]>
Commit | Line | Data |
---|---|---|
b6aeaded MS |
1 | /* |
2 | FUSE: Filesystem in Userspace | |
1729a16c | 3 | Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu> |
b6aeaded MS |
4 | |
5 | This program can be distributed under the terms of the GNU GPL. | |
6 | See the file COPYING. | |
7 | */ | |
8 | ||
9 | #include "fuse_i.h" | |
10 | ||
11 | #include <linux/pagemap.h> | |
12 | #include <linux/slab.h> | |
13 | #include <linux/kernel.h> | |
e8edc6e0 | 14 | #include <linux/sched.h> |
7a36094d | 15 | #include <linux/sched/signal.h> |
08cbf542 | 16 | #include <linux/module.h> |
d9d318d3 | 17 | #include <linux/compat.h> |
478e0841 | 18 | #include <linux/swap.h> |
3634a632 | 19 | #include <linux/falloc.h> |
e2e40f2c | 20 | #include <linux/uio.h> |
b6aeaded | 21 | |
4b6f5d20 | 22 | static const struct file_operations fuse_direct_io_file_operations; |
45323fb7 | 23 | |
91fe96b4 MS |
24 | static int fuse_send_open(struct fuse_conn *fc, u64 nodeid, struct file *file, |
25 | int opcode, struct fuse_open_out *outargp) | |
b6aeaded | 26 | { |
b6aeaded | 27 | struct fuse_open_in inarg; |
7078187a | 28 | FUSE_ARGS(args); |
fd72faac MS |
29 | |
30 | memset(&inarg, 0, sizeof(inarg)); | |
6ff958ed MS |
31 | inarg.flags = file->f_flags & ~(O_CREAT | O_EXCL | O_NOCTTY); |
32 | if (!fc->atomic_o_trunc) | |
33 | inarg.flags &= ~O_TRUNC; | |
7078187a MS |
34 | args.in.h.opcode = opcode; |
35 | args.in.h.nodeid = nodeid; | |
36 | args.in.numargs = 1; | |
37 | args.in.args[0].size = sizeof(inarg); | |
38 | args.in.args[0].value = &inarg; | |
39 | args.out.numargs = 1; | |
40 | args.out.args[0].size = sizeof(*outargp); | |
41 | args.out.args[0].value = outargp; | |
fd72faac | 42 | |
7078187a | 43 | return fuse_simple_request(fc, &args); |
fd72faac MS |
44 | } |
45 | ||
acf99433 | 46 | struct fuse_file *fuse_file_alloc(struct fuse_conn *fc) |
fd72faac MS |
47 | { |
48 | struct fuse_file *ff; | |
6b2db28a | 49 | |
68227c03 | 50 | ff = kzalloc(sizeof(struct fuse_file), GFP_KERNEL); |
6b2db28a TH |
51 | if (unlikely(!ff)) |
52 | return NULL; | |
53 | ||
da5e4714 | 54 | ff->fc = fc; |
4250c066 | 55 | ff->reserved_req = fuse_request_alloc(0); |
6b2db28a TH |
56 | if (unlikely(!ff->reserved_req)) { |
57 | kfree(ff); | |
58 | return NULL; | |
fd72faac | 59 | } |
6b2db28a TH |
60 | |
61 | INIT_LIST_HEAD(&ff->write_entry); | |
5d7bc7e8 | 62 | mutex_init(&ff->readdir.lock); |
4e8c2eb5 | 63 | refcount_set(&ff->count, 1); |
6b2db28a TH |
64 | RB_CLEAR_NODE(&ff->polled_node); |
65 | init_waitqueue_head(&ff->poll_wait); | |
66 | ||
67 | spin_lock(&fc->lock); | |
68 | ff->kh = ++fc->khctr; | |
69 | spin_unlock(&fc->lock); | |
70 | ||
fd72faac MS |
71 | return ff; |
72 | } | |
73 | ||
74 | void fuse_file_free(struct fuse_file *ff) | |
75 | { | |
33649c91 | 76 | fuse_request_free(ff->reserved_req); |
5d7bc7e8 | 77 | mutex_destroy(&ff->readdir.lock); |
fd72faac MS |
78 | kfree(ff); |
79 | } | |
80 | ||
267d8444 | 81 | static struct fuse_file *fuse_file_get(struct fuse_file *ff) |
c756e0a4 | 82 | { |
4e8c2eb5 | 83 | refcount_inc(&ff->count); |
c756e0a4 MS |
84 | return ff; |
85 | } | |
86 | ||
819c4b3b MS |
87 | static void fuse_release_end(struct fuse_conn *fc, struct fuse_req *req) |
88 | { | |
baebccbe | 89 | iput(req->misc.release.inode); |
819c4b3b MS |
90 | } |
91 | ||
2e64ff15 | 92 | static void fuse_file_put(struct fuse_file *ff, bool sync, bool isdir) |
c756e0a4 | 93 | { |
4e8c2eb5 | 94 | if (refcount_dec_and_test(&ff->count)) { |
c756e0a4 | 95 | struct fuse_req *req = ff->reserved_req; |
8b0797a4 | 96 | |
2e64ff15 | 97 | if (ff->fc->no_open && !isdir) { |
7678ac50 AG |
98 | /* |
99 | * Drop the release request when client does not | |
100 | * implement 'open' | |
101 | */ | |
825d6d33 | 102 | __clear_bit(FR_BACKGROUND, &req->flags); |
baebccbe | 103 | iput(req->misc.release.inode); |
7678ac50 AG |
104 | fuse_put_request(ff->fc, req); |
105 | } else if (sync) { | |
2e38bea9 | 106 | __set_bit(FR_FORCE, &req->flags); |
825d6d33 | 107 | __clear_bit(FR_BACKGROUND, &req->flags); |
5a18ec17 | 108 | fuse_request_send(ff->fc, req); |
baebccbe | 109 | iput(req->misc.release.inode); |
5a18ec17 MS |
110 | fuse_put_request(ff->fc, req); |
111 | } else { | |
112 | req->end = fuse_release_end; | |
825d6d33 | 113 | __set_bit(FR_BACKGROUND, &req->flags); |
5a18ec17 MS |
114 | fuse_request_send_background(ff->fc, req); |
115 | } | |
c756e0a4 MS |
116 | kfree(ff); |
117 | } | |
118 | } | |
119 | ||
08cbf542 TH |
120 | int fuse_do_open(struct fuse_conn *fc, u64 nodeid, struct file *file, |
121 | bool isdir) | |
91fe96b4 | 122 | { |
91fe96b4 | 123 | struct fuse_file *ff; |
91fe96b4 MS |
124 | int opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN; |
125 | ||
126 | ff = fuse_file_alloc(fc); | |
127 | if (!ff) | |
128 | return -ENOMEM; | |
129 | ||
7678ac50 AG |
130 | ff->fh = 0; |
131 | ff->open_flags = FOPEN_KEEP_CACHE; /* Default for no-open */ | |
132 | if (!fc->no_open || isdir) { | |
133 | struct fuse_open_out outarg; | |
134 | int err; | |
135 | ||
136 | err = fuse_send_open(fc, nodeid, file, opcode, &outarg); | |
137 | if (!err) { | |
138 | ff->fh = outarg.fh; | |
139 | ff->open_flags = outarg.open_flags; | |
140 | ||
141 | } else if (err != -ENOSYS || isdir) { | |
142 | fuse_file_free(ff); | |
143 | return err; | |
144 | } else { | |
145 | fc->no_open = 1; | |
146 | } | |
91fe96b4 MS |
147 | } |
148 | ||
149 | if (isdir) | |
7678ac50 | 150 | ff->open_flags &= ~FOPEN_DIRECT_IO; |
91fe96b4 | 151 | |
91fe96b4 | 152 | ff->nodeid = nodeid; |
267d8444 | 153 | file->private_data = ff; |
91fe96b4 MS |
154 | |
155 | return 0; | |
156 | } | |
08cbf542 | 157 | EXPORT_SYMBOL_GPL(fuse_do_open); |
91fe96b4 | 158 | |
650b22b9 PE |
159 | static void fuse_link_write_file(struct file *file) |
160 | { | |
161 | struct inode *inode = file_inode(file); | |
650b22b9 PE |
162 | struct fuse_inode *fi = get_fuse_inode(inode); |
163 | struct fuse_file *ff = file->private_data; | |
164 | /* | |
165 | * file may be written through mmap, so chain it onto the | |
166 | * inodes's write_file list | |
167 | */ | |
f15ecfef | 168 | spin_lock(&fi->lock); |
650b22b9 PE |
169 | if (list_empty(&ff->write_entry)) |
170 | list_add(&ff->write_entry, &fi->write_files); | |
f15ecfef | 171 | spin_unlock(&fi->lock); |
650b22b9 PE |
172 | } |
173 | ||
c7b7143c | 174 | void fuse_finish_open(struct inode *inode, struct file *file) |
fd72faac | 175 | { |
c7b7143c | 176 | struct fuse_file *ff = file->private_data; |
a0822c55 | 177 | struct fuse_conn *fc = get_fuse_conn(inode); |
c7b7143c MS |
178 | |
179 | if (ff->open_flags & FOPEN_DIRECT_IO) | |
fd72faac | 180 | file->f_op = &fuse_direct_io_file_operations; |
c7b7143c | 181 | if (!(ff->open_flags & FOPEN_KEEP_CACHE)) |
b1009979 | 182 | invalidate_inode_pages2(inode->i_mapping); |
c7b7143c | 183 | if (ff->open_flags & FOPEN_NONSEEKABLE) |
a7c1b990 | 184 | nonseekable_open(inode, file); |
a0822c55 KS |
185 | if (fc->atomic_o_trunc && (file->f_flags & O_TRUNC)) { |
186 | struct fuse_inode *fi = get_fuse_inode(inode); | |
187 | ||
f15ecfef | 188 | spin_lock(&fi->lock); |
4510d86f | 189 | fi->attr_version = atomic64_inc_return(&fc->attr_version); |
a0822c55 | 190 | i_size_write(inode, 0); |
f15ecfef | 191 | spin_unlock(&fi->lock); |
a0822c55 | 192 | fuse_invalidate_attr(inode); |
75caeecd MP |
193 | if (fc->writeback_cache) |
194 | file_update_time(file); | |
a0822c55 | 195 | } |
4d99ff8f PE |
196 | if ((file->f_mode & FMODE_WRITE) && fc->writeback_cache) |
197 | fuse_link_write_file(file); | |
fd72faac MS |
198 | } |
199 | ||
91fe96b4 | 200 | int fuse_open_common(struct inode *inode, struct file *file, bool isdir) |
fd72faac | 201 | { |
acf99433 | 202 | struct fuse_conn *fc = get_fuse_conn(inode); |
b6aeaded | 203 | int err; |
75caeecd MP |
204 | bool lock_inode = (file->f_flags & O_TRUNC) && |
205 | fc->atomic_o_trunc && | |
206 | fc->writeback_cache; | |
b6aeaded MS |
207 | |
208 | err = generic_file_open(inode, file); | |
209 | if (err) | |
210 | return err; | |
211 | ||
75caeecd | 212 | if (lock_inode) |
5955102c | 213 | inode_lock(inode); |
75caeecd | 214 | |
91fe96b4 | 215 | err = fuse_do_open(fc, get_node_id(inode), file, isdir); |
b6aeaded | 216 | |
75caeecd MP |
217 | if (!err) |
218 | fuse_finish_open(inode, file); | |
91fe96b4 | 219 | |
75caeecd | 220 | if (lock_inode) |
5955102c | 221 | inode_unlock(inode); |
75caeecd MP |
222 | |
223 | return err; | |
b6aeaded MS |
224 | } |
225 | ||
ebf84d0c KT |
226 | static void fuse_prepare_release(struct fuse_inode *fi, struct fuse_file *ff, |
227 | int flags, int opcode) | |
64c6d8ed | 228 | { |
8b0797a4 | 229 | struct fuse_conn *fc = ff->fc; |
33649c91 | 230 | struct fuse_req *req = ff->reserved_req; |
b57d4264 | 231 | struct fuse_release_in *inarg = &req->misc.release.in; |
b6aeaded | 232 | |
f15ecfef KT |
233 | /* Inode is NULL on error path of fuse_create_open() */ |
234 | if (likely(fi)) { | |
235 | spin_lock(&fi->lock); | |
236 | list_del(&ff->write_entry); | |
237 | spin_unlock(&fi->lock); | |
238 | } | |
8b0797a4 | 239 | spin_lock(&fc->lock); |
8b0797a4 MS |
240 | if (!RB_EMPTY_NODE(&ff->polled_node)) |
241 | rb_erase(&ff->polled_node, &fc->polled_files); | |
242 | spin_unlock(&fc->lock); | |
243 | ||
357ccf2b | 244 | wake_up_interruptible_all(&ff->poll_wait); |
8b0797a4 | 245 | |
b6aeaded | 246 | inarg->fh = ff->fh; |
fd72faac | 247 | inarg->flags = flags; |
51eb01e7 | 248 | req->in.h.opcode = opcode; |
c7b7143c | 249 | req->in.h.nodeid = ff->nodeid; |
b6aeaded MS |
250 | req->in.numargs = 1; |
251 | req->in.args[0].size = sizeof(struct fuse_release_in); | |
252 | req->in.args[0].value = inarg; | |
fd72faac MS |
253 | } |
254 | ||
2e64ff15 | 255 | void fuse_release_common(struct file *file, bool isdir) |
fd72faac | 256 | { |
ebf84d0c | 257 | struct fuse_inode *fi = get_fuse_inode(file_inode(file)); |
9a87ad3d MS |
258 | struct fuse_file *ff = file->private_data; |
259 | struct fuse_req *req = ff->reserved_req; | |
2e64ff15 | 260 | int opcode = isdir ? FUSE_RELEASEDIR : FUSE_RELEASE; |
6b2db28a | 261 | |
ebf84d0c | 262 | fuse_prepare_release(fi, ff, file->f_flags, opcode); |
6b2db28a | 263 | |
37fb3a30 MS |
264 | if (ff->flock) { |
265 | struct fuse_release_in *inarg = &req->misc.release.in; | |
266 | inarg->release_flags |= FUSE_RELEASE_FLOCK_UNLOCK; | |
267 | inarg->lock_owner = fuse_lock_owner_id(ff->fc, | |
268 | (fl_owner_t) file); | |
269 | } | |
baebccbe MS |
270 | /* Hold inode until release is finished */ |
271 | req->misc.release.inode = igrab(file_inode(file)); | |
6b2db28a | 272 | |
6b2db28a TH |
273 | /* |
274 | * Normally this will send the RELEASE request, however if | |
275 | * some asynchronous READ or WRITE requests are outstanding, | |
276 | * the sending will be delayed. | |
5a18ec17 MS |
277 | * |
278 | * Make the release synchronous if this is a fuseblk mount, | |
279 | * synchronous RELEASE is allowed (and desirable) in this case | |
280 | * because the server can be trusted not to screw up. | |
6b2db28a | 281 | */ |
2e64ff15 | 282 | fuse_file_put(ff, ff->fc->destroy_req != NULL, isdir); |
b6aeaded MS |
283 | } |
284 | ||
04730fef MS |
285 | static int fuse_open(struct inode *inode, struct file *file) |
286 | { | |
91fe96b4 | 287 | return fuse_open_common(inode, file, false); |
04730fef MS |
288 | } |
289 | ||
290 | static int fuse_release(struct inode *inode, struct file *file) | |
291 | { | |
e7cc133c PE |
292 | struct fuse_conn *fc = get_fuse_conn(inode); |
293 | ||
294 | /* see fuse_vma_close() for !writeback_cache case */ | |
295 | if (fc->writeback_cache) | |
1e18bda8 | 296 | write_inode_now(inode, 1); |
b0aa7606 | 297 | |
2e64ff15 | 298 | fuse_release_common(file, false); |
8b0797a4 MS |
299 | |
300 | /* return value is ignored by VFS */ | |
301 | return 0; | |
302 | } | |
303 | ||
ebf84d0c | 304 | void fuse_sync_release(struct fuse_inode *fi, struct fuse_file *ff, int flags) |
8b0797a4 | 305 | { |
4e8c2eb5 | 306 | WARN_ON(refcount_read(&ff->count) > 1); |
ebf84d0c | 307 | fuse_prepare_release(fi, ff, flags, FUSE_RELEASE); |
267d8444 MS |
308 | /* |
309 | * iput(NULL) is a no-op and since the refcount is 1 and everything's | |
310 | * synchronous, we are fine with not doing igrab() here" | |
311 | */ | |
2e64ff15 | 312 | fuse_file_put(ff, true, false); |
04730fef | 313 | } |
08cbf542 | 314 | EXPORT_SYMBOL_GPL(fuse_sync_release); |
04730fef | 315 | |
71421259 | 316 | /* |
9c8ef561 MS |
317 | * Scramble the ID space with XTEA, so that the value of the files_struct |
318 | * pointer is not exposed to userspace. | |
71421259 | 319 | */ |
f3332114 | 320 | u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id) |
71421259 | 321 | { |
9c8ef561 MS |
322 | u32 *k = fc->scramble_key; |
323 | u64 v = (unsigned long) id; | |
324 | u32 v0 = v; | |
325 | u32 v1 = v >> 32; | |
326 | u32 sum = 0; | |
327 | int i; | |
328 | ||
329 | for (i = 0; i < 32; i++) { | |
330 | v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]); | |
331 | sum += 0x9E3779B9; | |
332 | v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]); | |
333 | } | |
334 | ||
335 | return (u64) v0 + ((u64) v1 << 32); | |
71421259 MS |
336 | } |
337 | ||
2fe93bd4 MS |
338 | static struct fuse_req *fuse_find_writeback(struct fuse_inode *fi, |
339 | pgoff_t idx_from, pgoff_t idx_to) | |
340 | { | |
341 | struct fuse_req *req; | |
342 | ||
343 | list_for_each_entry(req, &fi->writepages, writepages_entry) { | |
344 | pgoff_t curr_index; | |
345 | ||
346 | WARN_ON(get_fuse_inode(req->inode) != fi); | |
347 | curr_index = req->misc.write.in.offset >> PAGE_SHIFT; | |
348 | if (idx_from < curr_index + req->num_pages && | |
349 | curr_index <= idx_to) { | |
350 | return req; | |
351 | } | |
352 | } | |
353 | return NULL; | |
354 | } | |
355 | ||
3be5a52b | 356 | /* |
ea8cd333 | 357 | * Check if any page in a range is under writeback |
3be5a52b MS |
358 | * |
359 | * This is currently done by walking the list of writepage requests | |
360 | * for the inode, which can be pretty inefficient. | |
361 | */ | |
ea8cd333 PE |
362 | static bool fuse_range_is_writeback(struct inode *inode, pgoff_t idx_from, |
363 | pgoff_t idx_to) | |
3be5a52b | 364 | { |
3be5a52b | 365 | struct fuse_inode *fi = get_fuse_inode(inode); |
2fe93bd4 | 366 | bool found; |
3be5a52b | 367 | |
f15ecfef | 368 | spin_lock(&fi->lock); |
2fe93bd4 | 369 | found = fuse_find_writeback(fi, idx_from, idx_to); |
f15ecfef | 370 | spin_unlock(&fi->lock); |
3be5a52b MS |
371 | |
372 | return found; | |
373 | } | |
374 | ||
ea8cd333 PE |
375 | static inline bool fuse_page_is_writeback(struct inode *inode, pgoff_t index) |
376 | { | |
377 | return fuse_range_is_writeback(inode, index, index); | |
378 | } | |
379 | ||
3be5a52b MS |
380 | /* |
381 | * Wait for page writeback to be completed. | |
382 | * | |
383 | * Since fuse doesn't rely on the VM writeback tracking, this has to | |
384 | * use some other means. | |
385 | */ | |
386 | static int fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index) | |
387 | { | |
388 | struct fuse_inode *fi = get_fuse_inode(inode); | |
389 | ||
390 | wait_event(fi->page_waitq, !fuse_page_is_writeback(inode, index)); | |
391 | return 0; | |
392 | } | |
393 | ||
fe38d7df MP |
394 | /* |
395 | * Wait for all pending writepages on the inode to finish. | |
396 | * | |
397 | * This is currently done by blocking further writes with FUSE_NOWRITE | |
