]>
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 | |
91fe96b4 MS |
18 | static int fuse_send_open(struct fuse_conn *fc, u64 nodeid, struct file *file, |
19 | int opcode, struct fuse_open_out *outargp) | |
b6aeaded | 20 | { |
b6aeaded | 21 | struct fuse_open_in inarg; |
fd72faac MS |
22 | struct fuse_req *req; |
23 | int err; | |
24 | ||
ce1d5a49 MS |
25 | req = fuse_get_req(fc); |
26 | if (IS_ERR(req)) | |
27 | return PTR_ERR(req); | |
fd72faac MS |
28 | |
29 | memset(&inarg, 0, sizeof(inarg)); | |
6ff958ed MS |
30 | inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY); |
31 | if (!fc->atomic_o_trunc) | |
32 | inarg.flags &= ~O_TRUNC; | |
91fe96b4 MS |
33 | req->in.h.opcode = opcode; |
34 | req->in.h.nodeid = nodeid; | |
fd72faac MS |
35 | req->in.numargs = 1; |
36 | req->in.args[0].size = sizeof(inarg); | |
37 | req->in.args[0].value = &inarg; | |
38 | req->out.numargs = 1; | |
39 | req->out.args[0].size = sizeof(*outargp); | |
40 | req->out.args[0].value = outargp; | |
b93f858a | 41 | fuse_request_send(fc, req); |
fd72faac MS |
42 | err = req->out.h.error; |
43 | fuse_put_request(fc, req); | |
44 | ||
45 | return err; | |
46 | } | |
47 | ||
acf99433 | 48 | struct fuse_file *fuse_file_alloc(struct fuse_conn *fc) |
fd72faac MS |
49 | { |
50 | struct fuse_file *ff; | |
6b2db28a | 51 | |
fd72faac | 52 | ff = kmalloc(sizeof(struct fuse_file), GFP_KERNEL); |
6b2db28a TH |
53 | if (unlikely(!ff)) |
54 | return NULL; | |
55 | ||
da5e4714 | 56 | ff->fc = fc; |
6b2db28a TH |
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 | ||
c7b7143c | 81 | struct fuse_file *fuse_file_get(struct fuse_file *ff) |
c756e0a4 MS |
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 | ||
91fe96b4 MS |
104 | static int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file, |
105 | bool isdir) | |
106 | { | |
107 | struct fuse_open_out outarg; | |
108 | struct fuse_file *ff; | |
109 | int err; | |
110 | int opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN; | |
111 | ||
112 | ff = fuse_file_alloc(fc); | |
113 | if (!ff) | |
114 | return -ENOMEM; | |
115 | ||
116 | err = fuse_send_open(fc, nodeid, file, opcode, &outarg); | |
117 | if (err) { | |
118 | fuse_file_free(ff); | |
119 | return err; | |
120 | } | |
121 | ||
122 | if (isdir) | |
123 | outarg.open_flags &= ~FOPEN_DIRECT_IO; | |
124 | ||
125 | ff->fh = outarg.fh; | |
126 | ff->nodeid = nodeid; | |
127 | ff->open_flags = outarg.open_flags; | |
128 | file->private_data = fuse_file_get(ff); | |
129 | ||
130 | return 0; | |
131 | } | |
132 | ||
c7b7143c | 133 | void fuse_finish_open(struct inode *inode, struct file *file) |
fd72faac | 134 | { |
c7b7143c MS |
135 | struct fuse_file *ff = file->private_data; |
136 | ||
137 | if (ff->open_flags & FOPEN_DIRECT_IO) | |
fd72faac | 138 | file->f_op = &fuse_direct_io_file_operations; |
c7b7143c | 139 | if (!(ff->open_flags & FOPEN_KEEP_CACHE)) |
b1009979 | 140 | invalidate_inode_pages2(inode->i_mapping); |
c7b7143c | 141 | if (ff->open_flags & FOPEN_NONSEEKABLE) |
a7c1b990 | 142 | nonseekable_open(inode, file); |
fd72faac MS |
143 | } |
144 | ||
91fe96b4 | 145 | int fuse_open_common(struct inode *inode, struct file *file, bool isdir) |
fd72faac | 146 | { |
acf99433 | 147 | struct fuse_conn *fc = get_fuse_conn(inode); |
b6aeaded | 148 | int err; |
b6aeaded | 149 | |
dd190d06 MS |
150 | /* VFS checks this, but only _after_ ->open() */ |
151 | if (file->f_flags & O_DIRECT) | |
152 | return -EINVAL; | |
153 | ||
b6aeaded MS |
154 | err = generic_file_open(inode, file); |
155 | if (err) | |
156 | return err; | |
157 | ||
91fe96b4 | 158 | err = fuse_do_open(fc, get_node_id(inode), file, isdir); |
fd72faac | 159 | if (err) |
91fe96b4 | 160 | return err; |
b6aeaded | 161 | |
91fe96b4 MS |
162 | fuse_finish_open(inode, file); |
163 | ||
164 | return 0; | |
b6aeaded MS |
165 | } |
166 | ||
c7b7143c | 167 | void fuse_release_fill(struct fuse_file *ff, int flags, int opcode) |
64c6d8ed | 168 | { |
33649c91 | 169 | struct fuse_req *req = ff->reserved_req; |
b57d4264 | 170 | struct fuse_release_in *inarg = &req->misc.release.in; |
b6aeaded MS |
171 | |
172 | inarg->fh = ff->fh; | |
fd72faac | 173 | inarg->flags = flags; |
51eb01e7 | 174 | req->in.h.opcode = opcode; |
c7b7143c | 175 | req->in.h.nodeid = ff->nodeid; |
b6aeaded MS |
176 | req->in.numargs = 1; |
177 | req->in.args[0].size = sizeof(struct fuse_release_in); | |
178 | req->in.args[0].value = inarg; | |
fd72faac MS |
179 | } |
180 | ||
181 | int fuse_release_common(struct inode *inode, struct file *file, int isdir) | |
182 | { | |
6b2db28a TH |
183 | struct fuse_conn *fc; |
184 | struct fuse_file *ff; | |
185 | struct fuse_req *req; | |
b6aeaded | 186 | |
6b2db28a TH |
187 | ff = file->private_data; |
188 | if (unlikely(!ff)) | |
189 | return 0; /* return value is ignored by VFS */ | |
190 | ||
191 | fc = get_fuse_conn(inode); | |
192 | req = ff->reserved_req; | |
193 | ||
c7b7143c | 194 | fuse_release_fill(ff, file->f_flags, |
6b2db28a TH |
195 | isdir ? FUSE_RELEASEDIR : FUSE_RELEASE); |
196 | ||
197 | /* Hold vfsmount and dentry until release is finished */ | |
b0be46eb MS |
198 | path_get(&file->f_path); |
199 | req->misc.release.path = file->f_path; | |
6b2db28a TH |
200 | |
201 | spin_lock(&fc->lock); | |
202 | list_del(&ff->write_entry); | |
203 | if (!RB_EMPTY_NODE(&ff->polled_node)) | |
204 | rb_erase(&ff->polled_node, &fc->polled_files); | |
205 | spin_unlock(&fc->lock); | |
206 | ||
207 | wake_up_interruptible_sync(&ff->poll_wait); | |
208 | /* | |
209 | * Normally this will send the RELEASE request, however if | |
210 | * some asynchronous READ or WRITE requests are outstanding, | |
211 | * the sending will be delayed. | |
212 | */ | |
213 | fuse_file_put(ff); | |
b6aeaded MS |
214 | return 0; |
215 | } | |
216 | ||
04730fef MS |
217 | static int fuse_open(struct inode *inode, struct file *file) |
218 | { | |
91fe96b4 | 219 | return fuse_open_common(inode, file, false); |
04730fef MS |
220 | } |
221 | ||
222 | static int fuse_release(struct inode *inode, struct file *file) | |
223 | { | |
224 | return fuse_release_common(inode, file, 0); | |
225 | } | |
226 | ||
71421259 | 227 | /* |
9c8ef561 MS |
228 | * Scramble the ID space with XTEA, so that the value of the files_struct |
229 | * pointer is not exposed to userspace. | |
71421259 | 230 | */ |
f3332114 | 231 | u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id) |
71421259 | 232 | { |
9c8ef561 MS |
233 | u32 *k = fc->scramble_key; |
234 | u64 v = (unsigned long) id; | |
235 | u32 v0 = v; | |
236 | u32 v1 = v >> 32; | |
237 | u32 sum = 0; | |
238 | int i; | |
239 | ||
240 | for (i = 0; i < 32; i++) { | |
241 | v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]); | |
242 | sum += 0x9E3779B9; | |
243 | v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]); | |
244 | } | |
245 | ||
246 | return (u64) v0 + ((u64) v1 << 32); | |
71421259 MS |
247 | } |
248 | ||
3be5a52b MS |
249 | /* |
250 | * Check if page is under writeback | |
251 | * | |
252 | * This is currently done by walking the list of writepage requests | |
253 | * for the inode, which can be pretty inefficient. | |
254 | */ | |
255 | static bool fuse_page_is_writeback(struct inode *inode, pgoff_t index) | |
256 | { | |
257 | struct fuse_conn *fc = get_fuse_conn(inode); | |
258 | struct fuse_inode *fi = get_fuse_inode(inode); | |
259 | struct fuse_req *req; | |
260 | bool found = false; | |
261 | ||
262 | spin_lock(&fc->lock); | |
263 | list_for_each_entry(req, &fi->writepages, writepages_entry) { | |
264 | pgoff_t curr_index; | |
265 | ||
266 | BUG_ON(req->inode != inode); | |
267 | curr_index = req->misc.write.in.offset >> PAGE_CACHE_SHIFT; | |
268 | if (curr_index == index) { | |
269 | found = true; | |
270 | break; | |
271 | } | |
272 | } | |
273 | spin_unlock(&fc->lock); | |
274 | ||
275 | return found; | |
276 | } | |
277 | ||
278 | /* | |
279 | * Wait for page writeback to be completed. | |
280 | * | |
281 | * Since fuse doesn't rely on the VM writeback tracking, this has to | |
282 | * use some other means. | |
283 | */ | |
284 | static int fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index) | |
285 | { | |
286 | struct fuse_inode *fi = get_fuse_inode(inode); | |
287 | ||
288 | wait_event(fi->page_waitq, !fuse_page_is_writeback(inode, index)); | |
289 | return 0; | |
290 | } | |
291 | ||
75e1fcc0 | 292 | static int fuse_flush(struct file *file, fl_owner_t id) |
b6aeaded | 293 | { |
7706a9d6 | 294 | struct inode *inode = file->f_path.dentry->d_inode; |
b6aeaded MS |
295 | struct fuse_conn *fc = get_fuse_conn(inode); |
296 | struct fuse_file *ff = file->private_data; | |
297 | struct fuse_req *req; | |
298 | struct fuse_flush_in inarg; | |
299 | int err; | |
