Commit | Line | Data |
---|---|---|
b6aeaded MS |
1 | /* |
2 | FUSE: Filesystem in Userspace | |
1729a16c | 3 | Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu> |
b6aeaded MS |
4 | |
5 | This program can be distributed under the terms of the GNU GPL. | |
6 | See the file COPYING. | |
7 | */ | |
8 | ||
9 | #include "fuse_i.h" | |
10 | ||
11 | #include <linux/pagemap.h> | |
12 | #include <linux/slab.h> | |
13 | #include <linux/kernel.h> | |
e8edc6e0 | 14 | #include <linux/sched.h> |
08cbf542 | 15 | #include <linux/module.h> |
d9d318d3 | 16 | #include <linux/compat.h> |
478e0841 | 17 | #include <linux/swap.h> |
b6aeaded | 18 | |
4b6f5d20 | 19 | static const struct file_operations fuse_direct_io_file_operations; |
45323fb7 | 20 | |
91fe96b4 MS |
21 | static int fuse_send_open(struct fuse_conn *fc, u64 nodeid, struct file *file, |
22 | int opcode, struct fuse_open_out *outargp) | |
b6aeaded | 23 | { |
b6aeaded | 24 | struct fuse_open_in inarg; |
fd72faac MS |
25 | struct fuse_req *req; |
26 | int err; | |
27 | ||
b111c8c0 | 28 | req = fuse_get_req_nopages(fc); |
ce1d5a49 MS |
29 | if (IS_ERR(req)) |
30 | return PTR_ERR(req); | |
fd72faac MS |
31 | |
32 | memset(&inarg, 0, sizeof(inarg)); | |
6ff958ed MS |
33 | inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY); |
34 | if (!fc->atomic_o_trunc) | |
35 | inarg.flags &= ~O_TRUNC; | |
91fe96b4 MS |
36 | req->in.h.opcode = opcode; |
37 | req->in.h.nodeid = nodeid; | |
fd72faac MS |
38 | req->in.numargs = 1; |
39 | req->in.args[0].size = sizeof(inarg); | |
40 | req->in.args[0].value = &inarg; | |
41 | req->out.numargs = 1; | |
42 | req->out.args[0].size = sizeof(*outargp); | |
43 | req->out.args[0].value = outargp; | |
b93f858a | 44 | fuse_request_send(fc, req); |
fd72faac MS |
45 | err = req->out.h.error; |
46 | fuse_put_request(fc, req); | |
47 | ||
48 | return err; | |
49 | } | |
50 | ||
acf99433 | 51 | struct fuse_file *fuse_file_alloc(struct fuse_conn *fc) |
fd72faac MS |
52 | { |
53 | struct fuse_file *ff; | |
6b2db28a | 54 | |
fd72faac | 55 | ff = kmalloc(sizeof(struct fuse_file), GFP_KERNEL); |
6b2db28a TH |
56 | if (unlikely(!ff)) |
57 | return NULL; | |
58 | ||
da5e4714 | 59 | ff->fc = fc; |
4250c066 | 60 | ff->reserved_req = fuse_request_alloc(0); |
6b2db28a TH |
61 | if (unlikely(!ff->reserved_req)) { |
62 | kfree(ff); | |
63 | return NULL; | |
fd72faac | 64 | } |
6b2db28a TH |
65 | |
66 | INIT_LIST_HEAD(&ff->write_entry); | |
67 | atomic_set(&ff->count, 0); | |
68 | RB_CLEAR_NODE(&ff->polled_node); | |
69 | init_waitqueue_head(&ff->poll_wait); | |
70 | ||
71 | spin_lock(&fc->lock); | |
72 | ff->kh = ++fc->khctr; | |
73 | spin_unlock(&fc->lock); | |
74 | ||
fd72faac MS |
75 | return ff; |
76 | } | |
77 | ||
78 | void fuse_file_free(struct fuse_file *ff) | |
79 | { | |
33649c91 | 80 | fuse_request_free(ff->reserved_req); |
fd72faac MS |
81 | kfree(ff); |
82 | } | |
83 | ||
c7b7143c | 84 | struct fuse_file *fuse_file_get(struct fuse_file *ff) |
c756e0a4 MS |
85 | { |
86 | atomic_inc(&ff->count); | |
87 | return ff; | |
88 | } | |
89 | ||
5a18ec17 MS |
90 | static void fuse_release_async(struct work_struct *work) |
91 | { | |
92 | struct fuse_req *req; | |
93 | struct fuse_conn *fc; | |
94 | struct path path; | |
95 | ||
96 | req = container_of(work, struct fuse_req, misc.release.work); | |
97 | path = req->misc.release.path; | |
98 | fc = get_fuse_conn(path.dentry->d_inode); | |
99 | ||
100 | fuse_put_request(fc, req); | |
101 | path_put(&path); | |
102 | } | |
103 | ||
819c4b3b MS |
104 | static void fuse_release_end(struct fuse_conn *fc, struct fuse_req *req) |
105 | { | |
5a18ec17 MS |
106 | if (fc->destroy_req) { |
107 | /* | |
108 | * If this is a fuseblk mount, then it's possible that | |
109 | * releasing the path will result in releasing the | |
110 | * super block and sending the DESTROY request. If | |
111 | * the server is single threaded, this would hang. | |
112 | * For this reason do the path_put() in a separate | |
113 | * thread. | |
114 | */ | |
115 | atomic_inc(&req->count); | |
116 | INIT_WORK(&req->misc.release.work, fuse_release_async); | |
117 | schedule_work(&req->misc.release.work); | |
118 | } else { | |
119 | path_put(&req->misc.release.path); | |
120 | } | |
819c4b3b MS |
121 | } |
122 | ||
5a18ec17 | 123 | static void fuse_file_put(struct fuse_file *ff, bool sync) |
c756e0a4 MS |
124 | { |
125 | if (atomic_dec_and_test(&ff->count)) { | |
126 | struct fuse_req *req = ff->reserved_req; | |
8b0797a4 | 127 | |
5a18ec17 | 128 | if (sync) { |
8b41e671 | 129 | req->background = 0; |
5a18ec17 MS |
130 | fuse_request_send(ff->fc, req); |
131 | path_put(&req->misc.release.path); | |
132 | fuse_put_request(ff->fc, req); | |
133 | } else { | |
134 | req->end = fuse_release_end; | |
8b41e671 | 135 | req->background = 1; |
5a18ec17 MS |
136 | fuse_request_send_background(ff->fc, req); |
137 | } | |
c756e0a4 MS |
138 | kfree(ff); |
139 | } | |
140 | } | |
141 | ||
08cbf542 TH |
142 | int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file, |
143 | bool isdir) | |
91fe96b4 MS |
144 | { |
145 | struct fuse_open_out outarg; | |
146 | struct fuse_file *ff; | |
147 | int err; | |
148 | int opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN; | |
149 | ||
150 | ff = fuse_file_alloc(fc); | |
151 | if (!ff) | |
152 | return -ENOMEM; | |
153 | ||
154 | err = fuse_send_open(fc, nodeid, file, opcode, &outarg); | |
155 | if (err) { | |
156 | fuse_file_free(ff); | |
157 | return err; | |
158 | } | |
159 | ||
160 | if (isdir) | |
161 | outarg.open_flags &= ~FOPEN_DIRECT_IO; | |
162 | ||
163 | ff->fh = outarg.fh; | |
164 | ff->nodeid = nodeid; | |
165 | ff->open_flags = outarg.open_flags; | |
166 | file->private_data = fuse_file_get(ff); | |
167 | ||
168 | return 0; | |
169 | } | |
08cbf542 | 170 | EXPORT_SYMBOL_GPL(fuse_do_open); |
91fe96b4 | 171 | |
c7b7143c | 172 | void fuse_finish_open(struct inode *inode, struct file *file) |
fd72faac | 173 | { |
c7b7143c | 174 | struct fuse_file *ff = file->private_data; |
a0822c55 | 175 | struct fuse_conn *fc = get_fuse_conn(inode); |
c7b7143c MS |
176 | |
177 | if (ff->open_flags & FOPEN_DIRECT_IO) | |
fd72faac | 178 | file->f_op = &fuse_direct_io_file_operations; |
c7b7143c | 179 | if (!(ff->open_flags & FOPEN_KEEP_CACHE)) |
b1009979 | 180 | invalidate_inode_pages2(inode->i_mapping); |
c7b7143c | 181 | if (ff->open_flags & FOPEN_NONSEEKABLE) |
a7c1b990 | 182 | nonseekable_open(inode, file); |
a0822c55 KS |
183 | if (fc->atomic_o_trunc && (file->f_flags & O_TRUNC)) { |
184 | struct fuse_inode *fi = get_fuse_inode(inode); | |
185 | ||
186 | spin_lock(&fc->lock); | |
187 | fi->attr_version = ++fc->attr_version; | |
188 | i_size_write(inode, 0); | |
189 | spin_unlock(&fc->lock); | |
190 | fuse_invalidate_attr(inode); | |
191 | } | |
fd72faac MS |
192 | } |
193 | ||
91fe96b4 | 194 | int fuse_open_common(struct inode *inode, struct file *file, bool isdir) |
fd72faac | 195 | { |
acf99433 | 196 | struct fuse_conn *fc = get_fuse_conn(inode); |
b6aeaded | 197 | int err; |
b6aeaded MS |
198 | |
199 | err = generic_file_open(inode, file); | |
200 | if (err) | |
201 | return err; | |
202 | ||
91fe96b4 | 203 | err = fuse_do_open(fc, get_node_id(inode), file, isdir); |
fd72faac | 204 | if (err) |
91fe96b4 | 205 | return err; |
b6aeaded | 206 | |
91fe96b4 MS |
207 | fuse_finish_open(inode, file); |
208 | ||
209 | return 0; | |
b6aeaded MS |
210 | } |
211 | ||
8b0797a4 | 212 | static void fuse_prepare_release(struct fuse_file *ff, int flags, int opcode) |
64c6d8ed | 213 | { |
8b0797a4 | 214 | struct fuse_conn *fc = ff->fc; |
33649c91 | 215 | struct fuse_req *req = ff->reserved_req; |
b57d4264 | 216 | struct fuse_release_in *inarg = &req->misc.release.in; |
b6aeaded | 217 | |
8b0797a4 MS |
218 | spin_lock(&fc->lock); |
219 | list_del(&ff->write_entry); | |
220 | if (!RB_EMPTY_NODE(&ff->polled_node)) | |
221 | rb_erase(&ff->polled_node, &fc->polled_files); | |
222 | spin_unlock(&fc->lock); | |
223 | ||
357ccf2b | 224 | wake_up_interruptible_all(&ff->poll_wait); |
8b0797a4 | 225 | |
b6aeaded | 226 | inarg->fh = ff->fh; |
fd72faac | 227 | inarg->flags = flags; |
51eb01e7 | 228 | req->in.h.opcode = opcode; |
c7b7143c | 229 | req->in.h.nodeid = ff->nodeid; |
b6aeaded MS |
230 | req->in.numargs = 1; |
231 | req->in.args[0].size = sizeof(struct fuse_release_in); | |
232 | req->in.args[0].value = inarg; | |
fd72faac MS |
233 | } |
234 | ||
8b0797a4 | 235 | void fuse_release_common(struct file *file, int opcode) |
fd72faac | 236 | { |
6b2db28a TH |
237 | struct fuse_file *ff; |
238 | struct fuse_req *req; | |
b6aeaded | 239 | |
6b2db28a TH |
240 | ff = file->private_data; |
241 | if (unlikely(!ff)) | |
8b0797a4 | 242 | return; |
6b2db28a | 243 | |
6b2db28a | 244 | req = ff->reserved_req; |
8b0797a4 | 245 | fuse_prepare_release(ff, file->f_flags, opcode); |
6b2db28a | 246 | |
37fb3a30 MS |
247 | if (ff->flock) { |
248 | struct fuse_release_in *inarg = &req->misc.release.in; | |
249 | inarg->release_flags |= FUSE_RELEASE_FLOCK_UNLOCK; | |
250 | inarg->lock_owner = fuse_lock_owner_id(ff->fc, | |
251 | (fl_owner_t) file); | |
252 | } | |
6b2db28a | 253 | /* Hold vfsmount and dentry until release is finished */ |
b0be46eb MS |
254 | path_get(&file->f_path); |
255 | req->misc.release.path = file->f_path; | |
6b2db28a | 256 | |
6b2db28a TH |
257 | /* |
258 | * Normally this will send the RELEASE request, however if | |
259 | * some asynchronous READ or WRITE requests are outstanding, | |
260 | * the sending will be delayed. | |
5a18ec17 MS |
261 | * |
262 | * Make the release synchronous if this is a fuseblk mount, | |
263 | * synchronous RELEASE is allowed (and desirable) in this case | |
264 | * because the server can be trusted not to screw up. | |
6b2db28a | 265 | */ |
5a18ec17 | 266 | fuse_file_put(ff, ff->fc->destroy_req != NULL); |
b6aeaded MS |
267 | } |
268 | ||
04730fef MS |
269 | static int fuse_open(struct inode *inode, struct file *file) |
270 | { | |
91fe96b4 | 271 | return fuse_open_common(inode, file, false); |
04730fef MS |
272 | } |
273 | ||
274 | static int fuse_release(struct inode *inode, struct file *file) | |
275 | { | |
8b0797a4 MS |
276 | fuse_release_common(file, FUSE_RELEASE); |
277 | ||
278 | /* return value is ignored by VFS */ | |
279 | return 0; | |
280 | } | |
281 | ||
282 | void fuse_sync_release(struct fuse_file *ff, int flags) | |
283 | { | |
284 | WARN_ON(atomic_read(&ff->count) > 1); | |
285 | fuse_prepare_release(ff, flags, FUSE_RELEASE); | |
286 | ff->reserved_req->force = 1; | |
8b41e671 | 287 | ff->reserved_req->background = 0; |
8b0797a4 MS |
288 | fuse_request_send(ff->fc, ff->reserved_req); |
289 | fuse_put_request(ff->fc, ff->reserved_req); | |
290 | kfree(ff); | |
04730fef | 291 | } |
08cbf542 | 292 | EXPORT_SYMBOL_GPL(fuse_sync_release); |
04730fef | 293 | |
71421259 | 294 | /* |
9c8ef561 MS |
295 | * Scramble the ID space with XTEA, so that the value of the files_struct |
296 | * pointer is not exposed to userspace. | |
71421259 | 297 | */ |
f3332114 | 298 | u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id) |
71421259 | 299 | { |
9c8ef561 MS |
300 | u32 *k = fc->scramble_key; |
301 | u64 v = (unsigned long) id; | |
302 | u32 v0 = v; | |
303 | u32 v1 = v >> 32; | |
304 | u32 sum = 0; | |
