]>
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> |
a27bb332 | 18 | #include <linux/aio.h> |
3634a632 | 19 | #include <linux/falloc.h> |
b6aeaded | 20 | |
4b6f5d20 | 21 | static const struct file_operations fuse_direct_io_file_operations; |
45323fb7 | 22 | |
91fe96b4 MS |
23 | static int fuse_send_open(struct fuse_conn *fc, u64 nodeid, struct file *file, |
24 | int opcode, struct fuse_open_out *outargp) | |
b6aeaded | 25 | { |
b6aeaded | 26 | struct fuse_open_in inarg; |
fd72faac MS |
27 | struct fuse_req *req; |
28 | int err; | |
29 | ||
b111c8c0 | 30 | req = fuse_get_req_nopages(fc); |
ce1d5a49 MS |
31 | if (IS_ERR(req)) |
32 | return PTR_ERR(req); | |
fd72faac MS |
33 | |
34 | memset(&inarg, 0, sizeof(inarg)); | |
6ff958ed MS |
35 | inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY); |
36 | if (!fc->atomic_o_trunc) | |
37 | inarg.flags &= ~O_TRUNC; | |
91fe96b4 MS |
38 | req->in.h.opcode = opcode; |
39 | req->in.h.nodeid = nodeid; | |
fd72faac MS |
40 | req->in.numargs = 1; |
41 | req->in.args[0].size = sizeof(inarg); | |
42 | req->in.args[0].value = &inarg; | |
43 | req->out.numargs = 1; | |
44 | req->out.args[0].size = sizeof(*outargp); | |
45 | req->out.args[0].value = outargp; | |
b93f858a | 46 | fuse_request_send(fc, req); |
fd72faac MS |
47 | err = req->out.h.error; |
48 | fuse_put_request(fc, req); | |
49 | ||
50 | return err; | |
51 | } | |
52 | ||
acf99433 | 53 | struct fuse_file *fuse_file_alloc(struct fuse_conn *fc) |
fd72faac MS |
54 | { |
55 | struct fuse_file *ff; | |
6b2db28a | 56 | |
fd72faac | 57 | ff = kmalloc(sizeof(struct fuse_file), GFP_KERNEL); |
6b2db28a TH |
58 | if (unlikely(!ff)) |
59 | return NULL; | |
60 | ||
da5e4714 | 61 | ff->fc = fc; |
4250c066 | 62 | ff->reserved_req = fuse_request_alloc(0); |
6b2db28a TH |
63 | if (unlikely(!ff->reserved_req)) { |
64 | kfree(ff); | |
65 | return NULL; | |
fd72faac | 66 | } |
6b2db28a TH |
67 | |
68 | INIT_LIST_HEAD(&ff->write_entry); | |
69 | atomic_set(&ff->count, 0); | |
70 | RB_CLEAR_NODE(&ff->polled_node); | |
71 | init_waitqueue_head(&ff->poll_wait); | |
72 | ||
73 | spin_lock(&fc->lock); | |
74 | ff->kh = ++fc->khctr; | |
75 | spin_unlock(&fc->lock); | |
76 | ||
fd72faac MS |
77 | return ff; |
78 | } | |
79 | ||
80 | void fuse_file_free(struct fuse_file *ff) | |
81 | { | |
33649c91 | 82 | fuse_request_free(ff->reserved_req); |
fd72faac MS |
83 | kfree(ff); |
84 | } | |
85 | ||
c7b7143c | 86 | struct fuse_file *fuse_file_get(struct fuse_file *ff) |
c756e0a4 MS |
87 | { |
88 | atomic_inc(&ff->count); | |
89 | return ff; | |
90 | } | |
91 | ||
5a18ec17 MS |
92 | static void fuse_release_async(struct work_struct *work) |
93 | { | |
94 | struct fuse_req *req; | |
95 | struct fuse_conn *fc; | |
96 | struct path path; | |
97 | ||
98 | req = container_of(work, struct fuse_req, misc.release.work); | |
99 | path = req->misc.release.path; | |
100 | fc = get_fuse_conn(path.dentry->d_inode); | |
101 | ||
102 | fuse_put_request(fc, req); | |
103 | path_put(&path); | |
104 | } | |
105 | ||
819c4b3b MS |
106 | static void fuse_release_end(struct fuse_conn *fc, struct fuse_req *req) |
107 | { | |
5a18ec17 MS |
108 | if (fc->destroy_req) { |
109 | /* | |
110 | * If this is a fuseblk mount, then it's possible that | |
111 | * releasing the path will result in releasing the | |
112 | * super block and sending the DESTROY request. If | |
113 | * the server is single threaded, this would hang. | |
114 | * For this reason do the path_put() in a separate | |
115 | * thread. | |
116 | */ | |
117 | atomic_inc(&req->count); | |
118 | INIT_WORK(&req->misc.release.work, fuse_release_async); | |
119 | schedule_work(&req->misc.release.work); | |
120 | } else { | |
121 | path_put(&req->misc.release.path); | |
122 | } | |
819c4b3b MS |
123 | } |
124 | ||
5a18ec17 | 125 | static void fuse_file_put(struct fuse_file *ff, bool sync) |
c756e0a4 MS |
126 | { |
127 | if (atomic_dec_and_test(&ff->count)) { | |
128 | struct fuse_req *req = ff->reserved_req; | |
8b0797a4 | 129 | |
7678ac50 AG |
130 | if (ff->fc->no_open) { |
131 | /* | |
132 | * Drop the release request when client does not | |
133 | * implement 'open' | |
134 | */ | |
135 | req->background = 0; | |
136 | path_put(&req->misc.release.path); | |
137 | fuse_put_request(ff->fc, req); | |
138 | } else if (sync) { | |
8b41e671 | 139 | req->background = 0; |
5a18ec17 MS |
140 | fuse_request_send(ff->fc, req); |
141 | path_put(&req->misc.release.path); | |
142 | fuse_put_request(ff->fc, req); | |
143 | } else { | |
144 | req->end = fuse_release_end; | |
8b41e671 | 145 | req->background = 1; |
5a18ec17 MS |
146 | fuse_request_send_background(ff->fc, req); |
147 | } | |
c756e0a4 MS |
148 | kfree(ff); |
149 | } | |
150 | } | |
151 | ||
08cbf542 TH |
152 | int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file, |
153 | bool isdir) | |
91fe96b4 | 154 | { |
91fe96b4 | 155 | struct fuse_file *ff; |
91fe96b4 MS |
156 | int opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN; |
157 | ||
158 | ff = fuse_file_alloc(fc); | |
159 | if (!ff) | |
160 | return -ENOMEM; | |
161 | ||
7678ac50 AG |
162 | ff->fh = 0; |
163 | ff->open_flags = FOPEN_KEEP_CACHE; /* Default for no-open */ | |
164 | if (!fc->no_open || isdir) { | |
165 | struct fuse_open_out outarg; | |
166 | int err; | |
167 | ||
168 | err = fuse_send_open(fc, nodeid, file, opcode, &outarg); | |
169 | if (!err) { | |
170 | ff->fh = outarg.fh; | |
171 | ff->open_flags = outarg.open_flags; | |
172 | ||
173 | } else if (err != -ENOSYS || isdir) { | |
174 | fuse_file_free(ff); | |
175 | return err; | |
176 | } else { | |
177 | fc->no_open = 1; | |
178 | } | |
91fe96b4 MS |
179 | } |
180 | ||
181 | if (isdir) | |
7678ac50 | 182 | ff->open_flags &= ~FOPEN_DIRECT_IO; |
91fe96b4 | 183 | |
91fe96b4 | 184 | ff->nodeid = nodeid; |
91fe96b4 MS |
185 | file->private_data = fuse_file_get(ff); |
186 | ||
187 | return 0; | |
188 | } | |
08cbf542 | 189 | EXPORT_SYMBOL_GPL(fuse_do_open); |
91fe96b4 | 190 | |
c7b7143c | 191 | void fuse_finish_open(struct inode *inode, struct file *file) |
fd72faac | 192 | { |
c7b7143c | 193 | struct fuse_file *ff = file->private_data; |
a0822c55 | 194 | struct fuse_conn *fc = get_fuse_conn(inode); |
c7b7143c MS |
195 | |
196 | if (ff->open_flags & FOPEN_DIRECT_IO) | |
fd72faac | 197 | file->f_op = &fuse_direct_io_file_operations; |
c7b7143c | 198 | if (!(ff->open_flags & FOPEN_KEEP_CACHE)) |
b1009979 | 199 | invalidate_inode_pages2(inode->i_mapping); |
c7b7143c | 200 | if (ff->open_flags & FOPEN_NONSEEKABLE) |
a7c1b990 | 201 | nonseekable_open(inode, file); |
a0822c55 KS |
202 | if (fc->atomic_o_trunc && (file->f_flags & O_TRUNC)) { |
203 | struct fuse_inode *fi = get_fuse_inode(inode); | |
204 | ||
205 | spin_lock(&fc->lock); | |
206 | fi->attr_version = ++fc->attr_version; | |
207 | i_size_write(inode, 0); | |
208 | spin_unlock(&fc->lock); | |
209 | fuse_invalidate_attr(inode); | |
210 | } | |
fd72faac MS |
211 | } |
212 | ||
91fe96b4 | 213 | int fuse_open_common(struct inode *inode, struct file *file, bool isdir) |
fd72faac | 214 | { |
acf99433 | 215 | struct fuse_conn *fc = get_fuse_conn(inode); |
b6aeaded | 216 | int err; |
b6aeaded MS |
217 | |
218 | err = generic_file_open(inode, file); | |
219 | if (err) | |
220 | return err; | |
221 | ||
91fe96b4 | 222 | err = fuse_do_open(fc, get_node_id(inode), file, isdir); |
fd72faac | 223 | if (err) |
91fe96b4 | 224 | return err; |
b6aeaded | 225 | |
91fe96b4 MS |
226 | fuse_finish_open(inode, file); |
227 | ||
228 | return 0; | |
b6aeaded MS |
229 | } |
230 | ||
8b0797a4 | 231 | static void fuse_prepare_release(struct fuse_file *ff, int flags, int opcode) |
64c6d8ed | 232 | { |
8b0797a4 | 233 | struct fuse_conn *fc = ff->fc; |
33649c91 | 234 | struct fuse_req *req = ff->reserved_req; |
b57d4264 | 235 | struct fuse_release_in *inarg = &req->misc.release.in; |
b6aeaded | 236 | |
8b0797a4 MS |
237 | spin_lock(&fc->lock); |
238 | list_del(&ff->write_entry); | |
239 | if (!RB_EMPTY_NODE(&ff->polled_node)) | |
240 | rb_erase(&ff->polled_node, &fc->polled_files); | |
241 | spin_unlock(&fc->lock); | |
242 | ||
357ccf2b | 243 | wake_up_interruptible_all(&ff->poll_wait); |
8b0797a4 | 244 | |
b6aeaded | 245 | inarg->fh = ff->fh; |
fd72faac | 246 | inarg->flags = flags; |
51eb01e7 | 247 | req->in.h.opcode = opcode; |
c7b7143c | 248 | req->in.h.nodeid = ff->nodeid; |
b6aeaded MS |
249 | req->in.numargs = 1; |
250 | req->in.args[0].size = sizeof(struct fuse_release_in); | |
251 | req->in.args[0].value = inarg; | |
fd72faac MS |
252 | } |
253 | ||
8b0797a4 | 254 | void fuse_release_common(struct file *file, int opcode) |
fd72faac | 255 | { |
6b2db28a TH |
256 | struct fuse_file *ff; |
257 | struct fuse_req *req; | |
b6aeaded | 258 | |
6b2db28a TH |
259 | ff = file->private_data; |
260 | if (unlikely(!ff)) | |
8b0797a4 | 261 | return; |
6b2db28a | 262 | |
6b2db28a | 263 | req = ff->reserved_req; |
8b0797a4 | 264 | fuse_prepare_release(ff, file->f_flags, opcode); |
6b2db28a | 265 | |
37fb3a30 MS |
266 | if (ff->flock) { |
267 | struct fuse_release_in *inarg = &req->misc.release.in; | |
268 | inarg->release_flags |= FUSE_RELEASE_FLOCK_UNLOCK; | |
269 | inarg->lock_owner = fuse_lock_owner_id(ff->fc, | |
270 | (fl_owner_t) file); | |
271 | } | |
6b2db28a | 272 | /* Hold vfsmount and dentry until release is finished */ |
b0be46eb MS |
273 | path_get(&file->f_path); |
274 | req->misc.release.path = file->f_path; | |
6b2db28a | 275 | |
6b2db28a TH |
276 | /* |
277 | * Normally this will send the RELEASE request, however if | |
278 | * some asynchronous READ or WRITE requests are outstanding, | |
279 | * the sending will be delayed. | |
5a18ec17 MS |
280 | * |
281 | * Make the release synchronous if this is a fuseblk mount, | |
282 | * synchronous RELEASE is allowed (and desirable) in this case | |
283 | * because the server can be trusted not to screw up. | |
6b2db28a | 284 | */ |
5a18ec17 | 285 | fuse_file_put(ff, ff->fc->destroy_req != NULL); |
b6aeaded MS |
286 | } |
287 | ||
04730fef MS |
288 | static int fuse_open(struct inode *inode, struct file *file) |
289 | { | |
91fe96b4 | 290 | return fuse_open_common(inode, file, false); |
04730fef MS |
291 | } |
292 | ||
293 | static int fuse_release(struct inode *inode, struct file *file) | |
294 | { | |
8b0797a4 MS |
295 | fuse_release_common(file, FUSE_RELEASE); |
296 | ||
297 | /* return value is ignored by VFS */ | |
298 | return 0; | |
299 | } | |
300 | ||
301 | void fuse_sync_release(struct fuse_file *ff, int flags) | |
302 | { | |
303 | WARN_ON(atomic_read(&ff->count) > 1); | |
304 | fuse_prepare_release(ff, flags, FUSE_RELEASE); | |
305 | ff->reserved_req->force = 1; | |
8b41e671 | 306 | ff->reserved_req->background = 0; |
8b0797a4 MS |
307 | fuse_request_send(ff->fc, ff->reserved_req); |
308 | fuse_put_request(ff->fc, ff->reserved_req); | |
309 | kfree(ff); | |
04730fef | 310 | } |
08cbf542 | 311 | EXPORT_SYMBOL_GPL(fuse_sync_release); |
04730fef | 312 | |
71421259 | 313 | /* |
9c8ef561 MS |
314 | * Scramble the ID space with XTEA, so that the value of the files_struct |
315 | * pointer is not exposed to userspace. | |
71421259 | 316 | */ |
f3332114 | 317 | u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id) |
71421259 | 318 | { |
9c8ef561 MS |
319 | u32 *k = fc->scramble_key; |
320 | u64 v = (unsigned long) id; | |
321 | u32 v0 = v; | |
322 | u32 v1 = v >> 32; | |
323 | u32 sum = 0; | |
324 | int i; | |
325 | ||
326 | for (i = 0; i < 32; i++) { | |
327 | v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]); | |
328 | sum += 0x9E3779B9; | |
329 | v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]); | |
330 | } | |
331 | ||
332 | return (u64) v0 + ((u64) v1 << 32); | |
71421259 MS |
333 | } |
334 | ||
3be5a52b MS |
335 | /* |
336 | * Check if page is under writeback | |
337 | * | |
338 | * This is currently done by walking the list of writepage requests | |
339 | * for the inode, which can be pretty inefficient. | |
340 | */ | |
341 | static bool fuse_page_is_writeback(struct inode *inode, pgoff_t index) | |
342 | { | |
343 | struct fuse_conn *fc = get_fuse_conn(inode); | |
344 | struct fuse_inode *fi = get_fuse_inode(inode); | |
345 | struct fuse_req *req; | |
346 | bool found = false; | |
347 | ||
348 | spin_lock(&fc->lock); | |
349 | list_for_each_entry(req, &fi->writepages, writepages_entry) { | |
350 | pgoff_t curr_index; | |
351 | ||
352 | BUG_ON(req->inode != inode); | |
353 | curr_index = req->misc.write.in.offset >> PAGE_CACHE_SHIFT; | |
385b1268 PE |
354 | if (curr_index <= index && |
355 | index < curr_index + req->num_pages) { | |
3be5a52b MS |
356 | found = true; |
357 | break; | |
358 | } | |
359 | } | |
360 | spin_unlock(&fc->lock); | |
361 | ||
362 | return found; | |
363 | } | |
364 | ||
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 | ||
75e1fcc0 | 379 | static int fuse_flush(struct file *file, fl_owner_t id) |
b6aeaded | 380 | { |
6131ffaa | 381 | struct inode *inode = file_inode(file); |
b6aeaded MS |
382 | struct fuse_conn *fc = get_fuse_conn(inode); |
383 | struct fuse_file *ff = file->private_data; | |
384 | struct fuse_req *req; | |
385 | struct fuse_flush_in inarg; | |
386 | int err; | |
387 | ||
248d86e8 MS |
388 | if (is_bad_inode(inode)) |
389 | return -EIO; | |
390 | ||
b6aeaded MS |
391 | if (fc->no_flush) |
392 | return 0; | |
393 | ||
b111c8c0 | 394 | req = fuse_get_req_nofail_nopages(fc, file); |
b6aeaded MS |
395 | memset(&inarg, 0, sizeof(inarg)); |
396 | inarg.fh = ff->fh; | |
9c8ef561 | 397 | inarg.lock_owner = fuse_lock_owner_id(fc, id); |