398 | * and waiting for all sent writes to complete. | |
399 | * | |
400 | * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage | |
401 | * could conflict with truncation. | |
402 | */ | |
403 | static void fuse_sync_writes(struct inode *inode) | |
404 | { | |
405 | fuse_set_nowrite(inode); | |
406 | fuse_release_nowrite(inode); | |
407 | } | |
408 | ||
75e1fcc0 | 409 | static int fuse_flush(struct file *file, fl_owner_t id) |
b6aeaded | 410 | { |
6131ffaa | 411 | struct inode *inode = file_inode(file); |
b6aeaded MS |
412 | struct fuse_conn *fc = get_fuse_conn(inode); |
413 | struct fuse_file *ff = file->private_data; | |
414 | struct fuse_req *req; | |
415 | struct fuse_flush_in inarg; | |
416 | int err; | |
417 | ||
248d86e8 MS |
418 | if (is_bad_inode(inode)) |
419 | return -EIO; | |
420 | ||
b6aeaded MS |
421 | if (fc->no_flush) |
422 | return 0; | |
423 | ||
1e18bda8 | 424 | err = write_inode_now(inode, 1); |
fe38d7df MP |
425 | if (err) |
426 | return err; | |
427 | ||
5955102c | 428 | inode_lock(inode); |
fe38d7df | 429 | fuse_sync_writes(inode); |
5955102c | 430 | inode_unlock(inode); |
fe38d7df | 431 | |
4a7f4e88 | 432 | err = filemap_check_errors(file->f_mapping); |
9ebce595 MP |
433 | if (err) |
434 | return err; | |
435 | ||
b111c8c0 | 436 | req = fuse_get_req_nofail_nopages(fc, file); |
b6aeaded MS |
437 | memset(&inarg, 0, sizeof(inarg)); |
438 | inarg.fh = ff->fh; | |
9c8ef561 | 439 | inarg.lock_owner = fuse_lock_owner_id(fc, id); |
b6aeaded MS |
440 | req->in.h.opcode = FUSE_FLUSH; |
441 | req->in.h.nodeid = get_node_id(inode); | |
b6aeaded MS |
442 | req->in.numargs = 1; |
443 | req->in.args[0].size = sizeof(inarg); | |
444 | req->in.args[0].value = &inarg; | |
825d6d33 | 445 | __set_bit(FR_FORCE, &req->flags); |
b93f858a | 446 | fuse_request_send(fc, req); |
b6aeaded MS |
447 | err = req->out.h.error; |
448 | fuse_put_request(fc, req); | |
449 | if (err == -ENOSYS) { | |
450 | fc->no_flush = 1; | |
451 | err = 0; | |
452 | } | |
453 | return err; | |
454 | } | |
455 | ||
02c24a82 | 456 | int fuse_fsync_common(struct file *file, loff_t start, loff_t end, |
a9c2d1e8 | 457 | int datasync, int opcode) |
b6aeaded | 458 | { |
7ea80859 | 459 | struct inode *inode = file->f_mapping->host; |
b6aeaded MS |
460 | struct fuse_conn *fc = get_fuse_conn(inode); |
461 | struct fuse_file *ff = file->private_data; | |
7078187a | 462 | FUSE_ARGS(args); |
b6aeaded | 463 | struct fuse_fsync_in inarg; |
a9c2d1e8 MS |
464 | |
465 | memset(&inarg, 0, sizeof(inarg)); | |
466 | inarg.fh = ff->fh; | |
467 | inarg.fsync_flags = datasync ? 1 : 0; | |
468 | args.in.h.opcode = opcode; | |
469 | args.in.h.nodeid = get_node_id(inode); | |
470 | args.in.numargs = 1; | |
471 | args.in.args[0].size = sizeof(inarg); | |
472 | args.in.args[0].value = &inarg; | |
473 | return fuse_simple_request(fc, &args); | |
474 | } | |
475 | ||
476 | static int fuse_fsync(struct file *file, loff_t start, loff_t end, | |
477 | int datasync) | |
478 | { | |
479 | struct inode *inode = file->f_mapping->host; | |
480 | struct fuse_conn *fc = get_fuse_conn(inode); | |
b6aeaded MS |
481 | int err; |
482 | ||
248d86e8 MS |
483 | if (is_bad_inode(inode)) |
484 | return -EIO; | |
485 | ||
5955102c | 486 | inode_lock(inode); |
02c24a82 | 487 | |
3be5a52b MS |
488 | /* |
489 | * Start writeback against all dirty pages of the inode, then | |
490 | * wait for all outstanding writes, before sending the FSYNC | |
491 | * request. | |
492 | */ | |
7e51fe1d | 493 | err = file_write_and_wait_range(file, start, end); |
3be5a52b | 494 | if (err) |
02c24a82 | 495 | goto out; |
3be5a52b MS |
496 | |
497 | fuse_sync_writes(inode); | |
ac7f052b AK |
498 | |
499 | /* | |
500 | * Due to implementation of fuse writeback | |
7e51fe1d | 501 | * file_write_and_wait_range() does not catch errors. |
ac7f052b AK |
502 | * We have to do this directly after fuse_sync_writes() |
503 | */ | |
7e51fe1d | 504 | err = file_check_and_advance_wb_err(file); |
ac7f052b AK |
505 | if (err) |
506 | goto out; | |
507 | ||
1e18bda8 MS |
508 | err = sync_inode_metadata(inode, 1); |
509 | if (err) | |
510 | goto out; | |
3be5a52b | 511 | |
a9c2d1e8 | 512 | if (fc->no_fsync) |
22401e7b | 513 | goto out; |
b0aa7606 | 514 | |
a9c2d1e8 | 515 | err = fuse_fsync_common(file, start, end, datasync, FUSE_FSYNC); |
b6aeaded | 516 | if (err == -ENOSYS) { |
a9c2d1e8 | 517 | fc->no_fsync = 1; |
b6aeaded MS |
518 | err = 0; |
519 | } | |
02c24a82 | 520 | out: |
5955102c | 521 | inode_unlock(inode); |
b6aeaded | 522 | |
a9c2d1e8 | 523 | return err; |
82547981 MS |
524 | } |
525 | ||
2106cb18 MS |
526 | void fuse_read_fill(struct fuse_req *req, struct file *file, loff_t pos, |
527 | size_t count, int opcode) | |
b6aeaded | 528 | { |
5c5c5e51 | 529 | struct fuse_read_in *inarg = &req->misc.read.in; |
a6643094 | 530 | struct fuse_file *ff = file->private_data; |
b6aeaded | 531 | |
361b1eb5 MS |
532 | inarg->fh = ff->fh; |
533 | inarg->offset = pos; | |
534 | inarg->size = count; | |
a6643094 | 535 | inarg->flags = file->f_flags; |
361b1eb5 | 536 | req->in.h.opcode = opcode; |
2106cb18 | 537 | req->in.h.nodeid = ff->nodeid; |
b6aeaded MS |
538 | req->in.numargs = 1; |
539 | req->in.args[0].size = sizeof(struct fuse_read_in); | |
c1aa96a5 | 540 | req->in.args[0].value = inarg; |
b6aeaded MS |
541 | req->out.argvar = 1; |
542 | req->out.numargs = 1; | |
543 | req->out.args[0].size = count; | |
b6aeaded MS |
544 | } |
545 | ||
8fba54ae | 546 | static void fuse_release_user_pages(struct fuse_req *req, bool should_dirty) |
187c5c36 MP |
547 | { |
548 | unsigned i; | |
549 | ||
550 | for (i = 0; i < req->num_pages; i++) { | |
551 | struct page *page = req->pages[i]; | |
8fba54ae | 552 | if (should_dirty) |
187c5c36 MP |
553 | set_page_dirty_lock(page); |
554 | put_page(page); | |
555 | } | |
556 | } | |
557 | ||
744742d6 SF |
558 | static void fuse_io_release(struct kref *kref) |
559 | { | |
560 | kfree(container_of(kref, struct fuse_io_priv, refcnt)); | |
561 | } | |
562 | ||
9d5722b7 CH |
563 | static ssize_t fuse_get_res_by_io(struct fuse_io_priv *io) |
564 | { | |
565 | if (io->err) | |
566 | return io->err; | |
567 | ||
568 | if (io->bytes >= 0 && io->write) | |
569 | return -EIO; | |
570 | ||
571 | return io->bytes < 0 ? io->size : io->bytes; | |
572 | } | |
573 | ||
01e9d11a MP |
574 | /** |
575 | * In case of short read, the caller sets 'pos' to the position of | |
576 | * actual end of fuse request in IO request. Otherwise, if bytes_requested | |
577 | * == bytes_transferred or rw == WRITE, the caller sets 'pos' to -1. | |
578 | * | |
579 | * An example: | |
580 | * User requested DIO read of 64K. It was splitted into two 32K fuse requests, | |
581 | * both submitted asynchronously. The first of them was ACKed by userspace as | |
582 | * fully completed (req->out.args[0].size == 32K) resulting in pos == -1. The | |
583 | * second request was ACKed as short, e.g. only 1K was read, resulting in | |
584 | * pos == 33K. | |
585 | * | |
586 | * Thus, when all fuse requests are completed, the minimal non-negative 'pos' | |
587 | * will be equal to the length of the longest contiguous fragment of | |
588 | * transferred data starting from the beginning of IO request. | |
589 | */ | |
590 | static void fuse_aio_complete(struct fuse_io_priv *io, int err, ssize_t pos) | |
591 | { | |
592 | int left; | |
593 | ||
594 | spin_lock(&io->lock); | |
595 | if (err) | |
596 | io->err = io->err ? : err; | |
597 | else if (pos >= 0 && (io->bytes < 0 || pos < io->bytes)) | |
598 | io->bytes = pos; | |
599 | ||
600 | left = --io->reqs; | |
7879c4e5 | 601 | if (!left && io->blocking) |
9d5722b7 | 602 | complete(io->done); |
01e9d11a MP |
603 | spin_unlock(&io->lock); |
604 | ||
7879c4e5 | 605 | if (!left && !io->blocking) { |
9d5722b7 | 606 | ssize_t res = fuse_get_res_by_io(io); |
01e9d11a | 607 | |
9d5722b7 CH |
608 | if (res >= 0) { |
609 | struct inode *inode = file_inode(io->iocb->ki_filp); | |
610 | struct fuse_conn *fc = get_fuse_conn(inode); | |
611 | struct fuse_inode *fi = get_fuse_inode(inode); | |
01e9d11a | 612 | |
f15ecfef | 613 | spin_lock(&fi->lock); |
4510d86f | 614 | fi->attr_version = atomic64_inc_return(&fc->attr_version); |
f15ecfef | 615 | spin_unlock(&fi->lock); |
01e9d11a MP |
616 | } |
617 | ||
04b2fa9f | 618 | io->iocb->ki_complete(io->iocb, res, 0); |
01e9d11a | 619 | } |
744742d6 SF |
620 | |
621 | kref_put(&io->refcnt, fuse_io_release); | |
01e9d11a MP |
622 | } |
623 | ||
624 | static void fuse_aio_complete_req(struct fuse_conn *fc, struct fuse_req *req) | |
625 | { | |
626 | struct fuse_io_priv *io = req->io; | |
627 | ssize_t pos = -1; | |
628 | ||
61c12b49 | 629 | fuse_release_user_pages(req, io->should_dirty); |
01e9d11a MP |
630 | |
631 | if (io->write) { | |
632 | if (req->misc.write.in.size != req->misc.write.out.size) | |
633 | pos = req->misc.write.in.offset - io->offset + | |
634 | req->misc.write.out.size; | |
635 | } else { | |
636 | if (req->misc.read.in.size != req->out.args[0].size) | |
637 | pos = req->misc.read.in.offset - io->offset + | |
638 | req->out.args[0].size; | |
639 | } | |
640 | ||
641 | fuse_aio_complete(io, req->out.h.error, pos); | |
642 | } | |
643 | ||
644 | static size_t fuse_async_req_send(struct fuse_conn *fc, struct fuse_req *req, | |
645 | size_t num_bytes, struct fuse_io_priv *io) | |
646 | { | |
647 | spin_lock(&io->lock); | |
744742d6 | 648 | kref_get(&io->refcnt); |
01e9d11a MP |
649 | io->size += num_bytes; |
650 | io->reqs++; | |
651 | spin_unlock(&io->lock); | |
652 | ||
653 | req->io = io; | |
654 | req->end = fuse_aio_complete_req; | |
655 | ||
36cf66ed | 656 | __fuse_get_request(req); |
01e9d11a MP |
657 | fuse_request_send_background(fc, req); |
658 | ||
659 | return num_bytes; | |
660 | } | |
661 | ||
36cf66ed | 662 | static size_t fuse_send_read(struct fuse_req *req, struct fuse_io_priv *io, |
2106cb18 | 663 | loff_t pos, size_t count, fl_owner_t owner) |
04730fef | 664 | { |
e1c0eecb | 665 | struct file *file = io->iocb->ki_filp; |
2106cb18 MS |
666 | struct fuse_file *ff = file->private_data; |
667 | struct fuse_conn *fc = ff->fc; | |
f3332114 | 668 | |
2106cb18 | 669 | fuse_read_fill(req, file, pos, count, FUSE_READ); |
f3332114 | 670 | if (owner != NULL) { |
5c5c5e51 | 671 | struct fuse_read_in *inarg = &req->misc.read.in; |
f3332114 MS |
672 | |
673 | inarg->read_flags |= FUSE_READ_LOCKOWNER; | |
674 | inarg->lock_owner = fuse_lock_owner_id(fc, owner); | |
675 | } | |
36cf66ed MP |
676 | |
677 | if (io->async) | |
678 | return fuse_async_req_send(fc, req, count, io); | |
679 | ||
b93f858a | 680 | fuse_request_send(fc, req); |
361b1eb5 | 681 | return req->out.args[0].size; |
04730fef MS |
682 | } |
683 | ||
5c5c5e51 MS |
684 | static void fuse_read_update_size(struct inode *inode, loff_t size, |
685 | u64 attr_ver) | |
686 | { | |
687 | struct fuse_conn *fc = get_fuse_conn(inode); | |
688 | struct fuse_inode *fi = get_fuse_inode(inode); | |
689 | ||
f15ecfef | 690 | spin_lock(&fi->lock); |
06a7c3c2 MP |
691 | if (attr_ver == fi->attr_version && size < inode->i_size && |
692 | !test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) { | |
4510d86f | 693 | fi->attr_version = atomic64_inc_return(&fc->attr_version); |
5c5c5e51 MS |
694 | i_size_write(inode, size); |
695 | } | |
f15ecfef | 696 | spin_unlock(&fi->lock); |
5c5c5e51 MS |
697 | } |
698 | ||
a92adc82 PE |
699 | static void fuse_short_read(struct fuse_req *req, struct inode *inode, |
700 | u64 attr_ver) | |
701 | { | |
702 | size_t num_read = req->out.args[0].size; | |
8373200b PE |
703 | struct fuse_conn *fc = get_fuse_conn(inode); |
704 | ||
705 | if (fc->writeback_cache) { | |
706 | /* | |
707 | * A hole in a file. Some data after the hole are in page cache, | |
708 | * but have not reached the client fs yet. So, the hole is not | |
709 | * present there. | |
710 | */ | |
711 | int i; | |
09cbfeaf KS |
712 | int start_idx = num_read >> PAGE_SHIFT; |
713 | size_t off = num_read & (PAGE_SIZE - 1); | |
a92adc82 | 714 | |
8373200b | 715 | for (i = start_idx; i < req->num_pages; i++) { |
09cbfeaf | 716 | zero_user_segment(req->pages[i], off, PAGE_SIZE); |
8373200b PE |
717 | off = 0; |
718 | } | |
719 | } else { | |
720 | loff_t pos = page_offset(req->pages[0]) + num_read; | |
721 | fuse_read_update_size(inode, pos, attr_ver); | |
722 | } | |
a92adc82 PE |
723 | } |
724 | ||
482fce55 | 725 | static int fuse_do_readpage(struct file *file, struct page *page) |
b6aeaded | 726 | { |
e1c0eecb MS |
727 | struct kiocb iocb; |
728 | struct fuse_io_priv io; | |
b6aeaded MS |
729 | struct inode *inode = page->mapping->host; |
730 | struct fuse_conn *fc = get_fuse_conn(inode); | |
248d86e8 | 731 | struct fuse_req *req; |
5c5c5e51 MS |
732 | size_t num_read; |
733 | loff_t pos = page_offset(page); | |
09cbfeaf | 734 | size_t count = PAGE_SIZE; |
5c5c5e51 | 735 | u64 attr_ver; |
248d86e8 MS |
736 | int err; |
737 | ||
3be5a52b | 738 | /* |
25985edc | 739 | * Page writeback can extend beyond the lifetime of the |
3be5a52b MS |
740 | * page-cache page, so make sure we read a properly synced |
741 | * page. | |
742 | */ | |
743 | fuse_wait_on_page_writeback(inode, page->index); | |
744 | ||
b111c8c0 | 745 | req = fuse_get_req(fc, 1); |
ce1d5a49 | 746 | if (IS_ERR(req)) |
482fce55 | 747 | return PTR_ERR(req); |
b6aeaded | 748 | |
5c5c5e51 MS |
749 | attr_ver = fuse_get_attr_version(fc); |
750 | ||
b6aeaded | 751 | req->out.page_zeroing = 1; |
f4975c67 | 752 | req->out.argpages = 1; |
b6aeaded MS |
753 | req->num_pages = 1; |
754 | req->pages[0] = page; | |
85f40aec | 755 | req->page_descs[0].length = count; |
e1c0eecb MS |
756 | init_sync_kiocb(&iocb, file); |
757 | io = (struct fuse_io_priv) FUSE_IO_PRIV_SYNC(&iocb); | |
36cf66ed | 758 | num_read = fuse_send_read(req, &io, pos, count, NULL); |
b6aeaded | 759 | err = req->out.h.error; |
5c5c5e51 MS |
760 | |
761 | if (!err) { | |
762 | /* | |
763 | * Short read means EOF. If file size is larger, truncate it | |
764 | */ | |
765 | if (num_read < count) | |
a92adc82 | 766 | fuse_short_read(req, inode, attr_ver); |
5c5c5e51 | 767 | |
b6aeaded | 768 | SetPageUptodate(page); |
5c5c5e51 MS |
769 | } |
770 | ||
a92adc82 | 771 | fuse_put_request(fc, req); |
482fce55 MP |
772 | |
773 | return err; | |
774 | } | |
775 | ||
776 | static int fuse_readpage(struct file *file, struct page *page) | |
777 | { | |
778 | struct inode *inode = page->mapping->host; | |
779 | int err; | |
780 | ||
781 | err = -EIO; | |
782 | if (is_bad_inode(inode)) | |
783 | goto out; | |
784 | ||
785 | err = fuse_do_readpage(file, page); | |
451418fc | 786 | fuse_invalidate_atime(inode); |
b6aeaded MS |
787 | out: |
788 | unlock_page(page); | |
789 | return err; | |
790 | } | |
791 | ||
c1aa96a5 | 792 | static void fuse_readpages_end(struct fuse_conn *fc, struct fuse_req *req) |
db50b96c | 793 | { |
c1aa96a5 | 794 | int i; |
5c5c5e51 MS |
795 | size_t count = req->misc.read.in.size; |
796 | size_t num_read = req->out.args[0].size; | |
ce534fb0 | 797 | struct address_space *mapping = NULL; |
c1aa96a5 | 798 | |
ce534fb0 MS |
799 | for (i = 0; mapping == NULL && i < req->num_pages; i++) |
800 | mapping = req->pages[i]->mapping; | |
5c5c5e51 | 801 | |
ce534fb0 MS |
802 | if (mapping) { |
803 | struct inode *inode = mapping->host; | |
804 | ||
805 | /* | |
806 | * Short read means EOF. If file size is larger, truncate it | |
807 | */ | |
a92adc82 PE |