300 | ||
248d86e8 MS |
301 | if (is_bad_inode(inode)) |
302 | return -EIO; | |
303 | ||
b6aeaded MS |
304 | if (fc->no_flush) |
305 | return 0; | |
306 | ||
33649c91 | 307 | req = fuse_get_req_nofail(fc, file); |
b6aeaded MS |
308 | memset(&inarg, 0, sizeof(inarg)); |
309 | inarg.fh = ff->fh; | |
9c8ef561 | 310 | inarg.lock_owner = fuse_lock_owner_id(fc, id); |
b6aeaded MS |
311 | req->in.h.opcode = FUSE_FLUSH; |
312 | req->in.h.nodeid = get_node_id(inode); | |
b6aeaded MS |
313 | req->in.numargs = 1; |
314 | req->in.args[0].size = sizeof(inarg); | |
315 | req->in.args[0].value = &inarg; | |
71421259 | 316 | req->force = 1; |
b93f858a | 317 | fuse_request_send(fc, req); |
b6aeaded MS |
318 | err = req->out.h.error; |
319 | fuse_put_request(fc, req); | |
320 | if (err == -ENOSYS) { | |
321 | fc->no_flush = 1; | |
322 | err = 0; | |
323 | } | |
324 | return err; | |
325 | } | |
326 | ||
3be5a52b MS |
327 | /* |
328 | * Wait for all pending writepages on the inode to finish. | |
329 | * | |
330 | * This is currently done by blocking further writes with FUSE_NOWRITE | |
331 | * and waiting for all sent writes to complete. | |
332 | * | |
333 | * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage | |
334 | * could conflict with truncation. | |
335 | */ | |
336 | static void fuse_sync_writes(struct inode *inode) | |
337 | { | |
338 | fuse_set_nowrite(inode); | |
339 | fuse_release_nowrite(inode); | |
340 | } | |
341 | ||
82547981 MS |
342 | int fuse_fsync_common(struct file *file, struct dentry *de, int datasync, |
343 | int isdir) | |
b6aeaded MS |
344 | { |
345 | struct inode *inode = de->d_inode; | |
346 | struct fuse_conn *fc = get_fuse_conn(inode); | |
347 | struct fuse_file *ff = file->private_data; | |
348 | struct fuse_req *req; | |
349 | struct fuse_fsync_in inarg; | |
350 | int err; | |
351 | ||
248d86e8 MS |
352 | if (is_bad_inode(inode)) |
353 | return -EIO; | |
354 | ||
82547981 | 355 | if ((!isdir && fc->no_fsync) || (isdir && fc->no_fsyncdir)) |
b6aeaded MS |
356 | return 0; |
357 | ||
3be5a52b MS |
358 | /* |
359 | * Start writeback against all dirty pages of the inode, then | |
360 | * wait for all outstanding writes, before sending the FSYNC | |
361 | * request. | |
362 | */ | |
363 | err = write_inode_now(inode, 0); | |
364 | if (err) | |
365 | return err; | |
366 | ||
367 | fuse_sync_writes(inode); | |
368 | ||
ce1d5a49 MS |
369 | req = fuse_get_req(fc); |
370 | if (IS_ERR(req)) | |
371 | return PTR_ERR(req); | |
b6aeaded MS |
372 | |
373 | memset(&inarg, 0, sizeof(inarg)); | |
374 | inarg.fh = ff->fh; | |
375 | inarg.fsync_flags = datasync ? 1 : 0; | |
82547981 | 376 | req->in.h.opcode = isdir ? FUSE_FSYNCDIR : FUSE_FSYNC; |
b6aeaded | 377 | req->in.h.nodeid = get_node_id(inode); |
b6aeaded MS |
378 | req->in.numargs = 1; |
379 | req->in.args[0].size = sizeof(inarg); | |
380 | req->in.args[0].value = &inarg; | |
b93f858a | 381 | fuse_request_send(fc, req); |
b6aeaded MS |
382 | err = req->out.h.error; |
383 | fuse_put_request(fc, req); | |
384 | if (err == -ENOSYS) { | |
82547981 MS |
385 | if (isdir) |
386 | fc->no_fsyncdir = 1; | |
387 | else | |
388 | fc->no_fsync = 1; | |
b6aeaded MS |
389 | err = 0; |
390 | } | |
391 | return err; | |
392 | } | |
393 | ||
82547981 MS |
394 | static int fuse_fsync(struct file *file, struct dentry *de, int datasync) |
395 | { | |
396 | return fuse_fsync_common(file, de, datasync, 0); | |
397 | } | |
398 | ||
2106cb18 MS |
399 | void fuse_read_fill(struct fuse_req *req, struct file *file, loff_t pos, |
400 | size_t count, int opcode) | |
b6aeaded | 401 | { |
5c5c5e51 | 402 | struct fuse_read_in *inarg = &req->misc.read.in; |
a6643094 | 403 | struct fuse_file *ff = file->private_data; |
b6aeaded | 404 | |
361b1eb5 MS |
405 | inarg->fh = ff->fh; |
406 | inarg->offset = pos; | |
407 | inarg->size = count; | |
a6643094 | 408 | inarg->flags = file->f_flags; |
361b1eb5 | 409 | req->in.h.opcode = opcode; |
2106cb18 | 410 | req->in.h.nodeid = ff->nodeid; |
b6aeaded MS |
411 | req->in.numargs = 1; |
412 | req->in.args[0].size = sizeof(struct fuse_read_in); | |
c1aa96a5 | 413 | req->in.args[0].value = inarg; |
b6aeaded MS |
414 | req->out.argvar = 1; |
415 | req->out.numargs = 1; | |
416 | req->out.args[0].size = count; | |
b6aeaded MS |
417 | } |
418 | ||
8bfc016d | 419 | static size_t fuse_send_read(struct fuse_req *req, struct file *file, |
2106cb18 | 420 | loff_t pos, size_t count, fl_owner_t owner) |
04730fef | 421 | { |
2106cb18 MS |
422 | struct fuse_file *ff = file->private_data; |
423 | struct fuse_conn *fc = ff->fc; | |
f3332114 | 424 | |
2106cb18 | 425 | fuse_read_fill(req, file, pos, count, FUSE_READ); |
f3332114 | 426 | if (owner != NULL) { |
5c5c5e51 | 427 | struct fuse_read_in *inarg = &req->misc.read.in; |
f3332114 MS |
428 | |
429 | inarg->read_flags |= FUSE_READ_LOCKOWNER; | |
430 | inarg->lock_owner = fuse_lock_owner_id(fc, owner); | |
431 | } | |
b93f858a | 432 | fuse_request_send(fc, req); |
361b1eb5 | 433 | return req->out.args[0].size; |
04730fef MS |
434 | } |
435 | ||
5c5c5e51 MS |
436 | static void fuse_read_update_size(struct inode *inode, loff_t size, |
437 | u64 attr_ver) | |
438 | { | |
439 | struct fuse_conn *fc = get_fuse_conn(inode); | |
440 | struct fuse_inode *fi = get_fuse_inode(inode); | |
441 | ||
442 | spin_lock(&fc->lock); | |
443 | if (attr_ver == fi->attr_version && size < inode->i_size) { | |
444 | fi->attr_version = ++fc->attr_version; | |
445 | i_size_write(inode, size); | |
446 | } | |
447 | spin_unlock(&fc->lock); | |
448 | } | |
449 | ||
b6aeaded MS |
450 | static int fuse_readpage(struct file *file, struct page *page) |
451 | { | |
452 | struct inode *inode = page->mapping->host; | |
453 | struct fuse_conn *fc = get_fuse_conn(inode); | |
248d86e8 | 454 | struct fuse_req *req; |
5c5c5e51 MS |
455 | size_t num_read; |
456 | loff_t pos = page_offset(page); | |
457 | size_t count = PAGE_CACHE_SIZE; | |
458 | u64 attr_ver; | |
248d86e8 MS |
459 | int err; |
460 | ||
461 | err = -EIO; | |
462 | if (is_bad_inode(inode)) | |
463 | goto out; | |
464 | ||
3be5a52b MS |
465 | /* |
466 | * Page writeback can extend beyond the liftime of the | |
467 | * page-cache page, so make sure we read a properly synced | |
468 | * page. | |
469 | */ | |
470 | fuse_wait_on_page_writeback(inode, page->index); | |
471 | ||
ce1d5a49 MS |
472 | req = fuse_get_req(fc); |
473 | err = PTR_ERR(req); | |
474 | if (IS_ERR(req)) | |
b6aeaded MS |
475 | goto out; |
476 | ||
5c5c5e51 MS |
477 | attr_ver = fuse_get_attr_version(fc); |
478 | ||
b6aeaded | 479 | req->out.page_zeroing = 1; |
f4975c67 | 480 | req->out.argpages = 1; |
b6aeaded MS |
481 | req->num_pages = 1; |
482 | req->pages[0] = page; | |
2106cb18 | 483 | num_read = fuse_send_read(req, file, pos, count, NULL); |
b6aeaded MS |
484 | err = req->out.h.error; |
485 | fuse_put_request(fc, req); | |
5c5c5e51 MS |
486 | |
487 | if (!err) { | |
488 | /* | |
489 | * Short read means EOF. If file size is larger, truncate it | |
490 | */ | |
491 | if (num_read < count) | |
492 | fuse_read_update_size(inode, pos + num_read, attr_ver); | |
493 | ||
b6aeaded | 494 | SetPageUptodate(page); |
5c5c5e51 MS |
495 | } |
496 | ||
b36c31ba | 497 | fuse_invalidate_attr(inode); /* atime changed */ |
b6aeaded MS |
498 | out: |
499 | unlock_page(page); | |
500 | return err; | |
501 | } | |
502 | ||
c1aa96a5 | 503 | static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req) |
db50b96c | 504 | { |
c1aa96a5 | 505 | int i; |
5c5c5e51 MS |
506 | size_t count = req->misc.read.in.size; |
507 | size_t num_read = req->out.args[0].size; | |
508 | struct inode *inode = req->pages[0]->mapping->host; | |
c1aa96a5 | 509 | |
5c5c5e51 MS |
510 | /* |
511 | * Short read means EOF. If file size is larger, truncate it | |
512 | */ | |
513 | if (!req->out.h.error && num_read < count) { | |
514 | loff_t pos = page_offset(req->pages[0]) + num_read; | |
515 | fuse_read_update_size(inode, pos, req->misc.read.attr_ver); | |
516 | } | |
517 | ||
518 | fuse_invalidate_attr(inode); /* atime changed */ | |
c1aa96a5 | 519 | |
db50b96c MS |
520 | for (i = 0; i < req->num_pages; i++) { |
521 | struct page *page = req->pages[i]; | |
522 | if (!req->out.h.error) | |
523 | SetPageUptodate(page); | |
c1aa96a5 MS |
524 | else |
525 | SetPageError(page); | |
db50b96c MS |
526 | unlock_page(page); |
527 | } | |
c756e0a4 MS |
528 | if (req->ff) |
529 | fuse_file_put(req->ff); | |
c1aa96a5 MS |
530 | } |
531 | ||
2106cb18 | 532 | static void fuse_send_readpages(struct fuse_req *req, struct file *file) |
c1aa96a5 | 533 | { |
2106cb18 MS |
534 | struct fuse_file *ff = file->private_data; |
535 | struct fuse_conn *fc = ff->fc; | |
c1aa96a5 MS |
536 | loff_t pos = page_offset(req->pages[0]); |
