431 | ||
15316263 | 1432 | static ssize_t fuse_direct_write_iter(struct kiocb *iocb, struct iov_iter *from) |
4273b793 | 1433 | { |
e1c0eecb MS |
1434 | struct inode *inode = file_inode(iocb->ki_filp); |
1435 | struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb); | |
15316263 | 1436 | ssize_t res; |
4273b793 AA |
1437 | |
1438 | if (is_bad_inode(inode)) | |
1439 | return -EIO; | |
1440 | ||
1441 | /* Don't allow parallel writes to the same file */ | |
5955102c | 1442 | inode_lock(inode); |
3309dd04 AV |
1443 | res = generic_write_checks(iocb, from); |
1444 | if (res > 0) | |
812408fb | 1445 | res = fuse_direct_io(&io, from, &iocb->ki_pos, FUSE_DIO_WRITE); |
812408fb | 1446 | fuse_invalidate_attr(inode); |
bcba24cc | 1447 | if (res > 0) |
15316263 | 1448 | fuse_write_update_size(inode, iocb->ki_pos); |
5955102c | 1449 | inode_unlock(inode); |
4273b793 AA |
1450 | |
1451 | return res; | |
1452 | } | |
1453 | ||
3be5a52b | 1454 | static void fuse_writepage_free(struct fuse_conn *fc, struct fuse_req *req) |
b6aeaded | 1455 | { |
385b1268 PE |
1456 | int i; |
1457 | ||
1458 | for (i = 0; i < req->num_pages; i++) | |
1459 | __free_page(req->pages[i]); | |
8b284dc4 MS |
1460 | |
1461 | if (req->ff) | |
1462 | fuse_file_put(req->ff, false); | |
3be5a52b MS |
1463 | } |
1464 | ||
1465 | static void fuse_writepage_finish(struct fuse_conn *fc, struct fuse_req *req) | |
1466 | { | |
1467 | struct inode *inode = req->inode; | |
1468 | struct fuse_inode *fi = get_fuse_inode(inode); | |
de1414a6 | 1469 | struct backing_dev_info *bdi = inode_to_bdi(inode); |
385b1268 | 1470 | int i; |
3be5a52b MS |
1471 | |
1472 | list_del(&req->writepages_entry); | |
385b1268 | 1473 | for (i = 0; i < req->num_pages; i++) { |
93f78d88 | 1474 | dec_wb_stat(&bdi->wb, WB_WRITEBACK); |
11fb9989 | 1475 | dec_node_page_state(req->pages[i], NR_WRITEBACK_TEMP); |
93f78d88 | 1476 | wb_writeout_inc(&bdi->wb); |
385b1268 | 1477 | } |
3be5a52b MS |
1478 | wake_up(&fi->page_waitq); |
1479 | } | |
1480 | ||
1481 | /* Called under fc->lock, may release and reacquire it */ | |
6eaf4782 MP |
1482 | static void fuse_send_writepage(struct fuse_conn *fc, struct fuse_req *req, |
1483 | loff_t size) | |
b9ca67b2 MS |
1484 | __releases(fc->lock) |
1485 | __acquires(fc->lock) | |
3be5a52b MS |
1486 | { |
1487 | struct fuse_inode *fi = get_fuse_inode(req->inode); | |
3be5a52b | 1488 | struct fuse_write_in *inarg = &req->misc.write.in; |
09cbfeaf | 1489 | __u64 data_size = req->num_pages * PAGE_SIZE; |
3be5a52b MS |
1490 | |
1491 | if (!fc->connected) | |
1492 | goto out_free; | |
1493 | ||
385b1268 PE |
1494 | if (inarg->offset + data_size <= size) { |
1495 | inarg->size = data_size; | |
3be5a52b | 1496 | } else if (inarg->offset < size) { |
385b1268 | 1497 | inarg->size = size - inarg->offset; |
3be5a52b MS |
1498 | } else { |
1499 | /* Got truncated off completely */ | |
1500 | goto out_free; | |
b6aeaded | 1501 | } |
3be5a52b MS |
1502 | |
1503 | req->in.args[1].size = inarg->size; | |
1504 | fi->writectr++; | |
b93f858a | 1505 | fuse_request_send_background_locked(fc, req); |
3be5a52b MS |
1506 | return; |
1507 | ||
1508 | out_free: | |
1509 | fuse_writepage_finish(fc, req); | |
1510 | spin_unlock(&fc->lock); | |
1511 | fuse_writepage_free(fc, req); | |
e9bb09dd | 1512 | fuse_put_request(fc, req); |
3be5a52b | 1513 | spin_lock(&fc->lock); |
b6aeaded MS |
1514 | } |
1515 | ||
3be5a52b MS |
1516 | /* |
1517 | * If fi->writectr is positive (no truncate or fsync going on) send | |
1518 | * all queued writepage requests. | |
1519 | * | |
1520 | * Called with fc->lock | |
1521 | */ | |
1522 | void fuse_flush_writepages(struct inode *inode) | |
b9ca67b2 MS |
1523 | __releases(fc->lock) |
1524 | __acquires(fc->lock) | |
b6aeaded | 1525 | { |
3be5a52b MS |
1526 | struct fuse_conn *fc = get_fuse_conn(inode); |
1527 | struct fuse_inode *fi = get_fuse_inode(inode); | |
6eaf4782 | 1528 | size_t crop = i_size_read(inode); |
3be5a52b MS |
1529 | struct fuse_req *req; |
1530 | ||
1531 | while (fi->writectr >= 0 && !list_empty(&fi->queued_writes)) { | |
1532 | req = list_entry(fi->queued_writes.next, struct fuse_req, list); | |
1533 | list_del_init(&req->list); | |
6eaf4782 | 1534 | fuse_send_writepage(fc, req, crop); |
3be5a52b MS |
1535 | } |
1536 | } | |
1537 | ||
1538 | static void fuse_writepage_end(struct fuse_conn *fc, struct fuse_req *req) | |
1539 | { | |
1540 | struct inode *inode = req->inode; | |
1541 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1542 | ||
1543 | mapping_set_error(inode->i_mapping, req->out.h.error); | |
1544 | spin_lock(&fc->lock); | |
8b284dc4 | 1545 | while (req->misc.write.next) { |
6eaf4782 MP |
1546 | struct fuse_conn *fc = get_fuse_conn(inode); |
1547 | struct fuse_write_in *inarg = &req->misc.write.in; | |
8b284dc4 MS |
1548 | struct fuse_req *next = req->misc.write.next; |
1549 | req->misc.write.next = next->misc.write.next; | |
1550 | next->misc.write.next = NULL; | |
ce128de6 | 1551 | next->ff = fuse_file_get(req->ff); |
8b284dc4 | 1552 | list_add(&next->writepages_entry, &fi->writepages); |
6eaf4782 MP |
1553 | |
1554 | /* | |
1555 | * Skip fuse_flush_writepages() to make it easy to crop requests | |
1556 | * based on primary request size. | |
1557 | * | |
1558 | * 1st case (trivial): there are no concurrent activities using | |
1559 | * fuse_set/release_nowrite. Then we're on safe side because | |
1560 | * fuse_flush_writepages() would call fuse_send_writepage() | |
1561 | * anyway. | |
1562 | * | |
1563 | * 2nd case: someone called fuse_set_nowrite and it is waiting | |
1564 | * now for completion of all in-flight requests. This happens | |
1565 | * rarely and no more than once per page, so this should be | |
1566 | * okay. | |
1567 | * | |
1568 | * 3rd case: someone (e.g. fuse_do_setattr()) is in the middle | |
1569 | * of fuse_set_nowrite..fuse_release_nowrite section. The fact | |
1570 | * that fuse_set_nowrite returned implies that all in-flight | |
1571 | * requests were completed along with all of their secondary | |
1572 | * requests. Further primary requests are blocked by negative | |
1573 | * writectr. Hence there cannot be any in-flight requests and | |
1574 | * no invocations of fuse_writepage_end() while we're in | |
1575 | * fuse_set_nowrite..fuse_release_nowrite section. | |
1576 | */ | |
1577 | fuse_send_writepage(fc, next, inarg->offset + inarg->size); | |
8b284dc4 | 1578 | } |
3be5a52b MS |
1579 | fi->writectr--; |
1580 | fuse_writepage_finish(fc, req); | |
1581 | spin_unlock(&fc->lock); | |
1582 | fuse_writepage_free(fc, req); | |
1583 | } | |
1584 | ||
1e18bda8 MS |
1585 | static struct fuse_file *__fuse_write_file_get(struct fuse_conn *fc, |
1586 | struct fuse_inode *fi) | |
adcadfa8 | 1587 | { |
72523425 | 1588 | struct fuse_file *ff = NULL; |
adcadfa8 PE |
1589 | |
1590 | spin_lock(&fc->lock); | |
1e18bda8 | 1591 | if (!list_empty(&fi->write_files)) { |
72523425 MS |
1592 | ff = list_entry(fi->write_files.next, struct fuse_file, |
1593 | write_entry); | |
1594 | fuse_file_get(ff); | |
1595 | } | |
adcadfa8 PE |
1596 | spin_unlock(&fc->lock); |
1597 | ||
1598 | return ff; | |
1599 | } | |
1600 | ||
1e18bda8 MS |
1601 | static struct fuse_file *fuse_write_file_get(struct fuse_conn *fc, |
1602 | struct fuse_inode *fi) | |
1603 | { | |
1604 | struct fuse_file *ff = __fuse_write_file_get(fc, fi); | |
1605 | WARN_ON(!ff); | |
1606 | return ff; | |
1607 | } | |
1608 | ||
1609 | int fuse_write_inode(struct inode *inode, struct writeback_control *wbc) | |
1610 | { | |
1611 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1612 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1613 | struct fuse_file *ff; | |
1614 | int err; | |
1615 | ||
1616 | ff = __fuse_write_file_get(fc, fi); | |
ab9e13f7 | 1617 | err = fuse_flush_times(inode, ff); |
1e18bda8 MS |
1618 | if (ff) |
1619 | fuse_file_put(ff, 0); | |