808 | if (!req->out.h.error && num_read < count) |
809 | fuse_short_read(req, inode, req->misc.read.attr_ver); | |
ce534fb0 | 810 | |
451418fc | 811 | fuse_invalidate_atime(inode); |
ce534fb0 | 812 | } |
c1aa96a5 | 813 | |
db50b96c MS |
814 | for (i = 0; i < req->num_pages; i++) { |
815 | struct page *page = req->pages[i]; | |
816 | if (!req->out.h.error) | |
817 | SetPageUptodate(page); | |
c1aa96a5 MS |
818 | else |
819 | SetPageError(page); | |
db50b96c | 820 | unlock_page(page); |
09cbfeaf | 821 | put_page(page); |
db50b96c | 822 | } |
c756e0a4 | 823 | if (req->ff) |
2e64ff15 | 824 | fuse_file_put(req->ff, false, false); |
c1aa96a5 MS |
825 | } |
826 | ||
2106cb18 | 827 | static void fuse_send_readpages(struct fuse_req *req, struct file *file) |
c1aa96a5 | 828 | { |
2106cb18 MS |
829 | struct fuse_file *ff = file->private_data; |
830 | struct fuse_conn *fc = ff->fc; | |
c1aa96a5 | 831 | loff_t pos = page_offset(req->pages[0]); |
09cbfeaf | 832 | size_t count = req->num_pages << PAGE_SHIFT; |
f4975c67 MS |
833 | |
834 | req->out.argpages = 1; | |
c1aa96a5 | 835 | req->out.page_zeroing = 1; |
ce534fb0 | 836 | req->out.page_replace = 1; |
2106cb18 | 837 | fuse_read_fill(req, file, pos, count, FUSE_READ); |
5c5c5e51 | 838 | req->misc.read.attr_ver = fuse_get_attr_version(fc); |
9cd68455 | 839 | if (fc->async_read) { |
c756e0a4 | 840 | req->ff = fuse_file_get(ff); |
9cd68455 | 841 | req->end = fuse_readpages_end; |
b93f858a | 842 | fuse_request_send_background(fc, req); |
9cd68455 | 843 | } else { |
b93f858a | 844 | fuse_request_send(fc, req); |
9cd68455 | 845 | fuse_readpages_end(fc, req); |
e9bb09dd | 846 | fuse_put_request(fc, req); |
9cd68455 | 847 | } |
db50b96c MS |
848 | } |
849 | ||
c756e0a4 | 850 | struct fuse_fill_data { |
db50b96c | 851 | struct fuse_req *req; |
a6643094 | 852 | struct file *file; |
db50b96c | 853 | struct inode *inode; |
f8dbdf81 | 854 | unsigned nr_pages; |
db50b96c MS |
855 | }; |
856 | ||
857 | static int fuse_readpages_fill(void *_data, struct page *page) | |
858 | { | |
c756e0a4 | 859 | struct fuse_fill_data *data = _data; |
db50b96c MS |
860 | struct fuse_req *req = data->req; |
861 | struct inode *inode = data->inode; | |
862 | struct fuse_conn *fc = get_fuse_conn(inode); | |
863 | ||
3be5a52b MS |
864 | fuse_wait_on_page_writeback(inode, page->index); |
865 | ||
db50b96c | 866 | if (req->num_pages && |
5da784cc | 867 | (req->num_pages == fc->max_pages || |
09cbfeaf | 868 | (req->num_pages + 1) * PAGE_SIZE > fc->max_read || |
db50b96c | 869 | req->pages[req->num_pages - 1]->index + 1 != page->index)) { |
5da784cc CS |
870 | unsigned int nr_alloc = min_t(unsigned int, data->nr_pages, |
871 | fc->max_pages); | |
2106cb18 | 872 | fuse_send_readpages(req, data->file); |
8b41e671 MP |
873 | if (fc->async_read) |
874 | req = fuse_get_req_for_background(fc, nr_alloc); | |
875 | else | |
876 | req = fuse_get_req(fc, nr_alloc); | |
877 | ||
878 | data->req = req; | |
ce1d5a49 | 879 | if (IS_ERR(req)) { |
db50b96c | 880 | unlock_page(page); |
ce1d5a49 | 881 | return PTR_ERR(req); |
db50b96c | 882 | } |
db50b96c | 883 | } |
f8dbdf81 MP |
884 | |
885 | if (WARN_ON(req->num_pages >= req->max_pages)) { | |
109728cc | 886 | unlock_page(page); |
f8dbdf81 MP |
887 | fuse_put_request(fc, req); |
888 | return -EIO; | |
889 | } | |
890 | ||
09cbfeaf | 891 | get_page(page); |
db50b96c | 892 | req->pages[req->num_pages] = page; |
85f40aec | 893 | req->page_descs[req->num_pages].length = PAGE_SIZE; |
1729a16c | 894 | req->num_pages++; |
f8dbdf81 | 895 | data->nr_pages--; |
db50b96c MS |
896 | return 0; |
897 | } | |
898 | ||
899 | static int fuse_readpages(struct file *file, struct address_space *mapping, | |
900 | struct list_head *pages, unsigned nr_pages) | |
901 | { | |
902 | struct inode *inode = mapping->host; | |
903 | struct fuse_conn *fc = get_fuse_conn(inode); | |
c756e0a4 | 904 | struct fuse_fill_data data; |
db50b96c | 905 | int err; |
5da784cc | 906 | unsigned int nr_alloc = min_t(unsigned int, nr_pages, fc->max_pages); |
248d86e8 | 907 | |
1d7ea732 | 908 | err = -EIO; |
248d86e8 | 909 | if (is_bad_inode(inode)) |
2e990021 | 910 | goto out; |
248d86e8 | 911 | |
a6643094 | 912 | data.file = file; |
db50b96c | 913 | data.inode = inode; |
8b41e671 MP |
914 | if (fc->async_read) |
915 | data.req = fuse_get_req_for_background(fc, nr_alloc); | |
916 | else | |
917 | data.req = fuse_get_req(fc, nr_alloc); | |
f8dbdf81 | 918 | data.nr_pages = nr_pages; |
1d7ea732 | 919 | err = PTR_ERR(data.req); |
ce1d5a49 | 920 | if (IS_ERR(data.req)) |
2e990021 | 921 | goto out; |
db50b96c MS |
922 | |
923 | err = read_cache_pages(mapping, pages, fuse_readpages_fill, &data); | |
d3406ffa MS |
924 | if (!err) { |
925 | if (data.req->num_pages) | |
2106cb18 | 926 | fuse_send_readpages(data.req, file); |
d3406ffa MS |
927 | else |
928 | fuse_put_request(fc, data.req); | |
929 | } | |
2e990021 | 930 | out: |
1d7ea732 | 931 | return err; |
db50b96c MS |
932 | } |
933 | ||
37c20f16 | 934 | static ssize_t fuse_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
bcb4be80 MS |
935 | { |
936 | struct inode *inode = iocb->ki_filp->f_mapping->host; | |
a8894274 | 937 | struct fuse_conn *fc = get_fuse_conn(inode); |
bcb4be80 | 938 | |
a8894274 BF |
939 | /* |
940 | * In auto invalidate mode, always update attributes on read. | |
941 | * Otherwise, only update if we attempt to read past EOF (to ensure | |
942 | * i_size is up to date). | |
943 | */ | |
944 | if (fc->auto_inval_data || | |
37c20f16 | 945 | (iocb->ki_pos + iov_iter_count(to) > i_size_read(inode))) { |
bcb4be80 | 946 | int err; |
5b97eeac | 947 | err = fuse_update_attributes(inode, iocb->ki_filp); |
bcb4be80 MS |
948 | if (err) |
949 | return err; | |
950 | } | |
951 | ||
37c20f16 | 952 | return generic_file_read_iter(iocb, to); |
bcb4be80 MS |
953 | } |
954 | ||
2d698b07 | 955 | static void fuse_write_fill(struct fuse_req *req, struct fuse_file *ff, |
2106cb18 | 956 | loff_t pos, size_t count) |
b6aeaded | 957 | { |
b25e82e5 MS |
958 | struct fuse_write_in *inarg = &req->misc.write.in; |
959 | struct fuse_write_out *outarg = &req->misc.write.out; | |
b6aeaded | 960 | |
b25e82e5 MS |
961 | inarg->fh = ff->fh; |
962 | inarg->offset = pos; | |
963 | inarg->size = count; | |
b6aeaded | 964 | req->in.h.opcode = FUSE_WRITE; |
2106cb18 | 965 | req->in.h.nodeid = ff->nodeid; |
b6aeaded | 966 | req->in.numargs = 2; |
2106cb18 | 967 | if (ff->fc->minor < 9) |
f3332114 MS |
968 | req->in.args[0].size = FUSE_COMPAT_WRITE_IN_SIZE; |
969 | else | |
970 | req->in.args[0].size = sizeof(struct fuse_write_in); | |
b25e82e5 | 971 | req->in.args[0].value = inarg; |
b6aeaded MS |
972 | req->in.args[1].size = count; |
973 | req->out.numargs = 1; | |
974 | req->out.args[0].size = sizeof(struct fuse_write_out); | |
b25e82e5 MS |
975 | req->out.args[0].value = outarg; |
976 | } | |
977 | ||
36cf66ed | 978 | static size_t fuse_send_write(struct fuse_req *req, struct fuse_io_priv *io, |
2106cb18 | 979 | loff_t pos, size_t count, fl_owner_t owner) |
b25e82e5 | 980 | { |
e1c0eecb MS |
981 | struct kiocb *iocb = io->iocb; |
982 | struct file *file = iocb->ki_filp; | |
2106cb18 MS |
983 | struct fuse_file *ff = file->private_data; |
984 | struct fuse_conn *fc = ff->fc; | |
2d698b07 MS |
985 | struct fuse_write_in *inarg = &req->misc.write.in; |
986 | ||
2106cb18 | 987 | fuse_write_fill(req, ff, pos, count); |
2d698b07 | 988 | inarg->flags = file->f_flags; |
e1c0eecb MS |
989 | if (iocb->ki_flags & IOCB_DSYNC) |
990 | inarg->flags |= O_DSYNC; | |
991 | if (iocb->ki_flags & IOCB_SYNC) | |
992 | inarg->flags |= O_SYNC; | |
f3332114 | 993 | if (owner != NULL) { |
f3332114 MS |
994 | inarg->write_flags |= FUSE_WRITE_LOCKOWNER; |
995 | inarg->lock_owner = fuse_lock_owner_id(fc, owner); | |
996 | } | |
36cf66ed MP |
997 | |
998 | if (io->async) | |
999 | return fuse_async_req_send(fc, req, count, io); | |
1000 | ||
b93f858a | 1001 | fuse_request_send(fc, req); |
b25e82e5 | 1002 | return req->misc.write.out.size; |
b6aeaded MS |
1003 | } |
1004 | ||
b0aa7606 | 1005 | bool fuse_write_update_size(struct inode *inode, loff_t pos) |
854512ec MS |
1006 | { |
1007 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1008 | struct fuse_inode *fi = get_fuse_inode(inode); | |
b0aa7606 | 1009 | bool ret = false; |
854512ec | 1010 | |
f15ecfef | 1011 | spin_lock(&fi->lock); |
4510d86f | 1012 | fi->attr_version = atomic64_inc_return(&fc->attr_version); |
b0aa7606 | 1013 | if (pos > inode->i_size) { |
854512ec | 1014 | i_size_write(inode, pos); |
b0aa7606 MP |
1015 | ret = true; |
1016 | } | |
f15ecfef | 1017 | spin_unlock(&fi->lock); |
b0aa7606 MP |
1018 | |
1019 | return ret; | |
854512ec MS |
1020 | } |
1021 | ||
e1c0eecb | 1022 | static size_t fuse_send_write_pages(struct fuse_req *req, struct kiocb *iocb, |
ea9b9907 NP |
1023 | struct inode *inode, loff_t pos, |
1024 | size_t count) | |
1025 | { | |
1026 | size_t res; | |
1027 | unsigned offset; | |
1028 | unsigned i; | |
e1c0eecb | 1029 | struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb); |
ea9b9907 NP |
1030 | |
1031 | for (i = 0; i < req->num_pages; i++) | |
1032 | fuse_wait_on_page_writeback(inode, req->pages[i]->index); | |
1033 | ||
36cf66ed | 1034 | res = fuse_send_write(req, &io, pos, count, NULL); |
ea9b9907 | 1035 | |
b2430d75 | 1036 | offset = req->page_descs[0].offset; |
ea9b9907 NP |
1037 | count = res; |
1038 | for (i = 0; i < req->num_pages; i++) { | |
1039 | struct page *page = req->pages[i]; | |
1040 | ||
09cbfeaf | 1041 | if (!req->out.h.error && !offset && count >= PAGE_SIZE) |
ea9b9907 NP |
1042 | SetPageUptodate(page); |
1043 | ||
09cbfeaf KS |
1044 | if (count > PAGE_SIZE - offset) |
1045 | count -= PAGE_SIZE - offset; | |
ea9b9907 NP |
1046 | else |
1047 | count = 0; | |
1048 | offset = 0; | |
1049 | ||
1050 | unlock_page(page); | |
09cbfeaf | 1051 | put_page(page); |
ea9b9907 NP |
1052 | } |
1053 | ||
1054 | return res; | |
1055 | } | |
1056 | ||
1057 | static ssize_t fuse_fill_write_pages(struct fuse_req *req, | |
1058 | struct address_space *mapping, | |
1059 | struct iov_iter *ii, loff_t pos) | |
1060 | { | |
1061 | struct fuse_conn *fc = get_fuse_conn(mapping->host); | |
09cbfeaf | 1062 | unsigned offset = pos & (PAGE_SIZE - 1); |
ea9b9907 NP |
1063 | size_t count = 0; |
1064 | int err; | |
1065 | ||
f4975c67 | 1066 | req->in.argpages = 1; |
b2430d75 | 1067 | req->page_descs[0].offset = offset; |
ea9b9907 NP |
1068 | |
1069 | do { | |
1070 | size_t tmp; | |
1071 | struct page *page; | |
09cbfeaf KS |
1072 | pgoff_t index = pos >> PAGE_SHIFT; |
1073 | size_t bytes = min_t(size_t, PAGE_SIZE - offset, | |
ea9b9907 NP |
1074 | iov_iter_count(ii)); |
1075 | ||
1076 | bytes = min_t(size_t, bytes, fc->max_write - count); | |
1077 | ||
1078 | again: | |
1079 | err = -EFAULT; | |
1080 | if (iov_iter_fault_in_readable(ii, bytes)) | |
1081 | break; | |
1082 | ||
1083 | err = -ENOMEM; | |
54566b2c | 1084 | page = grab_cache_page_write_begin(mapping, index, 0); |
ea9b9907 NP |
1085 | if (!page) |
1086 | break; | |
1087 | ||
931e80e4 | 1088 | if (mapping_writably_mapped(mapping)) |
1089 | flush_dcache_page(page); | |
1090 | ||
ea9b9907 | 1091 | tmp = iov_iter_copy_from_user_atomic(page, ii, offset, bytes); |
ea9b9907 NP |
1092 | flush_dcache_page(page); |
1093 | ||
3ca8138f | 1094 | iov_iter_advance(ii, tmp); |
ea9b9907 NP |
1095 | if (!tmp) { |
1096 | unlock_page(page); | |
09cbfeaf | 1097 | put_page(page); |
ea9b9907 NP |
1098 | bytes = min(bytes, iov_iter_single_seg_count(ii)); |
1099 | goto again; | |
1100 | } | |
1101 | ||
1102 | err = 0; | |
1103 | req->pages[req->num_pages] = page; | |
85f40aec | 1104 | req->page_descs[req->num_pages].length = tmp; |
ea9b9907 NP |
1105 | req->num_pages++; |
1106 | ||
ea9b9907 NP |
1107 | count += tmp; |
1108 | pos += tmp; | |
1109 | offset += tmp; | |
09cbfeaf | 1110 | if (offset == PAGE_SIZE) |
ea9b9907 NP |
1111 | offset = 0; |
1112 | ||
78bb6cb9 MS |
1113 | if (!fc->big_writes) |
1114 | break; | |
ea9b9907 | 1115 | } while (iov_iter_count(ii) && count < fc->max_write && |
d07f09f5 | 1116 | req->num_pages < req->max_pages && offset == 0); |
ea9b9907 NP |
1117 | |
1118 | return count > 0 ? count : err; | |
1119 | } | |
1120 | ||
5da784cc CS |
1121 | static inline unsigned int fuse_wr_pages(loff_t pos, size_t len, |
1122 | unsigned int max_pages) | |
d07f09f5 | 1123 | { |
5da784cc | 1124 | return min_t(unsigned int, |
09cbfeaf KS |
1125 | ((pos + len - 1) >> PAGE_SHIFT) - |
1126 | (pos >> PAGE_SHIFT) + 1, | |
5da784cc | 1127 | max_pages); |
d07f09f5 MP |
1128 | } |
1129 | ||
e1c0eecb | 1130 | static ssize_t fuse_perform_write(struct kiocb *iocb, |
ea9b9907 NP |
1131 | struct address_space *mapping, |
1132 | struct iov_iter *ii, loff_t pos) | |
1133 | { | |
1134 | struct inode *inode = mapping->host; | |
1135 | struct fuse_conn *fc = get_fuse_conn(inode); | |
06a7c3c2 | 1136 | struct fuse_inode *fi = get_fuse_inode(inode); |
ea9b9907 NP |
1137 | int err = 0; |
1138 | ssize_t res = 0; | |
1139 | ||
1140 | if (is_bad_inode(inode)) | |
1141 | return -EIO; | |
1142 | ||
06a7c3c2 MP |
1143 | if (inode->i_size < pos + iov_iter_count(ii)) |
1144 | set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); | |
1145 | ||
ea9b9907 NP |
1146 | do { |
1147 | struct fuse_req *req; | |
1148 | ssize_t count; | |
5da784cc CS |
1149 | unsigned int nr_pages = fuse_wr_pages(pos, iov_iter_count(ii), |
1150 | fc->max_pages); | |
ea9b9907 | 1151 | |
d07f09f5 | 1152 | req = fuse_get_req(fc, nr_pages); |
ea9b9907 NP |
1153 | if (IS_ERR(req)) { |
1154 | err = PTR_ERR(req); | |
1155 | break; | |
1156 | } | |
1157 | ||
1158 | count = fuse_fill_write_pages(req, mapping, ii, pos); | |
1159 | if (count <= 0) { | |
1160 | err = count; | |
1161 | } else { | |
1162 | size_t num_written; | |
1163 | ||
e1c0eecb | 1164 | num_written = fuse_send_write_pages(req, iocb, inode, |
ea9b9907 NP |
1165 | pos, count); |
1166 | err = req->out.h.error; | |
1167 | if (!err) { | |
1168 | res += num_written; | |
1169 | pos += num_written; | |
1170 | ||
1171 | /* break out of the loop on short write */ | |
1172 | if (num_written != count) | |
1173 | err = -EIO; | |
1174 | } | |
1175 | } | |
1176 | fuse_put_request(fc, req); | |
1177 | } while (!err && iov_iter_count(ii)); | |
1178 | ||
1179 | if (res > 0) | |
1180 | fuse_write_update_size(inode, pos); | |
1181 | ||
06a7c3c2 | 1182 | clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); |
ea9b9907 NP |
1183 | fuse_invalidate_attr(inode); |
1184 | ||
1185 | return res > 0 ? res : err; | |
1186 | } | |
1187 | ||
84c3d55c | 1188 | static ssize_t fuse_file_write_iter(struct kiocb *iocb, struct iov_iter *from) |
ea9b9907 NP |
1189 | { |
1190 | struct file *file = iocb->ki_filp; | |
1191 | struct address_space *mapping = file->f_mapping; | |
ea9b9907 | 1192 | ssize_t written = 0; |
4273b793 | 1193 | ssize_t written_buffered = 0; |
ea9b9907 NP |
1194 | struct inode *inode = mapping->host; |
1195 | ssize_t err; | |
4273b793 | 1196 | loff_t endbyte = 0; |
ea9b9907 | 1197 | |
4d99ff8f PE |
1198 | if (get_fuse_conn(inode)->writeback_cache) { |
1199 | /* Update size (EOF optimization) and mode (SUID clearing) */ | |
5b97eeac | 1200 | err = fuse_update_attributes(mapping->host, file); |
4d99ff8f PE |
1201 | if (err) |
1202 | return err; | |
1203 | ||
84c3d55c | 1204 | return generic_file_write_iter(iocb, from); |
4d99ff8f PE |
