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> |
b6aeaded | 15 | |
4b6f5d20 | 16 | static const struct file_operations fuse_direct_io_file_operations; |
45323fb7 | 17 | |
fd72faac MS |
18 | static int fuse_send_open(struct inode *inode, struct file *file, int isdir, |
19 | struct fuse_open_out *outargp) | |
b6aeaded MS |
20 | { |
21 | struct fuse_conn *fc = get_fuse_conn(inode); | |
b6aeaded | 22 | struct fuse_open_in inarg; |
fd72faac MS |
23 | struct fuse_req *req; |
24 | int err; | |
25 | ||
ce1d5a49 MS |
26 | req = fuse_get_req(fc); |
27 | if (IS_ERR(req)) | |
28 | return PTR_ERR(req); | |
fd72faac MS |
29 | |
30 | memset(&inarg, 0, sizeof(inarg)); | |
6ff958ed MS |
31 | inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY); |
32 | if (!fc->atomic_o_trunc) | |
33 | inarg.flags &= ~O_TRUNC; | |
fd72faac MS |
34 | req->in.h.opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN; |
35 | req->in.h.nodeid = get_node_id(inode); | |
fd72faac MS |
36 | req->in.numargs = 1; |
37 | req->in.args[0].size = sizeof(inarg); | |
38 | req->in.args[0].value = &inarg; | |
39 | req->out.numargs = 1; | |
40 | req->out.args[0].size = sizeof(*outargp); | |
41 | req->out.args[0].value = outargp; | |
b93f858a | 42 | fuse_request_send(fc, req); |
fd72faac MS |
43 | err = req->out.h.error; |
44 | fuse_put_request(fc, req); | |
45 | ||
46 | return err; | |
47 | } | |
48 | ||
acf99433 | 49 | struct fuse_file *fuse_file_alloc(struct fuse_conn *fc) |
fd72faac MS |
50 | { |
51 | struct fuse_file *ff; | |
6b2db28a | 52 | |
fd72faac | 53 | ff = kmalloc(sizeof(struct fuse_file), GFP_KERNEL); |
6b2db28a TH |
54 | if (unlikely(!ff)) |
55 | return NULL; | |
56 | ||
57 | ff->reserved_req = fuse_request_alloc(); | |
58 | if (unlikely(!ff->reserved_req)) { | |
59 | kfree(ff); | |
60 | return NULL; | |
fd72faac | 61 | } |
6b2db28a TH |
62 | |
63 | INIT_LIST_HEAD(&ff->write_entry); | |
64 | atomic_set(&ff->count, 0); | |
65 | RB_CLEAR_NODE(&ff->polled_node); | |
66 | init_waitqueue_head(&ff->poll_wait); | |
67 | ||
68 | spin_lock(&fc->lock); | |
69 | ff->kh = ++fc->khctr; | |
70 | spin_unlock(&fc->lock); | |
71 | ||
fd72faac MS |
72 | return ff; |
73 | } | |
74 | ||
75 | void fuse_file_free(struct fuse_file *ff) | |
76 | { | |
33649c91 | 77 | fuse_request_free(ff->reserved_req); |
fd72faac MS |
78 | kfree(ff); |
79 | } | |
80 | ||
c756e0a4 MS |
81 | static struct fuse_file *fuse_file_get(struct fuse_file *ff) |
82 | { | |
83 | atomic_inc(&ff->count); | |
84 | return ff; | |
85 | } | |
86 | ||
819c4b3b MS |
87 | static void fuse_release_end(struct fuse_conn *fc, struct fuse_req *req) |
88 | { | |
b0be46eb | 89 | path_put(&req->misc.release.path); |
819c4b3b MS |
90 | } |
91 | ||
c756e0a4 MS |
92 | static void fuse_file_put(struct fuse_file *ff) |
93 | { | |
94 | if (atomic_dec_and_test(&ff->count)) { | |
95 | struct fuse_req *req = ff->reserved_req; | |
b0be46eb | 96 | struct inode *inode = req->misc.release.path.dentry->d_inode; |
b57d4264 | 97 | struct fuse_conn *fc = get_fuse_conn(inode); |
819c4b3b | 98 | req->end = fuse_release_end; |
b93f858a | 99 | fuse_request_send_background(fc, req); |
c756e0a4 MS |
100 | kfree(ff); |
101 | } | |
102 | } | |
103 | ||
fd72faac MS |
104 | void fuse_finish_open(struct inode *inode, struct file *file, |
105 | struct fuse_file *ff, struct fuse_open_out *outarg) | |
106 | { | |
107 | if (outarg->open_flags & FOPEN_DIRECT_IO) | |
108 | file->f_op = &fuse_direct_io_file_operations; | |
109 | if (!(outarg->open_flags & FOPEN_KEEP_CACHE)) | |
b1009979 | 110 | invalidate_inode_pages2(inode->i_mapping); |
a7c1b990 TH |
111 | if (outarg->open_flags & FOPEN_NONSEEKABLE) |
112 | nonseekable_open(inode, file); | |
fd72faac | 113 | ff->fh = outarg->fh; |
c756e0a4 | 114 | file->private_data = fuse_file_get(ff); |
fd72faac MS |
115 | } |
116 | ||
117 | int fuse_open_common(struct inode *inode, struct file *file, int isdir) | |
118 | { | |
acf99433 | 119 | struct fuse_conn *fc = get_fuse_conn(inode); |
b6aeaded MS |
120 | struct fuse_open_out outarg; |
121 | struct fuse_file *ff; | |
122 | int err; | |
b6aeaded | 123 | |
dd190d06 MS |
124 | /* VFS checks this, but only _after_ ->open() */ |
125 | if (file->f_flags & O_DIRECT) | |
126 | return -EINVAL; | |
127 | ||
b6aeaded MS |
128 | err = generic_file_open(inode, file); |
129 | if (err) | |
130 | return err; | |
131 | ||
acf99433 | 132 | ff = fuse_file_alloc(fc); |
b6aeaded | 133 | if (!ff) |
fd72faac | 134 | return -ENOMEM; |
b6aeaded | 135 | |
fd72faac MS |
136 | err = fuse_send_open(inode, file, isdir, &outarg); |
137 | if (err) | |
138 | fuse_file_free(ff); | |
139 | else { | |
140 | if (isdir) | |
141 | outarg.open_flags &= ~FOPEN_DIRECT_IO; | |
142 | fuse_finish_open(inode, file, ff, &outarg); | |
b6aeaded MS |
143 | } |
144 | ||
b6aeaded MS |
145 | return err; |
146 | } | |
147 | ||
c756e0a4 | 148 | void fuse_release_fill(struct fuse_file *ff, u64 nodeid, int flags, int opcode) |
64c6d8ed | 149 | { |
33649c91 | 150 | struct fuse_req *req = ff->reserved_req; |
b57d4264 | 151 | struct fuse_release_in *inarg = &req->misc.release.in; |
b6aeaded MS |
152 | |
153 | inarg->fh = ff->fh; | |
fd72faac | 154 | inarg->flags = flags; |
51eb01e7 | 155 | req->in.h.opcode = opcode; |
fd72faac | 156 | req->in.h.nodeid = nodeid; |
b6aeaded MS |
157 | req->in.numargs = 1; |
158 | req->in.args[0].size = sizeof(struct fuse_release_in); | |
159 | req->in.args[0].value = inarg; | |
fd72faac MS |
160 | } |
161 | ||
162 | int fuse_release_common(struct inode *inode, struct file *file, int isdir) | |
163 | { | |
6b2db28a TH |
164 | struct fuse_conn *fc; |
165 | struct fuse_file *ff; | |
166 | struct fuse_req *req; | |
b6aeaded | 167 | |
6b2db28a TH |
168 | ff = file->private_data; |
169 | if (unlikely(!ff)) | |
170 | return 0; /* return value is ignored by VFS */ | |
171 | ||
172 | fc = get_fuse_conn(inode); | |
173 | req = ff->reserved_req; | |
174 | ||
175 | fuse_release_fill(ff, get_node_id(inode), file->f_flags, | |
176 | isdir ? FUSE_RELEASEDIR : FUSE_RELEASE); | |
177 | ||
178 | /* Hold vfsmount and dentry until release is finished */ | |
b0be46eb MS |
179 | path_get(&file->f_path); |
180 | req->misc.release.path = file->f_path; | |
6b2db28a TH |
181 | |
182 | spin_lock(&fc->lock); | |
183 | list_del(&ff->write_entry); | |
184 | if (!RB_EMPTY_NODE(&ff->polled_node)) | |
185 | rb_erase(&ff->polled_node, &fc->polled_files); | |
186 | spin_unlock(&fc->lock); | |
187 | ||
188 | wake_up_interruptible_sync(&ff->poll_wait); | |
189 | /* | |
190 | * Normally this will send the RELEASE request, however if | |
191 | * some asynchronous READ or WRITE requests are outstanding, | |
192 | * the sending will be delayed. | |
193 | */ | |
194 | fuse_file_put(ff); | |
b6aeaded MS |
195 | return 0; |
196 | } | |
197 | ||
04730fef MS |
198 | static int fuse_open(struct inode *inode, struct file *file) |
199 | { | |
200 | return fuse_open_common(inode, file, 0); | |
201 | } | |
202 | ||
203 | static int fuse_release(struct inode *inode, struct file *file) | |
204 | { | |
205 | return fuse_release_common(inode, file, 0); | |
206 | } | |
207 | ||
71421259 | 208 | /* |
9c8ef561 MS |
209 | * Scramble the ID space with XTEA, so that the value of the files_struct |
210 | * pointer is not exposed to userspace. | |
71421259 | 211 | */ |
f3332114 | 212 | u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id) |
71421259 | 213 | { |
9c8ef561 MS |
214 | u32 *k = fc->scramble_key; |
215 | u64 v = (unsigned long) id; | |
216 | u32 v0 = v; | |
217 | u32 v1 = v >> 32; | |
218 | u32 sum = 0; | |
219 | int i; | |
220 | ||
221 | for (i = 0; i < 32; i++) { | |
222 | v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]); | |
223 | sum += 0x9E3779B9; | |
224 | v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]); | |
225 | } | |
226 | ||
227 | return (u64) v0 + ((u64) v1 << 32); | |
71421259 MS |
228 | } |
229 | ||
3be5a52b MS |
230 | /* |
231 | * Check if page is under writeback | |
232 | * | |
233 | * This is currently done by walking the list of writepage requests | |
234 | * for the inode, which can be pretty inefficient. | |
235 | */ | |
236 | static bool fuse_page_is_writeback(struct inode *inode, pgoff_t index) | |
237 | { | |
238 | struct fuse_conn *fc = get_fuse_conn(inode); | |
239 | struct fuse_inode *fi = get_fuse_inode(inode); | |
240 | struct fuse_req *req; | |
241 | bool found = false; | |
242 | ||
243 | spin_lock(&fc->lock); | |
244 | list_for_each_entry(req, &fi->writepages, writepages_entry) { | |
245 | pgoff_t curr_index; | |
246 | ||
247 | BUG_ON(req->inode != inode); | |
248 | curr_index = req->misc.write.in.offset >> PAGE_CACHE_SHIFT; | |
249 | if (curr_index == index) { | |
250 | found = true; | |
251 | break; | |
252 | } | |
253 | } | |
254 | spin_unlock(&fc->lock); | |
255 | ||
256 | return found; | |
257 | } | |
258 | ||
259 | /* | |
260 | * Wait for page writeback to be completed. | |