1620 | ||
1621 | return err; | |
1622 | } | |
1623 | ||
3be5a52b MS |
1624 | static int fuse_writepage_locked(struct page *page) |
1625 | { | |
1626 | struct address_space *mapping = page->mapping; | |
1627 | struct inode *inode = mapping->host; | |
1628 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1629 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1630 | struct fuse_req *req; | |
3be5a52b | 1631 | struct page *tmp_page; |
72523425 | 1632 | int error = -ENOMEM; |
3be5a52b MS |
1633 | |
1634 | set_page_writeback(page); | |
1635 | ||
4250c066 | 1636 | req = fuse_request_alloc_nofs(1); |
3be5a52b MS |
1637 | if (!req) |
1638 | goto err; | |
1639 | ||
825d6d33 MS |
1640 | /* writeback always goes to bg_queue */ |
1641 | __set_bit(FR_BACKGROUND, &req->flags); | |
3be5a52b MS |
1642 | tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); |
1643 | if (!tmp_page) | |
1644 | goto err_free; | |
1645 | ||
72523425 | 1646 | error = -EIO; |
26d614df | 1647 | req->ff = fuse_write_file_get(fc, fi); |
72523425 | 1648 | if (!req->ff) |
27f1b363 | 1649 | goto err_nofile; |
72523425 | 1650 | |
adcadfa8 | 1651 | fuse_write_fill(req, req->ff, page_offset(page), 0); |
3be5a52b MS |
1652 | |
1653 | copy_highpage(tmp_page, page); | |
2d698b07 | 1654 | req->misc.write.in.write_flags |= FUSE_WRITE_CACHE; |
8b284dc4 | 1655 | req->misc.write.next = NULL; |
f4975c67 | 1656 | req->in.argpages = 1; |
3be5a52b MS |
1657 | req->num_pages = 1; |
1658 | req->pages[0] = tmp_page; | |
b2430d75 | 1659 | req->page_descs[0].offset = 0; |
85f40aec | 1660 | req->page_descs[0].length = PAGE_SIZE; |
3be5a52b MS |
1661 | req->end = fuse_writepage_end; |
1662 | req->inode = inode; | |
1663 | ||
93f78d88 | 1664 | inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK); |
11fb9989 | 1665 | inc_node_page_state(tmp_page, NR_WRITEBACK_TEMP); |
3be5a52b MS |
1666 | |
1667 | spin_lock(&fc->lock); | |
1668 | list_add(&req->writepages_entry, &fi->writepages); | |
1669 | list_add_tail(&req->list, &fi->queued_writes); | |
1670 | fuse_flush_writepages(inode); | |
1671 | spin_unlock(&fc->lock); | |
1672 | ||
4a4ac4eb MP |
1673 | end_page_writeback(page); |
1674 | ||
3be5a52b MS |
1675 | return 0; |
1676 | ||
27f1b363 MP |
1677 | err_nofile: |
1678 | __free_page(tmp_page); | |
3be5a52b MS |
1679 | err_free: |
1680 | fuse_request_free(req); | |
1681 | err: | |
9183976e | 1682 | mapping_set_error(page->mapping, error); |
3be5a52b | 1683 | end_page_writeback(page); |
72523425 | 1684 | return error; |
3be5a52b MS |
1685 | } |
1686 | ||
1687 | static int fuse_writepage(struct page *page, struct writeback_control *wbc) | |
1688 | { | |
1689 | int err; | |
1690 | ||
ff17be08 MS |
1691 | if (fuse_page_is_writeback(page->mapping->host, page->index)) { |
1692 | /* | |
1693 | * ->writepages() should be called for sync() and friends. We | |
1694 | * should only get here on direct reclaim and then we are | |
1695 | * allowed to skip a page which is already in flight | |
1696 | */ | |
1697 | WARN_ON(wbc->sync_mode == WB_SYNC_ALL); | |
1698 | ||
1699 | redirty_page_for_writepage(wbc, page); | |
1700 | return 0; | |
1701 | } | |
1702 | ||
3be5a52b MS |
1703 | err = fuse_writepage_locked(page); |
1704 | unlock_page(page); | |
1705 | ||
1706 | return err; | |
1707 | } | |
1708 | ||
26d614df PE |
1709 | struct fuse_fill_wb_data { |
1710 | struct fuse_req *req; | |
1711 | struct fuse_file *ff; | |
1712 | struct inode *inode; | |
2d033eaa | 1713 | struct page **orig_pages; |
26d614df PE |
1714 | }; |
1715 | ||
1716 | static void fuse_writepages_send(struct fuse_fill_wb_data *data) | |
1717 | { | |
1718 | struct fuse_req *req = data->req; | |
1719 | struct inode *inode = data->inode; | |
1720 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1721 | struct fuse_inode *fi = get_fuse_inode(inode); | |
2d033eaa MP |
1722 | int num_pages = req->num_pages; |
1723 | int i; | |
26d614df PE |
1724 | |
1725 | req->ff = fuse_file_get(data->ff); | |
1726 | spin_lock(&fc->lock); | |
1727 | list_add_tail(&req->list, &fi->queued_writes); | |
1728 | fuse_flush_writepages(inode); | |
1729 | spin_unlock(&fc->lock); | |
2d033eaa MP |
1730 | |
1731 | for (i = 0; i < num_pages; i++) | |
1732 | end_page_writeback(data->orig_pages[i]); | |
26d614df PE |
1733 | } |
1734 | ||
8b284dc4 MS |
1735 | static bool fuse_writepage_in_flight(struct fuse_req *new_req, |
1736 | struct page *page) | |
1737 | { | |
1738 | struct fuse_conn *fc = get_fuse_conn(new_req->inode); | |
1739 | struct fuse_inode *fi = get_fuse_inode(new_req->inode); | |
1740 | struct fuse_req *tmp; | |
1741 | struct fuse_req *old_req; | |
1742 | bool found = false; | |
1743 | pgoff_t curr_index; | |
1744 | ||
1745 | BUG_ON(new_req->num_pages != 0); | |
1746 | ||
1747 | spin_lock(&fc->lock); | |
1748 | list_del(&new_req->writepages_entry); | |
8b284dc4 MS |
1749 | list_for_each_entry(old_req, &fi->writepages, writepages_entry) { |
1750 | BUG_ON(old_req->inode != new_req->inode); | |
09cbfeaf | 1751 | curr_index = old_req->misc.write.in.offset >> PAGE_SHIFT; |
8b284dc4 MS |
1752 | if (curr_index <= page->index && |
1753 | page->index < curr_index + old_req->num_pages) { | |
1754 | found = true; | |
1755 | break; | |
1756 | } | |
1757 | } | |
f6011081 MP |
1758 | if (!found) { |
1759 | list_add(&new_req->writepages_entry, &fi->writepages); | |
8b284dc4 | 1760 | goto out_unlock; |
f6011081 | 1761 | } |
8b284dc4 | 1762 | |
f6011081 | 1763 | new_req->num_pages = 1; |
8b284dc4 MS |
1764 | for (tmp = old_req; tmp != NULL; tmp = tmp->misc.write.next) { |
1765 | BUG_ON(tmp->inode != new_req->inode); | |
09cbfeaf | 1766 | curr_index = tmp->misc.write.in.offset >> PAGE_SHIFT; |
8b284dc4 MS |
1767 | if (tmp->num_pages == 1 && |
1768 | curr_index == page->index) { | |
1769 | old_req = tmp; | |
1770 | } | |
1771 | } | |
1772 | ||
33e14b4d | 1773 | if (old_req->num_pages == 1 && test_bit(FR_PENDING, &old_req->flags)) { |
de1414a6 | 1774 | struct backing_dev_info *bdi = inode_to_bdi(page->mapping->host); |
41b6e41f | 1775 | |
8b284dc4 MS |
1776 | copy_highpage(old_req->pages[0], page); |
1777 | spin_unlock(&fc->lock); | |
1778 | ||
93f78d88 | 1779 | dec_wb_stat(&bdi->wb, WB_WRITEBACK); |
11fb9989 | 1780 | dec_node_page_state(page, NR_WRITEBACK_TEMP); |
93f78d88 | 1781 | wb_writeout_inc(&bdi->wb); |
8b284dc4 MS |
1782 | fuse_writepage_free(fc, new_req); |
1783 | fuse_request_free(new_req); | |
1784 | goto out; | |
1785 | } else { | |
1786 | new_req->misc.write.next = old_req->misc.write.next; | |
1787 | old_req->misc.write.next = new_req; | |
1788 | } | |
1789 | out_unlock: | |
1790 | spin_unlock(&fc->lock); | |
1791 | out: | |
1792 | return found; | |
1793 | } | |
1794 | ||
26d614df PE |
1795 | static int fuse_writepages_fill(struct page *page, |
1796 | struct writeback_control *wbc, void *_data) | |
1797 | { | |
1798 | struct fuse_fill_wb_data *data = _data; | |
1799 | struct fuse_req *req = data->req; | |
1800 | struct inode *inode = data->inode; | |
1801 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1802 | struct page *tmp_page; | |
8b284dc4 | 1803 | bool is_writeback; |
26d614df PE |
1804 | int err; |
1805 | ||
1806 | if (!data->ff) { | |
1807 | err = -EIO; | |
1808 | data->ff = fuse_write_file_get(fc, get_fuse_inode(inode)); | |
1809 | if (!data->ff) | |
1810 | goto out_unlock; | |
1811 | } | |
1812 | ||
8b284dc4 MS |
1813 | /* |
1814 | * Being under writeback is unlikely but possible. For example direct | |
1815 | * read to an mmaped fuse file will set the page dirty twice; once when | |
1816 | * the pages are faulted with get_user_pages(), and then after the read | |
1817 | * completed. | |
1818 | */ | |
1819 | is_writeback = fuse_page_is_writeback(inode, page->index); | |
1820 | ||
1821 | if (req && req->num_pages && | |
1822 | (is_writeback || req->num_pages == FUSE_MAX_PAGES_PER_REQ || | |