305 | int i; | |
306 | ||
307 | for (i = 0; i < 32; i++) { | |
308 | v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]); | |
309 | sum += 0x9E3779B9; | |
310 | v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]); | |
311 | } | |
312 | ||
313 | return (u64) v0 + ((u64) v1 << 32); | |
71421259 MS |
314 | } |
315 | ||
3be5a52b MS |
316 | /* |
317 | * Check if page is under writeback | |
318 | * | |
319 | * This is currently done by walking the list of writepage requests | |
320 | * for the inode, which can be pretty inefficient. | |
321 | */ | |
322 | static bool fuse_page_is_writeback(struct inode *inode, pgoff_t index) | |
323 | { | |
324 | struct fuse_conn *fc = get_fuse_conn(inode); | |
325 | struct fuse_inode *fi = get_fuse_inode(inode); | |
326 | struct fuse_req *req; | |
327 | bool found = false; | |
328 | ||
329 | spin_lock(&fc->lock); | |
330 | list_for_each_entry(req, &fi->writepages, writepages_entry) { | |
331 | pgoff_t curr_index; | |
332 | ||
333 | BUG_ON(req->inode != inode); | |
334 | curr_index = req->misc.write.in.offset >> PAGE_CACHE_SHIFT; | |
335 | if (curr_index == index) { | |
336 | found = true; | |
337 | break; | |
338 | } | |
339 | } | |
340 | spin_unlock(&fc->lock); | |
341 | ||
342 | return found; | |
343 | } | |
344 | ||
345 | /* | |
346 | * Wait for page writeback to be completed. | |
347 | * | |
348 | * Since fuse doesn't rely on the VM writeback tracking, this has to | |
349 | * use some other means. | |
350 | */ | |
351 | static int fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index) | |
352 | { | |
353 | struct fuse_inode *fi = get_fuse_inode(inode); | |
354 | ||
355 | wait_event(fi->page_waitq, !fuse_page_is_writeback(inode, index)); | |
356 | return 0; | |
357 | } | |
358 | ||
75e1fcc0 | 359 | static int fuse_flush(struct file *file, fl_owner_t id) |
b6aeaded | 360 | { |
6131ffaa | 361 | struct inode *inode = file_inode(file); |
b6aeaded MS |
362 | struct fuse_conn *fc = get_fuse_conn(inode); |
363 | struct fuse_file *ff = file->private_data; | |
364 | struct fuse_req *req; | |
365 | struct fuse_flush_in inarg; | |
366 | int err; | |
367 | ||
248d86e8 MS |
368 | if (is_bad_inode(inode)) |
369 | return -EIO; | |
370 | ||
b6aeaded MS |
371 | if (fc->no_flush) |
372 | return 0; | |
373 | ||
b111c8c0 | 374 | req = fuse_get_req_nofail_nopages(fc, file); |
b6aeaded MS |
375 | memset(&inarg, 0, sizeof(inarg)); |
376 | inarg.fh = ff->fh; | |
9c8ef561 | 377 | inarg.lock_owner = fuse_lock_owner_id(fc, id); |
b6aeaded MS |
378 | req->in.h.opcode = FUSE_FLUSH; |
379 | req->in.h.nodeid = get_node_id(inode); | |
b6aeaded MS |
380 | req->in.numargs = 1; |
381 | req->in.args[0].size = sizeof(inarg); | |
382 | req->in.args[0].value = &inarg; | |
71421259 | 383 | req->force = 1; |
b93f858a | 384 | fuse_request_send(fc, req); |
b6aeaded MS |
385 | err = req->out.h.error; |
386 | fuse_put_request(fc, req); | |
387 | if (err == -ENOSYS) { | |
388 | fc->no_flush = 1; | |
389 | err = 0; | |
390 | } | |
391 | return err; | |
392 | } | |
393 | ||
3be5a52b MS |
394 | /* |
395 | * Wait for all pending writepages on the inode to finish. | |
396 | * | |
397 | * This is currently done by blocking further writes with FUSE_NOWRITE | |
398 | * and waiting for all sent writes to complete. | |
399 | * | |
400 | * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage | |
401 | * could conflict with truncation. | |
402 | */ | |
403 | static void fuse_sync_writes(struct inode *inode) | |
404 | { | |
405 | fuse_set_nowrite(inode); | |
406 | fuse_release_nowrite(inode); | |
407 | } | |
408 | ||
02c24a82 JB |
409 | int fuse_fsync_common(struct file *file, loff_t start, loff_t end, |
410 | int datasync, int isdir) | |
b6aeaded | 411 | { |
7ea80859 | 412 | struct inode *inode = file->f_mapping->host; |
b6aeaded MS |
413 | struct fuse_conn *fc = get_fuse_conn(inode); |
414 | struct fuse_file *ff = file->private_data; | |
415 | struct fuse_req *req; | |
416 | struct fuse_fsync_in inarg; | |
417 | int err; | |
418 | ||
248d86e8 MS |
419 | if (is_bad_inode(inode)) |
420 | return -EIO; | |
421 | ||
02c24a82 JB |
422 | err = filemap_write_and_wait_range(inode->i_mapping, start, end); |
423 | if (err) | |
424 | return err; | |
425 | ||
82547981 | 426 | if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir)) |
b6aeaded MS |
427 | return 0; |
428 | ||
02c24a82 JB |
429 | mutex_lock(&inode->i_mutex); |
430 | ||
3be5a52b MS |
431 | /* |
432 | * Start writeback against all dirty pages of the inode, then | |
433 | * wait for all outstanding writes, before sending the FSYNC | |
434 | * request. | |
435 | */ | |
436 | err = write_inode_now(inode, 0); | |
437 | if (err) | |
02c24a82 | 438 | goto out; |
3be5a52b MS |
439 | |
440 | fuse_sync_writes(inode); | |
441 | ||
b111c8c0 | 442 | req = fuse_get_req_nopages(fc); |
02c24a82 JB |
443 | if (IS_ERR(req)) { |
444 | err = PTR_ERR(req); | |
445 | goto out; | |
446 | } | |
b6aeaded MS |
447 | |
448 | memset(&inarg, 0, sizeof(inarg)); | |
449 | inarg.fh = ff->fh; | |
450 | inarg.fsync_flags = datasync ? 1 : 0; | |
82547981 | 451 | req->in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC; |
b6aeaded | 452 | req->in.h.nodeid = get_node_id(inode); |
b6aeaded MS |
453 | req->in.numargs = 1; |
454 | req->in.args[0].size = sizeof(inarg); | |
455 | req->in.args[0].value = &inarg; | |
b93f858a | 456 | fuse_request_send(fc, req); |
b6aeaded MS |
457 | err = req->out.h.error; |
458 | fuse_put_request(fc, req); | |
459 | if (err == -ENOSYS) { | |
82547981 MS |
460 | if (isdir) |
461 | fc->no_fsyncdir = 1; | |
462 | else | |
463 | fc->no_fsync = 1; | |
b6aeaded MS |
464 | err = 0; |
465 | } | |
02c24a82 JB |
466 | out: |
467 | mutex_unlock(&inode->i_mutex); | |
b6aeaded MS |
468 | return err; |
469 | } | |
470 | ||
02c24a82 JB |
471 | static int fuse_fsync(struct file *file, loff_t start, loff_t end, |
472 | int datasync) | |
82547981 | 473 | { |
02c24a82 | 474 | return fuse_fsync_common(file, start, end, datasync, 0); |
82547981 MS |
475 | } |
476 | ||
2106cb18 MS |
477 | void fuse_read_fill(struct fuse_req *req, struct file *file, loff_t pos, |
478 | size_t count, int opcode) | |
b6aeaded | 479 | { |
5c5c5e51 | 480 | struct fuse_read_in *inarg = &req->misc.read.in; |
a6643094 | 481 | struct fuse_file *ff = file->private_data; |
b6aeaded | 482 | |
361b1eb5 MS |
483 | inarg->fh = ff->fh; |
484 | inarg->offset = pos; | |
485 | inarg->size = count; | |
a6643094 | 486 | inarg->flags = file->f_flags; |
361b1eb5 | 487 | req->in.h.opcode = opcode; |
2106cb18 | 488 | req->in.h.nodeid = ff->nodeid; |
b6aeaded MS |
489 | req->in.numargs = 1; |
490 | req->in.args[0].size = sizeof(struct fuse_read_in); | |
c1aa96a5 | 491 | req->in.args[0].value = inarg; |
b6aeaded MS |
492 | req->out.argvar = 1; |
493 | req->out.numargs = 1; | |
494 | req->out.args[0].size = count; | |
b6aeaded MS |
495 | } |
496 | ||
187c5c36 MP |
497 | static void fuse_release_user_pages(struct fuse_req *req, int write) |
498 | { | |
499 | unsigned i; | |
500 | ||
501 | for (i = 0; i < req->num_pages; i++) { | |
502 | struct page *page = req->pages[i]; | |
503 | if (write) | |
504 | set_page_dirty_lock(page); | |
505 | put_page(page); | |
506 | } | |
507 | } | |
508 | ||
01e9d11a MP |
509 | /** |
510 | * In case of short read, the caller sets 'pos' to the position of | |
511 | * actual end of fuse request in IO request. Otherwise, if bytes_requested | |
512 | * == bytes_transferred or rw == WRITE, the caller sets 'pos' to -1. | |
513 | * | |
514 | * An example: | |
515 | * User requested DIO read of 64K. It was splitted into two 32K fuse requests, | |
516 | * both submitted asynchronously. The first of them was ACKed by userspace as | |
517 | * fully completed (req->out.args[0].size == 32K) resulting in pos == -1. The | |
518 | * second request was ACKed as short, e.g. only 1K was read, resulting in | |
519 | * pos == 33K. | |
520 | * | |
521 | * Thus, when all fuse requests are completed, the minimal non-negative 'pos' | |
522 | * will be equal to the length of the longest contiguous fragment of | |
523 | * transferred data starting from the beginning of IO request. | |
524 | */ | |
525 | static void fuse_aio_complete(struct fuse_io_priv *io, int err, ssize_t pos) | |
526 | { | |
527 | int left; | |
528 | ||
529 | spin_lock(&io->lock); | |
530 | if (err) | |
531 | io->err = io->err ? : err; | |
532 | else if (pos >= 0 && (io->bytes < 0 || pos < io->bytes)) | |
533 | io->bytes = pos; | |
534 | ||
535 | left = --io->reqs; | |
536 | spin_unlock(&io->lock); | |
537 | ||
538 | if (!left) { | |
539 | long res; | |
540 | ||
541 | if (io->err) | |
542 | res = io->err; | |
543 | else if (io->bytes >= 0 && io->write) | |
544 | res = -EIO; | |
545 | else { | |
546 | res = io->bytes < 0 ? io->size : io->bytes; | |
547 | ||
548 | if (!is_sync_kiocb(io->iocb)) { | |
549 | struct path *path = &io->iocb->ki_filp->f_path; | |
550 | struct inode *inode = path->dentry->d_inode; | |
551 | struct fuse_conn *fc = get_fuse_conn(inode); | |
552 | struct fuse_inode *fi = get_fuse_inode(inode); | |
553 | ||
554 | spin_lock(&fc->lock); | |
555 | fi->attr_version = ++fc->attr_version; | |
556 | spin_unlock(&fc->lock); | |
557 | } | |
558 | } | |
559 | ||
560 | aio_complete(io->iocb, res, 0); | |
561 | kfree(io); | |
562 | } | |
563 | } | |
564 | ||
565 | static void fuse_aio_complete_req(struct fuse_conn *fc, struct fuse_req *req) | |
566 | { | |
567 | struct fuse_io_priv *io = req->io; | |
568 | ssize_t pos = -1; | |
569 | ||
570 | fuse_release_user_pages(req, !io->write); | |
571 | ||
572 | if (io->write) { | |
573 | if (req->misc.write.in.size != req->misc.write.out.size) | |
574 | pos = req->misc.write.in.offset - io->offset + | |
575 | req->misc.write.out.size; | |
576 | } else { | |
577 | if (req->misc.read.in.size != req->out.args[0].size) | |
578 | pos = req->misc.read.in.offset - io->offset + | |
579 | req->out.args[0].size; | |
580 | } | |
581 | ||
582 | fuse_aio_complete(io, req->out.h.error, pos); | |
583 | } | |
584 | ||
585 | static size_t fuse_async_req_send(struct fuse_conn *fc, struct fuse_req *req, | |
586 | size_t num_bytes, struct fuse_io_priv *io) | |
587 | { | |
588 | spin_lock(&io->lock); | |
589 | io->size += num_bytes; | |
590 | io->reqs++; | |
591 | spin_unlock(&io->lock); | |
592 | ||
593 | req->io = io; | |
594 | req->end = fuse_aio_complete_req; | |
595 | ||
596 | fuse_request_send_background(fc, req); | |
597 | ||
598 | return num_bytes; | |
599 | } | |
600 | ||
8bfc016d | 601 | static size_t fuse_send_read(struct fuse_req *req, struct file *file, |
2106cb18 | 602 | loff_t pos, size_t count, fl_owner_t owner) |
04730fef | 603 | { |
2106cb18 MS |
604 | struct fuse_file *ff = file->private_data; |
605 | struct fuse_conn *fc = ff->fc; | |
f3332114 | 606 | |
2106cb18 | 607 | fuse_read_fill(req, file, pos, count, FUSE_READ); |
f3332114 | 608 | if (owner != NULL) { |
5c5c5e51 | 609 | struct fuse_read_in *inarg = &req->misc.read.in; |
f3332114 MS |
610 | |
611 | inarg->read_flags |= FUSE_READ_LOCKOWNER; | |
612 | inarg->lock_owner = fuse_lock_owner_id(fc, owner); | |
613 | } | |
b93f858a | 614 | fuse_request_send(fc, req); |
361b1eb5 | 615 | return req->out.args[0].size; |
04730fef MS |
616 | } |
617 | ||
5c5c5e51 MS |