261 | * | |
262 | * Since fuse doesn't rely on the VM writeback tracking, this has to | |
263 | * use some other means. | |
264 | */ | |
265 | static int fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index) | |
266 | { | |
267 | struct fuse_inode *fi = get_fuse_inode(inode); | |
268 | ||
269 | wait_event(fi->page_waitq, !fuse_page_is_writeback(inode, index)); | |
270 | return 0; | |
271 | } | |
272 | ||
75e1fcc0 | 273 | static int fuse_flush(struct file *file, fl_owner_t id) |
b6aeaded | 274 | { |
7706a9d6 | 275 | struct inode *inode = file->f_path.dentry->d_inode; |
b6aeaded MS |
276 | struct fuse_conn *fc = get_fuse_conn(inode); |
277 | struct fuse_file *ff = file->private_data; | |
278 | struct fuse_req *req; | |
279 | struct fuse_flush_in inarg; | |
280 | int err; | |
281 | ||
248d86e8 MS |
282 | if (is_bad_inode(inode)) |
283 | return -EIO; | |
284 | ||
b6aeaded MS |
285 | if (fc->no_flush) |
286 | return 0; | |
287 | ||
33649c91 | 288 | req = fuse_get_req_nofail(fc, file); |
b6aeaded MS |
289 | memset(&inarg, 0, sizeof(inarg)); |
290 | inarg.fh = ff->fh; | |
9c8ef561 | 291 | inarg.lock_owner = fuse_lock_owner_id(fc, id); |
b6aeaded MS |
292 | req->in.h.opcode = FUSE_FLUSH; |
293 | req->in.h.nodeid = get_node_id(inode); | |
b6aeaded MS |
294 | req->in.numargs = 1; |
295 | req->in.args[0].size = sizeof(inarg); | |
296 | req->in.args[0].value = &inarg; | |
71421259 | 297 | req->force = 1; |
b93f858a | 298 | fuse_request_send(fc, req); |
b6aeaded MS |
299 | err = req->out.h.error; |
300 | fuse_put_request(fc, req); | |
301 | if (err == -ENOSYS) { | |
302 | fc->no_flush = 1; | |
303 | err = 0; | |
304 | } | |
305 | return err; | |
306 | } | |
307 | ||
3be5a52b MS |
308 | /* |
309 | * Wait for all pending writepages on the inode to finish. | |
310 | * | |
311 | * This is currently done by blocking further writes with FUSE_NOWRITE | |
312 | * and waiting for all sent writes to complete. | |
313 | * | |
314 | * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage | |
315 | * could conflict with truncation. | |
316 | */ | |
317 | static void fuse_sync_writes(struct inode *inode) | |
318 | { | |
319 | fuse_set_nowrite(inode); | |
320 | fuse_release_nowrite(inode); | |
321 | } | |
322 | ||
82547981 MS |
323 | int fuse_fsync_common(struct file *file, struct dentry *de, int datasync, |
324 | int isdir) | |
b6aeaded MS |
325 | { |
326 | struct inode *inode = de->d_inode; | |
327 | struct fuse_conn *fc = get_fuse_conn(inode); | |
328 | struct fuse_file *ff = file->private_data; | |
329 | struct fuse_req *req; | |
330 | struct fuse_fsync_in inarg; | |
331 | int err; | |
332 | ||
248d86e8 MS |
333 | if (is_bad_inode(inode)) |
334 | return -EIO; | |
335 | ||
82547981 | 336 | if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir)) |
b6aeaded MS |
337 | return 0; |
338 | ||
3be5a52b MS |
339 | /* |
340 | * Start writeback against all dirty pages of the inode, then | |
341 | * wait for all outstanding writes, before sending the FSYNC | |
342 | * request. | |
343 | */ | |
344 | err = write_inode_now(inode, 0); | |
345 | if (err) | |
346 | return err; | |
347 | ||
348 | fuse_sync_writes(inode); | |
349 | ||
ce1d5a49 MS |
350 | req = fuse_get_req(fc); |
351 | if (IS_ERR(req)) | |
352 | return PTR_ERR(req); | |
b6aeaded MS |
353 | |
354 | memset(&inarg, 0, sizeof(inarg)); | |
355 | inarg.fh = ff->fh; | |
356 | inarg.fsync_flags = datasync ? 1 : 0; | |
82547981 | 357 | req->in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC; |
b6aeaded | 358 | req->in.h.nodeid = get_node_id(inode); |
b6aeaded MS |
359 | req->in.numargs = 1; |
360 | req->in.args[0].size = sizeof(inarg); | |
361 | req->in.args[0].value = &inarg; | |
b93f858a | 362 | fuse_request_send(fc, req); |
b6aeaded MS |
363 | err = req->out.h.error; |
364 | fuse_put_request(fc, req); | |
365 | if (err == -ENOSYS) { | |
82547981 MS |
366 | if (isdir) |
367 | fc->no_fsyncdir = 1; | |
368 | else | |
369 | fc->no_fsync = 1; | |
b6aeaded MS |
370 | err = 0; |
371 | } | |
372 | return err; | |
373 | } | |
374 | ||
82547981 MS |
375 | static int fuse_fsync(struct file *file, struct dentry *de, int datasync) |
376 | { | |
377 | return fuse_fsync_common(file, de, datasync, 0); | |
378 | } | |
379 | ||
a6643094 | 380 | void fuse_read_fill(struct fuse_req *req, struct file *file, |
361b1eb5 | 381 | struct inode *inode, loff_t pos, size_t count, int opcode) |
b6aeaded | 382 | { |
5c5c5e51 | 383 | struct fuse_read_in *inarg = &req->misc.read.in; |
a6643094 | 384 | struct fuse_file *ff = file->private_data; |
b6aeaded | 385 | |
361b1eb5 MS |
386 | inarg->fh = ff->fh; |
387 | inarg->offset = pos; | |
388 | inarg->size = count; | |
a6643094 | 389 | inarg->flags = file->f_flags; |
361b1eb5 | 390 | req->in.h.opcode = opcode; |
b6aeaded | 391 | req->in.h.nodeid = get_node_id(inode); |
b6aeaded MS |
392 | req->in.numargs = 1; |
393 | req->in.args[0].size = sizeof(struct fuse_read_in); | |
c1aa96a5 | 394 | req->in.args[0].value = inarg; |
b6aeaded MS |
395 | req->out.argvar = 1; |
396 | req->out.numargs = 1; | |
397 | req->out.args[0].size = count; | |
b6aeaded MS |
398 | } |
399 | ||
8bfc016d | 400 | static size_t fuse_send_read(struct fuse_req *req, struct file *file, |
f3332114 MS |
401 | struct inode *inode, loff_t pos, size_t count, |
402 | fl_owner_t owner) | |
04730fef | 403 | { |
361b1eb5 | 404 | struct fuse_conn *fc = get_fuse_conn(inode); |
f3332114 | 405 | |
a6643094 | 406 | fuse_read_fill(req, file, inode, pos, count, FUSE_READ); |
f3332114 | 407 | if (owner != NULL) { |
5c5c5e51 | 408 | struct fuse_read_in *inarg = &req->misc.read.in; |
f3332114 MS |
409 | |
410 | inarg->read_flags |= FUSE_READ_LOCKOWNER; | |
411 | inarg->lock_owner = fuse_lock_owner_id(fc, owner); | |
412 | } | |
b93f858a | 413 | fuse_request_send(fc, req); |
361b1eb5 | 414 | return req->out.args[0].size; |
04730fef MS |
415 | } |
416 | ||
5c5c5e51 MS |
417 | static void fuse_read_update_size(struct inode *inode, loff_t size, |
418 | u64 attr_ver) | |
419 | { | |
420 | struct fuse_conn *fc = get_fuse_conn(inode); | |
421 | struct fuse_inode *fi = get_fuse_inode(inode); | |
422 | ||
423 | spin_lock(&fc->lock); | |
424 | if (attr_ver == fi->attr_version && size < inode->i_size) { | |
425 | fi->attr_version = ++fc->attr_version; | |
426 | i_size_write(inode, size); | |
427 | } | |
428 | spin_unlock(&fc->lock); | |
429 | } | |
430 | ||
b6aeaded MS |
431 | static int fuse_readpage(struct file *file, struct page *page) |
432 | { | |
433 | struct inode *inode = page->mapping->host; | |
434 | struct fuse_conn *fc = get_fuse_conn(inode); | |
248d86e8 | 435 | struct fuse_req *req; |
5c5c5e51 MS |
436 | size_t num_read; |
437 | loff_t pos = page_offset(page); | |
438 | size_t count = PAGE_CACHE_SIZE; | |
439 | u64 attr_ver; | |
248d86e8 MS |
440 | int err; |
441 | ||
442 | err = -EIO; | |
443 | if (is_bad_inode(inode)) | |
444 | goto out; | |
445 | ||
3be5a52b MS |
446 | /* |
447 | * Page writeback can extend beyond the liftime of the | |
448 | * page-cache page, so make sure we read a properly synced | |
449 | * page. | |
450 | */ | |
451 | fuse_wait_on_page_writeback(inode, page->index); | |
452 | ||
ce1d5a49 MS |
453 | req = fuse_get_req(fc); |
454 | err = PTR_ERR(req); | |
455 | if (IS_ERR(req)) | |
b6aeaded MS |
456 | goto out; |
457 | ||
5c5c5e51 MS |
458 | attr_ver = fuse_get_attr_version(fc); |
459 | ||
b6aeaded | 460 | req->out.page_zeroing = 1; |
f4975c67 | 461 | req->out.argpages = 1; |
b6aeaded MS |
462 | req->num_pages = 1; |
463 | req->pages[0] = page; | |
5c5c5e51 | 464 | num_read = fuse_send_read(req, file, inode, pos, count, NULL); |
b6aeaded MS |
465 | err = req->out.h.error; |
466 | fuse_put_request(fc, req); | |
5c5c5e51 MS |
467 | |
468 | if (!err) { | |
469 | /* | |
470 | * Short read means EOF. If file size is larger, truncate it | |
471 | */ | |
472 | if (num_read < count) | |
473 | fuse_read_update_size(inode, pos + num_read, attr_ver); | |
474 | ||
b6aeaded | 475 | SetPageUptodate(page); |
5c5c5e51 MS |
476 | } |
477 | ||
b36c31ba | 478 | fuse_invalidate_attr(inode); /* atime changed */ |
b6aeaded MS |
479 | out: |
480 | unlock_page(page); | |
481 | return err; | |
482 | } | |
483 | ||
c1aa96a5 | 484 | static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req) |
db50b96c | 485 | { |
c1aa96a5 | 486 | int i; |
5c5c5e51 MS |
487 | size_t count = req->misc.read.in.size; |
488 | size_t num_read = req->out.args[0].size; | |
489 | struct inode *inode = req->pages[0]->mapping->host; | |
c1aa96a5 | 490 | |
5c5c5e51 MS |
491 | /* |
492 | * Short read means EOF. If file size is larger, truncate it | |
493 | */ | |
494 | if (!req->out.h.error && num_read < count) { | |
495 | loff_t pos = page_offset(req->pages[0]) + num_read; | |
496 | fuse_read_update_size(inode, pos, req->misc.read.attr_ver); | |
497 | } | |
498 | ||
499 | fuse_invalidate_attr(inode); /* atime changed */ | |
