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> |
478e0841 | 17 | #include <linux/swap.h> |
3634a632 | 18 | #include <linux/falloc.h> |
e2e40f2c | 19 | #include <linux/uio.h> |
31070f6c | 20 | #include <linux/fs.h> |
5970e15d | 21 | #include <linux/filelock.h> |
705bcfcb | 22 | #include <linux/splice.h> |
2e3f7dd0 | 23 | #include <linux/task_io_accounting_ops.h> |
b6aeaded | 24 | |
b9d54c6f MS |
25 | static int fuse_send_open(struct fuse_mount *fm, u64 nodeid, |
26 | unsigned int open_flags, int opcode, | |
27 | struct fuse_open_out *outargp) | |
b6aeaded | 28 | { |
b6aeaded | 29 | struct fuse_open_in inarg; |
7078187a | 30 | FUSE_ARGS(args); |
fd72faac MS |
31 | |
32 | memset(&inarg, 0, sizeof(inarg)); | |
b9d54c6f | 33 | inarg.flags = open_flags & ~(O_CREAT | O_EXCL | O_NOCTTY); |
fcee216b | 34 | if (!fm->fc->atomic_o_trunc) |
6ff958ed | 35 | inarg.flags &= ~O_TRUNC; |
643a666a VG |
36 | |
37 | if (fm->fc->handle_killpriv_v2 && | |
38 | (inarg.flags & O_TRUNC) && !capable(CAP_FSETID)) { | |
39 | inarg.open_flags |= FUSE_OPEN_KILL_SUIDGID; | |
40 | } | |
41 | ||
d5b48543 MS |
42 | args.opcode = opcode; |
43 | args.nodeid = nodeid; | |
44 | args.in_numargs = 1; | |
45 | args.in_args[0].size = sizeof(inarg); | |
46 | args.in_args[0].value = &inarg; | |
47 | args.out_numargs = 1; | |
48 | args.out_args[0].size = sizeof(*outargp); | |
49 | args.out_args[0].value = outargp; | |
fd72faac | 50 | |
fcee216b | 51 | return fuse_simple_request(fm, &args); |
fd72faac MS |
52 | } |
53 | ||
e26ee4ef | 54 | struct fuse_file *fuse_file_alloc(struct fuse_mount *fm, bool release) |
fd72faac MS |
55 | { |
56 | struct fuse_file *ff; | |
6b2db28a | 57 | |
dc69e98c | 58 | ff = kzalloc(sizeof(struct fuse_file), GFP_KERNEL_ACCOUNT); |
6b2db28a TH |
59 | if (unlikely(!ff)) |
60 | return NULL; | |
61 | ||
fcee216b | 62 | ff->fm = fm; |
e26ee4ef | 63 | if (release) { |
fc8ff397 AG |
64 | ff->args = kzalloc(sizeof(*ff->args), GFP_KERNEL_ACCOUNT); |
65 | if (!ff->args) { | |
e26ee4ef AG |
66 | kfree(ff); |
67 | return NULL; | |
68 | } | |
fd72faac | 69 | } |
6b2db28a TH |
70 | |
71 | INIT_LIST_HEAD(&ff->write_entry); | |
5d7bc7e8 | 72 | mutex_init(&ff->readdir.lock); |
4e8c2eb5 | 73 | refcount_set(&ff->count, 1); |
6b2db28a TH |
74 | RB_CLEAR_NODE(&ff->polled_node); |
75 | init_waitqueue_head(&ff->poll_wait); | |
76 | ||
fcee216b | 77 | ff->kh = atomic64_inc_return(&fm->fc->khctr); |
6b2db28a | 78 | |
fd72faac MS |
79 | return ff; |
80 | } | |
81 | ||
82 | void fuse_file_free(struct fuse_file *ff) | |
83 | { | |
fc8ff397 | 84 | kfree(ff->args); |
5d7bc7e8 | 85 | mutex_destroy(&ff->readdir.lock); |
fd72faac MS |
86 | kfree(ff); |
87 | } | |
88 | ||
267d8444 | 89 | static struct fuse_file *fuse_file_get(struct fuse_file *ff) |
c756e0a4 | 90 | { |
4e8c2eb5 | 91 | refcount_inc(&ff->count); |
c756e0a4 MS |
92 | return ff; |
93 | } | |
94 | ||
fcee216b | 95 | static void fuse_release_end(struct fuse_mount *fm, struct fuse_args *args, |
4cb54866 | 96 | int error) |
819c4b3b | 97 | { |
4cb54866 MS |
98 | struct fuse_release_args *ra = container_of(args, typeof(*ra), args); |
99 | ||
100 | iput(ra->inode); | |
101 | kfree(ra); | |
819c4b3b MS |
102 | } |
103 | ||
e26ee4ef | 104 | static void fuse_file_put(struct fuse_file *ff, bool sync) |
c756e0a4 | 105 | { |
4e8c2eb5 | 106 | if (refcount_dec_and_test(&ff->count)) { |
fc8ff397 | 107 | struct fuse_release_args *ra = &ff->args->release_args; |
e26ee4ef | 108 | struct fuse_args *args = (ra ? &ra->args : NULL); |
8b0797a4 | 109 | |
cb098dd2 AG |
110 | if (ra && ra->inode) |
111 | fuse_file_io_release(ff, ra->inode); | |
112 | ||
e26ee4ef AG |
113 | if (!args) { |
114 | /* Do nothing when server does not implement 'open' */ | |
7678ac50 | 115 | } else if (sync) { |
fcee216b MR |
116 | fuse_simple_request(ff->fm, args); |
117 | fuse_release_end(ff->fm, args, 0); | |
5a18ec17 | 118 | } else { |
4cb54866 | 119 | args->end = fuse_release_end; |
fcee216b | 120 | if (fuse_simple_background(ff->fm, args, |
4cb54866 | 121 | GFP_KERNEL | __GFP_NOFAIL)) |
fcee216b | 122 | fuse_release_end(ff->fm, args, -ENOTCONN); |
5a18ec17 | 123 | } |
c756e0a4 MS |
124 | kfree(ff); |
125 | } | |
126 | } | |
127 | ||
b9d54c6f MS |
128 | struct fuse_file *fuse_file_open(struct fuse_mount *fm, u64 nodeid, |
129 | unsigned int open_flags, bool isdir) | |
91fe96b4 | 130 | { |
fcee216b | 131 | struct fuse_conn *fc = fm->fc; |
91fe96b4 | 132 | struct fuse_file *ff; |
91fe96b4 | 133 | int opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN; |
e26ee4ef | 134 | bool open = isdir ? !fc->no_opendir : !fc->no_open; |
91fe96b4 | 135 | |
e26ee4ef | 136 | ff = fuse_file_alloc(fm, open); |
91fe96b4 | 137 | if (!ff) |
b9d54c6f | 138 | return ERR_PTR(-ENOMEM); |
91fe96b4 | 139 | |
7678ac50 | 140 | ff->fh = 0; |
fabf7e02 CA |
141 | /* Default for no-open */ |
142 | ff->open_flags = FOPEN_KEEP_CACHE | (isdir ? FOPEN_CACHE_DIR : 0); | |
e26ee4ef | 143 | if (open) { |
fc8ff397 AG |
144 | /* Store outarg for fuse_finish_open() */ |
145 | struct fuse_open_out *outargp = &ff->args->open_outarg; | |
7678ac50 AG |
146 | int err; |
147 | ||
fc8ff397 | 148 | err = fuse_send_open(fm, nodeid, open_flags, opcode, outargp); |
7678ac50 | 149 | if (!err) { |
fc8ff397 AG |
150 | ff->fh = outargp->fh; |
151 | ff->open_flags = outargp->open_flags; | |
d9a9ea94 | 152 | } else if (err != -ENOSYS) { |
7678ac50 | 153 | fuse_file_free(ff); |
b9d54c6f | 154 | return ERR_PTR(err); |
7678ac50 | 155 | } else { |
e26ee4ef | 156 | /* No release needed */ |
fc8ff397 AG |
157 | kfree(ff->args); |
158 | ff->args = NULL; | |
d9a9ea94 CA |
159 | if (isdir) |
160 | fc->no_opendir = 1; | |
161 | else | |
162 | fc->no_open = 1; | |
7678ac50 | 163 | } |
91fe96b4 MS |
164 | } |
165 | ||
166 | if (isdir) | |
7678ac50 | 167 | ff->open_flags &= ~FOPEN_DIRECT_IO; |
91fe96b4 | 168 | |
91fe96b4 | 169 | ff->nodeid = nodeid; |
91fe96b4 | 170 | |
b9d54c6f MS |
171 | return ff; |
172 | } | |
173 | ||
174 | int fuse_do_open(struct fuse_mount *fm, u64 nodeid, struct file *file, | |
175 | bool isdir) | |
176 | { | |
177 | struct fuse_file *ff = fuse_file_open(fm, nodeid, file->f_flags, isdir); | |
178 | ||
179 | if (!IS_ERR(ff)) | |
180 | file->private_data = ff; | |
181 | ||
182 | return PTR_ERR_OR_ZERO(ff); | |
91fe96b4 | 183 | } |
08cbf542 | 184 | EXPORT_SYMBOL_GPL(fuse_do_open); |
91fe96b4 | 185 | |
650b22b9 PE |
186 | static void fuse_link_write_file(struct file *file) |
187 | { | |
188 | struct inode *inode = file_inode(file); | |
650b22b9 PE |
189 | struct fuse_inode *fi = get_fuse_inode(inode); |
190 | struct fuse_file *ff = file->private_data; | |
191 | /* | |
192 | * file may be written through mmap, so chain it onto the | |
193 | * inodes's write_file list | |
194 | */ | |
f15ecfef | 195 | spin_lock(&fi->lock); |
650b22b9 PE |
196 | if (list_empty(&ff->write_entry)) |
197 | list_add(&ff->write_entry, &fi->write_files); | |
f15ecfef | 198 | spin_unlock(&fi->lock); |
650b22b9 PE |
199 | } |
200 | ||
d2c487f1 | 201 | int fuse_finish_open(struct inode *inode, struct file *file) |
fd72faac | 202 | { |
c7b7143c | 203 | struct fuse_file *ff = file->private_data; |
a0822c55 | 204 | struct fuse_conn *fc = get_fuse_conn(inode); |
cb098dd2 AG |
205 | int err; |
206 | ||
207 | err = fuse_file_io_open(file, inode); | |
208 | if (err) | |
209 | return err; | |
c7b7143c | 210 | |
bbd84f33 KS |
211 | if (ff->open_flags & FOPEN_STREAM) |
212 | stream_open(inode, file); | |
213 | else if (ff->open_flags & FOPEN_NONSEEKABLE) | |
a7c1b990 | 214 | nonseekable_open(inode, file); |
76224355 | 215 | |
4d99ff8f PE |
216 | if ((file->f_mode & FMODE_WRITE) && fc->writeback_cache) |
217 | fuse_link_write_file(file); | |
d2c487f1 AG |
218 | |
219 | return 0; | |
fd72faac MS |
220 | } |
221 | ||
0c9d7089 AG |
222 | static void fuse_truncate_update_attr(struct inode *inode, struct file *file) |
223 | { | |
224 | struct fuse_conn *fc = get_fuse_conn(inode); | |
225 | struct fuse_inode *fi = get_fuse_inode(inode); | |
226 | ||
227 | spin_lock(&fi->lock); | |
228 | fi->attr_version = atomic64_inc_return(&fc->attr_version); | |
229 | i_size_write(inode, 0); | |
230 | spin_unlock(&fi->lock); | |
231 | file_update_time(file); | |
232 | fuse_invalidate_attr_mask(inode, FUSE_STATX_MODSIZE); | |
233 | } | |
234 | ||
7de64d52 | 235 | static int fuse_open(struct inode *inode, struct file *file) |
fd72faac | 236 | { |
fcee216b | 237 | struct fuse_mount *fm = get_fuse_mount(inode); |
d2c487f1 | 238 | struct fuse_inode *fi = get_fuse_inode(inode); |
fcee216b | 239 | struct fuse_conn *fc = fm->fc; |
d2c487f1 | 240 | struct fuse_file *ff; |
b6aeaded | 241 | int err; |
0c9d7089 AG |
242 | bool is_truncate = (file->f_flags & O_TRUNC) && fc->atomic_o_trunc; |
243 | bool is_wb_truncate = is_truncate && fc->writeback_cache; | |
244 | bool dax_truncate = is_truncate && FUSE_IS_DAX(inode); | |
b6aeaded | 245 | |
5d069dbe MS |
246 | if (fuse_is_bad(inode)) |
247 | return -EIO; | |
248 | ||
b6aeaded MS |
249 | err = generic_file_open(inode, file); |
250 | if (err) | |
251 | return err; | |
252 | ||
2fdbb8dd | 253 | if (is_wb_truncate || dax_truncate) |
5955102c | 254 | inode_lock(inode); |
75caeecd | 255 | |
6ae330ca | 256 | if (dax_truncate) { |
8bcbbe9c | 257 | filemap_invalidate_lock(inode->i_mapping); |
6ae330ca VG |
258 | err = fuse_dax_break_layouts(inode, 0, 0); |
259 | if (err) | |
2fdbb8dd | 260 | goto out_inode_unlock; |
6ae330ca | 261 | } |
b6aeaded | 262 | |
2fdbb8dd MS |
263 | if (is_wb_truncate || dax_truncate) |
264 | fuse_set_nowrite(inode); | |
265 | ||
7de64d52 | 266 | err = fuse_do_open(fm, get_node_id(inode), file, false); |
0c9d7089 | 267 | if (!err) { |
d2c487f1 AG |
268 | ff = file->private_data; |
269 | err = fuse_finish_open(inode, file); | |
270 | if (err) | |
271 | fuse_sync_release(fi, ff, file->f_flags); | |
272 | else if (is_truncate) | |
0c9d7089 AG |
273 | fuse_truncate_update_attr(inode, file); |
274 | } | |
91fe96b4 | 275 | |
2fdbb8dd MS |
276 | if (is_wb_truncate || dax_truncate) |
277 | fuse_release_nowrite(inode); | |
278 | if (!err) { | |
0c9d7089 | 279 | if (is_truncate) |
2fdbb8dd MS |
280 | truncate_pagecache(inode, 0); |
281 | else if (!(ff->open_flags & FOPEN_KEEP_CACHE)) | |
282 | invalidate_inode_pages2(inode->i_mapping); | |
283 | } | |
6ae330ca | 284 | if (dax_truncate) |
8bcbbe9c | 285 | filemap_invalidate_unlock(inode->i_mapping); |
2fdbb8dd MS |
286 | out_inode_unlock: |
287 | if (is_wb_truncate || dax_truncate) | |
5955102c | 288 | inode_unlock(inode); |
75caeecd MP |
289 | |
290 | return err; | |
b6aeaded MS |
291 | } |
292 | ||
ebf84d0c | 293 | static void fuse_prepare_release(struct fuse_inode *fi, struct fuse_file *ff, |
e26ee4ef | 294 | unsigned int flags, int opcode, bool sync) |
64c6d8ed | 295 | { |
fcee216b | 296 | struct fuse_conn *fc = ff->fm->fc; |
fc8ff397 | 297 | struct fuse_release_args *ra = &ff->args->release_args; |
b6aeaded | 298 | |
4a90451b AG |
299 | if (fuse_file_passthrough(ff)) |
300 | fuse_passthrough_release(ff, fuse_inode_backing(fi)); | |
301 | ||
f15ecfef KT |
302 | /* Inode is NULL on error path of fuse_create_open() */ |
303 | if (likely(fi)) { | |
304 | spin_lock(&fi->lock); | |
305 | list_del(&ff->write_entry); | |
306 | spin_unlock(&fi->lock); | |
307 | } | |
8b0797a4 | 308 | spin_lock(&fc->lock); |
8b0797a4 MS |
309 | if (!RB_EMPTY_NODE(&ff->polled_node)) |
310 | rb_erase(&ff->polled_node, &fc->polled_files); | |
311 | spin_unlock(&fc->lock); | |
312 | ||
357ccf2b | 313 | wake_up_interruptible_all(&ff->poll_wait); |
8b0797a4 | 314 | |
e26ee4ef AG |
315 | if (!ra) |
316 | return; | |
317 | ||
fc8ff397 AG |
318 | /* ff->args was used for open outarg */ |
319 | memset(ff->args, 0, sizeof(*ff->args)); | |
4cb54866 MS |
320 | ra->inarg.fh = ff->fh; |
321 | ra->inarg.flags = flags; | |
322 | ra->args.in_numargs = 1; | |
323 | ra->args.in_args[0].size = sizeof(struct fuse_release_in); | |
324 | ra->args.in_args[0].value = &ra->inarg; | |
325 | ra->args.opcode = opcode; | |
326 | ra->args.nodeid = ff->nodeid; | |
327 | ra->args.force = true; | |
328 | ra->args.nocreds = true; | |
e26ee4ef AG |
329 | |
330 | /* | |
331 | * Hold inode until release is finished. | |
332 | * From fuse_sync_release() the refcount is 1 and everything's | |
333 | * synchronous, so we are fine with not doing igrab() here. | |
334 | */ | |
335 | ra->inode = sync ? NULL : igrab(&fi->inode); | |
fd72faac MS |
336 | } |
337 | ||
b9d54c6f MS |
338 | void fuse_file_release(struct inode *inode, struct fuse_file *ff, |
339 | unsigned int open_flags, fl_owner_t id, bool isdir) | |
fd72faac | 340 | { |
b9d54c6f | 341 | struct fuse_inode *fi = get_fuse_inode(inode); |
fc8ff397 | 342 | struct fuse_release_args *ra = &ff->args->release_args; |
2e64ff15 | 343 | int opcode = isdir ? FUSE_RELEASEDIR : FUSE_RELEASE; |
6b2db28a | 344 | |
e26ee4ef | 345 | fuse_prepare_release(fi, ff, open_flags, opcode, false); |
6b2db28a | 346 | |
e26ee4ef | 347 | if (ra && ff->flock) { |
4cb54866 | 348 | ra->inarg.release_flags |= FUSE_RELEASE_FLOCK_UNLOCK; |
b9d54c6f | 349 | ra->inarg.lock_owner = fuse_lock_owner_id(ff->fm->fc, id); |
37fb3a30 | 350 | } |
6b2db28a | 351 | |
6b2db28a TH |
352 | /* |
353 | * Normally this will send the RELEASE request, however if | |
354 | * some asynchronous READ or WRITE requests are outstanding, | |
355 | * the sending will be delayed. | |
5a18ec17 MS |
356 | * |
357 | * Make the release synchronous if this is a fuseblk mount, | |
358 | * synchronous RELEASE is allowed (and desirable) in this case | |
359 | * because the server can be trusted not to screw up. | |
6b2db28a | 360 | */ |
e26ee4ef | 361 | fuse_file_put(ff, ff->fm->fc->destroy); |
b6aeaded MS |
362 | } |
363 | ||
b9d54c6f MS |
364 | void fuse_release_common(struct file *file, bool isdir) |
365 | { | |
366 | fuse_file_release(file_inode(file), file->private_data, file->f_flags, | |
367 | (fl_owner_t) file, isdir); | |
368 | } | |
369 | ||
04730fef MS |
370 | static int fuse_release(struct inode *inode, struct file *file) |
371 | { | |
035ff33c MS |
372 | struct fuse_conn *fc = get_fuse_conn(inode); |
373 | ||
374 | /* | |
375 | * Dirty pages might remain despite write_inode_now() call from | |
376 | * fuse_flush() due to writes racing with the close. | |
377 | */ | |
378 | if (fc->writeback_cache) | |
379 | write_inode_now(inode, 1); | |
380 | ||
2e64ff15 | 381 | fuse_release_common(file, false); |
8b0797a4 MS |
382 | |
383 | /* return value is ignored by VFS */ | |
384 | return 0; | |
385 | } | |
386 | ||
54d601cb MS |
387 | void fuse_sync_release(struct fuse_inode *fi, struct fuse_file *ff, |
388 | unsigned int flags) | |
8b0797a4 | 389 | { |
4e8c2eb5 | 390 | WARN_ON(refcount_read(&ff->count) > 1); |
e26ee4ef AG |
391 | fuse_prepare_release(fi, ff, flags, FUSE_RELEASE, true); |
392 | fuse_file_put(ff, true); | |
04730fef | 393 | } |
08cbf542 | 394 | EXPORT_SYMBOL_GPL(fuse_sync_release); |
04730fef | 395 | |
71421259 | 396 | /* |
9c8ef561 MS |
397 | * Scramble the ID space with XTEA, so that the value of the files_struct |
398 | * pointer is not exposed to userspace. | |
71421259 | 399 | */ |
f3332114 | 400 | u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id) |
71421259 | 401 | { |
9c8ef561 MS |
402 | u32 *k = fc->scramble_key; |