1205 | } |
1206 | ||
5955102c | 1207 | inode_lock(inode); |
ea9b9907 NP |
1208 | |
1209 | /* We can write back this queue in page reclaim */ | |
de1414a6 | 1210 | current->backing_dev_info = inode_to_bdi(inode); |
ea9b9907 | 1211 | |
3309dd04 AV |
1212 | err = generic_write_checks(iocb, from); |
1213 | if (err <= 0) | |
ea9b9907 NP |
1214 | goto out; |
1215 | ||
5fa8e0a1 | 1216 | err = file_remove_privs(file); |
ea9b9907 NP |
1217 | if (err) |
1218 | goto out; | |
1219 | ||
c3b2da31 JB |
1220 | err = file_update_time(file); |
1221 | if (err) | |
1222 | goto out; | |
ea9b9907 | 1223 | |
2ba48ce5 | 1224 | if (iocb->ki_flags & IOCB_DIRECT) { |
3309dd04 | 1225 | loff_t pos = iocb->ki_pos; |
1af5bb49 | 1226 | written = generic_file_direct_write(iocb, from); |
84c3d55c | 1227 | if (written < 0 || !iov_iter_count(from)) |
4273b793 AA |
1228 | goto out; |
1229 | ||
1230 | pos += written; | |
ea9b9907 | 1231 | |
e1c0eecb | 1232 | written_buffered = fuse_perform_write(iocb, mapping, from, pos); |
4273b793 AA |
1233 | if (written_buffered < 0) { |
1234 | err = written_buffered; | |
1235 | goto out; | |
1236 | } | |
1237 | endbyte = pos + written_buffered - 1; | |
1238 | ||
1239 | err = filemap_write_and_wait_range(file->f_mapping, pos, | |
1240 | endbyte); | |
1241 | if (err) | |
1242 | goto out; | |
1243 | ||
1244 | invalidate_mapping_pages(file->f_mapping, | |
09cbfeaf KS |
1245 | pos >> PAGE_SHIFT, |
1246 | endbyte >> PAGE_SHIFT); | |
4273b793 AA |
1247 | |
1248 | written += written_buffered; | |
1249 | iocb->ki_pos = pos + written_buffered; | |
1250 | } else { | |
e1c0eecb | 1251 | written = fuse_perform_write(iocb, mapping, from, iocb->ki_pos); |
4273b793 | 1252 | if (written >= 0) |
3309dd04 | 1253 | iocb->ki_pos += written; |
4273b793 | 1254 | } |
ea9b9907 NP |
1255 | out: |
1256 | current->backing_dev_info = NULL; | |
5955102c | 1257 | inode_unlock(inode); |
e1c0eecb MS |
1258 | if (written > 0) |
1259 | written = generic_write_sync(iocb, written); | |
ea9b9907 NP |
1260 | |
1261 | return written ? written : err; | |
1262 | } | |
1263 | ||
7c190c8b MP |
1264 | static inline void fuse_page_descs_length_init(struct fuse_req *req, |
1265 | unsigned index, unsigned nr_pages) | |
85f40aec MP |
1266 | { |
1267 | int i; | |
1268 | ||
7c190c8b | 1269 | for (i = index; i < index + nr_pages; i++) |
85f40aec MP |
1270 | req->page_descs[i].length = PAGE_SIZE - |
1271 | req->page_descs[i].offset; | |
1272 | } | |
1273 | ||
7c190c8b MP |
1274 | static inline unsigned long fuse_get_user_addr(const struct iov_iter *ii) |
1275 | { | |
1276 | return (unsigned long)ii->iov->iov_base + ii->iov_offset; | |
1277 | } | |
1278 | ||
1279 | static inline size_t fuse_get_frag_size(const struct iov_iter *ii, | |
1280 | size_t max_size) | |
1281 | { | |
1282 | return min(iov_iter_single_seg_count(ii), max_size); | |
1283 | } | |
1284 | ||
b98d023a | 1285 | static int fuse_get_user_pages(struct fuse_req *req, struct iov_iter *ii, |
ce60a2f1 | 1286 | size_t *nbytesp, int write) |
413ef8cb | 1287 | { |
7c190c8b | 1288 | size_t nbytes = 0; /* # bytes already packed in req */ |
742f9927 | 1289 | ssize_t ret = 0; |
b98d023a | 1290 | |
f4975c67 | 1291 | /* Special case for kernel I/O: can copy directly into the buffer */ |
00e23707 | 1292 | if (iov_iter_is_kvec(ii)) { |
7c190c8b MP |
1293 | unsigned long user_addr = fuse_get_user_addr(ii); |
1294 | size_t frag_size = fuse_get_frag_size(ii, *nbytesp); | |
1295 | ||
f4975c67 MS |
1296 | if (write) |
1297 | req->in.args[1].value = (void *) user_addr; | |
1298 | else | |
1299 | req->out.args[0].value = (void *) user_addr; | |
1300 | ||
b98d023a MP |
1301 | iov_iter_advance(ii, frag_size); |
1302 | *nbytesp = frag_size; | |
f4975c67 MS |
1303 | return 0; |
1304 | } | |
413ef8cb | 1305 | |
5565a9d8 | 1306 | while (nbytes < *nbytesp && req->num_pages < req->max_pages) { |
7c190c8b | 1307 | unsigned npages; |
f67da30c | 1308 | size_t start; |
742f9927 | 1309 | ret = iov_iter_get_pages(ii, &req->pages[req->num_pages], |
2c80929c | 1310 | *nbytesp - nbytes, |
c7f3888a AV |
1311 | req->max_pages - req->num_pages, |
1312 | &start); | |
7c190c8b | 1313 | if (ret < 0) |
742f9927 | 1314 | break; |
7c190c8b | 1315 | |
c9c37e2e AV |
1316 | iov_iter_advance(ii, ret); |
1317 | nbytes += ret; | |
7c190c8b | 1318 | |
c9c37e2e AV |
1319 | ret += start; |
1320 | npages = (ret + PAGE_SIZE - 1) / PAGE_SIZE; | |
7c190c8b | 1321 | |
c9c37e2e | 1322 | req->page_descs[req->num_pages].offset = start; |
7c190c8b MP |
1323 | fuse_page_descs_length_init(req, req->num_pages, npages); |
1324 | ||
1325 | req->num_pages += npages; | |
1326 | req->page_descs[req->num_pages - 1].length -= | |
c9c37e2e | 1327 | (PAGE_SIZE - ret) & (PAGE_SIZE - 1); |
7c190c8b | 1328 | } |
f4975c67 MS |
1329 | |
1330 | if (write) | |
1331 | req->in.argpages = 1; | |
1332 | else | |
1333 | req->out.argpages = 1; | |
1334 | ||
7c190c8b | 1335 | *nbytesp = nbytes; |
f4975c67 | 1336 | |
2c932d4c | 1337 | return ret < 0 ? ret : 0; |
413ef8cb MS |
1338 | } |
1339 | ||
d22a943f AV |
1340 | ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter, |
1341 | loff_t *ppos, int flags) | |
413ef8cb | 1342 | { |
ea8cd333 PE |
1343 | int write = flags & FUSE_DIO_WRITE; |
1344 | int cuse = flags & FUSE_DIO_CUSE; | |
e1c0eecb | 1345 | struct file *file = io->iocb->ki_filp; |
ea8cd333 | 1346 | struct inode *inode = file->f_mapping->host; |
2106cb18 MS |
1347 | struct fuse_file *ff = file->private_data; |
1348 | struct fuse_conn *fc = ff->fc; | |
413ef8cb MS |
1349 | size_t nmax = write ? fc->max_write : fc->max_read; |
1350 | loff_t pos = *ppos; | |
d22a943f | 1351 | size_t count = iov_iter_count(iter); |
09cbfeaf KS |
1352 | pgoff_t idx_from = pos >> PAGE_SHIFT; |
1353 | pgoff_t idx_to = (pos + count - 1) >> PAGE_SHIFT; | |
413ef8cb | 1354 | ssize_t res = 0; |
248d86e8 | 1355 | struct fuse_req *req; |
742f9927 | 1356 | int err = 0; |
248d86e8 | 1357 | |
de82b923 | 1358 | if (io->async) |
5da784cc CS |
1359 | req = fuse_get_req_for_background(fc, iov_iter_npages(iter, |
1360 | fc->max_pages)); | |
de82b923 | 1361 | else |
5da784cc | 1362 | req = fuse_get_req(fc, iov_iter_npages(iter, fc->max_pages)); |
ce1d5a49 MS |
1363 | if (IS_ERR(req)) |
1364 | return PTR_ERR(req); | |
413ef8cb | 1365 | |
ea8cd333 PE |
1366 | if (!cuse && fuse_range_is_writeback(inode, idx_from, idx_to)) { |
1367 | if (!write) | |
5955102c | 1368 | inode_lock(inode); |
ea8cd333 PE |
1369 | fuse_sync_writes(inode); |
1370 | if (!write) | |
5955102c | 1371 | inode_unlock(inode); |
ea8cd333 PE |
1372 | } |
1373 | ||
61c12b49 | 1374 | io->should_dirty = !write && iter_is_iovec(iter); |
413ef8cb | 1375 | while (count) { |
413ef8cb | 1376 | size_t nres; |
2106cb18 | 1377 | fl_owner_t owner = current->files; |
f4975c67 | 1378 | size_t nbytes = min(count, nmax); |
742f9927 AS |
1379 | err = fuse_get_user_pages(req, iter, &nbytes, write); |
1380 | if (err && !nbytes) | |
413ef8cb | 1381 | break; |
f4975c67 | 1382 | |
413ef8cb | 1383 | if (write) |
36cf66ed | 1384 | nres = fuse_send_write(req, io, pos, nbytes, owner); |
413ef8cb | 1385 | else |
36cf66ed | 1386 | nres = fuse_send_read(req, io, pos, nbytes, owner); |
2106cb18 | 1387 | |
36cf66ed | 1388 | if (!io->async) |
61c12b49 | 1389 | fuse_release_user_pages(req, io->should_dirty); |
413ef8cb | 1390 | if (req->out.h.error) { |
742f9927 | 1391 | err = req->out.h.error; |
413ef8cb MS |
1392 | break; |
1393 | } else if (nres > nbytes) { | |
742f9927 AS |
1394 | res = 0; |
1395 | err = -EIO; | |
413ef8cb MS |
1396 | break; |
1397 | } | |
1398 | count -= nres; | |
1399 | res += nres; | |
1400 | pos += nres; | |
413ef8cb MS |
1401 | if (nres != nbytes) |
1402 | break; | |
56cf34ff MS |
1403 | if (count) { |
1404 | fuse_put_request(fc, req); | |
de82b923 BF |
1405 | if (io->async) |
1406 | req = fuse_get_req_for_background(fc, | |
5da784cc | 1407 | iov_iter_npages(iter, fc->max_pages)); |
de82b923 | 1408 | else |
5da784cc CS |
1409 | req = fuse_get_req(fc, iov_iter_npages(iter, |
1410 | fc->max_pages)); | |
56cf34ff MS |
1411 | if (IS_ERR(req)) |
1412 | break; | |
1413 | } | |
413ef8cb | 1414 | } |
f60311d5 AA |
1415 | if (!IS_ERR(req)) |
1416 | fuse_put_request(fc, req); | |
d09cb9d7 | 1417 | if (res > 0) |
413ef8cb | 1418 | *ppos = pos; |
413ef8cb | 1419 | |
742f9927 | 1420 | return res > 0 ? res : err; |
413ef8cb | 1421 | } |
08cbf542 | 1422 | EXPORT_SYMBOL_GPL(fuse_direct_io); |
413ef8cb | 1423 | |
36cf66ed | 1424 | static ssize_t __fuse_direct_read(struct fuse_io_priv *io, |
d22a943f AV |
1425 | struct iov_iter *iter, |
1426 | loff_t *ppos) | |
413ef8cb | 1427 | { |
d09cb9d7 | 1428 | ssize_t res; |
e1c0eecb | 1429 | struct inode *inode = file_inode(io->iocb->ki_filp); |
d09cb9d7 MS |
1430 | |
1431 | if (is_bad_inode(inode)) | |
1432 | return -EIO; | |
1433 | ||
d22a943f | 1434 | res = fuse_direct_io(io, iter, ppos, 0); |
d09cb9d7 | 1435 | |
9a2eb24d | 1436 | fuse_invalidate_atime(inode); |
d09cb9d7 MS |
1437 | |
1438 | return res; | |
413ef8cb MS |
1439 | } |
1440 | ||
15316263 | 1441 | static ssize_t fuse_direct_read_iter(struct kiocb *iocb, struct iov_iter *to) |
b98d023a | 1442 | { |
e1c0eecb | 1443 | struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb); |
15316263 | 1444 | return __fuse_direct_read(&io, to, &iocb->ki_pos); |
b98d023a MP |
1445 | } |
1446 | ||
15316263 | 1447 | static ssize_t fuse_direct_write_iter(struct kiocb *iocb, struct iov_iter *from) |
4273b793 | 1448 | { |
e1c0eecb MS |
1449 | struct inode *inode = file_inode(iocb->ki_filp); |
1450 | struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb); | |
15316263 | 1451 | ssize_t res; |
4273b793 AA |
1452 | |
1453 | if (is_bad_inode(inode)) | |
1454 | return -EIO; | |
1455 | ||
1456 | /* Don't allow parallel writes to the same file */ | |
5955102c | 1457 | inode_lock(inode); |
3309dd04 AV |
1458 | res = generic_write_checks(iocb, from); |
1459 | if (res > 0) | |
812408fb | 1460 | res = fuse_direct_io(&io, from, &iocb->ki_pos, FUSE_DIO_WRITE); |
812408fb | 1461 | fuse_invalidate_attr(inode); |
bcba24cc | 1462 | if (res > 0) |
15316263 | 1463 | fuse_write_update_size(inode, iocb->ki_pos); |
5955102c | 1464 | inode_unlock(inode); |
4273b793 AA |
1465 | |
1466 | return res; | |
1467 | } | |
1468 | ||
3be5a52b | 1469 | static void fuse_writepage_free(struct fuse_conn *fc, struct fuse_req *req) |
b6aeaded | 1470 | { |
385b1268 PE |
1471 | int i; |
1472 | ||
1473 | for (i = 0; i < req->num_pages; i++) | |
1474 | __free_page(req->pages[i]); | |
8b284dc4 MS |
1475 | |
1476 | if (req->ff) | |
2e64ff15 | 1477 | fuse_file_put(req->ff, false, false); |
3be5a52b MS |
1478 | } |
1479 | ||
1480 | static void fuse_writepage_finish(struct fuse_conn *fc, struct fuse_req *req) | |
1481 | { | |
1482 | struct inode *inode = req->inode; | |
1483 | struct fuse_inode *fi = get_fuse_inode(inode); | |
de1414a6 | 1484 | struct backing_dev_info *bdi = inode_to_bdi(inode); |
385b1268 | 1485 | int i; |
3be5a52b MS |
1486 | |
1487 | list_del(&req->writepages_entry); | |
385b1268 | 1488 | for (i = 0; i < req->num_pages; i++) { |
93f78d88 | 1489 | dec_wb_stat(&bdi->wb, WB_WRITEBACK); |
11fb9989 | 1490 | dec_node_page_state(req->pages[i], NR_WRITEBACK_TEMP); |
93f78d88 | 1491 | wb_writeout_inc(&bdi->wb); |
385b1268 | 1492 | } |
3be5a52b MS |
1493 | wake_up(&fi->page_waitq); |
1494 | } | |
1495 | ||
f15ecfef | 1496 | /* Called under fi->lock, may release and reacquire it */ |
6eaf4782 MP |
1497 | static void fuse_send_writepage(struct fuse_conn *fc, struct fuse_req *req, |
1498 | loff_t size) | |
f15ecfef KT |
1499 | __releases(fi->lock) |
1500 | __acquires(fi->lock) | |
3be5a52b | 1501 | { |
e2653bd5 | 1502 | struct fuse_req *aux, *next; |
3be5a52b | 1503 | struct fuse_inode *fi = get_fuse_inode(req->inode); |
3be5a52b | 1504 | struct fuse_write_in *inarg = &req->misc.write.in; |
09cbfeaf | 1505 | __u64 data_size = req->num_pages * PAGE_SIZE; |
63825b4e | 1506 | bool queued; |
3be5a52b | 1507 | |
385b1268 PE |
1508 | if (inarg->offset + data_size <= size) { |
1509 | inarg->size = data_size; | |
3be5a52b | 1510 | } else if (inarg->offset < size) { |
385b1268 | 1511 | inarg->size = size - inarg->offset; |
3be5a52b MS |
1512 | } else { |
1513 | /* Got truncated off completely */ | |
1514 | goto out_free; | |
b6aeaded | 1515 | } |
3be5a52b MS |
1516 | |
1517 | req->in.args[1].size = inarg->size; | |
63825b4e | 1518 | queued = fuse_request_queue_background(fc, req); |
f15ecfef KT |
1519 | /* Fails on broken connection only */ |
1520 | if (unlikely(!queued)) | |
1521 | goto out_free; | |
1522 | ||
1523 | fi->writectr++; | |
3be5a52b MS |
1524 | return; |
1525 | ||
1526 | out_free: | |
1527 | fuse_writepage_finish(fc, req); | |
f15ecfef | 1528 | spin_unlock(&fi->lock); |
e2653bd5 MS |
1529 | |
1530 | /* After fuse_writepage_finish() aux request list is private */ | |
1531 | for (aux = req->misc.write.next; aux; aux = next) { | |
1532 | next = aux->misc.write.next; | |
1533 | aux->misc.write.next = NULL; | |
1534 | fuse_writepage_free(fc, aux); | |
1535 | fuse_put_request(fc, aux); | |
1536 | } | |
1537 | ||
3be5a52b | 1538 | fuse_writepage_free(fc, req); |
e9bb09dd | 1539 | fuse_put_request(fc, req); |
f15ecfef | 1540 | spin_lock(&fi->lock); |
b6aeaded MS |
1541 | } |
1542 | ||
3be5a52b MS |
1543 | /* |
1544 | * If fi->writectr is positive (no truncate or fsync going on) send | |
1545 | * all queued writepage requests. | |
1546 | * | |
f15ecfef | 1547 | * Called with fi->lock |
3be5a52b MS |
1548 | */ |
1549 | void fuse_flush_writepages(struct inode *inode) | |
f15ecfef KT |
1550 | __releases(fi->lock) |
1551 | __acquires(fi->lock) | |
b6aeaded | 1552 | { |
3be5a52b MS |
1553 | struct fuse_conn *fc = get_fuse_conn(inode); |
1554 | struct fuse_inode *fi = get_fuse_inode(inode); | |
6eaf4782 | 1555 | size_t crop = i_size_read(inode); |
3be5a52b MS |
1556 | struct fuse_req *req; |
1557 | ||
1558 | while (fi->writectr >= 0 && !list_empty(&fi->queued_writes)) { | |
1559 | req = list_entry(fi->queued_writes.next, struct fuse_req, list); | |
1560 | list_del_init(&req->list); | |
6eaf4782 | 1561 | fuse_send_writepage(fc, req, crop); |
3be5a52b MS |
1562 | } |
1563 | } | |
1564 | ||
1565 | static void fuse_writepage_end(struct fuse_conn *fc, struct fuse_req *req) | |
1566 | { | |
1567 | struct inode *inode = req->inode; | |
1568 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1569 | ||
1570 | mapping_set_error(inode->i_mapping, req->out.h.error); | |
f15ecfef | 1571 | spin_lock(&fi->lock); |
8b284dc4 | 1572 | while (req->misc.write.next) { |
6eaf4782 MP |
1573 | struct fuse_conn *fc = get_fuse_conn(inode); |
1574 | struct fuse_write_in *inarg = &req->misc.write.in; | |
8b284dc4 MS |
1575 | struct fuse_req *next = req->misc.write.next; |
1576 | req->misc.write.next = next->misc.write.next; | |
1577 | next->misc.write.next = NULL; | |
ce128de6 | 1578 | next->ff = fuse_file_get(req->ff); |
8b284dc4 | 1579 | list_add(&next->writepages_entry, &fi->writepages); |
6eaf4782 MP |
1580 | |
1581 | /* | |
1582 | * Skip fuse_flush_writepages() to make it easy to crop requests | |
1583 | * based on primary request size. | |
1584 | * | |
1585 | * 1st case (trivial): there are no concurrent activities using | |
1586 | * fuse_set/release_nowrite. Then we're on safe side because | |
1587 | * fuse_flush_writepages() would call fuse_send_writepage() | |
1588 | * anyway. | |
1589 | * | |