c1aa96a5 | 500 | |
db50b96c MS |
501 | for (i = 0; i < req->num_pages; i++) { |
502 | struct page *page = req->pages[i]; | |
503 | if (!req->out.h.error) | |
504 | SetPageUptodate(page); | |
c1aa96a5 MS |
505 | else |
506 | SetPageError(page); | |
db50b96c MS |
507 | unlock_page(page); |
508 | } | |
c756e0a4 MS |
509 | if (req->ff) |
510 | fuse_file_put(req->ff); | |
c1aa96a5 MS |
511 | } |
512 | ||
a6643094 | 513 | static void fuse_send_readpages(struct fuse_req *req, struct file *file, |
c1aa96a5 MS |
514 | struct inode *inode) |
515 | { | |
516 | struct fuse_conn *fc = get_fuse_conn(inode); | |
517 | loff_t pos = page_offset(req->pages[0]); | |
518 | size_t count = req->num_pages << PAGE_CACHE_SHIFT; | |
f4975c67 MS |
519 | |
520 | req->out.argpages = 1; | |
c1aa96a5 | 521 | req->out.page_zeroing = 1; |
a6643094 | 522 | fuse_read_fill(req, file, inode, pos, count, FUSE_READ); |
5c5c5e51 | 523 | req->misc.read.attr_ver = fuse_get_attr_version(fc); |
9cd68455 | 524 | if (fc->async_read) { |
a6643094 | 525 | struct fuse_file *ff = file->private_data; |
c756e0a4 | 526 | req->ff = fuse_file_get(ff); |
9cd68455 | 527 | req->end = fuse_readpages_end; |
b93f858a | 528 | fuse_request_send_background(fc, req); |
9cd68455 | 529 | } else { |
b93f858a | 530 | fuse_request_send(fc, req); |
9cd68455 | 531 | fuse_readpages_end(fc, req); |
e9bb09dd | 532 | fuse_put_request(fc, req); |
9cd68455 | 533 | } |
db50b96c MS |
534 | } |
535 | ||
c756e0a4 | 536 | struct fuse_fill_data { |
db50b96c | 537 | struct fuse_req *req; |
a6643094 | 538 | struct file *file; |
db50b96c MS |
539 | struct inode *inode; |
540 | }; | |
541 | ||
542 | static int fuse_readpages_fill(void *_data, struct page *page) | |
543 | { | |
c756e0a4 | 544 | struct fuse_fill_data *data = _data; |
db50b96c MS |
545 | struct fuse_req *req = data->req; |
546 | struct inode *inode = data->inode; | |
547 | struct fuse_conn *fc = get_fuse_conn(inode); | |
548 | ||
3be5a52b MS |
549 | fuse_wait_on_page_writeback(inode, page->index); |
550 | ||
db50b96c MS |
551 | if (req->num_pages && |
552 | (req->num_pages == FUSE_MAX_PAGES_PER_REQ || | |
553 | (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_read || | |
554 | req->pages[req->num_pages - 1]->index + 1 != page->index)) { | |
a6643094 | 555 | fuse_send_readpages(req, data->file, inode); |
ce1d5a49 MS |
556 | data->req = req = fuse_get_req(fc); |
557 | if (IS_ERR(req)) { | |
db50b96c | 558 | unlock_page(page); |
ce1d5a49 | 559 | return PTR_ERR(req); |
db50b96c | 560 | } |
db50b96c MS |
561 | } |
562 | req->pages[req->num_pages] = page; | |
1729a16c | 563 | req->num_pages++; |
db50b96c MS |
564 | return 0; |
565 | } | |
566 | ||
567 | static int fuse_readpages(struct file *file, struct address_space *mapping, | |
568 | struct list_head *pages, unsigned nr_pages) | |
569 | { | |
570 | struct inode *inode = mapping->host; | |
571 | struct fuse_conn *fc = get_fuse_conn(inode); | |
c756e0a4 | 572 | struct fuse_fill_data data; |
db50b96c | 573 | int err; |
248d86e8 | 574 | |
1d7ea732 | 575 | err = -EIO; |
248d86e8 | 576 | if (is_bad_inode(inode)) |
2e990021 | 577 | goto out; |
248d86e8 | 578 | |
a6643094 | 579 | data.file = file; |
db50b96c | 580 | data.inode = inode; |
ce1d5a49 | 581 | data.req = fuse_get_req(fc); |
1d7ea732 | 582 | err = PTR_ERR(data.req); |
ce1d5a49 | 583 | if (IS_ERR(data.req)) |
2e990021 | 584 | goto out; |
db50b96c MS |
585 | |
586 | err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data); | |
d3406ffa MS |
587 | if (!err) { |
588 | if (data.req->num_pages) | |
a6643094 | 589 | fuse_send_readpages(data.req, file, inode); |
d3406ffa MS |
590 | else |
591 | fuse_put_request(fc, data.req); | |
592 | } | |
2e990021 | 593 | out: |
1d7ea732 | 594 | return err; |
db50b96c MS |
595 | } |
596 | ||
bcb4be80 MS |
597 | static ssize_t fuse_file_aio_read(struct kiocb *iocb, const struct iovec *iov, |
598 | unsigned long nr_segs, loff_t pos) | |
599 | { | |
600 | struct inode *inode = iocb->ki_filp->f_mapping->host; | |
601 | ||
602 | if (pos + iov_length(iov, nr_segs) > i_size_read(inode)) { | |
603 | int err; | |
604 | /* | |
605 | * If trying to read past EOF, make sure the i_size | |
606 | * attribute is up-to-date. | |
607 | */ | |
608 | err = fuse_update_attributes(inode, NULL, iocb->ki_filp, NULL); | |
609 | if (err) | |
610 | return err; | |
611 | } | |
612 | ||
613 | return generic_file_aio_read(iocb, iov, nr_segs, pos); | |
614 | } | |
615 | ||
2d698b07 MS |
616 | static void fuse_write_fill(struct fuse_req *req, struct fuse_file *ff, |
617 | struct inode *inode, loff_t pos, size_t count) | |
b6aeaded | 618 | { |
f3332114 | 619 | struct fuse_conn *fc = get_fuse_conn(inode); |
b25e82e5 MS |
620 | struct fuse_write_in *inarg = &req->misc.write.in; |
621 | struct fuse_write_out *outarg = &req->misc.write.out; | |
b6aeaded | 622 | |
b25e82e5 MS |
623 | inarg->fh = ff->fh; |
624 | inarg->offset = pos; | |
625 | inarg->size = count; | |
b6aeaded MS |
626 | req->in.h.opcode = FUSE_WRITE; |
627 | req->in.h.nodeid = get_node_id(inode); | |
b6aeaded | 628 | req->in.numargs = 2; |
f3332114 MS |
629 | if (fc->minor < 9) |
630 | req->in.args[0].size = FUSE_COMPAT_WRITE_IN_SIZE; | |
631 | else | |
632 | req->in.args[0].size = sizeof(struct fuse_write_in); | |
b25e82e5 | 633 | req->in.args[0].value = inarg; |
b6aeaded MS |
634 | req->in.args[1].size = count; |
635 | req->out.numargs = 1; | |
636 | req->out.args[0].size = sizeof(struct fuse_write_out); | |
b25e82e5 MS |
637 | req->out.args[0].value = outarg; |
638 | } | |
639 | ||
640 | static size_t fuse_send_write(struct fuse_req *req, struct file *file, | |
f3332114 MS |
641 | struct inode *inode, loff_t pos, size_t count, |
642 | fl_owner_t owner) | |
b25e82e5 MS |
643 | { |
644 | struct fuse_conn *fc = get_fuse_conn(inode); | |
2d698b07 MS |
645 | struct fuse_write_in *inarg = &req->misc.write.in; |
646 | ||
647 | fuse_write_fill(req, file->private_data, inode, pos, count); | |
648 | inarg->flags = file->f_flags; | |
f3332114 | 649 | if (owner != NULL) { |
f3332114 MS |
650 | inarg->write_flags |= FUSE_WRITE_LOCKOWNER; |
651 | inarg->lock_owner = fuse_lock_owner_id(fc, owner); | |
652 | } | |
b93f858a | 653 | fuse_request_send(fc, req); |
b25e82e5 | 654 | return req->misc.write.out.size; |
b6aeaded MS |
655 | } |
656 | ||
5e6f58a1 NP |
657 | static int fuse_write_begin(struct file *file, struct address_space *mapping, |
658 | loff_t pos, unsigned len, unsigned flags, | |
659 | struct page **pagep, void **fsdata) | |
b6aeaded | 660 | { |
5e6f58a1 NP |
661 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
662 | ||
54566b2c | 663 | *pagep = grab_cache_page_write_begin(mapping, index, flags); |
5e6f58a1 NP |
664 | if (!*pagep) |
665 | return -ENOMEM; | |
b6aeaded MS |
666 | return 0; |
667 | } | |
668 | ||
854512ec MS |
669 | static void fuse_write_update_size(struct inode *inode, loff_t pos) |
670 | { | |
671 | struct fuse_conn *fc = get_fuse_conn(inode); | |
672 | struct fuse_inode *fi = get_fuse_inode(inode); | |
673 | ||
674 | spin_lock(&fc->lock); | |
675 | fi->attr_version = ++fc->attr_version; | |
676 | if (pos > inode->i_size) | |
677 | i_size_write(inode, pos); | |
678 | spin_unlock(&fc->lock); | |
679 | } | |
680 | ||
5e6f58a1 NP |
681 | static int fuse_buffered_write(struct file *file, struct inode *inode, |
682 | loff_t pos, unsigned count, struct page *page) | |
b6aeaded MS |
683 | { |
684 | int err; | |
04730fef | 685 | size_t nres; |
b6aeaded | 686 | struct fuse_conn *fc = get_fuse_conn(inode); |
5e6f58a1 | 687 | unsigned offset = pos & (PAGE_CACHE_SIZE - 1); |
248d86e8 MS |
688 | struct fuse_req *req; |
689 | ||
690 | if (is_bad_inode(inode)) | |
691 | return -EIO; | |
692 | ||
3be5a52b MS |
693 | /* |
694 | * Make sure writepages on the same page are not mixed up with | |
695 | * plain writes. | |
696 | */ | |
697 | fuse_wait_on_page_writeback(inode, page->index); | |
698 | ||
ce1d5a49 MS |
699 | req = fuse_get_req(fc); |
700 | if (IS_ERR(req)) | |
701 | return PTR_ERR(req); | |
b6aeaded | 702 | |
f4975c67 | 703 | req->in.argpages = 1; |
b6aeaded MS |
704 | req->num_pages = 1; |
705 | req->pages[0] = page; | |
706 | req->page_offset = offset; | |
f3332114 | 707 | nres = fuse_send_write(req, file, inode, pos, count, NULL); |
b6aeaded MS |
708 | err = req->out.h.error; |
709 | fuse_put_request(fc, req); | |
5e6f58a1 | 710 | if (!err && !nres) |
b6aeaded MS |
711 | err = -EIO; |
712 | if (!err) { | |
5e6f58a1 | 713 | pos += nres; |
854512ec | 714 | fuse_write_update_size(inode, pos); |
5e6f58a1 | 715 | if (count == PAGE_CACHE_SIZE) |
b6aeaded | 716 | SetPageUptodate(page); |
b36c31ba MS |
717 | } |
718 | fuse_invalidate_attr(inode); | |
5e6f58a1 NP |
719 | return err ? err : nres; |
720 | } | |
721 | ||
722 | static int fuse_write_end(struct file *file, struct address_space *mapping, | |