403 | u64 v = (unsigned long) id; | |
404 | u32 v0 = v; | |
405 | u32 v1 = v >> 32; | |
406 | u32 sum = 0; | |
407 | int i; | |
408 | ||
409 | for (i = 0; i < 32; i++) { | |
410 | v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]); | |
411 | sum += 0x9E3779B9; | |
412 | v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]); | |
413 | } | |
414 | ||
415 | return (u64) v0 + ((u64) v1 << 32); | |
71421259 MS |
416 | } |
417 | ||
33826ebb MS |
418 | struct fuse_writepage_args { |
419 | struct fuse_io_args ia; | |
6b2fb799 | 420 | struct rb_node writepages_entry; |
33826ebb MS |
421 | struct list_head queue_entry; |
422 | struct fuse_writepage_args *next; | |
423 | struct inode *inode; | |
660585b5 | 424 | struct fuse_sync_bucket *bucket; |
33826ebb MS |
425 | }; |
426 | ||
427 | static struct fuse_writepage_args *fuse_find_writeback(struct fuse_inode *fi, | |
2fe93bd4 MS |
428 | pgoff_t idx_from, pgoff_t idx_to) |
429 | { | |
6b2fb799 MP |
430 | struct rb_node *n; |
431 | ||
432 | n = fi->writepages.rb_node; | |
2fe93bd4 | 433 | |
6b2fb799 MP |
434 | while (n) { |
435 | struct fuse_writepage_args *wpa; | |
2fe93bd4 MS |
436 | pgoff_t curr_index; |
437 | ||
6b2fb799 | 438 | wpa = rb_entry(n, struct fuse_writepage_args, writepages_entry); |
33826ebb MS |
439 | WARN_ON(get_fuse_inode(wpa->inode) != fi); |
440 | curr_index = wpa->ia.write.in.offset >> PAGE_SHIFT; | |
6b2fb799 MP |
441 | if (idx_from >= curr_index + wpa->ia.ap.num_pages) |
442 | n = n->rb_right; | |
443 | else if (idx_to < curr_index) | |
444 | n = n->rb_left; | |
445 | else | |
33826ebb | 446 | return wpa; |
2fe93bd4 MS |
447 | } |
448 | return NULL; | |
449 | } | |
450 | ||
3be5a52b | 451 | /* |
ea8cd333 | 452 | * Check if any page in a range is under writeback |
3be5a52b MS |
453 | * |
454 | * This is currently done by walking the list of writepage requests | |
455 | * for the inode, which can be pretty inefficient. | |
456 | */ | |
ea8cd333 PE |
457 | static bool fuse_range_is_writeback(struct inode *inode, pgoff_t idx_from, |
458 | pgoff_t idx_to) | |
3be5a52b | 459 | { |
3be5a52b | 460 | struct fuse_inode *fi = get_fuse_inode(inode); |
2fe93bd4 | 461 | bool found; |
3be5a52b | 462 | |
f15ecfef | 463 | spin_lock(&fi->lock); |
2fe93bd4 | 464 | found = fuse_find_writeback(fi, idx_from, idx_to); |
f15ecfef | 465 | spin_unlock(&fi->lock); |
3be5a52b MS |
466 | |
467 | return found; | |
468 | } | |
469 | ||
ea8cd333 PE |
470 | static inline bool fuse_page_is_writeback(struct inode *inode, pgoff_t index) |
471 | { | |
472 | return fuse_range_is_writeback(inode, index, index); | |
473 | } | |
474 | ||
3be5a52b MS |
475 | /* |
476 | * Wait for page writeback to be completed. | |
477 | * | |
478 | * Since fuse doesn't rely on the VM writeback tracking, this has to | |
479 | * use some other means. | |
480 | */ | |
17b2cbe2 | 481 | static void fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index) |
3be5a52b MS |
482 | { |
483 | struct fuse_inode *fi = get_fuse_inode(inode); | |
484 | ||
485 | wait_event(fi->page_waitq, !fuse_page_is_writeback(inode, index)); | |
3be5a52b MS |
486 | } |
487 | ||
fe38d7df MP |
488 | /* |
489 | * Wait for all pending writepages on the inode to finish. | |
490 | * | |
491 | * This is currently done by blocking further writes with FUSE_NOWRITE | |
492 | * and waiting for all sent writes to complete. | |
493 | * | |
494 | * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage | |
495 | * could conflict with truncation. | |
496 | */ | |
497 | static void fuse_sync_writes(struct inode *inode) | |
498 | { | |
499 | fuse_set_nowrite(inode); | |
500 | fuse_release_nowrite(inode); | |
501 | } | |
502 | ||
91ec6c85 | 503 | static int fuse_flush(struct file *file, fl_owner_t id) |
5a8bee63 | 504 | { |
91ec6c85 | 505 | struct inode *inode = file_inode(file); |
5a8bee63 | 506 | struct fuse_mount *fm = get_fuse_mount(inode); |
91ec6c85 MS |
507 | struct fuse_file *ff = file->private_data; |
508 | struct fuse_flush_in inarg; | |
509 | FUSE_ARGS(args); | |
510 | int err; | |
511 | ||
512 | if (fuse_is_bad(inode)) | |
513 | return -EIO; | |
514 | ||
515 | if (ff->open_flags & FOPEN_NOFLUSH && !fm->fc->writeback_cache) | |
516 | return 0; | |
a390ccb3 | 517 | |
1e18bda8 | 518 | err = write_inode_now(inode, 1); |
fe38d7df | 519 | if (err) |
91ec6c85 | 520 | return err; |
fe38d7df | 521 | |
5955102c | 522 | inode_lock(inode); |
fe38d7df | 523 | fuse_sync_writes(inode); |
5955102c | 524 | inode_unlock(inode); |
fe38d7df | 525 | |
91ec6c85 | 526 | err = filemap_check_errors(file->f_mapping); |
9ebce595 | 527 | if (err) |
91ec6c85 | 528 | return err; |
9ebce595 | 529 | |
614c026e | 530 | err = 0; |
fcee216b | 531 | if (fm->fc->no_flush) |
614c026e MS |
532 | goto inval_attr_out; |
533 | ||
91ec6c85 MS |
534 | memset(&inarg, 0, sizeof(inarg)); |
535 | inarg.fh = ff->fh; | |
536 | inarg.lock_owner = fuse_lock_owner_id(fm->fc, id); | |
537 | args.opcode = FUSE_FLUSH; | |
538 | args.nodeid = get_node_id(inode); | |
539 | args.in_numargs = 1; | |
540 | args.in_args[0].size = sizeof(inarg); | |
541 | args.in_args[0].value = &inarg; | |
542 | args.force = true; | |
543 | ||
544 | err = fuse_simple_request(fm, &args); | |
b6aeaded | 545 | if (err == -ENOSYS) { |
fcee216b | 546 | fm->fc->no_flush = 1; |
b6aeaded MS |
547 | err = 0; |
548 | } | |
cf576c58 EG |
549 | |
550 | inval_attr_out: | |
551 | /* | |
552 | * In memory i_blocks is not maintained by fuse, if writeback cache is | |
553 | * enabled, i_blocks from cached attr may not be accurate. | |
554 | */ | |
fcee216b | 555 | if (!err && fm->fc->writeback_cache) |
fa5eee57 | 556 | fuse_invalidate_attr_mask(inode, STATX_BLOCKS); |
b6aeaded MS |
557 | return err; |
558 | } | |
559 | ||
02c24a82 | 560 | int fuse_fsync_common(struct file *file, loff_t start, loff_t end, |
a9c2d1e8 | 561 | int datasync, int opcode) |
b6aeaded | 562 | { |
7ea80859 | 563 | struct inode *inode = file->f_mapping->host; |
fcee216b | 564 | struct fuse_mount *fm = get_fuse_mount(inode); |
b6aeaded | 565 | struct fuse_file *ff = file->private_data; |
7078187a | 566 | FUSE_ARGS(args); |
b6aeaded | 567 | struct fuse_fsync_in inarg; |
a9c2d1e8 MS |
568 | |
569 | memset(&inarg, 0, sizeof(inarg)); | |
570 | inarg.fh = ff->fh; | |
154603fe | 571 | inarg.fsync_flags = datasync ? FUSE_FSYNC_FDATASYNC : 0; |
d5b48543 MS |
572 | args.opcode = opcode; |
573 | args.nodeid = get_node_id(inode); | |
574 | args.in_numargs = 1; | |
575 | args.in_args[0].size = sizeof(inarg); | |
576 | args.in_args[0].value = &inarg; | |
fcee216b | 577 | return fuse_simple_request(fm, &args); |
a9c2d1e8 MS |
578 | } |
579 | ||
580 | static int fuse_fsync(struct file *file, loff_t start, loff_t end, | |
581 | int datasync) | |
582 | { | |
583 | struct inode *inode = file->f_mapping->host; | |
584 | struct fuse_conn *fc = get_fuse_conn(inode); | |
b6aeaded MS |
585 | int err; |
586 | ||
5d069dbe | 587 | if (fuse_is_bad(inode)) |
248d86e8 MS |
588 | return -EIO; |
589 | ||
5955102c | 590 | inode_lock(inode); |
02c24a82 | 591 | |
3be5a52b MS |
592 | /* |
593 | * Start writeback against all dirty pages of the inode, then | |
594 | * wait for all outstanding writes, before sending the FSYNC | |
595 | * request. | |
596 | */ | |
7e51fe1d | 597 | err = file_write_and_wait_range(file, start, end); |
3be5a52b | 598 | if (err) |
02c24a82 | 599 | goto out; |
3be5a52b MS |
600 | |
601 | fuse_sync_writes(inode); | |
ac7f052b AK |
602 | |
603 | /* | |
604 | * Due to implementation of fuse writeback | |
7e51fe1d | 605 | * file_write_and_wait_range() does not catch errors. |
ac7f052b AK |
606 | * We have to do this directly after fuse_sync_writes() |
607 | */ | |
7e51fe1d | 608 | err = file_check_and_advance_wb_err(file); |
ac7f052b AK |
609 | if (err) |
610 | goto out; | |
611 | ||
1e18bda8 MS |
612 | err = sync_inode_metadata(inode, 1); |
613 | if (err) | |
614 | goto out; | |
3be5a52b | 615 | |
a9c2d1e8 | 616 | if (fc->no_fsync) |
22401e7b | 617 | goto out; |
b0aa7606 | 618 | |
a9c2d1e8 | 619 | err = fuse_fsync_common(file, start, end, datasync, FUSE_FSYNC); |
b6aeaded | 620 | if (err == -ENOSYS) { |
a9c2d1e8 | 621 | fc->no_fsync = 1; |
b6aeaded MS |
622 | err = 0; |
623 | } | |
02c24a82 | 624 | out: |
5955102c | 625 | inode_unlock(inode); |
b6aeaded | 626 | |
a9c2d1e8 | 627 | return err; |
82547981 MS |
628 | } |
629 | ||
00793ca5 MS |
630 | void fuse_read_args_fill(struct fuse_io_args *ia, struct file *file, loff_t pos, |
631 | size_t count, int opcode) | |
632 | { | |
633 | struct fuse_file *ff = file->private_data; | |
634 | struct fuse_args *args = &ia->ap.args; | |
635 | ||
636 | ia->read.in.fh = ff->fh; | |
637 | ia->read.in.offset = pos; | |
638 | ia->read.in.size = count; | |
639 | ia->read.in.flags = file->f_flags; | |
640 | args->opcode = opcode; | |
641 | args->nodeid = ff->nodeid; | |
642 | args->in_numargs = 1; | |
643 | args->in_args[0].size = sizeof(ia->read.in); | |
644 | args->in_args[0].value = &ia->read.in; | |
645 | args->out_argvar = true; | |
646 | args->out_numargs = 1; | |
647 | args->out_args[0].size = count; | |
648 | } | |
649 | ||
45ac96ed MS |
650 | static void fuse_release_user_pages(struct fuse_args_pages *ap, |
651 | bool should_dirty) | |
187c5c36 | 652 | { |
45ac96ed | 653 | unsigned int i; |
187c5c36 | 654 | |
45ac96ed | 655 | for (i = 0; i < ap->num_pages; i++) { |
8fba54ae | 656 | if (should_dirty) |
45ac96ed MS |
657 | set_page_dirty_lock(ap->pages[i]); |
658 | put_page(ap->pages[i]); | |
187c5c36 MP |
659 | } |
660 | } | |
661 | ||
744742d6 SF |
662 | static void fuse_io_release(struct kref *kref) |
663 | { | |
664 | kfree(container_of(kref, struct fuse_io_priv, refcnt)); | |
665 | } | |
666 | ||
9d5722b7 CH |
667 | static ssize_t fuse_get_res_by_io(struct fuse_io_priv *io) |
668 | { | |
669 | if (io->err) | |
670 | return io->err; | |
671 | ||
672 | if (io->bytes >= 0 && io->write) | |
673 | return -EIO; | |
674 | ||
675 | return io->bytes < 0 ? io->size : io->bytes; | |
676 | } | |
677 | ||
06bbb761 | 678 | /* |
01e9d11a MP |
679 | * In case of short read, the caller sets 'pos' to the position of |
680 | * actual end of fuse request in IO request. Otherwise, if bytes_requested | |
681 | * == bytes_transferred or rw == WRITE, the caller sets 'pos' to -1. | |
682 | * | |
683 | * An example: | |
c4e0cd4e | 684 | * User requested DIO read of 64K. It was split into two 32K fuse requests, |
01e9d11a MP |
685 | * both submitted asynchronously. The first of them was ACKed by userspace as |
686 | * fully completed (req->out.args[0].size == 32K) resulting in pos == -1. The | |
687 | * second request was ACKed as short, e.g. only 1K was read, resulting in | |
688 | * pos == 33K. | |
689 | * | |
690 | * Thus, when all fuse requests are completed, the minimal non-negative 'pos' | |
691 | * will be equal to the length of the longest contiguous fragment of | |
692 | * transferred data starting from the beginning of IO request. | |
693 | */ | |
694 | static void fuse_aio_complete(struct fuse_io_priv *io, int err, ssize_t pos) | |
695 | { | |
696 | int left; | |
697 | ||
698 | spin_lock(&io->lock); | |
699 | if (err) | |
700 | io->err = io->err ? : err; | |
701 | else if (pos >= 0 && (io->bytes < 0 || pos < io->bytes)) | |
702 | io->bytes = pos; | |
703 | ||
704 | left = --io->reqs; | |
7879c4e5 | 705 | if (!left && io->blocking) |
9d5722b7 | 706 | complete(io->done); |
01e9d11a MP |
707 | spin_unlock(&io->lock); |
708 | ||
7879c4e5 | 709 | if (!left && !io->blocking) { |
9d5722b7 | 710 | ssize_t res = fuse_get_res_by_io(io); |
01e9d11a | 711 | |
9d5722b7 CH |
712 | if (res >= 0) { |
713 | struct inode *inode = file_inode(io->iocb->ki_filp); | |
714 | struct fuse_conn *fc = get_fuse_conn(inode); | |
715 | struct fuse_inode *fi = get_fuse_inode(inode); | |
01e9d11a | 716 | |
f15ecfef | 717 | spin_lock(&fi->lock); |
4510d86f | 718 | fi->attr_version = atomic64_inc_return(&fc->attr_version); |
f15ecfef | 719 | spin_unlock(&fi->lock); |
01e9d11a MP |
720 | } |
721 | ||
6b19b766 | 722 | io->iocb->ki_complete(io->iocb, res); |
01e9d11a | 723 | } |
744742d6 SF |
724 | |
725 | kref_put(&io->refcnt, fuse_io_release); | |
01e9d11a MP |
726 | } |
727 | ||
45ac96ed MS |
728 | static struct fuse_io_args *fuse_io_alloc(struct fuse_io_priv *io, |
729 | unsigned int npages) | |
730 | { | |
731 | struct fuse_io_args *ia; | |
732 | ||
733 | ia = kzalloc(sizeof(*ia), GFP_KERNEL); | |
734 | if (ia) { | |
735 | ia->io = io; | |
736 | ia->ap.pages = fuse_pages_alloc(npages, GFP_KERNEL, | |
737 | &ia->ap.descs); | |
738 | if (!ia->ap.pages) { | |
739 | kfree(ia); | |
740 | ia = NULL; | |
741 | } | |
742 | } | |
743 | return ia; | |
744 | } | |
745 | ||
746 | static void fuse_io_free(struct fuse_io_args *ia) | |
01e9d11a | 747 | { |
45ac96ed MS |
748 | kfree(ia->ap.pages); |
749 | kfree(ia); | |
750 | } | |
751 | ||
fcee216b | 752 | static void fuse_aio_complete_req(struct fuse_mount *fm, struct fuse_args *args, |
45ac96ed MS |
753 | int err) |
754 | { | |
755 | struct fuse_io_args *ia = container_of(args, typeof(*ia), ap.args); | |
756 | struct fuse_io_priv *io = ia->io; | |
01e9d11a MP |
757 | ssize_t pos = -1; |
758 | ||
45ac96ed | 759 | fuse_release_user_pages(&ia->ap, io->should_dirty); |
01e9d11a | 760 | |
45ac96ed MS |
761 | if (err) { |
762 | /* Nothing */ | |
763 | } else if (io->write) { | |
764 | if (ia->write.out.size > ia->write.in.size) { | |
765 | err = -EIO; | |
766 | } else if (ia->write.in.size != ia->write.out.size) { | |
767 | pos = ia->write.in.offset - io->offset + | |
768 | ia->write.out.size; | |
769 | } | |
01e9d11a | 770 | } else { |
45ac96ed MS |
771 | u32 outsize = args->out_args[0].size; |
772 | ||
773 | if (ia->read.in.size != outsize) | |
774 | pos = ia->read.in.offset - io->offset + outsize; | |
01e9d11a MP |
775 | } |
776 | ||
45ac96ed MS |
777 | fuse_aio_complete(io, err, pos); |
778 | fuse_io_free(ia); | |
01e9d11a MP |
779 | } |
780 | ||
fcee216b | 781 | static ssize_t fuse_async_req_send(struct fuse_mount *fm, |
45ac96ed | 782 | struct fuse_io_args *ia, size_t num_bytes) |
01e9d11a | 783 | { |
45ac96ed MS |
784 | ssize_t err; |
785 | struct fuse_io_priv *io = ia->io; | |
786 | ||
01e9d11a | 787 | spin_lock(&io->lock); |
744742d6 | 788 | kref_get(&io->refcnt); |
01e9d11a MP |
789 | io->size += num_bytes; |
790 | io->reqs++; | |
791 | spin_unlock(&io->lock); | |
792 | ||
45ac96ed | 793 | ia->ap.args.end = fuse_aio_complete_req; |
bb737bbe | 794 | ia->ap.args.may_block = io->should_dirty; |
fcee216b | 795 | err = fuse_simple_background(fm, &ia->ap.args, GFP_KERNEL); |
f1ebdeff | 796 | if (err) |
fcee216b | 797 | fuse_aio_complete_req(fm, &ia->ap.args, err); |
01e9d11a | 798 | |
f1ebdeff | 799 | return num_bytes; |
01e9d11a MP |
800 | } |
801 | ||
45ac96ed MS |
802 | static ssize_t fuse_send_read(struct fuse_io_args *ia, loff_t pos, size_t count, |
803 | fl_owner_t owner) | |
04730fef | 804 | { |
45ac96ed | 805 | struct file *file = ia->io->iocb->ki_filp; |
2106cb18 | 806 | struct fuse_file *ff = file->private_data; |
fcee216b | 807 | struct fuse_mount *fm = ff->fm; |
f3332114 | 808 | |
45ac96ed | 809 | fuse_read_args_fill(ia, file, pos, count, FUSE_READ); |
f3332114 | 810 | if (owner != NULL) { |