618 | static void fuse_read_update_size(struct inode *inode, loff_t size, |
619 | u64 attr_ver) | |
620 | { | |
621 | struct fuse_conn *fc = get_fuse_conn(inode); | |
622 | struct fuse_inode *fi = get_fuse_inode(inode); | |
623 | ||
624 | spin_lock(&fc->lock); | |
625 | if (attr_ver == fi->attr_version && size < inode->i_size) { | |
626 | fi->attr_version = ++fc->attr_version; | |
627 | i_size_write(inode, size); | |
628 | } | |
629 | spin_unlock(&fc->lock); | |
630 | } | |
631 | ||
b6aeaded MS |
632 | static int fuse_readpage(struct file *file, struct page *page) |
633 | { | |
634 | struct inode *inode = page->mapping->host; | |
635 | struct fuse_conn *fc = get_fuse_conn(inode); | |
248d86e8 | 636 | struct fuse_req *req; |
5c5c5e51 MS |
637 | size_t num_read; |
638 | loff_t pos = page_offset(page); | |
639 | size_t count = PAGE_CACHE_SIZE; | |
640 | u64 attr_ver; | |
248d86e8 MS |
641 | int err; |
642 | ||
643 | err = -EIO; | |
644 | if (is_bad_inode(inode)) | |
645 | goto out; | |
646 | ||
3be5a52b | 647 | /* |
25985edc | 648 | * Page writeback can extend beyond the lifetime of the |
3be5a52b MS |
649 | * page-cache page, so make sure we read a properly synced |
650 | * page. | |
651 | */ | |
652 | fuse_wait_on_page_writeback(inode, page->index); | |
653 | ||
b111c8c0 | 654 | req = fuse_get_req(fc, 1); |
ce1d5a49 MS |
655 | err = PTR_ERR(req); |
656 | if (IS_ERR(req)) | |
b6aeaded MS |
657 | goto out; |
658 | ||
5c5c5e51 MS |
659 | attr_ver = fuse_get_attr_version(fc); |
660 | ||
b6aeaded | 661 | req->out.page_zeroing = 1; |
f4975c67 | 662 | req->out.argpages = 1; |
b6aeaded MS |
663 | req->num_pages = 1; |
664 | req->pages[0] = page; | |
85f40aec | 665 | req->page_descs[0].length = count; |
2106cb18 | 666 | num_read = fuse_send_read(req, file, pos, count, NULL); |
b6aeaded MS |
667 | err = req->out.h.error; |
668 | fuse_put_request(fc, req); | |
5c5c5e51 MS |
669 | |
670 | if (!err) { | |
671 | /* | |
672 | * Short read means EOF. If file size is larger, truncate it | |
673 | */ | |
674 | if (num_read < count) | |
675 | fuse_read_update_size(inode, pos + num_read, attr_ver); | |
676 | ||
b6aeaded | 677 | SetPageUptodate(page); |
5c5c5e51 MS |
678 | } |
679 | ||
b36c31ba | 680 | fuse_invalidate_attr(inode); /* atime changed */ |
b6aeaded MS |
681 | out: |
682 | unlock_page(page); | |
683 | return err; | |
684 | } | |
685 | ||
c1aa96a5 | 686 | static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req) |
db50b96c | 687 | { |
c1aa96a5 | 688 | int i; |
5c5c5e51 MS |
689 | size_t count = req->misc.read.in.size; |
690 | size_t num_read = req->out.args[0].size; | |
ce534fb0 | 691 | struct address_space *mapping = NULL; |
c1aa96a5 | 692 | |
ce534fb0 MS |
693 | for (i = 0; mapping == NULL && i < req->num_pages; i++) |
694 | mapping = req->pages[i]->mapping; | |
5c5c5e51 | 695 | |
ce534fb0 MS |
696 | if (mapping) { |
697 | struct inode *inode = mapping->host; | |
698 | ||
699 | /* | |
700 | * Short read means EOF. If file size is larger, truncate it | |
701 | */ | |
702 | if (!req->out.h.error && num_read < count) { | |
703 | loff_t pos; | |
704 | ||
705 | pos = page_offset(req->pages[0]) + num_read; | |
706 | fuse_read_update_size(inode, pos, | |
707 | req->misc.read.attr_ver); | |
708 | } | |
709 | fuse_invalidate_attr(inode); /* atime changed */ | |
710 | } | |
c1aa96a5 | 711 | |
db50b96c MS |
712 | for (i = 0; i < req->num_pages; i++) { |
713 | struct page *page = req->pages[i]; | |
714 | if (!req->out.h.error) | |
715 | SetPageUptodate(page); | |
c1aa96a5 MS |
716 | else |
717 | SetPageError(page); | |
db50b96c | 718 | unlock_page(page); |
b5dd3285 | 719 | page_cache_release(page); |
db50b96c | 720 | } |
c756e0a4 | 721 | if (req->ff) |
5a18ec17 | 722 | fuse_file_put(req->ff, false); |
c1aa96a5 MS |
723 | } |
724 | ||
2106cb18 | 725 | static void fuse_send_readpages(struct fuse_req *req, struct file *file) |
c1aa96a5 | 726 | { |
2106cb18 MS |
727 | struct fuse_file *ff = file->private_data; |
728 | struct fuse_conn *fc = ff->fc; | |
c1aa96a5 MS |
729 | loff_t pos = page_offset(req->pages[0]); |
730 | size_t count = req->num_pages << PAGE_CACHE_SHIFT; | |
f4975c67 MS |
731 | |
732 | req->out.argpages = 1; | |
c1aa96a5 | 733 | req->out.page_zeroing = 1; |
ce534fb0 | 734 | req->out.page_replace = 1; |
2106cb18 | 735 | fuse_read_fill(req, file, pos, count, FUSE_READ); |
5c5c5e51 | 736 | req->misc.read.attr_ver = fuse_get_attr_version(fc); |
9cd68455 | 737 | if (fc->async_read) { |
c756e0a4 | 738 | req->ff = fuse_file_get(ff); |
9cd68455 | 739 | req->end = fuse_readpages_end; |
b93f858a | 740 | fuse_request_send_background(fc, req); |
9cd68455 | 741 | } else { |
b93f858a | 742 | fuse_request_send(fc, req); |
9cd68455 | 743 | fuse_readpages_end(fc, req); |
e9bb09dd | 744 | fuse_put_request(fc, req); |
9cd68455 | 745 | } |
db50b96c MS |
746 | } |
747 | ||
c756e0a4 | 748 | struct fuse_fill_data { |
db50b96c | 749 | struct fuse_req *req; |
a6643094 | 750 | struct file *file; |
db50b96c | 751 | struct inode *inode; |
f8dbdf81 | 752 | unsigned nr_pages; |
db50b96c MS |
753 | }; |
754 | ||
755 | static int fuse_readpages_fill(void *_data, struct page *page) | |
756 | { | |
c756e0a4 | 757 | struct fuse_fill_data *data = _data; |
db50b96c MS |
758 | struct fuse_req *req = data->req; |
759 | struct inode *inode = data->inode; | |
760 | struct fuse_conn *fc = get_fuse_conn(inode); | |
761 | ||
3be5a52b MS |
762 | fuse_wait_on_page_writeback(inode, page->index); |
763 | ||
db50b96c MS |
764 | if (req->num_pages && |
765 | (req->num_pages == FUSE_MAX_PAGES_PER_REQ || | |
766 | (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_read || | |
767 | req->pages[req->num_pages - 1]->index + 1 != page->index)) { | |
f8dbdf81 MP |
768 | int nr_alloc = min_t(unsigned, data->nr_pages, |
769 | FUSE_MAX_PAGES_PER_REQ); | |
2106cb18 | 770 | fuse_send_readpages(req, data->file); |
8b41e671 MP |
771 | if (fc->async_read) |
772 | req = fuse_get_req_for_background(fc, nr_alloc); | |
773 | else | |
774 | req = fuse_get_req(fc, nr_alloc); | |
775 | ||
776 | data->req = req; | |
ce1d5a49 | 777 | if (IS_ERR(req)) { |
db50b96c | 778 | unlock_page(page); |
ce1d5a49 | 779 | return PTR_ERR(req); |
db50b96c | 780 | } |
db50b96c | 781 | } |
f8dbdf81 MP |
782 | |
783 | if (WARN_ON(req->num_pages >= req->max_pages)) { | |
784 | fuse_put_request(fc, req); | |
785 | return -EIO; | |
786 | } | |
787 | ||
b5dd3285 | 788 | page_cache_get(page); |
db50b96c | 789 | req->pages[req->num_pages] = page; |
85f40aec | 790 | req->page_descs[req->num_pages].length = PAGE_SIZE; |
1729a16c | 791 | req->num_pages++; |
f8dbdf81 | 792 | data->nr_pages--; |
db50b96c MS |
793 | return 0; |
794 | } | |
795 | ||
796 | static int fuse_readpages(struct file *file, struct address_space *mapping, | |
797 | struct list_head *pages, unsigned nr_pages) | |
798 | { | |
799 | struct inode *inode = mapping->host; | |
800 | struct fuse_conn *fc = get_fuse_conn(inode); | |
c756e0a4 | 801 | struct fuse_fill_data data; |
db50b96c | 802 | int err; |
f8dbdf81 | 803 | int nr_alloc = min_t(unsigned, nr_pages, FUSE_MAX_PAGES_PER_REQ); |
248d86e8 | 804 | |
1d7ea732 | 805 | err = -EIO; |
248d86e8 | 806 | if (is_bad_inode(inode)) |
2e990021 | 807 | goto out; |
248d86e8 | 808 | |
a6643094 | 809 | data.file = file; |
db50b96c | 810 | data.inode = inode; |
8b41e671 MP |
811 | if (fc->async_read) |
812 | data.req = fuse_get_req_for_background(fc, nr_alloc); | |
813 | else | |
814 | data.req = fuse_get_req(fc, nr_alloc); | |
f8dbdf81 | 815 | data.nr_pages = nr_pages; |
1d7ea732 | 816 | err = PTR_ERR(data.req); |
ce1d5a49 | 817 | if (IS_ERR(data.req)) |
2e990021 | 818 | goto out; |
db50b96c MS |
819 | |
820 | err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data); | |
d3406ffa MS |
821 | if (!err) { |
822 | if (data.req->num_pages) | |
2106cb18 | 823 | fuse_send_readpages(data.req, file); |
d3406ffa MS |
824 | else |
825 | fuse_put_request(fc, data.req); | |
826 | } | |
2e990021 | 827 | out: |
1d7ea732 | 828 | return err; |
db50b96c MS |
829 | } |
830 | ||
bcb4be80 MS |
831 | static ssize_t fuse_file_aio_read(struct kiocb *iocb, const struct iovec *iov, |
832 | unsigned long nr_segs, loff_t pos) | |
833 | { | |
834 | struct inode *inode = iocb->ki_filp->f_mapping->host; | |
a8894274 | 835 | struct fuse_conn *fc = get_fuse_conn(inode); |
bcb4be80 | 836 | |
a8894274 BF |
837 | /* |
838 | * In auto invalidate mode, always update attributes on read. | |
839 | * Otherwise, only update if we attempt to read past EOF (to ensure | |
840 | * i_size is up to date). | |
841 | */ | |
842 | if (fc->auto_inval_data || | |
843 | (pos + iov_length(iov, nr_segs) > i_size_read(inode))) { | |
bcb4be80 | 844 | int err; |
bcb4be80 MS |
845 | err = fuse_update_attributes(inode, NULL, iocb->ki_filp, NULL); |
846 | if (err) | |
847 | return err; | |
848 | } | |
849 | ||
850 | return generic_file_aio_read(iocb, iov, nr_segs, pos); | |
851 | } | |
852 | ||
2d698b07 | 853 | static void fuse_write_fill(struct fuse_req *req, struct fuse_file *ff, |
2106cb18 | 854 | loff_t pos, size_t count) |
b6aeaded | 855 | { |
b25e82e5 MS |
856 | struct fuse_write_in *inarg = &req->misc.write.in; |
857 | struct fuse_write_out *outarg = &req->misc.write.out; | |
b6aeaded | 858 | |
b25e82e5 MS |
859 | inarg->fh = ff->fh; |
860 | inarg->offset = pos; | |
861 | inarg->size = count; | |
b6aeaded | 862 | req->in.h.opcode = FUSE_WRITE; |
2106cb18 | 863 | req->in.h.nodeid = ff->nodeid; |
b6aeaded | 864 | req->in.numargs = 2; |
2106cb18 | 865 | if (ff->fc->minor < 9) |
f3332114 MS |
866 | req->in.args[0].size = FUSE_COMPAT_WRITE_IN_SIZE; |
867 | else | |
868 | req->in.args[0].size = sizeof(struct fuse_write_in); | |
b25e82e5 | 869 | req->in.args[0].value = inarg; |
b6aeaded MS |
870 | req->in.args[1].size = count; |
871 | req->out.numargs = 1; | |
872 | req->out.args[0].size = sizeof(struct fuse_write_out); | |
b25e82e5 MS |
873 | req->out.args[0].value = outarg; |
874 | } | |
875 | ||
876 | static size_t fuse_send_write(struct fuse_req *req, struct file *file, | |
2106cb18 | 877 | loff_t pos, size_t count, fl_owner_t owner) |
b25e82e5 | 878 | { |
2106cb18 MS |
879 | struct fuse_file *ff = file->private_data; |
880 | struct fuse_conn *fc = ff->fc; | |
2d698b07 MS |
881 | struct fuse_write_in *inarg = &req->misc.write.in; |
882 | ||
2106cb18 | 883 | fuse_write_fill(req, ff, pos, count); |
2d698b07 | 884 | inarg->flags = file->f_flags; |
f3332114 | 885 | if (owner != NULL) { |
f3332114 MS |
886 | inarg->write_flags |= FUSE_WRITE_LOCKOWNER; |
887 | inarg->lock_owner = fuse_lock_owner_id(fc, owner); | |
888 | } | |
b93f858a | 889 | fuse_request_send(fc, req); |
b25e82e5 | 890 | return req->misc.write.out.size; |
b6aeaded MS |
891 | } |
892 | ||
a1d75f25 | 893 | void fuse_write_update_size(struct inode *inode, loff_t pos) |
854512ec MS |
894 | { |
895 | struct fuse_conn *fc = get_fuse_conn(inode); | |
896 | struct fuse_inode *fi = get_fuse_inode(inode); | |
897 | ||
898 | spin_lock(&fc->lock); | |
899 | fi->attr_version = ++fc->attr_version; | |
900 | if (pos > inode->i_size) | |
901 | i_size_write(inode, pos); | |
902 | spin_unlock(&fc->lock); | |
903 | } | |
904 | ||
ea9b9907 NP |
905 | static size_t fuse_send_write_pages(struct fuse_req *req, struct file *file, |