b6aeaded MS |
398 | req->in.h.opcode = FUSE_FLUSH; |
399 | req->in.h.nodeid = get_node_id(inode); | |
b6aeaded MS |
400 | req->in.numargs = 1; |
401 | req->in.args[0].size = sizeof(inarg); | |
402 | req->in.args[0].value = &inarg; | |
71421259 | 403 | req->force = 1; |
b93f858a | 404 | fuse_request_send(fc, req); |
b6aeaded MS |
405 | err = req->out.h.error; |
406 | fuse_put_request(fc, req); | |
407 | if (err == -ENOSYS) { | |
408 | fc->no_flush = 1; | |
409 | err = 0; | |
410 | } | |
411 | return err; | |
412 | } | |
413 | ||
3be5a52b MS |
414 | /* |
415 | * Wait for all pending writepages on the inode to finish. | |
416 | * | |
417 | * This is currently done by blocking further writes with FUSE_NOWRITE | |
418 | * and waiting for all sent writes to complete. | |
419 | * | |
420 | * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage | |
421 | * could conflict with truncation. | |
422 | */ | |
423 | static void fuse_sync_writes(struct inode *inode) | |
424 | { | |
425 | fuse_set_nowrite(inode); | |
426 | fuse_release_nowrite(inode); | |
427 | } | |
428 | ||
02c24a82 JB |
429 | int fuse_fsync_common(struct file *file, loff_t start, loff_t end, |
430 | int datasync, int isdir) | |
b6aeaded | 431 | { |
7ea80859 | 432 | struct inode *inode = file->f_mapping->host; |
b6aeaded MS |
433 | struct fuse_conn *fc = get_fuse_conn(inode); |
434 | struct fuse_file *ff = file->private_data; | |
435 | struct fuse_req *req; | |
436 | struct fuse_fsync_in inarg; | |
437 | int err; | |
438 | ||
248d86e8 MS |
439 | if (is_bad_inode(inode)) |
440 | return -EIO; | |
441 | ||
02c24a82 JB |
442 | err = filemap_write_and_wait_range(inode->i_mapping, start, end); |
443 | if (err) | |
444 | return err; | |
445 | ||
82547981 | 446 | if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir)) |
b6aeaded MS |
447 | return 0; |
448 | ||
02c24a82 JB |
449 | mutex_lock(&inode->i_mutex); |
450 | ||
3be5a52b MS |
451 | /* |
452 | * Start writeback against all dirty pages of the inode, then | |
453 | * wait for all outstanding writes, before sending the FSYNC | |
454 | * request. | |
455 | */ | |
456 | err = write_inode_now(inode, 0); | |
457 | if (err) | |
02c24a82 | 458 | goto out; |
3be5a52b MS |
459 | |
460 | fuse_sync_writes(inode); | |
461 | ||
b111c8c0 | 462 | req = fuse_get_req_nopages(fc); |
02c24a82 JB |
463 | if (IS_ERR(req)) { |
464 | err = PTR_ERR(req); | |
465 | goto out; | |
466 | } | |
b6aeaded MS |
467 | |
468 | memset(&inarg, 0, sizeof(inarg)); | |
469 | inarg.fh = ff->fh; | |
470 | inarg.fsync_flags = datasync ? 1 : 0; | |
82547981 | 471 | req->in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC; |
b6aeaded | 472 | req->in.h.nodeid = get_node_id(inode); |
b6aeaded MS |
473 | req->in.numargs = 1; |
474 | req->in.args[0].size = sizeof(inarg); | |
475 | req->in.args[0].value = &inarg; | |
b93f858a | 476 | fuse_request_send(fc, req); |
b6aeaded MS |
477 | err = req->out.h.error; |
478 | fuse_put_request(fc, req); | |
479 | if (err == -ENOSYS) { | |
82547981 MS |
480 | if (isdir) |
481 | fc->no_fsyncdir = 1; | |
482 | else | |
483 | fc->no_fsync = 1; | |
b6aeaded MS |
484 | err = 0; |
485 | } | |
02c24a82 JB |
486 | out: |
487 | mutex_unlock(&inode->i_mutex); | |
b6aeaded MS |
488 | return err; |
489 | } | |
490 | ||
02c24a82 JB |
491 | static int fuse_fsync(struct file *file, loff_t start, loff_t end, |
492 | int datasync) | |
82547981 | 493 | { |
02c24a82 | 494 | return fuse_fsync_common(file, start, end, datasync, 0); |
82547981 MS |
495 | } |
496 | ||
2106cb18 MS |
497 | void fuse_read_fill(struct fuse_req *req, struct file *file, loff_t pos, |
498 | size_t count, int opcode) | |
b6aeaded | 499 | { |
5c5c5e51 | 500 | struct fuse_read_in *inarg = &req->misc.read.in; |
a6643094 | 501 | struct fuse_file *ff = file->private_data; |
b6aeaded | 502 | |
361b1eb5 MS |
503 | inarg->fh = ff->fh; |
504 | inarg->offset = pos; | |
505 | inarg->size = count; | |
a6643094 | 506 | inarg->flags = file->f_flags; |
361b1eb5 | 507 | req->in.h.opcode = opcode; |
2106cb18 | 508 | req->in.h.nodeid = ff->nodeid; |
b6aeaded MS |
509 | req->in.numargs = 1; |
510 | req->in.args[0].size = sizeof(struct fuse_read_in); | |
c1aa96a5 | 511 | req->in.args[0].value = inarg; |
b6aeaded MS |
512 | req->out.argvar = 1; |
513 | req->out.numargs = 1; | |
514 | req->out.args[0].size = count; | |
b6aeaded MS |
515 | } |
516 | ||
187c5c36 MP |
517 | static void fuse_release_user_pages(struct fuse_req *req, int write) |
518 | { | |
519 | unsigned i; | |
520 | ||
521 | for (i = 0; i < req->num_pages; i++) { | |
522 | struct page *page = req->pages[i]; | |
523 | if (write) | |
524 | set_page_dirty_lock(page); | |
525 | put_page(page); | |
526 | } | |
527 | } | |
528 | ||
01e9d11a MP |
529 | /** |
530 | * In case of short read, the caller sets 'pos' to the position of | |
531 | * actual end of fuse request in IO request. Otherwise, if bytes_requested | |
532 | * == bytes_transferred or rw == WRITE, the caller sets 'pos' to -1. | |
533 | * | |
534 | * An example: | |
535 | * User requested DIO read of 64K. It was splitted into two 32K fuse requests, | |
536 | * both submitted asynchronously. The first of them was ACKed by userspace as | |
537 | * fully completed (req->out.args[0].size == 32K) resulting in pos == -1. The | |
538 | * second request was ACKed as short, e.g. only 1K was read, resulting in | |
539 | * pos == 33K. | |
540 | * | |
541 | * Thus, when all fuse requests are completed, the minimal non-negative 'pos' | |
542 | * will be equal to the length of the longest contiguous fragment of | |
543 | * transferred data starting from the beginning of IO request. | |
544 | */ | |
545 | static void fuse_aio_complete(struct fuse_io_priv *io, int err, ssize_t pos) | |
546 | { | |
547 | int left; | |
548 | ||
549 | spin_lock(&io->lock); | |
550 | if (err) | |
551 | io->err = io->err ? : err; | |
552 | else if (pos >= 0 && (io->bytes < 0 || pos < io->bytes)) | |
553 | io->bytes = pos; | |
554 | ||
555 | left = --io->reqs; | |
556 | spin_unlock(&io->lock); | |
557 | ||
558 | if (!left) { | |
559 | long res; | |
560 | ||
561 | if (io->err) | |
562 | res = io->err; | |
563 | else if (io->bytes >= 0 && io->write) | |
564 | res = -EIO; | |
565 | else { | |
566 | res = io->bytes < 0 ? io->size : io->bytes; | |
567 | ||
568 | if (!is_sync_kiocb(io->iocb)) { | |
cb5e05d1 | 569 | struct inode *inode = file_inode(io->iocb->ki_filp); |
01e9d11a MP |
570 | struct fuse_conn *fc = get_fuse_conn(inode); |
571 | struct fuse_inode *fi = get_fuse_inode(inode); | |
572 | ||
573 | spin_lock(&fc->lock); | |
574 | fi->attr_version = ++fc->attr_version; | |
575 | spin_unlock(&fc->lock); | |
576 | } | |
577 | } | |
578 | ||
579 | aio_complete(io->iocb, res, 0); | |
580 | kfree(io); | |
581 | } | |
582 | } | |
583 | ||
584 | static void fuse_aio_complete_req(struct fuse_conn *fc, struct fuse_req *req) | |
585 | { | |
586 | struct fuse_io_priv *io = req->io; | |
587 | ssize_t pos = -1; | |
588 | ||
589 | fuse_release_user_pages(req, !io->write); | |
590 | ||
591 | if (io->write) { | |
592 | if (req->misc.write.in.size != req->misc.write.out.size) | |
593 | pos = req->misc.write.in.offset - io->offset + | |
594 | req->misc.write.out.size; | |
595 | } else { | |
596 | if (req->misc.read.in.size != req->out.args[0].size) | |
597 | pos = req->misc.read.in.offset - io->offset + | |
598 | req->out.args[0].size; | |
599 | } | |
600 | ||
601 | fuse_aio_complete(io, req->out.h.error, pos); | |
602 | } | |
603 | ||
604 | static size_t fuse_async_req_send(struct fuse_conn *fc, struct fuse_req *req, | |
605 | size_t num_bytes, struct fuse_io_priv *io) | |
606 | { | |
607 | spin_lock(&io->lock); | |
608 | io->size += num_bytes; | |
609 | io->reqs++; | |
610 | spin_unlock(&io->lock); | |
611 | ||
612 | req->io = io; | |
613 | req->end = fuse_aio_complete_req; | |
614 | ||
36cf66ed | 615 | __fuse_get_request(req); |
01e9d11a MP |
616 | fuse_request_send_background(fc, req); |
617 | ||
618 | return num_bytes; | |
619 | } | |
620 | ||
36cf66ed | 621 | static size_t fuse_send_read(struct fuse_req *req, struct fuse_io_priv *io, |
2106cb18 | 622 | loff_t pos, size_t count, fl_owner_t owner) |
04730fef | 623 | { |
36cf66ed | 624 | struct file *file = io->file; |
2106cb18 MS |
625 | struct fuse_file *ff = file->private_data; |
626 | struct fuse_conn *fc = ff->fc; | |
f3332114 | 627 | |
2106cb18 | 628 | fuse_read_fill(req, file, pos, count, FUSE_READ); |
f3332114 | 629 | if (owner != NULL) { |
5c5c5e51 | 630 | struct fuse_read_in *inarg = &req->misc.read.in; |
f3332114 MS |
631 | |
632 | inarg->read_flags |= FUSE_READ_LOCKOWNER; | |
633 | inarg->lock_owner = fuse_lock_owner_id(fc, owner); | |
634 | } | |
36cf66ed MP |
635 | |
636 | if (io->async) | |
637 | return fuse_async_req_send(fc, req, count, io); | |
638 | ||
b93f858a | 639 | fuse_request_send(fc, req); |
361b1eb5 | 640 | return req->out.args[0].size; |
04730fef MS |
641 | } |
642 | ||
5c5c5e51 MS |
643 | static void fuse_read_update_size(struct inode *inode, loff_t size, |
644 | u64 attr_ver) | |
645 | { | |
646 | struct fuse_conn *fc = get_fuse_conn(inode); | |
647 | struct fuse_inode *fi = get_fuse_inode(inode); | |
648 | ||
649 | spin_lock(&fc->lock); | |
06a7c3c2 MP |
650 | if (attr_ver == fi->attr_version && size < inode->i_size && |
651 | !test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) { | |
5c5c5e51 MS |
652 | fi->attr_version = ++fc->attr_version; |
653 | i_size_write(inode, size); | |
654 | } | |
655 | spin_unlock(&fc->lock); | |
656 | } | |
657 | ||
b6aeaded MS |
658 | static int fuse_readpage(struct file *file, struct page *page) |
659 | { | |
36cf66ed | 660 | struct fuse_io_priv io = { .async = 0, .file = file }; |
b6aeaded MS |
661 | struct inode *inode = page->mapping->host; |
662 | struct fuse_conn *fc = get_fuse_conn(inode); | |
248d86e8 | 663 | struct fuse_req *req; |
5c5c5e51 MS |
664 | size_t num_read; |
665 | loff_t pos = page_offset(page); | |
666 | size_t count = PAGE_CACHE_SIZE; | |
667 | u64 attr_ver; | |
248d86e8 MS |
668 | int err; |
669 | ||
670 | err = -EIO; | |
671 | if (is_bad_inode(inode)) | |
672 | goto out; | |
673 | ||
3be5a52b | 674 | /* |
25985edc | 675 | * Page writeback can extend beyond the lifetime of the |
3be5a52b MS |
676 | * page-cache page, so make sure we read a properly synced |
677 | * page. | |
678 | */ | |
679 | fuse_wait_on_page_writeback(inode, page->index); | |
680 | ||
b111c8c0 | 681 | req = fuse_get_req(fc, 1); |
ce1d5a49 MS |
682 | err = PTR_ERR(req); |
683 | if (IS_ERR(req)) | |
b6aeaded MS |
684 | goto out; |
685 | ||
5c5c5e51 MS |
686 | attr_ver = fuse_get_attr_version(fc); |
687 | ||
b6aeaded | 688 | req->out.page_zeroing = 1; |
f4975c67 | 689 | req->out.argpages = 1; |
b6aeaded MS |
690 | req->num_pages = 1; |
691 | req->pages[0] = page; | |
85f40aec | 692 | req->page_descs[0].length = count; |
36cf66ed | 693 | num_read = fuse_send_read(req, &io, pos, count, NULL); |
b6aeaded MS |
694 | err = req->out.h.error; |
695 | fuse_put_request(fc, req); | |
5c5c5e51 MS |
696 | |
697 | if (!err) { | |
698 | /* | |
699 | * Short read means EOF. If file size is larger, truncate it | |
700 | */ | |
701 | if (num_read < count) | |
702 | fuse_read_update_size(inode, pos + num_read, attr_ver); | |
703 | ||
b6aeaded | 704 | SetPageUptodate(page); |
5c5c5e51 MS |
705 | } |
706 | ||
451418fc | 707 | fuse_invalidate_atime(inode); |
b6aeaded MS |
708 | out: |
709 | unlock_page(page); | |
710 | return err; | |
711 | } | |
712 | ||
c1aa96a5 | 713 | static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req) |
db50b96c | 714 | { |
c1aa96a5 | 715 | int i; |
5c5c5e51 MS |
716 | size_t count = req->misc.read.in.size; |
717 | size_t num_read = req->out.args[0].size; | |
ce534fb0 | 718 | struct address_space *mapping = NULL; |
c1aa96a5 | 719 | |
ce534fb0 MS |
720 | for (i = 0; mapping == NULL && i < req->num_pages; i++) |
721 | mapping = req->pages[i]->mapping; | |
5c5c5e51 | 722 | |
ce534fb0 MS |
723 | if (mapping) { |
724 | struct inode *inode = mapping->host; | |
725 | ||
726 | /* | |
727 | * Short read means EOF. If file size is larger, truncate it | |
728 | */ | |
729 | if (!req->out.h.error && num_read < count) { | |
730 | loff_t pos; | |
731 | ||
732 | pos = page_offset(req->pages[0]) + num_read; | |
733 | fuse_read_update_size(inode, pos, | |
734 | req->misc.read.attr_ver); | |
735 | } | |
451418fc | 736 | fuse_invalidate_atime(inode); |
ce534fb0 | 737 | } |
c1aa96a5 | 738 | |
db50b96c MS |
739 | for (i = 0; i < req->num_pages; i++) { |
740 | struct page *page = req->pages[i]; | |
741 | if (!req->out.h.error) | |
742 | SetPageUptodate(page); | |
c1aa96a5 MS |
743 | else |
744 | SetPageError(page); | |
db50b96c | 745 | unlock_page(page); |
b5dd3285 | 746 | page_cache_release(page); |
db50b96c | 747 | } |
c756e0a4 | 748 | if (req->ff) |
5a18ec17 | 749 | fuse_file_put(req->ff, false); |
c1aa96a5 MS |
750 | } |
751 | ||
2106cb18 | 752 | static void fuse_send_readpages(struct fuse_req *req, struct file *file) |
c1aa96a5 | 753 | { |
2106cb18 MS |
754 | struct fuse_file *ff = file->private_data; |
755 | struct fuse_conn *fc = ff->fc; | |
c1aa96a5 MS |
756 | loff_t pos = page_offset(req->pages[0]); |
757 | size_t count = req->num_pages << PAGE_CACHE_SHIFT; | |
f4975c67 MS |
758 | |
759 | req->out.argpages = 1; | |
c1aa96a5 | 760 | req->out.page_zeroing = 1; |
ce534fb0 | 761 | req->out.page_replace = 1; |
2106cb18 | 762 | fuse_read_fill(req, file, pos, count, FUSE_READ); |
5c5c5e51 | 763 | req->misc.read.attr_ver = fuse_get_attr_version(fc); |