537 | size_t count = req->num_pages << PAGE_CACHE_SHIFT; | |
f4975c67 MS |
538 | |
539 | req->out.argpages = 1; | |
c1aa96a5 | 540 | req->out.page_zeroing = 1; |
2106cb18 | 541 | fuse_read_fill(req, file, pos, count, FUSE_READ); |
5c5c5e51 | 542 | req->misc.read.attr_ver = fuse_get_attr_version(fc); |
9cd68455 | 543 | if (fc->async_read) { |
c756e0a4 | 544 | req->ff = fuse_file_get(ff); |
9cd68455 | 545 | req->end = fuse_readpages_end; |
b93f858a | 546 | fuse_request_send_background(fc, req); |
9cd68455 | 547 | } else { |
b93f858a | 548 | fuse_request_send(fc, req); |
9cd68455 | 549 | fuse_readpages_end(fc, req); |
e9bb09dd | 550 | fuse_put_request(fc, req); |
9cd68455 | 551 | } |
db50b96c MS |
552 | } |
553 | ||
c756e0a4 | 554 | struct fuse_fill_data { |
db50b96c | 555 | struct fuse_req *req; |
a6643094 | 556 | struct file *file; |
db50b96c MS |
557 | struct inode *inode; |
558 | }; | |
559 | ||
560 | static int fuse_readpages_fill(void *_data, struct page *page) | |
561 | { | |
c756e0a4 | 562 | struct fuse_fill_data *data = _data; |
db50b96c MS |
563 | struct fuse_req *req = data->req; |
564 | struct inode *inode = data->inode; | |
565 | struct fuse_conn *fc = get_fuse_conn(inode); | |
566 | ||
3be5a52b MS |
567 | fuse_wait_on_page_writeback(inode, page->index); |
568 | ||
db50b96c MS |
569 | if (req->num_pages && |
570 | (req->num_pages == FUSE_MAX_PAGES_PER_REQ || | |
571 | (req->num_pages + 1) * PAGE_CACHE_SIZE > fc->max_read || | |
572 | req->pages[req->num_pages - 1]->index + 1 != page->index)) { | |
2106cb18 | 573 | fuse_send_readpages(req, data->file); |
ce1d5a49 MS |
574 | data->req = req = fuse_get_req(fc); |
575 | if (IS_ERR(req)) { | |
db50b96c | 576 | unlock_page(page); |
ce1d5a49 | 577 | return PTR_ERR(req); |
db50b96c | 578 | } |
db50b96c MS |
579 | } |
580 | req->pages[req->num_pages] = page; | |
1729a16c | 581 | req->num_pages++; |
db50b96c MS |
582 | return 0; |
583 | } | |
584 | ||
585 | static int fuse_readpages(struct file *file, struct address_space *mapping, | |
586 | struct list_head *pages, unsigned nr_pages) | |
587 | { | |
588 | struct inode *inode = mapping->host; | |
589 | struct fuse_conn *fc = get_fuse_conn(inode); | |
c756e0a4 | 590 | struct fuse_fill_data data; |
db50b96c | 591 | int err; |
248d86e8 | 592 | |
1d7ea732 | 593 | err = -EIO; |
248d86e8 | 594 | if (is_bad_inode(inode)) |
2e990021 | 595 | goto out; |
248d86e8 | 596 | |
a6643094 | 597 | data.file = file; |
db50b96c | 598 | data.inode = inode; |
ce1d5a49 | 599 | data.req = fuse_get_req(fc); |
1d7ea732 | 600 | err = PTR_ERR(data.req); |
ce1d5a49 | 601 | if (IS_ERR(data.req)) |
2e990021 | 602 | goto out; |
db50b96c MS |
603 | |
604 | err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data); | |
d3406ffa MS |
605 | if (!err) { |
606 | if (data.req->num_pages) | |
2106cb18 | 607 | fuse_send_readpages(data.req, file); |
d3406ffa MS |
608 | else |
609 | fuse_put_request(fc, data.req); | |
610 | } | |
2e990021 | 611 | out: |
1d7ea732 | 612 | return err; |
db50b96c MS |
613 | } |
614 | ||
bcb4be80 MS |
615 | static ssize_t fuse_file_aio_read(struct kiocb *iocb, const struct iovec *iov, |
616 | unsigned long nr_segs, loff_t pos) | |
617 | { | |
618 | struct inode *inode = iocb->ki_filp->f_mapping->host; | |
619 | ||
620 | if (pos + iov_length(iov, nr_segs) > i_size_read(inode)) { | |
621 | int err; | |
622 | /* | |
623 | * If trying to read past EOF, make sure the i_size | |
624 | * attribute is up-to-date. | |
625 | */ | |
626 | err = fuse_update_attributes(inode, NULL, iocb->ki_filp, NULL); | |
627 | if (err) | |
628 | return err; | |
629 | } | |
630 | ||
631 | return generic_file_aio_read(iocb, iov, nr_segs, pos); | |
632 | } | |
633 | ||
2d698b07 | 634 | static void fuse_write_fill(struct fuse_req *req, struct fuse_file *ff, |
2106cb18 | 635 | loff_t pos, size_t count) |
b6aeaded | 636 | { |
b25e82e5 MS |
637 | struct fuse_write_in *inarg = &req->misc.write.in; |
638 | struct fuse_write_out *outarg = &req->misc.write.out; | |
b6aeaded | 639 | |
b25e82e5 MS |
640 | inarg->fh = ff->fh; |
641 | inarg->offset = pos; | |
642 | inarg->size = count; | |
b6aeaded | 643 | req->in.h.opcode = FUSE_WRITE; |
2106cb18 | 644 | req->in.h.nodeid = ff->nodeid; |
b6aeaded | 645 | req->in.numargs = 2; |
2106cb18 | 646 | if (ff->fc->minor < 9) |
f3332114 MS |
647 | req->in.args[0].size = FUSE_COMPAT_WRITE_IN_SIZE; |
648 | else | |
649 | req->in.args[0].size = sizeof(struct fuse_write_in); | |
b25e82e5 | 650 | req->in.args[0].value = inarg; |
b6aeaded MS |
651 | req->in.args[1].size = count; |
652 | req->out.numargs = 1; | |
653 | req->out.args[0].size = sizeof(struct fuse_write_out); | |
b25e82e5 MS |
654 | req->out.args[0].value = outarg; |
655 | } | |
656 | ||
657 | static size_t fuse_send_write(struct fuse_req *req, struct file *file, | |
2106cb18 | 658 | loff_t pos, size_t count, fl_owner_t owner) |
b25e82e5 | 659 | { |
2106cb18 MS |
660 | struct fuse_file *ff = file->private_data; |
661 | struct fuse_conn *fc = ff->fc; | |
2d698b07 MS |
662 | struct fuse_write_in *inarg = &req->misc.write.in; |
663 | ||
2106cb18 | 664 | fuse_write_fill(req, ff, pos, count); |
2d698b07 | 665 | inarg->flags = file->f_flags; |
f3332114 | 666 | if (owner != NULL) { |
f3332114 MS |
667 | inarg->write_flags |= FUSE_WRITE_LOCKOWNER; |
668 | inarg->lock_owner = fuse_lock_owner_id(fc, owner); | |
669 | } | |
b93f858a | 670 | fuse_request_send(fc, req); |
b25e82e5 | 671 | return req->misc.write.out.size; |
b6aeaded MS |
672 | } |
673 | ||
5e6f58a1 NP |
674 | static int fuse_write_begin(struct file *file, struct address_space *mapping, |
675 | loff_t pos, unsigned len, unsigned flags, | |
676 | struct page **pagep, void **fsdata) | |
b6aeaded | 677 | { |
5e6f58a1 NP |
678 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
679 | ||
54566b2c | 680 | *pagep = grab_cache_page_write_begin(mapping, index, flags); |
5e6f58a1 NP |
681 | if (!*pagep) |
682 | return -ENOMEM; | |
b6aeaded MS |
683 | return 0; |
684 | } | |
685 | ||
854512ec MS |
686 | static void fuse_write_update_size(struct inode *inode, loff_t pos) |
687 | { | |
688 | struct fuse_conn *fc = get_fuse_conn(inode); | |
689 | struct fuse_inode *fi = get_fuse_inode(inode); | |
690 | ||
691 | spin_lock(&fc->lock); | |
692 | fi->attr_version = ++fc->attr_version; | |
693 | if (pos > inode->i_size) | |
694 | i_size_write(inode, pos); | |
695 | spin_unlock(&fc->lock); | |
696 | } | |
697 | ||
5e6f58a1 NP |
698 | static int fuse_buffered_write(struct file *file, struct inode *inode, |
699 | loff_t pos, unsigned count, struct page *page) | |
b6aeaded MS |
700 | { |
701 | int err; | |
04730fef | 702 | size_t nres; |
b6aeaded | 703 | struct fuse_conn *fc = get_fuse_conn(inode); |
5e6f58a1 | 704 | unsigned offset = pos & (PAGE_CACHE_SIZE - 1); |
248d86e8 MS |
705 | struct fuse_req *req; |
706 | ||
707 | if (is_bad_inode(inode)) | |
708 | return -EIO; | |
709 | ||
3be5a52b MS |
710 | /* |
711 | * Make sure writepages on the same page are not mixed up with | |
712 | * plain writes. | |
713 | */ | |
714 | fuse_wait_on_page_writeback(inode, page->index); | |
715 | ||
ce1d5a49 MS |
716 | req = fuse_get_req(fc); |
717 | if (IS_ERR(req)) | |
718 | return PTR_ERR(req); | |
b6aeaded | 719 | |
f4975c67 | 720 | req->in.argpages = 1; |
b6aeaded MS |
721 | req->num_pages = 1; |
722 | req->pages[0] = page; | |
723 | req->page_offset = offset; | |
2106cb18 | 724 | nres = fuse_send_write(req, file, pos, count, NULL); |
b6aeaded MS |
725 | err = req->out.h.error; |
726 | fuse_put_request(fc, req); | |
5e6f58a1 | 727 | if (!err && !nres) |
b6aeaded MS |
728 | err = -EIO; |
729 | if (!err) { | |
5e6f58a1 | 730 | pos += nres; |
854512ec | 731 | fuse_write_update_size(inode, pos); |
5e6f58a1 | 732 | if (count == PAGE_CACHE_SIZE) |
b6aeaded | 733 | SetPageUptodate(page); |
b36c31ba MS |
734 | } |
735 | fuse_invalidate_attr(inode); | |
5e6f58a1 NP |
736 | return err ? err : nres; |
737 | } | |
738 | ||
739 | static int fuse_write_end(struct file *file, struct address_space *mapping, | |
740 | loff_t pos, unsigned len, unsigned copied, | |
741 | struct page *page, void *fsdata) | |
742 | { | |
743 | struct inode *inode = mapping->host; | |
744 | int res = 0; | |
745 | ||
746 | if (copied) | |
747 | res = fuse_buffered_write(file, inode, pos, copied, page); | |
748 | ||
749 | unlock_page(page); | |
750 | page_cache_release(page); | |
751 | return res; | |
b6aeaded MS |
752 | } |
753 | ||
ea9b9907 NP |
754 | static size_t fuse_send_write_pages(struct fuse_req *req, struct file *file, |
755 | struct inode *inode, loff_t pos, | |
756 | size_t count) | |
757 | { | |
758 | size_t res; | |
759 | unsigned offset; | |
760 | unsigned i; | |
761 | ||