1590 | * 2nd case: someone called fuse_set_nowrite and it is waiting | |
1591 | * now for completion of all in-flight requests. This happens | |
1592 | * rarely and no more than once per page, so this should be | |
1593 | * okay. | |
1594 | * | |
1595 | * 3rd case: someone (e.g. fuse_do_setattr()) is in the middle | |
1596 | * of fuse_set_nowrite..fuse_release_nowrite section. The fact | |
1597 | * that fuse_set_nowrite returned implies that all in-flight | |
1598 | * requests were completed along with all of their secondary | |
1599 | * requests. Further primary requests are blocked by negative | |
1600 | * writectr. Hence there cannot be any in-flight requests and | |
1601 | * no invocations of fuse_writepage_end() while we're in | |
1602 | * fuse_set_nowrite..fuse_release_nowrite section. | |
1603 | */ | |
1604 | fuse_send_writepage(fc, next, inarg->offset + inarg->size); | |
8b284dc4 | 1605 | } |
3be5a52b MS |
1606 | fi->writectr--; |
1607 | fuse_writepage_finish(fc, req); | |
f15ecfef | 1608 | spin_unlock(&fi->lock); |
3be5a52b MS |
1609 | fuse_writepage_free(fc, req); |
1610 | } | |
1611 | ||
1e18bda8 MS |
1612 | static struct fuse_file *__fuse_write_file_get(struct fuse_conn *fc, |
1613 | struct fuse_inode *fi) | |
adcadfa8 | 1614 | { |
72523425 | 1615 | struct fuse_file *ff = NULL; |
adcadfa8 | 1616 | |
f15ecfef | 1617 | spin_lock(&fi->lock); |
1e18bda8 | 1618 | if (!list_empty(&fi->write_files)) { |
72523425 MS |
1619 | ff = list_entry(fi->write_files.next, struct fuse_file, |
1620 | write_entry); | |
1621 | fuse_file_get(ff); | |
1622 | } | |
f15ecfef | 1623 | spin_unlock(&fi->lock); |
adcadfa8 PE |
1624 | |
1625 | return ff; | |
1626 | } | |
1627 | ||
1e18bda8 MS |
1628 | static struct fuse_file *fuse_write_file_get(struct fuse_conn *fc, |
1629 | struct fuse_inode *fi) | |
1630 | { | |
1631 | struct fuse_file *ff = __fuse_write_file_get(fc, fi); | |
1632 | WARN_ON(!ff); | |
1633 | return ff; | |
1634 | } | |
1635 | ||
1636 | int fuse_write_inode(struct inode *inode, struct writeback_control *wbc) | |
1637 | { | |
1638 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1639 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1640 | struct fuse_file *ff; | |
1641 | int err; | |
1642 | ||
1643 | ff = __fuse_write_file_get(fc, fi); | |
ab9e13f7 | 1644 | err = fuse_flush_times(inode, ff); |
1e18bda8 | 1645 | if (ff) |
2e64ff15 | 1646 | fuse_file_put(ff, false, false); |
1e18bda8 MS |
1647 | |
1648 | return err; | |
1649 | } | |
1650 | ||
3be5a52b MS |
1651 | static int fuse_writepage_locked(struct page *page) |
1652 | { | |
1653 | struct address_space *mapping = page->mapping; | |
1654 | struct inode *inode = mapping->host; | |
1655 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1656 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1657 | struct fuse_req *req; | |
3be5a52b | 1658 | struct page *tmp_page; |
72523425 | 1659 | int error = -ENOMEM; |
3be5a52b MS |
1660 | |
1661 | set_page_writeback(page); | |
1662 | ||
4250c066 | 1663 | req = fuse_request_alloc_nofs(1); |
3be5a52b MS |
1664 | if (!req) |
1665 | goto err; | |
1666 | ||
825d6d33 MS |
1667 | /* writeback always goes to bg_queue */ |
1668 | __set_bit(FR_BACKGROUND, &req->flags); | |
3be5a52b MS |
1669 | tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); |
1670 | if (!tmp_page) | |
1671 | goto err_free; | |
1672 | ||
72523425 | 1673 | error = -EIO; |
26d614df | 1674 | req->ff = fuse_write_file_get(fc, fi); |
72523425 | 1675 | if (!req->ff) |
27f1b363 | 1676 | goto err_nofile; |
72523425 | 1677 | |
adcadfa8 | 1678 | fuse_write_fill(req, req->ff, page_offset(page), 0); |
3be5a52b MS |
1679 | |
1680 | copy_highpage(tmp_page, page); | |
2d698b07 | 1681 | req->misc.write.in.write_flags |= FUSE_WRITE_CACHE; |
8b284dc4 | 1682 | req->misc.write.next = NULL; |
f4975c67 | 1683 | req->in.argpages = 1; |
3be5a52b MS |
1684 | req->num_pages = 1; |
1685 | req->pages[0] = tmp_page; | |
b2430d75 | 1686 | req->page_descs[0].offset = 0; |
85f40aec | 1687 | req->page_descs[0].length = PAGE_SIZE; |
3be5a52b MS |
1688 | req->end = fuse_writepage_end; |
1689 | req->inode = inode; | |
1690 | ||
93f78d88 | 1691 | inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK); |
11fb9989 | 1692 | inc_node_page_state(tmp_page, NR_WRITEBACK_TEMP); |
3be5a52b | 1693 | |
f15ecfef | 1694 | spin_lock(&fi->lock); |
3be5a52b MS |
1695 | list_add(&req->writepages_entry, &fi->writepages); |
1696 | list_add_tail(&req->list, &fi->queued_writes); | |
1697 | fuse_flush_writepages(inode); | |
f15ecfef | 1698 | spin_unlock(&fi->lock); |
3be5a52b | 1699 | |
4a4ac4eb MP |
1700 | end_page_writeback(page); |
1701 | ||
3be5a52b MS |
1702 | return 0; |
1703 | ||
27f1b363 MP |
1704 | err_nofile: |
1705 | __free_page(tmp_page); | |
3be5a52b MS |
1706 | err_free: |
1707 | fuse_request_free(req); | |
1708 | err: | |
9183976e | 1709 | mapping_set_error(page->mapping, error); |
3be5a52b | 1710 | end_page_writeback(page); |
72523425 | 1711 | return error; |
3be5a52b MS |
1712 | } |
1713 | ||
1714 | static int fuse_writepage(struct page *page, struct writeback_control *wbc) | |
1715 | { | |
1716 | int err; | |
1717 | ||
ff17be08 MS |
1718 | if (fuse_page_is_writeback(page->mapping->host, page->index)) { |
1719 | /* | |
1720 | * ->writepages() should be called for sync() and friends. We | |
1721 | * should only get here on direct reclaim and then we are | |
1722 | * allowed to skip a page which is already in flight | |
1723 | */ | |
1724 | WARN_ON(wbc->sync_mode == WB_SYNC_ALL); | |
1725 | ||
1726 | redirty_page_for_writepage(wbc, page); | |
1727 | return 0; | |
1728 | } | |
1729 | ||
3be5a52b MS |
1730 | err = fuse_writepage_locked(page); |
1731 | unlock_page(page); | |
1732 | ||
1733 | return err; | |
1734 | } | |
1735 | ||
26d614df PE |
1736 | struct fuse_fill_wb_data { |
1737 | struct fuse_req *req; | |
1738 | struct fuse_file *ff; | |
1739 | struct inode *inode; | |
2d033eaa | 1740 | struct page **orig_pages; |
26d614df PE |
1741 | }; |
1742 | ||
1743 | static void fuse_writepages_send(struct fuse_fill_wb_data *data) | |
1744 | { | |
1745 | struct fuse_req *req = data->req; | |
1746 | struct inode *inode = data->inode; | |
26d614df | 1747 | struct fuse_inode *fi = get_fuse_inode(inode); |
2d033eaa MP |
1748 | int num_pages = req->num_pages; |
1749 | int i; | |
26d614df PE |
1750 | |
1751 | req->ff = fuse_file_get(data->ff); | |
f15ecfef | 1752 | spin_lock(&fi->lock); |
26d614df PE |
1753 | list_add_tail(&req->list, &fi->queued_writes); |
1754 | fuse_flush_writepages(inode); | |
f15ecfef | 1755 | spin_unlock(&fi->lock); |
2d033eaa MP |
1756 | |
1757 | for (i = 0; i < num_pages; i++) | |
1758 | end_page_writeback(data->orig_pages[i]); | |
26d614df PE |
1759 | } |
1760 | ||
7f305ca1 MS |
1761 | /* |
1762 | * First recheck under fi->lock if the offending offset is still under | |
419234d5 MS |
1763 | * writeback. If yes, then iterate auxiliary write requests, to see if there's |
1764 | * one already added for a page at this offset. If there's none, then insert | |
1765 | * this new request onto the auxiliary list, otherwise reuse the existing one by | |
7f305ca1 MS |
1766 | * copying the new page contents over to the old temporary page. |
1767 | */ | |
8b284dc4 MS |
1768 | static bool fuse_writepage_in_flight(struct fuse_req *new_req, |
1769 | struct page *page) | |
1770 | { | |
1771 | struct fuse_conn *fc = get_fuse_conn(new_req->inode); | |
1772 | struct fuse_inode *fi = get_fuse_inode(new_req->inode); | |
1773 | struct fuse_req *tmp; | |
1774 | struct fuse_req *old_req; | |
8b284dc4 | 1775 | |
7f305ca1 | 1776 | WARN_ON(new_req->num_pages != 0); |
8b284dc4 | 1777 | |
f15ecfef | 1778 | spin_lock(&fi->lock); |
8b284dc4 | 1779 | list_del(&new_req->writepages_entry); |
2fe93bd4 MS |
1780 | old_req = fuse_find_writeback(fi, page->index, page->index); |
1781 | if (!old_req) { | |
f6011081 | 1782 | list_add(&new_req->writepages_entry, &fi->writepages); |
f15ecfef | 1783 | spin_unlock(&fi->lock); |
2fe93bd4 | 1784 | return false; |
f6011081 | 1785 | } |
8b284dc4 | 1786 | |
f6011081 | 1787 | new_req->num_pages = 1; |
419234d5 | 1788 | for (tmp = old_req->misc.write.next; tmp; tmp = tmp->misc.write.next) { |
7f305ca1 MS |
1789 | pgoff_t curr_index; |
1790 | ||
1791 | WARN_ON(tmp->inode != new_req->inode); | |
09cbfeaf | 1792 | curr_index = tmp->misc.write.in.offset >> PAGE_SHIFT; |
419234d5 MS |
1793 | if (curr_index == page->index) { |
1794 | WARN_ON(tmp->num_pages != 1); | |
1795 | WARN_ON(!test_bit(FR_PENDING, &tmp->flags)); | |
c5de16cc | 1796 | swap(tmp->pages[0], new_req->pages[0]); |
7f305ca1 | 1797 | break; |
8b284dc4 MS |
1798 | } |
1799 | } | |
1800 | ||
7f305ca1 MS |
1801 | if (!tmp) { |
1802 | new_req->misc.write.next = old_req->misc.write.next; | |
1803 | old_req->misc.write.next = new_req; | |
1804 | } | |
41b6e41f | 1805 | |
f15ecfef | 1806 | spin_unlock(&fi->lock); |
7f305ca1 MS |
1807 | |
1808 | if (tmp) { | |
1809 | struct backing_dev_info *bdi = inode_to_bdi(new_req->inode); | |
8b284dc4 | 1810 | |
93f78d88 | 1811 | dec_wb_stat(&bdi->wb, WB_WRITEBACK); |
a2ebba82 | 1812 | dec_node_page_state(new_req->pages[0], NR_WRITEBACK_TEMP); |
93f78d88 | 1813 | wb_writeout_inc(&bdi->wb); |
8b284dc4 MS |
1814 | fuse_writepage_free(fc, new_req); |
1815 | fuse_request_free(new_req); | |
8b284dc4 | 1816 | } |
7f305ca1 | 1817 | |
2fe93bd4 | 1818 | return true; |
8b284dc4 MS |
1819 | } |
1820 | ||
26d614df PE |
1821 | static int fuse_writepages_fill(struct page *page, |
1822 | struct writeback_control *wbc, void *_data) | |
1823 | { | |
1824 | struct fuse_fill_wb_data *data = _data; | |
1825 | struct fuse_req *req = data->req; | |
1826 | struct inode *inode = data->inode; | |
f15ecfef | 1827 | struct fuse_inode *fi = get_fuse_inode(inode); |
26d614df PE |
1828 | struct fuse_conn *fc = get_fuse_conn(inode); |
1829 | struct page *tmp_page; | |
8b284dc4 | 1830 | bool is_writeback; |
26d614df PE |
1831 | int err; |
1832 | ||
1833 | if (!data->ff) { | |
1834 | err = -EIO; | |
1835 | data->ff = fuse_write_file_get(fc, get_fuse_inode(inode)); | |
1836 | if (!data->ff) | |
1837 | goto out_unlock; | |
1838 | } | |
1839 | ||
8b284dc4 MS |
1840 | /* |
1841 | * Being under writeback is unlikely but possible. For example direct | |
1842 | * read to an mmaped fuse file will set the page dirty twice; once when | |
1843 | * the pages are faulted with get_user_pages(), and then after the read | |
1844 | * completed. | |
1845 | */ | |
1846 | is_writeback = fuse_page_is_writeback(inode, page->index); | |
1847 | ||
1848 | if (req && req->num_pages && | |
5da784cc | 1849 | (is_writeback || req->num_pages == fc->max_pages || |
09cbfeaf | 1850 | (req->num_pages + 1) * PAGE_SIZE > fc->max_write || |
8b284dc4 MS |
1851 | data->orig_pages[req->num_pages - 1]->index + 1 != page->index)) { |
1852 | fuse_writepages_send(data); | |
1853 | data->req = NULL; | |
e52a8250 MS |
1854 | } else if (req && req->num_pages == req->max_pages) { |
1855 | if (!fuse_req_realloc_pages(fc, req, GFP_NOFS)) { | |
1856 | fuse_writepages_send(data); | |
1857 | req = data->req = NULL; | |
1858 | } | |
26d614df | 1859 | } |
e52a8250 | 1860 | |
26d614df PE |
1861 | err = -ENOMEM; |
1862 | tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
1863 | if (!tmp_page) | |
1864 | goto out_unlock; | |
1865 | ||
1866 | /* | |
1867 | * The page must not be redirtied until the writeout is completed | |
1868 | * (i.e. userspace has sent a reply to the write request). Otherwise | |
1869 | * there could be more than one temporary page instance for each real | |
1870 | * page. | |
1871 | * | |
1872 | * This is ensured by holding the page lock in page_mkwrite() while | |
1873 | * checking fuse_page_is_writeback(). We already hold the page lock | |
1874 | * since clear_page_dirty_for_io() and keep it held until we add the | |
1875 | * request to the fi->writepages list and increment req->num_pages. | |
1876 | * After this fuse_page_is_writeback() will indicate that the page is | |
1877 | * under writeback, so we can release the page lock. | |
1878 | */ | |
1879 | if (data->req == NULL) { | |
1880 | struct fuse_inode *fi = get_fuse_inode(inode); | |
1881 | ||
1882 | err = -ENOMEM; | |
e52a8250 | 1883 | req = fuse_request_alloc_nofs(FUSE_REQ_INLINE_PAGES); |
26d614df PE |
1884 | if (!req) { |
1885 | __free_page(tmp_page); | |
1886 | goto out_unlock; | |
1887 | } | |
1888 | ||
1889 | fuse_write_fill(req, data->ff, page_offset(page), 0); | |
1890 | req->misc.write.in.write_flags |= FUSE_WRITE_CACHE; | |
8b284dc4 | 1891 | req->misc.write.next = NULL; |
26d614df | 1892 | req->in.argpages = 1; |
825d6d33 | 1893 | __set_bit(FR_BACKGROUND, &req->flags); |
26d614df PE |
1894 | req->num_pages = 0; |
1895 | req->end = fuse_writepage_end; | |
1896 | req->inode = inode; | |
1897 | ||
f15ecfef | 1898 | spin_lock(&fi->lock); |
26d614df | 1899 | list_add(&req->writepages_entry, &fi->writepages); |
f15ecfef | 1900 | spin_unlock(&fi->lock); |
26d614df PE |
1901 | |
1902 | data->req = req; | |
1903 | } | |
1904 | set_page_writeback(page); | |
1905 | ||
1906 | copy_highpage(tmp_page, page); | |
1907 | req->pages[req->num_pages] = tmp_page; | |
1908 | req->page_descs[req->num_pages].offset = 0; | |
1909 | req->page_descs[req->num_pages].length = PAGE_SIZE; | |
1910 | ||
93f78d88 | 1911 | inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK); |
11fb9989 | 1912 | inc_node_page_state(tmp_page, NR_WRITEBACK_TEMP); |
8b284dc4 MS |
1913 | |
1914 | err = 0; | |
1915 | if (is_writeback && fuse_writepage_in_flight(req, page)) { | |
1916 | end_page_writeback(page); | |
1917 | data->req = NULL; | |
1918 | goto out_unlock; | |
1919 | } | |
2d033eaa | 1920 | data->orig_pages[req->num_pages] = page; |
26d614df PE |
1921 | |
1922 | /* | |
f15ecfef | 1923 | * Protected by fi->lock against concurrent access by |
26d614df PE |
1924 | * fuse_page_is_writeback(). |
1925 | */ | |
f15ecfef | 1926 | spin_lock(&fi->lock); |
26d614df | 1927 | req->num_pages++; |
f15ecfef | 1928 | spin_unlock(&fi->lock); |
26d614df | 1929 | |
26d614df PE |
1930 | out_unlock: |
1931 | unlock_page(page); | |
1932 | ||
1933 | return err; | |
1934 | } | |
1935 | ||
1936 | static int fuse_writepages(struct address_space *mapping, | |
1937 | struct writeback_control *wbc) | |
1938 | { | |
1939 | struct inode *inode = mapping->host; | |
5da784cc | 1940 | struct fuse_conn *fc = get_fuse_conn(inode); |
26d614df PE |
1941 | struct fuse_fill_wb_data data; |
1942 | int err; | |
1943 | ||
1944 | err = -EIO; | |
1945 | if (is_bad_inode(inode)) | |
1946 | goto out; | |
1947 | ||
1948 | data.inode = inode; | |
1949 | data.req = NULL; | |
1950 | data.ff = NULL; | |
1951 | ||
2d033eaa | 1952 | err = -ENOMEM; |
5da784cc | 1953 | data.orig_pages = kcalloc(fc->max_pages, |
f2b3455e | 1954 | sizeof(struct page *), |
2d033eaa MP |
1955 | GFP_NOFS); |
1956 | if (!data.orig_pages) | |
1957 | goto out; | |
1958 | ||
26d614df PE |
1959 | err = write_cache_pages(mapping, wbc, fuse_writepages_fill, &data); |
1960 | if (data.req) { | |
1961 | /* Ignore errors if we can write at least one page */ | |
1962 | BUG_ON(!data.req->num_pages); | |
1963 | fuse_writepages_send(&data); | |
1964 | err = 0; | |
1965 | } | |
1966 | if (data.ff) | |
2e64ff15 | 1967 | fuse_file_put(data.ff, false, false); |
2d033eaa MP |
1968 | |
1969 | kfree(data.orig_pages); | |
26d614df PE |
1970 | out: |
1971 | return err; | |