9cd68455 | 764 | if (fc->async_read) { |
c756e0a4 | 765 | req->ff = fuse_file_get(ff); |
9cd68455 | 766 | req->end = fuse_readpages_end; |
b93f858a | 767 | fuse_request_send_background(fc, req); |
9cd68455 | 768 | } else { |
b93f858a | 769 | fuse_request_send(fc, req); |
9cd68455 | 770 | fuse_readpages_end(fc, req); |
e9bb09dd | 771 | fuse_put_request(fc, req); |
9cd68455 | 772 | } |
db50b96c MS |
773 | } |
774 | ||
c756e0a4 | 775 | struct fuse_fill_data { |
db50b96c | 776 | struct fuse_req *req; |
a6643094 | 777 | struct file *file; |
db50b96c | 778 | struct inode *inode; |
f8dbdf81 | 779 | unsigned nr_pages; |
db50b96c MS |
780 | }; |
781 | ||
782 | static int fuse_readpages_fill(void *_data, struct page *page) | |
783 | { | |
c756e0a4 | 784 | struct fuse_fill_data *data = _data; |
db50b96c MS |
785 | struct fuse_req *req = data->req; |
786 | struct inode *inode = data->inode; | |
787 | struct fuse_conn *fc = get_fuse_conn(inode); | |
788 | ||
3be5a52b MS |
789 | fuse_wait_on_page_writeback(inode, page->index); |
790 | ||
db50b96c MS |
791 | if (req->num_pages && |
792 | (req->num_pages == FUSE_MAX_PAGES_PER_REQ || | |
793 | (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_read || | |
794 | req->pages[req->num_pages - 1]->index + 1 != page->index)) { | |
f8dbdf81 MP |
795 | int nr_alloc = min_t(unsigned, data->nr_pages, |
796 | FUSE_MAX_PAGES_PER_REQ); | |
2106cb18 | 797 | fuse_send_readpages(req, data->file); |
8b41e671 MP |
798 | if (fc->async_read) |
799 | req = fuse_get_req_for_background(fc, nr_alloc); | |
800 | else | |
801 | req = fuse_get_req(fc, nr_alloc); | |
802 | ||
803 | data->req = req; | |
ce1d5a49 | 804 | if (IS_ERR(req)) { |
db50b96c | 805 | unlock_page(page); |
ce1d5a49 | 806 | return PTR_ERR(req); |
db50b96c | 807 | } |
db50b96c | 808 | } |
f8dbdf81 MP |
809 | |
810 | if (WARN_ON(req->num_pages >= req->max_pages)) { | |
811 | fuse_put_request(fc, req); | |
812 | return -EIO; | |
813 | } | |
814 | ||
b5dd3285 | 815 | page_cache_get(page); |
db50b96c | 816 | req->pages[req->num_pages] = page; |
85f40aec | 817 | req->page_descs[req->num_pages].length = PAGE_SIZE; |
1729a16c | 818 | req->num_pages++; |
f8dbdf81 | 819 | data->nr_pages--; |
db50b96c MS |
820 | return 0; |
821 | } | |
822 | ||
823 | static int fuse_readpages(struct file *file, struct address_space *mapping, | |
824 | struct list_head *pages, unsigned nr_pages) | |
825 | { | |
826 | struct inode *inode = mapping->host; | |
827 | struct fuse_conn *fc = get_fuse_conn(inode); | |
c756e0a4 | 828 | struct fuse_fill_data data; |
db50b96c | 829 | int err; |
f8dbdf81 | 830 | int nr_alloc = min_t(unsigned, nr_pages, FUSE_MAX_PAGES_PER_REQ); |
248d86e8 | 831 | |
1d7ea732 | 832 | err = -EIO; |
248d86e8 | 833 | if (is_bad_inode(inode)) |
2e990021 | 834 | goto out; |
248d86e8 | 835 | |
a6643094 | 836 | data.file = file; |
db50b96c | 837 | data.inode = inode; |
8b41e671 MP |
838 | if (fc->async_read) |
839 | data.req = fuse_get_req_for_background(fc, nr_alloc); | |
840 | else | |
841 | data.req = fuse_get_req(fc, nr_alloc); | |
f8dbdf81 | 842 | data.nr_pages = nr_pages; |
1d7ea732 | 843 | err = PTR_ERR(data.req); |
ce1d5a49 | 844 | if (IS_ERR(data.req)) |
2e990021 | 845 | goto out; |
db50b96c MS |
846 | |
847 | err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data); | |
d3406ffa MS |
848 | if (!err) { |
849 | if (data.req->num_pages) | |
2106cb18 | 850 | fuse_send_readpages(data.req, file); |
d3406ffa MS |
851 | else |
852 | fuse_put_request(fc, data.req); | |
853 | } | |
2e990021 | 854 | out: |
1d7ea732 | 855 | return err; |
db50b96c MS |
856 | } |
857 | ||
bcb4be80 MS |
858 | static ssize_t fuse_file_aio_read(struct kiocb *iocb, const struct iovec *iov, |
859 | unsigned long nr_segs, loff_t pos) | |
860 | { | |
861 | struct inode *inode = iocb->ki_filp->f_mapping->host; | |
a8894274 | 862 | struct fuse_conn *fc = get_fuse_conn(inode); |
bcb4be80 | 863 | |
a8894274 BF |
864 | /* |
865 | * In auto invalidate mode, always update attributes on read. | |
866 | * Otherwise, only update if we attempt to read past EOF (to ensure | |
867 | * i_size is up to date). | |
868 | */ | |
869 | if (fc->auto_inval_data || | |
870 | (pos + iov_length(iov, nr_segs) > i_size_read(inode))) { | |
bcb4be80 | 871 | int err; |
bcb4be80 MS |
872 | err = fuse_update_attributes(inode, NULL, iocb->ki_filp, NULL); |
873 | if (err) | |
874 | return err; | |
875 | } | |
876 | ||
877 | return generic_file_aio_read(iocb, iov, nr_segs, pos); | |
878 | } | |
879 | ||
2d698b07 | 880 | static void fuse_write_fill(struct fuse_req *req, struct fuse_file *ff, |
2106cb18 | 881 | loff_t pos, size_t count) |
b6aeaded | 882 | { |
b25e82e5 MS |
883 | struct fuse_write_in *inarg = &req->misc.write.in; |
884 | struct fuse_write_out *outarg = &req->misc.write.out; | |
b6aeaded | 885 | |
b25e82e5 MS |
886 | inarg->fh = ff->fh; |
887 | inarg->offset = pos; | |
888 | inarg->size = count; | |
b6aeaded | 889 | req->in.h.opcode = FUSE_WRITE; |
2106cb18 | 890 | req->in.h.nodeid = ff->nodeid; |
b6aeaded | 891 | req->in.numargs = 2; |
2106cb18 | 892 | if (ff->fc->minor < 9) |
f3332114 MS |
893 | req->in.args[0].size = FUSE_COMPAT_WRITE_IN_SIZE; |
894 | else | |
895 | req->in.args[0].size = sizeof(struct fuse_write_in); | |
b25e82e5 | 896 | req->in.args[0].value = inarg; |
b6aeaded MS |
897 | req->in.args[1].size = count; |
898 | req->out.numargs = 1; | |
899 | req->out.args[0].size = sizeof(struct fuse_write_out); | |
b25e82e5 MS |
900 | req->out.args[0].value = outarg; |
901 | } | |
902 | ||
36cf66ed | 903 | static size_t fuse_send_write(struct fuse_req *req, struct fuse_io_priv *io, |
2106cb18 | 904 | loff_t pos, size_t count, fl_owner_t owner) |
b25e82e5 | 905 | { |
36cf66ed | 906 | struct file *file = io->file; |
2106cb18 MS |
907 | struct fuse_file *ff = file->private_data; |
908 | struct fuse_conn *fc = ff->fc; | |
2d698b07 MS |
909 | struct fuse_write_in *inarg = &req->misc.write.in; |
910 | ||
2106cb18 | 911 | fuse_write_fill(req, ff, pos, count); |
2d698b07 | 912 | inarg->flags = file->f_flags; |
f3332114 | 913 | if (owner != NULL) { |
f3332114 MS |
914 | inarg->write_flags |= FUSE_WRITE_LOCKOWNER; |
915 | inarg->lock_owner = fuse_lock_owner_id(fc, owner); | |
916 | } | |
36cf66ed MP |
917 | |
918 | if (io->async) | |
919 | return fuse_async_req_send(fc, req, count, io); | |
920 | ||
b93f858a | 921 | fuse_request_send(fc, req); |
b25e82e5 | 922 | return req->misc.write.out.size; |
b6aeaded MS |
923 | } |
924 | ||
a1d75f25 | 925 | void fuse_write_update_size(struct inode *inode, loff_t pos) |
854512ec MS |
926 | { |
927 | struct fuse_conn *fc = get_fuse_conn(inode); | |
928 | struct fuse_inode *fi = get_fuse_inode(inode); | |
929 | ||
930 | spin_lock(&fc->lock); | |
931 | fi->attr_version = ++fc->attr_version; | |
932 | if (pos > inode->i_size) | |
933 | i_size_write(inode, pos); | |
934 | spin_unlock(&fc->lock); | |
935 | } | |
936 | ||
ea9b9907 NP |
937 | static size_t fuse_send_write_pages(struct fuse_req *req, struct file *file, |
938 | struct inode *inode, loff_t pos, | |
939 | size_t count) | |
940 | { | |
941 | size_t res; | |
942 | unsigned offset; | |
943 | unsigned i; | |
36cf66ed | 944 | struct fuse_io_priv io = { .async = 0, .file = file }; |
ea9b9907 NP |
945 | |
946 | for (i = 0; i < req->num_pages; i++) | |
947 | fuse_wait_on_page_writeback(inode, req->pages[i]->index); | |
948 | ||
36cf66ed | 949 | res = fuse_send_write(req, &io, pos, count, NULL); |
ea9b9907 | 950 | |
b2430d75 | 951 | offset = req->page_descs[0].offset; |
ea9b9907 NP |
952 | count = res; |
953 | for (i = 0; i < req->num_pages; i++) { | |
954 | struct page *page = req->pages[i]; | |
955 | ||
956 | if (!req->out.h.error && !offset && count >= PAGE_CACHE_SIZE) | |
957 | SetPageUptodate(page); | |
958 | ||
959 | if (count > PAGE_CACHE_SIZE - offset) | |
960 | count -= PAGE_CACHE_SIZE - offset; | |
961 | else | |
962 | count = 0; | |
963 | offset = 0; | |
964 | ||
965 | unlock_page(page); | |
966 | page_cache_release(page); | |
967 | } | |
968 | ||
969 | return res; | |
970 | } | |
971 | ||
972 | static ssize_t fuse_fill_write_pages(struct fuse_req *req, | |
973 | struct address_space *mapping, | |
974 | struct iov_iter *ii, loff_t pos) | |
975 | { | |
976 | struct fuse_conn *fc = get_fuse_conn(mapping->host); | |
977 | unsigned offset = pos & (PAGE_CACHE_SIZE - 1); | |
978 | size_t count = 0; | |
979 | int err; | |
980 | ||
f4975c67 | 981 | req->in.argpages = 1; |
b2430d75 | 982 | req->page_descs[0].offset = offset; |
ea9b9907 NP |
983 | |
984 | do { | |
985 | size_t tmp; | |
986 | struct page *page; | |
987 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; | |
988 | size_t bytes = min_t(size_t, PAGE_CACHE_SIZE - offset, | |
989 | iov_iter_count(ii)); | |
990 | ||
991 | bytes = min_t(size_t, bytes, fc->max_write - count); | |
992 | ||
993 | again: | |
994 | err = -EFAULT; | |
995 | if (iov_iter_fault_in_readable(ii, bytes)) | |
996 | break; | |
997 | ||
998 | err = -ENOMEM; | |
54566b2c | 999 | page = grab_cache_page_write_begin(mapping, index, 0); |
ea9b9907 NP |
1000 | if (!page) |
1001 | break; | |
1002 | ||
931e80e4 | 1003 | if (mapping_writably_mapped(mapping)) |
1004 | flush_dcache_page(page); | |
1005 | ||
ea9b9907 NP |
1006 | pagefault_disable(); |
1007 | tmp = iov_iter_copy_from_user_atomic(page, ii, offset, bytes); | |
1008 | pagefault_enable(); | |
1009 | flush_dcache_page(page); | |
1010 | ||
478e0841 JW |
1011 | mark_page_accessed(page); |
1012 | ||
ea9b9907 NP |
1013 | if (!tmp) { |
1014 | unlock_page(page); | |
1015 | page_cache_release(page); | |
1016 | bytes = min(bytes, iov_iter_single_seg_count(ii)); | |
1017 | goto again; | |
1018 | } | |
1019 | ||
1020 | err = 0; | |
1021 | req->pages[req->num_pages] = page; | |
85f40aec | 1022 | req->page_descs[req->num_pages].length = tmp; |
ea9b9907 NP |
1023 | req->num_pages++; |
1024 | ||
1025 | iov_iter_advance(ii, tmp); | |
1026 | count += tmp; | |
1027 | pos += tmp; | |
1028 | offset += tmp; | |
1029 | if (offset == PAGE_CACHE_SIZE) | |
1030 | offset = 0; | |
1031 | ||
78bb6cb9 MS |
1032 | if (!fc->big_writes) |
1033 | break; | |
ea9b9907 | 1034 | } while (iov_iter_count(ii) && count < fc->max_write && |
d07f09f5 | 1035 | req->num_pages < req->max_pages && offset == 0); |
ea9b9907 NP |
1036 | |
1037 | return count > 0 ? count : err; | |
1038 | } | |
1039 | ||
d07f09f5 MP |
1040 | static inline unsigned fuse_wr_pages(loff_t pos, size_t len) |
1041 | { | |
1042 | return min_t(unsigned, | |
1043 | ((pos + len - 1) >> PAGE_CACHE_SHIFT) - | |
1044 | (pos >> PAGE_CACHE_SHIFT) + 1, | |
1045 | FUSE_MAX_PAGES_PER_REQ); | |
1046 | } | |
1047 | ||
ea9b9907 NP |
1048 | static ssize_t fuse_perform_write(struct file *file, |
1049 | struct address_space *mapping, | |
1050 | struct iov_iter *ii, loff_t pos) | |
1051 | { | |
1052 | struct inode *inode = mapping->host; | |
1053 | struct fuse_conn *fc = get_fuse_conn(inode); | |
06a7c3c2 | 1054 | struct fuse_inode *fi = get_fuse_inode(inode); |
ea9b9907 NP |
1055 | int err = 0; |
1056 | ssize_t res = 0; | |
1057 | ||
1058 | if (is_bad_inode(inode)) | |
1059 | return -EIO; | |
1060 | ||
06a7c3c2 MP |
1061 | if (inode->i_size < pos + iov_iter_count(ii)) |
1062 | set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); | |
1063 | ||
ea9b9907 NP |
1064 | do { |
1065 | struct fuse_req *req; | |
1066 | ssize_t count; | |
d07f09f5 | 1067 | unsigned nr_pages = fuse_wr_pages(pos, iov_iter_count(ii)); |
ea9b9907 | 1068 | |
d07f09f5 | 1069 | req = fuse_get_req(fc, nr_pages); |
ea9b9907 NP |
1070 | if (IS_ERR(req)) { |
1071 | err = PTR_ERR(req); | |
1072 | break; | |
1073 | } | |
1074 | ||
1075 | count = fuse_fill_write_pages(req, mapping, ii, pos); | |
1076 | if (count <= 0) { | |
1077 | err = count; | |
1078 | } else { | |
1079 | size_t num_written; | |
1080 | ||
1081 | num_written = fuse_send_write_pages(req, file, inode, | |
1082 | pos, count); | |
1083 | err = req->out.h.error; | |
1084 | if (!err) { | |
1085 | res += num_written; | |
1086 | pos += num_written; | |
1087 | ||
1088 | /* break out of the loop on short write */ | |
1089 | if (num_written != count) | |
1090 | err = -EIO; | |
1091 | } | |
1092 | } | |
1093 | fuse_put_request(fc, req); | |
1094 | } while (!err && iov_iter_count(ii)); | |
1095 | ||
1096 | if (res > 0) | |
1097 | fuse_write_update_size(inode, pos); | |
1098 | ||
06a7c3c2 | 1099 | clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); |
ea9b9907 NP |
1100 | fuse_invalidate_attr(inode); |
1101 | ||
1102 | return res > 0 ? res : err; | |
1103 | } | |
1104 | ||
1105 | static ssize_t fuse_file_aio_write(struct kiocb *iocb, const struct iovec *iov, | |
1106 | unsigned long nr_segs, loff_t pos) | |
1107 | { | |
1108 | struct file *file = iocb->ki_filp; | |
1109 | struct address_space *mapping = file->f_mapping; | |
1110 | size_t count = 0; | |
4273b793 | 1111 | size_t ocount = 0; |
ea9b9907 | 1112 | ssize_t written = 0; |
4273b793 | 1113 | ssize_t written_buffered = 0; |
ea9b9907 NP |
1114 | struct inode *inode = mapping->host; |
1115 | ssize_t err; | |
1116 | struct iov_iter i; | |
4273b793 | 1117 | loff_t endbyte = 0; |
ea9b9907 NP |
1118 | |
1119 | WARN_ON(iocb->ki_pos != pos); | |
1120 | ||
4273b793 AA |
1121 | ocount = 0; |
1122 | err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); | |
ea9b9907 NP |
1123 | if (err) |
1124 | return err; | |
1125 | ||
4273b793 | 1126 | count = ocount; |
ea9b9907 | 1127 | mutex_lock(&inode->i_mutex); |
ea9b9907 NP |
1128 | |
1129 | /* We can write back this queue in page reclaim */ | |