906 | struct inode *inode, loff_t pos, | |
907 | size_t count) | |
908 | { | |
909 | size_t res; | |
910 | unsigned offset; | |
911 | unsigned i; | |
912 | ||
913 | for (i = 0; i < req->num_pages; i++) | |
914 | fuse_wait_on_page_writeback(inode, req->pages[i]->index); | |
915 | ||
2106cb18 | 916 | res = fuse_send_write(req, file, pos, count, NULL); |
ea9b9907 | 917 | |
b2430d75 | 918 | offset = req->page_descs[0].offset; |
ea9b9907 NP |
919 | count = res; |
920 | for (i = 0; i < req->num_pages; i++) { | |
921 | struct page *page = req->pages[i]; | |
922 | ||
923 | if (!req->out.h.error && !offset && count >= PAGE_CACHE_SIZE) | |
924 | SetPageUptodate(page); | |
925 | ||
926 | if (count > PAGE_CACHE_SIZE - offset) | |
927 | count -= PAGE_CACHE_SIZE - offset; | |
928 | else | |
929 | count = 0; | |
930 | offset = 0; | |
931 | ||
932 | unlock_page(page); | |
933 | page_cache_release(page); | |
934 | } | |
935 | ||
936 | return res; | |
937 | } | |
938 | ||
939 | static ssize_t fuse_fill_write_pages(struct fuse_req *req, | |
940 | struct address_space *mapping, | |
941 | struct iov_iter *ii, loff_t pos) | |
942 | { | |
943 | struct fuse_conn *fc = get_fuse_conn(mapping->host); | |
944 | unsigned offset = pos & (PAGE_CACHE_SIZE - 1); | |
945 | size_t count = 0; | |
946 | int err; | |
947 | ||
f4975c67 | 948 | req->in.argpages = 1; |
b2430d75 | 949 | req->page_descs[0].offset = offset; |
ea9b9907 NP |
950 | |
951 | do { | |
952 | size_t tmp; | |
953 | struct page *page; | |
954 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; | |
955 | size_t bytes = min_t(size_t, PAGE_CACHE_SIZE - offset, | |
956 | iov_iter_count(ii)); | |
957 | ||
958 | bytes = min_t(size_t, bytes, fc->max_write - count); | |
959 | ||
960 | again: | |
961 | err = -EFAULT; | |
962 | if (iov_iter_fault_in_readable(ii, bytes)) | |
963 | break; | |
964 | ||
965 | err = -ENOMEM; | |
54566b2c | 966 | page = grab_cache_page_write_begin(mapping, index, 0); |
ea9b9907 NP |
967 | if (!page) |
968 | break; | |
969 | ||
931e80e4 | 970 | if (mapping_writably_mapped(mapping)) |
971 | flush_dcache_page(page); | |
972 | ||
ea9b9907 NP |
973 | pagefault_disable(); |
974 | tmp = iov_iter_copy_from_user_atomic(page, ii, offset, bytes); | |
975 | pagefault_enable(); | |
976 | flush_dcache_page(page); | |
977 | ||
478e0841 JW |
978 | mark_page_accessed(page); |
979 | ||
ea9b9907 NP |
980 | if (!tmp) { |
981 | unlock_page(page); | |
982 | page_cache_release(page); | |
983 | bytes = min(bytes, iov_iter_single_seg_count(ii)); | |
984 | goto again; | |
985 | } | |
986 | ||
987 | err = 0; | |
988 | req->pages[req->num_pages] = page; | |
85f40aec | 989 | req->page_descs[req->num_pages].length = tmp; |
ea9b9907 NP |
990 | req->num_pages++; |
991 | ||
992 | iov_iter_advance(ii, tmp); | |
993 | count += tmp; | |
994 | pos += tmp; | |
995 | offset += tmp; | |
996 | if (offset == PAGE_CACHE_SIZE) | |
997 | offset = 0; | |
998 | ||
78bb6cb9 MS |
999 | if (!fc->big_writes) |
1000 | break; | |
ea9b9907 | 1001 | } while (iov_iter_count(ii) && count < fc->max_write && |
d07f09f5 | 1002 | req->num_pages < req->max_pages && offset == 0); |
ea9b9907 NP |
1003 | |
1004 | return count > 0 ? count : err; | |
1005 | } | |
1006 | ||
d07f09f5 MP |
1007 | static inline unsigned fuse_wr_pages(loff_t pos, size_t len) |
1008 | { | |
1009 | return min_t(unsigned, | |
1010 | ((pos + len - 1) >> PAGE_CACHE_SHIFT) - | |
1011 | (pos >> PAGE_CACHE_SHIFT) + 1, | |
1012 | FUSE_MAX_PAGES_PER_REQ); | |
1013 | } | |
1014 | ||
ea9b9907 NP |
1015 | static ssize_t fuse_perform_write(struct file *file, |
1016 | struct address_space *mapping, | |
1017 | struct iov_iter *ii, loff_t pos) | |
1018 | { | |
1019 | struct inode *inode = mapping->host; | |
1020 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1021 | int err = 0; | |
1022 | ssize_t res = 0; | |
1023 | ||
1024 | if (is_bad_inode(inode)) | |
1025 | return -EIO; | |
1026 | ||
1027 | do { | |
1028 | struct fuse_req *req; | |
1029 | ssize_t count; | |
d07f09f5 | 1030 | unsigned nr_pages = fuse_wr_pages(pos, iov_iter_count(ii)); |
ea9b9907 | 1031 | |
d07f09f5 | 1032 | req = fuse_get_req(fc, nr_pages); |
ea9b9907 NP |
1033 | if (IS_ERR(req)) { |
1034 | err = PTR_ERR(req); | |
1035 | break; | |
1036 | } | |
1037 | ||
1038 | count = fuse_fill_write_pages(req, mapping, ii, pos); | |
1039 | if (count <= 0) { | |
1040 | err = count; | |
1041 | } else { | |
1042 | size_t num_written; | |
1043 | ||
1044 | num_written = fuse_send_write_pages(req, file, inode, | |
1045 | pos, count); | |
1046 | err = req->out.h.error; | |
1047 | if (!err) { | |
1048 | res += num_written; | |
1049 | pos += num_written; | |
1050 | ||
1051 | /* break out of the loop on short write */ | |
1052 | if (num_written != count) | |
1053 | err = -EIO; | |
1054 | } | |
1055 | } | |
1056 | fuse_put_request(fc, req); | |
1057 | } while (!err && iov_iter_count(ii)); | |
1058 | ||
1059 | if (res > 0) | |
1060 | fuse_write_update_size(inode, pos); | |
1061 | ||
1062 | fuse_invalidate_attr(inode); | |
1063 | ||
1064 | return res > 0 ? res : err; | |
1065 | } | |
1066 | ||
1067 | static ssize_t fuse_file_aio_write(struct kiocb *iocb, const struct iovec *iov, | |
1068 | unsigned long nr_segs, loff_t pos) | |
1069 | { | |
1070 | struct file *file = iocb->ki_filp; | |
1071 | struct address_space *mapping = file->f_mapping; | |
1072 | size_t count = 0; | |
4273b793 | 1073 | size_t ocount = 0; |
ea9b9907 | 1074 | ssize_t written = 0; |
4273b793 | 1075 | ssize_t written_buffered = 0; |
ea9b9907 NP |
1076 | struct inode *inode = mapping->host; |
1077 | ssize_t err; | |
1078 | struct iov_iter i; | |
4273b793 | 1079 | loff_t endbyte = 0; |
ea9b9907 NP |
1080 | |
1081 | WARN_ON(iocb->ki_pos != pos); | |
1082 | ||
4273b793 AA |
1083 | ocount = 0; |
1084 | err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); | |
ea9b9907 NP |
1085 | if (err) |
1086 | return err; | |
1087 | ||
4273b793 | 1088 | count = ocount; |
58ef6a75 | 1089 | sb_start_write(inode->i_sb); |
ea9b9907 | 1090 | mutex_lock(&inode->i_mutex); |
ea9b9907 NP |
1091 | |
1092 | /* We can write back this queue in page reclaim */ | |
1093 | current->backing_dev_info = mapping->backing_dev_info; | |
1094 | ||
1095 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); | |
1096 | if (err) | |
1097 | goto out; | |
1098 | ||
1099 | if (count == 0) | |
1100 | goto out; | |
1101 | ||
2f1936b8 | 1102 | err = file_remove_suid(file); |
ea9b9907 NP |
1103 | if (err) |
1104 | goto out; | |
1105 | ||
c3b2da31 JB |
1106 | err = file_update_time(file); |
1107 | if (err) | |
1108 | goto out; | |
ea9b9907 | 1109 | |
4273b793 AA |
1110 | if (file->f_flags & O_DIRECT) { |
1111 | written = generic_file_direct_write(iocb, iov, &nr_segs, | |
1112 | pos, &iocb->ki_pos, | |
1113 | count, ocount); | |
1114 | if (written < 0 || written == count) | |
1115 | goto out; | |
1116 | ||
1117 | pos += written; | |
1118 | count -= written; | |
ea9b9907 | 1119 | |
4273b793 AA |
1120 | iov_iter_init(&i, iov, nr_segs, count, written); |
1121 | written_buffered = fuse_perform_write(file, mapping, &i, pos); | |
1122 | if (written_buffered < 0) { | |
1123 | err = written_buffered; | |
1124 | goto out; | |
1125 | } | |
1126 | endbyte = pos + written_buffered - 1; | |
1127 | ||
1128 | err = filemap_write_and_wait_range(file->f_mapping, pos, | |
1129 | endbyte); | |
1130 | if (err) | |
1131 | goto out; | |
1132 | ||
1133 | invalidate_mapping_pages(file->f_mapping, | |
1134 | pos >> PAGE_CACHE_SHIFT, | |
1135 | endbyte >> PAGE_CACHE_SHIFT); | |
1136 | ||
1137 | written += written_buffered; | |
1138 | iocb->ki_pos = pos + written_buffered; | |
1139 | } else { | |
1140 | iov_iter_init(&i, iov, nr_segs, count, 0); | |
1141 | written = fuse_perform_write(file, mapping, &i, pos); | |
1142 | if (written >= 0) | |
1143 | iocb->ki_pos = pos + written; | |
1144 | } | |
ea9b9907 NP |
1145 | out: |
1146 | current->backing_dev_info = NULL; | |
1147 | mutex_unlock(&inode->i_mutex); | |
58ef6a75 | 1148 | sb_end_write(inode->i_sb); |
ea9b9907 NP |
1149 | |
1150 | return written ? written : err; | |
1151 | } | |
1152 | ||
7c190c8b MP |
1153 | static inline void fuse_page_descs_length_init(struct fuse_req *req, |
1154 | unsigned index, unsigned nr_pages) | |
85f40aec MP |
1155 | { |
1156 | int i; | |
1157 | ||
7c190c8b | 1158 | for (i = index; i < index + nr_pages; i++) |
85f40aec MP |
1159 | req->page_descs[i].length = PAGE_SIZE - |
1160 | req->page_descs[i].offset; | |
1161 | } | |
1162 | ||
7c190c8b MP |
1163 | static inline unsigned long fuse_get_user_addr(const struct iov_iter *ii) |
1164 | { | |
1165 | return (unsigned long)ii->iov->iov_base + ii->iov_offset; | |
1166 | } | |
1167 | ||
1168 | static inline size_t fuse_get_frag_size(const struct iov_iter *ii, | |
1169 | size_t max_size) | |
1170 | { | |
1171 | return min(iov_iter_single_seg_count(ii), max_size); | |
1172 | } | |
1173 | ||
b98d023a | 1174 | static int fuse_get_user_pages(struct fuse_req *req, struct iov_iter *ii, |
ce60a2f1 | 1175 | size_t *nbytesp, int write) |
413ef8cb | 1176 | { |
7c190c8b | 1177 | size_t nbytes = 0; /* # bytes already packed in req */ |
b98d023a | 1178 | |
f4975c67 MS |
1179 | /* Special case for kernel I/O: can copy directly into the buffer */ |
1180 | if (segment_eq(get_fs(), KERNEL_DS)) { | |
7c190c8b MP |
1181 | unsigned long user_addr = fuse_get_user_addr(ii); |
1182 | size_t frag_size = fuse_get_frag_size(ii, *nbytesp); | |
1183 | ||
f4975c67 MS |
1184 | if (write) |
1185 | req->in.args[1].value = (void *) user_addr; | |
1186 | else | |
1187 | req->out.args[0].value = (void *) user_addr; | |
1188 | ||
b98d023a MP |
1189 | iov_iter_advance(ii, frag_size); |
1190 | *nbytesp = frag_size; | |
f4975c67 MS |
1191 | return 0; |
1192 | } | |
413ef8cb | 1193 | |
5565a9d8 | 1194 | while (nbytes < *nbytesp && req->num_pages < req->max_pages) { |
7c190c8b MP |
1195 | unsigned npages; |
1196 | unsigned long user_addr = fuse_get_user_addr(ii); | |
1197 | unsigned offset = user_addr & ~PAGE_MASK; | |
1198 | size_t frag_size = fuse_get_frag_size(ii, *nbytesp - nbytes); | |
1199 | int ret; | |
413ef8cb | 1200 | |
5565a9d8 | 1201 | unsigned n = req->max_pages - req->num_pages; |
7c190c8b MP |
1202 | frag_size = min_t(size_t, frag_size, n << PAGE_SHIFT); |
1203 | ||
1204 | npages = (frag_size + offset + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1205 | npages = clamp(npages, 1U, n); | |
1206 | ||
1207 | ret = get_user_pages_fast(user_addr, npages, !write, | |
1208 | &req->pages[req->num_pages]); | |
1209 | if (ret < 0) | |
1210 | return ret; | |
1211 | ||
1212 | npages = ret; | |
1213 | frag_size = min_t(size_t, frag_size, | |
1214 | (npages << PAGE_SHIFT) - offset); | |
1215 | iov_iter_advance(ii, frag_size); | |
1216 | ||
1217 | req->page_descs[req->num_pages].offset = offset; | |
1218 | fuse_page_descs_length_init(req, req->num_pages, npages); | |
1219 | ||
1220 | req->num_pages += npages; | |
1221 | req->page_descs[req->num_pages - 1].length -= | |
1222 | (npages << PAGE_SHIFT) - offset - frag_size; | |
1223 | ||
1224 | nbytes += frag_size; | |
1225 | } | |
f4975c67 MS |
1226 | |
1227 | if (write) | |
1228 | req->in.argpages = 1; | |
1229 | else | |
1230 | req->out.argpages = 1; | |
1231 | ||