45ac96ed | 811 | ia->read.in.read_flags |= FUSE_READ_LOCKOWNER; |
fcee216b | 812 | ia->read.in.lock_owner = fuse_lock_owner_id(fm->fc, owner); |
f3332114 | 813 | } |
36cf66ed | 814 | |
45ac96ed | 815 | if (ia->io->async) |
fcee216b | 816 | return fuse_async_req_send(fm, ia, count); |
36cf66ed | 817 | |
fcee216b | 818 | return fuse_simple_request(fm, &ia->ap.args); |
04730fef MS |
819 | } |
820 | ||
5c5c5e51 MS |
821 | static void fuse_read_update_size(struct inode *inode, loff_t size, |
822 | u64 attr_ver) | |
823 | { | |
824 | struct fuse_conn *fc = get_fuse_conn(inode); | |
825 | struct fuse_inode *fi = get_fuse_inode(inode); | |
826 | ||
f15ecfef | 827 | spin_lock(&fi->lock); |
484ce657 | 828 | if (attr_ver >= fi->attr_version && size < inode->i_size && |
06a7c3c2 | 829 | !test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) { |
4510d86f | 830 | fi->attr_version = atomic64_inc_return(&fc->attr_version); |
5c5c5e51 MS |
831 | i_size_write(inode, size); |
832 | } | |
f15ecfef | 833 | spin_unlock(&fi->lock); |
5c5c5e51 MS |
834 | } |
835 | ||
a0d45d84 | 836 | static void fuse_short_read(struct inode *inode, u64 attr_ver, size_t num_read, |
134831e3 | 837 | struct fuse_args_pages *ap) |
a92adc82 | 838 | { |
8373200b PE |
839 | struct fuse_conn *fc = get_fuse_conn(inode); |
840 | ||
a73d47f5 MS |
841 | /* |
842 | * If writeback_cache is enabled, a short read means there's a hole in | |
843 | * the file. Some data after the hole is in page cache, but has not | |
844 | * reached the client fs yet. So the hole is not present there. | |
845 | */ | |
846 | if (!fc->writeback_cache) { | |
134831e3 | 847 | loff_t pos = page_offset(ap->pages[0]) + num_read; |
8373200b PE |
848 | fuse_read_update_size(inode, pos, attr_ver); |
849 | } | |
a92adc82 PE |
850 | } |
851 | ||
482fce55 | 852 | static int fuse_do_readpage(struct file *file, struct page *page) |
b6aeaded MS |
853 | { |
854 | struct inode *inode = page->mapping->host; | |
fcee216b | 855 | struct fuse_mount *fm = get_fuse_mount(inode); |
5c5c5e51 | 856 | loff_t pos = page_offset(page); |
00793ca5 MS |
857 | struct fuse_page_desc desc = { .length = PAGE_SIZE }; |
858 | struct fuse_io_args ia = { | |
859 | .ap.args.page_zeroing = true, | |
860 | .ap.args.out_pages = true, | |
861 | .ap.num_pages = 1, | |
862 | .ap.pages = &page, | |
863 | .ap.descs = &desc, | |
864 | }; | |
865 | ssize_t res; | |
5c5c5e51 | 866 | u64 attr_ver; |
248d86e8 | 867 | |
3be5a52b | 868 | /* |
25985edc | 869 | * Page writeback can extend beyond the lifetime of the |
3be5a52b MS |
870 | * page-cache page, so make sure we read a properly synced |
871 | * page. | |
872 | */ | |
873 | fuse_wait_on_page_writeback(inode, page->index); | |
874 | ||
fcee216b | 875 | attr_ver = fuse_get_attr_version(fm->fc); |
5c5c5e51 | 876 | |
2f139829 MS |
877 | /* Don't overflow end offset */ |
878 | if (pos + (desc.length - 1) == LLONG_MAX) | |
879 | desc.length--; | |
880 | ||
00793ca5 | 881 | fuse_read_args_fill(&ia, file, pos, desc.length, FUSE_READ); |
fcee216b | 882 | res = fuse_simple_request(fm, &ia.ap.args); |
00793ca5 MS |
883 | if (res < 0) |
884 | return res; | |
885 | /* | |
886 | * Short read means EOF. If file size is larger, truncate it | |
887 | */ | |
888 | if (res < desc.length) | |
134831e3 | 889 | fuse_short_read(inode, attr_ver, res, &ia.ap); |
5c5c5e51 | 890 | |
00793ca5 | 891 | SetPageUptodate(page); |
482fce55 | 892 | |
00793ca5 | 893 | return 0; |
482fce55 MP |
894 | } |
895 | ||
5efd00e4 | 896 | static int fuse_read_folio(struct file *file, struct folio *folio) |
482fce55 | 897 | { |
5efd00e4 | 898 | struct page *page = &folio->page; |
482fce55 MP |
899 | struct inode *inode = page->mapping->host; |
900 | int err; | |
901 | ||
902 | err = -EIO; | |
5d069dbe | 903 | if (fuse_is_bad(inode)) |
482fce55 MP |
904 | goto out; |
905 | ||
906 | err = fuse_do_readpage(file, page); | |
451418fc | 907 | fuse_invalidate_atime(inode); |
b6aeaded MS |
908 | out: |
909 | unlock_page(page); | |
910 | return err; | |
911 | } | |
912 | ||
fcee216b | 913 | static void fuse_readpages_end(struct fuse_mount *fm, struct fuse_args *args, |
134831e3 | 914 | int err) |
db50b96c | 915 | { |
c1aa96a5 | 916 | int i; |
134831e3 MS |
917 | struct fuse_io_args *ia = container_of(args, typeof(*ia), ap.args); |
918 | struct fuse_args_pages *ap = &ia->ap; | |
919 | size_t count = ia->read.in.size; | |
920 | size_t num_read = args->out_args[0].size; | |
ce534fb0 | 921 | struct address_space *mapping = NULL; |
c1aa96a5 | 922 | |
134831e3 MS |
923 | for (i = 0; mapping == NULL && i < ap->num_pages; i++) |
924 | mapping = ap->pages[i]->mapping; | |
5c5c5e51 | 925 | |
ce534fb0 MS |
926 | if (mapping) { |
927 | struct inode *inode = mapping->host; | |
928 | ||
929 | /* | |
930 | * Short read means EOF. If file size is larger, truncate it | |
931 | */ | |
134831e3 MS |
932 | if (!err && num_read < count) |
933 | fuse_short_read(inode, ia->read.attr_ver, num_read, ap); | |
ce534fb0 | 934 | |
451418fc | 935 | fuse_invalidate_atime(inode); |
ce534fb0 | 936 | } |
c1aa96a5 | 937 | |
134831e3 MS |
938 | for (i = 0; i < ap->num_pages; i++) { |
939 | struct page *page = ap->pages[i]; | |
940 | ||
941 | if (!err) | |
db50b96c | 942 | SetPageUptodate(page); |
c1aa96a5 MS |
943 | else |
944 | SetPageError(page); | |
db50b96c | 945 | unlock_page(page); |
09cbfeaf | 946 | put_page(page); |
db50b96c | 947 | } |
134831e3 | 948 | if (ia->ff) |
e26ee4ef | 949 | fuse_file_put(ia->ff, false); |
134831e3 MS |
950 | |
951 | fuse_io_free(ia); | |
c1aa96a5 MS |
952 | } |
953 | ||
134831e3 | 954 | static void fuse_send_readpages(struct fuse_io_args *ia, struct file *file) |
c1aa96a5 | 955 | { |
2106cb18 | 956 | struct fuse_file *ff = file->private_data; |
fcee216b | 957 | struct fuse_mount *fm = ff->fm; |
134831e3 MS |
958 | struct fuse_args_pages *ap = &ia->ap; |
959 | loff_t pos = page_offset(ap->pages[0]); | |
960 | size_t count = ap->num_pages << PAGE_SHIFT; | |
7df1e988 | 961 | ssize_t res; |
134831e3 MS |
962 | int err; |
963 | ||
964 | ap->args.out_pages = true; | |
965 | ap->args.page_zeroing = true; | |
966 | ap->args.page_replace = true; | |
2f139829 MS |
967 | |
968 | /* Don't overflow end offset */ | |
969 | if (pos + (count - 1) == LLONG_MAX) { | |
970 | count--; | |
971 | ap->descs[ap->num_pages - 1].length--; | |
972 | } | |
973 | WARN_ON((loff_t) (pos + count) < 0); | |
974 | ||
134831e3 | 975 | fuse_read_args_fill(ia, file, pos, count, FUSE_READ); |
fcee216b MR |
976 | ia->read.attr_ver = fuse_get_attr_version(fm->fc); |
977 | if (fm->fc->async_read) { | |
134831e3 MS |
978 | ia->ff = fuse_file_get(ff); |
979 | ap->args.end = fuse_readpages_end; | |
fcee216b | 980 | err = fuse_simple_background(fm, &ap->args, GFP_KERNEL); |
134831e3 MS |
981 | if (!err) |
982 | return; | |
9cd68455 | 983 | } else { |
fcee216b | 984 | res = fuse_simple_request(fm, &ap->args); |
7df1e988 | 985 | err = res < 0 ? res : 0; |
9cd68455 | 986 | } |
fcee216b | 987 | fuse_readpages_end(fm, &ap->args, err); |
db50b96c MS |
988 | } |
989 | ||
76a0294e | 990 | static void fuse_readahead(struct readahead_control *rac) |
db50b96c | 991 | { |
76a0294e | 992 | struct inode *inode = rac->mapping->host; |
db50b96c | 993 | struct fuse_conn *fc = get_fuse_conn(inode); |
76a0294e | 994 | unsigned int i, max_pages, nr_pages = 0; |
db50b96c | 995 | |
5d069dbe | 996 | if (fuse_is_bad(inode)) |
76a0294e | 997 | return; |
248d86e8 | 998 | |
76a0294e MWO |
999 | max_pages = min_t(unsigned int, fc->max_pages, |
1000 | fc->max_read / PAGE_SIZE); | |
db50b96c | 1001 | |
76a0294e MWO |
1002 | for (;;) { |
1003 | struct fuse_io_args *ia; | |
1004 | struct fuse_args_pages *ap; | |
1005 | ||
670d21c6 N |
1006 | if (fc->num_background >= fc->congestion_threshold && |
1007 | rac->ra->async_size >= readahead_count(rac)) | |
1008 | /* | |
1009 | * Congested and only async pages left, so skip the | |
1010 | * rest. | |
1011 | */ | |
1012 | break; | |
1013 | ||
76a0294e MWO |
1014 | nr_pages = readahead_count(rac) - nr_pages; |
1015 | if (nr_pages > max_pages) | |
1016 | nr_pages = max_pages; | |
1017 | if (nr_pages == 0) | |
1018 | break; | |
1019 | ia = fuse_io_alloc(NULL, nr_pages); | |
1020 | if (!ia) | |
1021 | return; | |
1022 | ap = &ia->ap; | |
1023 | nr_pages = __readahead_batch(rac, ap->pages, nr_pages); | |
1024 | for (i = 0; i < nr_pages; i++) { | |
1025 | fuse_wait_on_page_writeback(inode, | |
1026 | readahead_index(rac) + i); | |
1027 | ap->descs[i].length = PAGE_SIZE; | |
1028 | } | |
1029 | ap->num_pages = nr_pages; | |
1030 | fuse_send_readpages(ia, rac->file); | |
d3406ffa | 1031 | } |
db50b96c MS |
1032 | } |
1033 | ||
55752a3a | 1034 | static ssize_t fuse_cache_read_iter(struct kiocb *iocb, struct iov_iter *to) |
bcb4be80 MS |
1035 | { |
1036 | struct inode *inode = iocb->ki_filp->f_mapping->host; | |
a8894274 | 1037 | struct fuse_conn *fc = get_fuse_conn(inode); |
bcb4be80 | 1038 | |
a8894274 BF |
1039 | /* |
1040 | * In auto invalidate mode, always update attributes on read. | |
1041 | * Otherwise, only update if we attempt to read past EOF (to ensure | |
1042 | * i_size is up to date). | |
1043 | */ | |
1044 | if (fc->auto_inval_data || | |
37c20f16 | 1045 | (iocb->ki_pos + iov_iter_count(to) > i_size_read(inode))) { |
bcb4be80 | 1046 | int err; |
c6c745b8 | 1047 | err = fuse_update_attributes(inode, iocb->ki_filp, STATX_SIZE); |
bcb4be80 MS |
1048 | if (err) |
1049 | return err; | |
1050 | } | |
1051 | ||
37c20f16 | 1052 | return generic_file_read_iter(iocb, to); |
bcb4be80 MS |
1053 | } |
1054 | ||
338f2e3f MS |
1055 | static void fuse_write_args_fill(struct fuse_io_args *ia, struct fuse_file *ff, |
1056 | loff_t pos, size_t count) | |
1057 | { | |
1058 | struct fuse_args *args = &ia->ap.args; | |
1059 | ||
1060 | ia->write.in.fh = ff->fh; | |
1061 | ia->write.in.offset = pos; | |
1062 | ia->write.in.size = count; | |
1063 | args->opcode = FUSE_WRITE; | |
1064 | args->nodeid = ff->nodeid; | |
1065 | args->in_numargs = 2; | |
fcee216b | 1066 | if (ff->fm->fc->minor < 9) |
338f2e3f MS |
1067 | args->in_args[0].size = FUSE_COMPAT_WRITE_IN_SIZE; |
1068 | else | |
1069 | args->in_args[0].size = sizeof(ia->write.in); | |
1070 | args->in_args[0].value = &ia->write.in; | |
1071 | args->in_args[1].size = count; | |
1072 | args->out_numargs = 1; | |
1073 | args->out_args[0].size = sizeof(ia->write.out); | |
1074 | args->out_args[0].value = &ia->write.out; | |
1075 | } | |
1076 | ||
1077 | static unsigned int fuse_write_flags(struct kiocb *iocb) | |
1078 | { | |
1079 | unsigned int flags = iocb->ki_filp->f_flags; | |
1080 | ||
91b94c5d | 1081 | if (iocb_is_dsync(iocb)) |
338f2e3f MS |
1082 | flags |= O_DSYNC; |
1083 | if (iocb->ki_flags & IOCB_SYNC) | |
1084 | flags |= O_SYNC; | |
1085 | ||
1086 | return flags; | |
1087 | } | |
1088 | ||
45ac96ed MS |
1089 | static ssize_t fuse_send_write(struct fuse_io_args *ia, loff_t pos, |
1090 | size_t count, fl_owner_t owner) | |
b25e82e5 | 1091 | { |
45ac96ed | 1092 | struct kiocb *iocb = ia->io->iocb; |
e1c0eecb | 1093 | struct file *file = iocb->ki_filp; |
2106cb18 | 1094 | struct fuse_file *ff = file->private_data; |
fcee216b | 1095 | struct fuse_mount *fm = ff->fm; |
45ac96ed MS |
1096 | struct fuse_write_in *inarg = &ia->write.in; |
1097 | ssize_t err; | |
2d698b07 | 1098 | |
45ac96ed | 1099 | fuse_write_args_fill(ia, ff, pos, count); |
338f2e3f | 1100 | inarg->flags = fuse_write_flags(iocb); |
f3332114 | 1101 | if (owner != NULL) { |
f3332114 | 1102 | inarg->write_flags |= FUSE_WRITE_LOCKOWNER; |
fcee216b | 1103 | inarg->lock_owner = fuse_lock_owner_id(fm->fc, owner); |
f3332114 | 1104 | } |
36cf66ed | 1105 | |
45ac96ed | 1106 | if (ia->io->async) |
fcee216b | 1107 | return fuse_async_req_send(fm, ia, count); |
45ac96ed | 1108 | |
fcee216b | 1109 | err = fuse_simple_request(fm, &ia->ap.args); |
45ac96ed MS |
1110 | if (!err && ia->write.out.size > count) |
1111 | err = -EIO; | |
36cf66ed | 1112 | |
45ac96ed | 1113 | return err ?: ia->write.out.size; |
b6aeaded MS |
1114 | } |
1115 | ||
d347739a | 1116 | bool fuse_write_update_attr(struct inode *inode, loff_t pos, ssize_t written) |
854512ec MS |
1117 | { |
1118 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1119 | struct fuse_inode *fi = get_fuse_inode(inode); | |
b0aa7606 | 1120 | bool ret = false; |
854512ec | 1121 | |
f15ecfef | 1122 | spin_lock(&fi->lock); |
4510d86f | 1123 | fi->attr_version = atomic64_inc_return(&fc->attr_version); |
d347739a | 1124 | if (written > 0 && pos > inode->i_size) { |
854512ec | 1125 | i_size_write(inode, pos); |
b0aa7606 MP |
1126 | ret = true; |
1127 | } | |
f15ecfef | 1128 | spin_unlock(&fi->lock); |
b0aa7606 | 1129 | |
d347739a MS |
1130 | fuse_invalidate_attr_mask(inode, FUSE_STATX_MODSIZE); |
1131 | ||
b0aa7606 | 1132 | return ret; |
854512ec MS |
1133 | } |
1134 | ||
338f2e3f MS |
1135 | static ssize_t fuse_send_write_pages(struct fuse_io_args *ia, |
1136 | struct kiocb *iocb, struct inode *inode, | |
1137 | loff_t pos, size_t count) | |
ea9b9907 | 1138 | { |
338f2e3f MS |
1139 | struct fuse_args_pages *ap = &ia->ap; |
1140 | struct file *file = iocb->ki_filp; | |
1141 | struct fuse_file *ff = file->private_data; | |
fcee216b | 1142 | struct fuse_mount *fm = ff->fm; |
338f2e3f | 1143 | unsigned int offset, i; |
4f06dd92 | 1144 | bool short_write; |
338f2e3f | 1145 | int err; |
ea9b9907 | 1146 | |
338f2e3f MS |
1147 | for (i = 0; i < ap->num_pages; i++) |
1148 | fuse_wait_on_page_writeback(inode, ap->pages[i]->index); | |
ea9b9907 | 1149 | |
338f2e3f MS |
1150 | fuse_write_args_fill(ia, ff, pos, count); |
1151 | ia->write.in.flags = fuse_write_flags(iocb); | |
b8667395 VG |
1152 | if (fm->fc->handle_killpriv_v2 && !capable(CAP_FSETID)) |
1153 | ia->write.in.write_flags |= FUSE_WRITE_KILL_SUIDGID; | |
ea9b9907 | 1154 | |
fcee216b | 1155 | err = fuse_simple_request(fm, &ap->args); |
8aab336b MS |
1156 | if (!err && ia->write.out.size > count) |
1157 | err = -EIO; | |
338f2e3f | 1158 | |
4f06dd92 | 1159 | short_write = ia->write.out.size < count; |
338f2e3f MS |
1160 | offset = ap->descs[0].offset; |
1161 | count = ia->write.out.size; | |
1162 | for (i = 0; i < ap->num_pages; i++) { | |
1163 | struct page *page = ap->pages[i]; | |
ea9b9907 | 1164 | |
4f06dd92 VG |
1165 | if (err) { |
1166 | ClearPageUptodate(page); | |
1167 | } else { | |
1168 | if (count >= PAGE_SIZE - offset) | |
1169 | count -= PAGE_SIZE - offset; | |
1170 | else { | |
1171 | if (short_write) | |
1172 | ClearPageUptodate(page); | |
1173 | count = 0; | |
1174 | } | |
1175 | offset = 0; | |
1176 | } | |
1177 | if (ia->write.page_locked && (i == ap->num_pages - 1)) | |
1178 | unlock_page(page); | |
09cbfeaf | 1179 | put_page(page); |
ea9b9907 NP |
1180 | } |
1181 | ||
338f2e3f | 1182 | return err; |
ea9b9907 NP |
1183 | } |
1184 | ||
4f06dd92 | 1185 | static ssize_t fuse_fill_write_pages(struct fuse_io_args *ia, |
338f2e3f MS |
1186 | struct address_space *mapping, |
1187 | struct iov_iter *ii, loff_t pos, | |
1188 | unsigned int max_pages) | |
ea9b9907 | 1189 | { |
4f06dd92 | 1190 | struct fuse_args_pages *ap = &ia->ap; |
ea9b9907 | 1191 | struct fuse_conn *fc = get_fuse_conn(mapping->host); |
09cbfeaf | 1192 | unsigned offset = pos & (PAGE_SIZE - 1); |
ea9b9907 NP |
1193 | size_t count = 0; |
1194 | int err; | |