1972 | } | |
1973 | ||
6b12c1b3 PE |
1974 | /* |
1975 | * It's worthy to make sure that space is reserved on disk for the write, | |
1976 | * but how to implement it without killing performance need more thinking. | |
1977 | */ | |
1978 | static int fuse_write_begin(struct file *file, struct address_space *mapping, | |
1979 | loff_t pos, unsigned len, unsigned flags, | |
1980 | struct page **pagep, void **fsdata) | |
1981 | { | |
09cbfeaf | 1982 | pgoff_t index = pos >> PAGE_SHIFT; |
a455589f | 1983 | struct fuse_conn *fc = get_fuse_conn(file_inode(file)); |
6b12c1b3 PE |
1984 | struct page *page; |
1985 | loff_t fsize; | |
1986 | int err = -ENOMEM; | |
1987 | ||
1988 | WARN_ON(!fc->writeback_cache); | |
1989 | ||
1990 | page = grab_cache_page_write_begin(mapping, index, flags); | |
1991 | if (!page) | |
1992 | goto error; | |
1993 | ||
1994 | fuse_wait_on_page_writeback(mapping->host, page->index); | |
1995 | ||
09cbfeaf | 1996 | if (PageUptodate(page) || len == PAGE_SIZE) |
6b12c1b3 PE |
1997 | goto success; |
1998 | /* | |
1999 | * Check if the start this page comes after the end of file, in which | |
2000 | * case the readpage can be optimized away. | |
2001 | */ | |
2002 | fsize = i_size_read(mapping->host); | |
09cbfeaf KS |
2003 | if (fsize <= (pos & PAGE_MASK)) { |
2004 | size_t off = pos & ~PAGE_MASK; | |
6b12c1b3 PE |
2005 | if (off) |
2006 | zero_user_segment(page, 0, off); | |
2007 | goto success; | |
2008 | } | |
2009 | err = fuse_do_readpage(file, page); | |
2010 | if (err) | |
2011 | goto cleanup; | |
2012 | success: | |
2013 | *pagep = page; | |
2014 | return 0; | |
2015 | ||
2016 | cleanup: | |
2017 | unlock_page(page); | |
09cbfeaf | 2018 | put_page(page); |
6b12c1b3 PE |
2019 | error: |
2020 | return err; | |
2021 | } | |
2022 | ||
2023 | static int fuse_write_end(struct file *file, struct address_space *mapping, | |
2024 | loff_t pos, unsigned len, unsigned copied, | |
2025 | struct page *page, void *fsdata) | |
2026 | { | |
2027 | struct inode *inode = page->mapping->host; | |
2028 | ||
59c3b76c MS |
2029 | /* Haven't copied anything? Skip zeroing, size extending, dirtying. */ |
2030 | if (!copied) | |
2031 | goto unlock; | |
2032 | ||
6b12c1b3 PE |
2033 | if (!PageUptodate(page)) { |
2034 | /* Zero any unwritten bytes at the end of the page */ | |
09cbfeaf | 2035 | size_t endoff = (pos + copied) & ~PAGE_MASK; |
6b12c1b3 | 2036 | if (endoff) |
09cbfeaf | 2037 | zero_user_segment(page, endoff, PAGE_SIZE); |
6b12c1b3 PE |
2038 | SetPageUptodate(page); |
2039 | } | |
2040 | ||
2041 | fuse_write_update_size(inode, pos + copied); | |
2042 | set_page_dirty(page); | |
59c3b76c MS |
2043 | |
2044 | unlock: | |
6b12c1b3 | 2045 | unlock_page(page); |
09cbfeaf | 2046 | put_page(page); |
6b12c1b3 PE |
2047 | |
2048 | return copied; | |
2049 | } | |
2050 | ||
3be5a52b MS |
2051 | static int fuse_launder_page(struct page *page) |
2052 | { | |
2053 | int err = 0; | |
2054 | if (clear_page_dirty_for_io(page)) { | |
2055 | struct inode *inode = page->mapping->host; | |
2056 | err = fuse_writepage_locked(page); | |
2057 | if (!err) | |
2058 | fuse_wait_on_page_writeback(inode, page->index); | |
2059 | } | |
2060 | return err; | |
2061 | } | |
2062 | ||
2063 | /* | |
2064 | * Write back dirty pages now, because there may not be any suitable | |
2065 | * open files later | |
2066 | */ | |
2067 | static void fuse_vma_close(struct vm_area_struct *vma) | |
2068 | { | |
2069 | filemap_write_and_wait(vma->vm_file->f_mapping); | |
2070 | } | |
2071 | ||
2072 | /* | |
2073 | * Wait for writeback against this page to complete before allowing it | |
2074 | * to be marked dirty again, and hence written back again, possibly | |
2075 | * before the previous writepage completed. | |
2076 | * | |
2077 | * Block here, instead of in ->writepage(), so that the userspace fs | |
2078 | * can only block processes actually operating on the filesystem. | |
2079 | * | |
2080 | * Otherwise unprivileged userspace fs would be able to block | |
2081 | * unrelated: | |
2082 | * | |
2083 | * - page migration | |
2084 | * - sync(2) | |
2085 | * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER | |
2086 | */ | |
46fb504a | 2087 | static vm_fault_t fuse_page_mkwrite(struct vm_fault *vmf) |
3be5a52b | 2088 | { |
c2ec175c | 2089 | struct page *page = vmf->page; |
11bac800 | 2090 | struct inode *inode = file_inode(vmf->vma->vm_file); |
cca24370 | 2091 | |
11bac800 | 2092 | file_update_time(vmf->vma->vm_file); |
cca24370 MS |
2093 | lock_page(page); |
2094 | if (page->mapping != inode->i_mapping) { | |
2095 | unlock_page(page); | |
2096 | return VM_FAULT_NOPAGE; | |
2097 | } | |
3be5a52b MS |
2098 | |
2099 | fuse_wait_on_page_writeback(inode, page->index); | |
cca24370 | 2100 | return VM_FAULT_LOCKED; |
3be5a52b MS |
2101 | } |
2102 | ||
f0f37e2f | 2103 | static const struct vm_operations_struct fuse_file_vm_ops = { |
3be5a52b MS |
2104 | .close = fuse_vma_close, |
2105 | .fault = filemap_fault, | |
f1820361 | 2106 | .map_pages = filemap_map_pages, |
3be5a52b MS |
2107 | .page_mkwrite = fuse_page_mkwrite, |
2108 | }; | |
2109 | ||
2110 | static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma) | |
2111 | { | |
650b22b9 PE |
2112 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) |
2113 | fuse_link_write_file(file); | |
2114 | ||
3be5a52b MS |
2115 | file_accessed(file); |
2116 | vma->vm_ops = &fuse_file_vm_ops; | |
b6aeaded MS |
2117 | return 0; |
2118 | } | |
2119 | ||
fc280c96 MS |
2120 | static int fuse_direct_mmap(struct file *file, struct vm_area_struct *vma) |
2121 | { | |
2122 | /* Can't provide the coherency needed for MAP_SHARED */ | |
2123 | if (vma->vm_flags & VM_MAYSHARE) | |
2124 | return -ENODEV; | |
2125 | ||
3121bfe7 MS |
2126 | invalidate_inode_pages2(file->f_mapping); |
2127 | ||
fc280c96 MS |
2128 | return generic_file_mmap(file, vma); |
2129 | } | |
2130 | ||
0b6e9ea0 SF |
2131 | static int convert_fuse_file_lock(struct fuse_conn *fc, |
2132 | const struct fuse_file_lock *ffl, | |
71421259 MS |
2133 | struct file_lock *fl) |
2134 | { | |
2135 | switch (ffl->type) { | |
2136 | case F_UNLCK: | |
2137 | break; | |
2138 | ||
2139 | case F_RDLCK: | |
2140 | case F_WRLCK: | |
2141 | if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX || | |
2142 | ffl->end < ffl->start) | |
2143 | return -EIO; | |
2144 | ||
2145 | fl->fl_start = ffl->start; | |
2146 | fl->fl_end = ffl->end; | |
0b6e9ea0 SF |
2147 | |
2148 | /* | |
9d5b86ac BC |
2149 | * Convert pid into init's pid namespace. The locks API will |
2150 | * translate it into the caller's pid namespace. | |
0b6e9ea0 SF |
2151 | */ |
2152 | rcu_read_lock(); | |
9d5b86ac | 2153 | fl->fl_pid = pid_nr_ns(find_pid_ns(ffl->pid, fc->pid_ns), &init_pid_ns); |
0b6e9ea0 | 2154 | rcu_read_unlock(); |
71421259 MS |
2155 | break; |
2156 | ||
2157 | default: | |
2158 | return -EIO; | |
2159 | } | |
2160 | fl->fl_type = ffl->type; | |
2161 | return 0; | |
2162 | } | |
2163 | ||
7078187a | 2164 | static void fuse_lk_fill(struct fuse_args *args, struct file *file, |
a9ff4f87 | 2165 | const struct file_lock *fl, int opcode, pid_t pid, |
7078187a | 2166 | int flock, struct fuse_lk_in *inarg) |
71421259 | 2167 | { |
6131ffaa | 2168 | struct inode *inode = file_inode(file); |
9c8ef561 | 2169 | struct fuse_conn *fc = get_fuse_conn(inode); |
71421259 | 2170 | struct fuse_file *ff = file->private_data; |
7078187a MS |
2171 | |
2172 | memset(inarg, 0, sizeof(*inarg)); | |
2173 | inarg->fh = ff->fh; | |
2174 | inarg->owner = fuse_lock_owner_id(fc, fl->fl_owner); | |
2175 | inarg->lk.start = fl->fl_start; | |
2176 | inarg->lk.end = fl->fl_end; | |
2177 | inarg->lk.type = fl->fl_type; | |
2178 | inarg->lk.pid = pid; | |
a9ff4f87 | 2179 | if (flock) |
7078187a MS |
2180 | inarg->lk_flags |= FUSE_LK_FLOCK; |
2181 | args->in.h.opcode = opcode; | |
2182 | args->in.h.nodeid = get_node_id(inode); | |
2183 | args->in.numargs = 1; | |
2184 | args->in.args[0].size = sizeof(*inarg); | |
2185 | args->in.args[0].value = inarg; | |
71421259 MS |
2186 | } |
2187 | ||
2188 | static int fuse_getlk(struct file *file, struct file_lock *fl) | |
2189 | { | |
6131ffaa | 2190 | struct inode *inode = file_inode(file); |
71421259 | 2191 | struct fuse_conn *fc = get_fuse_conn(inode); |
7078187a MS |
2192 | FUSE_ARGS(args); |
2193 | struct fuse_lk_in inarg; | |
71421259 MS |
2194 | struct fuse_lk_out outarg; |
2195 | int err; | |
2196 | ||
7078187a MS |
2197 | fuse_lk_fill(&args, file, fl, FUSE_GETLK, 0, 0, &inarg); |
2198 | args.out.numargs = 1; | |
2199 | args.out.args[0].size = sizeof(outarg); | |
2200 | args.out.args[0].value = &outarg; | |
2201 | err = fuse_simple_request(fc, &args); | |
71421259 | 2202 | if (!err) |
0b6e9ea0 | 2203 | err = convert_fuse_file_lock(fc, &outarg.lk, fl); |
71421259 MS |
2204 | |
2205 | return err; | |
2206 | } | |
2207 | ||
a9ff4f87 | 2208 | static int fuse_setlk(struct file *file, struct file_lock *fl, int flock) |
71421259 | 2209 | { |
6131ffaa | 2210 | struct inode *inode = file_inode(file); |
71421259 | 2211 | struct fuse_conn *fc = get_fuse_conn(inode); |
7078187a MS |
2212 | FUSE_ARGS(args); |
2213 | struct fuse_lk_in inarg; | |
71421259 | 2214 | int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK; |
0b6e9ea0 SF |
2215 | struct pid *pid = fl->fl_type != F_UNLCK ? task_tgid(current) : NULL; |
2216 | pid_t pid_nr = pid_nr_ns(pid, fc->pid_ns); | |
71421259 MS |
2217 | int err; |
2218 | ||
8fb47a4f | 2219 | if (fl->fl_lmops && fl->fl_lmops->lm_grant) { |
48e90761 MS |
2220 | /* NLM needs asynchronous locks, which we don't support yet */ |
2221 | return -ENOLCK; | |
2222 | } | |
2223 | ||
71421259 | 2224 | /* Unlock on close is handled by the flush method */ |
50f2112c | 2225 | if ((fl->fl_flags & FL_CLOSE_POSIX) == FL_CLOSE_POSIX) |
71421259 MS |
2226 | return 0; |
2227 | ||
0b6e9ea0 | 2228 | fuse_lk_fill(&args, file, fl, opcode, pid_nr, flock, &inarg); |
7078187a | 2229 | err = fuse_simple_request(fc, &args); |
71421259 | 2230 | |
a4d27e75 MS |
2231 | /* locking is restartable */ |
2232 | if (err == -EINTR) | |
2233 | err = -ERESTARTSYS; | |
7078187a | 2234 | |
71421259 MS |
2235 | return err; |
2236 | } | |
2237 | ||
2238 | static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl) | |
2239 | { | |
6131ffaa | 2240 | struct inode *inode = file_inode(file); |
71421259 MS |
2241 | struct fuse_conn *fc = get_fuse_conn(inode); |
2242 | int err; | |
2243 | ||
48e90761 MS |
2244 | if (cmd == F_CANCELLK) { |
2245 | err = 0; | |
2246 | } else if (cmd == F_GETLK) { | |
71421259 | 2247 | if (fc->no_lock) { |
9d6a8c5c | 2248 | posix_test_lock(file, fl); |
71421259 MS |
2249 | err = 0; |
2250 | } else | |
2251 | err = fuse_getlk(file, fl); | |
2252 | } else { | |
2253 | if (fc->no_lock) | |
48e90761 | 2254 | err = posix_lock_file(file, fl, NULL); |
71421259 | 2255 | else |
a9ff4f87 | 2256 | err = fuse_setlk(file, fl, 0); |
71421259 MS |
2257 | } |
2258 | return err; | |
2259 | } | |
2260 | ||
a9ff4f87 MS |
2261 | static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl) |
2262 | { | |
6131ffaa | 2263 | struct inode *inode = file_inode(file); |
a9ff4f87 MS |
2264 | struct fuse_conn *fc = get_fuse_conn(inode); |
2265 | int err; | |
2266 | ||
37fb3a30 | 2267 | if (fc->no_flock) { |
4f656367 | 2268 | err = locks_lock_file_wait(file, fl); |
a9ff4f87 | 2269 | } else { |
37fb3a30 MS |
2270 | struct fuse_file *ff = file->private_data; |
2271 | ||
a9ff4f87 | 2272 | /* emulate flock with POSIX locks */ |
37fb3a30 | 2273 | ff->flock = true; |
a9ff4f87 MS |
2274 | err = fuse_setlk(file, fl, 1); |
2275 | } | |
2276 | ||
2277 | return err; | |
2278 | } | |
2279 | ||
b2d2272f MS |
2280 | static sector_t fuse_bmap(struct address_space *mapping, sector_t block) |
2281 | { | |
2282 | struct inode *inode = mapping->host; | |
2283 | struct fuse_conn *fc = get_fuse_conn(inode); | |
7078187a | 2284 | FUSE_ARGS(args); |
b2d2272f MS |
2285 | struct fuse_bmap_in inarg; |
2286 | struct fuse_bmap_out outarg; | |
2287 | int err; | |
2288 | ||
2289 | if (!inode->i_sb->s_bdev || fc->no_bmap) | |
2290 | return 0; | |
2291 | ||
b2d2272f MS |
2292 | memset(&inarg, 0, sizeof(inarg)); |
2293 | inarg.block = block; | |
2294 | inarg.blocksize = inode->i_sb->s_blocksize; | |
7078187a MS |
2295 | args.in.h.opcode = FUSE_BMAP; |
2296 | args.in.h.nodeid = get_node_id(inode); | |
2297 | args.in.numargs = 1; | |
2298 | args.in.args[0].size = sizeof(inarg); | |
2299 | args.in.args[0].value = &inarg; | |
2300 | args.out.numargs = 1; | |
2301 | args.out.args[0].size = sizeof(outarg); | |
2302 | args.out.args[0].value = &outarg; | |
2303 | err = fuse_simple_request(fc, &args); | |
b2d2272f MS |
2304 | if (err == -ENOSYS) |
2305 | fc->no_bmap = 1; | |
2306 | ||
2307 | return err ? 0 : outarg.block; | |
2308 | } | |
2309 | ||
0b5da8db R |
2310 | static loff_t fuse_lseek(struct file *file, loff_t offset, int whence) |
2311 | { | |
2312 | struct inode *inode = file->f_mapping->host; | |
2313 | struct fuse_conn *fc = get_fuse_conn(inode); | |
2314 | struct fuse_file *ff = file->private_data; | |
2315 | FUSE_ARGS(args); | |
2316 | struct fuse_lseek_in inarg = { | |
2317 | .fh = ff->fh, | |
2318 | .offset = offset, | |
2319 | .whence = whence | |
2320 | }; | |
2321 | struct fuse_lseek_out outarg; | |
2322 | int err; | |
2323 | ||
2324 | if (fc->no_lseek) | |
2325 | goto fallback; | |
2326 | ||
2327 | args.in.h.opcode = FUSE_LSEEK; | |
2328 | args.in.h.nodeid = ff->nodeid; | |
2329 | args.in.numargs = 1; | |
2330 | args.in.args[0].size = sizeof(inarg); | |
2331 | args.in.args[0].value = &inarg; | |
2332 | args.out.numargs = 1; | |
2333 | args.out.args[0].size = sizeof(outarg); | |
2334 | args.out.args[0].value = &outarg; | |
2335 | err = fuse_simple_request(fc, &args); | |
2336 | if (err) { | |
2337 | if (err == -ENOSYS) { | |
2338 | fc->no_lseek = 1; | |
2339 | goto fallback; | |
2340 | } | |
2341 | return err; | |
2342 | } | |
2343 | ||
2344 | return vfs_setpos(file, outarg.offset, inode->i_sb->s_maxbytes); | |
2345 | ||
2346 | fallback: | |
5b97eeac | 2347 | err = fuse_update_attributes(inode, file); |
0b5da8db R |
2348 | if (!err) |
2349 | return generic_file_llseek(file, offset, whence); | |
2350 | else | |
2351 | return err; | |
2352 | } | |
2353 | ||
965c8e59 | 2354 | static loff_t fuse_file_llseek(struct file *file, loff_t offset, int whence) |
5559b8f4 MS |
2355 | { |
2356 | loff_t retval; | |
6131ffaa | 2357 | struct inode *inode = file_inode(file); |
5559b8f4 | 2358 | |
0b5da8db R |
2359 | switch (whence) { |
2360 | case SEEK_SET: | |
2361 | case SEEK_CUR: | |
2362 | /* No i_mutex protection necessary for SEEK_CUR and SEEK_SET */ | |
965c8e59 | 2363 | retval = generic_file_llseek(file, offset, whence); |
0b5da8db R |
2364 | break; |
2365 | case SEEK_END: | |
5955102c | 2366 | inode_lock(inode); |
5b97eeac | 2367 | retval = fuse_update_attributes(inode, file); |
0b5da8db R |
2368 | if (!retval) |
2369 | retval = generic_file_llseek(file, offset, whence); | |
5955102c | 2370 | inode_unlock(inode); |
0b5da8db R |
2371 | break; |
2372 | case SEEK_HOLE: | |
2373 | case SEEK_DATA: | |
5955102c | 2374 | inode_lock(inode); |
0b5da8db | 2375 | retval = fuse_lseek(file, offset, whence); |
5955102c | 2376 | inode_unlock(inode); |
0b5da8db R |
2377 | break; |
2378 | default: | |
2379 | retval = -EINVAL; | |
2380 | } | |
c07c3d19 | 2381 | |
5559b8f4 MS |