7c190c8b | 1232 | *nbytesp = nbytes; |
f4975c67 | 1233 | |
413ef8cb MS |
1234 | return 0; |
1235 | } | |
1236 | ||
5565a9d8 MP |
1237 | static inline int fuse_iter_npages(const struct iov_iter *ii_p) |
1238 | { | |
1239 | struct iov_iter ii = *ii_p; | |
1240 | int npages = 0; | |
1241 | ||
1242 | while (iov_iter_count(&ii) && npages < FUSE_MAX_PAGES_PER_REQ) { | |
1243 | unsigned long user_addr = fuse_get_user_addr(&ii); | |
1244 | unsigned offset = user_addr & ~PAGE_MASK; | |
1245 | size_t frag_size = iov_iter_single_seg_count(&ii); | |
1246 | ||
1247 | npages += (frag_size + offset + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1248 | iov_iter_advance(&ii, frag_size); | |
1249 | } | |
1250 | ||
1251 | return min(npages, FUSE_MAX_PAGES_PER_REQ); | |
1252 | } | |
1253 | ||
fb05f41f MS |
1254 | ssize_t fuse_direct_io(struct file *file, const struct iovec *iov, |
1255 | unsigned long nr_segs, size_t count, loff_t *ppos, | |
1256 | int write) | |
413ef8cb | 1257 | { |
2106cb18 MS |
1258 | struct fuse_file *ff = file->private_data; |
1259 | struct fuse_conn *fc = ff->fc; | |
413ef8cb MS |
1260 | size_t nmax = write ? fc->max_write : fc->max_read; |
1261 | loff_t pos = *ppos; | |
1262 | ssize_t res = 0; | |
248d86e8 | 1263 | struct fuse_req *req; |
b98d023a MP |
1264 | struct iov_iter ii; |
1265 | ||
1266 | iov_iter_init(&ii, iov, nr_segs, count, 0); | |
248d86e8 | 1267 | |
5565a9d8 | 1268 | req = fuse_get_req(fc, fuse_iter_npages(&ii)); |
ce1d5a49 MS |
1269 | if (IS_ERR(req)) |
1270 | return PTR_ERR(req); | |
413ef8cb MS |
1271 | |
1272 | while (count) { | |
413ef8cb | 1273 | size_t nres; |
2106cb18 | 1274 | fl_owner_t owner = current->files; |
f4975c67 | 1275 | size_t nbytes = min(count, nmax); |
b98d023a | 1276 | int err = fuse_get_user_pages(req, &ii, &nbytes, write); |
413ef8cb MS |
1277 | if (err) { |
1278 | res = err; | |
1279 | break; | |
1280 | } | |
f4975c67 | 1281 | |
413ef8cb | 1282 | if (write) |
2106cb18 | 1283 | nres = fuse_send_write(req, file, pos, nbytes, owner); |
413ef8cb | 1284 | else |
2106cb18 MS |
1285 | nres = fuse_send_read(req, file, pos, nbytes, owner); |
1286 | ||
413ef8cb MS |
1287 | fuse_release_user_pages(req, !write); |
1288 | if (req->out.h.error) { | |
1289 | if (!res) | |
1290 | res = req->out.h.error; | |
1291 | break; | |
1292 | } else if (nres > nbytes) { | |
1293 | res = -EIO; | |
1294 | break; | |
1295 | } | |
1296 | count -= nres; | |
1297 | res += nres; | |
1298 | pos += nres; | |
413ef8cb MS |
1299 | if (nres != nbytes) |
1300 | break; | |
56cf34ff MS |
1301 | if (count) { |
1302 | fuse_put_request(fc, req); | |
5565a9d8 | 1303 | req = fuse_get_req(fc, fuse_iter_npages(&ii)); |
56cf34ff MS |
1304 | if (IS_ERR(req)) |
1305 | break; | |
1306 | } | |
413ef8cb | 1307 | } |
f60311d5 AA |
1308 | if (!IS_ERR(req)) |
1309 | fuse_put_request(fc, req); | |
d09cb9d7 | 1310 | if (res > 0) |
413ef8cb | 1311 | *ppos = pos; |
413ef8cb MS |
1312 | |
1313 | return res; | |
1314 | } | |
08cbf542 | 1315 | EXPORT_SYMBOL_GPL(fuse_direct_io); |
413ef8cb | 1316 | |
b98d023a MP |
1317 | static ssize_t __fuse_direct_read(struct file *file, const struct iovec *iov, |
1318 | unsigned long nr_segs, loff_t *ppos) | |
413ef8cb | 1319 | { |
d09cb9d7 | 1320 | ssize_t res; |
6131ffaa | 1321 | struct inode *inode = file_inode(file); |
d09cb9d7 MS |
1322 | |
1323 | if (is_bad_inode(inode)) | |
1324 | return -EIO; | |
1325 | ||
fb05f41f MS |
1326 | res = fuse_direct_io(file, iov, nr_segs, iov_length(iov, nr_segs), |
1327 | ppos, 0); | |
d09cb9d7 MS |
1328 | |
1329 | fuse_invalidate_attr(inode); | |
1330 | ||
1331 | return res; | |
413ef8cb MS |
1332 | } |
1333 | ||
b98d023a MP |
1334 | static ssize_t fuse_direct_read(struct file *file, char __user *buf, |
1335 | size_t count, loff_t *ppos) | |
1336 | { | |
fb05f41f | 1337 | struct iovec iov = { .iov_base = buf, .iov_len = count }; |
b98d023a MP |
1338 | return __fuse_direct_read(file, &iov, 1, ppos); |
1339 | } | |
1340 | ||
1341 | static ssize_t __fuse_direct_write(struct file *file, const struct iovec *iov, | |
1342 | unsigned long nr_segs, loff_t *ppos) | |
413ef8cb | 1343 | { |
6131ffaa | 1344 | struct inode *inode = file_inode(file); |
b98d023a | 1345 | size_t count = iov_length(iov, nr_segs); |
413ef8cb | 1346 | ssize_t res; |
d09cb9d7 | 1347 | |
889f7848 | 1348 | res = generic_write_checks(file, ppos, &count, 0); |
d09cb9d7 | 1349 | if (!res) { |
fb05f41f | 1350 | res = fuse_direct_io(file, iov, nr_segs, count, ppos, 1); |
d09cb9d7 MS |
1351 | if (res > 0) |
1352 | fuse_write_update_size(inode, *ppos); | |
1353 | } | |
d09cb9d7 MS |
1354 | |
1355 | fuse_invalidate_attr(inode); | |
1356 | ||
413ef8cb MS |
1357 | return res; |
1358 | } | |
1359 | ||
4273b793 AA |
1360 | static ssize_t fuse_direct_write(struct file *file, const char __user *buf, |
1361 | size_t count, loff_t *ppos) | |
1362 | { | |
fb05f41f | 1363 | struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = count }; |
6131ffaa | 1364 | struct inode *inode = file_inode(file); |
4273b793 AA |
1365 | ssize_t res; |
1366 | ||
1367 | if (is_bad_inode(inode)) | |
1368 | return -EIO; | |
1369 | ||
1370 | /* Don't allow parallel writes to the same file */ | |
1371 | mutex_lock(&inode->i_mutex); | |
b98d023a | 1372 | res = __fuse_direct_write(file, &iov, 1, ppos); |
4273b793 AA |
1373 | mutex_unlock(&inode->i_mutex); |
1374 | ||
1375 | return res; | |
1376 | } | |
1377 | ||
3be5a52b | 1378 | static void fuse_writepage_free(struct fuse_conn *fc, struct fuse_req *req) |
b6aeaded | 1379 | { |
3be5a52b | 1380 | __free_page(req->pages[0]); |
5a18ec17 | 1381 | fuse_file_put(req->ff, false); |
3be5a52b MS |
1382 | } |
1383 | ||
1384 | static void fuse_writepage_finish(struct fuse_conn *fc, struct fuse_req *req) | |
1385 | { | |
1386 | struct inode *inode = req->inode; | |
1387 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1388 | struct backing_dev_info *bdi = inode->i_mapping->backing_dev_info; | |
1389 | ||
1390 | list_del(&req->writepages_entry); | |
1391 | dec_bdi_stat(bdi, BDI_WRITEBACK); | |
1392 | dec_zone_page_state(req->pages[0], NR_WRITEBACK_TEMP); | |
1393 | bdi_writeout_inc(bdi); | |
1394 | wake_up(&fi->page_waitq); | |
1395 | } | |
1396 | ||
1397 | /* Called under fc->lock, may release and reacquire it */ | |
1398 | static void fuse_send_writepage(struct fuse_conn *fc, struct fuse_req *req) | |
b9ca67b2 MS |
1399 | __releases(fc->lock) |
1400 | __acquires(fc->lock) | |
3be5a52b MS |
1401 | { |
1402 | struct fuse_inode *fi = get_fuse_inode(req->inode); | |
1403 | loff_t size = i_size_read(req->inode); | |
1404 | struct fuse_write_in *inarg = &req->misc.write.in; | |
1405 | ||
1406 | if (!fc->connected) | |
1407 | goto out_free; | |
1408 | ||
1409 | if (inarg->offset + PAGE_CACHE_SIZE <= size) { | |
1410 | inarg->size = PAGE_CACHE_SIZE; | |
1411 | } else if (inarg->offset < size) { | |
1412 | inarg->size = size & (PAGE_CACHE_SIZE - 1); | |
1413 | } else { | |
1414 | /* Got truncated off completely */ | |
1415 | goto out_free; | |
b6aeaded | 1416 | } |
3be5a52b MS |
1417 | |
1418 | req->in.args[1].size = inarg->size; | |
1419 | fi->writectr++; | |
b93f858a | 1420 | fuse_request_send_background_locked(fc, req); |
3be5a52b MS |
1421 | return; |
1422 | ||
1423 | out_free: | |
1424 | fuse_writepage_finish(fc, req); | |
1425 | spin_unlock(&fc->lock); | |
1426 | fuse_writepage_free(fc, req); | |
e9bb09dd | 1427 | fuse_put_request(fc, req); |
3be5a52b | 1428 | spin_lock(&fc->lock); |
b6aeaded MS |
1429 | } |
1430 | ||
3be5a52b MS |
1431 | /* |
1432 | * If fi->writectr is positive (no truncate or fsync going on) send | |
1433 | * all queued writepage requests. | |
1434 | * | |
1435 | * Called with fc->lock | |
1436 | */ | |
1437 | void fuse_flush_writepages(struct inode *inode) | |
b9ca67b2 MS |
1438 | __releases(fc->lock) |
1439 | __acquires(fc->lock) | |
b6aeaded | 1440 | { |
3be5a52b MS |
1441 | struct fuse_conn *fc = get_fuse_conn(inode); |
1442 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1443 | struct fuse_req *req; | |
1444 | ||
1445 | while (fi->writectr >= 0 && !list_empty(&fi->queued_writes)) { | |
1446 | req = list_entry(fi->queued_writes.next, struct fuse_req, list); | |
1447 | list_del_init(&req->list); | |
1448 | fuse_send_writepage(fc, req); | |
1449 | } | |
1450 | } | |
1451 | ||
1452 | static void fuse_writepage_end(struct fuse_conn *fc, struct fuse_req *req) | |
1453 | { | |
1454 | struct inode *inode = req->inode; | |
1455 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1456 | ||
1457 | mapping_set_error(inode->i_mapping, req->out.h.error); | |
1458 | spin_lock(&fc->lock); | |
1459 | fi->writectr--; | |
1460 | fuse_writepage_finish(fc, req); | |
1461 | spin_unlock(&fc->lock); | |
1462 | fuse_writepage_free(fc, req); | |
1463 | } | |
1464 | ||
1465 | static int fuse_writepage_locked(struct page *page) | |
1466 | { | |
1467 | struct address_space *mapping = page->mapping; | |
1468 | struct inode *inode = mapping->host; | |
1469 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1470 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1471 | struct fuse_req *req; | |
1472 | struct fuse_file *ff; | |
1473 | struct page *tmp_page; | |
1474 | ||
1475 | set_page_writeback(page); | |
1476 | ||
4250c066 | 1477 | req = fuse_request_alloc_nofs(1); |
3be5a52b MS |
1478 | if (!req) |
1479 | goto err; | |
1480 | ||
8b41e671 | 1481 | req->background = 1; /* writeback always goes to bg_queue */ |
3be5a52b MS |
1482 | tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); |
1483 | if (!tmp_page) | |
1484 | goto err_free; | |
1485 | ||
1486 | spin_lock(&fc->lock); | |
1487 | BUG_ON(list_empty(&fi->write_files)); | |
1488 | ff = list_entry(fi->write_files.next, struct fuse_file, write_entry); | |
1489 | req->ff = fuse_file_get(ff); | |
1490 | spin_unlock(&fc->lock); | |
1491 | ||
2106cb18 | 1492 | fuse_write_fill(req, ff, page_offset(page), 0); |
3be5a52b MS |
1493 | |
1494 | copy_highpage(tmp_page, page); | |
2d698b07 | 1495 | req->misc.write.in.write_flags |= FUSE_WRITE_CACHE; |
f4975c67 | 1496 | req->in.argpages = 1; |
3be5a52b MS |
1497 | req->num_pages = 1; |
1498 | req->pages[0] = tmp_page; | |
b2430d75 | 1499 | req->page_descs[0].offset = 0; |
85f40aec | 1500 | req->page_descs[0].length = PAGE_SIZE; |
3be5a52b MS |
1501 | req->end = fuse_writepage_end; |
1502 | req->inode = inode; | |
1503 | ||
1504 | inc_bdi_stat(mapping->backing_dev_info, BDI_WRITEBACK); | |
1505 | inc_zone_page_state(tmp_page, NR_WRITEBACK_TEMP); | |
1506 | end_page_writeback(page); | |
1507 | ||
1508 | spin_lock(&fc->lock); | |
1509 | list_add(&req->writepages_entry, &fi->writepages); | |
1510 | list_add_tail(&req->list, &fi->queued_writes); | |
1511 | fuse_flush_writepages(inode); | |
1512 | spin_unlock(&fc->lock); | |
1513 | ||
1514 | return 0; | |
1515 | ||
1516 | err_free: | |
1517 | fuse_request_free(req); | |
1518 | err: | |
1519 | end_page_writeback(page); | |
1520 | return -ENOMEM; | |
1521 | } | |
1522 | ||
1523 | static int fuse_writepage(struct page *page, struct writeback_control *wbc) | |
1524 | { | |
1525 | int err; | |
1526 | ||
1527 | err = fuse_writepage_locked(page); | |
1528 | unlock_page(page); | |
1529 | ||
1530 | return err; | |
1531 | } | |
1532 | ||
1533 | static int fuse_launder_page(struct page *page) | |