09cbfeaf | 1823 | (req->num_pages + 1) * PAGE_SIZE > fc->max_write || |
8b284dc4 MS |
1824 | data->orig_pages[req->num_pages - 1]->index + 1 != page->index)) { |
1825 | fuse_writepages_send(data); | |
1826 | data->req = NULL; | |
26d614df PE |
1827 | } |
1828 | err = -ENOMEM; | |
1829 | tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
1830 | if (!tmp_page) | |
1831 | goto out_unlock; | |
1832 | ||
1833 | /* | |
1834 | * The page must not be redirtied until the writeout is completed | |
1835 | * (i.e. userspace has sent a reply to the write request). Otherwise | |
1836 | * there could be more than one temporary page instance for each real | |
1837 | * page. | |
1838 | * | |
1839 | * This is ensured by holding the page lock in page_mkwrite() while | |
1840 | * checking fuse_page_is_writeback(). We already hold the page lock | |
1841 | * since clear_page_dirty_for_io() and keep it held until we add the | |
1842 | * request to the fi->writepages list and increment req->num_pages. | |
1843 | * After this fuse_page_is_writeback() will indicate that the page is | |
1844 | * under writeback, so we can release the page lock. | |
1845 | */ | |
1846 | if (data->req == NULL) { | |
1847 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1848 | ||
1849 | err = -ENOMEM; | |
1850 | req = fuse_request_alloc_nofs(FUSE_MAX_PAGES_PER_REQ); | |
1851 | if (!req) { | |
1852 | __free_page(tmp_page); | |
1853 | goto out_unlock; | |
1854 | } | |
1855 | ||
1856 | fuse_write_fill(req, data->ff, page_offset(page), 0); | |
1857 | req->misc.write.in.write_flags |= FUSE_WRITE_CACHE; | |
8b284dc4 | 1858 | req->misc.write.next = NULL; |
26d614df | 1859 | req->in.argpages = 1; |
825d6d33 | 1860 | __set_bit(FR_BACKGROUND, &req->flags); |
26d614df PE |
1861 | req->num_pages = 0; |
1862 | req->end = fuse_writepage_end; | |
1863 | req->inode = inode; | |
1864 | ||
1865 | spin_lock(&fc->lock); | |
1866 | list_add(&req->writepages_entry, &fi->writepages); | |
1867 | spin_unlock(&fc->lock); | |
1868 | ||
1869 | data->req = req; | |
1870 | } | |
1871 | set_page_writeback(page); | |
1872 | ||
1873 | copy_highpage(tmp_page, page); | |
1874 | req->pages[req->num_pages] = tmp_page; | |
1875 | req->page_descs[req->num_pages].offset = 0; | |
1876 | req->page_descs[req->num_pages].length = PAGE_SIZE; | |
1877 | ||
93f78d88 | 1878 | inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK); |
11fb9989 | 1879 | inc_node_page_state(tmp_page, NR_WRITEBACK_TEMP); |
8b284dc4 MS |
1880 | |
1881 | err = 0; | |
1882 | if (is_writeback && fuse_writepage_in_flight(req, page)) { | |
1883 | end_page_writeback(page); | |
1884 | data->req = NULL; | |
1885 | goto out_unlock; | |
1886 | } | |
2d033eaa | 1887 | data->orig_pages[req->num_pages] = page; |
26d614df PE |
1888 | |
1889 | /* | |
1890 | * Protected by fc->lock against concurrent access by | |
1891 | * fuse_page_is_writeback(). | |
1892 | */ | |
1893 | spin_lock(&fc->lock); | |
1894 | req->num_pages++; | |
1895 | spin_unlock(&fc->lock); | |
1896 | ||
26d614df PE |
1897 | out_unlock: |
1898 | unlock_page(page); | |
1899 | ||
1900 | return err; | |
1901 | } | |
1902 | ||
1903 | static int fuse_writepages(struct address_space *mapping, | |
1904 | struct writeback_control *wbc) | |
1905 | { | |
1906 | struct inode *inode = mapping->host; | |
1907 | struct fuse_fill_wb_data data; | |
1908 | int err; | |
1909 | ||
1910 | err = -EIO; | |
1911 | if (is_bad_inode(inode)) | |
1912 | goto out; | |
1913 | ||
1914 | data.inode = inode; | |
1915 | data.req = NULL; | |
1916 | data.ff = NULL; | |
1917 | ||
2d033eaa | 1918 | err = -ENOMEM; |
f2b3455e FF |
1919 | data.orig_pages = kcalloc(FUSE_MAX_PAGES_PER_REQ, |
1920 | sizeof(struct page *), | |
2d033eaa MP |
1921 | GFP_NOFS); |
1922 | if (!data.orig_pages) | |
1923 | goto out; | |
1924 | ||
26d614df PE |
1925 | err = write_cache_pages(mapping, wbc, fuse_writepages_fill, &data); |
1926 | if (data.req) { | |
1927 | /* Ignore errors if we can write at least one page */ | |
1928 | BUG_ON(!data.req->num_pages); | |
1929 | fuse_writepages_send(&data); | |
1930 | err = 0; | |
1931 | } | |
1932 | if (data.ff) | |
1933 | fuse_file_put(data.ff, false); | |
2d033eaa MP |
1934 | |
1935 | kfree(data.orig_pages); | |
26d614df PE |
1936 | out: |
1937 | return err; | |
1938 | } | |
1939 | ||
6b12c1b3 PE |
1940 | /* |
1941 | * It's worthy to make sure that space is reserved on disk for the write, | |
1942 | * but how to implement it without killing performance need more thinking. | |
1943 | */ | |
1944 | static int fuse_write_begin(struct file *file, struct address_space *mapping, | |
1945 | loff_t pos, unsigned len, unsigned flags, | |
1946 | struct page **pagep, void **fsdata) | |
1947 | { | |
09cbfeaf | 1948 | pgoff_t index = pos >> PAGE_SHIFT; |
a455589f | 1949 | struct fuse_conn *fc = get_fuse_conn(file_inode(file)); |
6b12c1b3 PE |
1950 | struct page *page; |
1951 | loff_t fsize; | |
1952 | int err = -ENOMEM; | |
1953 | ||
1954 | WARN_ON(!fc->writeback_cache); | |
1955 | ||
1956 | page = grab_cache_page_write_begin(mapping, index, flags); | |
1957 | if (!page) | |
1958 | goto error; | |
1959 | ||
1960 | fuse_wait_on_page_writeback(mapping->host, page->index); | |
1961 | ||
09cbfeaf | 1962 | if (PageUptodate(page) || len == PAGE_SIZE) |
6b12c1b3 PE |
1963 | goto success; |
1964 | /* | |
1965 | * Check if the start this page comes after the end of file, in which | |
1966 | * case the readpage can be optimized away. | |
1967 | */ | |
1968 | fsize = i_size_read(mapping->host); | |
09cbfeaf KS |
1969 | if (fsize <= (pos & PAGE_MASK)) { |
1970 | size_t off = pos & ~PAGE_MASK; | |
6b12c1b3 PE |
1971 | if (off) |
1972 | zero_user_segment(page, 0, off); | |
1973 | goto success; | |
1974 | } | |
1975 | err = fuse_do_readpage(file, page); | |
1976 | if (err) | |
1977 | goto cleanup; | |
1978 | success: | |
1979 | *pagep = page; | |
1980 | return 0; | |
1981 | ||
1982 | cleanup: | |
1983 | unlock_page(page); | |
09cbfeaf | 1984 | put_page(page); |
6b12c1b3 PE |
1985 | error: |
1986 | return err; | |
1987 | } | |
1988 | ||
1989 | static int fuse_write_end(struct file *file, struct address_space *mapping, | |
1990 | loff_t pos, unsigned len, unsigned copied, | |
1991 | struct page *page, void *fsdata) | |
1992 | { | |
1993 | struct inode *inode = page->mapping->host; | |
1994 | ||
59c3b76c MS |
1995 | /* Haven't copied anything? Skip zeroing, size extending, dirtying. */ |
1996 | if (!copied) | |
1997 | goto unlock; | |
1998 | ||
6b12c1b3 PE |
1999 | if (!PageUptodate(page)) { |
2000 | /* Zero any unwritten bytes at the end of the page */ | |
09cbfeaf | 2001 | size_t endoff = (pos + copied) & ~PAGE_MASK; |
6b12c1b3 | 2002 | if (endoff) |
09cbfeaf | 2003 | zero_user_segment(page, endoff, PAGE_SIZE); |
6b12c1b3 PE |
2004 | SetPageUptodate(page); |
2005 | } | |
2006 | ||
2007 | fuse_write_update_size(inode, pos + copied); | |
2008 | set_page_dirty(page); | |
59c3b76c MS |
2009 | |
2010 | unlock: | |
6b12c1b3 | 2011 | unlock_page(page); |
09cbfeaf | 2012 | put_page(page); |
6b12c1b3 PE |
2013 | |
2014 | return copied; | |
2015 | } | |
2016 | ||
3be5a52b MS |
2017 | static int fuse_launder_page(struct page *page) |
2018 | { | |
2019 | int err = 0; | |
2020 | if (clear_page_dirty_for_io(page)) { | |
2021 | struct inode *inode = page->mapping->host; | |
2022 | err = fuse_writepage_locked(page); | |
2023 | if (!err) | |
2024 | fuse_wait_on_page_writeback(inode, page->index); | |
2025 | } | |
2026 | return err; | |
2027 | } | |
2028 | ||
2029 | /* | |
2030 | * Write back dirty pages now, because there may not be any suitable | |
2031 | * open files later | |
2032 | */ | |
2033 | static void fuse_vma_close(struct vm_area_struct *vma) | |
2034 | { | |
2035 | filemap_write_and_wait(vma->vm_file->f_mapping); | |
2036 | } | |
2037 | ||
2038 | /* | |
2039 | * Wait for writeback against this page to complete before allowing it | |