762 | for (i = 0; i < req->num_pages; i++) | |
763 | fuse_wait_on_page_writeback(inode, req->pages[i]->index); | |
764 | ||
2106cb18 | 765 | res = fuse_send_write(req, file, pos, count, NULL); |
ea9b9907 NP |
766 | |
767 | offset = req->page_offset; | |
768 | count = res; | |
769 | for (i = 0; i < req->num_pages; i++) { | |
770 | struct page *page = req->pages[i]; | |
771 | ||
772 | if (!req->out.h.error && !offset && count >= PAGE_CACHE_SIZE) | |
773 | SetPageUptodate(page); | |
774 | ||
775 | if (count > PAGE_CACHE_SIZE - offset) | |
776 | count -= PAGE_CACHE_SIZE - offset; | |
777 | else | |
778 | count = 0; | |
779 | offset = 0; | |
780 | ||
781 | unlock_page(page); | |
782 | page_cache_release(page); | |
783 | } | |
784 | ||
785 | return res; | |
786 | } | |
787 | ||
788 | static ssize_t fuse_fill_write_pages(struct fuse_req *req, | |
789 | struct address_space *mapping, | |
790 | struct iov_iter *ii, loff_t pos) | |
791 | { | |
792 | struct fuse_conn *fc = get_fuse_conn(mapping->host); | |
793 | unsigned offset = pos & (PAGE_CACHE_SIZE - 1); | |
794 | size_t count = 0; | |
795 | int err; | |
796 | ||
f4975c67 | 797 | req->in.argpages = 1; |
ea9b9907 NP |
798 | req->page_offset = offset; |
799 | ||
800 | do { | |
801 | size_t tmp; | |
802 | struct page *page; | |
803 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; | |
804 | size_t bytes = min_t(size_t, PAGE_CACHE_SIZE - offset, | |
805 | iov_iter_count(ii)); | |
806 | ||
807 | bytes = min_t(size_t, bytes, fc->max_write - count); | |
808 | ||
809 | again: | |
810 | err = -EFAULT; | |
811 | if (iov_iter_fault_in_readable(ii, bytes)) | |
812 | break; | |
813 | ||
814 | err = -ENOMEM; | |
54566b2c | 815 | page = grab_cache_page_write_begin(mapping, index, 0); |
ea9b9907 NP |
816 | if (!page) |
817 | break; | |
818 | ||
819 | pagefault_disable(); | |
820 | tmp = iov_iter_copy_from_user_atomic(page, ii, offset, bytes); | |
821 | pagefault_enable(); | |
822 | flush_dcache_page(page); | |
823 | ||
824 | if (!tmp) { | |
825 | unlock_page(page); | |
826 | page_cache_release(page); | |
827 | bytes = min(bytes, iov_iter_single_seg_count(ii)); | |
828 | goto again; | |
829 | } | |
830 | ||
831 | err = 0; | |
832 | req->pages[req->num_pages] = page; | |
833 | req->num_pages++; | |
834 | ||
835 | iov_iter_advance(ii, tmp); | |
836 | count += tmp; | |
837 | pos += tmp; | |
838 | offset += tmp; | |
839 | if (offset == PAGE_CACHE_SIZE) | |
840 | offset = 0; | |
841 | ||
78bb6cb9 MS |
842 | if (!fc->big_writes) |
843 | break; | |
ea9b9907 NP |
844 | } while (iov_iter_count(ii) && count < fc->max_write && |
845 | req->num_pages < FUSE_MAX_PAGES_PER_REQ && offset == 0); | |
846 | ||
847 | return count > 0 ? count : err; | |
848 | } | |
849 | ||
850 | static ssize_t fuse_perform_write(struct file *file, | |
851 | struct address_space *mapping, | |
852 | struct iov_iter *ii, loff_t pos) | |
853 | { | |
854 | struct inode *inode = mapping->host; | |
855 | struct fuse_conn *fc = get_fuse_conn(inode); | |
856 | int err = 0; | |
857 | ssize_t res = 0; | |
858 | ||
859 | if (is_bad_inode(inode)) | |
860 | return -EIO; | |
861 | ||
862 | do { | |
863 | struct fuse_req *req; | |
864 | ssize_t count; | |
865 | ||
866 | req = fuse_get_req(fc); | |
867 | if (IS_ERR(req)) { | |
868 | err = PTR_ERR(req); | |
869 | break; | |
870 | } | |
871 | ||
872 | count = fuse_fill_write_pages(req, mapping, ii, pos); | |
873 | if (count <= 0) { | |
874 | err = count; | |
875 | } else { | |
876 | size_t num_written; | |
877 | ||
878 | num_written = fuse_send_write_pages(req, file, inode, | |
879 | pos, count); | |
880 | err = req->out.h.error; | |
881 | if (!err) { | |
882 | res += num_written; | |
883 | pos += num_written; | |
884 | ||
885 | /* break out of the loop on short write */ | |
886 | if (num_written != count) | |
887 | err = -EIO; | |
888 | } | |
889 | } | |
890 | fuse_put_request(fc, req); | |
891 | } while (!err && iov_iter_count(ii)); | |
892 | ||
893 | if (res > 0) | |
894 | fuse_write_update_size(inode, pos); | |
895 | ||
896 | fuse_invalidate_attr(inode); | |
897 | ||
898 | return res > 0 ? res : err; | |
899 | } | |
900 | ||
901 | static ssize_t fuse_file_aio_write(struct kiocb *iocb, const struct iovec *iov, | |
902 | unsigned long nr_segs, loff_t pos) | |
903 | { | |
904 | struct file *file = iocb->ki_filp; | |
905 | struct address_space *mapping = file->f_mapping; | |
906 | size_t count = 0; | |
907 | ssize_t written = 0; | |
908 | struct inode *inode = mapping->host; | |
909 | ssize_t err; | |
910 | struct iov_iter i; | |
911 | ||
912 | WARN_ON(iocb->ki_pos != pos); | |
913 | ||
914 | err = generic_segment_checks(iov, &nr_segs, &count, VERIFY_READ); | |
915 | if (err) | |
916 | return err; | |
917 | ||
918 | mutex_lock(&inode->i_mutex); | |
919 | vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); | |
920 | ||
921 | /* We can write back this queue in page reclaim */ | |
922 | current->backing_dev_info = mapping->backing_dev_info; | |
923 | ||
924 | err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); | |
925 | if (err) | |
926 | goto out; | |
927 | ||
928 | if (count == 0) | |
929 | goto out; | |
930 | ||
2f1936b8 | 931 | err = file_remove_suid(file); |
ea9b9907 NP |
932 | if (err) |
933 | goto out; | |
934 | ||
935 | file_update_time(file); | |
936 | ||
937 | iov_iter_init(&i, iov, nr_segs, count, 0); | |
938 | written = fuse_perform_write(file, mapping, &i, pos); | |
939 | if (written >= 0) | |
940 | iocb->ki_pos = pos + written; | |
941 | ||
942 | out: | |
943 | current->backing_dev_info = NULL; | |
944 | mutex_unlock(&inode->i_mutex); | |
945 | ||
946 | return written ? written : err; | |
947 | } | |
948 | ||
413ef8cb MS |
949 | static void fuse_release_user_pages(struct fuse_req *req, int write) |
950 | { | |
951 | unsigned i; | |
952 | ||
953 | for (i = 0; i < req->num_pages; i++) { | |
954 | struct page *page = req->pages[i]; | |
955 | if (write) | |
956 | set_page_dirty_lock(page); | |
957 | put_page(page); | |
958 | } | |
959 | } | |
960 | ||
961 | static int fuse_get_user_pages(struct fuse_req *req, const char __user *buf, | |
ce60a2f1 | 962 | size_t *nbytesp, int write) |
413ef8cb | 963 | { |
ce60a2f1 | 964 | size_t nbytes = *nbytesp; |
413ef8cb MS |
965 | unsigned long user_addr = (unsigned long) buf; |
966 | unsigned offset = user_addr & ~PAGE_MASK; | |
967 | int npages; | |
968 | ||
f4975c67 MS |
969 | /* Special case for kernel I/O: can copy directly into the buffer */ |
970 | if (segment_eq(get_fs(), KERNEL_DS)) { | |
971 | if (write) | |
972 | req->in.args[1].value = (void *) user_addr; | |
973 | else | |
974 | req->out.args[0].value = (void *) user_addr; | |
975 | ||
976 | return 0; | |
977 | } | |
413ef8cb | 978 | |
ce60a2f1 | 979 | nbytes = min_t(size_t, nbytes, FUSE_MAX_PAGES_PER_REQ << PAGE_SHIFT); |
413ef8cb | 980 | npages = (nbytes + offset + PAGE_SIZE - 1) >> PAGE_SHIFT; |
bd730967 | 981 | npages = clamp(npages, 1, FUSE_MAX_PAGES_PER_REQ); |
413ef8cb | 982 | down_read(¤t->mm->mmap_sem); |
f4975c67 | 983 | npages = get_user_pages(current, current->mm, user_addr, npages, !write, |
413ef8cb MS |
984 | 0, req->pages, NULL); |
985 | up_read(¤t->mm->mmap_sem); | |
986 | if (npages < 0) | |
987 | return npages; | |
988 | ||
989 | req->num_pages = npages; | |
990 | req->page_offset = offset; | |
f4975c67 MS |
991 | |
992 | if (write) | |
993 | req->in.argpages = 1; | |
994 | else | |
995 | req->out.argpages = 1; | |
996 | ||
997 | nbytes = (req->num_pages << PAGE_SHIFT) - req->page_offset; | |
998 | *nbytesp = min(*nbytesp, nbytes); | |
999 | ||
413ef8cb MS |
1000 | return 0; |
1001 | } | |
1002 | ||
1003 | static ssize_t fuse_direct_io(struct file *file, const char __user *buf, | |
1004 | size_t count, loff_t *ppos, int write) | |
1005 | { | |
2106cb18 MS |
1006 | struct fuse_file *ff = file->private_data; |
1007 | struct fuse_conn *fc = ff->fc; | |
413ef8cb MS |
1008 | size_t nmax = write ? fc->max_write : fc->max_read; |
1009 | loff_t pos = *ppos; | |
1010 | ssize_t res = 0; | |
248d86e8 MS |
1011 | struct fuse_req *req; |
1012 | ||
ce1d5a49 MS |
1013 | req = fuse_get_req(fc); |
1014 | if (IS_ERR(req)) | |
1015 | return PTR_ERR(req); | |
413ef8cb MS |
1016 | |
1017 | while (count) { | |
413ef8cb | 1018 | size_t nres; |
2106cb18 | 1019 | fl_owner_t owner = current->files; |
f4975c67 MS |
1020 | size_t nbytes = min(count, nmax); |
1021 | int err = fuse_get_user_pages(req, buf, &nbytes, write); | |
413ef8cb MS |
1022 | if (err) { |
1023 | res = err; | |
1024 | break; | |
1025 | } | |
f4975c67 | 1026 | |
413ef8cb | 1027 | if (write) |
2106cb18 | 1028 | nres = fuse_send_write(req, file, pos, nbytes, owner); |
413ef8cb | 1029 | else |
2106cb18 MS |
1030 | nres = fuse_send_read(req, file, pos, nbytes, owner); |