1534 | { | |
1535 | int err = 0; | |
1536 | if (clear_page_dirty_for_io(page)) { | |
1537 | struct inode *inode = page->mapping->host; | |
1538 | err = fuse_writepage_locked(page); | |
1539 | if (!err) | |
1540 | fuse_wait_on_page_writeback(inode, page->index); | |
1541 | } | |
1542 | return err; | |
1543 | } | |
1544 | ||
1545 | /* | |
1546 | * Write back dirty pages now, because there may not be any suitable | |
1547 | * open files later | |
1548 | */ | |
1549 | static void fuse_vma_close(struct vm_area_struct *vma) | |
1550 | { | |
1551 | filemap_write_and_wait(vma->vm_file->f_mapping); | |
1552 | } | |
1553 | ||
1554 | /* | |
1555 | * Wait for writeback against this page to complete before allowing it | |
1556 | * to be marked dirty again, and hence written back again, possibly | |
1557 | * before the previous writepage completed. | |
1558 | * | |
1559 | * Block here, instead of in ->writepage(), so that the userspace fs | |
1560 | * can only block processes actually operating on the filesystem. | |
1561 | * | |
1562 | * Otherwise unprivileged userspace fs would be able to block | |
1563 | * unrelated: | |
1564 | * | |
1565 | * - page migration | |
1566 | * - sync(2) | |
1567 | * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER | |
1568 | */ | |
c2ec175c | 1569 | static int fuse_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
3be5a52b | 1570 | { |
c2ec175c | 1571 | struct page *page = vmf->page; |
3be5a52b MS |
1572 | /* |
1573 | * Don't use page->mapping as it may become NULL from a | |
1574 | * concurrent truncate. | |
1575 | */ | |
1576 | struct inode *inode = vma->vm_file->f_mapping->host; | |
1577 | ||
1578 | fuse_wait_on_page_writeback(inode, page->index); | |
1579 | return 0; | |
1580 | } | |
1581 | ||
f0f37e2f | 1582 | static const struct vm_operations_struct fuse_file_vm_ops = { |
3be5a52b MS |
1583 | .close = fuse_vma_close, |
1584 | .fault = filemap_fault, | |
1585 | .page_mkwrite = fuse_page_mkwrite, | |
0b173bc4 | 1586 | .remap_pages = generic_file_remap_pages, |
3be5a52b MS |
1587 | }; |
1588 | ||
1589 | static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma) | |
1590 | { | |
1591 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) { | |
6131ffaa | 1592 | struct inode *inode = file_inode(file); |
3be5a52b MS |
1593 | struct fuse_conn *fc = get_fuse_conn(inode); |
1594 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1595 | struct fuse_file *ff = file->private_data; | |
1596 | /* | |
1597 | * file may be written through mmap, so chain it onto the | |
1598 | * inodes's write_file list | |
1599 | */ | |
1600 | spin_lock(&fc->lock); | |
1601 | if (list_empty(&ff->write_entry)) | |
1602 | list_add(&ff->write_entry, &fi->write_files); | |
1603 | spin_unlock(&fc->lock); | |
1604 | } | |
1605 | file_accessed(file); | |
1606 | vma->vm_ops = &fuse_file_vm_ops; | |
b6aeaded MS |
1607 | return 0; |
1608 | } | |
1609 | ||
fc280c96 MS |
1610 | static int fuse_direct_mmap(struct file *file, struct vm_area_struct *vma) |
1611 | { | |
1612 | /* Can't provide the coherency needed for MAP_SHARED */ | |
1613 | if (vma->vm_flags & VM_MAYSHARE) | |
1614 | return -ENODEV; | |
1615 | ||
3121bfe7 MS |
1616 | invalidate_inode_pages2(file->f_mapping); |
1617 | ||
fc280c96 MS |
1618 | return generic_file_mmap(file, vma); |
1619 | } | |
1620 | ||
71421259 MS |
1621 | static int convert_fuse_file_lock(const struct fuse_file_lock *ffl, |
1622 | struct file_lock *fl) | |
1623 | { | |
1624 | switch (ffl->type) { | |
1625 | case F_UNLCK: | |
1626 | break; | |
1627 | ||
1628 | case F_RDLCK: | |
1629 | case F_WRLCK: | |
1630 | if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX || | |
1631 | ffl->end < ffl->start) | |
1632 | return -EIO; | |
1633 | ||
1634 | fl->fl_start = ffl->start; | |
1635 | fl->fl_end = ffl->end; | |
1636 | fl->fl_pid = ffl->pid; | |
1637 | break; | |
1638 | ||
1639 | default: | |
1640 | return -EIO; | |
1641 | } | |
1642 | fl->fl_type = ffl->type; | |
1643 | return 0; | |
1644 | } | |
1645 | ||
1646 | static void fuse_lk_fill(struct fuse_req *req, struct file *file, | |
a9ff4f87 MS |
1647 | const struct file_lock *fl, int opcode, pid_t pid, |
1648 | int flock) | |
71421259 | 1649 | { |
6131ffaa | 1650 | struct inode *inode = file_inode(file); |
9c8ef561 | 1651 | struct fuse_conn *fc = get_fuse_conn(inode); |
71421259 MS |
1652 | struct fuse_file *ff = file->private_data; |
1653 | struct fuse_lk_in *arg = &req->misc.lk_in; | |
1654 | ||
1655 | arg->fh = ff->fh; | |
9c8ef561 | 1656 | arg->owner = fuse_lock_owner_id(fc, fl->fl_owner); |
71421259 MS |
1657 | arg->lk.start = fl->fl_start; |
1658 | arg->lk.end = fl->fl_end; | |
1659 | arg->lk.type = fl->fl_type; | |
1660 | arg->lk.pid = pid; | |
a9ff4f87 MS |
1661 | if (flock) |
1662 | arg->lk_flags |= FUSE_LK_FLOCK; | |
71421259 MS |
1663 | req->in.h.opcode = opcode; |
1664 | req->in.h.nodeid = get_node_id(inode); | |
1665 | req->in.numargs = 1; | |
1666 | req->in.args[0].size = sizeof(*arg); | |
1667 | req->in.args[0].value = arg; | |
1668 | } | |
1669 | ||
1670 | static int fuse_getlk(struct file *file, struct file_lock *fl) | |
1671 | { | |
6131ffaa | 1672 | struct inode *inode = file_inode(file); |
71421259 MS |
1673 | struct fuse_conn *fc = get_fuse_conn(inode); |
1674 | struct fuse_req *req; | |
1675 | struct fuse_lk_out outarg; | |
1676 | int err; | |
1677 | ||
b111c8c0 | 1678 | req = fuse_get_req_nopages(fc); |
71421259 MS |
1679 | if (IS_ERR(req)) |
1680 | return PTR_ERR(req); | |
1681 | ||
a9ff4f87 | 1682 | fuse_lk_fill(req, file, fl, FUSE_GETLK, 0, 0); |
71421259 MS |
1683 | req->out.numargs = 1; |
1684 | req->out.args[0].size = sizeof(outarg); | |
1685 | req->out.args[0].value = &outarg; | |
b93f858a | 1686 | fuse_request_send(fc, req); |
71421259 MS |
1687 | err = req->out.h.error; |
1688 | fuse_put_request(fc, req); | |
1689 | if (!err) | |
1690 | err = convert_fuse_file_lock(&outarg.lk, fl); | |
1691 | ||
1692 | return err; | |
1693 | } | |
1694 | ||
a9ff4f87 | 1695 | static int fuse_setlk(struct file *file, struct file_lock *fl, int flock) |
71421259 | 1696 | { |
6131ffaa | 1697 | struct inode *inode = file_inode(file); |
71421259 MS |
1698 | struct fuse_conn *fc = get_fuse_conn(inode); |
1699 | struct fuse_req *req; | |
1700 | int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK; | |
1701 | pid_t pid = fl->fl_type != F_UNLCK ? current->tgid : 0; | |
1702 | int err; | |
1703 | ||
8fb47a4f | 1704 | if (fl->fl_lmops && fl->fl_lmops->lm_grant) { |
48e90761 MS |
1705 | /* NLM needs asynchronous locks, which we don't support yet */ |
1706 | return -ENOLCK; | |
1707 | } | |
1708 | ||
71421259 MS |
1709 | /* Unlock on close is handled by the flush method */ |
1710 | if (fl->fl_flags & FL_CLOSE) | |
1711 | return 0; | |
1712 | ||
b111c8c0 | 1713 | req = fuse_get_req_nopages(fc); |
71421259 MS |
1714 | if (IS_ERR(req)) |
1715 | return PTR_ERR(req); | |
1716 | ||
a9ff4f87 | 1717 | fuse_lk_fill(req, file, fl, opcode, pid, flock); |
b93f858a | 1718 | fuse_request_send(fc, req); |
71421259 | 1719 | err = req->out.h.error; |
a4d27e75 MS |
1720 | /* locking is restartable */ |
1721 | if (err == -EINTR) | |
1722 | err = -ERESTARTSYS; | |
71421259 MS |
1723 | fuse_put_request(fc, req); |
1724 | return err; | |
1725 | } | |
1726 | ||
1727 | static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl) | |
1728 | { | |
6131ffaa | 1729 | struct inode *inode = file_inode(file); |
71421259 MS |
1730 | struct fuse_conn *fc = get_fuse_conn(inode); |
1731 | int err; | |
1732 | ||
48e90761 MS |
1733 | if (cmd == F_CANCELLK) { |
1734 | err = 0; | |
1735 | } else if (cmd == F_GETLK) { | |
71421259 | 1736 | if (fc->no_lock) { |
9d6a8c5c | 1737 | posix_test_lock(file, fl); |
71421259 MS |
1738 | err = 0; |
1739 | } else | |
1740 | err = fuse_getlk(file, fl); | |
1741 | } else { | |
1742 | if (fc->no_lock) | |
48e90761 | 1743 | err = posix_lock_file(file, fl, NULL); |
71421259 | 1744 | else |
a9ff4f87 | 1745 | err = fuse_setlk(file, fl, 0); |
71421259 MS |
1746 | } |
1747 | return err; | |
1748 | } | |
1749 | ||
a9ff4f87 MS |
1750 | static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl) |
1751 | { | |
6131ffaa | 1752 | struct inode *inode = file_inode(file); |
a9ff4f87 MS |
1753 | struct fuse_conn *fc = get_fuse_conn(inode); |
1754 | int err; | |
1755 | ||
37fb3a30 | 1756 | if (fc->no_flock) { |
a9ff4f87 MS |
1757 | err = flock_lock_file_wait(file, fl); |
1758 | } else { | |
37fb3a30 MS |
1759 | struct fuse_file *ff = file->private_data; |
1760 | ||
a9ff4f87 MS |
1761 | /* emulate flock with POSIX locks */ |
1762 | fl->fl_owner = (fl_owner_t) file; | |
37fb3a30 | 1763 | ff->flock = true; |
a9ff4f87 MS |
1764 | err = fuse_setlk(file, fl, 1); |
1765 | } | |
1766 | ||
1767 | return err; | |
1768 | } | |
1769 | ||
b2d2272f MS |
1770 | static sector_t fuse_bmap(struct address_space *mapping, sector_t block) |
1771 | { | |
1772 | struct inode *inode = mapping->host; | |
1773 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1774 | struct fuse_req *req; | |
1775 | struct fuse_bmap_in inarg; | |
1776 | struct fuse_bmap_out outarg; | |
1777 | int err; | |
1778 | ||
1779 | if (!inode->i_sb->s_bdev || fc->no_bmap) | |
1780 | return 0; | |
1781 | ||
b111c8c0 | 1782 | req = fuse_get_req_nopages(fc); |
b2d2272f MS |
1783 | if (IS_ERR(req)) |
1784 | return 0; | |
1785 | ||
1786 | memset(&inarg, 0, sizeof(inarg)); | |
1787 | inarg.block = block; | |
1788 | inarg.blocksize = inode->i_sb->s_blocksize; | |
1789 | req->in.h.opcode = FUSE_BMAP; | |
1790 | req->in.h.nodeid = get_node_id(inode); | |
1791 | req->in.numargs = 1; | |
1792 | req->in.args[0].size = sizeof(inarg); | |
1793 | req->in.args[0].value = &inarg; | |
1794 | req->out.numargs = 1; | |
1795 | req->out.args[0].size = sizeof(outarg); | |
1796 | req->out.args[0].value = &outarg; | |
b93f858a | 1797 | fuse_request_send(fc, req); |
b2d2272f MS |
1798 | err = req->out.h.error; |
1799 | fuse_put_request(fc, req); | |
1800 | if (err == -ENOSYS) | |
1801 | fc->no_bmap = 1; | |
1802 | ||
1803 | return err ? 0 : outarg.block; | |
1804 | } | |
1805 | ||
965c8e59 | 1806 | static loff_t fuse_file_llseek(struct file *file, loff_t offset, int whence) |
5559b8f4 MS |
1807 | { |
1808 | loff_t retval; | |
6131ffaa | 1809 | struct inode *inode = file_inode(file); |
5559b8f4 | 1810 | |
c07c3d19 | 1811 | /* No i_mutex protection necessary for SEEK_CUR and SEEK_SET */ |
965c8e59 AM |
1812 | if (whence == SEEK_CUR || whence == SEEK_SET) |
1813 | return generic_file_llseek(file, offset, whence); | |
06222e49 | 1814 | |
c07c3d19 MS |
1815 | mutex_lock(&inode->i_mutex); |
1816 | retval = fuse_update_attributes(inode, NULL, file, NULL); | |
1817 | if (!retval) | |
965c8e59 | 1818 | retval = generic_file_llseek(file, offset, whence); |
5559b8f4 | 1819 | mutex_unlock(&inode->i_mutex); |
c07c3d19 | 1820 | |
5559b8f4 MS |
1821 | return retval; |
1822 | } | |
1823 | ||
59efec7b TH |
1824 | static int fuse_ioctl_copy_user(struct page **pages, struct iovec *iov, |
1825 | unsigned int nr_segs, size_t bytes, bool to_user) | |
1826 | { | |
1827 | struct iov_iter ii; | |
1828 | int page_idx = 0; | |
1829 | ||
1830 | if (!bytes) | |
1831 | return 0; | |
1832 | ||
1833 | iov_iter_init(&ii, iov, nr_segs, bytes, 0); | |