2040 | * to be marked dirty again, and hence written back again, possibly | |
2041 | * before the previous writepage completed. | |
2042 | * | |
2043 | * Block here, instead of in ->writepage(), so that the userspace fs | |
2044 | * can only block processes actually operating on the filesystem. | |
2045 | * | |
2046 | * Otherwise unprivileged userspace fs would be able to block | |
2047 | * unrelated: | |
2048 | * | |
2049 | * - page migration | |
2050 | * - sync(2) | |
2051 | * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER | |
2052 | */ | |
46fb504a | 2053 | static vm_fault_t fuse_page_mkwrite(struct vm_fault *vmf) |
3be5a52b | 2054 | { |
c2ec175c | 2055 | struct page *page = vmf->page; |
11bac800 | 2056 | struct inode *inode = file_inode(vmf->vma->vm_file); |
cca24370 | 2057 | |
11bac800 | 2058 | file_update_time(vmf->vma->vm_file); |
cca24370 MS |
2059 | lock_page(page); |
2060 | if (page->mapping != inode->i_mapping) { | |
2061 | unlock_page(page); | |
2062 | return VM_FAULT_NOPAGE; | |
2063 | } | |
3be5a52b MS |
2064 | |
2065 | fuse_wait_on_page_writeback(inode, page->index); | |
cca24370 | 2066 | return VM_FAULT_LOCKED; |
3be5a52b MS |
2067 | } |
2068 | ||
f0f37e2f | 2069 | static const struct vm_operations_struct fuse_file_vm_ops = { |
3be5a52b MS |
2070 | .close = fuse_vma_close, |
2071 | .fault = filemap_fault, | |
f1820361 | 2072 | .map_pages = filemap_map_pages, |
3be5a52b MS |
2073 | .page_mkwrite = fuse_page_mkwrite, |
2074 | }; | |
2075 | ||
2076 | static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma) | |
2077 | { | |
650b22b9 PE |
2078 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) |
2079 | fuse_link_write_file(file); | |
2080 | ||
3be5a52b MS |
2081 | file_accessed(file); |
2082 | vma->vm_ops = &fuse_file_vm_ops; | |
b6aeaded MS |
2083 | return 0; |
2084 | } | |
2085 | ||
fc280c96 MS |
2086 | static int fuse_direct_mmap(struct file *file, struct vm_area_struct *vma) |
2087 | { | |
2088 | /* Can't provide the coherency needed for MAP_SHARED */ | |
2089 | if (vma->vm_flags & VM_MAYSHARE) | |
2090 | return -ENODEV; | |
2091 | ||
3121bfe7 MS |
2092 | invalidate_inode_pages2(file->f_mapping); |
2093 | ||
fc280c96 MS |
2094 | return generic_file_mmap(file, vma); |
2095 | } | |
2096 | ||
0b6e9ea0 SF |
2097 | static int convert_fuse_file_lock(struct fuse_conn *fc, |
2098 | const struct fuse_file_lock *ffl, | |
71421259 MS |
2099 | struct file_lock *fl) |
2100 | { | |
2101 | switch (ffl->type) { | |
2102 | case F_UNLCK: | |
2103 | break; | |
2104 | ||
2105 | case F_RDLCK: | |
2106 | case F_WRLCK: | |
2107 | if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX || | |
2108 | ffl->end < ffl->start) | |
2109 | return -EIO; | |
2110 | ||
2111 | fl->fl_start = ffl->start; | |
2112 | fl->fl_end = ffl->end; | |
0b6e9ea0 SF |
2113 | |
2114 | /* | |
9d5b86ac BC |
2115 | * Convert pid into init's pid namespace. The locks API will |
2116 | * translate it into the caller's pid namespace. | |
0b6e9ea0 SF |
2117 | */ |
2118 | rcu_read_lock(); | |
9d5b86ac | 2119 | fl->fl_pid = pid_nr_ns(find_pid_ns(ffl->pid, fc->pid_ns), &init_pid_ns); |
0b6e9ea0 | 2120 | rcu_read_unlock(); |
71421259 MS |
2121 | break; |
2122 | ||
2123 | default: | |
2124 | return -EIO; | |
2125 | } | |
2126 | fl->fl_type = ffl->type; | |
2127 | return 0; | |
2128 | } | |
2129 | ||
7078187a | 2130 | static void fuse_lk_fill(struct fuse_args *args, struct file *file, |
a9ff4f87 | 2131 | const struct file_lock *fl, int opcode, pid_t pid, |
7078187a | 2132 | int flock, struct fuse_lk_in *inarg) |
71421259 | 2133 | { |
6131ffaa | 2134 | struct inode *inode = file_inode(file); |
9c8ef561 | 2135 | struct fuse_conn *fc = get_fuse_conn(inode); |
71421259 | 2136 | struct fuse_file *ff = file->private_data; |
7078187a MS |
2137 | |
2138 | memset(inarg, 0, sizeof(*inarg)); | |
2139 | inarg->fh = ff->fh; | |
2140 | inarg->owner = fuse_lock_owner_id(fc, fl->fl_owner); | |
2141 | inarg->lk.start = fl->fl_start; | |
2142 | inarg->lk.end = fl->fl_end; | |
2143 | inarg->lk.type = fl->fl_type; | |
2144 | inarg->lk.pid = pid; | |
a9ff4f87 | 2145 | if (flock) |
7078187a MS |
2146 | inarg->lk_flags |= FUSE_LK_FLOCK; |
2147 | args->in.h.opcode = opcode; | |
2148 | args->in.h.nodeid = get_node_id(inode); | |
2149 | args->in.numargs = 1; | |
2150 | args->in.args[0].size = sizeof(*inarg); | |
2151 | args->in.args[0].value = inarg; | |
71421259 MS |
2152 | } |
2153 | ||
2154 | static int fuse_getlk(struct file *file, struct file_lock *fl) | |
2155 | { | |
6131ffaa | 2156 | struct inode *inode = file_inode(file); |
71421259 | 2157 | struct fuse_conn *fc = get_fuse_conn(inode); |
7078187a MS |
2158 | FUSE_ARGS(args); |
2159 | struct fuse_lk_in inarg; | |
71421259 MS |
2160 | struct fuse_lk_out outarg; |
2161 | int err; | |
2162 | ||
7078187a MS |
2163 | fuse_lk_fill(&args, file, fl, FUSE_GETLK, 0, 0, &inarg); |
2164 | args.out.numargs = 1; | |
2165 | args.out.args[0].size = sizeof(outarg); | |
2166 | args.out.args[0].value = &outarg; | |
2167 | err = fuse_simple_request(fc, &args); | |
71421259 | 2168 | if (!err) |
0b6e9ea0 | 2169 | err = convert_fuse_file_lock(fc, &outarg.lk, fl); |
71421259 MS |
2170 | |
2171 | return err; | |
2172 | } | |
2173 | ||
a9ff4f87 | 2174 | static int fuse_setlk(struct file *file, struct file_lock *fl, int flock) |
71421259 | 2175 | { |
6131ffaa | 2176 | struct inode *inode = file_inode(file); |
71421259 | 2177 | struct fuse_conn *fc = get_fuse_conn(inode); |
7078187a MS |
2178 | FUSE_ARGS(args); |
2179 | struct fuse_lk_in inarg; | |
71421259 | 2180 | int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK; |
0b6e9ea0 SF |
2181 | struct pid *pid = fl->fl_type != F_UNLCK ? task_tgid(current) : NULL; |
2182 | pid_t pid_nr = pid_nr_ns(pid, fc->pid_ns); | |
71421259 MS |
2183 | int err; |
2184 | ||
8fb47a4f | 2185 | if (fl->fl_lmops && fl->fl_lmops->lm_grant) { |
48e90761 MS |
2186 | /* NLM needs asynchronous locks, which we don't support yet */ |
2187 | return -ENOLCK; | |
2188 | } | |
2189 | ||
71421259 | 2190 | /* Unlock on close is handled by the flush method */ |
50f2112c | 2191 | if ((fl->fl_flags & FL_CLOSE_POSIX) == FL_CLOSE_POSIX) |
71421259 MS |
2192 | return 0; |
2193 | ||
0b6e9ea0 | 2194 | fuse_lk_fill(&args, file, fl, opcode, pid_nr, flock, &inarg); |
7078187a | 2195 | err = fuse_simple_request(fc, &args); |
71421259 | 2196 | |
a4d27e75 MS |
2197 | /* locking is restartable */ |
2198 | if (err == -EINTR) | |
2199 | err = -ERESTARTSYS; | |
7078187a | 2200 | |
71421259 MS |
2201 | return err; |
2202 | } | |
2203 | ||
2204 | static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl) | |
2205 | { | |
6131ffaa | 2206 | struct inode *inode = file_inode(file); |
71421259 MS |
2207 | struct fuse_conn *fc = get_fuse_conn(inode); |
2208 | int err; | |
2209 | ||
48e90761 MS |
2210 | if (cmd == F_CANCELLK) { |
2211 | err = 0; | |
2212 | } else if (cmd == F_GETLK) { | |
71421259 | 2213 | if (fc->no_lock) { |
9d6a8c5c | 2214 | posix_test_lock(file, fl); |
71421259 MS |
2215 | err = 0; |
2216 | } else | |
2217 | err = fuse_getlk(file, fl); | |
2218 | } else { | |
2219 | if (fc->no_lock) | |
48e90761 | 2220 | err = posix_lock_file(file, fl, NULL); |
71421259 | 2221 | else |
a9ff4f87 | 2222 | err = fuse_setlk(file, fl, 0); |
71421259 MS |
2223 | } |
2224 | return err; | |
2225 | } | |
2226 | ||
a9ff4f87 MS |
2227 | static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl) |
2228 | { | |
6131ffaa | 2229 | struct inode *inode = file_inode(file); |
a9ff4f87 MS |
2230 | struct fuse_conn *fc = get_fuse_conn(inode); |
2231 | int err; | |
2232 | ||
37fb3a30 | 2233 | if (fc->no_flock) { |
4f656367 | 2234 | err = locks_lock_file_wait(file, fl); |
a9ff4f87 | 2235 | } else { |
37fb3a30 MS |
2236 | struct fuse_file *ff = file->private_data; |
2237 | ||
a9ff4f87 | 2238 | /* emulate flock with POSIX locks */ |