723 | loff_t pos, unsigned len, unsigned copied, | |
724 | struct page *page, void *fsdata) | |
725 | { | |
726 | struct inode *inode = mapping->host; | |
727 | int res = 0; | |
728 | ||
729 | if (copied) | |
730 | res = fuse_buffered_write(file, inode, pos, copied, page); | |
731 | ||
732 | unlock_page(page); | |
733 | page_cache_release(page); | |
734 | return res; | |
b6aeaded MS |
735 | } |
736 | ||
ea9b9907 NP |
737 | static size_t fuse_send_write_pages(struct fuse_req *req, struct file *file, |
738 | struct inode *inode, loff_t pos, | |
739 | size_t count) | |
740 | { | |
741 | size_t res; | |
742 | unsigned offset; | |
743 | unsigned i; | |
744 | ||
745 | for (i = 0; i < req->num_pages; i++) | |
746 | fuse_wait_on_page_writeback(inode, req->pages[i]->index); | |
747 | ||
748 | res = fuse_send_write(req, file, inode, pos, count, NULL); | |
749 | ||
750 | offset = req->page_offset; | |
751 | count = res; | |
752 | for (i = 0; i < req->num_pages; i++) { | |
753 | struct page *page = req->pages[i]; | |
754 | ||
755 | if (!req->out.h.error && !offset && count >= PAGE_CACHE_SIZE) | |
756 | SetPageUptodate(page); | |
757 | ||
758 | if (count > PAGE_CACHE_SIZE - offset) | |
759 | count -= PAGE_CACHE_SIZE - offset; | |
760 | else | |
761 | count = 0; | |
762 | offset = 0; | |
763 | ||
764 | unlock_page(page); | |
765 | page_cache_release(page); | |
766 | } | |
767 | ||
768 | return res; | |
769 | } | |
770 | ||
771 | static ssize_t fuse_fill_write_pages(struct fuse_req *req, | |
772 | struct address_space *mapping, | |
773 | struct iov_iter *ii, loff_t pos) | |
774 | { | |
775 | struct fuse_conn *fc = get_fuse_conn(mapping->host); | |
776 | unsigned offset = pos & (PAGE_CACHE_SIZE - 1); | |
777 | size_t count = 0; | |
778 | int err; | |
779 | ||
f4975c67 | 780 | req->in.argpages = 1; |
ea9b9907 NP |
781 | req->page_offset = offset; |
782 | ||
783 | do { | |
784 | size_t tmp; | |
785 | struct page *page; | |
786 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; | |
787 | size_t bytes = min_t(size_t, PAGE_CACHE_SIZE - offset, | |
788 | iov_iter_count(ii)); | |
789 | ||
790 | bytes = min_t(size_t, bytes, fc->max_write - count); | |
791 | ||
792 | again: | |
793 | err = -EFAULT; | |
794 | if (iov_iter_fault_in_readable(ii, bytes)) | |
795 | break; | |
796 | ||
797 | err = -ENOMEM; | |
54566b2c | 798 | page = grab_cache_page_write_begin(mapping, index, 0); |
ea9b9907 NP |
799 | if (!page) |
800 | break; | |
801 | ||
802 | pagefault_disable(); | |
803 | tmp = iov_iter_copy_from_user_atomic(page, ii, offset, bytes); | |
804 | pagefault_enable(); | |
805 | flush_dcache_page(page); | |
806 | ||
807 | if (!tmp) { | |
808 | unlock_page(page); | |
809 | page_cache_release(page); | |
810 | bytes = min(bytes, iov_iter_single_seg_count(ii)); | |
811 | goto again; | |
812 | } | |
813 | ||
814 | err = 0; | |
815 | req->pages[req->num_pages] = page; | |
816 | req->num_pages++; | |
817 | ||
818 | iov_iter_advance(ii, tmp); | |
819 | count += tmp; | |
820 | pos += tmp; | |
821 | offset += tmp; | |
822 | if (offset == PAGE_CACHE_SIZE) | |
823 | offset = 0; | |
824 | ||
78bb6cb9 MS |
825 | if (!fc->big_writes) |
826 | break; | |
ea9b9907 NP |
827 | } while (iov_iter_count(ii) && count < fc->max_write && |
828 | req->num_pages < FUSE_MAX_PAGES_PER_REQ && offset == 0); | |
829 | ||
830 | return count > 0 ? count : err; | |
831 | } | |
832 | ||
833 | static ssize_t fuse_perform_write(struct file *file, | |
834 | struct address_space *mapping, | |
835 | struct iov_iter *ii, loff_t pos) | |
836 | { | |
837 | struct inode *inode = mapping->host; | |
838 | struct fuse_conn *fc = get_fuse_conn(inode); | |
839 | int err = 0; | |
840 | ssize_t res = 0; | |
841 | ||
842 | if (is_bad_inode(inode)) | |
843 | return -EIO; | |
844 | ||
845 | do { | |
846 | struct fuse_req *req; | |
847 | ssize_t count; | |
848 | ||
849 | req = fuse_get_req(fc); | |
850 | if (IS_ERR(req)) { | |
851 | err = PTR_ERR(req); | |
852 | break; | |
853 | } | |
854 | ||
855 | count = fuse_fill_write_pages(req, mapping, ii, pos); | |
856 | if (count <= 0) { | |
857 | err = count; | |
858 | } else { | |
859 | size_t num_written; | |
860 | ||
861 | num_written = fuse_send_write_pages(req, file, inode, | |
862 | pos, count); | |
863 | err = req->out.h.error; | |
864 | if (!err) { | |
865 | res += num_written; | |
866 | pos += num_written; | |
867 | ||
868 | /* break out of the loop on short write */ | |
869 | if (num_written != count) | |
870 | err = -EIO; | |
871 | } | |
872 | } | |
873 | fuse_put_request(fc, req); | |
874 | } while (!err && iov_iter_count(ii)); | |
875 | ||
876 | if (res > 0) | |
877 | fuse_write_update_size(inode, pos); | |
878 | ||
879 | fuse_invalidate_attr(inode); | |
880 | ||
881 | return res > 0 ? res : err; | |
882 | } | |
883 | ||
884 | static ssize_t fuse_file_aio_write(struct kiocb *iocb, const struct iovec *iov, | |
885 | unsigned long nr_segs, loff_t pos) | |
886 | { | |
887 | struct file *file = iocb->ki_filp; | |
888 | struct address_space *mapping = file->f_mapping; | |
889 | size_t count = 0; | |
890 | ssize_t written = 0; | |
891 | struct inode *inode = mapping->host; | |
892 | ssize_t err; | |
893 | struct iov_iter i; | |
894 | ||
895 | WARN_ON(iocb->ki_pos != pos); | |
896 | ||
897 | err = generic_segment_checks(iov, &nr_segs, &count, VERIFY_READ); | |
898 | if (err) | |
899 | return err; | |
900 | ||
901 | mutex_lock(&inode->i_mutex); | |
902 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); | |
903 | ||
904 | /* We can write back this queue in page reclaim */ | |
905 | current->backing_dev_info = mapping->backing_dev_info; | |
906 | ||
907 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); | |
908 | if (err) | |
909 | goto out; | |
910 | ||
911 | if (count == 0) | |
912 | goto out; | |
913 | ||
2f1936b8 | 914 | err = file_remove_suid(file); |
ea9b9907 NP |
915 | if (err) |
916 | goto out; | |
917 | ||
918 | file_update_time(file); | |
919 | ||
920 | iov_iter_init(&i, iov, nr_segs, count, 0); | |
921 | written = fuse_perform_write(file, mapping, &i, pos); | |
922 | if (written >= 0) | |
923 | iocb->ki_pos = pos + written; | |
924 | ||
925 | out: | |
926 | current->backing_dev_info = NULL; | |
927 | mutex_unlock(&inode->i_mutex); | |
928 | ||
929 | return written ? written : err; | |
930 | } | |
931 | ||
413ef8cb MS |
932 | static void fuse_release_user_pages(struct fuse_req *req, int write) |
933 | { | |
934 | unsigned i; | |
935 | ||
936 | for (i = 0; i < req->num_pages; i++) { | |
937 | struct page *page = req->pages[i]; | |
938 | if (write) | |
939 | set_page_dirty_lock(page); | |
940 | put_page(page); | |
941 | } | |
942 | } | |
943 | ||
944 | static int fuse_get_user_pages(struct fuse_req *req, const char __user *buf, | |
ce60a2f1 | 945 | size_t *nbytesp, int write) |
413ef8cb | 946 | { |
ce60a2f1 | 947 | size_t nbytes = *nbytesp; |
413ef8cb MS |
948 | unsigned long user_addr = (unsigned long) buf; |
949 | unsigned offset = user_addr & ~PAGE_MASK; | |
950 | int npages; | |
951 | ||
f4975c67 MS |
952 | /* Special case for kernel I/O: can copy directly into the buffer */ |
953 | if (segment_eq(get_fs(), KERNEL_DS)) { | |
954 | if (write) | |
955 | req->in.args[1].value = (void *) user_addr; | |
956 | else | |
957 | req->out.args[0].value = (void *) user_addr; | |
958 | ||
959 | return 0; | |
960 | } | |
413ef8cb | 961 | |
ce60a2f1 | 962 | nbytes = min_t(size_t, nbytes, FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT); |
413ef8cb | 963 | npages = (nbytes + offset + PAGE_SIZE - 1) >> PAGE_SHIFT; |
bd730967 | 964 | npages = clamp(npages, 1, FUSE_MAX_PAGES_PER_REQ); |
413ef8cb | 965 | down_read(¤t->mm->mmap_sem); |
f4975c67 | 966 | npages = get_user_pages(current, current->mm, user_addr, npages, !write, |
413ef8cb MS |
967 | 0, req->pages, NULL); |
968 | up_read(¤t->mm->mmap_sem); | |
969 | if (npages < 0) | |
970 | return npages; | |
971 | ||
972 | req->num_pages = npages; | |
973 | req->page_offset = offset; | |
f4975c67 MS |
974 | |
975 | if (write) | |
976 | req->in.argpages = 1; | |
977 | else | |
978 | req->out.argpages = 1; | |
979 | ||
980 | nbytes = (req->num_pages << PAGE_SHIFT) - req->page_offset; | |
981 | *nbytesp = min(*nbytesp, nbytes); | |
982 | ||
413ef8cb MS |
983 | return 0; |
984 | } | |
985 | ||
986 | static ssize_t fuse_direct_io(struct file *file, const char __user *buf, | |
987 | size_t count, loff_t *ppos, int write) | |
988 | { | |
7706a9d6 | 989 | struct inode *inode = file->f_path.dentry->d_inode; |
413ef8cb MS |
990 | struct fuse_conn *fc = get_fuse_conn(inode); |
991 | size_t nmax = write ? fc->max_write : fc->max_read; | |
992 | loff_t pos = *ppos; | |
993 | ssize_t res = 0; | |
248d86e8 MS |
994 | struct fuse_req *req; |
995 | ||
ce1d5a49 MS |
996 | req = fuse_get_req(fc); |
997 | if (IS_ERR(req)) | |
998 | return PTR_ERR(req); | |
413ef8cb MS |
999 | |
1000 | while (count) { | |
413ef8cb | 1001 | size_t nres; |