2382 | return retval; |
2383 | } | |
2384 | ||
d9d318d3 MS |
2385 | /* |
2386 | * CUSE servers compiled on 32bit broke on 64bit kernels because the | |
2387 | * ABI was defined to be 'struct iovec' which is different on 32bit | |
2388 | * and 64bit. Fortunately we can determine which structure the server | |
2389 | * used from the size of the reply. | |
2390 | */ | |
1baa26b2 MS |
2391 | static int fuse_copy_ioctl_iovec_old(struct iovec *dst, void *src, |
2392 | size_t transferred, unsigned count, | |
2393 | bool is_compat) | |
d9d318d3 MS |
2394 | { |
2395 | #ifdef CONFIG_COMPAT | |
2396 | if (count * sizeof(struct compat_iovec) == transferred) { | |
2397 | struct compat_iovec *ciov = src; | |
2398 | unsigned i; | |
2399 | ||
2400 | /* | |
2401 | * With this interface a 32bit server cannot support | |
2402 | * non-compat (i.e. ones coming from 64bit apps) ioctl | |
2403 | * requests | |
2404 | */ | |
2405 | if (!is_compat) | |
2406 | return -EINVAL; | |
2407 | ||
2408 | for (i = 0; i < count; i++) { | |
2409 | dst[i].iov_base = compat_ptr(ciov[i].iov_base); | |
2410 | dst[i].iov_len = ciov[i].iov_len; | |
2411 | } | |
2412 | return 0; | |
2413 | } | |
2414 | #endif | |
2415 | ||
2416 | if (count * sizeof(struct iovec) != transferred) | |
2417 | return -EIO; | |
2418 | ||
2419 | memcpy(dst, src, transferred); | |
2420 | return 0; | |
2421 | } | |
2422 | ||
7572777e | 2423 | /* Make sure iov_length() won't overflow */ |
5da784cc CS |
2424 | static int fuse_verify_ioctl_iov(struct fuse_conn *fc, struct iovec *iov, |
2425 | size_t count) | |
7572777e MS |
2426 | { |
2427 | size_t n; | |
5da784cc | 2428 | u32 max = fc->max_pages << PAGE_SHIFT; |
7572777e | 2429 | |
fb6ccff6 | 2430 | for (n = 0; n < count; n++, iov++) { |
7572777e MS |
2431 | if (iov->iov_len > (size_t) max) |
2432 | return -ENOMEM; | |
2433 | max -= iov->iov_len; | |
2434 | } | |
2435 | return 0; | |
2436 | } | |
2437 | ||
1baa26b2 MS |
2438 | static int fuse_copy_ioctl_iovec(struct fuse_conn *fc, struct iovec *dst, |
2439 | void *src, size_t transferred, unsigned count, | |
2440 | bool is_compat) | |
2441 | { | |
2442 | unsigned i; | |
2443 | struct fuse_ioctl_iovec *fiov = src; | |
2444 | ||
2445 | if (fc->minor < 16) { | |
2446 | return fuse_copy_ioctl_iovec_old(dst, src, transferred, | |
2447 | count, is_compat); | |
2448 | } | |
2449 | ||
2450 | if (count * sizeof(struct fuse_ioctl_iovec) != transferred) | |
2451 | return -EIO; | |
2452 | ||
2453 | for (i = 0; i < count; i++) { | |
2454 | /* Did the server supply an inappropriate value? */ | |
2455 | if (fiov[i].base != (unsigned long) fiov[i].base || | |
2456 | fiov[i].len != (unsigned long) fiov[i].len) | |
2457 | return -EIO; | |
2458 | ||
2459 | dst[i].iov_base = (void __user *) (unsigned long) fiov[i].base; | |
2460 | dst[i].iov_len = (size_t) fiov[i].len; | |
2461 | ||
2462 | #ifdef CONFIG_COMPAT | |
2463 | if (is_compat && | |
2464 | (ptr_to_compat(dst[i].iov_base) != fiov[i].base || | |
2465 | (compat_size_t) dst[i].iov_len != fiov[i].len)) | |
2466 | return -EIO; | |
2467 | #endif | |
2468 | } | |
2469 | ||
2470 | return 0; | |
2471 | } | |
2472 | ||
2473 | ||
59efec7b TH |
2474 | /* |
2475 | * For ioctls, there is no generic way to determine how much memory | |
2476 | * needs to be read and/or written. Furthermore, ioctls are allowed | |
2477 | * to dereference the passed pointer, so the parameter requires deep | |
2478 | * copying but FUSE has no idea whatsoever about what to copy in or | |
2479 | * out. | |
2480 | * | |
2481 | * This is solved by allowing FUSE server to retry ioctl with | |
2482 | * necessary in/out iovecs. Let's assume the ioctl implementation | |
2483 | * needs to read in the following structure. | |
2484 | * | |
2485 | * struct a { | |
2486 | * char *buf; | |
2487 | * size_t buflen; | |
2488 | * } | |
2489 | * | |
2490 | * On the first callout to FUSE server, inarg->in_size and | |
2491 | * inarg->out_size will be NULL; then, the server completes the ioctl | |
2492 | * with FUSE_IOCTL_RETRY set in out->flags, out->in_iovs set to 1 and | |
2493 | * the actual iov array to | |
2494 | * | |
2495 | * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) } } | |
2496 | * | |
2497 | * which tells FUSE to copy in the requested area and retry the ioctl. | |
2498 | * On the second round, the server has access to the structure and | |
2499 | * from that it can tell what to look for next, so on the invocation, | |
2500 | * it sets FUSE_IOCTL_RETRY, out->in_iovs to 2 and iov array to | |
2501 | * | |
2502 | * { { .iov_base = inarg.arg, .iov_len = sizeof(struct a) }, | |
2503 | * { .iov_base = a.buf, .iov_len = a.buflen } } | |
2504 | * | |
2505 | * FUSE will copy both struct a and the pointed buffer from the | |
2506 | * process doing the ioctl and retry ioctl with both struct a and the | |
2507 | * buffer. | |
2508 | * | |
2509 | * This time, FUSE server has everything it needs and completes ioctl | |
2510 | * without FUSE_IOCTL_RETRY which finishes the ioctl call. | |
2511 | * | |
2512 | * Copying data out works the same way. | |
2513 | * | |
2514 | * Note that if FUSE_IOCTL_UNRESTRICTED is clear, the kernel | |
2515 | * automatically initializes in and out iovs by decoding @cmd with | |
2516 | * _IOC_* macros and the server is not allowed to request RETRY. This | |
2517 | * limits ioctl data transfers to well-formed ioctls and is the forced | |
2518 | * behavior for all FUSE servers. | |
2519 | */ | |
08cbf542 TH |
2520 | long fuse_do_ioctl(struct file *file, unsigned int cmd, unsigned long arg, |
2521 | unsigned int flags) | |
59efec7b | 2522 | { |
59efec7b | 2523 | struct fuse_file *ff = file->private_data; |
d36f2487 | 2524 | struct fuse_conn *fc = ff->fc; |
59efec7b TH |
2525 | struct fuse_ioctl_in inarg = { |
2526 | .fh = ff->fh, | |
2527 | .cmd = cmd, | |
2528 | .arg = arg, | |
2529 | .flags = flags | |
2530 | }; | |
2531 | struct fuse_ioctl_out outarg; | |
2532 | struct fuse_req *req = NULL; | |
2533 | struct page **pages = NULL; | |
8ac83505 | 2534 | struct iovec *iov_page = NULL; |
59efec7b TH |
2535 | struct iovec *in_iov = NULL, *out_iov = NULL; |
2536 | unsigned int in_iovs = 0, out_iovs = 0, num_pages = 0, max_pages; | |
acbe5fda MS |
2537 | size_t in_size, out_size, transferred, c; |
2538 | int err, i; | |
2539 | struct iov_iter ii; | |
59efec7b | 2540 | |
1baa26b2 MS |
2541 | #if BITS_PER_LONG == 32 |
2542 | inarg.flags |= FUSE_IOCTL_32BIT; | |
2543 | #else | |
2544 | if (flags & FUSE_IOCTL_COMPAT) | |
2545 | inarg.flags |= FUSE_IOCTL_32BIT; | |
2546 | #endif | |
2547 | ||
59efec7b | 2548 | /* assume all the iovs returned by client always fits in a page */ |
1baa26b2 | 2549 | BUILD_BUG_ON(sizeof(struct fuse_ioctl_iovec) * FUSE_IOCTL_MAX_IOV > PAGE_SIZE); |
59efec7b | 2550 | |
59efec7b | 2551 | err = -ENOMEM; |
5da784cc | 2552 | pages = kcalloc(fc->max_pages, sizeof(pages[0]), GFP_KERNEL); |
8ac83505 | 2553 | iov_page = (struct iovec *) __get_free_page(GFP_KERNEL); |
59efec7b TH |
2554 | if (!pages || !iov_page) |
2555 | goto out; | |
2556 | ||
2557 | /* | |
2558 | * If restricted, initialize IO parameters as encoded in @cmd. | |
2559 | * RETRY from server is not allowed. | |
2560 | */ | |
2561 | if (!(flags & FUSE_IOCTL_UNRESTRICTED)) { | |
8ac83505 | 2562 | struct iovec *iov = iov_page; |
59efec7b | 2563 | |
c9f0523d | 2564 | iov->iov_base = (void __user *)arg; |
59efec7b TH |
2565 | iov->iov_len = _IOC_SIZE(cmd); |
2566 | ||
2567 | if (_IOC_DIR(cmd) & _IOC_WRITE) { | |
2568 | in_iov = iov; | |
2569 | in_iovs = 1; | |
2570 | } | |
2571 | ||
2572 | if (_IOC_DIR(cmd) & _IOC_READ) { | |
2573 | out_iov = iov; | |
2574 | out_iovs = 1; | |
2575 | } | |
2576 | } | |
2577 | ||
2578 | retry: | |
2579 | inarg.in_size = in_size = iov_length(in_iov, in_iovs); | |
2580 | inarg.out_size = out_size = iov_length(out_iov, out_iovs); | |
2581 | ||
2582 | /* | |
2583 | * Out data can be used either for actual out data or iovs, | |
2584 | * make sure there always is at least one page. | |
2585 | */ | |
2586 | out_size = max_t(size_t, out_size, PAGE_SIZE); | |
2587 | max_pages = DIV_ROUND_UP(max(in_size, out_size), PAGE_SIZE); | |
2588 | ||
2589 | /* make sure there are enough buffer pages and init request with them */ | |
2590 | err = -ENOMEM; | |
5da784cc | 2591 | if (max_pages > fc->max_pages) |
59efec7b TH |
2592 | goto out; |
2593 | while (num_pages < max_pages) { | |
2594 | pages[num_pages] = alloc_page(GFP_KERNEL | __GFP_HIGHMEM); | |
2595 | if (!pages[num_pages]) | |
2596 | goto out; | |
2597 | num_pages++; | |
2598 | } | |
2599 | ||
54b96670 | 2600 | req = fuse_get_req(fc, num_pages); |
59efec7b TH |
2601 | if (IS_ERR(req)) { |
2602 | err = PTR_ERR(req); | |
2603 | req = NULL; | |
2604 | goto out; | |
2605 | } | |
2606 | memcpy(req->pages, pages, sizeof(req->pages[0]) * num_pages); | |
2607 | req->num_pages = num_pages; | |
7c190c8b | 2608 | fuse_page_descs_length_init(req, 0, req->num_pages); |
59efec7b TH |
2609 | |
2610 | /* okay, let's send it to the client */ | |
2611 | req->in.h.opcode = FUSE_IOCTL; | |
d36f2487 | 2612 | req->in.h.nodeid = ff->nodeid; |
59efec7b TH |
2613 | req->in.numargs = 1; |
2614 | req->in.args[0].size = sizeof(inarg); | |
2615 | req->in.args[0].value = &inarg; | |
2616 | if (in_size) { | |
2617 | req->in.numargs++; | |
2618 | req->in.args[1].size = in_size; | |
2619 | req->in.argpages = 1; | |
2620 | ||
acbe5fda MS |
2621 | err = -EFAULT; |
2622 | iov_iter_init(&ii, WRITE, in_iov, in_iovs, in_size); | |
2623 | for (i = 0; iov_iter_count(&ii) && !WARN_ON(i >= num_pages); i++) { | |
2624 | c = copy_page_from_iter(pages[i], 0, PAGE_SIZE, &ii); | |
2625 | if (c != PAGE_SIZE && iov_iter_count(&ii)) | |
2626 | goto out; | |
2627 | } | |
59efec7b TH |
2628 | } |
2629 | ||
2630 | req->out.numargs = 2; | |
2631 | req->out.args[0].size = sizeof(outarg); | |
2632 | req->out.args[0].value = &outarg; | |
2633 | req->out.args[1].size = out_size; | |
2634 | req->out.argpages = 1; | |
2635 | req->out.argvar = 1; | |
2636 | ||
b93f858a | 2637 | fuse_request_send(fc, req); |
59efec7b TH |
2638 | err = req->out.h.error; |
2639 | transferred = req->out.args[1].size; | |
2640 | fuse_put_request(fc, req); | |
2641 | req = NULL; | |
2642 | if (err) | |
2643 | goto out; | |
2644 | ||
2645 | /* did it ask for retry? */ | |
2646 | if (outarg.flags & FUSE_IOCTL_RETRY) { | |
8ac83505 | 2647 | void *vaddr; |
59efec7b TH |
2648 | |
2649 | /* no retry if in restricted mode */ | |
2650 | err = -EIO; | |
2651 | if (!(flags & FUSE_IOCTL_UNRESTRICTED)) | |
2652 | goto out; | |
2653 | ||
2654 | in_iovs = outarg.in_iovs; | |
2655 | out_iovs = outarg.out_iovs; | |
2656 | ||
2657 | /* | |
2658 | * Make sure things are in boundary, separate checks | |
2659 | * are to protect against overflow. | |
2660 | */ | |
2661 | err = -ENOMEM; | |
2662 | if (in_iovs > FUSE_IOCTL_MAX_IOV || | |
2663 | out_iovs > FUSE_IOCTL_MAX_IOV || | |
2664 | in_iovs + out_iovs > FUSE_IOCTL_MAX_IOV) | |
2665 | goto out; | |
2666 | ||
2408f6ef | 2667 | vaddr = kmap_atomic(pages[0]); |
1baa26b2 | 2668 | err = fuse_copy_ioctl_iovec(fc, iov_page, vaddr, |
d9d318d3 MS |
2669 | transferred, in_iovs + out_iovs, |
2670 | (flags & FUSE_IOCTL_COMPAT) != 0); | |
2408f6ef | 2671 | kunmap_atomic(vaddr); |
d9d318d3 MS |
2672 | if (err) |
2673 | goto out; | |
59efec7b | 2674 | |
8ac83505 | 2675 | in_iov = iov_page; |
59efec7b TH |
2676 | out_iov = in_iov + in_iovs; |
2677 | ||
5da784cc | 2678 | err = fuse_verify_ioctl_iov(fc, in_iov, in_iovs); |
7572777e MS |
2679 | if (err) |
2680 | goto out; | |
2681 | ||
5da784cc | 2682 | err = fuse_verify_ioctl_iov(fc, out_iov, out_iovs); |
7572777e MS |
2683 | if (err) |
2684 | goto out; | |
2685 | ||
59efec7b TH |
2686 | goto retry; |
2687 | } | |
2688 | ||
2689 | err = -EIO; | |
2690 | if (transferred > inarg.out_size) | |
2691 | goto out; | |
2692 | ||
acbe5fda MS |
2693 | err = -EFAULT; |
2694 | iov_iter_init(&ii, READ, out_iov, out_iovs, transferred); | |
2695 | for (i = 0; iov_iter_count(&ii) && !WARN_ON(i >= num_pages); i++) { | |
2696 | c = copy_page_to_iter(pages[i], 0, PAGE_SIZE, &ii); | |
2697 | if (c != PAGE_SIZE && iov_iter_count(&ii)) | |
2698 | goto out; | |
2699 | } | |
2700 | err = 0; | |
59efec7b TH |
2701 | out: |
2702 | if (req) | |
2703 | fuse_put_request(fc, req); | |
8ac83505 | 2704 | free_page((unsigned long) iov_page); |
59efec7b TH |
2705 | while (num_pages) |
2706 | __free_page(pages[--num_pages]); | |
2707 | kfree(pages); | |
2708 | ||
2709 | return err ? err : outarg.result; | |
2710 | } | |
08cbf542 | 2711 | EXPORT_SYMBOL_GPL(fuse_do_ioctl); |
59efec7b | 2712 | |
b18da0c5 MS |
2713 | long fuse_ioctl_common(struct file *file, unsigned int cmd, |
2714 | unsigned long arg, unsigned int flags) | |
d36f2487 | 2715 | { |
6131ffaa | 2716 | struct inode *inode = file_inode(file); |
d36f2487 MS |
2717 | struct fuse_conn *fc = get_fuse_conn(inode); |
2718 | ||
c2132c1b | 2719 | if (!fuse_allow_current_process(fc)) |
d36f2487 MS |
2720 | return -EACCES; |
2721 | ||
2722 | if (is_bad_inode(inode)) | |
2723 | return -EIO; | |
2724 | ||
2725 | return fuse_do_ioctl(file, cmd, arg, flags); | |
2726 | } | |
2727 | ||
59efec7b TH |
2728 | static long fuse_file_ioctl(struct file *file, unsigned int cmd, |
2729 | unsigned long arg) | |
2730 | { | |
b18da0c5 | 2731 | return fuse_ioctl_common(file, cmd, arg, 0); |
59efec7b TH |
2732 | } |
2733 | ||
2734 | static long fuse_file_compat_ioctl(struct file *file, unsigned int cmd, | |
2735 | unsigned long arg) | |
2736 | { | |
b18da0c5 | 2737 | return fuse_ioctl_common(file, cmd, arg, FUSE_IOCTL_COMPAT); |
59efec7b TH |
2738 | } |
2739 | ||
95668a69 TH |
2740 | /* |
2741 | * All files which have been polled are linked to RB tree | |
2742 | * fuse_conn->polled_files which is indexed by kh. Walk the tree and | |
2743 | * find the matching one. | |
2744 | */ | |
2745 | static struct rb_node **fuse_find_polled_node(struct fuse_conn *fc, u64 kh, | |
2746 | struct rb_node **parent_out) | |
2747 | { | |
2748 | struct rb_node **link = &fc->polled_files.rb_node; | |
2749 | struct rb_node *last = NULL; | |
2750 | ||
2751 | while (*link) { | |
2752 | struct fuse_file *ff; | |
2753 | ||
2754 | last = *link; | |
2755 | ff = rb_entry(last, struct fuse_file, polled_node); | |
2756 | ||
2757 | if (kh < ff->kh) | |
2758 | link = &last->rb_left; | |
2759 | else if (kh > ff->kh) | |
2760 | link = &last->rb_right; | |
2761 | else | |
2762 | return link; | |
2763 | } | |
2764 | ||
2765 | if (parent_out) | |
2766 | *parent_out = last; | |
2767 | return link; | |
2768 | } | |
2769 | ||
2770 | /* | |
2771 | * The file is about to be polled. Make sure it's on the polled_files | |
2772 | * RB tree. Note that files once added to the polled_files tree are | |
2773 | * not removed before the file is released. This is because a file | |
2774 | * polled once is likely to be polled again. | |
2775 | */ | |
2776 | static void fuse_register_polled_file(struct fuse_conn *fc, | |
2777 | struct fuse_file *ff) | |
2778 | { | |
2779 | spin_lock(&fc->lock); | |
2780 | if (RB_EMPTY_NODE(&ff->polled_node)) { | |
f3846266 | 2781 | struct rb_node **link, *uninitialized_var(parent); |
95668a69 TH |
2782 | |
2783 | link = fuse_find_polled_node(fc, ff->kh, &parent); | |
2784 | BUG_ON(*link); | |
2785 | rb_link_node(&ff->polled_node, parent, link); | |
2786 | rb_insert_color(&ff->polled_node, &fc->polled_files); | |
2787 | } | |