1130 | current->backing_dev_info = mapping->backing_dev_info; | |
1131 | ||
1132 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); | |
1133 | if (err) | |
1134 | goto out; | |
1135 | ||
1136 | if (count == 0) | |
1137 | goto out; | |
1138 | ||
2f1936b8 | 1139 | err = file_remove_suid(file); |
ea9b9907 NP |
1140 | if (err) |
1141 | goto out; | |
1142 | ||
c3b2da31 JB |
1143 | err = file_update_time(file); |
1144 | if (err) | |
1145 | goto out; | |
ea9b9907 | 1146 | |
4273b793 AA |
1147 | if (file->f_flags & O_DIRECT) { |
1148 | written = generic_file_direct_write(iocb, iov, &nr_segs, | |
1149 | pos, &iocb->ki_pos, | |
1150 | count, ocount); | |
1151 | if (written < 0 || written == count) | |
1152 | goto out; | |
1153 | ||
1154 | pos += written; | |
1155 | count -= written; | |
ea9b9907 | 1156 | |
4273b793 AA |
1157 | iov_iter_init(&i, iov, nr_segs, count, written); |
1158 | written_buffered = fuse_perform_write(file, mapping, &i, pos); | |
1159 | if (written_buffered < 0) { | |
1160 | err = written_buffered; | |
1161 | goto out; | |
1162 | } | |
1163 | endbyte = pos + written_buffered - 1; | |
1164 | ||
1165 | err = filemap_write_and_wait_range(file->f_mapping, pos, | |
1166 | endbyte); | |
1167 | if (err) | |
1168 | goto out; | |
1169 | ||
1170 | invalidate_mapping_pages(file->f_mapping, | |
1171 | pos >> PAGE_CACHE_SHIFT, | |
1172 | endbyte >> PAGE_CACHE_SHIFT); | |
1173 | ||
1174 | written += written_buffered; | |
1175 | iocb->ki_pos = pos + written_buffered; | |
1176 | } else { | |
1177 | iov_iter_init(&i, iov, nr_segs, count, 0); | |
1178 | written = fuse_perform_write(file, mapping, &i, pos); | |
1179 | if (written >= 0) | |
1180 | iocb->ki_pos = pos + written; | |
1181 | } | |
ea9b9907 NP |
1182 | out: |
1183 | current->backing_dev_info = NULL; | |
1184 | mutex_unlock(&inode->i_mutex); | |
1185 | ||
1186 | return written ? written : err; | |
1187 | } | |
1188 | ||
7c190c8b MP |
1189 | static inline void fuse_page_descs_length_init(struct fuse_req *req, |
1190 | unsigned index, unsigned nr_pages) | |
85f40aec MP |
1191 | { |
1192 | int i; | |
1193 | ||
7c190c8b | 1194 | for (i = index; i < index + nr_pages; i++) |
85f40aec MP |
1195 | req->page_descs[i].length = PAGE_SIZE - |
1196 | req->page_descs[i].offset; | |
1197 | } | |
1198 | ||
7c190c8b MP |
1199 | static inline unsigned long fuse_get_user_addr(const struct iov_iter *ii) |
1200 | { | |
1201 | return (unsigned long)ii->iov->iov_base + ii->iov_offset; | |
1202 | } | |
1203 | ||
1204 | static inline size_t fuse_get_frag_size(const struct iov_iter *ii, | |
1205 | size_t max_size) | |
1206 | { | |
1207 | return min(iov_iter_single_seg_count(ii), max_size); | |
1208 | } | |
1209 | ||
b98d023a | 1210 | static int fuse_get_user_pages(struct fuse_req *req, struct iov_iter *ii, |
ce60a2f1 | 1211 | size_t *nbytesp, int write) |
413ef8cb | 1212 | { |
7c190c8b | 1213 | size_t nbytes = 0; /* # bytes already packed in req */ |
b98d023a | 1214 | |
f4975c67 MS |
1215 | /* Special case for kernel I/O: can copy directly into the buffer */ |
1216 | if (segment_eq(get_fs(), KERNEL_DS)) { | |
7c190c8b MP |
1217 | unsigned long user_addr = fuse_get_user_addr(ii); |
1218 | size_t frag_size = fuse_get_frag_size(ii, *nbytesp); | |
1219 | ||
f4975c67 MS |
1220 | if (write) |
1221 | req->in.args[1].value = (void *) user_addr; | |
1222 | else | |
1223 | req->out.args[0].value = (void *) user_addr; | |
1224 | ||
b98d023a MP |
1225 | iov_iter_advance(ii, frag_size); |
1226 | *nbytesp = frag_size; | |
f4975c67 MS |
1227 | return 0; |
1228 | } | |
413ef8cb | 1229 | |
5565a9d8 | 1230 | while (nbytes < *nbytesp && req->num_pages < req->max_pages) { |
7c190c8b MP |
1231 | unsigned npages; |
1232 | unsigned long user_addr = fuse_get_user_addr(ii); | |
1233 | unsigned offset = user_addr & ~PAGE_MASK; | |
1234 | size_t frag_size = fuse_get_frag_size(ii, *nbytesp - nbytes); | |
1235 | int ret; | |
413ef8cb | 1236 | |
5565a9d8 | 1237 | unsigned n = req->max_pages - req->num_pages; |
7c190c8b MP |
1238 | frag_size = min_t(size_t, frag_size, n << PAGE_SHIFT); |
1239 | ||
1240 | npages = (frag_size + offset + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1241 | npages = clamp(npages, 1U, n); | |
1242 | ||
1243 | ret = get_user_pages_fast(user_addr, npages, !write, | |
1244 | &req->pages[req->num_pages]); | |
1245 | if (ret < 0) | |
1246 | return ret; | |
1247 | ||
1248 | npages = ret; | |
1249 | frag_size = min_t(size_t, frag_size, | |
1250 | (npages << PAGE_SHIFT) - offset); | |
1251 | iov_iter_advance(ii, frag_size); | |
1252 | ||
1253 | req->page_descs[req->num_pages].offset = offset; | |
1254 | fuse_page_descs_length_init(req, req->num_pages, npages); | |
1255 | ||
1256 | req->num_pages += npages; | |
1257 | req->page_descs[req->num_pages - 1].length -= | |
1258 | (npages << PAGE_SHIFT) - offset - frag_size; | |
1259 | ||
1260 | nbytes += frag_size; | |
1261 | } | |
f4975c67 MS |
1262 | |
1263 | if (write) | |
1264 | req->in.argpages = 1; | |
1265 | else | |
1266 | req->out.argpages = 1; | |
1267 | ||
7c190c8b | 1268 | *nbytesp = nbytes; |
f4975c67 | 1269 | |
413ef8cb MS |
1270 | return 0; |
1271 | } | |
1272 | ||
5565a9d8 MP |
1273 | static inline int fuse_iter_npages(const struct iov_iter *ii_p) |
1274 | { | |
1275 | struct iov_iter ii = *ii_p; | |
1276 | int npages = 0; | |
1277 | ||
1278 | while (iov_iter_count(&ii) && npages < FUSE_MAX_PAGES_PER_REQ) { | |
1279 | unsigned long user_addr = fuse_get_user_addr(&ii); | |
1280 | unsigned offset = user_addr & ~PAGE_MASK; | |
1281 | size_t frag_size = iov_iter_single_seg_count(&ii); | |
1282 | ||
1283 | npages += (frag_size + offset + PAGE_SIZE - 1) >> PAGE_SHIFT; | |
1284 | iov_iter_advance(&ii, frag_size); | |
1285 | } | |
1286 | ||
1287 | return min(npages, FUSE_MAX_PAGES_PER_REQ); | |
1288 | } | |
1289 | ||
36cf66ed | 1290 | ssize_t fuse_direct_io(struct fuse_io_priv *io, const struct iovec *iov, |
fb05f41f MS |
1291 | unsigned long nr_segs, size_t count, loff_t *ppos, |
1292 | int write) | |
413ef8cb | 1293 | { |
36cf66ed | 1294 | struct file *file = io->file; |
2106cb18 MS |
1295 | struct fuse_file *ff = file->private_data; |
1296 | struct fuse_conn *fc = ff->fc; | |
413ef8cb MS |
1297 | size_t nmax = write ? fc->max_write : fc->max_read; |
1298 | loff_t pos = *ppos; | |
1299 | ssize_t res = 0; | |
248d86e8 | 1300 | struct fuse_req *req; |
b98d023a MP |
1301 | struct iov_iter ii; |
1302 | ||
1303 | iov_iter_init(&ii, iov, nr_segs, count, 0); | |
248d86e8 | 1304 | |
de82b923 BF |
1305 | if (io->async) |
1306 | req = fuse_get_req_for_background(fc, fuse_iter_npages(&ii)); | |
1307 | else | |
1308 | req = fuse_get_req(fc, fuse_iter_npages(&ii)); | |
ce1d5a49 MS |
1309 | if (IS_ERR(req)) |
1310 | return PTR_ERR(req); | |
413ef8cb MS |
1311 | |
1312 | while (count) { | |
413ef8cb | 1313 | size_t nres; |
2106cb18 | 1314 | fl_owner_t owner = current->files; |
f4975c67 | 1315 | size_t nbytes = min(count, nmax); |
b98d023a | 1316 | int err = fuse_get_user_pages(req, &ii, &nbytes, write); |
413ef8cb MS |
1317 | if (err) { |
1318 | res = err; | |
1319 | break; | |
1320 | } | |
f4975c67 | 1321 | |
413ef8cb | 1322 | if (write) |
36cf66ed | 1323 | nres = fuse_send_write(req, io, pos, nbytes, owner); |
413ef8cb | 1324 | else |
36cf66ed | 1325 | nres = fuse_send_read(req, io, pos, nbytes, owner); |
2106cb18 | 1326 | |
36cf66ed MP |
1327 | if (!io->async) |
1328 | fuse_release_user_pages(req, !write); | |
413ef8cb MS |
1329 | if (req->out.h.error) { |
1330 | if (!res) | |
1331 | res = req->out.h.error; | |
1332 | break; | |
1333 | } else if (nres > nbytes) { | |
1334 | res = -EIO; | |
1335 | break; | |
1336 | } | |
1337 | count -= nres; | |
1338 | res += nres; | |
1339 | pos += nres; | |
413ef8cb MS |
1340 | if (nres != nbytes) |
1341 | break; | |
56cf34ff MS |
1342 | if (count) { |
1343 | fuse_put_request(fc, req); | |
de82b923 BF |
1344 | if (io->async) |
1345 | req = fuse_get_req_for_background(fc, | |
1346 | fuse_iter_npages(&ii)); | |
1347 | else | |
1348 | req = fuse_get_req(fc, fuse_iter_npages(&ii)); | |
56cf34ff MS |
1349 | if (IS_ERR(req)) |
1350 | break; | |
1351 | } | |
413ef8cb | 1352 | } |
f60311d5 AA |
1353 | if (!IS_ERR(req)) |
1354 | fuse_put_request(fc, req); | |
d09cb9d7 | 1355 | if (res > 0) |
413ef8cb | 1356 | *ppos = pos; |
413ef8cb MS |
1357 | |
1358 | return res; | |
1359 | } | |
08cbf542 | 1360 | EXPORT_SYMBOL_GPL(fuse_direct_io); |
413ef8cb | 1361 | |
36cf66ed MP |
1362 | static ssize_t __fuse_direct_read(struct fuse_io_priv *io, |
1363 | const struct iovec *iov, | |
439ee5f0 MP |
1364 | unsigned long nr_segs, loff_t *ppos, |
1365 | size_t count) | |
413ef8cb | 1366 | { |
d09cb9d7 | 1367 | ssize_t res; |
36cf66ed | 1368 | struct file *file = io->file; |
6131ffaa | 1369 | struct inode *inode = file_inode(file); |
d09cb9d7 MS |
1370 | |
1371 | if (is_bad_inode(inode)) | |
1372 | return -EIO; | |
1373 | ||
439ee5f0 | 1374 | res = fuse_direct_io(io, iov, nr_segs, count, ppos, 0); |
d09cb9d7 MS |
1375 | |
1376 | fuse_invalidate_attr(inode); | |
1377 | ||
1378 | return res; | |
413ef8cb MS |
1379 | } |
1380 | ||
b98d023a MP |
1381 | static ssize_t fuse_direct_read(struct file *file, char __user *buf, |
1382 | size_t count, loff_t *ppos) | |
1383 | { | |
36cf66ed | 1384 | struct fuse_io_priv io = { .async = 0, .file = file }; |
fb05f41f | 1385 | struct iovec iov = { .iov_base = buf, .iov_len = count }; |
439ee5f0 | 1386 | return __fuse_direct_read(&io, &iov, 1, ppos, count); |
b98d023a MP |
1387 | } |
1388 | ||
36cf66ed MP |
1389 | static ssize_t __fuse_direct_write(struct fuse_io_priv *io, |
1390 | const struct iovec *iov, | |
b98d023a | 1391 | unsigned long nr_segs, loff_t *ppos) |
413ef8cb | 1392 | { |
36cf66ed | 1393 | struct file *file = io->file; |
6131ffaa | 1394 | struct inode *inode = file_inode(file); |
b98d023a | 1395 | size_t count = iov_length(iov, nr_segs); |
413ef8cb | 1396 | ssize_t res; |
d09cb9d7 | 1397 | |
889f7848 | 1398 | res = generic_write_checks(file, ppos, &count, 0); |
bcba24cc | 1399 | if (!res) |
36cf66ed | 1400 | res = fuse_direct_io(io, iov, nr_segs, count, ppos, 1); |
d09cb9d7 MS |
1401 | |
1402 | fuse_invalidate_attr(inode); | |
1403 | ||
413ef8cb MS |
1404 | return res; |
1405 | } | |
1406 | ||
4273b793 AA |
1407 | static ssize_t fuse_direct_write(struct file *file, const char __user *buf, |
1408 | size_t count, loff_t *ppos) | |
1409 | { | |
fb05f41f | 1410 | struct iovec iov = { .iov_base = (void __user *)buf, .iov_len = count }; |
6131ffaa | 1411 | struct inode *inode = file_inode(file); |
4273b793 | 1412 | ssize_t res; |
36cf66ed | 1413 | struct fuse_io_priv io = { .async = 0, .file = file }; |
4273b793 AA |
1414 | |
1415 | if (is_bad_inode(inode)) | |
1416 | return -EIO; | |
1417 | ||
1418 | /* Don't allow parallel writes to the same file */ | |
1419 | mutex_lock(&inode->i_mutex); | |
36cf66ed | 1420 | res = __fuse_direct_write(&io, &iov, 1, ppos); |
bcba24cc MP |
1421 | if (res > 0) |
1422 | fuse_write_update_size(inode, *ppos); | |
4273b793 AA |
1423 | mutex_unlock(&inode->i_mutex); |
1424 | ||
1425 | return res; | |
1426 | } | |
1427 | ||
3be5a52b | 1428 | static void fuse_writepage_free(struct fuse_conn *fc, struct fuse_req *req) |
b6aeaded | 1429 | { |
385b1268 PE |
1430 | int i; |
1431 | ||
1432 | for (i = 0; i < req->num_pages; i++) | |
1433 | __free_page(req->pages[i]); | |
8b284dc4 MS |
1434 | |
1435 | if (req->ff) | |
1436 | fuse_file_put(req->ff, false); | |
3be5a52b MS |
1437 | } |
1438 | ||
1439 | static void fuse_writepage_finish(struct fuse_conn *fc, struct fuse_req *req) | |
1440 | { | |
1441 | struct inode *inode = req->inode; | |
1442 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1443 | struct backing_dev_info *bdi = inode->i_mapping->backing_dev_info; | |
385b1268 | 1444 | int i; |
3be5a52b MS |
1445 | |
1446 | list_del(&req->writepages_entry); | |
385b1268 PE |
1447 | for (i = 0; i < req->num_pages; i++) { |
1448 | dec_bdi_stat(bdi, BDI_WRITEBACK); | |
1449 | dec_zone_page_state(req->pages[i], NR_WRITEBACK_TEMP); | |
1450 | bdi_writeout_inc(bdi); | |
1451 | } | |
3be5a52b MS |
1452 | wake_up(&fi->page_waitq); |
1453 | } | |
1454 | ||
1455 | /* Called under fc->lock, may release and reacquire it */ | |
6eaf4782 MP |
1456 | static void fuse_send_writepage(struct fuse_conn *fc, struct fuse_req *req, |
1457 | loff_t size) | |
b9ca67b2 MS |
1458 | __releases(fc->lock) |
1459 | __acquires(fc->lock) | |
3be5a52b MS |
1460 | { |
1461 | struct fuse_inode *fi = get_fuse_inode(req->inode); | |
3be5a52b | 1462 | struct fuse_write_in *inarg = &req->misc.write.in; |
385b1268 | 1463 | __u64 data_size = req->num_pages * PAGE_CACHE_SIZE; |
3be5a52b MS |
1464 | |
1465 | if (!fc->connected) | |
1466 | goto out_free; | |
1467 | ||
385b1268 PE |
1468 | if (inarg->offset + data_size <= size) { |
1469 | inarg->size = data_size; | |
3be5a52b | 1470 | } else if (inarg->offset < size) { |
385b1268 | 1471 | inarg->size = size - inarg->offset; |
3be5a52b MS |
1472 | } else { |
1473 | /* Got truncated off completely */ | |
1474 | goto out_free; | |
b6aeaded | 1475 | } |
3be5a52b MS |
1476 | |
1477 | req->in.args[1].size = inarg->size; | |
1478 | fi->writectr++; | |
b93f858a | 1479 | fuse_request_send_background_locked(fc, req); |
3be5a52b MS |
1480 | return; |
1481 | ||
1482 | out_free: | |
1483 | fuse_writepage_finish(fc, req); | |
1484 | spin_unlock(&fc->lock); | |
1485 | fuse_writepage_free(fc, req); | |
e9bb09dd | 1486 | fuse_put_request(fc, req); |
3be5a52b | 1487 | spin_lock(&fc->lock); |
b6aeaded MS |
1488 | } |
1489 | ||
3be5a52b MS |
1490 | /* |