1195 | ||
338f2e3f MS |
1196 | ap->args.in_pages = true; |
1197 | ap->descs[0].offset = offset; | |
ea9b9907 NP |
1198 | |
1199 | do { | |
1200 | size_t tmp; | |
1201 | struct page *page; | |
09cbfeaf KS |
1202 | pgoff_t index = pos >> PAGE_SHIFT; |
1203 | size_t bytes = min_t(size_t, PAGE_SIZE - offset, | |
ea9b9907 NP |
1204 | iov_iter_count(ii)); |
1205 | ||
1206 | bytes = min_t(size_t, bytes, fc->max_write - count); | |
1207 | ||
1208 | again: | |
1209 | err = -EFAULT; | |
a6294593 | 1210 | if (fault_in_iov_iter_readable(ii, bytes)) |
ea9b9907 NP |
1211 | break; |
1212 | ||
1213 | err = -ENOMEM; | |
b7446e7c | 1214 | page = grab_cache_page_write_begin(mapping, index); |
ea9b9907 NP |
1215 | if (!page) |
1216 | break; | |
1217 | ||
931e80e4 | 1218 | if (mapping_writably_mapped(mapping)) |
1219 | flush_dcache_page(page); | |
1220 | ||
f0b65f39 | 1221 | tmp = copy_page_from_iter_atomic(page, offset, bytes, ii); |
ea9b9907 NP |
1222 | flush_dcache_page(page); |
1223 | ||
1224 | if (!tmp) { | |
1225 | unlock_page(page); | |
09cbfeaf | 1226 | put_page(page); |
ea9b9907 NP |
1227 | goto again; |
1228 | } | |
1229 | ||
1230 | err = 0; | |
338f2e3f MS |
1231 | ap->pages[ap->num_pages] = page; |
1232 | ap->descs[ap->num_pages].length = tmp; | |
1233 | ap->num_pages++; | |
ea9b9907 | 1234 | |
ea9b9907 NP |
1235 | count += tmp; |
1236 | pos += tmp; | |
1237 | offset += tmp; | |
09cbfeaf | 1238 | if (offset == PAGE_SIZE) |
ea9b9907 NP |
1239 | offset = 0; |
1240 | ||
4f06dd92 VG |
1241 | /* If we copied full page, mark it uptodate */ |
1242 | if (tmp == PAGE_SIZE) | |
1243 | SetPageUptodate(page); | |
1244 | ||
1245 | if (PageUptodate(page)) { | |
1246 | unlock_page(page); | |
1247 | } else { | |
1248 | ia->write.page_locked = true; | |
1249 | break; | |
1250 | } | |
78bb6cb9 MS |
1251 | if (!fc->big_writes) |
1252 | break; | |
ea9b9907 | 1253 | } while (iov_iter_count(ii) && count < fc->max_write && |
338f2e3f | 1254 | ap->num_pages < max_pages && offset == 0); |
ea9b9907 NP |
1255 | |
1256 | return count > 0 ? count : err; | |
1257 | } | |
1258 | ||
5da784cc CS |
1259 | static inline unsigned int fuse_wr_pages(loff_t pos, size_t len, |
1260 | unsigned int max_pages) | |
d07f09f5 | 1261 | { |
5da784cc | 1262 | return min_t(unsigned int, |
09cbfeaf KS |
1263 | ((pos + len - 1) >> PAGE_SHIFT) - |
1264 | (pos >> PAGE_SHIFT) + 1, | |
5da784cc | 1265 | max_pages); |
d07f09f5 MP |
1266 | } |
1267 | ||
596df33d | 1268 | static ssize_t fuse_perform_write(struct kiocb *iocb, struct iov_iter *ii) |
ea9b9907 | 1269 | { |
596df33d | 1270 | struct address_space *mapping = iocb->ki_filp->f_mapping; |
ea9b9907 NP |
1271 | struct inode *inode = mapping->host; |
1272 | struct fuse_conn *fc = get_fuse_conn(inode); | |
06a7c3c2 | 1273 | struct fuse_inode *fi = get_fuse_inode(inode); |
596df33d | 1274 | loff_t pos = iocb->ki_pos; |
ea9b9907 NP |
1275 | int err = 0; |
1276 | ssize_t res = 0; | |
1277 | ||
06a7c3c2 MP |
1278 | if (inode->i_size < pos + iov_iter_count(ii)) |
1279 | set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); | |
1280 | ||
ea9b9907 | 1281 | do { |
ea9b9907 | 1282 | ssize_t count; |
338f2e3f MS |
1283 | struct fuse_io_args ia = {}; |
1284 | struct fuse_args_pages *ap = &ia.ap; | |
5da784cc CS |
1285 | unsigned int nr_pages = fuse_wr_pages(pos, iov_iter_count(ii), |
1286 | fc->max_pages); | |
ea9b9907 | 1287 | |
338f2e3f MS |
1288 | ap->pages = fuse_pages_alloc(nr_pages, GFP_KERNEL, &ap->descs); |
1289 | if (!ap->pages) { | |
1290 | err = -ENOMEM; | |
ea9b9907 NP |
1291 | break; |
1292 | } | |
1293 | ||
4f06dd92 | 1294 | count = fuse_fill_write_pages(&ia, mapping, ii, pos, nr_pages); |
ea9b9907 NP |
1295 | if (count <= 0) { |
1296 | err = count; | |
1297 | } else { | |
338f2e3f MS |
1298 | err = fuse_send_write_pages(&ia, iocb, inode, |
1299 | pos, count); | |
ea9b9907 | 1300 | if (!err) { |
338f2e3f MS |
1301 | size_t num_written = ia.write.out.size; |
1302 | ||
ea9b9907 NP |
1303 | res += num_written; |
1304 | pos += num_written; | |
1305 | ||
1306 | /* break out of the loop on short write */ | |
1307 | if (num_written != count) | |
1308 | err = -EIO; | |
1309 | } | |
1310 | } | |
338f2e3f | 1311 | kfree(ap->pages); |
ea9b9907 NP |
1312 | } while (!err && iov_iter_count(ii)); |
1313 | ||
d347739a | 1314 | fuse_write_update_attr(inode, pos, res); |
06a7c3c2 | 1315 | clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); |
ea9b9907 | 1316 | |
70e986c3 CH |
1317 | if (!res) |
1318 | return err; | |
1319 | iocb->ki_pos += res; | |
1320 | return res; | |
ea9b9907 NP |
1321 | } |
1322 | ||
699cf824 BS |
1323 | static bool fuse_io_past_eof(struct kiocb *iocb, struct iov_iter *iter) |
1324 | { | |
1325 | struct inode *inode = file_inode(iocb->ki_filp); | |
1326 | ||
1327 | return iocb->ki_pos + iov_iter_count(iter) > i_size_read(inode); | |
1328 | } | |
1329 | ||
1330 | /* | |
1331 | * @return true if an exclusive lock for direct IO writes is needed | |
1332 | */ | |
1333 | static bool fuse_dio_wr_exclusive_lock(struct kiocb *iocb, struct iov_iter *from) | |
1334 | { | |
1335 | struct file *file = iocb->ki_filp; | |
1336 | struct fuse_file *ff = file->private_data; | |
1337 | struct inode *inode = file_inode(iocb->ki_filp); | |
205c1d80 | 1338 | struct fuse_inode *fi = get_fuse_inode(inode); |
699cf824 BS |
1339 | |
1340 | /* Server side has to advise that it supports parallel dio writes. */ | |
1341 | if (!(ff->open_flags & FOPEN_PARALLEL_DIRECT_WRITES)) | |
1342 | return true; | |
1343 | ||
1344 | /* | |
1345 | * Append will need to know the eventual EOF - always needs an | |
1346 | * exclusive lock. | |
1347 | */ | |
1348 | if (iocb->ki_flags & IOCB_APPEND) | |
1349 | return true; | |
1350 | ||
205c1d80 AG |
1351 | /* shared locks are not allowed with parallel page cache IO */ |
1352 | if (test_bit(FUSE_I_CACHE_IO_MODE, &fi->state)) | |
1353 | return false; | |
699cf824 BS |
1354 | |
1355 | /* Parallel dio beyond EOF is not supported, at least for now. */ | |
1356 | if (fuse_io_past_eof(iocb, from)) | |
1357 | return true; | |
1358 | ||
1359 | return false; | |
1360 | } | |
1361 | ||
9bbb6717 BS |
1362 | static void fuse_dio_lock(struct kiocb *iocb, struct iov_iter *from, |
1363 | bool *exclusive) | |
1364 | { | |
1365 | struct inode *inode = file_inode(iocb->ki_filp); | |
205c1d80 | 1366 | struct fuse_file *ff = iocb->ki_filp->private_data; |
9bbb6717 BS |
1367 | |
1368 | *exclusive = fuse_dio_wr_exclusive_lock(iocb, from); | |
1369 | if (*exclusive) { | |
1370 | inode_lock(inode); | |
1371 | } else { | |
1372 | inode_lock_shared(inode); | |
1373 | /* | |
205c1d80 AG |
1374 | * New parallal dio allowed only if inode is not in caching |
1375 | * mode and denies new opens in caching mode. This check | |
1376 | * should be performed only after taking shared inode lock. | |
1377 | * Previous past eof check was without inode lock and might | |
1378 | * have raced, so check it again. | |
9bbb6717 | 1379 | */ |
205c1d80 | 1380 | if (fuse_io_past_eof(iocb, from) || |
4a90451b | 1381 | fuse_file_uncached_io_start(inode, ff, NULL) != 0) { |
9bbb6717 BS |
1382 | inode_unlock_shared(inode); |
1383 | inode_lock(inode); | |
1384 | *exclusive = true; | |
1385 | } | |
1386 | } | |
1387 | } | |
1388 | ||
205c1d80 | 1389 | static void fuse_dio_unlock(struct kiocb *iocb, bool exclusive) |
9bbb6717 | 1390 | { |
205c1d80 AG |
1391 | struct inode *inode = file_inode(iocb->ki_filp); |
1392 | struct fuse_file *ff = iocb->ki_filp->private_data; | |
1393 | ||
9bbb6717 BS |
1394 | if (exclusive) { |
1395 | inode_unlock(inode); | |
1396 | } else { | |
205c1d80 AG |
1397 | /* Allow opens in caching mode after last parallel dio end */ |
1398 | fuse_file_uncached_io_end(inode, ff); | |
9bbb6717 BS |
1399 | inode_unlock_shared(inode); |
1400 | } | |
1401 | } | |
1402 | ||
55752a3a | 1403 | static ssize_t fuse_cache_write_iter(struct kiocb *iocb, struct iov_iter *from) |
ea9b9907 NP |
1404 | { |
1405 | struct file *file = iocb->ki_filp; | |
1406 | struct address_space *mapping = file->f_mapping; | |
ea9b9907 NP |
1407 | ssize_t written = 0; |
1408 | struct inode *inode = mapping->host; | |
2e3f7dd0 | 1409 | ssize_t err, count; |
8981bdfd | 1410 | struct fuse_conn *fc = get_fuse_conn(inode); |
ea9b9907 | 1411 | |
8981bdfd | 1412 | if (fc->writeback_cache) { |
4d99ff8f | 1413 | /* Update size (EOF optimization) and mode (SUID clearing) */ |
c6c745b8 MS |
1414 | err = fuse_update_attributes(mapping->host, file, |
1415 | STATX_SIZE | STATX_MODE); | |
4d99ff8f PE |
1416 | if (err) |
1417 | return err; | |
1418 | ||
8981bdfd | 1419 | if (fc->handle_killpriv_v2 && |
9452e93e CB |
1420 | setattr_should_drop_suidgid(&nop_mnt_idmap, |
1421 | file_inode(file))) { | |
8981bdfd VG |
1422 | goto writethrough; |
1423 | } | |
1424 | ||
84c3d55c | 1425 | return generic_file_write_iter(iocb, from); |
4d99ff8f PE |
1426 | } |
1427 | ||
8981bdfd | 1428 | writethrough: |
5955102c | 1429 | inode_lock(inode); |
ea9b9907 | 1430 | |
2e3f7dd0 | 1431 | err = count = generic_write_checks(iocb, from); |
3309dd04 | 1432 | if (err <= 0) |
ea9b9907 NP |
1433 | goto out; |
1434 | ||
2e3f7dd0 ZJ |
1435 | task_io_account_write(count); |
1436 | ||
5fa8e0a1 | 1437 | err = file_remove_privs(file); |
ea9b9907 NP |
1438 | if (err) |
1439 | goto out; | |
1440 | ||
c3b2da31 JB |
1441 | err = file_update_time(file); |
1442 | if (err) | |
1443 | goto out; | |
ea9b9907 | 1444 | |
2ba48ce5 | 1445 | if (iocb->ki_flags & IOCB_DIRECT) { |
1af5bb49 | 1446 | written = generic_file_direct_write(iocb, from); |
84c3d55c | 1447 | if (written < 0 || !iov_iter_count(from)) |
4273b793 | 1448 | goto out; |
64d1b4dd CH |
1449 | written = direct_write_fallback(iocb, from, written, |
1450 | fuse_perform_write(iocb, from)); | |
4273b793 | 1451 | } else { |
596df33d | 1452 | written = fuse_perform_write(iocb, from); |
4273b793 | 1453 | } |
ea9b9907 | 1454 | out: |
5955102c | 1455 | inode_unlock(inode); |
e1c0eecb MS |
1456 | if (written > 0) |
1457 | written = generic_write_sync(iocb, written); | |
ea9b9907 NP |
1458 | |
1459 | return written ? written : err; | |
1460 | } | |
1461 | ||
7c190c8b MP |
1462 | static inline unsigned long fuse_get_user_addr(const struct iov_iter *ii) |
1463 | { | |
de4f5fed | 1464 | return (unsigned long)iter_iov(ii)->iov_base + ii->iov_offset; |
7c190c8b MP |
1465 | } |
1466 | ||
1467 | static inline size_t fuse_get_frag_size(const struct iov_iter *ii, | |
1468 | size_t max_size) | |
1469 | { | |
1470 | return min(iov_iter_single_seg_count(ii), max_size); | |
1471 | } | |
1472 | ||
45ac96ed MS |
1473 | static int fuse_get_user_pages(struct fuse_args_pages *ap, struct iov_iter *ii, |
1474 | size_t *nbytesp, int write, | |
1475 | unsigned int max_pages) | |
413ef8cb | 1476 | { |
7c190c8b | 1477 | size_t nbytes = 0; /* # bytes already packed in req */ |
742f9927 | 1478 | ssize_t ret = 0; |
b98d023a | 1479 | |
f4975c67 | 1480 | /* Special case for kernel I/O: can copy directly into the buffer */ |
00e23707 | 1481 | if (iov_iter_is_kvec(ii)) { |
7c190c8b MP |
1482 | unsigned long user_addr = fuse_get_user_addr(ii); |
1483 | size_t frag_size = fuse_get_frag_size(ii, *nbytesp); | |
1484 | ||
f4975c67 | 1485 | if (write) |
45ac96ed | 1486 | ap->args.in_args[1].value = (void *) user_addr; |
f4975c67 | 1487 | else |
45ac96ed | 1488 | ap->args.out_args[0].value = (void *) user_addr; |
f4975c67 | 1489 | |
b98d023a MP |
1490 | iov_iter_advance(ii, frag_size); |
1491 | *nbytesp = frag_size; | |
f4975c67 MS |
1492 | return 0; |
1493 | } | |
413ef8cb | 1494 | |
45ac96ed | 1495 | while (nbytes < *nbytesp && ap->num_pages < max_pages) { |
7c190c8b | 1496 | unsigned npages; |
f67da30c | 1497 | size_t start; |
1ef255e2 | 1498 | ret = iov_iter_get_pages2(ii, &ap->pages[ap->num_pages], |
2c80929c | 1499 | *nbytesp - nbytes, |
45ac96ed | 1500 | max_pages - ap->num_pages, |
c7f3888a | 1501 | &start); |
7c190c8b | 1502 | if (ret < 0) |
742f9927 | 1503 | break; |
7c190c8b | 1504 | |
c9c37e2e | 1505 | nbytes += ret; |
7c190c8b | 1506 | |
c9c37e2e | 1507 | ret += start; |
6c88632b | 1508 | npages = DIV_ROUND_UP(ret, PAGE_SIZE); |
7c190c8b | 1509 | |
45ac96ed MS |
1510 | ap->descs[ap->num_pages].offset = start; |
1511 | fuse_page_descs_length_init(ap->descs, ap->num_pages, npages); | |
7c190c8b | 1512 | |
45ac96ed MS |
1513 | ap->num_pages += npages; |
1514 | ap->descs[ap->num_pages - 1].length -= | |
c9c37e2e | 1515 | (PAGE_SIZE - ret) & (PAGE_SIZE - 1); |
7c190c8b | 1516 | } |
f4975c67 | 1517 | |
0c4bcfde | 1518 | ap->args.user_pages = true; |
f4975c67 | 1519 | if (write) |
cabdb4fa | 1520 | ap->args.in_pages = true; |
f4975c67 | 1521 | else |
cabdb4fa | 1522 | ap->args.out_pages = true; |
f4975c67 | 1523 | |
7c190c8b | 1524 | *nbytesp = nbytes; |
f4975c67 | 1525 | |
2c932d4c | 1526 | return ret < 0 ? ret : 0; |
413ef8cb MS |
1527 | } |
1528 | ||
d22a943f AV |
1529 | ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter, |
1530 | loff_t *ppos, int flags) | |
413ef8cb | 1531 | { |
ea8cd333 PE |
1532 | int write = flags & FUSE_DIO_WRITE; |
1533 | int cuse = flags & FUSE_DIO_CUSE; | |
e1c0eecb | 1534 | struct file *file = io->iocb->ki_filp; |
80e4f252 HX |
1535 | struct address_space *mapping = file->f_mapping; |
1536 | struct inode *inode = mapping->host; | |
2106cb18 | 1537 | struct fuse_file *ff = file->private_data; |
fcee216b | 1538 | struct fuse_conn *fc = ff->fm->fc; |
413ef8cb MS |
1539 | size_t nmax = write ? fc->max_write : fc->max_read; |
1540 | loff_t pos = *ppos; | |
d22a943f | 1541 | size_t count = iov_iter_count(iter); |
09cbfeaf KS |
1542 | pgoff_t idx_from = pos >> PAGE_SHIFT; |
1543 | pgoff_t idx_to = (pos + count - 1) >> PAGE_SHIFT; | |
413ef8cb | 1544 | ssize_t res = 0; |
742f9927 | 1545 | int err = 0; |
45ac96ed MS |
1546 | struct fuse_io_args *ia; |
1547 | unsigned int max_pages; | |
80e4f252 | 1548 | bool fopen_direct_io = ff->open_flags & FOPEN_DIRECT_IO; |
248d86e8 | 1549 | |
45ac96ed MS |
1550 | max_pages = iov_iter_npages(iter, fc->max_pages); |
1551 | ia = fuse_io_alloc(io, max_pages); | |
1552 | if (!ia) | |
1553 | return -ENOMEM; | |
413ef8cb | 1554 | |
c55e0a55 | 1555 | if (fopen_direct_io && fc->direct_io_allow_mmap) { |
b5a2a3a0 HX |
1556 | res = filemap_write_and_wait_range(mapping, pos, pos + count - 1); |
1557 | if (res) { | |
1558 | fuse_io_free(ia); | |
1559 | return res; | |
1560 | } | |
1561 | } | |
ea8cd333 PE |
1562 | if (!cuse && fuse_range_is_writeback(inode, idx_from, idx_to)) { |
1563 | if (!write) | |
5955102c | 1564 | inode_lock(inode); |
ea8cd333 PE |
1565 | fuse_sync_writes(inode); |
1566 | if (!write) | |
5955102c | 1567 | inode_unlock(inode); |
ea8cd333 PE |
1568 | } |
1569 | ||
80e4f252 HX |
1570 | if (fopen_direct_io && write) { |
1571 | res = invalidate_inode_pages2_range(mapping, idx_from, idx_to); | |
1572 | if (res) { | |
1573 | fuse_io_free(ia); | |
1574 | return res; | |
1575 | } | |
1576 | } | |
1577 | ||
fcb14cb1 | 1578 | io->should_dirty = !write && user_backed_iter(iter); |
413ef8cb | 1579 | while (count) { |
45ac96ed | 1580 | ssize_t nres; |
2106cb18 | 1581 | fl_owner_t owner = current->files; |
f4975c67 | 1582 | size_t nbytes = min(count, nmax); |
45ac96ed MS |
1583 | |
1584 | err = fuse_get_user_pages(&ia->ap, iter, &nbytes, write, | |