37fb3a30 | 2239 | ff->flock = true; |
a9ff4f87 MS |
2240 | err = fuse_setlk(file, fl, 1); |
2241 | } | |
2242 | ||
2243 | return err; | |
2244 | } | |
2245 | ||
b2d2272f MS |
2246 | static sector_t fuse_bmap(struct address_space *mapping, sector_t block) |
2247 | { | |
2248 | struct inode *inode = mapping->host; | |
2249 | struct fuse_conn *fc = get_fuse_conn(inode); | |
7078187a | 2250 | FUSE_ARGS(args); |
b2d2272f MS |
2251 | struct fuse_bmap_in inarg; |
2252 | struct fuse_bmap_out outarg; | |
2253 | int err; | |
2254 | ||
2255 | if (!inode->i_sb->s_bdev || fc->no_bmap) | |
2256 | return 0; | |
2257 | ||
b2d2272f MS |
2258 | memset(&inarg, 0, sizeof(inarg)); |
2259 | inarg.block = block; | |
2260 | inarg.blocksize = inode->i_sb->s_blocksize; | |
7078187a MS |
2261 | args.in.h.opcode = FUSE_BMAP; |
2262 | args.in.h.nodeid = get_node_id(inode); | |
2263 | args.in.numargs = 1; | |
2264 | args.in.args[0].size = sizeof(inarg); | |
2265 | args.in.args[0].value = &inarg; | |
2266 | args.out.numargs = 1; | |
2267 | args.out.args[0].size = sizeof(outarg); | |
2268 | args.out.args[0].value = &outarg; | |
2269 | err = fuse_simple_request(fc, &args); | |
b2d2272f MS |
2270 | if (err == -ENOSYS) |
2271 | fc->no_bmap = 1; | |
2272 | ||
2273 | return err ? 0 : outarg.block; | |
2274 | } | |
2275 | ||
0b5da8db R |
2276 | static loff_t fuse_lseek(struct file *file, loff_t offset, int whence) |
2277 | { | |
2278 | struct inode *inode = file->f_mapping->host; | |
2279 | struct fuse_conn *fc = get_fuse_conn(inode); | |
2280 | struct fuse_file *ff = file->private_data; | |
2281 | FUSE_ARGS(args); | |
2282 | struct fuse_lseek_in inarg = { | |
2283 | .fh = ff->fh, | |
2284 | .offset = offset, | |
2285 | .whence = whence | |
2286 | }; | |
2287 | struct fuse_lseek_out outarg; | |
2288 | int err; | |
2289 | ||
2290 | if (fc->no_lseek) | |
2291 | goto fallback; | |
2292 | ||
2293 | args.in.h.opcode = FUSE_LSEEK; | |
2294 | args.in.h.nodeid = ff->nodeid; | |
2295 | args.in.numargs = 1; | |
2296 | args.in.args[0].size = sizeof(inarg); | |
2297 | args.in.args[0].value = &inarg; | |
2298 | args.out.numargs = 1; | |
2299 | args.out.args[0].size = sizeof(outarg); | |
2300 | args.out.args[0].value = &outarg; | |
2301 | err = fuse_simple_request(fc, &args); | |
2302 | if (err) { | |
2303 | if (err == -ENOSYS) { | |
2304 | fc->no_lseek = 1; | |
2305 | goto fallback; | |
2306 | } | |
2307 | return err; | |
2308 | } | |
2309 | ||
2310 | return vfs_setpos(file, outarg.offset, inode->i_sb->s_maxbytes); | |
2311 | ||
2312 | fallback: | |
5b97eeac | 2313 | err = fuse_update_attributes(inode, file); |
0b5da8db R |
2314 | if (!err) |
2315 | return generic_file_llseek(file, offset, whence); | |
2316 | else | |
2317 | return err; | |
2318 | } | |
2319 | ||
965c8e59 | 2320 | static loff_t fuse_file_llseek(struct file *file, loff_t offset, int whence) |
5559b8f4 MS |
2321 | { |
2322 | loff_t retval; | |
6131ffaa | 2323 | struct inode *inode = file_inode(file); |
5559b8f4 | 2324 | |
0b5da8db R |
2325 | switch (whence) { |
2326 | case SEEK_SET: | |
2327 | case SEEK_CUR: | |
2328 | /* No i_mutex protection necessary for SEEK_CUR and SEEK_SET */ | |
965c8e59 | 2329 | retval = generic_file_llseek(file, offset, whence); |
0b5da8db R |
2330 | break; |
2331 | case SEEK_END: | |
5955102c | 2332 | inode_lock(inode); |
5b97eeac | 2333 | retval = fuse_update_attributes(inode, file); |
0b5da8db R |
2334 | if (!retval) |
2335 | retval = generic_file_llseek(file, offset, whence); | |
5955102c | 2336 | inode_unlock(inode); |
0b5da8db R |
2337 | break; |
2338 | case SEEK_HOLE: | |
2339 | case SEEK_DATA: | |
5955102c | 2340 | inode_lock(inode); |
0b5da8db | 2341 | retval = fuse_lseek(file, offset, whence); |
5955102c | 2342 | inode_unlock(inode); |
0b5da8db R |
2343 | break; |
2344 | default: | |
2345 | retval = -EINVAL; | |
2346 | } | |
c07c3d19 | 2347 | |
5559b8f4 MS |
2348 | return retval; |
2349 | } | |
2350 | ||
d9d318d3 MS |
2351 | /* |
2352 | * CUSE servers compiled on 32bit broke on 64bit kernels because the | |
2353 | * ABI was defined to be 'struct iovec' which is different on 32bit | |
2354 | * and 64bit. Fortunately we can determine which structure the server | |
2355 | * used from the size of the reply. | |
2356 | */ | |
1baa26b2 MS |
2357 | static int fuse_copy_ioctl_iovec_old(struct iovec *dst, void *src, |
2358 | size_t transferred, unsigned count, | |
2359 | bool is_compat) | |
d9d318d3 MS |
2360 | { |
2361 | #ifdef CONFIG_COMPAT | |
2362 | if (count * sizeof(struct compat_iovec) == transferred) { | |
2363 | struct compat_iovec *ciov = src; | |
2364 | unsigned i; | |
2365 | ||
2366 | /* | |
2367 | * With this interface a 32bit server cannot support | |
2368 | * non-compat (i.e. ones coming from 64bit apps) ioctl | |
2369 | * requests | |
2370 | */ | |
2371 | if (!is_compat) | |
2372 | return -EINVAL; | |
2373 | ||
2374 | for (i = 0; i < count; i++) { | |
2375 | dst[i].iov_base = compat_ptr(ciov[i].iov_base); | |
2376 | dst[i].iov_len = ciov[i].iov_len; | |
2377 | } | |
2378 | return 0; | |
2379 | } | |
2380 | #endif | |
2381 | ||
2382 | if (count * sizeof(struct iovec) != transferred) | |
2383 | return -EIO; | |
2384 | ||
2385 | memcpy(dst, src, transferred); | |
2386 | return 0; | |
2387 | } | |
2388 | ||
7572777e MS |
2389 | /* Make sure iov_length() won't overflow */ |
2390 | static int fuse_verify_ioctl_iov(struct iovec *iov, size_t count) | |
2391 | { | |
2392 | size_t n; | |
2393 | u32 max = FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT; | |
2394 | ||
fb6ccff6 | 2395 | for (n = 0; n < count; n++, iov++) { |
7572777e MS |
2396 | if (iov->iov_len > (size_t) max) |
2397 | return -ENOMEM; | |
2398 | max -= iov->iov_len; | |
2399 | } | |
2400 | return 0; | |
2401 | } | |
2402 | ||
1baa26b2 MS |
2403 | static int fuse_copy_ioctl_iovec(struct fuse_conn *fc, struct iovec *dst, |
2404 | void *src, size_t transferred, unsigned count, | |
2405 | bool is_compat) | |
2406 | { | |
2407 | unsigned i; | |
2408 | struct fuse_ioctl_iovec *fiov = src; | |
2409 | ||
2410 | if (fc->minor < 16) { | |
2411 | return fuse_copy_ioctl_iovec_old(dst, src, transferred, | |
2412 | count, is_compat); | |
2413 | } | |
2414 | ||
2415 | if (count * sizeof(struct fuse_ioctl_iovec) != transferred) | |
2416 | return -EIO; | |
2417 | ||
2418 | for (i = 0; i < count; i++) { | |
2419 | /* Did the server supply an inappropriate value? */ | |
2420 | if (fiov[i].base != (unsigned long) fiov[i].base || | |
2421 | fiov[i].len != (unsigned long) fiov[i].len) | |
2422 | return -EIO; | |
2423 | ||
2424 | dst[i].iov_base = (void __user *) (unsigned long) fiov[i].base; | |
2425 | dst[i].iov_len = (size_t) fiov[i].len; | |
2426 | ||
2427 | #ifdef CONFIG_COMPAT | |
2428 | if (is_compat && | |
2429 | (ptr_to_compat(dst[i].iov_base) != fiov[i].base || | |
2430 | (compat_size_t) dst[i].iov_len != fiov[i].len)) | |
2431 | return -EIO; | |
2432 | #endif | |
2433 | } | |
2434 | ||
2435 | return 0; | |
2436 | } | |
2437 | ||
2438 | ||
59efec7b TH |
2439 | /* |
2440 | * For ioctls, there is no generic way to determine how much memory | |
2441 | * needs to be read and/or written. Furthermore, ioctls are allowed | |
2442 | * to dereference the passed pointer, so the parameter requires deep | |
2443 | * copying but FUSE has no idea whatsoever about what to copy in or | |
2444 | * out. | |
2445 | * | |
2446 | * This is solved by allowing FUSE server to retry ioctl with | |
2447 | * necessary in/out iovecs. Let's assume the ioctl implementation | |
2448 | * needs to read in the following structure. | |
2449 | * | |
2450 | * struct a { | |
2451 | * char *buf; | |
2452 | * size_t buflen; | |
2453 | * } | |
2454 | * | |
2455 | * On the first callout to FUSE server, inarg->in_size and | |
2456 | * inarg->out_size will be NULL; then, the server completes the ioctl | |
2457 | * with FUSE_IOCTL_RETRY set in out->flags, out->in_iovs set to 1 and | |