1031 | ||
413ef8cb MS |
1032 | fuse_release_user_pages(req, !write); |
1033 | if (req->out.h.error) { | |
1034 | if (!res) | |
1035 | res = req->out.h.error; | |
1036 | break; | |
1037 | } else if (nres > nbytes) { | |
1038 | res = -EIO; | |
1039 | break; | |
1040 | } | |
1041 | count -= nres; | |
1042 | res += nres; | |
1043 | pos += nres; | |
1044 | buf += nres; | |
1045 | if (nres != nbytes) | |
1046 | break; | |
56cf34ff MS |
1047 | if (count) { |
1048 | fuse_put_request(fc, req); | |
1049 | req = fuse_get_req(fc); | |
1050 | if (IS_ERR(req)) | |
1051 | break; | |
1052 | } | |
413ef8cb MS |
1053 | } |
1054 | fuse_put_request(fc, req); | |
d09cb9d7 | 1055 | if (res > 0) |
413ef8cb | 1056 | *ppos = pos; |
413ef8cb MS |
1057 | |
1058 | return res; | |
1059 | } | |
1060 | ||
1061 | static ssize_t fuse_direct_read(struct file *file, char __user *buf, | |
1062 | size_t count, loff_t *ppos) | |
1063 | { | |
d09cb9d7 MS |
1064 | ssize_t res; |
1065 | struct inode *inode = file->f_path.dentry->d_inode; | |
1066 | ||
1067 | if (is_bad_inode(inode)) | |
1068 | return -EIO; | |
1069 | ||
1070 | res = fuse_direct_io(file, buf, count, ppos, 0); | |
1071 | ||
1072 | fuse_invalidate_attr(inode); | |
1073 | ||
1074 | return res; | |
413ef8cb MS |
1075 | } |
1076 | ||
1077 | static ssize_t fuse_direct_write(struct file *file, const char __user *buf, | |
1078 | size_t count, loff_t *ppos) | |
1079 | { | |
7706a9d6 | 1080 | struct inode *inode = file->f_path.dentry->d_inode; |
413ef8cb | 1081 | ssize_t res; |
d09cb9d7 MS |
1082 | |
1083 | if (is_bad_inode(inode)) | |
1084 | return -EIO; | |
1085 | ||
413ef8cb | 1086 | /* Don't allow parallel writes to the same file */ |
1b1dcc1b | 1087 | mutex_lock(&inode->i_mutex); |
889f7848 | 1088 | res = generic_write_checks(file, ppos, &count, 0); |
d09cb9d7 | 1089 | if (!res) { |
889f7848 | 1090 | res = fuse_direct_io(file, buf, count, ppos, 1); |
d09cb9d7 MS |
1091 | if (res > 0) |
1092 | fuse_write_update_size(inode, *ppos); | |
1093 | } | |
1b1dcc1b | 1094 | mutex_unlock(&inode->i_mutex); |
d09cb9d7 MS |
1095 | |
1096 | fuse_invalidate_attr(inode); | |
1097 | ||
413ef8cb MS |
1098 | return res; |
1099 | } | |
1100 | ||
3be5a52b | 1101 | static void fuse_writepage_free(struct fuse_conn *fc, struct fuse_req *req) |
b6aeaded | 1102 | { |
3be5a52b MS |
1103 | __free_page(req->pages[0]); |
1104 | fuse_file_put(req->ff); | |
3be5a52b MS |
1105 | } |
1106 | ||
1107 | static void fuse_writepage_finish(struct fuse_conn *fc, struct fuse_req *req) | |
1108 | { | |
1109 | struct inode *inode = req->inode; | |
1110 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1111 | struct backing_dev_info *bdi = inode->i_mapping->backing_dev_info; | |
1112 | ||
1113 | list_del(&req->writepages_entry); | |
1114 | dec_bdi_stat(bdi, BDI_WRITEBACK); | |
1115 | dec_zone_page_state(req->pages[0], NR_WRITEBACK_TEMP); | |
1116 | bdi_writeout_inc(bdi); | |
1117 | wake_up(&fi->page_waitq); | |
1118 | } | |
1119 | ||
1120 | /* Called under fc->lock, may release and reacquire it */ | |
1121 | static void fuse_send_writepage(struct fuse_conn *fc, struct fuse_req *req) | |
5d9ec854 HH |
1122 | __releases(&fc->lock) |
1123 | __acquires(&fc->lock) | |
3be5a52b MS |
1124 | { |
1125 | struct fuse_inode *fi = get_fuse_inode(req->inode); | |
1126 | loff_t size = i_size_read(req->inode); | |
1127 | struct fuse_write_in *inarg = &req->misc.write.in; | |
1128 | ||
1129 | if (!fc->connected) | |
1130 | goto out_free; | |
1131 | ||
1132 | if (inarg->offset + PAGE_CACHE_SIZE <= size) { | |
1133 | inarg->size = PAGE_CACHE_SIZE; | |
1134 | } else if (inarg->offset < size) { | |
1135 | inarg->size = size & (PAGE_CACHE_SIZE - 1); | |
1136 | } else { | |
1137 | /* Got truncated off completely */ | |
1138 | goto out_free; | |
b6aeaded | 1139 | } |
3be5a52b MS |
1140 | |
1141 | req->in.args[1].size = inarg->size; | |
1142 | fi->writectr++; | |
b93f858a | 1143 | fuse_request_send_background_locked(fc, req); |
3be5a52b MS |
1144 | return; |
1145 | ||
1146 | out_free: | |
1147 | fuse_writepage_finish(fc, req); | |
1148 | spin_unlock(&fc->lock); | |
1149 | fuse_writepage_free(fc, req); | |
e9bb09dd | 1150 | fuse_put_request(fc, req); |
3be5a52b | 1151 | spin_lock(&fc->lock); |
b6aeaded MS |
1152 | } |
1153 | ||
3be5a52b MS |
1154 | /* |
1155 | * If fi->writectr is positive (no truncate or fsync going on) send | |
1156 | * all queued writepage requests. | |
1157 | * | |
1158 | * Called with fc->lock | |
1159 | */ | |
1160 | void fuse_flush_writepages(struct inode *inode) | |
5d9ec854 HH |
1161 | __releases(&fc->lock) |
1162 | __acquires(&fc->lock) | |
b6aeaded | 1163 | { |
3be5a52b MS |
1164 | struct fuse_conn *fc = get_fuse_conn(inode); |
1165 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1166 | struct fuse_req *req; | |
1167 | ||
1168 | while (fi->writectr >= 0 && !list_empty(&fi->queued_writes)) { | |
1169 | req = list_entry(fi->queued_writes.next, struct fuse_req, list); | |
1170 | list_del_init(&req->list); | |
1171 | fuse_send_writepage(fc, req); | |
1172 | } | |
1173 | } | |
1174 | ||
1175 | static void fuse_writepage_end(struct fuse_conn *fc, struct fuse_req *req) | |
1176 | { | |
1177 | struct inode *inode = req->inode; | |
1178 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1179 | ||
1180 | mapping_set_error(inode->i_mapping, req->out.h.error); | |
1181 | spin_lock(&fc->lock); | |
1182 | fi->writectr--; | |
1183 | fuse_writepage_finish(fc, req); | |
1184 | spin_unlock(&fc->lock); | |
1185 | fuse_writepage_free(fc, req); | |
1186 | } | |
1187 | ||
1188 | static int fuse_writepage_locked(struct page *page) | |
1189 | { | |
1190 | struct address_space *mapping = page->mapping; | |
1191 | struct inode *inode = mapping->host; | |
1192 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1193 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1194 | struct fuse_req *req; | |
1195 | struct fuse_file *ff; | |
1196 | struct page *tmp_page; | |
1197 | ||
1198 | set_page_writeback(page); | |
1199 | ||
1200 | req = fuse_request_alloc_nofs(); | |
1201 | if (!req) | |
1202 | goto err; | |
1203 | ||
1204 | tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
1205 | if (!tmp_page) | |
1206 | goto err_free; | |
1207 | ||
1208 | spin_lock(&fc->lock); | |
1209 | BUG_ON(list_empty(&fi->write_files)); | |
1210 | ff = list_entry(fi->write_files.next, struct fuse_file, write_entry); | |
1211 | req->ff = fuse_file_get(ff); | |
1212 | spin_unlock(&fc->lock); | |
1213 | ||
2106cb18 | 1214 | fuse_write_fill(req, ff, page_offset(page), 0); |
3be5a52b MS |
1215 | |
1216 | copy_highpage(tmp_page, page); | |
2d698b07 | 1217 | req->misc.write.in.write_flags |= FUSE_WRITE_CACHE; |
f4975c67 | 1218 | req->in.argpages = 1; |
3be5a52b MS |
1219 | req->num_pages = 1; |
1220 | req->pages[0] = tmp_page; | |
1221 | req->page_offset = 0; | |
1222 | req->end = fuse_writepage_end; | |
1223 | req->inode = inode; | |
1224 | ||
1225 | inc_bdi_stat(mapping->backing_dev_info, BDI_WRITEBACK); | |
1226 | inc_zone_page_state(tmp_page, NR_WRITEBACK_TEMP); | |
1227 | end_page_writeback(page); | |
1228 | ||
1229 | spin_lock(&fc->lock); | |
1230 | list_add(&req->writepages_entry, &fi->writepages); | |
1231 | list_add_tail(&req->list, &fi->queued_writes); | |
1232 | fuse_flush_writepages(inode); | |
1233 | spin_unlock(&fc->lock); | |
1234 | ||
1235 | return 0; | |
1236 | ||
1237 | err_free: | |
1238 | fuse_request_free(req); | |
1239 | err: | |
1240 | end_page_writeback(page); | |
1241 | return -ENOMEM; | |
1242 | } | |
1243 | ||
1244 | static int fuse_writepage(struct page *page, struct writeback_control *wbc) | |
1245 | { | |
1246 | int err; | |
1247 | ||
1248 | err = fuse_writepage_locked(page); | |
1249 | unlock_page(page); | |
1250 | ||
1251 | return err; | |
1252 | } | |
1253 | ||
1254 | static int fuse_launder_page(struct page *page) | |
1255 | { | |
1256 | int err = 0; | |
1257 | if (clear_page_dirty_for_io(page)) { | |
1258 | struct inode *inode = page->mapping->host; | |
1259 | err = fuse_writepage_locked(page); | |
1260 | if (!err) | |
1261 | fuse_wait_on_page_writeback(inode, page->index); | |
1262 | } | |
1263 | return err; | |
1264 | } | |
1265 | ||
1266 | /* | |
1267 | * Write back dirty pages now, because there may not be any suitable | |
1268 | * open files later | |
1269 | */ | |
1270 | static void fuse_vma_close(struct vm_area_struct *vma) | |
1271 | { | |
1272 | filemap_write_and_wait(vma->vm_file->f_mapping); | |
1273 | } | |
1274 | ||
1275 | /* | |
1276 | * Wait for writeback against this page to complete before allowing it | |
1277 | * to be marked dirty again, and hence written back again, possibly | |
1278 | * before the previous writepage completed. | |