1491 | * If fi->writectr is positive (no truncate or fsync going on) send | |
1492 | * all queued writepage requests. | |
1493 | * | |
1494 | * Called with fc->lock | |
1495 | */ | |
1496 | void fuse_flush_writepages(struct inode *inode) | |
b9ca67b2 MS |
1497 | __releases(fc->lock) |
1498 | __acquires(fc->lock) | |
b6aeaded | 1499 | { |
3be5a52b MS |
1500 | struct fuse_conn *fc = get_fuse_conn(inode); |
1501 | struct fuse_inode *fi = get_fuse_inode(inode); | |
6eaf4782 | 1502 | size_t crop = i_size_read(inode); |
3be5a52b MS |
1503 | struct fuse_req *req; |
1504 | ||
1505 | while (fi->writectr >= 0 && !list_empty(&fi->queued_writes)) { | |
1506 | req = list_entry(fi->queued_writes.next, struct fuse_req, list); | |
1507 | list_del_init(&req->list); | |
6eaf4782 | 1508 | fuse_send_writepage(fc, req, crop); |
3be5a52b MS |
1509 | } |
1510 | } | |
1511 | ||
1512 | static void fuse_writepage_end(struct fuse_conn *fc, struct fuse_req *req) | |
1513 | { | |
1514 | struct inode *inode = req->inode; | |
1515 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1516 | ||
1517 | mapping_set_error(inode->i_mapping, req->out.h.error); | |
1518 | spin_lock(&fc->lock); | |
8b284dc4 | 1519 | while (req->misc.write.next) { |
6eaf4782 MP |
1520 | struct fuse_conn *fc = get_fuse_conn(inode); |
1521 | struct fuse_write_in *inarg = &req->misc.write.in; | |
8b284dc4 MS |
1522 | struct fuse_req *next = req->misc.write.next; |
1523 | req->misc.write.next = next->misc.write.next; | |
1524 | next->misc.write.next = NULL; | |
ce128de6 | 1525 | next->ff = fuse_file_get(req->ff); |
8b284dc4 | 1526 | list_add(&next->writepages_entry, &fi->writepages); |
6eaf4782 MP |
1527 | |
1528 | /* | |
1529 | * Skip fuse_flush_writepages() to make it easy to crop requests | |
1530 | * based on primary request size. | |
1531 | * | |
1532 | * 1st case (trivial): there are no concurrent activities using | |
1533 | * fuse_set/release_nowrite. Then we're on safe side because | |
1534 | * fuse_flush_writepages() would call fuse_send_writepage() | |
1535 | * anyway. | |
1536 | * | |
1537 | * 2nd case: someone called fuse_set_nowrite and it is waiting | |
1538 | * now for completion of all in-flight requests. This happens | |
1539 | * rarely and no more than once per page, so this should be | |
1540 | * okay. | |
1541 | * | |
1542 | * 3rd case: someone (e.g. fuse_do_setattr()) is in the middle | |
1543 | * of fuse_set_nowrite..fuse_release_nowrite section. The fact | |
1544 | * that fuse_set_nowrite returned implies that all in-flight | |
1545 | * requests were completed along with all of their secondary | |
1546 | * requests. Further primary requests are blocked by negative | |
1547 | * writectr. Hence there cannot be any in-flight requests and | |
1548 | * no invocations of fuse_writepage_end() while we're in | |
1549 | * fuse_set_nowrite..fuse_release_nowrite section. | |
1550 | */ | |
1551 | fuse_send_writepage(fc, next, inarg->offset + inarg->size); | |
8b284dc4 | 1552 | } |
3be5a52b MS |
1553 | fi->writectr--; |
1554 | fuse_writepage_finish(fc, req); | |
1555 | spin_unlock(&fc->lock); | |
1556 | fuse_writepage_free(fc, req); | |
1557 | } | |
1558 | ||
26d614df PE |
1559 | static struct fuse_file *fuse_write_file_get(struct fuse_conn *fc, |
1560 | struct fuse_inode *fi) | |
adcadfa8 | 1561 | { |
72523425 | 1562 | struct fuse_file *ff = NULL; |
adcadfa8 PE |
1563 | |
1564 | spin_lock(&fc->lock); | |
72523425 MS |
1565 | if (!WARN_ON(list_empty(&fi->write_files))) { |
1566 | ff = list_entry(fi->write_files.next, struct fuse_file, | |
1567 | write_entry); | |
1568 | fuse_file_get(ff); | |
1569 | } | |
adcadfa8 PE |
1570 | spin_unlock(&fc->lock); |
1571 | ||
1572 | return ff; | |
1573 | } | |
1574 | ||
3be5a52b MS |
1575 | static int fuse_writepage_locked(struct page *page) |
1576 | { | |
1577 | struct address_space *mapping = page->mapping; | |
1578 | struct inode *inode = mapping->host; | |
1579 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1580 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1581 | struct fuse_req *req; | |
3be5a52b | 1582 | struct page *tmp_page; |
72523425 | 1583 | int error = -ENOMEM; |
3be5a52b MS |
1584 | |
1585 | set_page_writeback(page); | |
1586 | ||
4250c066 | 1587 | req = fuse_request_alloc_nofs(1); |
3be5a52b MS |
1588 | if (!req) |
1589 | goto err; | |
1590 | ||
8b41e671 | 1591 | req->background = 1; /* writeback always goes to bg_queue */ |
3be5a52b MS |
1592 | tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); |
1593 | if (!tmp_page) | |
1594 | goto err_free; | |
1595 | ||
72523425 | 1596 | error = -EIO; |
26d614df | 1597 | req->ff = fuse_write_file_get(fc, fi); |
72523425 MS |
1598 | if (!req->ff) |
1599 | goto err_free; | |
1600 | ||
adcadfa8 | 1601 | fuse_write_fill(req, req->ff, page_offset(page), 0); |
3be5a52b MS |
1602 | |
1603 | copy_highpage(tmp_page, page); | |
2d698b07 | 1604 | req->misc.write.in.write_flags |= FUSE_WRITE_CACHE; |
8b284dc4 | 1605 | req->misc.write.next = NULL; |
f4975c67 | 1606 | req->in.argpages = 1; |
3be5a52b MS |
1607 | req->num_pages = 1; |
1608 | req->pages[0] = tmp_page; | |
b2430d75 | 1609 | req->page_descs[0].offset = 0; |
85f40aec | 1610 | req->page_descs[0].length = PAGE_SIZE; |
3be5a52b MS |
1611 | req->end = fuse_writepage_end; |
1612 | req->inode = inode; | |
1613 | ||
1614 | inc_bdi_stat(mapping->backing_dev_info, BDI_WRITEBACK); | |
1615 | inc_zone_page_state(tmp_page, NR_WRITEBACK_TEMP); | |
3be5a52b MS |
1616 | |
1617 | spin_lock(&fc->lock); | |
1618 | list_add(&req->writepages_entry, &fi->writepages); | |
1619 | list_add_tail(&req->list, &fi->queued_writes); | |
1620 | fuse_flush_writepages(inode); | |
1621 | spin_unlock(&fc->lock); | |
1622 | ||
4a4ac4eb MP |
1623 | end_page_writeback(page); |
1624 | ||
3be5a52b MS |
1625 | return 0; |
1626 | ||
1627 | err_free: | |
1628 | fuse_request_free(req); | |
1629 | err: | |
1630 | end_page_writeback(page); | |
72523425 | 1631 | return error; |
3be5a52b MS |
1632 | } |
1633 | ||
1634 | static int fuse_writepage(struct page *page, struct writeback_control *wbc) | |
1635 | { | |
1636 | int err; | |
1637 | ||
ff17be08 MS |
1638 | if (fuse_page_is_writeback(page->mapping->host, page->index)) { |
1639 | /* | |
1640 | * ->writepages() should be called for sync() and friends. We | |
1641 | * should only get here on direct reclaim and then we are | |
1642 | * allowed to skip a page which is already in flight | |
1643 | */ | |
1644 | WARN_ON(wbc->sync_mode == WB_SYNC_ALL); | |
1645 | ||
1646 | redirty_page_for_writepage(wbc, page); | |
1647 | return 0; | |
1648 | } | |
1649 | ||
3be5a52b MS |
1650 | err = fuse_writepage_locked(page); |
1651 | unlock_page(page); | |
1652 | ||
1653 | return err; | |
1654 | } | |
1655 | ||
26d614df PE |
1656 | struct fuse_fill_wb_data { |
1657 | struct fuse_req *req; | |
1658 | struct fuse_file *ff; | |
1659 | struct inode *inode; | |
2d033eaa | 1660 | struct page **orig_pages; |
26d614df PE |
1661 | }; |
1662 | ||
1663 | static void fuse_writepages_send(struct fuse_fill_wb_data *data) | |
1664 | { | |
1665 | struct fuse_req *req = data->req; | |
1666 | struct inode *inode = data->inode; | |
1667 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1668 | struct fuse_inode *fi = get_fuse_inode(inode); | |
2d033eaa MP |
1669 | int num_pages = req->num_pages; |
1670 | int i; | |
26d614df PE |
1671 | |
1672 | req->ff = fuse_file_get(data->ff); | |
1673 | spin_lock(&fc->lock); | |
1674 | list_add_tail(&req->list, &fi->queued_writes); | |
1675 | fuse_flush_writepages(inode); | |
1676 | spin_unlock(&fc->lock); | |
2d033eaa MP |
1677 | |
1678 | for (i = 0; i < num_pages; i++) | |
1679 | end_page_writeback(data->orig_pages[i]); | |
26d614df PE |
1680 | } |
1681 | ||
8b284dc4 MS |
1682 | static bool fuse_writepage_in_flight(struct fuse_req *new_req, |
1683 | struct page *page) | |
1684 | { | |
1685 | struct fuse_conn *fc = get_fuse_conn(new_req->inode); | |
1686 | struct fuse_inode *fi = get_fuse_inode(new_req->inode); | |
1687 | struct fuse_req *tmp; | |
1688 | struct fuse_req *old_req; | |
1689 | bool found = false; | |
1690 | pgoff_t curr_index; | |
1691 | ||
1692 | BUG_ON(new_req->num_pages != 0); | |
1693 | ||
1694 | spin_lock(&fc->lock); | |
1695 | list_del(&new_req->writepages_entry); | |
8b284dc4 MS |
1696 | list_for_each_entry(old_req, &fi->writepages, writepages_entry) { |
1697 | BUG_ON(old_req->inode != new_req->inode); | |
1698 | curr_index = old_req->misc.write.in.offset >> PAGE_CACHE_SHIFT; | |
1699 | if (curr_index <= page->index && | |
1700 | page->index < curr_index + old_req->num_pages) { | |
1701 | found = true; | |
1702 | break; | |
1703 | } | |
1704 | } | |
f6011081 MP |
1705 | if (!found) { |
1706 | list_add(&new_req->writepages_entry, &fi->writepages); | |
8b284dc4 | 1707 | goto out_unlock; |
f6011081 | 1708 | } |
8b284dc4 | 1709 | |
f6011081 | 1710 | new_req->num_pages = 1; |
8b284dc4 MS |
1711 | for (tmp = old_req; tmp != NULL; tmp = tmp->misc.write.next) { |
1712 | BUG_ON(tmp->inode != new_req->inode); | |
1713 | curr_index = tmp->misc.write.in.offset >> PAGE_CACHE_SHIFT; | |
1714 | if (tmp->num_pages == 1 && | |
1715 | curr_index == page->index) { | |
1716 | old_req = tmp; | |
1717 | } | |
1718 | } | |
1719 | ||
1720 | if (old_req->num_pages == 1 && (old_req->state == FUSE_REQ_INIT || | |
1721 | old_req->state == FUSE_REQ_PENDING)) { | |
41b6e41f MP |
1722 | struct backing_dev_info *bdi = page->mapping->backing_dev_info; |
1723 | ||
8b284dc4 MS |
1724 | copy_highpage(old_req->pages[0], page); |
1725 | spin_unlock(&fc->lock); | |
1726 | ||
41b6e41f | 1727 | dec_bdi_stat(bdi, BDI_WRITEBACK); |
8b284dc4 | 1728 | dec_zone_page_state(page, NR_WRITEBACK_TEMP); |
41b6e41f | 1729 | bdi_writeout_inc(bdi); |
8b284dc4 MS |
1730 | fuse_writepage_free(fc, new_req); |
1731 | fuse_request_free(new_req); | |
1732 | goto out; | |
1733 | } else { | |
1734 | new_req->misc.write.next = old_req->misc.write.next; | |
1735 | old_req->misc.write.next = new_req; | |
1736 | } | |
1737 | out_unlock: | |
1738 | spin_unlock(&fc->lock); | |
1739 | out: | |
1740 | return found; | |
1741 | } | |
1742 | ||
26d614df PE |
1743 | static int fuse_writepages_fill(struct page *page, |
1744 | struct writeback_control *wbc, void *_data) | |
1745 | { | |
1746 | struct fuse_fill_wb_data *data = _data; | |
1747 | struct fuse_req *req = data->req; | |
1748 | struct inode *inode = data->inode; | |
1749 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1750 | struct page *tmp_page; | |
8b284dc4 | 1751 | bool is_writeback; |
26d614df PE |
1752 | int err; |
1753 | ||
1754 | if (!data->ff) { | |
1755 | err = -EIO; | |
1756 | data->ff = fuse_write_file_get(fc, get_fuse_inode(inode)); | |
1757 | if (!data->ff) | |
1758 | goto out_unlock; | |
1759 | } | |
1760 | ||
8b284dc4 MS |
1761 | /* |
1762 | * Being under writeback is unlikely but possible. For example direct | |
1763 | * read to an mmaped fuse file will set the page dirty twice; once when | |
1764 | * the pages are faulted with get_user_pages(), and then after the read | |
1765 | * completed. | |
1766 | */ | |
1767 | is_writeback = fuse_page_is_writeback(inode, page->index); | |
1768 | ||
1769 | if (req && req->num_pages && | |
1770 | (is_writeback || req->num_pages == FUSE_MAX_PAGES_PER_REQ || | |
1771 | (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_write || | |
1772 | data->orig_pages[req->num_pages - 1]->index + 1 != page->index)) { | |
1773 | fuse_writepages_send(data); | |
1774 | data->req = NULL; | |
26d614df PE |
1775 | } |
1776 | err = -ENOMEM; | |
1777 | tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
1778 | if (!tmp_page) | |
1779 | goto out_unlock; | |
1780 | ||
1781 | /* | |
1782 | * The page must not be redirtied until the writeout is completed | |
1783 | * (i.e. userspace has sent a reply to the write request). Otherwise | |
1784 | * there could be more than one temporary page instance for each real | |
1785 | * page. | |
1786 | * | |
1787 | * This is ensured by holding the page lock in page_mkwrite() while | |
1788 | * checking fuse_page_is_writeback(). We already hold the page lock | |
1789 | * since clear_page_dirty_for_io() and keep it held until we add the | |
1790 | * request to the fi->writepages list and increment req->num_pages. | |
1791 | * After this fuse_page_is_writeback() will indicate that the page is | |
1792 | * under writeback, so we can release the page lock. | |
1793 | */ | |
1794 | if (data->req == NULL) { | |
1795 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1796 | ||
1797 | err = -ENOMEM; | |
1798 | req = fuse_request_alloc_nofs(FUSE_MAX_PAGES_PER_REQ); | |
1799 | if (!req) { | |
1800 | __free_page(tmp_page); | |
1801 | goto out_unlock; | |
1802 | } | |
1803 | ||
1804 | fuse_write_fill(req, data->ff, page_offset(page), 0); | |
1805 | req->misc.write.in.write_flags |= FUSE_WRITE_CACHE; | |
8b284dc4 | 1806 | req->misc.write.next = NULL; |
26d614df PE |
1807 | req->in.argpages = 1; |
1808 | req->background = 1; | |
1809 | req->num_pages = 0; | |
1810 | req->end = fuse_writepage_end; | |
1811 | req->inode = inode; | |
1812 | ||
1813 | spin_lock(&fc->lock); | |
1814 | list_add(&req->writepages_entry, &fi->writepages); | |
1815 | spin_unlock(&fc->lock); | |
1816 | ||
1817 | data->req = req; | |
1818 | } | |
1819 | set_page_writeback(page); | |
1820 | ||
1821 | copy_highpage(tmp_page, page); | |
1822 | req->pages[req->num_pages] = tmp_page; | |
1823 | req->page_descs[req->num_pages].offset = 0; | |
1824 | req->page_descs[req->num_pages].length = PAGE_SIZE; | |
1825 | ||