1834 | ||
1835 | while (iov_iter_count(&ii)) { | |
1836 | struct page *page = pages[page_idx++]; | |
1837 | size_t todo = min_t(size_t, PAGE_SIZE, iov_iter_count(&ii)); | |
4aa0edd2 | 1838 | void *kaddr; |
59efec7b | 1839 | |
4aa0edd2 | 1840 | kaddr = kmap(page); |
59efec7b TH |
1841 | |
1842 | while (todo) { | |
1843 | char __user *uaddr = ii.iov->iov_base + ii.iov_offset; | |
1844 | size_t iov_len = ii.iov->iov_len - ii.iov_offset; | |
1845 | size_t copy = min(todo, iov_len); | |
1846 | size_t left; | |
1847 | ||
1848 | if (!to_user) | |
1849 | left = copy_from_user(kaddr, uaddr, copy); | |
1850 | else | |
1851 | left = copy_to_user(uaddr, kaddr, copy); | |
1852 | ||
1853 | if (unlikely(left)) | |
1854 | return -EFAULT; | |
1855 | ||
1856 | iov_iter_advance(&ii, copy); | |
1857 | todo -= copy; | |
1858 | kaddr += copy; | |
1859 | } | |
1860 | ||
0bd87182 | 1861 | kunmap(page); |
59efec7b TH |
1862 | } |
1863 | ||
1864 | return 0; | |
1865 | } | |
1866 | ||
d9d318d3 MS |
1867 | /* |
1868 | * CUSE servers compiled on 32bit broke on 64bit kernels because the | |
1869 | * ABI was defined to be 'struct iovec' which is different on 32bit | |
1870 | * and 64bit. Fortunately we can determine which structure the server | |
1871 | * used from the size of the reply. | |
1872 | */ | |
1baa26b2 MS |
1873 | static int fuse_copy_ioctl_iovec_old(struct iovec *dst, void *src, |
1874 | size_t transferred, unsigned count, | |
1875 | bool is_compat) | |
d9d318d3 MS |
1876 | { |
1877 | #ifdef CONFIG_COMPAT | |
1878 | if (count * sizeof(struct compat_iovec) == transferred) { | |
1879 | struct compat_iovec *ciov = src; | |
1880 | unsigned i; | |
1881 | ||
1882 | /* | |
1883 | * With this interface a 32bit server cannot support | |
1884 | * non-compat (i.e. ones coming from 64bit apps) ioctl | |
1885 | * requests | |
1886 | */ | |
1887 | if (!is_compat) | |
1888 | return -EINVAL; | |
1889 | ||
1890 | for (i = 0; i < count; i++) { | |
1891 | dst[i].iov_base = compat_ptr(ciov[i].iov_base); | |
1892 | dst[i].iov_len = ciov[i].iov_len; | |
1893 | } | |
1894 | return 0; | |
1895 | } | |
1896 | #endif | |
1897 | ||
1898 | if (count * sizeof(struct iovec) != transferred) | |
1899 | return -EIO; | |
1900 | ||
1901 | memcpy(dst, src, transferred); | |
1902 | return 0; | |
1903 | } | |
1904 | ||
7572777e MS |
1905 | /* Make sure iov_length() won't overflow */ |
1906 | static int fuse_verify_ioctl_iov(struct iovec *iov, size_t count) | |
1907 | { | |
1908 | size_t n; | |
1909 | u32 max = FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT; | |
1910 | ||
fb6ccff6 | 1911 | for (n = 0; n < count; n++, iov++) { |
7572777e MS |
1912 | if (iov->iov_len > (size_t) max) |
1913 | return -ENOMEM; | |
1914 | max -= iov->iov_len; | |
1915 | } | |
1916 | return 0; | |
1917 | } | |
1918 | ||
1baa26b2 MS |
1919 | static int fuse_copy_ioctl_iovec(struct fuse_conn *fc, struct iovec *dst, |
1920 | void *src, size_t transferred, unsigned count, | |
1921 | bool is_compat) | |
1922 | { | |
1923 | unsigned i; | |
1924 | struct fuse_ioctl_iovec *fiov = src; | |
1925 | ||
1926 | if (fc->minor < 16) { | |
1927 | return fuse_copy_ioctl_iovec_old(dst, src, transferred, | |
1928 | count, is_compat); | |
1929 | } | |
1930 | ||
1931 | if (count * sizeof(struct fuse_ioctl_iovec) != transferred) | |
1932 | return -EIO; | |
1933 | ||
1934 | for (i = 0; i < count; i++) { | |
1935 | /* Did the server supply an inappropriate value? */ | |
1936 | if (fiov[i].base != (unsigned long) fiov[i].base || | |
1937 | fiov[i].len != (unsigned long) fiov[i].len) | |
1938 | return -EIO; | |
1939 | ||
1940 | dst[i].iov_base = (void __user *) (unsigned long) fiov[i].base; | |
1941 | dst[i].iov_len = (size_t) fiov[i].len; | |
1942 | ||
1943 | #ifdef CONFIG_COMPAT | |
1944 | if (is_compat && | |
1945 | (ptr_to_compat(dst[i].iov_base) != fiov[i].base || | |
1946 | (compat_size_t) dst[i].iov_len != fiov[i].len)) | |
1947 | return -EIO; | |
1948 | #endif | |
1949 | } | |
1950 | ||
1951 | return 0; | |
1952 | } | |
1953 | ||
1954 | ||
59efec7b TH |
1955 | /* |
1956 | * For ioctls, there is no generic way to determine how much memory | |
1957 | * needs to be read and/or written. Furthermore, ioctls are allowed | |
1958 | * to dereference the passed pointer, so the parameter requires deep | |
1959 | * copying but FUSE has no idea whatsoever about what to copy in or | |
1960 | * out. | |
1961 | * | |
1962 | * This is solved by allowing FUSE server to retry ioctl with | |
1963 | * necessary in/out iovecs. Let's assume the ioctl implementation | |
1964 | * needs to read in the following structure. | |
1965 | * | |
1966 | * struct a { | |
1967 | * char *buf; | |
1968 | * size_t buflen; | |
1969 | * } | |
1970 | * | |
1971 | * On the first callout to FUSE server, inarg->in_size and | |
1972 | * inarg->out_size will be NULL; then, the server completes the ioctl | |
1973 | * with FUSE_IOCTL_RETRY set in out->flags, out->in_iovs set to 1 and | |
1974 | * the actual iov array to | |
1975 | * | |
1976 | * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) } } | |
1977 | * | |
1978 | * which tells FUSE to copy in the requested area and retry the ioctl. | |
1979 | * On the second round, the server has access to the structure and | |
1980 | * from that it can tell what to look for next, so on the invocation, | |
1981 | * it sets FUSE_IOCTL_RETRY, out->in_iovs to 2 and iov array to | |
1982 | * | |
1983 | * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) }, | |
1984 | * { .iov_base = a.buf, .iov_len = a.buflen } } | |
1985 | * | |
1986 | * FUSE will copy both struct a and the pointed buffer from the | |
1987 | * process doing the ioctl and retry ioctl with both struct a and the | |
1988 | * buffer. | |
1989 | * | |
1990 | * This time, FUSE server has everything it needs and completes ioctl | |
1991 | * without FUSE_IOCTL_RETRY which finishes the ioctl call. | |
1992 | * | |
1993 | * Copying data out works the same way. | |
1994 | * | |
1995 | * Note that if FUSE_IOCTL_UNRESTRICTED is clear, the kernel | |
1996 | * automatically initializes in and out iovs by decoding @cmd with | |
1997 | * _IOC_* macros and the server is not allowed to request RETRY. This | |
1998 | * limits ioctl data transfers to well-formed ioctls and is the forced | |
1999 | * behavior for all FUSE servers. | |
2000 | */ | |
08cbf542 TH |
2001 | long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg, |
2002 | unsigned int flags) | |
59efec7b | 2003 | { |
59efec7b | 2004 | struct fuse_file *ff = file->private_data; |
d36f2487 | 2005 | struct fuse_conn *fc = ff->fc; |
59efec7b TH |
2006 | struct fuse_ioctl_in inarg = { |
2007 | .fh = ff->fh, | |
2008 | .cmd = cmd, | |
2009 | .arg = arg, | |
2010 | .flags = flags | |
2011 | }; | |
2012 | struct fuse_ioctl_out outarg; | |
2013 | struct fuse_req *req = NULL; | |
2014 | struct page **pages = NULL; | |
8ac83505 | 2015 | struct iovec *iov_page = NULL; |
59efec7b TH |
2016 | struct iovec *in_iov = NULL, *out_iov = NULL; |
2017 | unsigned int in_iovs = 0, out_iovs = 0, num_pages = 0, max_pages; | |
2018 | size_t in_size, out_size, transferred; | |
2019 | int err; | |
2020 | ||
1baa26b2 MS |
2021 | #if BITS_PER_LONG == 32 |
2022 | inarg.flags |= FUSE_IOCTL_32BIT; | |
2023 | #else | |
2024 | if (flags & FUSE_IOCTL_COMPAT) | |
2025 | inarg.flags |= FUSE_IOCTL_32BIT; | |
2026 | #endif | |
2027 | ||
59efec7b | 2028 | /* assume all the iovs returned by client always fits in a page */ |
1baa26b2 | 2029 | BUILD_BUG_ON(sizeof(struct fuse_ioctl_iovec) * FUSE_IOCTL_MAX_IOV > PAGE_SIZE); |
59efec7b | 2030 | |
59efec7b | 2031 | err = -ENOMEM; |
c411cc88 | 2032 | pages = kcalloc(FUSE_MAX_PAGES_PER_REQ, sizeof(pages[0]), GFP_KERNEL); |
8ac83505 | 2033 | iov_page = (struct iovec *) __get_free_page(GFP_KERNEL); |
59efec7b TH |
2034 | if (!pages || !iov_page) |
2035 | goto out; | |
2036 | ||
2037 | /* | |
2038 | * If restricted, initialize IO parameters as encoded in @cmd. | |
2039 | * RETRY from server is not allowed. | |
2040 | */ | |
2041 | if (!(flags & FUSE_IOCTL_UNRESTRICTED)) { | |
8ac83505 | 2042 | struct iovec *iov = iov_page; |
59efec7b | 2043 | |
c9f0523d | 2044 | iov->iov_base = (void __user *)arg; |
59efec7b TH |
2045 | iov->iov_len = _IOC_SIZE(cmd); |
2046 | ||
2047 | if (_IOC_DIR(cmd) & _IOC_WRITE) { | |
2048 | in_iov = iov; | |
2049 | in_iovs = 1; | |
2050 | } | |
2051 | ||
2052 | if (_IOC_DIR(cmd) & _IOC_READ) { | |
2053 | out_iov = iov; | |
2054 | out_iovs = 1; | |
2055 | } | |
2056 | } | |
2057 | ||
2058 | retry: | |
2059 | inarg.in_size = in_size = iov_length(in_iov, in_iovs); | |
2060 | inarg.out_size = out_size = iov_length(out_iov, out_iovs); | |
2061 | ||
2062 | /* | |
2063 | * Out data can be used either for actual out data or iovs, | |
2064 | * make sure there always is at least one page. | |
2065 | */ | |
2066 | out_size = max_t(size_t, out_size, PAGE_SIZE); | |
2067 | max_pages = DIV_ROUND_UP(max(in_size, out_size), PAGE_SIZE); | |
2068 | ||
2069 | /* make sure there are enough buffer pages and init request with them */ | |
2070 | err = -ENOMEM; | |
2071 | if (max_pages > FUSE_MAX_PAGES_PER_REQ) | |
2072 | goto out; | |
2073 | while (num_pages < max_pages) { | |
2074 | pages[num_pages] = alloc_page(GFP_KERNEL | __GFP_HIGHMEM); | |
2075 | if (!pages[num_pages]) | |
2076 | goto out; | |
2077 | num_pages++; | |
2078 | } | |
2079 | ||
54b96670 | 2080 | req = fuse_get_req(fc, num_pages); |
59efec7b TH |
2081 | if (IS_ERR(req)) { |
2082 | err = PTR_ERR(req); | |
2083 | req = NULL; | |
2084 | goto out; | |
2085 | } | |
2086 | memcpy(req->pages, pages, sizeof(req->pages[0]) * num_pages); | |
2087 | req->num_pages = num_pages; | |
7c190c8b | 2088 | fuse_page_descs_length_init(req, 0, req->num_pages); |
59efec7b TH |
2089 | |
2090 | /* okay, let's send it to the client */ | |
2091 | req->in.h.opcode = FUSE_IOCTL; | |
d36f2487 | 2092 | req->in.h.nodeid = ff->nodeid; |
59efec7b TH |
2093 | req->in.numargs = 1; |
2094 | req->in.args[0].size = sizeof(inarg); | |
2095 | req->in.args[0].value = &inarg; | |
2096 | if (in_size) { | |
2097 | req->in.numargs++; | |
2098 | req->in.args[1].size = in_size; | |
2099 | req->in.argpages = 1; | |
2100 | ||
2101 | err = fuse_ioctl_copy_user(pages, in_iov, in_iovs, in_size, | |
2102 | false); | |
2103 | if (err) | |
2104 | goto out; | |
2105 | } | |
2106 | ||
2107 | req->out.numargs = 2; | |
2108 | req->out.args[0].size = sizeof(outarg); | |
2109 | req->out.args[0].value = &outarg; | |
2110 | req->out.args[1].size = out_size; | |
2111 | req->out.argpages = 1; | |
2112 | req->out.argvar = 1; | |
2113 | ||
b93f858a | 2114 | fuse_request_send(fc, req); |
59efec7b TH |
2115 | err = req->out.h.error; |
2116 | transferred = req->out.args[1].size; | |
2117 | fuse_put_request(fc, req); | |
2118 | req = NULL; | |
2119 | if (err) | |
2120 | goto out; | |
2121 | ||
2122 | /* did it ask for retry? */ | |
2123 | if (outarg.flags & FUSE_IOCTL_RETRY) { | |
8ac83505 | 2124 | void *vaddr; |
59efec7b TH |
2125 | |
2126 | /* no retry if in restricted mode */ | |
2127 | err = -EIO; | |
2128 | if (!(flags & FUSE_IOCTL_UNRESTRICTED)) | |
2129 | goto out; | |
2130 | ||
2131 | in_iovs = outarg.in_iovs; | |
2132 | out_iovs = outarg.out_iovs; | |
2133 | ||
2134 | /* | |