1585 | max_pages); | |
742f9927 | 1586 | if (err && !nbytes) |
413ef8cb | 1587 | break; |
f4975c67 | 1588 | |
4a2abf99 | 1589 | if (write) { |
45ac96ed | 1590 | if (!capable(CAP_FSETID)) |
10c52c84 | 1591 | ia->write.in.write_flags |= FUSE_WRITE_KILL_SUIDGID; |
4a2abf99 | 1592 | |
45ac96ed | 1593 | nres = fuse_send_write(ia, pos, nbytes, owner); |
4a2abf99 | 1594 | } else { |
45ac96ed | 1595 | nres = fuse_send_read(ia, pos, nbytes, owner); |
4a2abf99 | 1596 | } |
2106cb18 | 1597 | |
45ac96ed MS |
1598 | if (!io->async || nres < 0) { |
1599 | fuse_release_user_pages(&ia->ap, io->should_dirty); | |
1600 | fuse_io_free(ia); | |
1601 | } | |
1602 | ia = NULL; | |
1603 | if (nres < 0) { | |
f658adee | 1604 | iov_iter_revert(iter, nbytes); |
45ac96ed | 1605 | err = nres; |
413ef8cb MS |
1606 | break; |
1607 | } | |
45ac96ed MS |
1608 | WARN_ON(nres > nbytes); |
1609 | ||
413ef8cb MS |
1610 | count -= nres; |
1611 | res += nres; | |
1612 | pos += nres; | |
f658adee MS |
1613 | if (nres != nbytes) { |
1614 | iov_iter_revert(iter, nbytes - nres); | |
413ef8cb | 1615 | break; |
f658adee | 1616 | } |
56cf34ff | 1617 | if (count) { |
45ac96ed MS |
1618 | max_pages = iov_iter_npages(iter, fc->max_pages); |
1619 | ia = fuse_io_alloc(io, max_pages); | |
1620 | if (!ia) | |
56cf34ff MS |
1621 | break; |
1622 | } | |
413ef8cb | 1623 | } |
45ac96ed MS |
1624 | if (ia) |
1625 | fuse_io_free(ia); | |
d09cb9d7 | 1626 | if (res > 0) |
413ef8cb | 1627 | *ppos = pos; |
413ef8cb | 1628 | |
742f9927 | 1629 | return res > 0 ? res : err; |
413ef8cb | 1630 | } |
08cbf542 | 1631 | EXPORT_SYMBOL_GPL(fuse_direct_io); |
413ef8cb | 1632 | |
36cf66ed | 1633 | static ssize_t __fuse_direct_read(struct fuse_io_priv *io, |
d22a943f AV |
1634 | struct iov_iter *iter, |
1635 | loff_t *ppos) | |
413ef8cb | 1636 | { |
d09cb9d7 | 1637 | ssize_t res; |
e1c0eecb | 1638 | struct inode *inode = file_inode(io->iocb->ki_filp); |
d09cb9d7 | 1639 | |
d22a943f | 1640 | res = fuse_direct_io(io, iter, ppos, 0); |
d09cb9d7 | 1641 | |
9a2eb24d | 1642 | fuse_invalidate_atime(inode); |
d09cb9d7 MS |
1643 | |
1644 | return res; | |
413ef8cb MS |
1645 | } |
1646 | ||
23c94e1c MR |
1647 | static ssize_t fuse_direct_IO(struct kiocb *iocb, struct iov_iter *iter); |
1648 | ||
15316263 | 1649 | static ssize_t fuse_direct_read_iter(struct kiocb *iocb, struct iov_iter *to) |
b98d023a | 1650 | { |
23c94e1c MR |
1651 | ssize_t res; |
1652 | ||
1653 | if (!is_sync_kiocb(iocb) && iocb->ki_flags & IOCB_DIRECT) { | |
23c94e1c MR |
1654 | res = fuse_direct_IO(iocb, to); |
1655 | } else { | |
1656 | struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb); | |
1657 | ||
1658 | res = __fuse_direct_read(&io, to, &iocb->ki_pos); | |
1659 | } | |
1660 | ||
1661 | return res; | |
b98d023a MP |
1662 | } |
1663 | ||
15316263 | 1664 | static ssize_t fuse_direct_write_iter(struct kiocb *iocb, struct iov_iter *from) |
4273b793 | 1665 | { |
e1c0eecb MS |
1666 | struct inode *inode = file_inode(iocb->ki_filp); |
1667 | struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb); | |
15316263 | 1668 | ssize_t res; |
9bbb6717 | 1669 | bool exclusive; |
4273b793 | 1670 | |
9bbb6717 | 1671 | fuse_dio_lock(iocb, from, &exclusive); |
3309dd04 | 1672 | res = generic_write_checks(iocb, from); |
23c94e1c | 1673 | if (res > 0) { |
2e3f7dd0 | 1674 | task_io_account_write(res); |
23c94e1c MR |
1675 | if (!is_sync_kiocb(iocb) && iocb->ki_flags & IOCB_DIRECT) { |
1676 | res = fuse_direct_IO(iocb, from); | |
1677 | } else { | |
1678 | res = fuse_direct_io(&io, from, &iocb->ki_pos, | |
1679 | FUSE_DIO_WRITE); | |
d347739a | 1680 | fuse_write_update_attr(inode, iocb->ki_pos, res); |
23c94e1c MR |
1681 | } |
1682 | } | |
205c1d80 | 1683 | fuse_dio_unlock(iocb, exclusive); |
4273b793 AA |
1684 | |
1685 | return res; | |
1686 | } | |
1687 | ||
55752a3a MS |
1688 | static ssize_t fuse_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
1689 | { | |
2f7b6f5b MS |
1690 | struct file *file = iocb->ki_filp; |
1691 | struct fuse_file *ff = file->private_data; | |
c2d0ad00 | 1692 | struct inode *inode = file_inode(file); |
2f7b6f5b | 1693 | |
5d069dbe | 1694 | if (fuse_is_bad(inode)) |
2f7b6f5b | 1695 | return -EIO; |
55752a3a | 1696 | |
c2d0ad00 VG |
1697 | if (FUSE_IS_DAX(inode)) |
1698 | return fuse_dax_read_iter(iocb, to); | |
1699 | ||
57e1176e AG |
1700 | /* FOPEN_DIRECT_IO overrides FOPEN_PASSTHROUGH */ |
1701 | if (ff->open_flags & FOPEN_DIRECT_IO) | |
55752a3a | 1702 | return fuse_direct_read_iter(iocb, to); |
57e1176e AG |
1703 | else if (fuse_file_passthrough(ff)) |
1704 | return fuse_passthrough_read_iter(iocb, to); | |
1705 | else | |
1706 | return fuse_cache_read_iter(iocb, to); | |
55752a3a MS |
1707 | } |
1708 | ||
1709 | static ssize_t fuse_file_write_iter(struct kiocb *iocb, struct iov_iter *from) | |
1710 | { | |
2f7b6f5b MS |
1711 | struct file *file = iocb->ki_filp; |
1712 | struct fuse_file *ff = file->private_data; | |
c2d0ad00 | 1713 | struct inode *inode = file_inode(file); |
2f7b6f5b | 1714 | |
5d069dbe | 1715 | if (fuse_is_bad(inode)) |
2f7b6f5b | 1716 | return -EIO; |
55752a3a | 1717 | |
c2d0ad00 VG |
1718 | if (FUSE_IS_DAX(inode)) |
1719 | return fuse_dax_write_iter(iocb, from); | |
1720 | ||
57e1176e AG |
1721 | /* FOPEN_DIRECT_IO overrides FOPEN_PASSTHROUGH */ |
1722 | if (ff->open_flags & FOPEN_DIRECT_IO) | |
55752a3a | 1723 | return fuse_direct_write_iter(iocb, from); |
57e1176e AG |
1724 | else if (fuse_file_passthrough(ff)) |
1725 | return fuse_passthrough_write_iter(iocb, from); | |
1726 | else | |
1727 | return fuse_cache_write_iter(iocb, from); | |
55752a3a MS |
1728 | } |
1729 | ||
5ca73468 AG |
1730 | static ssize_t fuse_splice_read(struct file *in, loff_t *ppos, |
1731 | struct pipe_inode_info *pipe, size_t len, | |
1732 | unsigned int flags) | |
1733 | { | |
1734 | struct fuse_file *ff = in->private_data; | |
1735 | ||
1736 | /* FOPEN_DIRECT_IO overrides FOPEN_PASSTHROUGH */ | |
1737 | if (fuse_file_passthrough(ff) && !(ff->open_flags & FOPEN_DIRECT_IO)) | |
1738 | return fuse_passthrough_splice_read(in, ppos, pipe, len, flags); | |
1739 | else | |
1740 | return filemap_splice_read(in, ppos, pipe, len, flags); | |
1741 | } | |
1742 | ||
1743 | static ssize_t fuse_splice_write(struct pipe_inode_info *pipe, struct file *out, | |
1744 | loff_t *ppos, size_t len, unsigned int flags) | |
1745 | { | |
1746 | struct fuse_file *ff = out->private_data; | |
1747 | ||
1748 | /* FOPEN_DIRECT_IO overrides FOPEN_PASSTHROUGH */ | |
1749 | if (fuse_file_passthrough(ff) && !(ff->open_flags & FOPEN_DIRECT_IO)) | |
1750 | return fuse_passthrough_splice_write(pipe, out, ppos, len, flags); | |
1751 | else | |
1752 | return iter_file_splice_write(pipe, out, ppos, len, flags); | |
1753 | } | |
1754 | ||
33826ebb | 1755 | static void fuse_writepage_free(struct fuse_writepage_args *wpa) |
b6aeaded | 1756 | { |
33826ebb | 1757 | struct fuse_args_pages *ap = &wpa->ia.ap; |
385b1268 PE |
1758 | int i; |
1759 | ||
660585b5 MS |
1760 | if (wpa->bucket) |
1761 | fuse_sync_bucket_dec(wpa->bucket); | |
1762 | ||
33826ebb MS |
1763 | for (i = 0; i < ap->num_pages; i++) |
1764 | __free_page(ap->pages[i]); | |
1765 | ||
1766 | if (wpa->ia.ff) | |
e26ee4ef | 1767 | fuse_file_put(wpa->ia.ff, false); |
8b284dc4 | 1768 | |
33826ebb MS |
1769 | kfree(ap->pages); |
1770 | kfree(wpa); | |
3be5a52b MS |
1771 | } |
1772 | ||
fcee216b | 1773 | static void fuse_writepage_finish(struct fuse_mount *fm, |
33826ebb | 1774 | struct fuse_writepage_args *wpa) |
3be5a52b | 1775 | { |
33826ebb MS |
1776 | struct fuse_args_pages *ap = &wpa->ia.ap; |
1777 | struct inode *inode = wpa->inode; | |
3be5a52b | 1778 | struct fuse_inode *fi = get_fuse_inode(inode); |
de1414a6 | 1779 | struct backing_dev_info *bdi = inode_to_bdi(inode); |
385b1268 | 1780 | int i; |
3be5a52b | 1781 | |
33826ebb | 1782 | for (i = 0; i < ap->num_pages; i++) { |
93f78d88 | 1783 | dec_wb_stat(&bdi->wb, WB_WRITEBACK); |
33826ebb | 1784 | dec_node_page_state(ap->pages[i], NR_WRITEBACK_TEMP); |
93f78d88 | 1785 | wb_writeout_inc(&bdi->wb); |
385b1268 | 1786 | } |
3be5a52b MS |
1787 | wake_up(&fi->page_waitq); |
1788 | } | |
1789 | ||
f15ecfef | 1790 | /* Called under fi->lock, may release and reacquire it */ |
fcee216b | 1791 | static void fuse_send_writepage(struct fuse_mount *fm, |
33826ebb | 1792 | struct fuse_writepage_args *wpa, loff_t size) |
f15ecfef KT |
1793 | __releases(fi->lock) |
1794 | __acquires(fi->lock) | |
3be5a52b | 1795 | { |
33826ebb MS |
1796 | struct fuse_writepage_args *aux, *next; |
1797 | struct fuse_inode *fi = get_fuse_inode(wpa->inode); | |
1798 | struct fuse_write_in *inarg = &wpa->ia.write.in; | |
1799 | struct fuse_args *args = &wpa->ia.ap.args; | |
1800 | __u64 data_size = wpa->ia.ap.num_pages * PAGE_SIZE; | |
1801 | int err; | |
3be5a52b | 1802 | |
33826ebb | 1803 | fi->writectr++; |
385b1268 PE |
1804 | if (inarg->offset + data_size <= size) { |
1805 | inarg->size = data_size; | |
3be5a52b | 1806 | } else if (inarg->offset < size) { |
385b1268 | 1807 | inarg->size = size - inarg->offset; |
3be5a52b MS |
1808 | } else { |
1809 | /* Got truncated off completely */ | |
1810 | goto out_free; | |
b6aeaded | 1811 | } |
3be5a52b | 1812 | |
33826ebb MS |
1813 | args->in_args[1].size = inarg->size; |
1814 | args->force = true; | |
1815 | args->nocreds = true; | |
1816 | ||
fcee216b | 1817 | err = fuse_simple_background(fm, args, GFP_ATOMIC); |
33826ebb MS |
1818 | if (err == -ENOMEM) { |
1819 | spin_unlock(&fi->lock); | |
fcee216b | 1820 | err = fuse_simple_background(fm, args, GFP_NOFS | __GFP_NOFAIL); |
33826ebb MS |
1821 | spin_lock(&fi->lock); |
1822 | } | |
1823 | ||
f15ecfef | 1824 | /* Fails on broken connection only */ |
33826ebb | 1825 | if (unlikely(err)) |
f15ecfef KT |
1826 | goto out_free; |
1827 | ||
3be5a52b MS |
1828 | return; |
1829 | ||
1830 | out_free: | |
33826ebb | 1831 | fi->writectr--; |
69a6487a | 1832 | rb_erase(&wpa->writepages_entry, &fi->writepages); |
fcee216b | 1833 | fuse_writepage_finish(fm, wpa); |
f15ecfef | 1834 | spin_unlock(&fi->lock); |
e2653bd5 MS |
1835 | |
1836 | /* After fuse_writepage_finish() aux request list is private */ | |
33826ebb MS |
1837 | for (aux = wpa->next; aux; aux = next) { |
1838 | next = aux->next; | |
1839 | aux->next = NULL; | |
1840 | fuse_writepage_free(aux); | |
e2653bd5 MS |
1841 | } |
1842 | ||
33826ebb | 1843 | fuse_writepage_free(wpa); |
f15ecfef | 1844 | spin_lock(&fi->lock); |
b6aeaded MS |
1845 | } |
1846 | ||
3be5a52b MS |
1847 | /* |
1848 | * If fi->writectr is positive (no truncate or fsync going on) send | |
1849 | * all queued writepage requests. | |
1850 | * | |
f15ecfef | 1851 | * Called with fi->lock |
3be5a52b MS |
1852 | */ |
1853 | void fuse_flush_writepages(struct inode *inode) | |
f15ecfef KT |
1854 | __releases(fi->lock) |
1855 | __acquires(fi->lock) | |
b6aeaded | 1856 | { |
fcee216b | 1857 | struct fuse_mount *fm = get_fuse_mount(inode); |
3be5a52b | 1858 | struct fuse_inode *fi = get_fuse_inode(inode); |
9de5be06 | 1859 | loff_t crop = i_size_read(inode); |
33826ebb | 1860 | struct fuse_writepage_args *wpa; |
3be5a52b MS |
1861 | |
1862 | while (fi->writectr >= 0 && !list_empty(&fi->queued_writes)) { | |
33826ebb MS |
1863 | wpa = list_entry(fi->queued_writes.next, |
1864 | struct fuse_writepage_args, queue_entry); | |
1865 | list_del_init(&wpa->queue_entry); | |
fcee216b | 1866 | fuse_send_writepage(fm, wpa, crop); |
3be5a52b MS |
1867 | } |
1868 | } | |
1869 | ||
c146024e MS |
1870 | static struct fuse_writepage_args *fuse_insert_writeback(struct rb_root *root, |
1871 | struct fuse_writepage_args *wpa) | |
6b2fb799 MP |
1872 | { |
1873 | pgoff_t idx_from = wpa->ia.write.in.offset >> PAGE_SHIFT; | |
1874 | pgoff_t idx_to = idx_from + wpa->ia.ap.num_pages - 1; | |
1875 | struct rb_node **p = &root->rb_node; | |
1876 | struct rb_node *parent = NULL; | |
1877 | ||
1878 | WARN_ON(!wpa->ia.ap.num_pages); | |
1879 | while (*p) { | |
1880 | struct fuse_writepage_args *curr; | |
1881 | pgoff_t curr_index; | |
1882 | ||
1883 | parent = *p; | |
1884 | curr = rb_entry(parent, struct fuse_writepage_args, | |
1885 | writepages_entry); | |
1886 | WARN_ON(curr->inode != wpa->inode); | |
1887 | curr_index = curr->ia.write.in.offset >> PAGE_SHIFT; | |
1888 | ||
1889 | if (idx_from >= curr_index + curr->ia.ap.num_pages) | |
1890 | p = &(*p)->rb_right; | |
1891 | else if (idx_to < curr_index) | |
1892 | p = &(*p)->rb_left; | |
1893 | else | |
c146024e | 1894 | return curr; |
6b2fb799 MP |
1895 | } |
1896 | ||
1897 | rb_link_node(&wpa->writepages_entry, parent, p); | |
1898 | rb_insert_color(&wpa->writepages_entry, root); | |
c146024e MS |
1899 | return NULL; |
1900 | } | |
1901 | ||
1902 | static void tree_insert(struct rb_root *root, struct fuse_writepage_args *wpa) | |
1903 | { | |
1904 | WARN_ON(fuse_insert_writeback(root, wpa)); | |
6b2fb799 MP |
1905 | } |
1906 | ||
fcee216b | 1907 | static void fuse_writepage_end(struct fuse_mount *fm, struct fuse_args *args, |
33826ebb | 1908 | int error) |
3be5a52b | 1909 | { |
33826ebb MS |
1910 | struct fuse_writepage_args *wpa = |
1911 | container_of(args, typeof(*wpa), ia.ap.args); | |
1912 | struct inode *inode = wpa->inode; | |
3be5a52b | 1913 | struct fuse_inode *fi = get_fuse_inode(inode); |
3466958b | 1914 | struct fuse_conn *fc = get_fuse_conn(inode); |
3be5a52b | 1915 | |
33826ebb | 1916 | mapping_set_error(inode->i_mapping, error); |
3466958b VG |
1917 | /* |
1918 | * A writeback finished and this might have updated mtime/ctime on | |
1919 | * server making local mtime/ctime stale. Hence invalidate attrs. | |
1920 | * Do this only if writeback_cache is not enabled. If writeback_cache | |
1921 | * is enabled, we trust local ctime/mtime. | |
1922 | */ | |
1923 | if (!fc->writeback_cache) | |
fa5eee57 | 1924 | fuse_invalidate_attr_mask(inode, FUSE_STATX_MODIFY); |
f15ecfef | 1925 | spin_lock(&fi->lock); |
69a6487a | 1926 | rb_erase(&wpa->writepages_entry, &fi->writepages); |
33826ebb | 1927 | while (wpa->next) { |
fcee216b | 1928 | struct fuse_mount *fm = get_fuse_mount(inode); |
33826ebb MS |
1929 | struct fuse_write_in *inarg = &wpa->ia.write.in; |
1930 | struct fuse_writepage_args *next = wpa->next; | |
1931 | ||
1932 | wpa->next = next->next; | |
1933 | next->next = NULL; | |
1934 | next->ia.ff = fuse_file_get(wpa->ia.ff); | |
6b2fb799 | 1935 | tree_insert(&fi->writepages, next); |
6eaf4782 MP |
1936 | |
1937 | /* | |
1938 | * Skip fuse_flush_writepages() to make it easy to crop requests | |
1939 | * based on primary request size. | |
1940 | * | |
1941 | * 1st case (trivial): there are no concurrent activities using | |
1942 | * fuse_set/release_nowrite. Then we're on safe side because | |
1943 | * fuse_flush_writepages() would call fuse_send_writepage() | |
1944 | * anyway. | |
1945 | * | |
1946 | * 2nd case: someone called fuse_set_nowrite and it is waiting | |
1947 | * now for completion of all in-flight requests. This happens | |
1948 | * rarely and no more than once per page, so this should be | |
1949 | * okay. | |
1950 | * | |
1951 | * 3rd case: someone (e.g. fuse_do_setattr()) is in the middle | |
1952 | * of fuse_set_nowrite..fuse_release_nowrite section. The fact | |
1953 | * that fuse_set_nowrite returned implies that all in-flight | |