1826 | inc_bdi_stat(page->mapping->backing_dev_info, BDI_WRITEBACK); | |
1827 | inc_zone_page_state(tmp_page, NR_WRITEBACK_TEMP); | |
8b284dc4 MS |
1828 | |
1829 | err = 0; | |
1830 | if (is_writeback && fuse_writepage_in_flight(req, page)) { | |
1831 | end_page_writeback(page); | |
1832 | data->req = NULL; | |
1833 | goto out_unlock; | |
1834 | } | |
2d033eaa | 1835 | data->orig_pages[req->num_pages] = page; |
26d614df PE |
1836 | |
1837 | /* | |
1838 | * Protected by fc->lock against concurrent access by | |
1839 | * fuse_page_is_writeback(). | |
1840 | */ | |
1841 | spin_lock(&fc->lock); | |
1842 | req->num_pages++; | |
1843 | spin_unlock(&fc->lock); | |
1844 | ||
26d614df PE |
1845 | out_unlock: |
1846 | unlock_page(page); | |
1847 | ||
1848 | return err; | |
1849 | } | |
1850 | ||
1851 | static int fuse_writepages(struct address_space *mapping, | |
1852 | struct writeback_control *wbc) | |
1853 | { | |
1854 | struct inode *inode = mapping->host; | |
1855 | struct fuse_fill_wb_data data; | |
1856 | int err; | |
1857 | ||
1858 | err = -EIO; | |
1859 | if (is_bad_inode(inode)) | |
1860 | goto out; | |
1861 | ||
1862 | data.inode = inode; | |
1863 | data.req = NULL; | |
1864 | data.ff = NULL; | |
1865 | ||
2d033eaa MP |
1866 | err = -ENOMEM; |
1867 | data.orig_pages = kzalloc(sizeof(struct page *) * | |
1868 | FUSE_MAX_PAGES_PER_REQ, | |
1869 | GFP_NOFS); | |
1870 | if (!data.orig_pages) | |
1871 | goto out; | |
1872 | ||
26d614df PE |
1873 | err = write_cache_pages(mapping, wbc, fuse_writepages_fill, &data); |
1874 | if (data.req) { | |
1875 | /* Ignore errors if we can write at least one page */ | |
1876 | BUG_ON(!data.req->num_pages); | |
1877 | fuse_writepages_send(&data); | |
1878 | err = 0; | |
1879 | } | |
1880 | if (data.ff) | |
1881 | fuse_file_put(data.ff, false); | |
2d033eaa MP |
1882 | |
1883 | kfree(data.orig_pages); | |
26d614df PE |
1884 | out: |
1885 | return err; | |
1886 | } | |
1887 | ||
3be5a52b MS |
1888 | static int fuse_launder_page(struct page *page) |
1889 | { | |
1890 | int err = 0; | |
1891 | if (clear_page_dirty_for_io(page)) { | |
1892 | struct inode *inode = page->mapping->host; | |
1893 | err = fuse_writepage_locked(page); | |
1894 | if (!err) | |
1895 | fuse_wait_on_page_writeback(inode, page->index); | |
1896 | } | |
1897 | return err; | |
1898 | } | |
1899 | ||
1900 | /* | |
1901 | * Write back dirty pages now, because there may not be any suitable | |
1902 | * open files later | |
1903 | */ | |
1904 | static void fuse_vma_close(struct vm_area_struct *vma) | |
1905 | { | |
1906 | filemap_write_and_wait(vma->vm_file->f_mapping); | |
1907 | } | |
1908 | ||
1909 | /* | |
1910 | * Wait for writeback against this page to complete before allowing it | |
1911 | * to be marked dirty again, and hence written back again, possibly | |
1912 | * before the previous writepage completed. | |
1913 | * | |
1914 | * Block here, instead of in ->writepage(), so that the userspace fs | |
1915 | * can only block processes actually operating on the filesystem. | |
1916 | * | |
1917 | * Otherwise unprivileged userspace fs would be able to block | |
1918 | * unrelated: | |
1919 | * | |
1920 | * - page migration | |
1921 | * - sync(2) | |
1922 | * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER | |
1923 | */ | |
c2ec175c | 1924 | static int fuse_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
3be5a52b | 1925 | { |
c2ec175c | 1926 | struct page *page = vmf->page; |
cca24370 MS |
1927 | struct inode *inode = file_inode(vma->vm_file); |
1928 | ||
1929 | file_update_time(vma->vm_file); | |
1930 | lock_page(page); | |
1931 | if (page->mapping != inode->i_mapping) { | |
1932 | unlock_page(page); | |
1933 | return VM_FAULT_NOPAGE; | |
1934 | } | |
3be5a52b MS |
1935 | |
1936 | fuse_wait_on_page_writeback(inode, page->index); | |
cca24370 | 1937 | return VM_FAULT_LOCKED; |
3be5a52b MS |
1938 | } |
1939 | ||
f0f37e2f | 1940 | static const struct vm_operations_struct fuse_file_vm_ops = { |
3be5a52b MS |
1941 | .close = fuse_vma_close, |
1942 | .fault = filemap_fault, | |
1943 | .page_mkwrite = fuse_page_mkwrite, | |
0b173bc4 | 1944 | .remap_pages = generic_file_remap_pages, |
3be5a52b MS |
1945 | }; |
1946 | ||
1947 | static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma) | |
1948 | { | |
1949 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) { | |
6131ffaa | 1950 | struct inode *inode = file_inode(file); |
3be5a52b MS |
1951 | struct fuse_conn *fc = get_fuse_conn(inode); |
1952 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1953 | struct fuse_file *ff = file->private_data; | |
1954 | /* | |
1955 | * file may be written through mmap, so chain it onto the | |
1956 | * inodes's write_file list | |
1957 | */ | |
1958 | spin_lock(&fc->lock); | |
1959 | if (list_empty(&ff->write_entry)) | |
1960 | list_add(&ff->write_entry, &fi->write_files); | |
1961 | spin_unlock(&fc->lock); | |
1962 | } | |
1963 | file_accessed(file); | |
1964 | vma->vm_ops = &fuse_file_vm_ops; | |
b6aeaded MS |
1965 | return 0; |
1966 | } | |
1967 | ||
fc280c96 MS |
1968 | static int fuse_direct_mmap(struct file *file, struct vm_area_struct *vma) |
1969 | { | |
1970 | /* Can't provide the coherency needed for MAP_SHARED */ | |
1971 | if (vma->vm_flags & VM_MAYSHARE) | |
1972 | return -ENODEV; | |
1973 | ||
3121bfe7 MS |
1974 | invalidate_inode_pages2(file->f_mapping); |
1975 | ||
fc280c96 MS |
1976 | return generic_file_mmap(file, vma); |
1977 | } | |
1978 | ||
71421259 MS |
1979 | static int convert_fuse_file_lock(const struct fuse_file_lock *ffl, |
1980 | struct file_lock *fl) | |
1981 | { | |
1982 | switch (ffl->type) { | |
1983 | case F_UNLCK: | |
1984 | break; | |
1985 | ||
1986 | case F_RDLCK: | |
1987 | case F_WRLCK: | |
1988 | if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX || | |
1989 | ffl->end < ffl->start) | |
1990 | return -EIO; | |
1991 | ||
1992 | fl->fl_start = ffl->start; | |
1993 | fl->fl_end = ffl->end; | |
1994 | fl->fl_pid = ffl->pid; | |
1995 | break; | |
1996 | ||
1997 | default: | |
1998 | return -EIO; | |
1999 | } | |
2000 | fl->fl_type = ffl->type; | |
2001 | return 0; | |
2002 | } | |
2003 | ||
2004 | static void fuse_lk_fill(struct fuse_req *req, struct file *file, | |
a9ff4f87 MS |
2005 | const struct file_lock *fl, int opcode, pid_t pid, |
2006 | int flock) | |
71421259 | 2007 | { |
6131ffaa | 2008 | struct inode *inode = file_inode(file); |
9c8ef561 | 2009 | struct fuse_conn *fc = get_fuse_conn(inode); |
71421259 MS |
2010 | struct fuse_file *ff = file->private_data; |
2011 | struct fuse_lk_in *arg = &req->misc.lk_in; | |
2012 | ||
2013 | arg->fh = ff->fh; | |
9c8ef561 | 2014 | arg->owner = fuse_lock_owner_id(fc, fl->fl_owner); |
71421259 MS |
2015 | arg->lk.start = fl->fl_start; |
2016 | arg->lk.end = fl->fl_end; | |
2017 | arg->lk.type = fl->fl_type; | |
2018 | arg->lk.pid = pid; | |
a9ff4f87 MS |
2019 | if (flock) |
2020 | arg->lk_flags |= FUSE_LK_FLOCK; | |
71421259 MS |
2021 | req->in.h.opcode = opcode; |
2022 | req->in.h.nodeid = get_node_id(inode); | |
2023 | req->in.numargs = 1; | |
2024 | req->in.args[0].size = sizeof(*arg); | |
2025 | req->in.args[0].value = arg; | |
2026 | } | |
2027 | ||
2028 | static int fuse_getlk(struct file *file, struct file_lock *fl) | |
2029 | { | |
6131ffaa | 2030 | struct inode *inode = file_inode(file); |
71421259 MS |
2031 | struct fuse_conn *fc = get_fuse_conn(inode); |
2032 | struct fuse_req *req; | |
2033 | struct fuse_lk_out outarg; | |
2034 | int err; | |
2035 | ||
b111c8c0 | 2036 | req = fuse_get_req_nopages(fc); |
71421259 MS |
2037 | if (IS_ERR(req)) |
2038 | return PTR_ERR(req); | |
2039 | ||
a9ff4f87 | 2040 | fuse_lk_fill(req, file, fl, FUSE_GETLK, 0, 0); |
71421259 MS |
2041 | req->out.numargs = 1; |
2042 | req->out.args[0].size = sizeof(outarg); | |
2043 | req->out.args[0].value = &outarg; | |
b93f858a | 2044 | fuse_request_send(fc, req); |
71421259 MS |
2045 | err = req->out.h.error; |
2046 | fuse_put_request(fc, req); | |
2047 | if (!err) | |
2048 | err = convert_fuse_file_lock(&outarg.lk, fl); | |
2049 | ||
2050 | return err; | |
2051 | } | |
2052 | ||
a9ff4f87 | 2053 | static int fuse_setlk(struct file *file, struct file_lock *fl, int flock) |
71421259 | 2054 | { |
6131ffaa | 2055 | struct inode *inode = file_inode(file); |
71421259 MS |
2056 | struct fuse_conn *fc = get_fuse_conn(inode); |
2057 | struct fuse_req *req; | |
2058 | int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK; | |
2059 | pid_t pid = fl->fl_type != F_UNLCK ? current->tgid : 0; | |
2060 | int err; | |
2061 | ||
8fb47a4f | 2062 | if (fl->fl_lmops && fl->fl_lmops->lm_grant) { |
48e90761 MS |
2063 | /* NLM needs asynchronous locks, which we don't support yet */ |
2064 | return -ENOLCK; | |
2065 | } | |
2066 | ||
71421259 MS |
2067 | /* Unlock on close is handled by the flush method */ |
2068 | if (fl->fl_flags & FL_CLOSE) | |
2069 | return 0; | |
2070 | ||
b111c8c0 | 2071 | req = fuse_get_req_nopages(fc); |
71421259 MS |
2072 | if (IS_ERR(req)) |
2073 | return PTR_ERR(req); | |
2074 | ||
a9ff4f87 | 2075 | fuse_lk_fill(req, file, fl, opcode, pid, flock); |
b93f858a | 2076 | fuse_request_send(fc, req); |
71421259 | 2077 | err = req->out.h.error; |
a4d27e75 MS |
2078 | /* locking is restartable */ |
2079 | if (err == -EINTR) | |
2080 | err = -ERESTARTSYS; | |
71421259 MS |
2081 | fuse_put_request(fc, req); |
2082 | return err; | |
2083 | } | |
2084 | ||
2085 | static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl) | |
2086 | { | |
6131ffaa | 2087 | struct inode *inode = file_inode(file); |
71421259 MS |
2088 | struct fuse_conn *fc = get_fuse_conn(inode); |
2089 | int err; | |
2090 | ||
48e90761 MS |
2091 | if (cmd == F_CANCELLK) { |
2092 | err = 0; | |
2093 | } else if (cmd == F_GETLK) { | |
71421259 | 2094 | if (fc->no_lock) { |
9d6a8c5c | 2095 | posix_test_lock(file, fl); |
71421259 MS |
2096 | err = 0; |
2097 | } else | |
2098 | err = fuse_getlk(file, fl); | |
2099 | } else { | |
2100 | if (fc->no_lock) | |
48e90761 | 2101 | err = posix_lock_file(file, fl, NULL); |
71421259 | 2102 | else |
a9ff4f87 | 2103 | err = fuse_setlk(file, fl, 0); |
71421259 MS |
2104 | } |
2105 | return err; | |
2106 | } | |
2107 | ||
a9ff4f87 MS |
2108 | static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl) |
2109 | { | |
6131ffaa | 2110 | struct inode *inode = file_inode(file); |
a9ff4f87 MS |
2111 | struct fuse_conn *fc = get_fuse_conn(inode); |
2112 | int err; | |
2113 | ||
37fb3a30 | 2114 | if (fc->no_flock) { |
a9ff4f87 MS |
2115 | err = flock_lock_file_wait(file, fl); |
2116 | } else { | |
37fb3a30 MS |
2117 | struct fuse_file *ff = file->private_data; |
2118 | ||
a9ff4f87 MS |
2119 | /* emulate flock with POSIX locks */ |
2120 | fl->fl_owner = (fl_owner_t) file; | |
37fb3a30 | 2121 | ff->flock = true; |
a9ff4f87 MS |
2122 | err = fuse_setlk(file, fl, 1); |
2123 | } | |
2124 | ||
2125 | return err; | |
2126 | } | |
2127 | ||
b2d2272f MS |
2128 | static sector_t fuse_bmap(struct address_space *mapping, sector_t block) |
2129 | { | |
2130 | struct inode *inode = mapping->host; | |
2131 | struct fuse_conn *fc = get_fuse_conn(inode); | |
2132 | struct fuse_req *req; | |
2133 | struct fuse_bmap_in inarg; | |
2134 | struct fuse_bmap_out outarg; | |
2135 | int err; | |
2136 | ||
2137 | if (!inode->i_sb->s_bdev || fc->no_bmap) | |
2138 | return 0; | |
2139 | ||
b111c8c0 | 2140 | req = fuse_get_req_nopages(fc); |
b2d2272f MS |
2141 | if (IS_ERR(req)) |
2142 | return 0; | |
2143 | ||
2144 | memset(&inarg, 0, sizeof(inarg)); | |
2145 | inarg.block = block; | |
2146 | inarg.blocksize = inode->i_sb->s_blocksize; | |
2147 | req->in.h.opcode = FUSE_BMAP; | |
2148 | req->in.h.nodeid = get_node_id(inode); | |
2149 | req->in.numargs = 1; | |
2150 | req->in.args[0].size = sizeof(inarg); | |
2151 | req->in.args[0].value = &inarg; | |
2152 | req->out.numargs = 1; | |
2153 | req->out.args[0].size = sizeof(outarg); | |
2154 | req->out.args[0].value = &outarg; | |
b93f858a | 2155 | fuse_request_send(fc, req); |
b2d2272f MS |
2156 | err = req->out.h.error; |
2157 | fuse_put_request(fc, req); | |
2158 | if (err == -ENOSYS) | |
2159 | fc->no_bmap = 1; | |
2160 | ||
2161 | return err ? 0 : outarg.block; | |
2162 | } | |
2163 | ||
965c8e59 | 2164 | static loff_t fuse_file_llseek(struct file *file, loff_t offset, int whence) |
5559b8f4 MS |
2165 | { |
2166 | loff_t retval; | |
6131ffaa | 2167 | struct inode *inode = file_inode(file); |
5559b8f4 | 2168 | |
c07c3d19 | 2169 | /* No i_mutex protection necessary for SEEK_CUR and SEEK_SET */ |
965c8e59 AM |
2170 | if (whence == SEEK_CUR || whence == SEEK_SET) |
2171 | return generic_file_llseek(file, offset, whence); | |
06222e49 | 2172 | |
c07c3d19 MS |
2173 | mutex_lock(&inode->i_mutex); |
2174 | retval = fuse_update_attributes(inode, NULL, file, NULL); | |
2175 | if (!retval) | |
965c8e59 | 2176 | retval = generic_file_llseek(file, offset, whence); |
5559b8f4 | 2177 | mutex_unlock(&inode->i_mutex); |
c07c3d19 | 2178 | |
5559b8f4 MS |
2179 | return retval; |
2180 | } | |
2181 | ||
59efec7b TH |
2182 | static int fuse_ioctl_copy_user(struct page **pages, struct iovec *iov, |
2183 | unsigned int nr_segs, size_t bytes, bool to_user) | |
2184 | { | |
2185 | struct iov_iter ii; | |
2186 | int page_idx = 0; | |
2187 | ||
2188 | if (!bytes) | |
2189 | return 0; | |
2190 | ||
2191 | iov_iter_init(&ii, iov, nr_segs, bytes, 0); | |
2192 | ||
2193 | while (iov_iter_count(&ii)) { | |
2194 | struct page *page = pages[page_idx++]; | |
2195 | size_t todo = min_t(size_t, PAGE_SIZE, iov_iter_count(&ii)); | |