f4975c67 MS |
1002 | size_t nbytes = min(count, nmax); |
1003 | int err = fuse_get_user_pages(req, buf, &nbytes, write); | |
413ef8cb MS |
1004 | if (err) { |
1005 | res = err; | |
1006 | break; | |
1007 | } | |
f4975c67 | 1008 | |
413ef8cb | 1009 | if (write) |
f3332114 MS |
1010 | nres = fuse_send_write(req, file, inode, pos, nbytes, |
1011 | current->files); | |
413ef8cb | 1012 | else |
f3332114 MS |
1013 | nres = fuse_send_read(req, file, inode, pos, nbytes, |
1014 | current->files); | |
413ef8cb MS |
1015 | fuse_release_user_pages(req, !write); |
1016 | if (req->out.h.error) { | |
1017 | if (!res) | |
1018 | res = req->out.h.error; | |
1019 | break; | |
1020 | } else if (nres > nbytes) { | |
1021 | res = -EIO; | |
1022 | break; | |
1023 | } | |
1024 | count -= nres; | |
1025 | res += nres; | |
1026 | pos += nres; | |
1027 | buf += nres; | |
1028 | if (nres != nbytes) | |
1029 | break; | |
56cf34ff MS |
1030 | if (count) { |
1031 | fuse_put_request(fc, req); | |
1032 | req = fuse_get_req(fc); | |
1033 | if (IS_ERR(req)) | |
1034 | break; | |
1035 | } | |
413ef8cb MS |
1036 | } |
1037 | fuse_put_request(fc, req); | |
d09cb9d7 | 1038 | if (res > 0) |
413ef8cb | 1039 | *ppos = pos; |
413ef8cb MS |
1040 | |
1041 | return res; | |
1042 | } | |
1043 | ||
1044 | static ssize_t fuse_direct_read(struct file *file, char __user *buf, | |
1045 | size_t count, loff_t *ppos) | |
1046 | { | |
d09cb9d7 MS |
1047 | ssize_t res; |
1048 | struct inode *inode = file->f_path.dentry->d_inode; | |
1049 | ||
1050 | if (is_bad_inode(inode)) | |
1051 | return -EIO; | |
1052 | ||
1053 | res = fuse_direct_io(file, buf, count, ppos, 0); | |
1054 | ||
1055 | fuse_invalidate_attr(inode); | |
1056 | ||
1057 | return res; | |
413ef8cb MS |
1058 | } |
1059 | ||
1060 | static ssize_t fuse_direct_write(struct file *file, const char __user *buf, | |
1061 | size_t count, loff_t *ppos) | |
1062 | { | |
7706a9d6 | 1063 | struct inode *inode = file->f_path.dentry->d_inode; |
413ef8cb | 1064 | ssize_t res; |
d09cb9d7 MS |
1065 | |
1066 | if (is_bad_inode(inode)) | |
1067 | return -EIO; | |
1068 | ||
413ef8cb | 1069 | /* Don't allow parallel writes to the same file */ |
1b1dcc1b | 1070 | mutex_lock(&inode->i_mutex); |
889f7848 | 1071 | res = generic_write_checks(file, ppos, &count, 0); |
d09cb9d7 | 1072 | if (!res) { |
889f7848 | 1073 | res = fuse_direct_io(file, buf, count, ppos, 1); |
d09cb9d7 MS |
1074 | if (res > 0) |
1075 | fuse_write_update_size(inode, *ppos); | |
1076 | } | |
1b1dcc1b | 1077 | mutex_unlock(&inode->i_mutex); |
d09cb9d7 MS |
1078 | |
1079 | fuse_invalidate_attr(inode); | |
1080 | ||
413ef8cb MS |
1081 | return res; |
1082 | } | |
1083 | ||
3be5a52b | 1084 | static void fuse_writepage_free(struct fuse_conn *fc, struct fuse_req *req) |
b6aeaded | 1085 | { |
3be5a52b MS |
1086 | __free_page(req->pages[0]); |
1087 | fuse_file_put(req->ff); | |
3be5a52b MS |
1088 | } |
1089 | ||
1090 | static void fuse_writepage_finish(struct fuse_conn *fc, struct fuse_req *req) | |
1091 | { | |
1092 | struct inode *inode = req->inode; | |
1093 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1094 | struct backing_dev_info *bdi = inode->i_mapping->backing_dev_info; | |
1095 | ||
1096 | list_del(&req->writepages_entry); | |
1097 | dec_bdi_stat(bdi, BDI_WRITEBACK); | |
1098 | dec_zone_page_state(req->pages[0], NR_WRITEBACK_TEMP); | |
1099 | bdi_writeout_inc(bdi); | |
1100 | wake_up(&fi->page_waitq); | |
1101 | } | |
1102 | ||
1103 | /* Called under fc->lock, may release and reacquire it */ | |
1104 | static void fuse_send_writepage(struct fuse_conn *fc, struct fuse_req *req) | |
5d9ec854 HH |
1105 | __releases(&fc->lock) |
1106 | __acquires(&fc->lock) | |
3be5a52b MS |
1107 | { |
1108 | struct fuse_inode *fi = get_fuse_inode(req->inode); | |
1109 | loff_t size = i_size_read(req->inode); | |
1110 | struct fuse_write_in *inarg = &req->misc.write.in; | |
1111 | ||
1112 | if (!fc->connected) | |
1113 | goto out_free; | |
1114 | ||
1115 | if (inarg->offset + PAGE_CACHE_SIZE <= size) { | |
1116 | inarg->size = PAGE_CACHE_SIZE; | |
1117 | } else if (inarg->offset < size) { | |
1118 | inarg->size = size & (PAGE_CACHE_SIZE - 1); | |
1119 | } else { | |
1120 | /* Got truncated off completely */ | |
1121 | goto out_free; | |
b6aeaded | 1122 | } |
3be5a52b MS |
1123 | |
1124 | req->in.args[1].size = inarg->size; | |
1125 | fi->writectr++; | |
b93f858a | 1126 | fuse_request_send_background_locked(fc, req); |
3be5a52b MS |
1127 | return; |
1128 | ||
1129 | out_free: | |
1130 | fuse_writepage_finish(fc, req); | |
1131 | spin_unlock(&fc->lock); | |
1132 | fuse_writepage_free(fc, req); | |
e9bb09dd | 1133 | fuse_put_request(fc, req); |
3be5a52b | 1134 | spin_lock(&fc->lock); |
b6aeaded MS |
1135 | } |
1136 | ||
3be5a52b MS |
1137 | /* |
1138 | * If fi->writectr is positive (no truncate or fsync going on) send | |
1139 | * all queued writepage requests. | |
1140 | * | |
1141 | * Called with fc->lock | |
1142 | */ | |
1143 | void fuse_flush_writepages(struct inode *inode) | |
5d9ec854 HH |
1144 | __releases(&fc->lock) |
1145 | __acquires(&fc->lock) | |
b6aeaded | 1146 | { |
3be5a52b MS |
1147 | struct fuse_conn *fc = get_fuse_conn(inode); |
1148 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1149 | struct fuse_req *req; | |
1150 | ||
1151 | while (fi->writectr >= 0 && !list_empty(&fi->queued_writes)) { | |
1152 | req = list_entry(fi->queued_writes.next, struct fuse_req, list); | |
1153 | list_del_init(&req->list); | |
1154 | fuse_send_writepage(fc, req); | |
1155 | } | |
1156 | } | |
1157 | ||
1158 | static void fuse_writepage_end(struct fuse_conn *fc, struct fuse_req *req) | |
1159 | { | |
1160 | struct inode *inode = req->inode; | |
1161 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1162 | ||
1163 | mapping_set_error(inode->i_mapping, req->out.h.error); | |
1164 | spin_lock(&fc->lock); | |
1165 | fi->writectr--; | |
1166 | fuse_writepage_finish(fc, req); | |
1167 | spin_unlock(&fc->lock); | |
1168 | fuse_writepage_free(fc, req); | |
1169 | } | |
1170 | ||
1171 | static int fuse_writepage_locked(struct page *page) | |
1172 | { | |
1173 | struct address_space *mapping = page->mapping; | |
1174 | struct inode *inode = mapping->host; | |
1175 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1176 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1177 | struct fuse_req *req; | |
1178 | struct fuse_file *ff; | |
1179 | struct page *tmp_page; | |
1180 | ||
1181 | set_page_writeback(page); | |
1182 | ||
1183 | req = fuse_request_alloc_nofs(); | |
1184 | if (!req) | |
1185 | goto err; | |
1186 | ||
1187 | tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
1188 | if (!tmp_page) | |
1189 | goto err_free; | |
1190 | ||
1191 | spin_lock(&fc->lock); | |
1192 | BUG_ON(list_empty(&fi->write_files)); | |
1193 | ff = list_entry(fi->write_files.next, struct fuse_file, write_entry); | |
1194 | req->ff = fuse_file_get(ff); | |
1195 | spin_unlock(&fc->lock); | |
1196 | ||
2d698b07 | 1197 | fuse_write_fill(req, ff, inode, page_offset(page), 0); |
3be5a52b MS |
1198 | |
1199 | copy_highpage(tmp_page, page); | |
2d698b07 | 1200 | req->misc.write.in.write_flags |= FUSE_WRITE_CACHE; |
f4975c67 | 1201 | req->in.argpages = 1; |
3be5a52b MS |
1202 | req->num_pages = 1; |
1203 | req->pages[0] = tmp_page; | |
1204 | req->page_offset = 0; | |
1205 | req->end = fuse_writepage_end; | |
1206 | req->inode = inode; | |
1207 | ||
1208 | inc_bdi_stat(mapping->backing_dev_info, BDI_WRITEBACK); | |
1209 | inc_zone_page_state(tmp_page, NR_WRITEBACK_TEMP); | |
1210 | end_page_writeback(page); | |
1211 | ||
1212 | spin_lock(&fc->lock); | |
1213 | list_add(&req->writepages_entry, &fi->writepages); | |
1214 | list_add_tail(&req->list, &fi->queued_writes); | |
1215 | fuse_flush_writepages(inode); | |
1216 | spin_unlock(&fc->lock); | |
1217 | ||
1218 | return 0; | |
1219 | ||
1220 | err_free: | |
1221 | fuse_request_free(req); | |
1222 | err: | |
1223 | end_page_writeback(page); | |
1224 | return -ENOMEM; | |
1225 | } | |
1226 | ||
1227 | static int fuse_writepage(struct page *page, struct writeback_control *wbc) | |
1228 | { | |
1229 | int err; | |
1230 | ||
1231 | err = fuse_writepage_locked(page); | |
1232 | unlock_page(page); | |
1233 | ||
1234 | return err; | |
1235 | } | |
1236 | ||
1237 | static int fuse_launder_page(struct page *page) | |
1238 | { | |
1239 | int err = 0; | |
1240 | if (clear_page_dirty_for_io(page)) { | |
1241 | struct inode *inode = page->mapping->host; | |
1242 | err = fuse_writepage_locked(page); | |
1243 | if (!err) | |
1244 | fuse_wait_on_page_writeback(inode, page->index); | |
1245 | } | |
1246 | return err; | |
1247 | } | |
1248 | ||
1249 | /* | |
1250 | * Write back dirty pages now, because there may not be any suitable | |
1251 | * open files later | |
1252 | */ | |
1253 | static void fuse_vma_close(struct vm_area_struct *vma) | |