1954 | * requests were completed along with all of their secondary | |
1955 | * requests. Further primary requests are blocked by negative | |
1956 | * writectr. Hence there cannot be any in-flight requests and | |
1957 | * no invocations of fuse_writepage_end() while we're in | |
1958 | * fuse_set_nowrite..fuse_release_nowrite section. | |
1959 | */ | |
fcee216b | 1960 | fuse_send_writepage(fm, next, inarg->offset + inarg->size); |
8b284dc4 | 1961 | } |
3be5a52b | 1962 | fi->writectr--; |
fcee216b | 1963 | fuse_writepage_finish(fm, wpa); |
f15ecfef | 1964 | spin_unlock(&fi->lock); |
33826ebb | 1965 | fuse_writepage_free(wpa); |
3be5a52b MS |
1966 | } |
1967 | ||
a9667ac8 | 1968 | static struct fuse_file *__fuse_write_file_get(struct fuse_inode *fi) |
adcadfa8 | 1969 | { |
84840efc | 1970 | struct fuse_file *ff; |
adcadfa8 | 1971 | |
f15ecfef | 1972 | spin_lock(&fi->lock); |
84840efc MS |
1973 | ff = list_first_entry_or_null(&fi->write_files, struct fuse_file, |
1974 | write_entry); | |
1975 | if (ff) | |
72523425 | 1976 | fuse_file_get(ff); |
f15ecfef | 1977 | spin_unlock(&fi->lock); |
adcadfa8 PE |
1978 | |
1979 | return ff; | |
1980 | } | |
1981 | ||
a9667ac8 | 1982 | static struct fuse_file *fuse_write_file_get(struct fuse_inode *fi) |
1e18bda8 | 1983 | { |
a9667ac8 | 1984 | struct fuse_file *ff = __fuse_write_file_get(fi); |
1e18bda8 MS |
1985 | WARN_ON(!ff); |
1986 | return ff; | |
1987 | } | |
1988 | ||
1989 | int fuse_write_inode(struct inode *inode, struct writeback_control *wbc) | |
1990 | { | |
1e18bda8 MS |
1991 | struct fuse_inode *fi = get_fuse_inode(inode); |
1992 | struct fuse_file *ff; | |
1993 | int err; | |
1994 | ||
5c791fe1 MS |
1995 | /* |
1996 | * Inode is always written before the last reference is dropped and | |
1997 | * hence this should not be reached from reclaim. | |
1998 | * | |
1999 | * Writing back the inode from reclaim can deadlock if the request | |
2000 | * processing itself needs an allocation. Allocations triggering | |
2001 | * reclaim while serving a request can't be prevented, because it can | |
2002 | * involve any number of unrelated userspace processes. | |
2003 | */ | |
2004 | WARN_ON(wbc->for_reclaim); | |
2005 | ||
a9667ac8 | 2006 | ff = __fuse_write_file_get(fi); |
ab9e13f7 | 2007 | err = fuse_flush_times(inode, ff); |
1e18bda8 | 2008 | if (ff) |
e26ee4ef | 2009 | fuse_file_put(ff, false); |
1e18bda8 MS |
2010 | |
2011 | return err; | |
2012 | } | |
2013 | ||
33826ebb MS |
2014 | static struct fuse_writepage_args *fuse_writepage_args_alloc(void) |
2015 | { | |
2016 | struct fuse_writepage_args *wpa; | |
2017 | struct fuse_args_pages *ap; | |
2018 | ||
2019 | wpa = kzalloc(sizeof(*wpa), GFP_NOFS); | |
2020 | if (wpa) { | |
2021 | ap = &wpa->ia.ap; | |
2022 | ap->num_pages = 0; | |
2023 | ap->pages = fuse_pages_alloc(1, GFP_NOFS, &ap->descs); | |
2024 | if (!ap->pages) { | |
2025 | kfree(wpa); | |
2026 | wpa = NULL; | |
2027 | } | |
2028 | } | |
2029 | return wpa; | |
2030 | ||
2031 | } | |
2032 | ||
660585b5 MS |
2033 | static void fuse_writepage_add_to_bucket(struct fuse_conn *fc, |
2034 | struct fuse_writepage_args *wpa) | |
2035 | { | |
2036 | if (!fc->sync_fs) | |
2037 | return; | |
2038 | ||
2039 | rcu_read_lock(); | |
2040 | /* Prevent resurrection of dead bucket in unlikely race with syncfs */ | |
2041 | do { | |
2042 | wpa->bucket = rcu_dereference(fc->curr_bucket); | |
2043 | } while (unlikely(!atomic_inc_not_zero(&wpa->bucket->count))); | |
2044 | rcu_read_unlock(); | |
2045 | } | |
2046 | ||
e0887e09 | 2047 | static int fuse_writepage_locked(struct folio *folio) |
3be5a52b | 2048 | { |
e0887e09 | 2049 | struct address_space *mapping = folio->mapping; |
3be5a52b MS |
2050 | struct inode *inode = mapping->host; |
2051 | struct fuse_conn *fc = get_fuse_conn(inode); | |
2052 | struct fuse_inode *fi = get_fuse_inode(inode); | |
33826ebb MS |
2053 | struct fuse_writepage_args *wpa; |
2054 | struct fuse_args_pages *ap; | |
e0887e09 | 2055 | struct folio *tmp_folio; |
72523425 | 2056 | int error = -ENOMEM; |
3be5a52b | 2057 | |
e0887e09 | 2058 | folio_start_writeback(folio); |
3be5a52b | 2059 | |
33826ebb MS |
2060 | wpa = fuse_writepage_args_alloc(); |
2061 | if (!wpa) | |
3be5a52b | 2062 | goto err; |
33826ebb | 2063 | ap = &wpa->ia.ap; |
3be5a52b | 2064 | |
e0887e09 MWO |
2065 | tmp_folio = folio_alloc(GFP_NOFS | __GFP_HIGHMEM, 0); |
2066 | if (!tmp_folio) | |
3be5a52b MS |
2067 | goto err_free; |
2068 | ||
72523425 | 2069 | error = -EIO; |
a9667ac8 | 2070 | wpa->ia.ff = fuse_write_file_get(fi); |
33826ebb | 2071 | if (!wpa->ia.ff) |
27f1b363 | 2072 | goto err_nofile; |
72523425 | 2073 | |
660585b5 | 2074 | fuse_writepage_add_to_bucket(fc, wpa); |
e0887e09 | 2075 | fuse_write_args_fill(&wpa->ia, wpa->ia.ff, folio_pos(folio), 0); |
3be5a52b | 2076 | |
e0887e09 | 2077 | folio_copy(tmp_folio, folio); |
33826ebb MS |
2078 | wpa->ia.write.in.write_flags |= FUSE_WRITE_CACHE; |
2079 | wpa->next = NULL; | |
2080 | ap->args.in_pages = true; | |
2081 | ap->num_pages = 1; | |
e0887e09 | 2082 | ap->pages[0] = &tmp_folio->page; |
33826ebb MS |
2083 | ap->descs[0].offset = 0; |
2084 | ap->descs[0].length = PAGE_SIZE; | |
2085 | ap->args.end = fuse_writepage_end; | |
2086 | wpa->inode = inode; | |
3be5a52b | 2087 | |
93f78d88 | 2088 | inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK); |
e0887e09 | 2089 | node_stat_add_folio(tmp_folio, NR_WRITEBACK_TEMP); |
3be5a52b | 2090 | |
f15ecfef | 2091 | spin_lock(&fi->lock); |
6b2fb799 | 2092 | tree_insert(&fi->writepages, wpa); |
33826ebb | 2093 | list_add_tail(&wpa->queue_entry, &fi->queued_writes); |
3be5a52b | 2094 | fuse_flush_writepages(inode); |
f15ecfef | 2095 | spin_unlock(&fi->lock); |
3be5a52b | 2096 | |
e0887e09 | 2097 | folio_end_writeback(folio); |
4a4ac4eb | 2098 | |
3be5a52b MS |
2099 | return 0; |
2100 | ||
27f1b363 | 2101 | err_nofile: |
e0887e09 | 2102 | folio_put(tmp_folio); |
3be5a52b | 2103 | err_free: |
33826ebb | 2104 | kfree(wpa); |
3be5a52b | 2105 | err: |
e0887e09 MWO |
2106 | mapping_set_error(folio->mapping, error); |
2107 | folio_end_writeback(folio); | |
72523425 | 2108 | return error; |
3be5a52b MS |
2109 | } |
2110 | ||
26d614df | 2111 | struct fuse_fill_wb_data { |
33826ebb | 2112 | struct fuse_writepage_args *wpa; |
26d614df PE |
2113 | struct fuse_file *ff; |
2114 | struct inode *inode; | |
2d033eaa | 2115 | struct page **orig_pages; |
33826ebb | 2116 | unsigned int max_pages; |
26d614df PE |
2117 | }; |
2118 | ||
33826ebb MS |
2119 | static bool fuse_pages_realloc(struct fuse_fill_wb_data *data) |
2120 | { | |
2121 | struct fuse_args_pages *ap = &data->wpa->ia.ap; | |
2122 | struct fuse_conn *fc = get_fuse_conn(data->inode); | |
2123 | struct page **pages; | |
2124 | struct fuse_page_desc *descs; | |
2125 | unsigned int npages = min_t(unsigned int, | |
2126 | max_t(unsigned int, data->max_pages * 2, | |
2127 | FUSE_DEFAULT_MAX_PAGES_PER_REQ), | |
2128 | fc->max_pages); | |
2129 | WARN_ON(npages <= data->max_pages); | |
2130 | ||
2131 | pages = fuse_pages_alloc(npages, GFP_NOFS, &descs); | |
2132 | if (!pages) | |
2133 | return false; | |
2134 | ||
2135 | memcpy(pages, ap->pages, sizeof(struct page *) * ap->num_pages); | |
2136 | memcpy(descs, ap->descs, sizeof(struct fuse_page_desc) * ap->num_pages); | |
2137 | kfree(ap->pages); | |
2138 | ap->pages = pages; | |
2139 | ap->descs = descs; | |
2140 | data->max_pages = npages; | |
2141 | ||
2142 | return true; | |
2143 | } | |
2144 | ||
26d614df PE |
2145 | static void fuse_writepages_send(struct fuse_fill_wb_data *data) |
2146 | { | |
33826ebb | 2147 | struct fuse_writepage_args *wpa = data->wpa; |
26d614df | 2148 | struct inode *inode = data->inode; |
26d614df | 2149 | struct fuse_inode *fi = get_fuse_inode(inode); |
33826ebb | 2150 | int num_pages = wpa->ia.ap.num_pages; |
2d033eaa | 2151 | int i; |
26d614df | 2152 | |
33826ebb | 2153 | wpa->ia.ff = fuse_file_get(data->ff); |
f15ecfef | 2154 | spin_lock(&fi->lock); |
33826ebb | 2155 | list_add_tail(&wpa->queue_entry, &fi->queued_writes); |
26d614df | 2156 | fuse_flush_writepages(inode); |
f15ecfef | 2157 | spin_unlock(&fi->lock); |
2d033eaa MP |
2158 | |
2159 | for (i = 0; i < num_pages; i++) | |
2160 | end_page_writeback(data->orig_pages[i]); | |
26d614df PE |
2161 | } |
2162 | ||
7f305ca1 | 2163 | /* |
c146024e MS |
2164 | * Check under fi->lock if the page is under writeback, and insert it onto the |
2165 | * rb_tree if not. Otherwise iterate auxiliary write requests, to see if there's | |
419234d5 MS |
2166 | * one already added for a page at this offset. If there's none, then insert |
2167 | * this new request onto the auxiliary list, otherwise reuse the existing one by | |
c146024e | 2168 | * swapping the new temp page with the old one. |
7f305ca1 | 2169 | */ |
c146024e MS |
2170 | static bool fuse_writepage_add(struct fuse_writepage_args *new_wpa, |
2171 | struct page *page) | |
8b284dc4 | 2172 | { |
33826ebb MS |
2173 | struct fuse_inode *fi = get_fuse_inode(new_wpa->inode); |
2174 | struct fuse_writepage_args *tmp; | |
2175 | struct fuse_writepage_args *old_wpa; | |
2176 | struct fuse_args_pages *new_ap = &new_wpa->ia.ap; | |
8b284dc4 | 2177 | |
33826ebb | 2178 | WARN_ON(new_ap->num_pages != 0); |
c146024e | 2179 | new_ap->num_pages = 1; |
8b284dc4 | 2180 | |
f15ecfef | 2181 | spin_lock(&fi->lock); |
c146024e | 2182 | old_wpa = fuse_insert_writeback(&fi->writepages, new_wpa); |
33826ebb | 2183 | if (!old_wpa) { |
f15ecfef | 2184 | spin_unlock(&fi->lock); |
c146024e | 2185 | return true; |
f6011081 | 2186 | } |
8b284dc4 | 2187 | |
33826ebb | 2188 | for (tmp = old_wpa->next; tmp; tmp = tmp->next) { |
7f305ca1 MS |
2189 | pgoff_t curr_index; |
2190 | ||
33826ebb MS |
2191 | WARN_ON(tmp->inode != new_wpa->inode); |
2192 | curr_index = tmp->ia.write.in.offset >> PAGE_SHIFT; | |
419234d5 | 2193 | if (curr_index == page->index) { |
33826ebb MS |
2194 | WARN_ON(tmp->ia.ap.num_pages != 1); |
2195 | swap(tmp->ia.ap.pages[0], new_ap->pages[0]); | |
7f305ca1 | 2196 | break; |
8b284dc4 MS |
2197 | } |
2198 | } | |
2199 | ||
7f305ca1 | 2200 | if (!tmp) { |
33826ebb MS |
2201 | new_wpa->next = old_wpa->next; |
2202 | old_wpa->next = new_wpa; | |
7f305ca1 | 2203 | } |
41b6e41f | 2204 | |
f15ecfef | 2205 | spin_unlock(&fi->lock); |
7f305ca1 MS |
2206 | |
2207 | if (tmp) { | |
33826ebb | 2208 | struct backing_dev_info *bdi = inode_to_bdi(new_wpa->inode); |
8b284dc4 | 2209 | |
93f78d88 | 2210 | dec_wb_stat(&bdi->wb, WB_WRITEBACK); |
33826ebb | 2211 | dec_node_page_state(new_ap->pages[0], NR_WRITEBACK_TEMP); |
93f78d88 | 2212 | wb_writeout_inc(&bdi->wb); |
33826ebb | 2213 | fuse_writepage_free(new_wpa); |
8b284dc4 | 2214 | } |
7f305ca1 | 2215 | |
c146024e | 2216 | return false; |
8b284dc4 MS |
2217 | } |
2218 | ||
6ddf3af9 MS |
2219 | static bool fuse_writepage_need_send(struct fuse_conn *fc, struct page *page, |
2220 | struct fuse_args_pages *ap, | |
2221 | struct fuse_fill_wb_data *data) | |
2222 | { | |
2223 | WARN_ON(!ap->num_pages); | |
2224 | ||
2225 | /* | |
2226 | * Being under writeback is unlikely but possible. For example direct | |
2227 | * read to an mmaped fuse file will set the page dirty twice; once when | |
2228 | * the pages are faulted with get_user_pages(), and then after the read | |
2229 | * completed. | |
2230 | */ | |
2231 | if (fuse_page_is_writeback(data->inode, page->index)) | |
2232 | return true; | |
2233 | ||
2234 | /* Reached max pages */ | |
2235 | if (ap->num_pages == fc->max_pages) | |
2236 | return true; | |
2237 | ||
2238 | /* Reached max write bytes */ | |
2239 | if ((ap->num_pages + 1) * PAGE_SIZE > fc->max_write) | |
2240 | return true; | |
2241 | ||
2242 | /* Discontinuity */ | |
2243 | if (data->orig_pages[ap->num_pages - 1]->index + 1 != page->index) | |
2244 | return true; | |
2245 | ||
2246 | /* Need to grow the pages array? If so, did the expansion fail? */ | |
2247 | if (ap->num_pages == data->max_pages && !fuse_pages_realloc(data)) | |
2248 | return true; | |
2249 | ||
2250 | return false; | |
2251 | } | |
2252 | ||
d585bdbe | 2253 | static int fuse_writepages_fill(struct folio *folio, |
26d614df PE |
2254 | struct writeback_control *wbc, void *_data) |
2255 | { | |
2256 | struct fuse_fill_wb_data *data = _data; | |
33826ebb MS |
2257 | struct fuse_writepage_args *wpa = data->wpa; |
2258 | struct fuse_args_pages *ap = &wpa->ia.ap; | |
26d614df | 2259 | struct inode *inode = data->inode; |
f15ecfef | 2260 | struct fuse_inode *fi = get_fuse_inode(inode); |
26d614df PE |
2261 | struct fuse_conn *fc = get_fuse_conn(inode); |
2262 | struct page *tmp_page; | |
2263 | int err; | |
2264 | ||
2265 | if (!data->ff) { | |
2266 | err = -EIO; | |
a9667ac8 | 2267 | data->ff = fuse_write_file_get(fi); |
26d614df PE |
2268 | if (!data->ff) |
2269 | goto out_unlock; | |
2270 | } | |
2271 | ||
d585bdbe | 2272 | if (wpa && fuse_writepage_need_send(fc, &folio->page, ap, data)) { |
8b284dc4 | 2273 | fuse_writepages_send(data); |
33826ebb | 2274 | data->wpa = NULL; |
26d614df | 2275 | } |
e52a8250 | 2276 | |
26d614df PE |
2277 | err = -ENOMEM; |
2278 | tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
2279 | if (!tmp_page) | |
2280 | goto out_unlock; | |
2281 | ||
2282 | /* | |
2283 | * The page must not be redirtied until the writeout is completed | |
2284 | * (i.e. userspace has sent a reply to the write request). Otherwise | |
2285 | * there could be more than one temporary page instance for each real | |
2286 | * page. | |
2287 | * | |
2288 | * This is ensured by holding the page lock in page_mkwrite() while | |
2289 | * checking fuse_page_is_writeback(). We already hold the page lock | |
2290 | * since clear_page_dirty_for_io() and keep it held until we add the | |
33826ebb | 2291 | * request to the fi->writepages list and increment ap->num_pages. |
26d614df PE |
2292 | * After this fuse_page_is_writeback() will indicate that the page is |
2293 | * under writeback, so we can release the page lock. | |
2294 | */ | |
33826ebb | 2295 | if (data->wpa == NULL) { |
26d614df | 2296 | err = -ENOMEM; |
33826ebb MS |
2297 | wpa = fuse_writepage_args_alloc(); |
2298 | if (!wpa) { | |
26d614df PE |
2299 | __free_page(tmp_page); |
2300 | goto out_unlock; | |
2301 | } | |
660585b5 MS |
2302 | fuse_writepage_add_to_bucket(fc, wpa); |
2303 | ||
33826ebb | 2304 | data->max_pages = 1; |
26d614df | 2305 | |
33826ebb | 2306 | ap = &wpa->ia.ap; |
d585bdbe | 2307 | fuse_write_args_fill(&wpa->ia, data->ff, folio_pos(folio), 0); |
33826ebb MS |
2308 | wpa->ia.write.in.write_flags |= FUSE_WRITE_CACHE; |
2309 | wpa->next = NULL; | |
2310 | ap->args.in_pages = true; | |
2311 | ap->args.end = fuse_writepage_end; | |
2312 | ap->num_pages = 0; | |
2313 | wpa->inode = inode; | |
26d614df | 2314 | } |
d585bdbe | 2315 | folio_start_writeback(folio); |
26d614df | 2316 | |
d585bdbe | 2317 | copy_highpage(tmp_page, &folio->page); |
33826ebb MS |
2318 | ap->pages[ap->num_pages] = tmp_page; |
2319 | ap->descs[ap->num_pages].offset = 0; | |
2320 | ap->descs[ap->num_pages].length = PAGE_SIZE; | |
d585bdbe | 2321 | data->orig_pages[ap->num_pages] = &folio->page; |
26d614df | 2322 | |
93f78d88 | 2323 | inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK); |
11fb9989 | 2324 | inc_node_page_state(tmp_page, NR_WRITEBACK_TEMP); |