2135 | * Make sure things are in boundary, separate checks | |
2136 | * are to protect against overflow. | |
2137 | */ | |
2138 | err = -ENOMEM; | |
2139 | if (in_iovs > FUSE_IOCTL_MAX_IOV || | |
2140 | out_iovs > FUSE_IOCTL_MAX_IOV || | |
2141 | in_iovs + out_iovs > FUSE_IOCTL_MAX_IOV) | |
2142 | goto out; | |
2143 | ||
2408f6ef | 2144 | vaddr = kmap_atomic(pages[0]); |
1baa26b2 | 2145 | err = fuse_copy_ioctl_iovec(fc, iov_page, vaddr, |
d9d318d3 MS |
2146 | transferred, in_iovs + out_iovs, |
2147 | (flags & FUSE_IOCTL_COMPAT) != 0); | |
2408f6ef | 2148 | kunmap_atomic(vaddr); |
d9d318d3 MS |
2149 | if (err) |
2150 | goto out; | |
59efec7b | 2151 | |
8ac83505 | 2152 | in_iov = iov_page; |
59efec7b TH |
2153 | out_iov = in_iov + in_iovs; |
2154 | ||
7572777e MS |
2155 | err = fuse_verify_ioctl_iov(in_iov, in_iovs); |
2156 | if (err) | |
2157 | goto out; | |
2158 | ||
2159 | err = fuse_verify_ioctl_iov(out_iov, out_iovs); | |
2160 | if (err) | |
2161 | goto out; | |
2162 | ||
59efec7b TH |
2163 | goto retry; |
2164 | } | |
2165 | ||
2166 | err = -EIO; | |
2167 | if (transferred > inarg.out_size) | |
2168 | goto out; | |
2169 | ||
2170 | err = fuse_ioctl_copy_user(pages, out_iov, out_iovs, transferred, true); | |
2171 | out: | |
2172 | if (req) | |
2173 | fuse_put_request(fc, req); | |
8ac83505 | 2174 | free_page((unsigned long) iov_page); |
59efec7b TH |
2175 | while (num_pages) |
2176 | __free_page(pages[--num_pages]); | |
2177 | kfree(pages); | |
2178 | ||
2179 | return err ? err : outarg.result; | |
2180 | } | |
08cbf542 | 2181 | EXPORT_SYMBOL_GPL(fuse_do_ioctl); |
59efec7b | 2182 | |
b18da0c5 MS |
2183 | long fuse_ioctl_common(struct file *file, unsigned int cmd, |
2184 | unsigned long arg, unsigned int flags) | |
d36f2487 | 2185 | { |
6131ffaa | 2186 | struct inode *inode = file_inode(file); |
d36f2487 MS |
2187 | struct fuse_conn *fc = get_fuse_conn(inode); |
2188 | ||
c2132c1b | 2189 | if (!fuse_allow_current_process(fc)) |
d36f2487 MS |
2190 | return -EACCES; |
2191 | ||
2192 | if (is_bad_inode(inode)) | |
2193 | return -EIO; | |
2194 | ||
2195 | return fuse_do_ioctl(file, cmd, arg, flags); | |
2196 | } | |
2197 | ||
59efec7b TH |
2198 | static long fuse_file_ioctl(struct file *file, unsigned int cmd, |
2199 | unsigned long arg) | |
2200 | { | |
b18da0c5 | 2201 | return fuse_ioctl_common(file, cmd, arg, 0); |
59efec7b TH |
2202 | } |
2203 | ||
2204 | static long fuse_file_compat_ioctl(struct file *file, unsigned int cmd, | |
2205 | unsigned long arg) | |
2206 | { | |
b18da0c5 | 2207 | return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_COMPAT); |
59efec7b TH |
2208 | } |
2209 | ||
95668a69 TH |
2210 | /* |
2211 | * All files which have been polled are linked to RB tree | |
2212 | * fuse_conn->polled_files which is indexed by kh. Walk the tree and | |
2213 | * find the matching one. | |
2214 | */ | |
2215 | static struct rb_node **fuse_find_polled_node(struct fuse_conn *fc, u64 kh, | |
2216 | struct rb_node **parent_out) | |
2217 | { | |
2218 | struct rb_node **link = &fc->polled_files.rb_node; | |
2219 | struct rb_node *last = NULL; | |
2220 | ||
2221 | while (*link) { | |
2222 | struct fuse_file *ff; | |
2223 | ||
2224 | last = *link; | |
2225 | ff = rb_entry(last, struct fuse_file, polled_node); | |
2226 | ||
2227 | if (kh < ff->kh) | |
2228 | link = &last->rb_left; | |
2229 | else if (kh > ff->kh) | |
2230 | link = &last->rb_right; | |
2231 | else | |
2232 | return link; | |
2233 | } | |
2234 | ||
2235 | if (parent_out) | |
2236 | *parent_out = last; | |
2237 | return link; | |
2238 | } | |
2239 | ||
2240 | /* | |
2241 | * The file is about to be polled. Make sure it's on the polled_files | |
2242 | * RB tree. Note that files once added to the polled_files tree are | |
2243 | * not removed before the file is released. This is because a file | |
2244 | * polled once is likely to be polled again. | |
2245 | */ | |
2246 | static void fuse_register_polled_file(struct fuse_conn *fc, | |
2247 | struct fuse_file *ff) | |
2248 | { | |
2249 | spin_lock(&fc->lock); | |
2250 | if (RB_EMPTY_NODE(&ff->polled_node)) { | |
2251 | struct rb_node **link, *parent; | |
2252 | ||
2253 | link = fuse_find_polled_node(fc, ff->kh, &parent); | |
2254 | BUG_ON(*link); | |
2255 | rb_link_node(&ff->polled_node, parent, link); | |
2256 | rb_insert_color(&ff->polled_node, &fc->polled_files); | |
2257 | } | |
2258 | spin_unlock(&fc->lock); | |
2259 | } | |
2260 | ||
08cbf542 | 2261 | unsigned fuse_file_poll(struct file *file, poll_table *wait) |
95668a69 | 2262 | { |
95668a69 | 2263 | struct fuse_file *ff = file->private_data; |
797759aa | 2264 | struct fuse_conn *fc = ff->fc; |
95668a69 TH |
2265 | struct fuse_poll_in inarg = { .fh = ff->fh, .kh = ff->kh }; |
2266 | struct fuse_poll_out outarg; | |
2267 | struct fuse_req *req; | |
2268 | int err; | |
2269 | ||
2270 | if (fc->no_poll) | |
2271 | return DEFAULT_POLLMASK; | |
2272 | ||
2273 | poll_wait(file, &ff->poll_wait, wait); | |
0415d291 | 2274 | inarg.events = (__u32)poll_requested_events(wait); |
95668a69 TH |
2275 | |
2276 | /* | |
2277 | * Ask for notification iff there's someone waiting for it. | |
2278 | * The client may ignore the flag and always notify. | |
2279 | */ | |
2280 | if (waitqueue_active(&ff->poll_wait)) { | |
2281 | inarg.flags |= FUSE_POLL_SCHEDULE_NOTIFY; | |
2282 | fuse_register_polled_file(fc, ff); | |
2283 | } | |
2284 | ||
b111c8c0 | 2285 | req = fuse_get_req_nopages(fc); |
95668a69 | 2286 | if (IS_ERR(req)) |
201fa69a | 2287 | return POLLERR; |
95668a69 TH |
2288 | |
2289 | req->in.h.opcode = FUSE_POLL; | |
797759aa | 2290 | req->in.h.nodeid = ff->nodeid; |
95668a69 TH |
2291 | req->in.numargs = 1; |
2292 | req->in.args[0].size = sizeof(inarg); | |
2293 | req->in.args[0].value = &inarg; | |
2294 | req->out.numargs = 1; | |
2295 | req->out.args[0].size = sizeof(outarg); | |
2296 | req->out.args[0].value = &outarg; | |
b93f858a | 2297 | fuse_request_send(fc, req); |
95668a69 TH |
2298 | err = req->out.h.error; |
2299 | fuse_put_request(fc, req); | |
2300 | ||
2301 | if (!err) | |
2302 | return outarg.revents; | |
2303 | if (err == -ENOSYS) { | |
2304 | fc->no_poll = 1; | |
2305 | return DEFAULT_POLLMASK; | |
2306 | } | |
2307 | return POLLERR; | |
2308 | } | |
08cbf542 | 2309 | EXPORT_SYMBOL_GPL(fuse_file_poll); |
95668a69 TH |
2310 | |
2311 | /* | |
2312 | * This is called from fuse_handle_notify() on FUSE_NOTIFY_POLL and | |
2313 | * wakes up the poll waiters. | |
2314 | */ | |
2315 | int fuse_notify_poll_wakeup(struct fuse_conn *fc, | |
2316 | struct fuse_notify_poll_wakeup_out *outarg) | |
2317 | { | |
2318 | u64 kh = outarg->kh; | |
2319 | struct rb_node **link; | |
2320 | ||
2321 | spin_lock(&fc->lock); | |
2322 | ||
2323 | link = fuse_find_polled_node(fc, kh, NULL); | |
2324 | if (*link) { | |
2325 | struct fuse_file *ff; | |
2326 | ||
2327 | ff = rb_entry(*link, struct fuse_file, polled_node); | |
2328 | wake_up_interruptible_sync(&ff->poll_wait); | |
2329 | } | |
2330 | ||
2331 | spin_unlock(&fc->lock); | |
2332 | return 0; | |
2333 | } | |
2334 | ||
4273b793 AA |
2335 | static ssize_t |
2336 | fuse_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, | |
2337 | loff_t offset, unsigned long nr_segs) | |
2338 | { | |
2339 | ssize_t ret = 0; | |
2340 | struct file *file = NULL; | |
2341 | loff_t pos = 0; | |
2342 | ||
2343 | file = iocb->ki_filp; | |
2344 | pos = offset; | |
2345 | ||
b98d023a MP |
2346 | if (rw == WRITE) |
2347 | ret = __fuse_direct_write(file, iov, nr_segs, &pos); | |
2348 | else | |
2349 | ret = __fuse_direct_read(file, iov, nr_segs, &pos); | |
4273b793 AA |
2350 | |
2351 | return ret; | |
2352 | } | |
2353 | ||
cdadb11c MS |
2354 | static long fuse_file_fallocate(struct file *file, int mode, loff_t offset, |
2355 | loff_t length) | |
05ba1f08 AP |
2356 | { |
2357 | struct fuse_file *ff = file->private_data; | |
2358 | struct fuse_conn *fc = ff->fc; | |
2359 | struct fuse_req *req; | |
2360 | struct fuse_fallocate_in inarg = { | |
2361 | .fh = ff->fh, | |
2362 | .offset = offset, | |
2363 | .length = length, | |
2364 | .mode = mode | |
2365 | }; | |
2366 | int err; | |
2367 | ||
519c6040 MS |
2368 | if (fc->no_fallocate) |
2369 | return -EOPNOTSUPP; | |
2370 | ||
b111c8c0 | 2371 | req = fuse_get_req_nopages(fc); |
05ba1f08 AP |
2372 | if (IS_ERR(req)) |
2373 | return PTR_ERR(req); | |
2374 | ||
2375 | req->in.h.opcode = FUSE_FALLOCATE; | |
2376 | req->in.h.nodeid = ff->nodeid; | |
2377 | req->in.numargs = 1; | |
2378 | req->in.args[0].size = sizeof(inarg); | |
2379 | req->in.args[0].value = &inarg; | |
2380 | fuse_request_send(fc, req); | |
2381 | err = req->out.h.error; | |
519c6040 MS |
2382 | if (err == -ENOSYS) { |
2383 | fc->no_fallocate = 1; | |
2384 | err = -EOPNOTSUPP; | |
2385 | } | |
05ba1f08 AP |
2386 | fuse_put_request(fc, req); |
2387 | ||
2388 | return err; | |
2389 | } | |
05ba1f08 | 2390 | |
4b6f5d20 | 2391 | static const struct file_operations fuse_file_operations = { |
5559b8f4 | 2392 | .llseek = fuse_file_llseek, |
543ade1f | 2393 | .read = do_sync_read, |
bcb4be80 | 2394 | .aio_read = fuse_file_aio_read, |
543ade1f | 2395 | .write = do_sync_write, |
ea9b9907 | 2396 | .aio_write = fuse_file_aio_write, |
b6aeaded MS |
2397 | .mmap = fuse_file_mmap, |
2398 | .open = fuse_open, | |
2399 | .flush = fuse_flush, | |
2400 | .release = fuse_release, | |
2401 | .fsync = fuse_fsync, | |
71421259 | 2402 | .lock = fuse_file_lock, |
a9ff4f87 | 2403 | .flock = fuse_file_flock, |
5ffc4ef4 | 2404 | .splice_read = generic_file_splice_read, |
59efec7b TH |
2405 | .unlocked_ioctl = fuse_file_ioctl, |
2406 | .compat_ioctl = fuse_file_compat_ioctl, | |
95668a69 | 2407 | .poll = fuse_file_poll, |
05ba1f08 | 2408 | .fallocate = fuse_file_fallocate, |
b6aeaded MS |
2409 | }; |
2410 | ||
4b6f5d20 | 2411 | static const struct file_operations fuse_direct_io_file_operations = { |
5559b8f4 | 2412 | .llseek = fuse_file_llseek, |
413ef8cb MS |
2413 | .read = fuse_direct_read, |
2414 | .write = fuse_direct_write, | |
fc280c96 | 2415 | .mmap = fuse_direct_mmap, |
413ef8cb MS |
2416 | .open = fuse_open, |
2417 | .flush = fuse_flush, | |
2418 | .release = fuse_release, | |
2419 | .fsync = fuse_fsync, | |
71421259 | 2420 | .lock = fuse_file_lock, |
a9ff4f87 | 2421 | .flock = fuse_file_flock, |
59efec7b TH |
2422 | .unlocked_ioctl = fuse_file_ioctl, |
2423 | .compat_ioctl = fuse_file_compat_ioctl, | |
95668a69 | 2424 | .poll = fuse_file_poll, |
05ba1f08 | 2425 | .fallocate = fuse_file_fallocate, |
fc280c96 | 2426 | /* no splice_read */ |
413ef8cb MS |
2427 | }; |
2428 | ||
f5e54d6e | 2429 | static const struct address_space_operations fuse_file_aops = { |
b6aeaded | 2430 | .readpage = fuse_readpage, |
3be5a52b MS |
2431 | .writepage = fuse_writepage, |
2432 | .launder_page = fuse_launder_page, | |
db50b96c | 2433 | .readpages = fuse_readpages, |
3be5a52b | 2434 | .set_page_dirty = __set_page_dirty_nobuffers, |
b2d2272f | 2435 | .bmap = fuse_bmap, |
4273b793 | 2436 | .direct_IO = fuse_direct_IO, |
b6aeaded MS |
2437 | }; |
2438 | ||
2439 | void fuse_init_file_inode(struct inode *inode) | |
2440 | { | |
45323fb7 MS |
2441 | inode->i_fop = &fuse_file_operations; |
2442 | inode->i_data.a_ops = &fuse_file_aops; | |
b6aeaded | 2443 | } |