1254 | { | |
1255 | filemap_write_and_wait(vma->vm_file->f_mapping); | |
1256 | } | |
1257 | ||
1258 | /* | |
1259 | * Wait for writeback against this page to complete before allowing it | |
1260 | * to be marked dirty again, and hence written back again, possibly | |
1261 | * before the previous writepage completed. | |
1262 | * | |
1263 | * Block here, instead of in ->writepage(), so that the userspace fs | |
1264 | * can only block processes actually operating on the filesystem. | |
1265 | * | |
1266 | * Otherwise unprivileged userspace fs would be able to block | |
1267 | * unrelated: | |
1268 | * | |
1269 | * - page migration | |
1270 | * - sync(2) | |
1271 | * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER | |
1272 | */ | |
c2ec175c | 1273 | static int fuse_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
3be5a52b | 1274 | { |
c2ec175c | 1275 | struct page *page = vmf->page; |
3be5a52b MS |
1276 | /* |
1277 | * Don't use page->mapping as it may become NULL from a | |
1278 | * concurrent truncate. | |
1279 | */ | |
1280 | struct inode *inode = vma->vm_file->f_mapping->host; | |
1281 | ||
1282 | fuse_wait_on_page_writeback(inode, page->index); | |
1283 | return 0; | |
1284 | } | |
1285 | ||
1286 | static struct vm_operations_struct fuse_file_vm_ops = { | |
1287 | .close = fuse_vma_close, | |
1288 | .fault = filemap_fault, | |
1289 | .page_mkwrite = fuse_page_mkwrite, | |
1290 | }; | |
1291 | ||
1292 | static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma) | |
1293 | { | |
1294 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) { | |
1295 | struct inode *inode = file->f_dentry->d_inode; | |
1296 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1297 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1298 | struct fuse_file *ff = file->private_data; | |
1299 | /* | |
1300 | * file may be written through mmap, so chain it onto the | |
1301 | * inodes's write_file list | |
1302 | */ | |
1303 | spin_lock(&fc->lock); | |
1304 | if (list_empty(&ff->write_entry)) | |
1305 | list_add(&ff->write_entry, &fi->write_files); | |
1306 | spin_unlock(&fc->lock); | |
1307 | } | |
1308 | file_accessed(file); | |
1309 | vma->vm_ops = &fuse_file_vm_ops; | |
b6aeaded MS |
1310 | return 0; |
1311 | } | |
1312 | ||
fc280c96 MS |
1313 | static int fuse_direct_mmap(struct file *file, struct vm_area_struct *vma) |
1314 | { | |
1315 | /* Can't provide the coherency needed for MAP_SHARED */ | |
1316 | if (vma->vm_flags & VM_MAYSHARE) | |
1317 | return -ENODEV; | |
1318 | ||
3121bfe7 MS |
1319 | invalidate_inode_pages2(file->f_mapping); |
1320 | ||
fc280c96 MS |
1321 | return generic_file_mmap(file, vma); |
1322 | } | |
1323 | ||
71421259 MS |
1324 | static int convert_fuse_file_lock(const struct fuse_file_lock *ffl, |
1325 | struct file_lock *fl) | |
1326 | { | |
1327 | switch (ffl->type) { | |
1328 | case F_UNLCK: | |
1329 | break; | |
1330 | ||
1331 | case F_RDLCK: | |
1332 | case F_WRLCK: | |
1333 | if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX || | |
1334 | ffl->end < ffl->start) | |
1335 | return -EIO; | |
1336 | ||
1337 | fl->fl_start = ffl->start; | |
1338 | fl->fl_end = ffl->end; | |
1339 | fl->fl_pid = ffl->pid; | |
1340 | break; | |
1341 | ||
1342 | default: | |
1343 | return -EIO; | |
1344 | } | |
1345 | fl->fl_type = ffl->type; | |
1346 | return 0; | |
1347 | } | |
1348 | ||
1349 | static void fuse_lk_fill(struct fuse_req *req, struct file *file, | |
a9ff4f87 MS |
1350 | const struct file_lock *fl, int opcode, pid_t pid, |
1351 | int flock) | |
71421259 | 1352 | { |
7706a9d6 | 1353 | struct inode *inode = file->f_path.dentry->d_inode; |
9c8ef561 | 1354 | struct fuse_conn *fc = get_fuse_conn(inode); |
71421259 MS |
1355 | struct fuse_file *ff = file->private_data; |
1356 | struct fuse_lk_in *arg = &req->misc.lk_in; | |
1357 | ||
1358 | arg->fh = ff->fh; | |
9c8ef561 | 1359 | arg->owner = fuse_lock_owner_id(fc, fl->fl_owner); |
71421259 MS |
1360 | arg->lk.start = fl->fl_start; |
1361 | arg->lk.end = fl->fl_end; | |
1362 | arg->lk.type = fl->fl_type; | |
1363 | arg->lk.pid = pid; | |
a9ff4f87 MS |
1364 | if (flock) |
1365 | arg->lk_flags |= FUSE_LK_FLOCK; | |
71421259 MS |
1366 | req->in.h.opcode = opcode; |
1367 | req->in.h.nodeid = get_node_id(inode); | |
1368 | req->in.numargs = 1; | |
1369 | req->in.args[0].size = sizeof(*arg); | |
1370 | req->in.args[0].value = arg; | |
1371 | } | |
1372 | ||
1373 | static int fuse_getlk(struct file *file, struct file_lock *fl) | |
1374 | { | |
7706a9d6 | 1375 | struct inode *inode = file->f_path.dentry->d_inode; |
71421259 MS |
1376 | struct fuse_conn *fc = get_fuse_conn(inode); |
1377 | struct fuse_req *req; | |
1378 | struct fuse_lk_out outarg; | |
1379 | int err; | |
1380 | ||
1381 | req = fuse_get_req(fc); | |
1382 | if (IS_ERR(req)) | |
1383 | return PTR_ERR(req); | |
1384 | ||
a9ff4f87 | 1385 | fuse_lk_fill(req, file, fl, FUSE_GETLK, 0, 0); |
71421259 MS |
1386 | req->out.numargs = 1; |
1387 | req->out.args[0].size = sizeof(outarg); | |
1388 | req->out.args[0].value = &outarg; | |
b93f858a | 1389 | fuse_request_send(fc, req); |
71421259 MS |
1390 | err = req->out.h.error; |
1391 | fuse_put_request(fc, req); | |
1392 | if (!err) | |
1393 | err = convert_fuse_file_lock(&outarg.lk, fl); | |
1394 | ||
1395 | return err; | |
1396 | } | |
1397 | ||
a9ff4f87 | 1398 | static int fuse_setlk(struct file *file, struct file_lock *fl, int flock) |
71421259 | 1399 | { |
7706a9d6 | 1400 | struct inode *inode = file->f_path.dentry->d_inode; |
71421259 MS |
1401 | struct fuse_conn *fc = get_fuse_conn(inode); |
1402 | struct fuse_req *req; | |
1403 | int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK; | |
1404 | pid_t pid = fl->fl_type != F_UNLCK ? current->tgid : 0; | |
1405 | int err; | |
1406 | ||
48e90761 MS |
1407 | if (fl->fl_lmops && fl->fl_lmops->fl_grant) { |
1408 | /* NLM needs asynchronous locks, which we don't support yet */ | |
1409 | return -ENOLCK; | |
1410 | } | |
1411 | ||
71421259 MS |
1412 | /* Unlock on close is handled by the flush method */ |
1413 | if (fl->fl_flags & FL_CLOSE) | |
1414 | return 0; | |
1415 | ||
1416 | req = fuse_get_req(fc); | |
1417 | if (IS_ERR(req)) | |
1418 | return PTR_ERR(req); | |
1419 | ||
a9ff4f87 | 1420 | fuse_lk_fill(req, file, fl, opcode, pid, flock); |
b93f858a | 1421 | fuse_request_send(fc, req); |
71421259 | 1422 | err = req->out.h.error; |
a4d27e75 MS |
1423 | /* locking is restartable */ |
1424 | if (err == -EINTR) | |
1425 | err = -ERESTARTSYS; | |
71421259 MS |
1426 | fuse_put_request(fc, req); |
1427 | return err; | |
1428 | } | |
1429 | ||
1430 | static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl) | |
1431 | { | |
7706a9d6 | 1432 | struct inode *inode = file->f_path.dentry->d_inode; |
71421259 MS |
1433 | struct fuse_conn *fc = get_fuse_conn(inode); |
1434 | int err; | |
1435 | ||
48e90761 MS |
1436 | if (cmd == F_CANCELLK) { |
1437 | err = 0; | |
1438 | } else if (cmd == F_GETLK) { | |
71421259 | 1439 | if (fc->no_lock) { |
9d6a8c5c | 1440 | posix_test_lock(file, fl); |
71421259 MS |
1441 | err = 0; |
1442 | } else | |
1443 | err = fuse_getlk(file, fl); | |
1444 | } else { | |
1445 | if (fc->no_lock) | |
48e90761 | 1446 | err = posix_lock_file(file, fl, NULL); |
71421259 | 1447 | else |
a9ff4f87 | 1448 | err = fuse_setlk(file, fl, 0); |
71421259 MS |
1449 | } |
1450 | return err; | |
1451 | } | |
1452 | ||
a9ff4f87 MS |
1453 | static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl) |
1454 | { | |
1455 | struct inode *inode = file->f_path.dentry->d_inode; | |
1456 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1457 | int err; | |
1458 | ||
1459 | if (fc->no_lock) { | |
1460 | err = flock_lock_file_wait(file, fl); | |
1461 | } else { | |
1462 | /* emulate flock with POSIX locks */ | |
1463 | fl->fl_owner = (fl_owner_t) file; | |
1464 | err = fuse_setlk(file, fl, 1); | |
1465 | } | |
1466 | ||
1467 | return err; | |
1468 | } | |
1469 | ||
b2d2272f MS |
1470 | static sector_t fuse_bmap(struct address_space *mapping, sector_t block) |
1471 | { | |
1472 | struct inode *inode = mapping->host; | |
1473 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1474 | struct fuse_req *req; | |
1475 | struct fuse_bmap_in inarg; | |
1476 | struct fuse_bmap_out outarg; | |
1477 | int err; | |
1478 | ||
1479 | if (!inode->i_sb->s_bdev || fc->no_bmap) | |
1480 | return 0; | |
1481 | ||
1482 | req = fuse_get_req(fc); | |
1483 | if (IS_ERR(req)) | |
1484 | return 0; | |
1485 | ||
1486 | memset(&inarg, 0, sizeof(inarg)); | |
1487 | inarg.block = block; | |
1488 | inarg.blocksize = inode->i_sb->s_blocksize; | |
1489 | req->in.h.opcode = FUSE_BMAP; | |
1490 | req->in.h.nodeid = get_node_id(inode); | |
1491 | req->in.numargs = 1; | |
1492 | req->in.args[0].size = sizeof(inarg); | |
1493 | req->in.args[0].value = &inarg; | |
1494 | req->out.numargs = 1; | |