4aa0edd2 | 2196 | void *kaddr; |
59efec7b | 2197 | |
4aa0edd2 | 2198 | kaddr = kmap(page); |
59efec7b TH |
2199 | |
2200 | while (todo) { | |
2201 | char __user *uaddr = ii.iov->iov_base + ii.iov_offset; | |
2202 | size_t iov_len = ii.iov->iov_len - ii.iov_offset; | |
2203 | size_t copy = min(todo, iov_len); | |
2204 | size_t left; | |
2205 | ||
2206 | if (!to_user) | |
2207 | left = copy_from_user(kaddr, uaddr, copy); | |
2208 | else | |
2209 | left = copy_to_user(uaddr, kaddr, copy); | |
2210 | ||
2211 | if (unlikely(left)) | |
2212 | return -EFAULT; | |
2213 | ||
2214 | iov_iter_advance(&ii, copy); | |
2215 | todo -= copy; | |
2216 | kaddr += copy; | |
2217 | } | |
2218 | ||
0bd87182 | 2219 | kunmap(page); |
59efec7b TH |
2220 | } |
2221 | ||
2222 | return 0; | |
2223 | } | |
2224 | ||
d9d318d3 MS |
2225 | /* |
2226 | * CUSE servers compiled on 32bit broke on 64bit kernels because the | |
2227 | * ABI was defined to be 'struct iovec' which is different on 32bit | |
2228 | * and 64bit. Fortunately we can determine which structure the server | |
2229 | * used from the size of the reply. | |
2230 | */ | |
1baa26b2 MS |
2231 | static int fuse_copy_ioctl_iovec_old(struct iovec *dst, void *src, |
2232 | size_t transferred, unsigned count, | |
2233 | bool is_compat) | |
d9d318d3 MS |
2234 | { |
2235 | #ifdef CONFIG_COMPAT | |
2236 | if (count * sizeof(struct compat_iovec) == transferred) { | |
2237 | struct compat_iovec *ciov = src; | |
2238 | unsigned i; | |
2239 | ||
2240 | /* | |
2241 | * With this interface a 32bit server cannot support | |
2242 | * non-compat (i.e. ones coming from 64bit apps) ioctl | |
2243 | * requests | |
2244 | */ | |
2245 | if (!is_compat) | |
2246 | return -EINVAL; | |
2247 | ||
2248 | for (i = 0; i < count; i++) { | |
2249 | dst[i].iov_base = compat_ptr(ciov[i].iov_base); | |
2250 | dst[i].iov_len = ciov[i].iov_len; | |
2251 | } | |
2252 | return 0; | |
2253 | } | |
2254 | #endif | |
2255 | ||
2256 | if (count * sizeof(struct iovec) != transferred) | |
2257 | return -EIO; | |
2258 | ||
2259 | memcpy(dst, src, transferred); | |
2260 | return 0; | |
2261 | } | |
2262 | ||
7572777e MS |
2263 | /* Make sure iov_length() won't overflow */ |
2264 | static int fuse_verify_ioctl_iov(struct iovec *iov, size_t count) | |
2265 | { | |
2266 | size_t n; | |
2267 | u32 max = FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT; | |
2268 | ||
fb6ccff6 | 2269 | for (n = 0; n < count; n++, iov++) { |
7572777e MS |
2270 | if (iov->iov_len > (size_t) max) |
2271 | return -ENOMEM; | |
2272 | max -= iov->iov_len; | |
2273 | } | |
2274 | return 0; | |
2275 | } | |
2276 | ||
1baa26b2 MS |
2277 | static int fuse_copy_ioctl_iovec(struct fuse_conn *fc, struct iovec *dst, |
2278 | void *src, size_t transferred, unsigned count, | |
2279 | bool is_compat) | |
2280 | { | |
2281 | unsigned i; | |
2282 | struct fuse_ioctl_iovec *fiov = src; | |
2283 | ||
2284 | if (fc->minor < 16) { | |
2285 | return fuse_copy_ioctl_iovec_old(dst, src, transferred, | |
2286 | count, is_compat); | |
2287 | } | |
2288 | ||
2289 | if (count * sizeof(struct fuse_ioctl_iovec) != transferred) | |
2290 | return -EIO; | |
2291 | ||
2292 | for (i = 0; i < count; i++) { | |
2293 | /* Did the server supply an inappropriate value? */ | |
2294 | if (fiov[i].base != (unsigned long) fiov[i].base || | |
2295 | fiov[i].len != (unsigned long) fiov[i].len) | |
2296 | return -EIO; | |
2297 | ||
2298 | dst[i].iov_base = (void __user *) (unsigned long) fiov[i].base; | |
2299 | dst[i].iov_len = (size_t) fiov[i].len; | |
2300 | ||
2301 | #ifdef CONFIG_COMPAT | |
2302 | if (is_compat && | |
2303 | (ptr_to_compat(dst[i].iov_base) != fiov[i].base || | |
2304 | (compat_size_t) dst[i].iov_len != fiov[i].len)) | |
2305 | return -EIO; | |
2306 | #endif | |
2307 | } | |
2308 | ||
2309 | return 0; | |
2310 | } | |
2311 | ||
2312 | ||
59efec7b TH |
2313 | /* |
2314 | * For ioctls, there is no generic way to determine how much memory | |
2315 | * needs to be read and/or written. Furthermore, ioctls are allowed | |
2316 | * to dereference the passed pointer, so the parameter requires deep | |
2317 | * copying but FUSE has no idea whatsoever about what to copy in or | |
2318 | * out. | |
2319 | * | |
2320 | * This is solved by allowing FUSE server to retry ioctl with | |
2321 | * necessary in/out iovecs. Let's assume the ioctl implementation | |
2322 | * needs to read in the following structure. | |
2323 | * | |
2324 | * struct a { | |
2325 | * char *buf; | |
2326 | * size_t buflen; | |
2327 | * } | |
2328 | * | |
2329 | * On the first callout to FUSE server, inarg->in_size and | |
2330 | * inarg->out_size will be NULL; then, the server completes the ioctl | |
2331 | * with FUSE_IOCTL_RETRY set in out->flags, out->in_iovs set to 1 and | |
2332 | * the actual iov array to | |
2333 | * | |
2334 | * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) } } | |
2335 | * | |
2336 | * which tells FUSE to copy in the requested area and retry the ioctl. | |
2337 | * On the second round, the server has access to the structure and | |
2338 | * from that it can tell what to look for next, so on the invocation, | |
2339 | * it sets FUSE_IOCTL_RETRY, out->in_iovs to 2 and iov array to | |
2340 | * | |
2341 | * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) }, | |
2342 | * { .iov_base = a.buf, .iov_len = a.buflen } } | |
2343 | * | |
2344 | * FUSE will copy both struct a and the pointed buffer from the | |
2345 | * process doing the ioctl and retry ioctl with both struct a and the | |
2346 | * buffer. | |
2347 | * | |
2348 | * This time, FUSE server has everything it needs and completes ioctl | |
2349 | * without FUSE_IOCTL_RETRY which finishes the ioctl call. | |
2350 | * | |
2351 | * Copying data out works the same way. | |
2352 | * | |
2353 | * Note that if FUSE_IOCTL_UNRESTRICTED is clear, the kernel | |
2354 | * automatically initializes in and out iovs by decoding @cmd with | |
2355 | * _IOC_* macros and the server is not allowed to request RETRY. This | |
2356 | * limits ioctl data transfers to well-formed ioctls and is the forced | |
2357 | * behavior for all FUSE servers. | |
2358 | */ | |
08cbf542 TH |
2359 | long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg, |
2360 | unsigned int flags) | |
59efec7b | 2361 | { |
59efec7b | 2362 | struct fuse_file *ff = file->private_data; |
d36f2487 | 2363 | struct fuse_conn *fc = ff->fc; |
59efec7b TH |
2364 | struct fuse_ioctl_in inarg = { |
2365 | .fh = ff->fh, | |
2366 | .cmd = cmd, | |
2367 | .arg = arg, | |
2368 | .flags = flags | |
2369 | }; | |
2370 | struct fuse_ioctl_out outarg; | |
2371 | struct fuse_req *req = NULL; | |
2372 | struct page **pages = NULL; | |
8ac83505 | 2373 | struct iovec *iov_page = NULL; |
59efec7b TH |
2374 | struct iovec *in_iov = NULL, *out_iov = NULL; |
2375 | unsigned int in_iovs = 0, out_iovs = 0, num_pages = 0, max_pages; | |
2376 | size_t in_size, out_size, transferred; | |
2377 | int err; | |
2378 | ||
1baa26b2 MS |
2379 | #if BITS_PER_LONG == 32 |
2380 | inarg.flags |= FUSE_IOCTL_32BIT; | |
2381 | #else | |
2382 | if (flags & FUSE_IOCTL_COMPAT) | |
2383 | inarg.flags |= FUSE_IOCTL_32BIT; | |
2384 | #endif | |
2385 | ||
59efec7b | 2386 | /* assume all the iovs returned by client always fits in a page */ |
1baa26b2 | 2387 | BUILD_BUG_ON(sizeof(struct fuse_ioctl_iovec) * FUSE_IOCTL_MAX_IOV > PAGE_SIZE); |
59efec7b | 2388 | |
59efec7b | 2389 | err = -ENOMEM; |
c411cc88 | 2390 | pages = kcalloc(FUSE_MAX_PAGES_PER_REQ, sizeof(pages[0]), GFP_KERNEL); |
8ac83505 | 2391 | iov_page = (struct iovec *) __get_free_page(GFP_KERNEL); |
59efec7b TH |
2392 | if (!pages || !iov_page) |
2393 | goto out; | |
2394 | ||
2395 | /* | |
2396 | * If restricted, initialize IO parameters as encoded in @cmd. | |
2397 | * RETRY from server is not allowed. | |
2398 | */ | |
2399 | if (!(flags & FUSE_IOCTL_UNRESTRICTED)) { | |
8ac83505 | 2400 | struct iovec *iov = iov_page; |
59efec7b | 2401 | |
c9f0523d | 2402 | iov->iov_base = (void __user *)arg; |
59efec7b TH |
2403 | iov->iov_len = _IOC_SIZE(cmd); |
2404 | ||
2405 | if (_IOC_DIR(cmd) & _IOC_WRITE) { | |
2406 | in_iov = iov; | |
2407 | in_iovs = 1; | |
2408 | } | |
2409 | ||
2410 | if (_IOC_DIR(cmd) & _IOC_READ) { | |
2411 | out_iov = iov; | |
2412 | out_iovs = 1; | |
2413 | } | |
2414 | } | |
2415 | ||
2416 | retry: | |
2417 | inarg.in_size = in_size = iov_length(in_iov, in_iovs); | |
2418 | inarg.out_size = out_size = iov_length(out_iov, out_iovs); | |
2419 | ||
2420 | /* | |
2421 | * Out data can be used either for actual out data or iovs, | |
2422 | * make sure there always is at least one page. | |
2423 | */ | |
2424 | out_size = max_t(size_t, out_size, PAGE_SIZE); | |
2425 | max_pages = DIV_ROUND_UP(max(in_size, out_size), PAGE_SIZE); | |
2426 | ||
2427 | /* make sure there are enough buffer pages and init request with them */ | |
2428 | err = -ENOMEM; | |
2429 | if (max_pages > FUSE_MAX_PAGES_PER_REQ) | |
2430 | goto out; | |
2431 | while (num_pages < max_pages) { | |
2432 | pages[num_pages] = alloc_page(GFP_KERNEL | __GFP_HIGHMEM); | |
2433 | if (!pages[num_pages]) | |
2434 | goto out; | |
2435 | num_pages++; | |
2436 | } | |
2437 | ||
54b96670 | 2438 | req = fuse_get_req(fc, num_pages); |
59efec7b TH |
2439 | if (IS_ERR(req)) { |
2440 | err = PTR_ERR(req); | |
2441 | req = NULL; | |
2442 | goto out; | |
2443 | } | |
2444 | memcpy(req->pages, pages, sizeof(req->pages[0]) * num_pages); | |
2445 | req->num_pages = num_pages; | |
7c190c8b | 2446 | fuse_page_descs_length_init(req, 0, req->num_pages); |
59efec7b TH |
2447 | |
2448 | /* okay, let's send it to the client */ | |
2449 | req->in.h.opcode = FUSE_IOCTL; | |
d36f2487 | 2450 | req->in.h.nodeid = ff->nodeid; |
59efec7b TH |
2451 | req->in.numargs = 1; |
2452 | req->in.args[0].size = sizeof(inarg); | |
2453 | req->in.args[0].value = &inarg; | |
2454 | if (in_size) { | |
2455 | req->in.numargs++; | |
2456 | req->in.args[1].size = in_size; | |
2457 | req->in.argpages = 1; | |
2458 | ||
2459 | err = fuse_ioctl_copy_user(pages, in_iov, in_iovs, in_size, | |
2460 | false); | |
2461 | if (err) | |
2462 | goto out; | |
2463 | } | |
2464 | ||
2465 | req->out.numargs = 2; | |
2466 | req->out.args[0].size = sizeof(outarg); | |
2467 | req->out.args[0].value = &outarg; | |
2468 | req->out.args[1].size = out_size; | |
2469 | req->out.argpages = 1; | |
2470 | req->out.argvar = 1; | |
2471 | ||
b93f858a | 2472 | fuse_request_send(fc, req); |
59efec7b TH |
2473 | err = req->out.h.error; |
2474 | transferred = req->out.args[1].size; | |
2475 | fuse_put_request(fc, req); | |
2476 | req = NULL; | |
2477 | if (err) | |
2478 | goto out; | |
2479 | ||
2480 | /* did it ask for retry? */ | |
2481 | if (outarg.flags & FUSE_IOCTL_RETRY) { | |
8ac83505 | 2482 | void *vaddr; |
59efec7b TH |
2483 | |
2484 | /* no retry if in restricted mode */ | |
2485 | err = -EIO; | |
2486 | if (!(flags & FUSE_IOCTL_UNRESTRICTED)) | |
2487 | goto out; | |
2488 | ||
2489 | in_iovs = outarg.in_iovs; | |
2490 | out_iovs = outarg.out_iovs; | |
2491 | ||
2492 | /* | |
2493 | * Make sure things are in boundary, separate checks | |
2494 | * are to protect against overflow. | |
2495 | */ | |
2496 | err = -ENOMEM; | |
2497 | if (in_iovs > FUSE_IOCTL_MAX_IOV || | |
2498 | out_iovs > FUSE_IOCTL_MAX_IOV || | |
2499 | in_iovs + out_iovs > FUSE_IOCTL_MAX_IOV) | |
2500 | goto out; | |
2501 | ||
2408f6ef | 2502 | vaddr = kmap_atomic(pages[0]); |
1baa26b2 | 2503 | err = fuse_copy_ioctl_iovec(fc, iov_page, vaddr, |
d9d318d3 MS |
2504 | transferred, in_iovs + out_iovs, |
2505 | (flags & FUSE_IOCTL_COMPAT) != 0); | |
2408f6ef | 2506 | kunmap_atomic(vaddr); |
d9d318d3 MS |
2507 | if (err) |
2508 | goto out; | |
59efec7b | 2509 | |
8ac83505 | 2510 | in_iov = iov_page; |
59efec7b TH |
2511 | out_iov = in_iov + in_iovs; |
2512 | ||
7572777e MS |
2513 | err = fuse_verify_ioctl_iov(in_iov, in_iovs); |
2514 | if (err) | |
2515 | goto out; | |
2516 | ||
2517 | err = fuse_verify_ioctl_iov(out_iov, out_iovs); | |
2518 | if (err) | |
2519 | goto out; | |
2520 | ||
59efec7b TH |
2521 | goto retry; |
2522 | } | |
2523 | ||
2524 | err = -EIO; | |
2525 | if (transferred > inarg.out_size) | |
2526 | goto out; | |
2527 | ||
2528 | err = fuse_ioctl_copy_user(pages, out_iov, out_iovs, transferred, true); | |
2529 | out: | |
2530 | if (req) | |
2531 | fuse_put_request(fc, req); | |
8ac83505 | 2532 | free_page((unsigned long) iov_page); |
59efec7b TH |
2533 | while (num_pages) |
2534 | __free_page(pages[--num_pages]); | |
2535 | kfree(pages); | |
2536 | ||
2537 | return err ? err : outarg.result; | |
2538 | } | |
08cbf542 | 2539 | EXPORT_SYMBOL_GPL(fuse_do_ioctl); |
59efec7b | 2540 | |
b18da0c5 MS |
2541 | long fuse_ioctl_common(struct file *file, unsigned int cmd, |
2542 | unsigned long arg, unsigned int flags) | |
d36f2487 | 2543 | { |
6131ffaa | 2544 | struct inode *inode = file_inode(file); |
d36f2487 MS |
2545 | struct fuse_conn *fc = get_fuse_conn(inode); |
2546 | ||
c2132c1b | 2547 | if (!fuse_allow_current_process(fc)) |
d36f2487 MS |
2548 | return -EACCES; |
2549 | ||
2550 | if (is_bad_inode(inode)) | |
2551 | return -EIO; | |
2552 | ||
2553 | return fuse_do_ioctl(file, cmd, arg, flags); | |
2554 | } | |
2555 | ||
59efec7b TH |
2556 | static long fuse_file_ioctl(struct file *file, unsigned int cmd, |
2557 | unsigned long arg) | |
2558 | { | |
b18da0c5 | 2559 | return fuse_ioctl_common(file, cmd, arg, 0); |
59efec7b TH |
2560 | } |
2561 | ||
2562 | static long fuse_file_compat_ioctl(struct file *file, unsigned int cmd, | |