1279 | * | |
1280 | * Block here, instead of in ->writepage(), so that the userspace fs | |
1281 | * can only block processes actually operating on the filesystem. | |
1282 | * | |
1283 | * Otherwise unprivileged userspace fs would be able to block | |
1284 | * unrelated: | |
1285 | * | |
1286 | * - page migration | |
1287 | * - sync(2) | |
1288 | * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER | |
1289 | */ | |
c2ec175c | 1290 | static int fuse_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) |
3be5a52b | 1291 | { |
c2ec175c | 1292 | struct page *page = vmf->page; |
3be5a52b MS |
1293 | /* |
1294 | * Don't use page->mapping as it may become NULL from a | |
1295 | * concurrent truncate. | |
1296 | */ | |
1297 | struct inode *inode = vma->vm_file->f_mapping->host; | |
1298 | ||
1299 | fuse_wait_on_page_writeback(inode, page->index); | |
1300 | return 0; | |
1301 | } | |
1302 | ||
1303 | static struct vm_operations_struct fuse_file_vm_ops = { | |
1304 | .close = fuse_vma_close, | |
1305 | .fault = filemap_fault, | |
1306 | .page_mkwrite = fuse_page_mkwrite, | |
1307 | }; | |
1308 | ||
1309 | static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma) | |
1310 | { | |
1311 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) { | |
1312 | struct inode *inode = file->f_dentry->d_inode; | |
1313 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1314 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1315 | struct fuse_file *ff = file->private_data; | |
1316 | /* | |
1317 | * file may be written through mmap, so chain it onto the | |
1318 | * inodes's write_file list | |
1319 | */ | |
1320 | spin_lock(&fc->lock); | |
1321 | if (list_empty(&ff->write_entry)) | |
1322 | list_add(&ff->write_entry, &fi->write_files); | |
1323 | spin_unlock(&fc->lock); | |
1324 | } | |
1325 | file_accessed(file); | |
1326 | vma->vm_ops = &fuse_file_vm_ops; | |
b6aeaded MS |
1327 | return 0; |
1328 | } | |
1329 | ||
fc280c96 MS |
1330 | static int fuse_direct_mmap(struct file *file, struct vm_area_struct *vma) |
1331 | { | |
1332 | /* Can't provide the coherency needed for MAP_SHARED */ | |
1333 | if (vma->vm_flags & VM_MAYSHARE) | |
1334 | return -ENODEV; | |
1335 | ||
3121bfe7 MS |
1336 | invalidate_inode_pages2(file->f_mapping); |
1337 | ||
fc280c96 MS |
1338 | return generic_file_mmap(file, vma); |
1339 | } | |
1340 | ||
71421259 MS |
1341 | static int convert_fuse_file_lock(const struct fuse_file_lock *ffl, |
1342 | struct file_lock *fl) | |
1343 | { | |
1344 | switch (ffl->type) { | |
1345 | case F_UNLCK: | |
1346 | break; | |
1347 | ||
1348 | case F_RDLCK: | |
1349 | case F_WRLCK: | |
1350 | if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX || | |
1351 | ffl->end < ffl->start) | |
1352 | return -EIO; | |
1353 | ||
1354 | fl->fl_start = ffl->start; | |
1355 | fl->fl_end = ffl->end; | |
1356 | fl->fl_pid = ffl->pid; | |
1357 | break; | |
1358 | ||
1359 | default: | |
1360 | return -EIO; | |
1361 | } | |
1362 | fl->fl_type = ffl->type; | |
1363 | return 0; | |
1364 | } | |
1365 | ||
1366 | static void fuse_lk_fill(struct fuse_req *req, struct file *file, | |
a9ff4f87 MS |
1367 | const struct file_lock *fl, int opcode, pid_t pid, |
1368 | int flock) | |
71421259 | 1369 | { |
7706a9d6 | 1370 | struct inode *inode = file->f_path.dentry->d_inode; |
9c8ef561 | 1371 | struct fuse_conn *fc = get_fuse_conn(inode); |
71421259 MS |
1372 | struct fuse_file *ff = file->private_data; |
1373 | struct fuse_lk_in *arg = &req->misc.lk_in; | |
1374 | ||
1375 | arg->fh = ff->fh; | |
9c8ef561 | 1376 | arg->owner = fuse_lock_owner_id(fc, fl->fl_owner); |
71421259 MS |
1377 | arg->lk.start = fl->fl_start; |
1378 | arg->lk.end = fl->fl_end; | |
1379 | arg->lk.type = fl->fl_type; | |
1380 | arg->lk.pid = pid; | |
a9ff4f87 MS |
1381 | if (flock) |
1382 | arg->lk_flags |= FUSE_LK_FLOCK; | |
71421259 MS |
1383 | req->in.h.opcode = opcode; |
1384 | req->in.h.nodeid = get_node_id(inode); | |
1385 | req->in.numargs = 1; | |
1386 | req->in.args[0].size = sizeof(*arg); | |
1387 | req->in.args[0].value = arg; | |
1388 | } | |
1389 | ||
1390 | static int fuse_getlk(struct file *file, struct file_lock *fl) | |
1391 | { | |
7706a9d6 | 1392 | struct inode *inode = file->f_path.dentry->d_inode; |
71421259 MS |
1393 | struct fuse_conn *fc = get_fuse_conn(inode); |
1394 | struct fuse_req *req; | |
1395 | struct fuse_lk_out outarg; | |
1396 | int err; | |
1397 | ||
1398 | req = fuse_get_req(fc); | |
1399 | if (IS_ERR(req)) | |
1400 | return PTR_ERR(req); | |
1401 | ||
a9ff4f87 | 1402 | fuse_lk_fill(req, file, fl, FUSE_GETLK, 0, 0); |
71421259 MS |
1403 | req->out.numargs = 1; |
1404 | req->out.args[0].size = sizeof(outarg); | |
1405 | req->out.args[0].value = &outarg; | |
b93f858a | 1406 | fuse_request_send(fc, req); |
71421259 MS |
1407 | err = req->out.h.error; |
1408 | fuse_put_request(fc, req); | |
1409 | if (!err) | |
1410 | err = convert_fuse_file_lock(&outarg.lk, fl); | |
1411 | ||
1412 | return err; | |
1413 | } | |
1414 | ||
a9ff4f87 | 1415 | static int fuse_setlk(struct file *file, struct file_lock *fl, int flock) |
71421259 | 1416 | { |
7706a9d6 | 1417 | struct inode *inode = file->f_path.dentry->d_inode; |
71421259 MS |
1418 | struct fuse_conn *fc = get_fuse_conn(inode); |
1419 | struct fuse_req *req; | |
1420 | int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK; | |
1421 | pid_t pid = fl->fl_type != F_UNLCK ? current->tgid : 0; | |
1422 | int err; | |
1423 | ||
48e90761 MS |
1424 | if (fl->fl_lmops && fl->fl_lmops->fl_grant) { |
1425 | /* NLM needs asynchronous locks, which we don't support yet */ | |
1426 | return -ENOLCK; | |
1427 | } | |
1428 | ||
71421259 MS |
1429 | /* Unlock on close is handled by the flush method */ |
1430 | if (fl->fl_flags & FL_CLOSE) | |
1431 | return 0; | |
1432 | ||
1433 | req = fuse_get_req(fc); | |
1434 | if (IS_ERR(req)) | |
1435 | return PTR_ERR(req); | |
1436 | ||
a9ff4f87 | 1437 | fuse_lk_fill(req, file, fl, opcode, pid, flock); |
b93f858a | 1438 | fuse_request_send(fc, req); |
71421259 | 1439 | err = req->out.h.error; |
a4d27e75 MS |
1440 | /* locking is restartable */ |
1441 | if (err == -EINTR) | |
1442 | err = -ERESTARTSYS; | |
71421259 MS |
1443 | fuse_put_request(fc, req); |
1444 | return err; | |
1445 | } | |
1446 | ||
1447 | static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl) | |
1448 | { | |
7706a9d6 | 1449 | struct inode *inode = file->f_path.dentry->d_inode; |
71421259 MS |
1450 | struct fuse_conn *fc = get_fuse_conn(inode); |
1451 | int err; | |
1452 | ||
48e90761 MS |
1453 | if (cmd == F_CANCELLK) { |
1454 | err = 0; | |
1455 | } else if (cmd == F_GETLK) { | |
71421259 | 1456 | if (fc->no_lock) { |
9d6a8c5c | 1457 | posix_test_lock(file, fl); |
71421259 MS |
1458 | err = 0; |
1459 | } else | |
1460 | err = fuse_getlk(file, fl); | |
1461 | } else { | |
1462 | if (fc->no_lock) | |
48e90761 | 1463 | err = posix_lock_file(file, fl, NULL); |
71421259 | 1464 | else |
a9ff4f87 | 1465 | err = fuse_setlk(file, fl, 0); |
71421259 MS |
1466 | } |
1467 | return err; | |
1468 | } | |
1469 | ||
a9ff4f87 MS |
1470 | static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl) |
1471 | { | |
1472 | struct inode *inode = file->f_path.dentry->d_inode; | |
1473 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1474 | int err; | |
1475 | ||
1476 | if (fc->no_lock) { | |
1477 | err = flock_lock_file_wait(file, fl); | |
1478 | } else { | |
1479 | /* emulate flock with POSIX locks */ | |
1480 | fl->fl_owner = (fl_owner_t) file; | |
1481 | err = fuse_setlk(file, fl, 1); | |
1482 | } | |
1483 | ||
1484 | return err; | |
1485 | } | |
1486 | ||
b2d2272f MS |
1487 | static sector_t fuse_bmap(struct address_space *mapping, sector_t block) |
1488 | { | |
1489 | struct inode *inode = mapping->host; | |
1490 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1491 | struct fuse_req *req; | |
1492 | struct fuse_bmap_in inarg; | |
1493 | struct fuse_bmap_out outarg; | |
1494 | int err; | |
1495 | ||
1496 | if (!inode->i_sb->s_bdev || fc->no_bmap) | |
1497 | return 0; | |
1498 | ||
1499 | req = fuse_get_req(fc); | |
1500 | if (IS_ERR(req)) | |
1501 | return 0; | |
1502 | ||
1503 | memset(&inarg, 0, sizeof(inarg)); | |
1504 | inarg.block = block; | |
1505 | inarg.blocksize = inode->i_sb->s_blocksize; | |
1506 | req->in.h.opcode = FUSE_BMAP; | |
1507 | req->in.h.nodeid = get_node_id(inode); | |
1508 | req->in.numargs = 1; | |
1509 | req->in.args[0].size = sizeof(inarg); | |
1510 | req->in.args[0].value = &inarg; | |
1511 | req->out.numargs = 1; | |
1512 | req->out.args[0].size = sizeof(outarg); | |
1513 | req->out.args[0].value = &outarg; | |
b93f858a | 1514 | fuse_request_send(fc, req); |
b2d2272f MS |
1515 | err = req->out.h.error; |