2458 | * the actual iov array to | |
2459 | * | |
2460 | * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) } } | |
2461 | * | |
2462 | * which tells FUSE to copy in the requested area and retry the ioctl. | |
2463 | * On the second round, the server has access to the structure and | |
2464 | * from that it can tell what to look for next, so on the invocation, | |
2465 | * it sets FUSE_IOCTL_RETRY, out->in_iovs to 2 and iov array to | |
2466 | * | |
2467 | * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) }, | |
2468 | * { .iov_base = a.buf, .iov_len = a.buflen } } | |
2469 | * | |
2470 | * FUSE will copy both struct a and the pointed buffer from the | |
2471 | * process doing the ioctl and retry ioctl with both struct a and the | |
2472 | * buffer. | |
2473 | * | |
2474 | * This time, FUSE server has everything it needs and completes ioctl | |
2475 | * without FUSE_IOCTL_RETRY which finishes the ioctl call. | |
2476 | * | |
2477 | * Copying data out works the same way. | |
2478 | * | |
2479 | * Note that if FUSE_IOCTL_UNRESTRICTED is clear, the kernel | |
2480 | * automatically initializes in and out iovs by decoding @cmd with | |
2481 | * _IOC_* macros and the server is not allowed to request RETRY. This | |
2482 | * limits ioctl data transfers to well-formed ioctls and is the forced | |
2483 | * behavior for all FUSE servers. | |
2484 | */ | |
08cbf542 TH |
2485 | long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg, |
2486 | unsigned int flags) | |
59efec7b | 2487 | { |
59efec7b | 2488 | struct fuse_file *ff = file->private_data; |
d36f2487 | 2489 | struct fuse_conn *fc = ff->fc; |
59efec7b TH |
2490 | struct fuse_ioctl_in inarg = { |
2491 | .fh = ff->fh, | |
2492 | .cmd = cmd, | |
2493 | .arg = arg, | |
2494 | .flags = flags | |
2495 | }; | |
2496 | struct fuse_ioctl_out outarg; | |
2497 | struct fuse_req *req = NULL; | |
2498 | struct page **pages = NULL; | |
8ac83505 | 2499 | struct iovec *iov_page = NULL; |
59efec7b TH |
2500 | struct iovec *in_iov = NULL, *out_iov = NULL; |
2501 | unsigned int in_iovs = 0, out_iovs = 0, num_pages = 0, max_pages; | |
acbe5fda MS |
2502 | size_t in_size, out_size, transferred, c; |
2503 | int err, i; | |
2504 | struct iov_iter ii; | |
59efec7b | 2505 | |
1baa26b2 MS |
2506 | #if BITS_PER_LONG == 32 |
2507 | inarg.flags |= FUSE_IOCTL_32BIT; | |
2508 | #else | |
2509 | if (flags & FUSE_IOCTL_COMPAT) | |
2510 | inarg.flags |= FUSE_IOCTL_32BIT; | |
2511 | #endif | |
2512 | ||
59efec7b | 2513 | /* assume all the iovs returned by client always fits in a page */ |
1baa26b2 | 2514 | BUILD_BUG_ON(sizeof(struct fuse_ioctl_iovec) * FUSE_IOCTL_MAX_IOV > PAGE_SIZE); |
59efec7b | 2515 | |
59efec7b | 2516 | err = -ENOMEM; |
c411cc88 | 2517 | pages = kcalloc(FUSE_MAX_PAGES_PER_REQ, sizeof(pages[0]), GFP_KERNEL); |
8ac83505 | 2518 | iov_page = (struct iovec *) __get_free_page(GFP_KERNEL); |
59efec7b TH |
2519 | if (!pages || !iov_page) |
2520 | goto out; | |
2521 | ||
2522 | /* | |
2523 | * If restricted, initialize IO parameters as encoded in @cmd. | |
2524 | * RETRY from server is not allowed. | |
2525 | */ | |
2526 | if (!(flags & FUSE_IOCTL_UNRESTRICTED)) { | |
8ac83505 | 2527 | struct iovec *iov = iov_page; |
59efec7b | 2528 | |
c9f0523d | 2529 | iov->iov_base = (void __user *)arg; |
59efec7b TH |
2530 | iov->iov_len = _IOC_SIZE(cmd); |
2531 | ||
2532 | if (_IOC_DIR(cmd) & _IOC_WRITE) { | |
2533 | in_iov = iov; | |
2534 | in_iovs = 1; | |
2535 | } | |
2536 | ||
2537 | if (_IOC_DIR(cmd) & _IOC_READ) { | |
2538 | out_iov = iov; | |
2539 | out_iovs = 1; | |
2540 | } | |
2541 | } | |
2542 | ||
2543 | retry: | |
2544 | inarg.in_size = in_size = iov_length(in_iov, in_iovs); | |
2545 | inarg.out_size = out_size = iov_length(out_iov, out_iovs); | |
2546 | ||
2547 | /* | |
2548 | * Out data can be used either for actual out data or iovs, | |
2549 | * make sure there always is at least one page. | |
2550 | */ | |
2551 | out_size = max_t(size_t, out_size, PAGE_SIZE); | |
2552 | max_pages = DIV_ROUND_UP(max(in_size, out_size), PAGE_SIZE); | |
2553 | ||
2554 | /* make sure there are enough buffer pages and init request with them */ | |
2555 | err = -ENOMEM; | |
2556 | if (max_pages > FUSE_MAX_PAGES_PER_REQ) | |
2557 | goto out; | |
2558 | while (num_pages < max_pages) { | |
2559 | pages[num_pages] = alloc_page(GFP_KERNEL | __GFP_HIGHMEM); | |
2560 | if (!pages[num_pages]) | |
2561 | goto out; | |
2562 | num_pages++; | |
2563 | } | |
2564 | ||
54b96670 | 2565 | req = fuse_get_req(fc, num_pages); |
59efec7b TH |
2566 | if (IS_ERR(req)) { |
2567 | err = PTR_ERR(req); | |
2568 | req = NULL; | |
2569 | goto out; | |
2570 | } | |
2571 | memcpy(req->pages, pages, sizeof(req->pages[0]) * num_pages); | |
2572 | req->num_pages = num_pages; | |
7c190c8b | 2573 | fuse_page_descs_length_init(req, 0, req->num_pages); |
59efec7b TH |
2574 | |
2575 | /* okay, let's send it to the client */ | |
2576 | req->in.h.opcode = FUSE_IOCTL; | |
d36f2487 | 2577 | req->in.h.nodeid = ff->nodeid; |
59efec7b TH |
2578 | req->in.numargs = 1; |
2579 | req->in.args[0].size = sizeof(inarg); | |
2580 | req->in.args[0].value = &inarg; | |
2581 | if (in_size) { | |
2582 | req->in.numargs++; | |
2583 | req->in.args[1].size = in_size; | |
2584 | req->in.argpages = 1; | |
2585 | ||
acbe5fda MS |
2586 | err = -EFAULT; |
2587 | iov_iter_init(&ii, WRITE, in_iov, in_iovs, in_size); | |
2588 | for (i = 0; iov_iter_count(&ii) && !WARN_ON(i >= num_pages); i++) { | |
2589 | c = copy_page_from_iter(pages[i], 0, PAGE_SIZE, &ii); | |
2590 | if (c != PAGE_SIZE && iov_iter_count(&ii)) | |
2591 | goto out; | |
2592 | } | |
59efec7b TH |
2593 | } |
2594 | ||
2595 | req->out.numargs = 2; | |
2596 | req->out.args[0].size = sizeof(outarg); | |
2597 | req->out.args[0].value = &outarg; | |
2598 | req->out.args[1].size = out_size; | |
2599 | req->out.argpages = 1; | |
2600 | req->out.argvar = 1; | |
2601 | ||
b93f858a | 2602 | fuse_request_send(fc, req); |
59efec7b TH |
2603 | err = req->out.h.error; |
2604 | transferred = req->out.args[1].size; | |
2605 | fuse_put_request(fc, req); | |
2606 | req = NULL; | |
2607 | if (err) | |
2608 | goto out; | |
2609 | ||
2610 | /* did it ask for retry? */ | |
2611 | if (outarg.flags & FUSE_IOCTL_RETRY) { | |
8ac83505 | 2612 | void *vaddr; |
59efec7b TH |
2613 | |
2614 | /* no retry if in restricted mode */ | |
2615 | err = -EIO; | |
2616 | if (!(flags & FUSE_IOCTL_UNRESTRICTED)) | |
2617 | goto out; | |
2618 | ||
2619 | in_iovs = outarg.in_iovs; | |
2620 | out_iovs = outarg.out_iovs; | |
2621 | ||
2622 | /* | |
2623 | * Make sure things are in boundary, separate checks | |
2624 | * are to protect against overflow. | |
2625 | */ | |
2626 | err = -ENOMEM; | |
2627 | if (in_iovs > FUSE_IOCTL_MAX_IOV || | |
2628 | out_iovs > FUSE_IOCTL_MAX_IOV || | |
2629 | in_iovs + out_iovs > FUSE_IOCTL_MAX_IOV) | |
2630 | goto out; | |
2631 | ||
2408f6ef | 2632 | vaddr = kmap_atomic(pages[0]); |
1baa26b2 | 2633 | err = fuse_copy_ioctl_iovec(fc, iov_page, vaddr, |
d9d318d3 MS |
2634 | transferred, in_iovs + out_iovs, |
2635 | (flags & FUSE_IOCTL_COMPAT) != 0); | |
2408f6ef | 2636 | kunmap_atomic(vaddr); |
d9d318d3 MS |
2637 | if (err) |
2638 | goto out; | |
59efec7b | 2639 | |
8ac83505 | 2640 | in_iov = iov_page; |
59efec7b TH |
2641 | out_iov = in_iov + in_iovs; |
2642 | ||
7572777e MS |
2643 | err = fuse_verify_ioctl_iov(in_iov, in_iovs); |
2644 | if (err) | |
2645 | goto out; | |
2646 | ||
2647 | err = fuse_verify_ioctl_iov(out_iov, out_iovs); | |
2648 | if (err) | |
2649 | goto out; | |
2650 | ||
59efec7b TH |
2651 | goto retry; |
2652 | } | |
2653 | ||
2654 | err = -EIO; | |
2655 | if (transferred > inarg.out_size) | |
2656 | goto out; | |
2657 | ||
acbe5fda MS |
2658 | err = -EFAULT; |
2659 | iov_iter_init(&ii, READ, out_iov, out_iovs, transferred); | |