1495 | req->out.args[0].size = sizeof(outarg); | |
1496 | req->out.args[0].value = &outarg; | |
b93f858a | 1497 | fuse_request_send(fc, req); |
b2d2272f MS |
1498 | err = req->out.h.error; |
1499 | fuse_put_request(fc, req); | |
1500 | if (err == -ENOSYS) | |
1501 | fc->no_bmap = 1; | |
1502 | ||
1503 | return err ? 0 : outarg.block; | |
1504 | } | |
1505 | ||
5559b8f4 MS |
1506 | static loff_t fuse_file_llseek(struct file *file, loff_t offset, int origin) |
1507 | { | |
1508 | loff_t retval; | |
1509 | struct inode *inode = file->f_path.dentry->d_inode; | |
1510 | ||
1511 | mutex_lock(&inode->i_mutex); | |
1512 | switch (origin) { | |
1513 | case SEEK_END: | |
769415c6 MS |
1514 | retval = fuse_update_attributes(inode, NULL, file, NULL); |
1515 | if (retval) | |
5291658d | 1516 | goto exit; |
5559b8f4 MS |
1517 | offset += i_size_read(inode); |
1518 | break; | |
1519 | case SEEK_CUR: | |
1520 | offset += file->f_pos; | |
1521 | } | |
1522 | retval = -EINVAL; | |
1523 | if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) { | |
1524 | if (offset != file->f_pos) { | |
1525 | file->f_pos = offset; | |
1526 | file->f_version = 0; | |
1527 | } | |
1528 | retval = offset; | |
1529 | } | |
5291658d | 1530 | exit: |
5559b8f4 MS |
1531 | mutex_unlock(&inode->i_mutex); |
1532 | return retval; | |
1533 | } | |
1534 | ||
59efec7b TH |
1535 | static int fuse_ioctl_copy_user(struct page **pages, struct iovec *iov, |
1536 | unsigned int nr_segs, size_t bytes, bool to_user) | |
1537 | { | |
1538 | struct iov_iter ii; | |
1539 | int page_idx = 0; | |
1540 | ||
1541 | if (!bytes) | |
1542 | return 0; | |
1543 | ||
1544 | iov_iter_init(&ii, iov, nr_segs, bytes, 0); | |
1545 | ||
1546 | while (iov_iter_count(&ii)) { | |
1547 | struct page *page = pages[page_idx++]; | |
1548 | size_t todo = min_t(size_t, PAGE_SIZE, iov_iter_count(&ii)); | |
1549 | void *kaddr, *map; | |
1550 | ||
1551 | kaddr = map = kmap(page); | |
1552 | ||
1553 | while (todo) { | |
1554 | char __user *uaddr = ii.iov->iov_base + ii.iov_offset; | |
1555 | size_t iov_len = ii.iov->iov_len - ii.iov_offset; | |
1556 | size_t copy = min(todo, iov_len); | |
1557 | size_t left; | |
1558 | ||
1559 | if (!to_user) | |
1560 | left = copy_from_user(kaddr, uaddr, copy); | |
1561 | else | |
1562 | left = copy_to_user(uaddr, kaddr, copy); | |
1563 | ||
1564 | if (unlikely(left)) | |
1565 | return -EFAULT; | |
1566 | ||
1567 | iov_iter_advance(&ii, copy); | |
1568 | todo -= copy; | |
1569 | kaddr += copy; | |
1570 | } | |
1571 | ||
1572 | kunmap(map); | |
1573 | } | |
1574 | ||
1575 | return 0; | |
1576 | } | |
1577 | ||
1578 | /* | |
1579 | * For ioctls, there is no generic way to determine how much memory | |
1580 | * needs to be read and/or written. Furthermore, ioctls are allowed | |
1581 | * to dereference the passed pointer, so the parameter requires deep | |
1582 | * copying but FUSE has no idea whatsoever about what to copy in or | |
1583 | * out. | |
1584 | * | |
1585 | * This is solved by allowing FUSE server to retry ioctl with | |
1586 | * necessary in/out iovecs. Let's assume the ioctl implementation | |
1587 | * needs to read in the following structure. | |
1588 | * | |
1589 | * struct a { | |
1590 | * char *buf; | |
1591 | * size_t buflen; | |
1592 | * } | |
1593 | * | |
1594 | * On the first callout to FUSE server, inarg->in_size and | |
1595 | * inarg->out_size will be NULL; then, the server completes the ioctl | |
1596 | * with FUSE_IOCTL_RETRY set in out->flags, out->in_iovs set to 1 and | |
1597 | * the actual iov array to | |
1598 | * | |
1599 | * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) } } | |
1600 | * | |
1601 | * which tells FUSE to copy in the requested area and retry the ioctl. | |
1602 | * On the second round, the server has access to the structure and | |
1603 | * from that it can tell what to look for next, so on the invocation, | |
1604 | * it sets FUSE_IOCTL_RETRY, out->in_iovs to 2 and iov array to | |
1605 | * | |
1606 | * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) }, | |
1607 | * { .iov_base = a.buf, .iov_len = a.buflen } } | |
1608 | * | |
1609 | * FUSE will copy both struct a and the pointed buffer from the | |
1610 | * process doing the ioctl and retry ioctl with both struct a and the | |
1611 | * buffer. | |
1612 | * | |
1613 | * This time, FUSE server has everything it needs and completes ioctl | |
1614 | * without FUSE_IOCTL_RETRY which finishes the ioctl call. | |
1615 | * | |
1616 | * Copying data out works the same way. | |
1617 | * | |
1618 | * Note that if FUSE_IOCTL_UNRESTRICTED is clear, the kernel | |
1619 | * automatically initializes in and out iovs by decoding @cmd with | |
1620 | * _IOC_* macros and the server is not allowed to request RETRY. This | |
1621 | * limits ioctl data transfers to well-formed ioctls and is the forced | |
1622 | * behavior for all FUSE servers. | |
1623 | */ | |
1624 | static long fuse_file_do_ioctl(struct file *file, unsigned int cmd, | |
1625 | unsigned long arg, unsigned int flags) | |
1626 | { | |
1627 | struct inode *inode = file->f_dentry->d_inode; | |
1628 | struct fuse_file *ff = file->private_data; | |
1629 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1630 | struct fuse_ioctl_in inarg = { | |
1631 | .fh = ff->fh, | |
1632 | .cmd = cmd, | |
1633 | .arg = arg, | |
1634 | .flags = flags | |
1635 | }; | |
1636 | struct fuse_ioctl_out outarg; | |
1637 | struct fuse_req *req = NULL; | |
1638 | struct page **pages = NULL; | |
1639 | struct page *iov_page = NULL; | |
1640 | struct iovec *in_iov = NULL, *out_iov = NULL; | |
1641 | unsigned int in_iovs = 0, out_iovs = 0, num_pages = 0, max_pages; | |
1642 | size_t in_size, out_size, transferred; | |
1643 | int err; | |
1644 | ||
1645 | /* assume all the iovs returned by client always fits in a page */ | |
1646 | BUILD_BUG_ON(sizeof(struct iovec) * FUSE_IOCTL_MAX_IOV > PAGE_SIZE); | |
1647 | ||
1648 | if (!fuse_allow_task(fc, current)) | |
1649 | return -EACCES; | |
1650 | ||
1651 | err = -EIO; | |
1652 | if (is_bad_inode(inode)) | |
1653 | goto out; | |
1654 | ||
1655 | err = -ENOMEM; | |
1656 | pages = kzalloc(sizeof(pages[0]) * FUSE_MAX_PAGES_PER_REQ, GFP_KERNEL); | |
1657 | iov_page = alloc_page(GFP_KERNEL); | |
1658 | if (!pages || !iov_page) | |
1659 | goto out; | |
1660 | ||
1661 | /* | |
1662 | * If restricted, initialize IO parameters as encoded in @cmd. | |
1663 | * RETRY from server is not allowed. | |
1664 | */ | |
1665 | if (!(flags & FUSE_IOCTL_UNRESTRICTED)) { | |
1666 | struct iovec *iov = page_address(iov_page); | |
1667 | ||
c9f0523d | 1668 | iov->iov_base = (void __user *)arg; |
59efec7b TH |
1669 | iov->iov_len = _IOC_SIZE(cmd); |
1670 | ||
1671 | if (_IOC_DIR(cmd) & _IOC_WRITE) { | |
1672 | in_iov = iov; | |
1673 | in_iovs = 1; | |
1674 | } | |
1675 | ||
1676 | if (_IOC_DIR(cmd) & _IOC_READ) { | |
1677 | out_iov = iov; | |
1678 | out_iovs = 1; | |
1679 | } | |
1680 | } | |
1681 | ||
1682 | retry: | |
1683 | inarg.in_size = in_size = iov_length(in_iov, in_iovs); | |
1684 | inarg.out_size = out_size = iov_length(out_iov, out_iovs); | |
1685 | ||
1686 | /* | |
1687 | * Out data can be used either for actual out data or iovs, | |
1688 | * make sure there always is at least one page. | |
1689 | */ | |
1690 | out_size = max_t(size_t, out_size, PAGE_SIZE); | |
1691 | max_pages = DIV_ROUND_UP(max(in_size, out_size), PAGE_SIZE); | |
1692 | ||
1693 | /* make sure there are enough buffer pages and init request with them */ | |
1694 | err = -ENOMEM; | |
1695 | if (max_pages > FUSE_MAX_PAGES_PER_REQ) | |
1696 | goto out; | |
1697 | while (num_pages < max_pages) { | |
1698 | pages[num_pages] = alloc_page(GFP_KERNEL | __GFP_HIGHMEM); | |
1699 | if (!pages[num_pages]) | |
1700 | goto out; | |
1701 | num_pages++; | |
1702 | } | |
1703 | ||
1704 | req = fuse_get_req(fc); | |
1705 | if (IS_ERR(req)) { | |
1706 | err = PTR_ERR(req); | |
1707 | req = NULL; | |
1708 | goto out; | |
1709 | } | |
1710 | memcpy(req->pages, pages, sizeof(req->pages[0]) * num_pages); | |
1711 | req->num_pages = num_pages; | |
1712 | ||
1713 | /* okay, let's send it to the client */ | |
1714 | req->in.h.opcode = FUSE_IOCTL; | |
1715 | req->in.h.nodeid = get_node_id(inode); | |
1716 | req->in.numargs = 1; | |
1717 | req->in.args[0].size = sizeof(inarg); | |
1718 | req->in.args[0].value = &inarg; | |
1719 | if (in_size) { | |
1720 | req->in.numargs++; | |
1721 | req->in.args[1].size = in_size; | |
1722 | req->in.argpages = 1; | |
1723 | ||
1724 | err = fuse_ioctl_copy_user(pages, in_iov, in_iovs, in_size, | |
1725 | false); | |
1726 | if (err) | |
1727 | goto out; | |
1728 | } | |
1729 | ||
1730 | req->out.numargs = 2; | |
1731 | req->out.args[0].size = sizeof(outarg); | |
1732 | req->out.args[0].value = &outarg; | |
1733 | req->out.args[1].size = out_size; | |
1734 | req->out.argpages = 1; | |
1735 | req->out.argvar = 1; | |
1736 | ||