1516 | fuse_put_request(fc, req); | |
1517 | if (err == -ENOSYS) | |
1518 | fc->no_bmap = 1; | |
1519 | ||
1520 | return err ? 0 : outarg.block; | |
1521 | } | |
1522 | ||
5559b8f4 MS |
1523 | static loff_t fuse_file_llseek(struct file *file, loff_t offset, int origin) |
1524 | { | |
1525 | loff_t retval; | |
1526 | struct inode *inode = file->f_path.dentry->d_inode; | |
1527 | ||
1528 | mutex_lock(&inode->i_mutex); | |
1529 | switch (origin) { | |
1530 | case SEEK_END: | |
769415c6 MS |
1531 | retval = fuse_update_attributes(inode, NULL, file, NULL); |
1532 | if (retval) | |
5291658d | 1533 | goto exit; |
5559b8f4 MS |
1534 | offset += i_size_read(inode); |
1535 | break; | |
1536 | case SEEK_CUR: | |
1537 | offset += file->f_pos; | |
1538 | } | |
1539 | retval = -EINVAL; | |
1540 | if (offset >= 0 && offset <= inode->i_sb->s_maxbytes) { | |
1541 | if (offset != file->f_pos) { | |
1542 | file->f_pos = offset; | |
1543 | file->f_version = 0; | |
1544 | } | |
1545 | retval = offset; | |
1546 | } | |
5291658d | 1547 | exit: |
5559b8f4 MS |
1548 | mutex_unlock(&inode->i_mutex); |
1549 | return retval; | |
1550 | } | |
1551 | ||
59efec7b TH |
1552 | static int fuse_ioctl_copy_user(struct page **pages, struct iovec *iov, |
1553 | unsigned int nr_segs, size_t bytes, bool to_user) | |
1554 | { | |
1555 | struct iov_iter ii; | |
1556 | int page_idx = 0; | |
1557 | ||
1558 | if (!bytes) | |
1559 | return 0; | |
1560 | ||
1561 | iov_iter_init(&ii, iov, nr_segs, bytes, 0); | |
1562 | ||
1563 | while (iov_iter_count(&ii)) { | |
1564 | struct page *page = pages[page_idx++]; | |
1565 | size_t todo = min_t(size_t, PAGE_SIZE, iov_iter_count(&ii)); | |
1566 | void *kaddr, *map; | |
1567 | ||
1568 | kaddr = map = kmap(page); | |
1569 | ||
1570 | while (todo) { | |
1571 | char __user *uaddr = ii.iov->iov_base + ii.iov_offset; | |
1572 | size_t iov_len = ii.iov->iov_len - ii.iov_offset; | |
1573 | size_t copy = min(todo, iov_len); | |
1574 | size_t left; | |
1575 | ||
1576 | if (!to_user) | |
1577 | left = copy_from_user(kaddr, uaddr, copy); | |
1578 | else | |
1579 | left = copy_to_user(uaddr, kaddr, copy); | |
1580 | ||
1581 | if (unlikely(left)) | |
1582 | return -EFAULT; | |
1583 | ||
1584 | iov_iter_advance(&ii, copy); | |
1585 | todo -= copy; | |
1586 | kaddr += copy; | |
1587 | } | |
1588 | ||
1589 | kunmap(map); | |
1590 | } | |
1591 | ||
1592 | return 0; | |
1593 | } | |
1594 | ||
1595 | /* | |
1596 | * For ioctls, there is no generic way to determine how much memory | |
1597 | * needs to be read and/or written. Furthermore, ioctls are allowed | |
1598 | * to dereference the passed pointer, so the parameter requires deep | |
1599 | * copying but FUSE has no idea whatsoever about what to copy in or | |
1600 | * out. | |
1601 | * | |
1602 | * This is solved by allowing FUSE server to retry ioctl with | |
1603 | * necessary in/out iovecs. Let's assume the ioctl implementation | |
1604 | * needs to read in the following structure. | |
1605 | * | |
1606 | * struct a { | |
1607 | * char *buf; | |
1608 | * size_t buflen; | |
1609 | * } | |
1610 | * | |
1611 | * On the first callout to FUSE server, inarg->in_size and | |
1612 | * inarg->out_size will be NULL; then, the server completes the ioctl | |
1613 | * with FUSE_IOCTL_RETRY set in out->flags, out->in_iovs set to 1 and | |
1614 | * the actual iov array to | |
1615 | * | |
1616 | * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) } } | |
1617 | * | |
1618 | * which tells FUSE to copy in the requested area and retry the ioctl. | |
1619 | * On the second round, the server has access to the structure and | |
1620 | * from that it can tell what to look for next, so on the invocation, | |
1621 | * it sets FUSE_IOCTL_RETRY, out->in_iovs to 2 and iov array to | |
1622 | * | |
1623 | * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) }, | |
1624 | * { .iov_base = a.buf, .iov_len = a.buflen } } | |
1625 | * | |
1626 | * FUSE will copy both struct a and the pointed buffer from the | |
1627 | * process doing the ioctl and retry ioctl with both struct a and the | |
1628 | * buffer. | |
1629 | * | |
1630 | * This time, FUSE server has everything it needs and completes ioctl | |
1631 | * without FUSE_IOCTL_RETRY which finishes the ioctl call. | |
1632 | * | |
1633 | * Copying data out works the same way. | |
1634 | * | |
1635 | * Note that if FUSE_IOCTL_UNRESTRICTED is clear, the kernel | |
1636 | * automatically initializes in and out iovs by decoding @cmd with | |
1637 | * _IOC_* macros and the server is not allowed to request RETRY. This | |
1638 | * limits ioctl data transfers to well-formed ioctls and is the forced | |
1639 | * behavior for all FUSE servers. | |
1640 | */ | |
1641 | static long fuse_file_do_ioctl(struct file *file, unsigned int cmd, | |
1642 | unsigned long arg, unsigned int flags) | |
1643 | { | |
1644 | struct inode *inode = file->f_dentry->d_inode; | |
1645 | struct fuse_file *ff = file->private_data; | |
1646 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1647 | struct fuse_ioctl_in inarg = { | |
1648 | .fh = ff->fh, | |
1649 | .cmd = cmd, | |
1650 | .arg = arg, | |
1651 | .flags = flags | |
1652 | }; | |
1653 | struct fuse_ioctl_out outarg; | |
1654 | struct fuse_req *req = NULL; | |
1655 | struct page **pages = NULL; | |
1656 | struct page *iov_page = NULL; | |
1657 | struct iovec *in_iov = NULL, *out_iov = NULL; | |
1658 | unsigned int in_iovs = 0, out_iovs = 0, num_pages = 0, max_pages; | |
1659 | size_t in_size, out_size, transferred; | |
1660 | int err; | |
1661 | ||
1662 | /* assume all the iovs returned by client always fits in a page */ | |
1663 | BUILD_BUG_ON(sizeof(struct iovec) * FUSE_IOCTL_MAX_IOV > PAGE_SIZE); | |
1664 | ||
1665 | if (!fuse_allow_task(fc, current)) | |
1666 | return -EACCES; | |
1667 | ||
1668 | err = -EIO; | |
1669 | if (is_bad_inode(inode)) | |
1670 | goto out; | |
1671 | ||
1672 | err = -ENOMEM; | |
1673 | pages = kzalloc(sizeof(pages[0]) * FUSE_MAX_PAGES_PER_REQ, GFP_KERNEL); | |
1674 | iov_page = alloc_page(GFP_KERNEL); | |
1675 | if (!pages || !iov_page) | |
1676 | goto out; | |
1677 | ||
1678 | /* | |
1679 | * If restricted, initialize IO parameters as encoded in @cmd. | |
1680 | * RETRY from server is not allowed. | |
1681 | */ | |
1682 | if (!(flags & FUSE_IOCTL_UNRESTRICTED)) { | |
1683 | struct iovec *iov = page_address(iov_page); | |
1684 | ||
c9f0523d | 1685 | iov->iov_base = (void __user *)arg; |
59efec7b TH |
1686 | iov->iov_len = _IOC_SIZE(cmd); |
1687 | ||
1688 | if (_IOC_DIR(cmd) & _IOC_WRITE) { | |
1689 | in_iov = iov; | |
1690 | in_iovs = 1; | |
1691 | } | |
1692 | ||
1693 | if (_IOC_DIR(cmd) & _IOC_READ) { | |
1694 | out_iov = iov; | |
1695 | out_iovs = 1; | |
1696 | } | |
1697 | } | |
1698 | ||
1699 | retry: | |
1700 | inarg.in_size = in_size = iov_length(in_iov, in_iovs); | |
1701 | inarg.out_size = out_size = iov_length(out_iov, out_iovs); | |
1702 | ||
1703 | /* | |
1704 | * Out data can be used either for actual out data or iovs, | |
1705 | * make sure there always is at least one page. | |
1706 | */ | |
1707 | out_size = max_t(size_t, out_size, PAGE_SIZE); | |
1708 | max_pages = DIV_ROUND_UP(max(in_size, out_size), PAGE_SIZE); | |
1709 | ||
1710 | /* make sure there are enough buffer pages and init request with them */ | |
1711 | err = -ENOMEM; | |
1712 | if (max_pages > FUSE_MAX_PAGES_PER_REQ) | |
1713 | goto out; | |
1714 | while (num_pages < max_pages) { | |
1715 | pages[num_pages] = alloc_page(GFP_KERNEL | __GFP_HIGHMEM); | |
1716 | if (!pages[num_pages]) | |
1717 | goto out; | |
1718 | num_pages++; | |
1719 | } | |
1720 | ||
1721 | req = fuse_get_req(fc); | |
1722 | if (IS_ERR(req)) { | |
1723 | err = PTR_ERR(req); | |
1724 | req = NULL; | |
1725 | goto out; | |
1726 | } | |
1727 | memcpy(req->pages, pages, sizeof(req->pages[0]) * num_pages); | |
1728 | req->num_pages = num_pages; | |
1729 | ||
1730 | /* okay, let's send it to the client */ | |
1731 | req->in.h.opcode = FUSE_IOCTL; | |
1732 | req->in.h.nodeid = get_node_id(inode); | |
1733 | req->in.numargs = 1; | |
1734 | req->in.args[0].size = sizeof(inarg); | |
1735 | req->in.args[0].value = &inarg; | |
1736 | if (in_size) { | |
1737 | req->in.numargs++; | |
1738 | req->in.args[1].size = in_size; | |
1739 | req->in.argpages = 1; | |
1740 | ||
1741 | err = fuse_ioctl_copy_user(pages, in_iov, in_iovs, in_size, | |
1742 | false); | |
1743 | if (err) | |
1744 | goto out; | |
1745 | } | |
1746 | ||
1747 | req->out.numargs = 2; | |
1748 | req->out.args[0].size = sizeof(outarg); | |
1749 | req->out.args[0].value = &outarg; | |
1750 | req->out.args[1].size = out_size; | |
1751 | req->out.argpages = 1; | |
1752 | req->out.argvar = 1; | |
1753 | ||