8b284dc4 MS |
2325 | |
2326 | err = 0; | |
c146024e MS |
2327 | if (data->wpa) { |
2328 | /* | |
2329 | * Protected by fi->lock against concurrent access by | |
2330 | * fuse_page_is_writeback(). | |
2331 | */ | |
2332 | spin_lock(&fi->lock); | |
2333 | ap->num_pages++; | |
2334 | spin_unlock(&fi->lock); | |
d585bdbe | 2335 | } else if (fuse_writepage_add(wpa, &folio->page)) { |
c146024e MS |
2336 | data->wpa = wpa; |
2337 | } else { | |
d585bdbe | 2338 | folio_end_writeback(folio); |
8b284dc4 | 2339 | } |
26d614df | 2340 | out_unlock: |
d585bdbe | 2341 | folio_unlock(folio); |
26d614df PE |
2342 | |
2343 | return err; | |
2344 | } | |
2345 | ||
2346 | static int fuse_writepages(struct address_space *mapping, | |
2347 | struct writeback_control *wbc) | |
2348 | { | |
2349 | struct inode *inode = mapping->host; | |
5da784cc | 2350 | struct fuse_conn *fc = get_fuse_conn(inode); |
26d614df PE |
2351 | struct fuse_fill_wb_data data; |
2352 | int err; | |
2353 | ||
2354 | err = -EIO; | |
5d069dbe | 2355 | if (fuse_is_bad(inode)) |
26d614df PE |
2356 | goto out; |
2357 | ||
670d21c6 N |
2358 | if (wbc->sync_mode == WB_SYNC_NONE && |
2359 | fc->num_background >= fc->congestion_threshold) | |
2360 | return 0; | |
2361 | ||
26d614df | 2362 | data.inode = inode; |
33826ebb | 2363 | data.wpa = NULL; |
26d614df PE |
2364 | data.ff = NULL; |
2365 | ||
2d033eaa | 2366 | err = -ENOMEM; |
5da784cc | 2367 | data.orig_pages = kcalloc(fc->max_pages, |
f2b3455e | 2368 | sizeof(struct page *), |
2d033eaa MP |
2369 | GFP_NOFS); |
2370 | if (!data.orig_pages) | |
2371 | goto out; | |
2372 | ||
26d614df | 2373 | err = write_cache_pages(mapping, wbc, fuse_writepages_fill, &data); |
33826ebb | 2374 | if (data.wpa) { |
33826ebb | 2375 | WARN_ON(!data.wpa->ia.ap.num_pages); |
26d614df | 2376 | fuse_writepages_send(&data); |
26d614df PE |
2377 | } |
2378 | if (data.ff) | |
e26ee4ef | 2379 | fuse_file_put(data.ff, false); |
2d033eaa MP |
2380 | |
2381 | kfree(data.orig_pages); | |
26d614df PE |
2382 | out: |
2383 | return err; | |
2384 | } | |
2385 | ||
6b12c1b3 PE |
2386 | /* |
2387 | * It's worthy to make sure that space is reserved on disk for the write, | |
2388 | * but how to implement it without killing performance need more thinking. | |
2389 | */ | |
2390 | static int fuse_write_begin(struct file *file, struct address_space *mapping, | |
9d6b0cd7 | 2391 | loff_t pos, unsigned len, struct page **pagep, void **fsdata) |
6b12c1b3 | 2392 | { |
09cbfeaf | 2393 | pgoff_t index = pos >> PAGE_SHIFT; |
a455589f | 2394 | struct fuse_conn *fc = get_fuse_conn(file_inode(file)); |
6b12c1b3 PE |
2395 | struct page *page; |
2396 | loff_t fsize; | |
2397 | int err = -ENOMEM; | |
2398 | ||
2399 | WARN_ON(!fc->writeback_cache); | |
2400 | ||
b7446e7c | 2401 | page = grab_cache_page_write_begin(mapping, index); |
6b12c1b3 PE |
2402 | if (!page) |
2403 | goto error; | |
2404 | ||
2405 | fuse_wait_on_page_writeback(mapping->host, page->index); | |
2406 | ||
09cbfeaf | 2407 | if (PageUptodate(page) || len == PAGE_SIZE) |
6b12c1b3 PE |
2408 | goto success; |
2409 | /* | |
2410 | * Check if the start this page comes after the end of file, in which | |
2411 | * case the readpage can be optimized away. | |
2412 | */ | |
2413 | fsize = i_size_read(mapping->host); | |
09cbfeaf KS |
2414 | if (fsize <= (pos & PAGE_MASK)) { |
2415 | size_t off = pos & ~PAGE_MASK; | |
6b12c1b3 PE |
2416 | if (off) |
2417 | zero_user_segment(page, 0, off); | |
2418 | goto success; | |
2419 | } | |
2420 | err = fuse_do_readpage(file, page); | |
2421 | if (err) | |
2422 | goto cleanup; | |
2423 | success: | |
2424 | *pagep = page; | |
2425 | return 0; | |
2426 | ||
2427 | cleanup: | |
2428 | unlock_page(page); | |
09cbfeaf | 2429 | put_page(page); |
6b12c1b3 PE |
2430 | error: |
2431 | return err; | |
2432 | } | |
2433 | ||
2434 | static int fuse_write_end(struct file *file, struct address_space *mapping, | |
2435 | loff_t pos, unsigned len, unsigned copied, | |
2436 | struct page *page, void *fsdata) | |
2437 | { | |
2438 | struct inode *inode = page->mapping->host; | |
2439 | ||
59c3b76c MS |
2440 | /* Haven't copied anything? Skip zeroing, size extending, dirtying. */ |
2441 | if (!copied) | |
2442 | goto unlock; | |
2443 | ||
8c56e03d | 2444 | pos += copied; |
6b12c1b3 PE |
2445 | if (!PageUptodate(page)) { |
2446 | /* Zero any unwritten bytes at the end of the page */ | |
8c56e03d | 2447 | size_t endoff = pos & ~PAGE_MASK; |
6b12c1b3 | 2448 | if (endoff) |
09cbfeaf | 2449 | zero_user_segment(page, endoff, PAGE_SIZE); |
6b12c1b3 PE |
2450 | SetPageUptodate(page); |
2451 | } | |
2452 | ||
8c56e03d MS |
2453 | if (pos > inode->i_size) |
2454 | i_size_write(inode, pos); | |
2455 | ||
6b12c1b3 | 2456 | set_page_dirty(page); |
59c3b76c MS |
2457 | |
2458 | unlock: | |
6b12c1b3 | 2459 | unlock_page(page); |
09cbfeaf | 2460 | put_page(page); |
6b12c1b3 PE |
2461 | |
2462 | return copied; | |
2463 | } | |
2464 | ||
2bf06b8e | 2465 | static int fuse_launder_folio(struct folio *folio) |
3be5a52b MS |
2466 | { |
2467 | int err = 0; | |
2bf06b8e MWO |
2468 | if (folio_clear_dirty_for_io(folio)) { |
2469 | struct inode *inode = folio->mapping->host; | |
3993382b MS |
2470 | |
2471 | /* Serialize with pending writeback for the same page */ | |
2bf06b8e | 2472 | fuse_wait_on_page_writeback(inode, folio->index); |
e0887e09 | 2473 | err = fuse_writepage_locked(folio); |
3be5a52b | 2474 | if (!err) |
2bf06b8e | 2475 | fuse_wait_on_page_writeback(inode, folio->index); |
3be5a52b MS |
2476 | } |
2477 | return err; | |
2478 | } | |
2479 | ||
2480 | /* | |
36ea2337 MS |
2481 | * Write back dirty data/metadata now (there may not be any suitable |
2482 | * open files later for data) | |
3be5a52b MS |
2483 | */ |
2484 | static void fuse_vma_close(struct vm_area_struct *vma) | |
2485 | { | |
36ea2337 MS |
2486 | int err; |
2487 | ||
2488 | err = write_inode_now(vma->vm_file->f_mapping->host, 1); | |
2489 | mapping_set_error(vma->vm_file->f_mapping, err); | |
3be5a52b MS |
2490 | } |
2491 | ||
2492 | /* | |
2493 | * Wait for writeback against this page to complete before allowing it | |
2494 | * to be marked dirty again, and hence written back again, possibly | |
2495 | * before the previous writepage completed. | |
2496 | * | |
2497 | * Block here, instead of in ->writepage(), so that the userspace fs | |
2498 | * can only block processes actually operating on the filesystem. | |
2499 | * | |
2500 | * Otherwise unprivileged userspace fs would be able to block | |
2501 | * unrelated: | |
2502 | * | |
2503 | * - page migration | |
2504 | * - sync(2) | |
2505 | * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER | |
2506 | */ | |
46fb504a | 2507 | static vm_fault_t fuse_page_mkwrite(struct vm_fault *vmf) |
3be5a52b | 2508 | { |
c2ec175c | 2509 | struct page *page = vmf->page; |
11bac800 | 2510 | struct inode *inode = file_inode(vmf->vma->vm_file); |
cca24370 | 2511 | |
11bac800 | 2512 | file_update_time(vmf->vma->vm_file); |
cca24370 MS |
2513 | lock_page(page); |
2514 | if (page->mapping != inode->i_mapping) { | |
2515 | unlock_page(page); | |
2516 | return VM_FAULT_NOPAGE; | |
2517 | } | |
3be5a52b MS |
2518 | |
2519 | fuse_wait_on_page_writeback(inode, page->index); | |
cca24370 | 2520 | return VM_FAULT_LOCKED; |
3be5a52b MS |
2521 | } |
2522 | ||
f0f37e2f | 2523 | static const struct vm_operations_struct fuse_file_vm_ops = { |
3be5a52b MS |
2524 | .close = fuse_vma_close, |
2525 | .fault = filemap_fault, | |
f1820361 | 2526 | .map_pages = filemap_map_pages, |
3be5a52b MS |
2527 | .page_mkwrite = fuse_page_mkwrite, |
2528 | }; | |
2529 | ||
2530 | static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma) | |
2531 | { | |
55752a3a | 2532 | struct fuse_file *ff = file->private_data; |
e78662e8 | 2533 | struct fuse_conn *fc = ff->fm->fc; |
fda0b98e | 2534 | struct inode *inode = file_inode(file); |
cb098dd2 | 2535 | int rc; |
55752a3a | 2536 | |
2a9a609a | 2537 | /* DAX mmap is superior to direct_io mmap */ |
fda0b98e | 2538 | if (FUSE_IS_DAX(inode)) |
2a9a609a SH |
2539 | return fuse_dax_mmap(file, vma); |
2540 | ||
fda0b98e AG |
2541 | /* |
2542 | * If inode is in passthrough io mode, because it has some file open | |
2543 | * in passthrough mode, either mmap to backing file or fail mmap, | |
2544 | * because mixing cached mmap and passthrough io mode is not allowed. | |
2545 | */ | |
4a90451b | 2546 | if (fuse_file_passthrough(ff)) |
fda0b98e AG |
2547 | return fuse_passthrough_mmap(file, vma); |
2548 | else if (fuse_inode_backing(get_fuse_inode(inode))) | |
4a90451b AG |
2549 | return -ENODEV; |
2550 | ||
205c1d80 AG |
2551 | /* |
2552 | * FOPEN_DIRECT_IO handling is special compared to O_DIRECT, | |
2553 | * as does not allow MAP_SHARED mmap without FUSE_DIRECT_IO_ALLOW_MMAP. | |
2554 | */ | |
55752a3a | 2555 | if (ff->open_flags & FOPEN_DIRECT_IO) { |
9511176b BS |
2556 | /* |
2557 | * Can't provide the coherency needed for MAP_SHARED | |
c55e0a55 | 2558 | * if FUSE_DIRECT_IO_ALLOW_MMAP isn't set. |
e78662e8 | 2559 | */ |
c55e0a55 | 2560 | if ((vma->vm_flags & VM_MAYSHARE) && !fc->direct_io_allow_mmap) |
55752a3a MS |
2561 | return -ENODEV; |
2562 | ||
2563 | invalidate_inode_pages2(file->f_mapping); | |
2564 | ||
9511176b BS |
2565 | if (!(vma->vm_flags & VM_MAYSHARE)) { |
2566 | /* MAP_PRIVATE */ | |
2567 | return generic_file_mmap(file, vma); | |
2568 | } | |
cb098dd2 | 2569 | |
205c1d80 AG |
2570 | /* |
2571 | * First mmap of direct_io file enters caching inode io mode. | |
2572 | * Also waits for parallel dio writers to go into serial mode | |
2573 | * (exclusive instead of shared lock). | |
2574 | */ | |
fda0b98e | 2575 | rc = fuse_file_cached_io_start(inode, ff); |
cb098dd2 AG |
2576 | if (rc) |
2577 | return rc; | |
55752a3a MS |
2578 | } |
2579 | ||
650b22b9 PE |
2580 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) |
2581 | fuse_link_write_file(file); | |
2582 | ||
3be5a52b MS |
2583 | file_accessed(file); |
2584 | vma->vm_ops = &fuse_file_vm_ops; | |
b6aeaded MS |
2585 | return 0; |
2586 | } | |
2587 | ||
0b6e9ea0 SF |
2588 | static int convert_fuse_file_lock(struct fuse_conn *fc, |
2589 | const struct fuse_file_lock *ffl, | |
71421259 MS |
2590 | struct file_lock *fl) |
2591 | { | |
2592 | switch (ffl->type) { | |
2593 | case F_UNLCK: | |
2594 | break; | |
2595 | ||
2596 | case F_RDLCK: | |
2597 | case F_WRLCK: | |
2598 | if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX || | |
2599 | ffl->end < ffl->start) | |
2600 | return -EIO; | |
2601 | ||
2602 | fl->fl_start = ffl->start; | |
2603 | fl->fl_end = ffl->end; | |
0b6e9ea0 SF |
2604 | |
2605 | /* | |
9d5b86ac BC |
2606 | * Convert pid into init's pid namespace. The locks API will |
2607 | * translate it into the caller's pid namespace. | |
0b6e9ea0 SF |
2608 | */ |
2609 | rcu_read_lock(); | |
9d5b86ac | 2610 | fl->fl_pid = pid_nr_ns(find_pid_ns(ffl->pid, fc->pid_ns), &init_pid_ns); |
0b6e9ea0 | 2611 | rcu_read_unlock(); |
71421259 MS |
2612 | break; |
2613 | ||
2614 | default: | |
2615 | return -EIO; | |
2616 | } | |
2617 | fl->fl_type = ffl->type; | |
2618 | return 0; | |
2619 | } | |
2620 | ||
7078187a | 2621 | static void fuse_lk_fill(struct fuse_args *args, struct file *file, |
a9ff4f87 | 2622 | const struct file_lock *fl, int opcode, pid_t pid, |
7078187a | 2623 | int flock, struct fuse_lk_in *inarg) |
71421259 | 2624 | { |
6131ffaa | 2625 | struct inode *inode = file_inode(file); |
9c8ef561 | 2626 | struct fuse_conn *fc = get_fuse_conn(inode); |
71421259 | 2627 | struct fuse_file *ff = file->private_data; |
7078187a MS |
2628 | |
2629 | memset(inarg, 0, sizeof(*inarg)); | |
2630 | inarg->fh = ff->fh; | |
2631 | inarg->owner = fuse_lock_owner_id(fc, fl->fl_owner); | |
2632 | inarg->lk.start = fl->fl_start; | |
2633 | inarg->lk.end = fl->fl_end; | |
2634 | inarg->lk.type = fl->fl_type; | |
2635 | inarg->lk.pid = pid; | |
a9ff4f87 | 2636 | if (flock) |
7078187a | 2637 | inarg->lk_flags |= FUSE_LK_FLOCK; |
d5b48543 MS |
2638 | args->opcode = opcode; |
2639 | args->nodeid = get_node_id(inode); | |
2640 | args->in_numargs = 1; | |
2641 | args->in_args[0].size = sizeof(*inarg); | |
2642 | args->in_args[0].value = inarg; | |
71421259 MS |
2643 | } |
2644 | ||
2645 | static int fuse_getlk(struct file *file, struct file_lock *fl) | |
2646 | { | |
6131ffaa | 2647 | struct inode *inode = file_inode(file); |
fcee216b | 2648 | struct fuse_mount *fm = get_fuse_mount(inode); |
7078187a MS |
2649 | FUSE_ARGS(args); |
2650 | struct fuse_lk_in inarg; | |
71421259 MS |
2651 | struct fuse_lk_out outarg; |
2652 | int err; | |
2653 | ||
7078187a | 2654 | fuse_lk_fill(&args, file, fl, FUSE_GETLK, 0, 0, &inarg); |
d5b48543 MS |
2655 | args.out_numargs = 1; |
2656 | args.out_args[0].size = sizeof(outarg); | |
2657 | args.out_args[0].value = &outarg; | |
fcee216b | 2658 | err = fuse_simple_request(fm, &args); |
71421259 | 2659 | if (!err) |
fcee216b | 2660 | err = convert_fuse_file_lock(fm->fc, &outarg.lk, fl); |
71421259 MS |
2661 | |
2662 | return err; | |
2663 | } | |
2664 | ||
a9ff4f87 | 2665 | static int fuse_setlk(struct file *file, struct file_lock *fl, int flock) |
71421259 | 2666 | { |
6131ffaa | 2667 | struct inode *inode = file_inode(file); |
fcee216b | 2668 | struct fuse_mount *fm = get_fuse_mount(inode); |
7078187a MS |
2669 | FUSE_ARGS(args); |
2670 | struct fuse_lk_in inarg; | |
71421259 | 2671 | int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK; |
0b6e9ea0 | 2672 | struct pid *pid = fl->fl_type != F_UNLCK ? task_tgid(current) : NULL; |
fcee216b | 2673 | pid_t pid_nr = pid_nr_ns(pid, fm->fc->pid_ns); |
71421259 MS |
2674 | int err; |
2675 | ||
8fb47a4f | 2676 | if (fl->fl_lmops && fl->fl_lmops->lm_grant) { |
48e90761 MS |
2677 | /* NLM needs asynchronous locks, which we don't support yet */ |
2678 | return -ENOLCK; | |
2679 | } | |
2680 | ||
0b6e9ea0 | 2681 | fuse_lk_fill(&args, file, fl, opcode, pid_nr, flock, &inarg); |
fcee216b | 2682 | err = fuse_simple_request(fm, &args); |
71421259 | 2683 | |
a4d27e75 MS |
2684 | /* locking is restartable */ |
2685 | if (err == -EINTR) | |
2686 | err = -ERESTARTSYS; | |
7078187a | 2687 | |
71421259 MS |
2688 | return err; |
2689 | } | |
2690 | ||
2691 | static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl) | |
2692 | { | |
6131ffaa | 2693 | struct inode *inode = file_inode(file); |
71421259 MS |
2694 | struct fuse_conn *fc = get_fuse_conn(inode); |
2695 | int err; | |
2696 | ||
48e90761 MS |
2697 | if (cmd == F_CANCELLK) { |
2698 | err = 0; | |
2699 | } else if (cmd == F_GETLK) { | |
71421259 | 2700 | if (fc->no_lock) { |
9d6a8c5c | 2701 | posix_test_lock(file, fl); |
71421259 MS |
2702 | err = 0; |
2703 | } else | |
2704 | err = fuse_getlk(file, fl); | |
2705 | } else { | |
2706 | if (fc->no_lock) | |
48e90761 | 2707 | err = posix_lock_file(file, fl, NULL); |
71421259 | 2708 | else |
a9ff4f87 | 2709 | err = fuse_setlk(file, fl, 0); |
71421259 MS |
2710 | } |
2711 | return err; | |
2712 | } | |
2713 | ||
a9ff4f87 MS |
2714 | static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl) |
2715 | { | |
6131ffaa | 2716 | struct inode *inode = file_inode(file); |
a9ff4f87 MS |
2717 | struct fuse_conn *fc = get_fuse_conn(inode); |
2718 | int err; | |