b93f858a | 1737 | fuse_request_send(fc, req); |
59efec7b TH |
1738 | err = req->out.h.error; |
1739 | transferred = req->out.args[1].size; | |
1740 | fuse_put_request(fc, req); | |
1741 | req = NULL; | |
1742 | if (err) | |
1743 | goto out; | |
1744 | ||
1745 | /* did it ask for retry? */ | |
1746 | if (outarg.flags & FUSE_IOCTL_RETRY) { | |
1747 | char *vaddr; | |
1748 | ||
1749 | /* no retry if in restricted mode */ | |
1750 | err = -EIO; | |
1751 | if (!(flags & FUSE_IOCTL_UNRESTRICTED)) | |
1752 | goto out; | |
1753 | ||
1754 | in_iovs = outarg.in_iovs; | |
1755 | out_iovs = outarg.out_iovs; | |
1756 | ||
1757 | /* | |
1758 | * Make sure things are in boundary, separate checks | |
1759 | * are to protect against overflow. | |
1760 | */ | |
1761 | err = -ENOMEM; | |
1762 | if (in_iovs > FUSE_IOCTL_MAX_IOV || | |
1763 | out_iovs > FUSE_IOCTL_MAX_IOV || | |
1764 | in_iovs + out_iovs > FUSE_IOCTL_MAX_IOV) | |
1765 | goto out; | |
1766 | ||
1767 | err = -EIO; | |
1768 | if ((in_iovs + out_iovs) * sizeof(struct iovec) != transferred) | |
1769 | goto out; | |
1770 | ||
1771 | /* okay, copy in iovs and retry */ | |
1772 | vaddr = kmap_atomic(pages[0], KM_USER0); | |
1773 | memcpy(page_address(iov_page), vaddr, transferred); | |
1774 | kunmap_atomic(vaddr, KM_USER0); | |
1775 | ||
1776 | in_iov = page_address(iov_page); | |
1777 | out_iov = in_iov + in_iovs; | |
1778 | ||
1779 | goto retry; | |
1780 | } | |
1781 | ||
1782 | err = -EIO; | |
1783 | if (transferred > inarg.out_size) | |
1784 | goto out; | |
1785 | ||
1786 | err = fuse_ioctl_copy_user(pages, out_iov, out_iovs, transferred, true); | |
1787 | out: | |
1788 | if (req) | |
1789 | fuse_put_request(fc, req); | |
1790 | if (iov_page) | |
1791 | __free_page(iov_page); | |
1792 | while (num_pages) | |
1793 | __free_page(pages[--num_pages]); | |
1794 | kfree(pages); | |
1795 | ||
1796 | return err ? err : outarg.result; | |
1797 | } | |
1798 | ||
1799 | static long fuse_file_ioctl(struct file *file, unsigned int cmd, | |
1800 | unsigned long arg) | |
1801 | { | |
1802 | return fuse_file_do_ioctl(file, cmd, arg, 0); | |
1803 | } | |
1804 | ||
1805 | static long fuse_file_compat_ioctl(struct file *file, unsigned int cmd, | |
1806 | unsigned long arg) | |
1807 | { | |
1808 | return fuse_file_do_ioctl(file, cmd, arg, FUSE_IOCTL_COMPAT); | |
1809 | } | |
1810 | ||
95668a69 TH |
1811 | /* |
1812 | * All files which have been polled are linked to RB tree | |
1813 | * fuse_conn->polled_files which is indexed by kh. Walk the tree and | |
1814 | * find the matching one. | |
1815 | */ | |
1816 | static struct rb_node **fuse_find_polled_node(struct fuse_conn *fc, u64 kh, | |
1817 | struct rb_node **parent_out) | |
1818 | { | |
1819 | struct rb_node **link = &fc->polled_files.rb_node; | |
1820 | struct rb_node *last = NULL; | |
1821 | ||
1822 | while (*link) { | |
1823 | struct fuse_file *ff; | |
1824 | ||
1825 | last = *link; | |
1826 | ff = rb_entry(last, struct fuse_file, polled_node); | |
1827 | ||
1828 | if (kh < ff->kh) | |
1829 | link = &last->rb_left; | |
1830 | else if (kh > ff->kh) | |
1831 | link = &last->rb_right; | |
1832 | else | |
1833 | return link; | |
1834 | } | |
1835 | ||
1836 | if (parent_out) | |
1837 | *parent_out = last; | |
1838 | return link; | |
1839 | } | |
1840 | ||
1841 | /* | |
1842 | * The file is about to be polled. Make sure it's on the polled_files | |
1843 | * RB tree. Note that files once added to the polled_files tree are | |
1844 | * not removed before the file is released. This is because a file | |
1845 | * polled once is likely to be polled again. | |
1846 | */ | |
1847 | static void fuse_register_polled_file(struct fuse_conn *fc, | |
1848 | struct fuse_file *ff) | |
1849 | { | |
1850 | spin_lock(&fc->lock); | |
1851 | if (RB_EMPTY_NODE(&ff->polled_node)) { | |
1852 | struct rb_node **link, *parent; | |
1853 | ||
1854 | link = fuse_find_polled_node(fc, ff->kh, &parent); | |
1855 | BUG_ON(*link); | |
1856 | rb_link_node(&ff->polled_node, parent, link); | |
1857 | rb_insert_color(&ff->polled_node, &fc->polled_files); | |
1858 | } | |
1859 | spin_unlock(&fc->lock); | |
1860 | } | |
1861 | ||
1862 | static unsigned fuse_file_poll(struct file *file, poll_table *wait) | |
1863 | { | |
1864 | struct inode *inode = file->f_dentry->d_inode; | |
1865 | struct fuse_file *ff = file->private_data; | |
1866 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1867 | struct fuse_poll_in inarg = { .fh = ff->fh, .kh = ff->kh }; | |
1868 | struct fuse_poll_out outarg; | |
1869 | struct fuse_req *req; | |
1870 | int err; | |
1871 | ||
1872 | if (fc->no_poll) | |
1873 | return DEFAULT_POLLMASK; | |
1874 | ||
1875 | poll_wait(file, &ff->poll_wait, wait); | |
1876 | ||
1877 | /* | |
1878 | * Ask for notification iff there's someone waiting for it. | |
1879 | * The client may ignore the flag and always notify. | |
1880 | */ | |
1881 | if (waitqueue_active(&ff->poll_wait)) { | |
1882 | inarg.flags |= FUSE_POLL_SCHEDULE_NOTIFY; | |
1883 | fuse_register_polled_file(fc, ff); | |
1884 | } | |
1885 | ||
1886 | req = fuse_get_req(fc); | |
1887 | if (IS_ERR(req)) | |
1888 | return PTR_ERR(req); | |
1889 | ||
1890 | req->in.h.opcode = FUSE_POLL; | |
1891 | req->in.h.nodeid = get_node_id(inode); | |
1892 | req->in.numargs = 1; | |
1893 | req->in.args[0].size = sizeof(inarg); | |
1894 | req->in.args[0].value = &inarg; | |
1895 | req->out.numargs = 1; | |
1896 | req->out.args[0].size = sizeof(outarg); | |
1897 | req->out.args[0].value = &outarg; | |
b93f858a | 1898 | fuse_request_send(fc, req); |
95668a69 TH |
1899 | err = req->out.h.error; |
1900 | fuse_put_request(fc, req); | |
1901 | ||
1902 | if (!err) | |
1903 | return outarg.revents; | |
1904 | if (err == -ENOSYS) { | |
1905 | fc->no_poll = 1; | |
1906 | return DEFAULT_POLLMASK; | |
1907 | } | |
1908 | return POLLERR; | |
1909 | } | |
1910 | ||
1911 | /* | |
1912 | * This is called from fuse_handle_notify() on FUSE_NOTIFY_POLL and | |
1913 | * wakes up the poll waiters. | |
1914 | */ | |
1915 | int fuse_notify_poll_wakeup(struct fuse_conn *fc, | |
1916 | struct fuse_notify_poll_wakeup_out *outarg) | |
1917 | { | |
1918 | u64 kh = outarg->kh; | |
1919 | struct rb_node **link; | |
1920 | ||
1921 | spin_lock(&fc->lock); | |
1922 | ||
1923 | link = fuse_find_polled_node(fc, kh, NULL); | |
1924 | if (*link) { | |
1925 | struct fuse_file *ff; | |
1926 | ||
1927 | ff = rb_entry(*link, struct fuse_file, polled_node); | |
1928 | wake_up_interruptible_sync(&ff->poll_wait); | |
1929 | } | |
1930 | ||
1931 | spin_unlock(&fc->lock); | |
1932 | return 0; | |
1933 | } | |
1934 | ||
4b6f5d20 | 1935 | static const struct file_operations fuse_file_operations = { |
5559b8f4 | 1936 | .llseek = fuse_file_llseek, |
543ade1f | 1937 | .read = do_sync_read, |
bcb4be80 | 1938 | .aio_read = fuse_file_aio_read, |
543ade1f | 1939 | .write = do_sync_write, |
ea9b9907 | 1940 | .aio_write = fuse_file_aio_write, |
b6aeaded MS |
1941 | .mmap = fuse_file_mmap, |
1942 | .open = fuse_open, | |
1943 | .flush = fuse_flush, | |
1944 | .release = fuse_release, | |
1945 | .fsync = fuse_fsync, | |
71421259 | 1946 | .lock = fuse_file_lock, |
a9ff4f87 | 1947 | .flock = fuse_file_flock, |
5ffc4ef4 | 1948 | .splice_read = generic_file_splice_read, |
59efec7b TH |
1949 | .unlocked_ioctl = fuse_file_ioctl, |
1950 | .compat_ioctl = fuse_file_compat_ioctl, | |
95668a69 | 1951 | .poll = fuse_file_poll, |
b6aeaded MS |
1952 | }; |
1953 | ||
4b6f5d20 | 1954 | static const struct file_operations fuse_direct_io_file_operations = { |
5559b8f4 | 1955 | .llseek = fuse_file_llseek, |
413ef8cb MS |
1956 | .read = fuse_direct_read, |
1957 | .write = fuse_direct_write, | |
fc280c96 | 1958 | .mmap = fuse_direct_mmap, |
413ef8cb MS |
1959 | .open = fuse_open, |
1960 | .flush = fuse_flush, | |
1961 | .release = fuse_release, | |
1962 | .fsync = fuse_fsync, | |
71421259 | 1963 | .lock = fuse_file_lock, |
a9ff4f87 | 1964 | .flock = fuse_file_flock, |
59efec7b TH |
1965 | .unlocked_ioctl = fuse_file_ioctl, |
1966 | .compat_ioctl = fuse_file_compat_ioctl, | |
95668a69 | 1967 | .poll = fuse_file_poll, |
fc280c96 | 1968 | /* no splice_read */ |
413ef8cb MS |
1969 | }; |
1970 | ||
f5e54d6e | 1971 | static const struct address_space_operations fuse_file_aops = { |
b6aeaded | 1972 | .readpage = fuse_readpage, |
3be5a52b MS |
1973 | .writepage = fuse_writepage, |
1974 | .launder_page = fuse_launder_page, | |
5e6f58a1 NP |
1975 | .write_begin = fuse_write_begin, |
1976 | .write_end = fuse_write_end, | |
db50b96c | 1977 | .readpages = fuse_readpages, |
3be5a52b | 1978 | .set_page_dirty = __set_page_dirty_nobuffers, |
b2d2272f | 1979 | .bmap = fuse_bmap, |
b6aeaded MS |
1980 | }; |
1981 | ||
1982 | void fuse_init_file_inode(struct inode *inode) | |
1983 | { | |
45323fb7 MS |
1984 | inode->i_fop = &fuse_file_operations; |
1985 | inode->i_data.a_ops = &fuse_file_aops; | |
b6aeaded | 1986 | } |