2660 | for (i = 0; iov_iter_count(&ii) && !WARN_ON(i >= num_pages); i++) { | |
2661 | c = copy_page_to_iter(pages[i], 0, PAGE_SIZE, &ii); | |
2662 | if (c != PAGE_SIZE && iov_iter_count(&ii)) | |
2663 | goto out; | |
2664 | } | |
2665 | err = 0; | |
59efec7b TH |
2666 | out: |
2667 | if (req) | |
2668 | fuse_put_request(fc, req); | |
8ac83505 | 2669 | free_page((unsigned long) iov_page); |
59efec7b TH |
2670 | while (num_pages) |
2671 | __free_page(pages[--num_pages]); | |
2672 | kfree(pages); | |
2673 | ||
2674 | return err ? err : outarg.result; | |
2675 | } | |
08cbf542 | 2676 | EXPORT_SYMBOL_GPL(fuse_do_ioctl); |
59efec7b | 2677 | |
b18da0c5 MS |
2678 | long fuse_ioctl_common(struct file *file, unsigned int cmd, |
2679 | unsigned long arg, unsigned int flags) | |
d36f2487 | 2680 | { |
6131ffaa | 2681 | struct inode *inode = file_inode(file); |
d36f2487 MS |
2682 | struct fuse_conn *fc = get_fuse_conn(inode); |
2683 | ||
c2132c1b | 2684 | if (!fuse_allow_current_process(fc)) |
d36f2487 MS |
2685 | return -EACCES; |
2686 | ||
2687 | if (is_bad_inode(inode)) | |
2688 | return -EIO; | |
2689 | ||
2690 | return fuse_do_ioctl(file, cmd, arg, flags); | |
2691 | } | |
2692 | ||
59efec7b TH |
2693 | static long fuse_file_ioctl(struct file *file, unsigned int cmd, |
2694 | unsigned long arg) | |
2695 | { | |
b18da0c5 | 2696 | return fuse_ioctl_common(file, cmd, arg, 0); |
59efec7b TH |
2697 | } |
2698 | ||
2699 | static long fuse_file_compat_ioctl(struct file *file, unsigned int cmd, | |
2700 | unsigned long arg) | |
2701 | { | |
b18da0c5 | 2702 | return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_COMPAT); |
59efec7b TH |
2703 | } |
2704 | ||
95668a69 TH |
2705 | /* |
2706 | * All files which have been polled are linked to RB tree | |
2707 | * fuse_conn->polled_files which is indexed by kh. Walk the tree and | |
2708 | * find the matching one. | |
2709 | */ | |
2710 | static struct rb_node **fuse_find_polled_node(struct fuse_conn *fc, u64 kh, | |
2711 | struct rb_node **parent_out) | |
2712 | { | |
2713 | struct rb_node **link = &fc->polled_files.rb_node; | |
2714 | struct rb_node *last = NULL; | |
2715 | ||
2716 | while (*link) { | |
2717 | struct fuse_file *ff; | |
2718 | ||
2719 | last = *link; | |
2720 | ff = rb_entry(last, struct fuse_file, polled_node); | |
2721 | ||
2722 | if (kh < ff->kh) | |
2723 | link = &last->rb_left; | |
2724 | else if (kh > ff->kh) | |
2725 | link = &last->rb_right; | |
2726 | else | |
2727 | return link; | |
2728 | } | |
2729 | ||
2730 | if (parent_out) | |
2731 | *parent_out = last; | |
2732 | return link; | |
2733 | } | |
2734 | ||
2735 | /* | |
2736 | * The file is about to be polled. Make sure it's on the polled_files | |
2737 | * RB tree. Note that files once added to the polled_files tree are | |
2738 | * not removed before the file is released. This is because a file | |
2739 | * polled once is likely to be polled again. | |
2740 | */ | |
2741 | static void fuse_register_polled_file(struct fuse_conn *fc, | |
2742 | struct fuse_file *ff) | |
2743 | { | |
2744 | spin_lock(&fc->lock); | |
2745 | if (RB_EMPTY_NODE(&ff->polled_node)) { | |
f3846266 | 2746 | struct rb_node **link, *uninitialized_var(parent); |
95668a69 TH |
2747 | |
2748 | link = fuse_find_polled_node(fc, ff->kh, &parent); | |
2749 | BUG_ON(*link); | |
2750 | rb_link_node(&ff->polled_node, parent, link); | |
2751 | rb_insert_color(&ff->polled_node, &fc->polled_files); | |
2752 | } | |
2753 | spin_unlock(&fc->lock); | |
2754 | } | |
2755 | ||
076ccb76 | 2756 | __poll_t fuse_file_poll(struct file *file, poll_table *wait) |
95668a69 | 2757 | { |
95668a69 | 2758 | struct fuse_file *ff = file->private_data; |
797759aa | 2759 | struct fuse_conn *fc = ff->fc; |
95668a69 TH |
2760 | struct fuse_poll_in inarg = { .fh = ff->fh, .kh = ff->kh }; |
2761 | struct fuse_poll_out outarg; | |
7078187a | 2762 | FUSE_ARGS(args); |
95668a69 TH |
2763 | int err; |
2764 | ||
2765 | if (fc->no_poll) | |
2766 | return DEFAULT_POLLMASK; | |
2767 | ||
2768 | poll_wait(file, &ff->poll_wait, wait); | |
c71d227f | 2769 | inarg.events = mangle_poll(poll_requested_events(wait)); |
95668a69 TH |
2770 | |
2771 | /* | |
2772 | * Ask for notification iff there's someone waiting for it. | |
2773 | * The client may ignore the flag and always notify. | |
2774 | */ | |
2775 | if (waitqueue_active(&ff->poll_wait)) { | |
2776 | inarg.flags |= FUSE_POLL_SCHEDULE_NOTIFY; | |
2777 | fuse_register_polled_file(fc, ff); | |
2778 | } | |
2779 | ||
7078187a MS |
2780 | args.in.h.opcode = FUSE_POLL; |
2781 | args.in.h.nodeid = ff->nodeid; | |
2782 | args.in.numargs = 1; | |
2783 | args.in.args[0].size = sizeof(inarg); | |
2784 | args.in.args[0].value = &inarg; | |
2785 | args.out.numargs = 1; | |
2786 | args.out.args[0].size = sizeof(outarg); | |
2787 | args.out.args[0].value = &outarg; | |
2788 | err = fuse_simple_request(fc, &args); | |
95668a69 TH |
2789 | |
2790 | if (!err) | |
c71d227f | 2791 | return demangle_poll(outarg.revents); |
95668a69 TH |
2792 | if (err == -ENOSYS) { |
2793 | fc->no_poll = 1; | |
2794 | return DEFAULT_POLLMASK; | |
2795 | } | |
a9a08845 | 2796 | return EPOLLERR; |
95668a69 | 2797 | } |
08cbf542 | 2798 | EXPORT_SYMBOL_GPL(fuse_file_poll); |
95668a69 TH |
2799 | |
2800 | /* | |
2801 | * This is called from fuse_handle_notify() on FUSE_NOTIFY_POLL and | |
2802 | * wakes up the poll waiters. | |
2803 | */ | |
2804 | int fuse_notify_poll_wakeup(struct fuse_conn *fc, | |
2805 | struct fuse_notify_poll_wakeup_out *outarg) | |
2806 | { | |
2807 | u64 kh = outarg->kh; | |
2808 | struct rb_node **link; | |
2809 | ||
2810 | spin_lock(&fc->lock); | |
2811 | ||
2812 | link = fuse_find_polled_node(fc, kh, NULL); | |
2813 | if (*link) { | |
2814 | struct fuse_file *ff; | |
2815 | ||
2816 | ff = rb_entry(*link, struct fuse_file, polled_node); | |
2817 | wake_up_interruptible_sync(&ff->poll_wait); | |
2818 | } | |
2819 | ||
2820 | spin_unlock(&fc->lock); | |
2821 | return 0; | |
2822 | } | |
2823 | ||
efb9fa9e MP |
2824 | static void fuse_do_truncate(struct file *file) |
2825 | { | |
2826 | struct inode *inode = file->f_mapping->host; | |
2827 | struct iattr attr; | |
2828 | ||
2829 | attr.ia_valid = ATTR_SIZE; | |
2830 | attr.ia_size = i_size_read(inode); | |
2831 | ||
2832 | attr.ia_file = file; | |
2833 | attr.ia_valid |= ATTR_FILE; | |
2834 | ||
62490330 | 2835 | fuse_do_setattr(file_dentry(file), &attr, file); |
efb9fa9e MP |
2836 | } |
2837 | ||
e5c5f05d MP |
2838 | static inline loff_t fuse_round_up(loff_t off) |
2839 | { | |
2840 | return round_up(off, FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT); | |
2841 | } | |
2842 | ||
4273b793 | 2843 | static ssize_t |
c8b8e32d | 2844 | fuse_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
4273b793 | 2845 | { |
9d5722b7 | 2846 | DECLARE_COMPLETION_ONSTACK(wait); |
4273b793 | 2847 | ssize_t ret = 0; |
60b9df7a MS |
2848 | struct file *file = iocb->ki_filp; |
2849 | struct fuse_file *ff = file->private_data; | |
e5c5f05d | 2850 | bool async_dio = ff->fc->async_dio; |
4273b793 | 2851 | loff_t pos = 0; |
bcba24cc MP |
2852 | struct inode *inode; |
2853 | loff_t i_size; | |
a6cbcd4a | 2854 | size_t count = iov_iter_count(iter); |
c8b8e32d | 2855 | loff_t offset = iocb->ki_pos; |
36cf66ed | 2856 | struct fuse_io_priv *io; |
4273b793 | 2857 | |
4273b793 | 2858 | pos = offset; |
bcba24cc MP |
2859 | inode = file->f_mapping->host; |
2860 | i_size = i_size_read(inode); | |
4273b793 | 2861 | |
6f673763 | 2862 | if ((iov_iter_rw(iter) == READ) && (offset > i_size)) |
9fe55eea SW |
2863 | return 0; |
2864 | ||
439ee5f0 | 2865 | /* optimization for short read */ |
6f673763 | 2866 | if (async_dio && iov_iter_rw(iter) != WRITE && offset + count > i_size) { |
439ee5f0 MP |
2867 | if (offset >= i_size) |
2868 | return 0; | |
6b775b18 AV |
2869 | iov_iter_truncate(iter, fuse_round_up(i_size - offset)); |