2563 | unsigned long arg) | |
2564 | { | |
b18da0c5 | 2565 | return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_COMPAT); |
59efec7b TH |
2566 | } |
2567 | ||
95668a69 TH |
2568 | /* |
2569 | * All files which have been polled are linked to RB tree | |
2570 | * fuse_conn->polled_files which is indexed by kh. Walk the tree and | |
2571 | * find the matching one. | |
2572 | */ | |
2573 | static struct rb_node **fuse_find_polled_node(struct fuse_conn *fc, u64 kh, | |
2574 | struct rb_node **parent_out) | |
2575 | { | |
2576 | struct rb_node **link = &fc->polled_files.rb_node; | |
2577 | struct rb_node *last = NULL; | |
2578 | ||
2579 | while (*link) { | |
2580 | struct fuse_file *ff; | |
2581 | ||
2582 | last = *link; | |
2583 | ff = rb_entry(last, struct fuse_file, polled_node); | |
2584 | ||
2585 | if (kh < ff->kh) | |
2586 | link = &last->rb_left; | |
2587 | else if (kh > ff->kh) | |
2588 | link = &last->rb_right; | |
2589 | else | |
2590 | return link; | |
2591 | } | |
2592 | ||
2593 | if (parent_out) | |
2594 | *parent_out = last; | |
2595 | return link; | |
2596 | } | |
2597 | ||
2598 | /* | |
2599 | * The file is about to be polled. Make sure it's on the polled_files | |
2600 | * RB tree. Note that files once added to the polled_files tree are | |
2601 | * not removed before the file is released. This is because a file | |
2602 | * polled once is likely to be polled again. | |
2603 | */ | |
2604 | static void fuse_register_polled_file(struct fuse_conn *fc, | |
2605 | struct fuse_file *ff) | |
2606 | { | |
2607 | spin_lock(&fc->lock); | |
2608 | if (RB_EMPTY_NODE(&ff->polled_node)) { | |
2609 | struct rb_node **link, *parent; | |
2610 | ||
2611 | link = fuse_find_polled_node(fc, ff->kh, &parent); | |
2612 | BUG_ON(*link); | |
2613 | rb_link_node(&ff->polled_node, parent, link); | |
2614 | rb_insert_color(&ff->polled_node, &fc->polled_files); | |
2615 | } | |
2616 | spin_unlock(&fc->lock); | |
2617 | } | |
2618 | ||
08cbf542 | 2619 | unsigned fuse_file_poll(struct file *file, poll_table *wait) |
95668a69 | 2620 | { |
95668a69 | 2621 | struct fuse_file *ff = file->private_data; |
797759aa | 2622 | struct fuse_conn *fc = ff->fc; |
95668a69 TH |
2623 | struct fuse_poll_in inarg = { .fh = ff->fh, .kh = ff->kh }; |
2624 | struct fuse_poll_out outarg; | |
2625 | struct fuse_req *req; | |
2626 | int err; | |
2627 | ||
2628 | if (fc->no_poll) | |
2629 | return DEFAULT_POLLMASK; | |
2630 | ||
2631 | poll_wait(file, &ff->poll_wait, wait); | |
0415d291 | 2632 | inarg.events = (__u32)poll_requested_events(wait); |
95668a69 TH |
2633 | |
2634 | /* | |
2635 | * Ask for notification iff there's someone waiting for it. | |
2636 | * The client may ignore the flag and always notify. | |
2637 | */ | |
2638 | if (waitqueue_active(&ff->poll_wait)) { | |
2639 | inarg.flags |= FUSE_POLL_SCHEDULE_NOTIFY; | |
2640 | fuse_register_polled_file(fc, ff); | |
2641 | } | |
2642 | ||
b111c8c0 | 2643 | req = fuse_get_req_nopages(fc); |
95668a69 | 2644 | if (IS_ERR(req)) |
201fa69a | 2645 | return POLLERR; |
95668a69 TH |
2646 | |
2647 | req->in.h.opcode = FUSE_POLL; | |
797759aa | 2648 | req->in.h.nodeid = ff->nodeid; |
95668a69 TH |
2649 | req->in.numargs = 1; |
2650 | req->in.args[0].size = sizeof(inarg); | |
2651 | req->in.args[0].value = &inarg; | |
2652 | req->out.numargs = 1; | |
2653 | req->out.args[0].size = sizeof(outarg); | |
2654 | req->out.args[0].value = &outarg; | |
b93f858a | 2655 | fuse_request_send(fc, req); |
95668a69 TH |
2656 | err = req->out.h.error; |
2657 | fuse_put_request(fc, req); | |
2658 | ||
2659 | if (!err) | |
2660 | return outarg.revents; | |
2661 | if (err == -ENOSYS) { | |
2662 | fc->no_poll = 1; | |
2663 | return DEFAULT_POLLMASK; | |
2664 | } | |
2665 | return POLLERR; | |
2666 | } | |
08cbf542 | 2667 | EXPORT_SYMBOL_GPL(fuse_file_poll); |
95668a69 TH |
2668 | |
2669 | /* | |
2670 | * This is called from fuse_handle_notify() on FUSE_NOTIFY_POLL and | |
2671 | * wakes up the poll waiters. | |
2672 | */ | |
2673 | int fuse_notify_poll_wakeup(struct fuse_conn *fc, | |
2674 | struct fuse_notify_poll_wakeup_out *outarg) | |
2675 | { | |
2676 | u64 kh = outarg->kh; | |
2677 | struct rb_node **link; | |
2678 | ||
2679 | spin_lock(&fc->lock); | |
2680 | ||
2681 | link = fuse_find_polled_node(fc, kh, NULL); | |
2682 | if (*link) { | |
2683 | struct fuse_file *ff; | |
2684 | ||
2685 | ff = rb_entry(*link, struct fuse_file, polled_node); | |
2686 | wake_up_interruptible_sync(&ff->poll_wait); | |
2687 | } | |
2688 | ||
2689 | spin_unlock(&fc->lock); | |
2690 | return 0; | |
2691 | } | |
2692 | ||
efb9fa9e MP |
2693 | static void fuse_do_truncate(struct file *file) |
2694 | { | |
2695 | struct inode *inode = file->f_mapping->host; | |
2696 | struct iattr attr; | |
2697 | ||
2698 | attr.ia_valid = ATTR_SIZE; | |
2699 | attr.ia_size = i_size_read(inode); | |
2700 | ||
2701 | attr.ia_file = file; | |
2702 | attr.ia_valid |= ATTR_FILE; | |
2703 | ||
2704 | fuse_do_setattr(inode, &attr, file); | |
2705 | } | |
2706 | ||
e5c5f05d MP |
2707 | static inline loff_t fuse_round_up(loff_t off) |
2708 | { | |
2709 | return round_up(off, FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT); | |
2710 | } | |
2711 | ||
4273b793 AA |
2712 | static ssize_t |
2713 | fuse_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, | |
2714 | loff_t offset, unsigned long nr_segs) | |
2715 | { | |
2716 | ssize_t ret = 0; | |
60b9df7a MS |
2717 | struct file *file = iocb->ki_filp; |
2718 | struct fuse_file *ff = file->private_data; | |
e5c5f05d | 2719 | bool async_dio = ff->fc->async_dio; |
4273b793 | 2720 | loff_t pos = 0; |
bcba24cc MP |
2721 | struct inode *inode; |
2722 | loff_t i_size; | |
2723 | size_t count = iov_length(iov, nr_segs); | |
36cf66ed | 2724 | struct fuse_io_priv *io; |
4273b793 | 2725 | |
4273b793 | 2726 | pos = offset; |
bcba24cc MP |
2727 | inode = file->f_mapping->host; |
2728 | i_size = i_size_read(inode); | |
4273b793 | 2729 | |
9fe55eea SW |
2730 | if ((rw == READ) && (offset > i_size)) |
2731 | return 0; | |
2732 | ||
439ee5f0 | 2733 | /* optimization for short read */ |
e5c5f05d | 2734 | if (async_dio && rw != WRITE && offset + count > i_size) { |
439ee5f0 MP |
2735 | if (offset >= i_size) |
2736 | return 0; | |
e5c5f05d | 2737 | count = min_t(loff_t, count, fuse_round_up(i_size - offset)); |
439ee5f0 MP |
2738 | } |
2739 | ||
bcba24cc | 2740 | io = kmalloc(sizeof(struct fuse_io_priv), GFP_KERNEL); |
36cf66ed MP |
2741 | if (!io) |
2742 | return -ENOMEM; | |
bcba24cc MP |
2743 | spin_lock_init(&io->lock); |
2744 | io->reqs = 1; | |
2745 | io->bytes = -1; | |
2746 | io->size = 0; | |
2747 | io->offset = offset; | |
2748 | io->write = (rw == WRITE); | |
2749 | io->err = 0; | |
36cf66ed | 2750 | io->file = file; |
bcba24cc MP |
2751 | /* |
2752 | * By default, we want to optimize all I/Os with async request | |
60b9df7a | 2753 | * submission to the client filesystem if supported. |
bcba24cc | 2754 | */ |
e5c5f05d | 2755 | io->async = async_dio; |
bcba24cc MP |
2756 | io->iocb = iocb; |
2757 | ||
2758 | /* | |
2759 | * We cannot asynchronously extend the size of a file. We have no method | |
2760 | * to wait on real async I/O requests, so we must submit this request | |
2761 | * synchronously. | |
2762 | */ | |
e5c5f05d | 2763 | if (!is_sync_kiocb(iocb) && (offset + count > i_size) && rw == WRITE) |
bcba24cc | 2764 | io->async = false; |
4273b793 | 2765 | |
b98d023a | 2766 | if (rw == WRITE) |
36cf66ed | 2767 | ret = __fuse_direct_write(io, iov, nr_segs, &pos); |
b98d023a | 2768 | else |
439ee5f0 | 2769 | ret = __fuse_direct_read(io, iov, nr_segs, &pos, count); |
36cf66ed | 2770 | |
bcba24cc MP |
2771 | if (io->async) { |
2772 | fuse_aio_complete(io, ret < 0 ? ret : 0, -1); | |
2773 | ||
2774 | /* we have a non-extending, async request, so return */ | |
c9ecf989 | 2775 | if (!is_sync_kiocb(iocb)) |
bcba24cc MP |
2776 | return -EIOCBQUEUED; |
2777 | ||
2778 | ret = wait_on_sync_kiocb(iocb); | |
2779 | } else { | |
2780 | kfree(io); | |
2781 | } | |
2782 | ||
efb9fa9e MP |
2783 | if (rw == WRITE) { |
2784 | if (ret > 0) | |
2785 | fuse_write_update_size(inode, pos); | |
2786 | else if (ret < 0 && offset + count > i_size) | |
2787 | fuse_do_truncate(file); | |
2788 | } | |
4273b793 AA |
2789 | |
2790 | return ret; | |
2791 | } | |
2792 | ||
cdadb11c MS |
2793 | static long fuse_file_fallocate(struct file *file, int mode, loff_t offset, |
2794 | loff_t length) | |
05ba1f08 AP |
2795 | { |
2796 | struct fuse_file *ff = file->private_data; | |
3634a632 | 2797 | struct inode *inode = file->f_inode; |
0ab08f57 | 2798 | struct fuse_inode *fi = get_fuse_inode(inode); |
05ba1f08 AP |
2799 | struct fuse_conn *fc = ff->fc; |
2800 | struct fuse_req *req; | |
2801 | struct fuse_fallocate_in inarg = { | |
2802 | .fh = ff->fh, | |
2803 | .offset = offset, | |
2804 | .length = length, | |
2805 | .mode = mode | |
2806 | }; | |
2807 | int err; | |
14c14414 MP |
2808 | bool lock_inode = !(mode & FALLOC_FL_KEEP_SIZE) || |
2809 | (mode & FALLOC_FL_PUNCH_HOLE); | |
05ba1f08 | 2810 | |
519c6040 MS |
2811 | if (fc->no_fallocate) |
2812 | return -EOPNOTSUPP; | |
2813 | ||
14c14414 | 2814 | if (lock_inode) { |
3634a632 | 2815 | mutex_lock(&inode->i_mutex); |
bde52788 MP |
2816 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
2817 | loff_t endbyte = offset + length - 1; | |
2818 | err = filemap_write_and_wait_range(inode->i_mapping, | |
2819 | offset, endbyte); | |
2820 | if (err) | |
2821 | goto out; | |
2822 | ||
2823 | fuse_sync_writes(inode); | |
2824 | } | |
3634a632 BF |
2825 | } |
2826 | ||
0ab08f57 MP |
2827 | if (!(mode & FALLOC_FL_KEEP_SIZE)) |
2828 | set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); | |
2829 | ||
b111c8c0 | 2830 | req = fuse_get_req_nopages(fc); |
3634a632 BF |
2831 | if (IS_ERR(req)) { |
2832 | err = PTR_ERR(req); | |
2833 | goto out; | |
2834 | } | |
05ba1f08 AP |
2835 | |
2836 | req->in.h.opcode = FUSE_FALLOCATE; | |
2837 | req->in.h.nodeid = ff->nodeid; | |
2838 | req->in.numargs = 1; | |
2839 | req->in.args[0].size = sizeof(inarg); | |
2840 | req->in.args[0].value = &inarg; | |
2841 | fuse_request_send(fc, req); | |
2842 | err = req->out.h.error; | |
519c6040 MS |
2843 | if (err == -ENOSYS) { |
2844 | fc->no_fallocate = 1; | |
2845 | err = -EOPNOTSUPP; | |
2846 | } | |
05ba1f08 AP |
2847 | fuse_put_request(fc, req); |
2848 | ||
bee6c307 BF |
2849 | if (err) |
2850 | goto out; | |
2851 | ||
2852 | /* we could have extended the file */ | |
2853 | if (!(mode & FALLOC_FL_KEEP_SIZE)) | |
2854 | fuse_write_update_size(inode, offset + length); | |
2855 | ||
2856 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
2857 | truncate_pagecache_range(inode, offset, offset + length - 1); | |
2858 | ||
2859 | fuse_invalidate_attr(inode); | |
2860 | ||
3634a632 | 2861 | out: |
0ab08f57 MP |
2862 | if (!(mode & FALLOC_FL_KEEP_SIZE)) |
2863 | clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); | |
2864 | ||
bde52788 | 2865 | if (lock_inode) |
3634a632 | 2866 | mutex_unlock(&inode->i_mutex); |
3634a632 | 2867 | |
05ba1f08 AP |
2868 | return err; |
2869 | } | |
05ba1f08 | 2870 | |
4b6f5d20 | 2871 | static const struct file_operations fuse_file_operations = { |
5559b8f4 | 2872 | .llseek = fuse_file_llseek, |
543ade1f | 2873 | .read = do_sync_read, |
bcb4be80 | 2874 | .aio_read = fuse_file_aio_read, |
543ade1f | 2875 | .write = do_sync_write, |
ea9b9907 | 2876 | .aio_write = fuse_file_aio_write, |
b6aeaded MS |
2877 | .mmap = fuse_file_mmap, |
2878 | .open = fuse_open, | |
2879 | .flush = fuse_flush, | |
2880 | .release = fuse_release, | |
2881 | .fsync = fuse_fsync, | |
71421259 | 2882 | .lock = fuse_file_lock, |
a9ff4f87 | 2883 | .flock = fuse_file_flock, |
5ffc4ef4 | 2884 | .splice_read = generic_file_splice_read, |
59efec7b TH |
2885 | .unlocked_ioctl = fuse_file_ioctl, |
2886 | .compat_ioctl = fuse_file_compat_ioctl, | |
95668a69 | 2887 | .poll = fuse_file_poll, |
05ba1f08 | 2888 | .fallocate = fuse_file_fallocate, |
b6aeaded MS |
2889 | }; |
2890 | ||
4b6f5d20 | 2891 | static const struct file_operations fuse_direct_io_file_operations = { |
5559b8f4 | 2892 | .llseek = fuse_file_llseek, |
413ef8cb MS |
2893 | .read = fuse_direct_read, |
2894 | .write = fuse_direct_write, | |
fc280c96 | 2895 | .mmap = fuse_direct_mmap, |
413ef8cb MS |
2896 | .open = fuse_open, |
2897 | .flush = fuse_flush, | |
2898 | .release = fuse_release, | |
2899 | .fsync = fuse_fsync, | |
71421259 | 2900 | .lock = fuse_file_lock, |
a9ff4f87 | 2901 | .flock = fuse_file_flock, |
59efec7b TH |
2902 | .unlocked_ioctl = fuse_file_ioctl, |
2903 | .compat_ioctl = fuse_file_compat_ioctl, | |
95668a69 | 2904 | .poll = fuse_file_poll, |
05ba1f08 | 2905 | .fallocate = fuse_file_fallocate, |
fc280c96 | 2906 | /* no splice_read */ |
413ef8cb MS |
2907 | }; |
2908 | ||
f5e54d6e | 2909 | static const struct address_space_operations fuse_file_aops = { |
b6aeaded | 2910 | .readpage = fuse_readpage, |
3be5a52b | 2911 | .writepage = fuse_writepage, |
26d614df | 2912 | .writepages = fuse_writepages, |
3be5a52b | 2913 | .launder_page = fuse_launder_page, |
db50b96c | 2914 | .readpages = fuse_readpages, |
3be5a52b | 2915 | .set_page_dirty = __set_page_dirty_nobuffers, |
b2d2272f | 2916 | .bmap = fuse_bmap, |
4273b793 | 2917 | .direct_IO = fuse_direct_IO, |
b6aeaded MS |
2918 | }; |
2919 | ||
2920 | void fuse_init_file_inode(struct inode *inode) | |
2921 | { | |
45323fb7 MS |
2922 | inode->i_fop = &fuse_file_operations; |
2923 | inode->i_data.a_ops = &fuse_file_aops; | |
b6aeaded | 2924 | } |