2788 | spin_unlock(&fc->lock); | |
2789 | } | |
2790 | ||
076ccb76 | 2791 | __poll_t fuse_file_poll(struct file *file, poll_table *wait) |
95668a69 | 2792 | { |
95668a69 | 2793 | struct fuse_file *ff = file->private_data; |
797759aa | 2794 | struct fuse_conn *fc = ff->fc; |
95668a69 TH |
2795 | struct fuse_poll_in inarg = { .fh = ff->fh, .kh = ff->kh }; |
2796 | struct fuse_poll_out outarg; | |
7078187a | 2797 | FUSE_ARGS(args); |
95668a69 TH |
2798 | int err; |
2799 | ||
2800 | if (fc->no_poll) | |
2801 | return DEFAULT_POLLMASK; | |
2802 | ||
2803 | poll_wait(file, &ff->poll_wait, wait); | |
c71d227f | 2804 | inarg.events = mangle_poll(poll_requested_events(wait)); |
95668a69 TH |
2805 | |
2806 | /* | |
2807 | * Ask for notification iff there's someone waiting for it. | |
2808 | * The client may ignore the flag and always notify. | |
2809 | */ | |
2810 | if (waitqueue_active(&ff->poll_wait)) { | |
2811 | inarg.flags |= FUSE_POLL_SCHEDULE_NOTIFY; | |
2812 | fuse_register_polled_file(fc, ff); | |
2813 | } | |
2814 | ||
7078187a MS |
2815 | args.in.h.opcode = FUSE_POLL; |
2816 | args.in.h.nodeid = ff->nodeid; | |
2817 | args.in.numargs = 1; | |
2818 | args.in.args[0].size = sizeof(inarg); | |
2819 | args.in.args[0].value = &inarg; | |
2820 | args.out.numargs = 1; | |
2821 | args.out.args[0].size = sizeof(outarg); | |
2822 | args.out.args[0].value = &outarg; | |
2823 | err = fuse_simple_request(fc, &args); | |
95668a69 TH |
2824 | |
2825 | if (!err) | |
c71d227f | 2826 | return demangle_poll(outarg.revents); |
95668a69 TH |
2827 | if (err == -ENOSYS) { |
2828 | fc->no_poll = 1; | |
2829 | return DEFAULT_POLLMASK; | |
2830 | } | |
a9a08845 | 2831 | return EPOLLERR; |
95668a69 | 2832 | } |
08cbf542 | 2833 | EXPORT_SYMBOL_GPL(fuse_file_poll); |
95668a69 TH |
2834 | |
2835 | /* | |
2836 | * This is called from fuse_handle_notify() on FUSE_NOTIFY_POLL and | |
2837 | * wakes up the poll waiters. | |
2838 | */ | |
2839 | int fuse_notify_poll_wakeup(struct fuse_conn *fc, | |
2840 | struct fuse_notify_poll_wakeup_out *outarg) | |
2841 | { | |
2842 | u64 kh = outarg->kh; | |
2843 | struct rb_node **link; | |
2844 | ||
2845 | spin_lock(&fc->lock); | |
2846 | ||
2847 | link = fuse_find_polled_node(fc, kh, NULL); | |
2848 | if (*link) { | |
2849 | struct fuse_file *ff; | |
2850 | ||
2851 | ff = rb_entry(*link, struct fuse_file, polled_node); | |
2852 | wake_up_interruptible_sync(&ff->poll_wait); | |
2853 | } | |
2854 | ||
2855 | spin_unlock(&fc->lock); | |
2856 | return 0; | |
2857 | } | |
2858 | ||
efb9fa9e MP |
2859 | static void fuse_do_truncate(struct file *file) |
2860 | { | |
2861 | struct inode *inode = file->f_mapping->host; | |
2862 | struct iattr attr; | |
2863 | ||
2864 | attr.ia_valid = ATTR_SIZE; | |
2865 | attr.ia_size = i_size_read(inode); | |
2866 | ||
2867 | attr.ia_file = file; | |
2868 | attr.ia_valid |= ATTR_FILE; | |
2869 | ||
62490330 | 2870 | fuse_do_setattr(file_dentry(file), &attr, file); |
efb9fa9e MP |
2871 | } |
2872 | ||
5da784cc | 2873 | static inline loff_t fuse_round_up(struct fuse_conn *fc, loff_t off) |
e5c5f05d | 2874 | { |
5da784cc | 2875 | return round_up(off, fc->max_pages << PAGE_SHIFT); |
e5c5f05d MP |
2876 | } |
2877 | ||
4273b793 | 2878 | static ssize_t |
c8b8e32d | 2879 | fuse_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
4273b793 | 2880 | { |
9d5722b7 | 2881 | DECLARE_COMPLETION_ONSTACK(wait); |
4273b793 | 2882 | ssize_t ret = 0; |
60b9df7a MS |
2883 | struct file *file = iocb->ki_filp; |
2884 | struct fuse_file *ff = file->private_data; | |
e5c5f05d | 2885 | bool async_dio = ff->fc->async_dio; |
4273b793 | 2886 | loff_t pos = 0; |
bcba24cc MP |
2887 | struct inode *inode; |
2888 | loff_t i_size; | |
a6cbcd4a | 2889 | size_t count = iov_iter_count(iter); |
c8b8e32d | 2890 | loff_t offset = iocb->ki_pos; |
36cf66ed | 2891 | struct fuse_io_priv *io; |
4273b793 | 2892 | |
4273b793 | 2893 | pos = offset; |
bcba24cc MP |
2894 | inode = file->f_mapping->host; |
2895 | i_size = i_size_read(inode); | |
4273b793 | 2896 | |
6f673763 | 2897 | if ((iov_iter_rw(iter) == READ) && (offset > i_size)) |
9fe55eea SW |
2898 | return 0; |
2899 | ||
439ee5f0 | 2900 | /* optimization for short read */ |
6f673763 | 2901 | if (async_dio && iov_iter_rw(iter) != WRITE && offset + count > i_size) { |
439ee5f0 MP |
2902 | if (offset >= i_size) |
2903 | return 0; | |
5da784cc | 2904 | iov_iter_truncate(iter, fuse_round_up(ff->fc, i_size - offset)); |
6b775b18 | 2905 | count = iov_iter_count(iter); |
439ee5f0 MP |
2906 | } |
2907 | ||
bcba24cc | 2908 | io = kmalloc(sizeof(struct fuse_io_priv), GFP_KERNEL); |
36cf66ed MP |
2909 | if (!io) |
2910 | return -ENOMEM; | |
bcba24cc | 2911 | spin_lock_init(&io->lock); |
744742d6 | 2912 | kref_init(&io->refcnt); |
bcba24cc MP |
2913 | io->reqs = 1; |
2914 | io->bytes = -1; | |
2915 | io->size = 0; | |
2916 | io->offset = offset; | |
6f673763 | 2917 | io->write = (iov_iter_rw(iter) == WRITE); |
bcba24cc | 2918 | io->err = 0; |
bcba24cc MP |
2919 | /* |
2920 | * By default, we want to optimize all I/Os with async request | |
60b9df7a | 2921 | * submission to the client filesystem if supported. |
bcba24cc | 2922 | */ |
e5c5f05d | 2923 | io->async = async_dio; |
bcba24cc | 2924 | io->iocb = iocb; |
7879c4e5 | 2925 | io->blocking = is_sync_kiocb(iocb); |
bcba24cc MP |
2926 | |
2927 | /* | |
7879c4e5 AS |
2928 | * We cannot asynchronously extend the size of a file. |
2929 | * In such case the aio will behave exactly like sync io. | |
bcba24cc | 2930 | */ |
7879c4e5 AS |
2931 | if ((offset + count > i_size) && iov_iter_rw(iter) == WRITE) |
2932 | io->blocking = true; | |
4273b793 | 2933 | |
7879c4e5 | 2934 | if (io->async && io->blocking) { |
744742d6 SF |
2935 | /* |
2936 | * Additional reference to keep io around after | |
2937 | * calling fuse_aio_complete() | |
2938 | */ | |
2939 | kref_get(&io->refcnt); | |
9d5722b7 | 2940 | io->done = &wait; |
744742d6 | 2941 | } |
9d5722b7 | 2942 | |
6f673763 | 2943 | if (iov_iter_rw(iter) == WRITE) { |
6b775b18 | 2944 | ret = fuse_direct_io(io, iter, &pos, FUSE_DIO_WRITE); |
812408fb AV |
2945 | fuse_invalidate_attr(inode); |
2946 | } else { | |
d22a943f | 2947 | ret = __fuse_direct_read(io, iter, &pos); |
812408fb | 2948 | } |
36cf66ed | 2949 | |
bcba24cc | 2950 | if (io->async) { |
ebacb812 LC |
2951 | bool blocking = io->blocking; |
2952 | ||
bcba24cc MP |
2953 | fuse_aio_complete(io, ret < 0 ? ret : 0, -1); |
2954 | ||
2955 | /* we have a non-extending, async request, so return */ | |
ebacb812 | 2956 | if (!blocking) |
bcba24cc MP |
2957 | return -EIOCBQUEUED; |
2958 | ||
9d5722b7 CH |
2959 | wait_for_completion(&wait); |
2960 | ret = fuse_get_res_by_io(io); | |
bcba24cc MP |
2961 | } |
2962 | ||
744742d6 | 2963 | kref_put(&io->refcnt, fuse_io_release); |
9d5722b7 | 2964 | |
6f673763 | 2965 | if (iov_iter_rw(iter) == WRITE) { |
efb9fa9e MP |
2966 | if (ret > 0) |
2967 | fuse_write_update_size(inode, pos); | |
2968 | else if (ret < 0 && offset + count > i_size) | |
2969 | fuse_do_truncate(file); | |
2970 | } | |
4273b793 AA |
2971 | |
2972 | return ret; | |
2973 | } | |
2974 | ||
cdadb11c MS |
2975 | static long fuse_file_fallocate(struct file *file, int mode, loff_t offset, |
2976 | loff_t length) | |
05ba1f08 AP |
2977 | { |
2978 | struct fuse_file *ff = file->private_data; | |
1c68271c | 2979 | struct inode *inode = file_inode(file); |
0ab08f57 | 2980 | struct fuse_inode *fi = get_fuse_inode(inode); |
05ba1f08 | 2981 | struct fuse_conn *fc = ff->fc; |
7078187a | 2982 | FUSE_ARGS(args); |
05ba1f08 AP |
2983 | struct fuse_fallocate_in inarg = { |
2984 | .fh = ff->fh, | |
2985 | .offset = offset, | |
2986 | .length = length, | |
2987 | .mode = mode | |
2988 | }; | |
2989 | int err; | |
14c14414 MP |
2990 | bool lock_inode = !(mode & FALLOC_FL_KEEP_SIZE) || |
2991 | (mode & FALLOC_FL_PUNCH_HOLE); | |
05ba1f08 | 2992 | |
4adb8302 MS |
2993 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) |
2994 | return -EOPNOTSUPP; | |
2995 | ||
519c6040 MS |
2996 | if (fc->no_fallocate) |
2997 | return -EOPNOTSUPP; | |
2998 | ||
14c14414 | 2999 | if (lock_inode) { |
5955102c | 3000 | inode_lock(inode); |
bde52788 MP |
3001 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
3002 | loff_t endbyte = offset + length - 1; | |
3003 | err = filemap_write_and_wait_range(inode->i_mapping, | |
3004 | offset, endbyte); | |
3005 | if (err) | |
3006 | goto out; | |
3007 | ||
3008 | fuse_sync_writes(inode); | |
3009 | } | |
3634a632 BF |
3010 | } |
3011 | ||
0ab08f57 MP |
3012 | if (!(mode & FALLOC_FL_KEEP_SIZE)) |
3013 | set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); | |
3014 | ||
7078187a MS |
3015 | args.in.h.opcode = FUSE_FALLOCATE; |
3016 | args.in.h.nodeid = ff->nodeid; | |
3017 | args.in.numargs = 1; | |
3018 | args.in.args[0].size = sizeof(inarg); | |
3019 | args.in.args[0].value = &inarg; | |
3020 | err = fuse_simple_request(fc, &args); | |
519c6040 MS |
3021 | if (err == -ENOSYS) { |
3022 | fc->no_fallocate = 1; | |
3023 | err = -EOPNOTSUPP; | |
3024 | } | |
bee6c307 BF |
3025 | if (err) |
3026 | goto out; | |
3027 | ||
3028 | /* we could have extended the file */ | |
b0aa7606 MP |
3029 | if (!(mode & FALLOC_FL_KEEP_SIZE)) { |
3030 | bool changed = fuse_write_update_size(inode, offset + length); | |
3031 | ||
93d2269d MS |
3032 | if (changed && fc->writeback_cache) |
3033 | file_update_time(file); | |
b0aa7606 | 3034 | } |
bee6c307 BF |
3035 | |
3036 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
3037 | truncate_pagecache_range(inode, offset, offset + length - 1); | |
3038 | ||
3039 | fuse_invalidate_attr(inode); | |
3040 | ||
3634a632 | 3041 | out: |
0ab08f57 MP |
3042 | if (!(mode & FALLOC_FL_KEEP_SIZE)) |
3043 | clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); | |
3044 | ||
bde52788 | 3045 | if (lock_inode) |
5955102c | 3046 | inode_unlock(inode); |
3634a632 | 3047 | |
05ba1f08 AP |
3048 | return err; |
3049 | } | |
05ba1f08 | 3050 | |
88bc7d50 NV |
3051 | static ssize_t fuse_copy_file_range(struct file *file_in, loff_t pos_in, |
3052 | struct file *file_out, loff_t pos_out, | |
3053 | size_t len, unsigned int flags) | |
3054 | { | |
3055 | struct fuse_file *ff_in = file_in->private_data; | |
3056 | struct fuse_file *ff_out = file_out->private_data; | |
3057 | struct inode *inode_out = file_inode(file_out); | |
3058 | struct fuse_inode *fi_out = get_fuse_inode(inode_out); | |
3059 | struct fuse_conn *fc = ff_in->fc; | |
3060 | FUSE_ARGS(args); | |
3061 | struct fuse_copy_file_range_in inarg = { | |
3062 | .fh_in = ff_in->fh, | |
3063 | .off_in = pos_in, | |
3064 | .nodeid_out = ff_out->nodeid, | |
3065 | .fh_out = ff_out->fh, | |
3066 | .off_out = pos_out, | |
3067 | .len = len, | |
3068 | .flags = flags | |
3069 | }; | |
3070 | struct fuse_write_out outarg; | |
3071 | ssize_t err; | |
3072 | /* mark unstable when write-back is not used, and file_out gets | |
3073 | * extended */ | |
3074 | bool is_unstable = (!fc->writeback_cache) && | |
3075 | ((pos_out + len) > inode_out->i_size); | |
3076 | ||
3077 | if (fc->no_copy_file_range) | |
3078 | return -EOPNOTSUPP; | |
3079 | ||
3080 | inode_lock(inode_out); | |
3081 | ||
3082 | if (fc->writeback_cache) { | |
3083 | err = filemap_write_and_wait_range(inode_out->i_mapping, | |
3084 | pos_out, pos_out + len); | |
3085 | if (err) | |
3086 | goto out; | |
3087 | ||
3088 | fuse_sync_writes(inode_out); | |
3089 | } | |
3090 | ||
3091 | if (is_unstable) | |
3092 | set_bit(FUSE_I_SIZE_UNSTABLE, &fi_out->state); | |
3093 | ||
3094 | args.in.h.opcode = FUSE_COPY_FILE_RANGE; | |
3095 | args.in.h.nodeid = ff_in->nodeid; | |
3096 | args.in.numargs = 1; | |
3097 | args.in.args[0].size = sizeof(inarg); | |
3098 | args.in.args[0].value = &inarg; | |
3099 | args.out.numargs = 1; | |
3100 | args.out.args[0].size = sizeof(outarg); | |
3101 | args.out.args[0].value = &outarg; | |
3102 | err = fuse_simple_request(fc, &args); | |
3103 | if (err == -ENOSYS) { | |
3104 | fc->no_copy_file_range = 1; | |
3105 | err = -EOPNOTSUPP; | |
3106 | } | |
3107 | if (err) | |
3108 | goto out; | |
3109 | ||
3110 | if (fc->writeback_cache) { | |
3111 | fuse_write_update_size(inode_out, pos_out + outarg.size); | |
3112 | file_update_time(file_out); | |
3113 | } | |
3114 | ||
3115 | fuse_invalidate_attr(inode_out); | |
3116 | ||
3117 | err = outarg.size; | |
3118 | out: | |
3119 | if (is_unstable) | |
3120 | clear_bit(FUSE_I_SIZE_UNSTABLE, &fi_out->state); | |
3121 | ||
3122 | inode_unlock(inode_out); | |
3123 | ||
3124 | return err; | |
3125 | } | |
3126 | ||
4b6f5d20 | 3127 | static const struct file_operations fuse_file_operations = { |
5559b8f4 | 3128 | .llseek = fuse_file_llseek, |
37c20f16 | 3129 | .read_iter = fuse_file_read_iter, |
84c3d55c | 3130 | .write_iter = fuse_file_write_iter, |
b6aeaded MS |
3131 | .mmap = fuse_file_mmap, |
3132 | .open = fuse_open, | |
3133 | .flush = fuse_flush, | |
3134 | .release = fuse_release, | |
3135 | .fsync = fuse_fsync, | |
71421259 | 3136 | .lock = fuse_file_lock, |
a9ff4f87 | 3137 | .flock = fuse_file_flock, |
5ffc4ef4 | 3138 | .splice_read = generic_file_splice_read, |
59efec7b TH |
3139 | .unlocked_ioctl = fuse_file_ioctl, |
3140 | .compat_ioctl = fuse_file_compat_ioctl, | |
95668a69 | 3141 | .poll = fuse_file_poll, |
05ba1f08 | 3142 | .fallocate = fuse_file_fallocate, |
88bc7d50 | 3143 | .copy_file_range = fuse_copy_file_range, |
b6aeaded MS |
3144 | }; |
3145 | ||
4b6f5d20 | 3146 | static const struct file_operations fuse_direct_io_file_operations = { |
5559b8f4 | 3147 | .llseek = fuse_file_llseek, |
15316263 | 3148 | .read_iter = fuse_direct_read_iter, |
15316263 | 3149 | .write_iter = fuse_direct_write_iter, |
fc280c96 | 3150 | .mmap = fuse_direct_mmap, |
413ef8cb MS |
3151 | .open = fuse_open, |
3152 | .flush = fuse_flush, | |
3153 | .release = fuse_release, | |
3154 | .fsync = fuse_fsync, | |
71421259 | 3155 | .lock = fuse_file_lock, |
a9ff4f87 | 3156 | .flock = fuse_file_flock, |
59efec7b TH |
3157 | .unlocked_ioctl = fuse_file_ioctl, |
3158 | .compat_ioctl = fuse_file_compat_ioctl, | |
95668a69 | 3159 | .poll = fuse_file_poll, |
05ba1f08 | 3160 | .fallocate = fuse_file_fallocate, |
fc280c96 | 3161 | /* no splice_read */ |
413ef8cb MS |
3162 | }; |
3163 | ||
f5e54d6e | 3164 | static const struct address_space_operations fuse_file_aops = { |
b6aeaded | 3165 | .readpage = fuse_readpage, |
3be5a52b | 3166 | .writepage = fuse_writepage, |
26d614df | 3167 | .writepages = fuse_writepages, |
3be5a52b | 3168 | .launder_page = fuse_launder_page, |
db50b96c | 3169 | .readpages = fuse_readpages, |
3be5a52b | 3170 | .set_page_dirty = __set_page_dirty_nobuffers, |
b2d2272f | 3171 | .bmap = fuse_bmap, |
4273b793 | 3172 | .direct_IO = fuse_direct_IO, |
6b12c1b3 PE |
3173 | .write_begin = fuse_write_begin, |
3174 | .write_end = fuse_write_end, | |
b6aeaded MS |
3175 | }; |
3176 | ||
3177 | void fuse_init_file_inode(struct inode *inode) | |
3178 | { | |
ab2257e9 MS |
3179 | struct fuse_inode *fi = get_fuse_inode(inode); |
3180 | ||
45323fb7 MS |
3181 | inode->i_fop = &fuse_file_operations; |
3182 | inode->i_data.a_ops = &fuse_file_aops; | |
ab2257e9 MS |
3183 | |
3184 | INIT_LIST_HEAD(&fi->write_files); | |
3185 | INIT_LIST_HEAD(&fi->queued_writes); | |
3186 | fi->writectr = 0; | |
3187 | init_waitqueue_head(&fi->page_waitq); | |
3188 | INIT_LIST_HEAD(&fi->writepages); | |
b6aeaded | 3189 | } |