2870 | count = iov_iter_count(iter); | |
439ee5f0 MP |
2871 | } |
2872 | ||
bcba24cc | 2873 | io = kmalloc(sizeof(struct fuse_io_priv), GFP_KERNEL); |
36cf66ed MP |
2874 | if (!io) |
2875 | return -ENOMEM; | |
bcba24cc | 2876 | spin_lock_init(&io->lock); |
744742d6 | 2877 | kref_init(&io->refcnt); |
bcba24cc MP |
2878 | io->reqs = 1; |
2879 | io->bytes = -1; | |
2880 | io->size = 0; | |
2881 | io->offset = offset; | |
6f673763 | 2882 | io->write = (iov_iter_rw(iter) == WRITE); |
bcba24cc | 2883 | io->err = 0; |
bcba24cc MP |
2884 | /* |
2885 | * By default, we want to optimize all I/Os with async request | |
60b9df7a | 2886 | * submission to the client filesystem if supported. |
bcba24cc | 2887 | */ |
e5c5f05d | 2888 | io->async = async_dio; |
bcba24cc | 2889 | io->iocb = iocb; |
7879c4e5 | 2890 | io->blocking = is_sync_kiocb(iocb); |
bcba24cc MP |
2891 | |
2892 | /* | |
7879c4e5 AS |
2893 | * We cannot asynchronously extend the size of a file. |
2894 | * In such case the aio will behave exactly like sync io. | |
bcba24cc | 2895 | */ |
7879c4e5 AS |
2896 | if ((offset + count > i_size) && iov_iter_rw(iter) == WRITE) |
2897 | io->blocking = true; | |
4273b793 | 2898 | |
7879c4e5 | 2899 | if (io->async && io->blocking) { |
744742d6 SF |
2900 | /* |
2901 | * Additional reference to keep io around after | |
2902 | * calling fuse_aio_complete() | |
2903 | */ | |
2904 | kref_get(&io->refcnt); | |
9d5722b7 | 2905 | io->done = &wait; |
744742d6 | 2906 | } |
9d5722b7 | 2907 | |
6f673763 | 2908 | if (iov_iter_rw(iter) == WRITE) { |
6b775b18 | 2909 | ret = fuse_direct_io(io, iter, &pos, FUSE_DIO_WRITE); |
812408fb AV |
2910 | fuse_invalidate_attr(inode); |
2911 | } else { | |
d22a943f | 2912 | ret = __fuse_direct_read(io, iter, &pos); |
812408fb | 2913 | } |
36cf66ed | 2914 | |
bcba24cc MP |
2915 | if (io->async) { |
2916 | fuse_aio_complete(io, ret < 0 ? ret : 0, -1); | |
2917 | ||
2918 | /* we have a non-extending, async request, so return */ | |
7879c4e5 | 2919 | if (!io->blocking) |
bcba24cc MP |
2920 | return -EIOCBQUEUED; |
2921 | ||
9d5722b7 CH |
2922 | wait_for_completion(&wait); |
2923 | ret = fuse_get_res_by_io(io); | |
bcba24cc MP |
2924 | } |
2925 | ||
744742d6 | 2926 | kref_put(&io->refcnt, fuse_io_release); |
9d5722b7 | 2927 | |
6f673763 | 2928 | if (iov_iter_rw(iter) == WRITE) { |
efb9fa9e MP |
2929 | if (ret > 0) |
2930 | fuse_write_update_size(inode, pos); | |
2931 | else if (ret < 0 && offset + count > i_size) | |
2932 | fuse_do_truncate(file); | |
2933 | } | |
4273b793 AA |
2934 | |
2935 | return ret; | |
2936 | } | |
2937 | ||
cdadb11c MS |
2938 | static long fuse_file_fallocate(struct file *file, int mode, loff_t offset, |
2939 | loff_t length) | |
05ba1f08 AP |
2940 | { |
2941 | struct fuse_file *ff = file->private_data; | |
1c68271c | 2942 | struct inode *inode = file_inode(file); |
0ab08f57 | 2943 | struct fuse_inode *fi = get_fuse_inode(inode); |
05ba1f08 | 2944 | struct fuse_conn *fc = ff->fc; |
7078187a | 2945 | FUSE_ARGS(args); |
05ba1f08 AP |
2946 | struct fuse_fallocate_in inarg = { |
2947 | .fh = ff->fh, | |
2948 | .offset = offset, | |
2949 | .length = length, | |
2950 | .mode = mode | |
2951 | }; | |
2952 | int err; | |
14c14414 MP |
2953 | bool lock_inode = !(mode & FALLOC_FL_KEEP_SIZE) || |
2954 | (mode & FALLOC_FL_PUNCH_HOLE); | |
05ba1f08 | 2955 | |
4adb8302 MS |
2956 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
2957 | return -EOPNOTSUPP; | |
2958 | ||
519c6040 MS |
2959 | if (fc->no_fallocate) |
2960 | return -EOPNOTSUPP; | |
2961 | ||
14c14414 | 2962 | if (lock_inode) { |
5955102c | 2963 | inode_lock(inode); |
bde52788 MP |
2964 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
2965 | loff_t endbyte = offset + length - 1; | |
2966 | err = filemap_write_and_wait_range(inode->i_mapping, | |
2967 | offset, endbyte); | |
2968 | if (err) | |
2969 | goto out; | |
2970 | ||
2971 | fuse_sync_writes(inode); | |
2972 | } | |
3634a632 BF |
2973 | } |
2974 | ||
0ab08f57 MP |
2975 | if (!(mode & FALLOC_FL_KEEP_SIZE)) |
2976 | set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); | |
2977 | ||
7078187a MS |
2978 | args.in.h.opcode = FUSE_FALLOCATE; |
2979 | args.in.h.nodeid = ff->nodeid; | |
2980 | args.in.numargs = 1; | |
2981 | args.in.args[0].size = sizeof(inarg); | |
2982 | args.in.args[0].value = &inarg; | |
2983 | err = fuse_simple_request(fc, &args); | |
519c6040 MS |
2984 | if (err == -ENOSYS) { |
2985 | fc->no_fallocate = 1; | |
2986 | err = -EOPNOTSUPP; | |
2987 | } | |
bee6c307 BF |
2988 | if (err) |
2989 | goto out; | |
2990 | ||
2991 | /* we could have extended the file */ | |
b0aa7606 MP |
2992 | if (!(mode & FALLOC_FL_KEEP_SIZE)) { |
2993 | bool changed = fuse_write_update_size(inode, offset + length); | |
2994 | ||
93d2269d MS |
2995 | if (changed && fc->writeback_cache) |
2996 | file_update_time(file); | |
b0aa7606 | 2997 | } |
bee6c307 BF |
2998 | |
2999 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
3000 | truncate_pagecache_range(inode, offset, offset + length - 1); | |
3001 | ||
3002 | fuse_invalidate_attr(inode); | |
3003 | ||
3634a632 | 3004 | out: |
0ab08f57 MP |
3005 | if (!(mode & FALLOC_FL_KEEP_SIZE)) |
3006 | clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); | |
3007 | ||
bde52788 | 3008 | if (lock_inode) |
5955102c | 3009 | inode_unlock(inode); |
3634a632 | 3010 | |
05ba1f08 AP |
3011 | return err; |
3012 | } | |
05ba1f08 | 3013 | |
4b6f5d20 | 3014 | static const struct file_operations fuse_file_operations = { |
5559b8f4 | 3015 | .llseek = fuse_file_llseek, |
37c20f16 | 3016 | .read_iter = fuse_file_read_iter, |
84c3d55c | 3017 | .write_iter = fuse_file_write_iter, |
b6aeaded MS |
3018 | .mmap = fuse_file_mmap, |
3019 | .open = fuse_open, | |
3020 | .flush = fuse_flush, | |
3021 | .release = fuse_release, | |
3022 | .fsync = fuse_fsync, | |
71421259 | 3023 | .lock = fuse_file_lock, |
a9ff4f87 | 3024 | .flock = fuse_file_flock, |
5ffc4ef4 | 3025 | .splice_read = generic_file_splice_read, |
59efec7b TH |
3026 | .unlocked_ioctl = fuse_file_ioctl, |
3027 | .compat_ioctl = fuse_file_compat_ioctl, | |
95668a69 | 3028 | .poll = fuse_file_poll, |
05ba1f08 | 3029 | .fallocate = fuse_file_fallocate, |
b6aeaded MS |
3030 | }; |
3031 | ||
4b6f5d20 | 3032 | static const struct file_operations fuse_direct_io_file_operations = { |
5559b8f4 | 3033 | .llseek = fuse_file_llseek, |
15316263 | 3034 | .read_iter = fuse_direct_read_iter, |
15316263 | 3035 | .write_iter = fuse_direct_write_iter, |
fc280c96 | 3036 | .mmap = fuse_direct_mmap, |
413ef8cb MS |
3037 | .open = fuse_open, |
3038 | .flush = fuse_flush, | |
3039 | .release = fuse_release, | |
3040 | .fsync = fuse_fsync, | |
71421259 | 3041 | .lock = fuse_file_lock, |
a9ff4f87 | 3042 | .flock = fuse_file_flock, |
59efec7b TH |
3043 | .unlocked_ioctl = fuse_file_ioctl, |
3044 | .compat_ioctl = fuse_file_compat_ioctl, | |
95668a69 | 3045 | .poll = fuse_file_poll, |
05ba1f08 | 3046 | .fallocate = fuse_file_fallocate, |
fc280c96 | 3047 | /* no splice_read */ |
413ef8cb MS |
3048 | }; |
3049 | ||
f5e54d6e | 3050 | static const struct address_space_operations fuse_file_aops = { |
b6aeaded | 3051 | .readpage = fuse_readpage, |
3be5a52b | 3052 | .writepage = fuse_writepage, |
26d614df | 3053 | .writepages = fuse_writepages, |
3be5a52b | 3054 | .launder_page = fuse_launder_page, |
db50b96c | 3055 | .readpages = fuse_readpages, |
3be5a52b | 3056 | .set_page_dirty = __set_page_dirty_nobuffers, |
b2d2272f | 3057 | .bmap = fuse_bmap, |
4273b793 | 3058 | .direct_IO = fuse_direct_IO, |
6b12c1b3 PE |
3059 | .write_begin = fuse_write_begin, |
3060 | .write_end = fuse_write_end, | |
b6aeaded MS |
3061 | }; |
3062 | ||
3063 | void fuse_init_file_inode(struct inode *inode) | |
3064 | { | |
45323fb7 MS |
3065 | inode->i_fop = &fuse_file_operations; |
3066 | inode->i_data.a_ops = &fuse_file_aops; | |
b6aeaded | 3067 | } |