b93f858a | 1754 | fuse_request_send(fc, req); |
59efec7b TH |
1755 | err = req->out.h.error; |
1756 | transferred = req->out.args[1].size; | |
1757 | fuse_put_request(fc, req); | |
1758 | req = NULL; | |
1759 | if (err) | |
1760 | goto out; | |
1761 | ||
1762 | /* did it ask for retry? */ | |
1763 | if (outarg.flags & FUSE_IOCTL_RETRY) { | |
1764 | char *vaddr; | |
1765 | ||
1766 | /* no retry if in restricted mode */ | |
1767 | err = -EIO; | |
1768 | if (!(flags & FUSE_IOCTL_UNRESTRICTED)) | |
1769 | goto out; | |
1770 | ||
1771 | in_iovs = outarg.in_iovs; | |
1772 | out_iovs = outarg.out_iovs; | |
1773 | ||
1774 | /* | |
1775 | * Make sure things are in boundary, separate checks | |
1776 | * are to protect against overflow. | |
1777 | */ | |
1778 | err = -ENOMEM; | |
1779 | if (in_iovs > FUSE_IOCTL_MAX_IOV || | |
1780 | out_iovs > FUSE_IOCTL_MAX_IOV || | |
1781 | in_iovs + out_iovs > FUSE_IOCTL_MAX_IOV) | |
1782 | goto out; | |
1783 | ||
1784 | err = -EIO; | |
1785 | if ((in_iovs + out_iovs) * sizeof(struct iovec) != transferred) | |
1786 | goto out; | |
1787 | ||
1788 | /* okay, copy in iovs and retry */ | |
1789 | vaddr = kmap_atomic(pages[0], KM_USER0); | |
1790 | memcpy(page_address(iov_page), vaddr, transferred); | |
1791 | kunmap_atomic(vaddr, KM_USER0); | |
1792 | ||
1793 | in_iov = page_address(iov_page); | |
1794 | out_iov = in_iov + in_iovs; | |
1795 | ||
1796 | goto retry; | |
1797 | } | |
1798 | ||
1799 | err = -EIO; | |
1800 | if (transferred > inarg.out_size) | |
1801 | goto out; | |
1802 | ||
1803 | err = fuse_ioctl_copy_user(pages, out_iov, out_iovs, transferred, true); | |
1804 | out: | |
1805 | if (req) | |
1806 | fuse_put_request(fc, req); | |
1807 | if (iov_page) | |
1808 | __free_page(iov_page); | |
1809 | while (num_pages) | |
1810 | __free_page(pages[--num_pages]); | |
1811 | kfree(pages); | |
1812 | ||
1813 | return err ? err : outarg.result; | |
1814 | } | |
1815 | ||
1816 | static long fuse_file_ioctl(struct file *file, unsigned int cmd, | |
1817 | unsigned long arg) | |
1818 | { | |
1819 | return fuse_file_do_ioctl(file, cmd, arg, 0); | |
1820 | } | |
1821 | ||
1822 | static long fuse_file_compat_ioctl(struct file *file, unsigned int cmd, | |
1823 | unsigned long arg) | |
1824 | { | |
1825 | return fuse_file_do_ioctl(file, cmd, arg, FUSE_IOCTL_COMPAT); | |
1826 | } | |
1827 | ||
95668a69 TH |
1828 | /* |
1829 | * All files which have been polled are linked to RB tree | |
1830 | * fuse_conn->polled_files which is indexed by kh. Walk the tree and | |
1831 | * find the matching one. | |
1832 | */ | |
1833 | static struct rb_node **fuse_find_polled_node(struct fuse_conn *fc, u64 kh, | |
1834 | struct rb_node **parent_out) | |
1835 | { | |
1836 | struct rb_node **link = &fc->polled_files.rb_node; | |
1837 | struct rb_node *last = NULL; | |
1838 | ||
1839 | while (*link) { | |
1840 | struct fuse_file *ff; | |
1841 | ||
1842 | last = *link; | |
1843 | ff = rb_entry(last, struct fuse_file, polled_node); | |
1844 | ||
1845 | if (kh < ff->kh) | |
1846 | link = &last->rb_left; | |
1847 | else if (kh > ff->kh) | |
1848 | link = &last->rb_right; | |
1849 | else | |
1850 | return link; | |
1851 | } | |
1852 | ||
1853 | if (parent_out) | |
1854 | *parent_out = last; | |
1855 | return link; | |
1856 | } | |
1857 | ||
1858 | /* | |
1859 | * The file is about to be polled. Make sure it's on the polled_files | |
1860 | * RB tree. Note that files once added to the polled_files tree are | |
1861 | * not removed before the file is released. This is because a file | |
1862 | * polled once is likely to be polled again. | |
1863 | */ | |
1864 | static void fuse_register_polled_file(struct fuse_conn *fc, | |
1865 | struct fuse_file *ff) | |
1866 | { | |
1867 | spin_lock(&fc->lock); | |
1868 | if (RB_EMPTY_NODE(&ff->polled_node)) { | |
1869 | struct rb_node **link, *parent; | |
1870 | ||
1871 | link = fuse_find_polled_node(fc, ff->kh, &parent); | |
1872 | BUG_ON(*link); | |
1873 | rb_link_node(&ff->polled_node, parent, link); | |
1874 | rb_insert_color(&ff->polled_node, &fc->polled_files); | |
1875 | } | |
1876 | spin_unlock(&fc->lock); | |
1877 | } | |
1878 | ||
1879 | static unsigned fuse_file_poll(struct file *file, poll_table *wait) | |
1880 | { | |
1881 | struct inode *inode = file->f_dentry->d_inode; | |
1882 | struct fuse_file *ff = file->private_data; | |
1883 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1884 | struct fuse_poll_in inarg = { .fh = ff->fh, .kh = ff->kh }; | |
1885 | struct fuse_poll_out outarg; | |
1886 | struct fuse_req *req; | |
1887 | int err; | |
1888 | ||
1889 | if (fc->no_poll) | |
1890 | return DEFAULT_POLLMASK; | |
1891 | ||
1892 | poll_wait(file, &ff->poll_wait, wait); | |
1893 | ||
1894 | /* | |
1895 | * Ask for notification iff there's someone waiting for it. | |
1896 | * The client may ignore the flag and always notify. | |
1897 | */ | |
1898 | if (waitqueue_active(&ff->poll_wait)) { | |
1899 | inarg.flags |= FUSE_POLL_SCHEDULE_NOTIFY; | |
1900 | fuse_register_polled_file(fc, ff); | |
1901 | } | |
1902 | ||
1903 | req = fuse_get_req(fc); | |
1904 | if (IS_ERR(req)) | |
1905 | return PTR_ERR(req); | |
1906 | ||
1907 | req->in.h.opcode = FUSE_POLL; | |
1908 | req->in.h.nodeid = get_node_id(inode); | |
1909 | req->in.numargs = 1; | |
1910 | req->in.args[0].size = sizeof(inarg); | |
1911 | req->in.args[0].value = &inarg; | |
1912 | req->out.numargs = 1; | |
1913 | req->out.args[0].size = sizeof(outarg); | |
1914 | req->out.args[0].value = &outarg; | |
b93f858a | 1915 | fuse_request_send(fc, req); |
95668a69 TH |
1916 | err = req->out.h.error; |
1917 | fuse_put_request(fc, req); | |
1918 | ||
1919 | if (!err) | |
1920 | return outarg.revents; | |
1921 | if (err == -ENOSYS) { | |
1922 | fc->no_poll = 1; | |
1923 | return DEFAULT_POLLMASK; | |
1924 | } | |
1925 | return POLLERR; | |
1926 | } | |
1927 | ||
1928 | /* | |
1929 | * This is called from fuse_handle_notify() on FUSE_NOTIFY_POLL and | |
1930 | * wakes up the poll waiters. | |
1931 | */ | |
1932 | int fuse_notify_poll_wakeup(struct fuse_conn *fc, | |
1933 | struct fuse_notify_poll_wakeup_out *outarg) | |
1934 | { | |
1935 | u64 kh = outarg->kh; | |
1936 | struct rb_node **link; | |
1937 | ||
1938 | spin_lock(&fc->lock); | |
1939 | ||
1940 | link = fuse_find_polled_node(fc, kh, NULL); | |
1941 | if (*link) { | |
1942 | struct fuse_file *ff; | |
1943 | ||
1944 | ff = rb_entry(*link, struct fuse_file, polled_node); | |
1945 | wake_up_interruptible_sync(&ff->poll_wait); | |
1946 | } | |
1947 | ||
1948 | spin_unlock(&fc->lock); | |
1949 | return 0; | |
1950 | } | |
1951 | ||
4b6f5d20 | 1952 | static const struct file_operations fuse_file_operations = { |
5559b8f4 | 1953 | .llseek = fuse_file_llseek, |
543ade1f | 1954 | .read = do_sync_read, |
bcb4be80 | 1955 | .aio_read = fuse_file_aio_read, |
543ade1f | 1956 | .write = do_sync_write, |
ea9b9907 | 1957 | .aio_write = fuse_file_aio_write, |
b6aeaded MS |
1958 | .mmap = fuse_file_mmap, |
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, |
5ffc4ef4 | 1965 | .splice_read = generic_file_splice_read, |
59efec7b TH |
1966 | .unlocked_ioctl = fuse_file_ioctl, |
1967 | .compat_ioctl = fuse_file_compat_ioctl, | |
95668a69 | 1968 | .poll = fuse_file_poll, |
b6aeaded MS |
1969 | }; |
1970 | ||
4b6f5d20 | 1971 | static const struct file_operations fuse_direct_io_file_operations = { |
5559b8f4 | 1972 | .llseek = fuse_file_llseek, |
413ef8cb MS |
1973 | .read = fuse_direct_read, |
1974 | .write = fuse_direct_write, | |
fc280c96 | 1975 | .mmap = fuse_direct_mmap, |
413ef8cb MS |
1976 | .open = fuse_open, |
1977 | .flush = fuse_flush, | |
1978 | .release = fuse_release, | |
1979 | .fsync = fuse_fsync, | |
71421259 | 1980 | .lock = fuse_file_lock, |
a9ff4f87 | 1981 | .flock = fuse_file_flock, |
59efec7b TH |
1982 | .unlocked_ioctl = fuse_file_ioctl, |
1983 | .compat_ioctl = fuse_file_compat_ioctl, | |
95668a69 | 1984 | .poll = fuse_file_poll, |
fc280c96 | 1985 | /* no splice_read */ |
413ef8cb MS |
1986 | }; |
1987 | ||
f5e54d6e | 1988 | static const struct address_space_operations fuse_file_aops = { |
b6aeaded | 1989 | .readpage = fuse_readpage, |
3be5a52b MS |
1990 | .writepage = fuse_writepage, |
1991 | .launder_page = fuse_launder_page, | |
5e6f58a1 NP |
1992 | .write_begin = fuse_write_begin, |
1993 | .write_end = fuse_write_end, | |
db50b96c | 1994 | .readpages = fuse_readpages, |
3be5a52b | 1995 | .set_page_dirty = __set_page_dirty_nobuffers, |
b2d2272f | 1996 | .bmap = fuse_bmap, |
b6aeaded MS |
1997 | }; |
1998 | ||
1999 | void fuse_init_file_inode(struct inode *inode) | |
2000 | { | |
45323fb7 MS |
2001 | inode->i_fop = &fuse_file_operations; |
2002 | inode->i_data.a_ops = &fuse_file_aops; | |
b6aeaded | 2003 | } |