2719 | ||
37fb3a30 | 2720 | if (fc->no_flock) { |
4f656367 | 2721 | err = locks_lock_file_wait(file, fl); |
a9ff4f87 | 2722 | } else { |
37fb3a30 MS |
2723 | struct fuse_file *ff = file->private_data; |
2724 | ||
a9ff4f87 | 2725 | /* emulate flock with POSIX locks */ |
37fb3a30 | 2726 | ff->flock = true; |
a9ff4f87 MS |
2727 | err = fuse_setlk(file, fl, 1); |
2728 | } | |
2729 | ||
2730 | return err; | |
2731 | } | |
2732 | ||
b2d2272f MS |
2733 | static sector_t fuse_bmap(struct address_space *mapping, sector_t block) |
2734 | { | |
2735 | struct inode *inode = mapping->host; | |
fcee216b | 2736 | struct fuse_mount *fm = get_fuse_mount(inode); |
7078187a | 2737 | FUSE_ARGS(args); |
b2d2272f MS |
2738 | struct fuse_bmap_in inarg; |
2739 | struct fuse_bmap_out outarg; | |
2740 | int err; | |
2741 | ||
fcee216b | 2742 | if (!inode->i_sb->s_bdev || fm->fc->no_bmap) |
b2d2272f MS |
2743 | return 0; |
2744 | ||
b2d2272f MS |
2745 | memset(&inarg, 0, sizeof(inarg)); |
2746 | inarg.block = block; | |
2747 | inarg.blocksize = inode->i_sb->s_blocksize; | |
d5b48543 MS |
2748 | args.opcode = FUSE_BMAP; |
2749 | args.nodeid = get_node_id(inode); | |
2750 | args.in_numargs = 1; | |
2751 | args.in_args[0].size = sizeof(inarg); | |
2752 | args.in_args[0].value = &inarg; | |
2753 | args.out_numargs = 1; | |
2754 | args.out_args[0].size = sizeof(outarg); | |
2755 | args.out_args[0].value = &outarg; | |
fcee216b | 2756 | err = fuse_simple_request(fm, &args); |
b2d2272f | 2757 | if (err == -ENOSYS) |
fcee216b | 2758 | fm->fc->no_bmap = 1; |
b2d2272f MS |
2759 | |
2760 | return err ? 0 : outarg.block; | |
2761 | } | |
2762 | ||
0b5da8db R |
2763 | static loff_t fuse_lseek(struct file *file, loff_t offset, int whence) |
2764 | { | |
2765 | struct inode *inode = file->f_mapping->host; | |
fcee216b | 2766 | struct fuse_mount *fm = get_fuse_mount(inode); |
0b5da8db R |
2767 | struct fuse_file *ff = file->private_data; |
2768 | FUSE_ARGS(args); | |
2769 | struct fuse_lseek_in inarg = { | |
2770 | .fh = ff->fh, | |
2771 | .offset = offset, | |
2772 | .whence = whence | |
2773 | }; | |
2774 | struct fuse_lseek_out outarg; | |
2775 | int err; | |
2776 | ||
fcee216b | 2777 | if (fm->fc->no_lseek) |
0b5da8db R |
2778 | goto fallback; |
2779 | ||
d5b48543 MS |
2780 | args.opcode = FUSE_LSEEK; |
2781 | args.nodeid = ff->nodeid; | |
2782 | args.in_numargs = 1; | |
2783 | args.in_args[0].size = sizeof(inarg); | |
2784 | args.in_args[0].value = &inarg; | |
2785 | args.out_numargs = 1; | |
2786 | args.out_args[0].size = sizeof(outarg); | |
2787 | args.out_args[0].value = &outarg; | |
fcee216b | 2788 | err = fuse_simple_request(fm, &args); |
0b5da8db R |
2789 | if (err) { |
2790 | if (err == -ENOSYS) { | |
fcee216b | 2791 | fm->fc->no_lseek = 1; |
0b5da8db R |
2792 | goto fallback; |
2793 | } | |
2794 | return err; | |
2795 | } | |
2796 | ||
2797 | return vfs_setpos(file, outarg.offset, inode->i_sb->s_maxbytes); | |
2798 | ||
2799 | fallback: | |
c6c745b8 | 2800 | err = fuse_update_attributes(inode, file, STATX_SIZE); |
0b5da8db R |
2801 | if (!err) |
2802 | return generic_file_llseek(file, offset, whence); | |
2803 | else | |
2804 | return err; | |
2805 | } | |
2806 | ||
965c8e59 | 2807 | static loff_t fuse_file_llseek(struct file *file, loff_t offset, int whence) |
5559b8f4 MS |
2808 | { |
2809 | loff_t retval; | |
6131ffaa | 2810 | struct inode *inode = file_inode(file); |
5559b8f4 | 2811 | |
0b5da8db R |
2812 | switch (whence) { |
2813 | case SEEK_SET: | |
2814 | case SEEK_CUR: | |
2815 | /* No i_mutex protection necessary for SEEK_CUR and SEEK_SET */ | |
965c8e59 | 2816 | retval = generic_file_llseek(file, offset, whence); |
0b5da8db R |
2817 | break; |
2818 | case SEEK_END: | |
5955102c | 2819 | inode_lock(inode); |
c6c745b8 | 2820 | retval = fuse_update_attributes(inode, file, STATX_SIZE); |
0b5da8db R |
2821 | if (!retval) |
2822 | retval = generic_file_llseek(file, offset, whence); | |
5955102c | 2823 | inode_unlock(inode); |
0b5da8db R |
2824 | break; |
2825 | case SEEK_HOLE: | |
2826 | case SEEK_DATA: | |
5955102c | 2827 | inode_lock(inode); |
0b5da8db | 2828 | retval = fuse_lseek(file, offset, whence); |
5955102c | 2829 | inode_unlock(inode); |
0b5da8db R |
2830 | break; |
2831 | default: | |
2832 | retval = -EINVAL; | |
2833 | } | |
c07c3d19 | 2834 | |
5559b8f4 MS |
2835 | return retval; |
2836 | } | |
2837 | ||
95668a69 TH |
2838 | /* |
2839 | * All files which have been polled are linked to RB tree | |
2840 | * fuse_conn->polled_files which is indexed by kh. Walk the tree and | |
2841 | * find the matching one. | |
2842 | */ | |
2843 | static struct rb_node **fuse_find_polled_node(struct fuse_conn *fc, u64 kh, | |
2844 | struct rb_node **parent_out) | |
2845 | { | |
2846 | struct rb_node **link = &fc->polled_files.rb_node; | |
2847 | struct rb_node *last = NULL; | |
2848 | ||
2849 | while (*link) { | |
2850 | struct fuse_file *ff; | |
2851 | ||
2852 | last = *link; | |
2853 | ff = rb_entry(last, struct fuse_file, polled_node); | |
2854 | ||
2855 | if (kh < ff->kh) | |
2856 | link = &last->rb_left; | |
2857 | else if (kh > ff->kh) | |
2858 | link = &last->rb_right; | |
2859 | else | |
2860 | return link; | |
2861 | } | |
2862 | ||
2863 | if (parent_out) | |
2864 | *parent_out = last; | |
2865 | return link; | |
2866 | } | |
2867 | ||
2868 | /* | |
2869 | * The file is about to be polled. Make sure it's on the polled_files | |
2870 | * RB tree. Note that files once added to the polled_files tree are | |
2871 | * not removed before the file is released. This is because a file | |
2872 | * polled once is likely to be polled again. | |
2873 | */ | |
2874 | static void fuse_register_polled_file(struct fuse_conn *fc, | |
2875 | struct fuse_file *ff) | |
2876 | { | |
2877 | spin_lock(&fc->lock); | |
2878 | if (RB_EMPTY_NODE(&ff->polled_node)) { | |
3f649ab7 | 2879 | struct rb_node **link, *parent; |
95668a69 TH |
2880 | |
2881 | link = fuse_find_polled_node(fc, ff->kh, &parent); | |
2882 | BUG_ON(*link); | |
2883 | rb_link_node(&ff->polled_node, parent, link); | |
2884 | rb_insert_color(&ff->polled_node, &fc->polled_files); | |
2885 | } | |
2886 | spin_unlock(&fc->lock); | |
2887 | } | |
2888 | ||
076ccb76 | 2889 | __poll_t fuse_file_poll(struct file *file, poll_table *wait) |
95668a69 | 2890 | { |
95668a69 | 2891 | struct fuse_file *ff = file->private_data; |
fcee216b | 2892 | struct fuse_mount *fm = ff->fm; |
95668a69 TH |
2893 | struct fuse_poll_in inarg = { .fh = ff->fh, .kh = ff->kh }; |
2894 | struct fuse_poll_out outarg; | |
7078187a | 2895 | FUSE_ARGS(args); |
95668a69 TH |
2896 | int err; |
2897 | ||
fcee216b | 2898 | if (fm->fc->no_poll) |
95668a69 TH |
2899 | return DEFAULT_POLLMASK; |
2900 | ||
2901 | poll_wait(file, &ff->poll_wait, wait); | |
c71d227f | 2902 | inarg.events = mangle_poll(poll_requested_events(wait)); |
95668a69 TH |
2903 | |
2904 | /* | |
2905 | * Ask for notification iff there's someone waiting for it. | |
2906 | * The client may ignore the flag and always notify. | |
2907 | */ | |
2908 | if (waitqueue_active(&ff->poll_wait)) { | |
2909 | inarg.flags |= FUSE_POLL_SCHEDULE_NOTIFY; | |
fcee216b | 2910 | fuse_register_polled_file(fm->fc, ff); |
95668a69 TH |
2911 | } |
2912 | ||
d5b48543 MS |
2913 | args.opcode = FUSE_POLL; |
2914 | args.nodeid = ff->nodeid; | |
2915 | args.in_numargs = 1; | |
2916 | args.in_args[0].size = sizeof(inarg); | |
2917 | args.in_args[0].value = &inarg; | |
2918 | args.out_numargs = 1; | |
2919 | args.out_args[0].size = sizeof(outarg); | |
2920 | args.out_args[0].value = &outarg; | |
fcee216b | 2921 | err = fuse_simple_request(fm, &args); |
95668a69 TH |
2922 | |
2923 | if (!err) | |
c71d227f | 2924 | return demangle_poll(outarg.revents); |
95668a69 | 2925 | if (err == -ENOSYS) { |
fcee216b | 2926 | fm->fc->no_poll = 1; |
95668a69 TH |
2927 | return DEFAULT_POLLMASK; |
2928 | } | |
a9a08845 | 2929 | return EPOLLERR; |
95668a69 | 2930 | } |
08cbf542 | 2931 | EXPORT_SYMBOL_GPL(fuse_file_poll); |
95668a69 TH |
2932 | |
2933 | /* | |
2934 | * This is called from fuse_handle_notify() on FUSE_NOTIFY_POLL and | |
2935 | * wakes up the poll waiters. | |
2936 | */ | |
2937 | int fuse_notify_poll_wakeup(struct fuse_conn *fc, | |
2938 | struct fuse_notify_poll_wakeup_out *outarg) | |
2939 | { | |
2940 | u64 kh = outarg->kh; | |
2941 | struct rb_node **link; | |
2942 | ||
2943 | spin_lock(&fc->lock); | |
2944 | ||
2945 | link = fuse_find_polled_node(fc, kh, NULL); | |
2946 | if (*link) { | |
2947 | struct fuse_file *ff; | |
2948 | ||
2949 | ff = rb_entry(*link, struct fuse_file, polled_node); | |
2950 | wake_up_interruptible_sync(&ff->poll_wait); | |
2951 | } | |
2952 | ||
2953 | spin_unlock(&fc->lock); | |
2954 | return 0; | |
2955 | } | |
2956 | ||
efb9fa9e MP |
2957 | static void fuse_do_truncate(struct file *file) |
2958 | { | |
2959 | struct inode *inode = file->f_mapping->host; | |
2960 | struct iattr attr; | |
2961 | ||
2962 | attr.ia_valid = ATTR_SIZE; | |
2963 | attr.ia_size = i_size_read(inode); | |
2964 | ||
2965 | attr.ia_file = file; | |
2966 | attr.ia_valid |= ATTR_FILE; | |
2967 | ||
62490330 | 2968 | fuse_do_setattr(file_dentry(file), &attr, file); |
efb9fa9e MP |
2969 | } |
2970 | ||
5da784cc | 2971 | static inline loff_t fuse_round_up(struct fuse_conn *fc, loff_t off) |
e5c5f05d | 2972 | { |
5da784cc | 2973 | return round_up(off, fc->max_pages << PAGE_SHIFT); |
e5c5f05d MP |
2974 | } |
2975 | ||
4273b793 | 2976 | static ssize_t |
c8b8e32d | 2977 | fuse_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
4273b793 | 2978 | { |
9d5722b7 | 2979 | DECLARE_COMPLETION_ONSTACK(wait); |
4273b793 | 2980 | ssize_t ret = 0; |
60b9df7a MS |
2981 | struct file *file = iocb->ki_filp; |
2982 | struct fuse_file *ff = file->private_data; | |
4273b793 | 2983 | loff_t pos = 0; |
bcba24cc MP |
2984 | struct inode *inode; |
2985 | loff_t i_size; | |
933a3752 | 2986 | size_t count = iov_iter_count(iter), shortened = 0; |
c8b8e32d | 2987 | loff_t offset = iocb->ki_pos; |
36cf66ed | 2988 | struct fuse_io_priv *io; |
4273b793 | 2989 | |
4273b793 | 2990 | pos = offset; |
bcba24cc MP |
2991 | inode = file->f_mapping->host; |
2992 | i_size = i_size_read(inode); | |
4273b793 | 2993 | |
933a3752 | 2994 | if ((iov_iter_rw(iter) == READ) && (offset >= i_size)) |
9fe55eea SW |
2995 | return 0; |
2996 | ||
bcba24cc | 2997 | io = kmalloc(sizeof(struct fuse_io_priv), GFP_KERNEL); |
36cf66ed MP |
2998 | if (!io) |
2999 | return -ENOMEM; | |
bcba24cc | 3000 | spin_lock_init(&io->lock); |
744742d6 | 3001 | kref_init(&io->refcnt); |
bcba24cc MP |
3002 | io->reqs = 1; |
3003 | io->bytes = -1; | |
3004 | io->size = 0; | |
3005 | io->offset = offset; | |
6f673763 | 3006 | io->write = (iov_iter_rw(iter) == WRITE); |
bcba24cc | 3007 | io->err = 0; |
bcba24cc MP |
3008 | /* |
3009 | * By default, we want to optimize all I/Os with async request | |
60b9df7a | 3010 | * submission to the client filesystem if supported. |
bcba24cc | 3011 | */ |
69456535 | 3012 | io->async = ff->fm->fc->async_dio; |
bcba24cc | 3013 | io->iocb = iocb; |
7879c4e5 | 3014 | io->blocking = is_sync_kiocb(iocb); |
bcba24cc | 3015 | |
933a3752 AV |
3016 | /* optimization for short read */ |
3017 | if (io->async && !io->write && offset + count > i_size) { | |
69456535 | 3018 | iov_iter_truncate(iter, fuse_round_up(ff->fm->fc, i_size - offset)); |
933a3752 AV |
3019 | shortened = count - iov_iter_count(iter); |
3020 | count -= shortened; | |
3021 | } | |
3022 | ||
bcba24cc | 3023 | /* |
7879c4e5 AS |
3024 | * We cannot asynchronously extend the size of a file. |
3025 | * In such case the aio will behave exactly like sync io. | |
bcba24cc | 3026 | */ |
933a3752 | 3027 | if ((offset + count > i_size) && io->write) |
7879c4e5 | 3028 | io->blocking = true; |
4273b793 | 3029 | |
7879c4e5 | 3030 | if (io->async && io->blocking) { |
744742d6 SF |
3031 | /* |
3032 | * Additional reference to keep io around after | |
3033 | * calling fuse_aio_complete() | |
3034 | */ | |
3035 | kref_get(&io->refcnt); | |
9d5722b7 | 3036 | io->done = &wait; |
744742d6 | 3037 | } |
9d5722b7 | 3038 | |
6f673763 | 3039 | if (iov_iter_rw(iter) == WRITE) { |
6b775b18 | 3040 | ret = fuse_direct_io(io, iter, &pos, FUSE_DIO_WRITE); |
fa5eee57 | 3041 | fuse_invalidate_attr_mask(inode, FUSE_STATX_MODSIZE); |
812408fb | 3042 | } else { |
d22a943f | 3043 | ret = __fuse_direct_read(io, iter, &pos); |
812408fb | 3044 | } |
933a3752 | 3045 | iov_iter_reexpand(iter, iov_iter_count(iter) + shortened); |
36cf66ed | 3046 | |
bcba24cc | 3047 | if (io->async) { |
ebacb812 LC |
3048 | bool blocking = io->blocking; |
3049 | ||
bcba24cc MP |
3050 | fuse_aio_complete(io, ret < 0 ? ret : 0, -1); |
3051 | ||
3052 | /* we have a non-extending, async request, so return */ | |
ebacb812 | 3053 | if (!blocking) |
bcba24cc MP |
3054 | return -EIOCBQUEUED; |
3055 | ||
9d5722b7 CH |
3056 | wait_for_completion(&wait); |
3057 | ret = fuse_get_res_by_io(io); | |
bcba24cc MP |
3058 | } |
3059 | ||
744742d6 | 3060 | kref_put(&io->refcnt, fuse_io_release); |
9d5722b7 | 3061 | |
6f673763 | 3062 | if (iov_iter_rw(iter) == WRITE) { |
d347739a | 3063 | fuse_write_update_attr(inode, pos, ret); |
15352405 | 3064 | /* For extending writes we already hold exclusive lock */ |
d347739a | 3065 | if (ret < 0 && offset + count > i_size) |
efb9fa9e MP |
3066 | fuse_do_truncate(file); |
3067 | } | |
4273b793 AA |
3068 | |
3069 | return ret; | |
3070 | } | |
3071 | ||
26eb3bae MS |
3072 | static int fuse_writeback_range(struct inode *inode, loff_t start, loff_t end) |
3073 | { | |
e388164e | 3074 | int err = filemap_write_and_wait_range(inode->i_mapping, start, LLONG_MAX); |
26eb3bae MS |
3075 | |
3076 | if (!err) | |
3077 | fuse_sync_writes(inode); | |
3078 | ||
3079 | return err; | |
3080 | } | |
3081 | ||
cdadb11c MS |
3082 | static long fuse_file_fallocate(struct file *file, int mode, loff_t offset, |
3083 | loff_t length) | |
05ba1f08 AP |
3084 | { |
3085 | struct fuse_file *ff = file->private_data; | |
1c68271c | 3086 | struct inode *inode = file_inode(file); |
0ab08f57 | 3087 | struct fuse_inode *fi = get_fuse_inode(inode); |
fcee216b | 3088 | struct fuse_mount *fm = ff->fm; |
7078187a | 3089 | FUSE_ARGS(args); |
05ba1f08 AP |
3090 | struct fuse_fallocate_in inarg = { |
3091 | .fh = ff->fh, | |
3092 | .offset = offset, | |
3093 | .length = length, | |
3094 | .mode = mode | |
3095 | }; | |
3096 | int err; | |
44361e8c MS |
3097 | bool block_faults = FUSE_IS_DAX(inode) && |
3098 | (!(mode & FALLOC_FL_KEEP_SIZE) || | |
3099 | (mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))); | |
6ae330ca | 3100 | |
6b1bdb56 RJ |
3101 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | |
3102 | FALLOC_FL_ZERO_RANGE)) | |
4adb8302 MS |
3103 | return -EOPNOTSUPP; |
3104 | ||
fcee216b | 3105 | if (fm->fc->no_fallocate) |
519c6040 MS |
3106 | return -EOPNOTSUPP; |
3107 | ||
44361e8c MS |
3108 | inode_lock(inode); |
3109 | if (block_faults) { | |
3110 | filemap_invalidate_lock(inode->i_mapping); | |
3111 | err = fuse_dax_break_layouts(inode, 0, 0); | |
3112 | if (err) | |
3113 | goto out; | |
3114 | } | |
6ae330ca | 3115 | |
44361e8c MS |
3116 | if (mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) { |
3117 | loff_t endbyte = offset + length - 1; | |
26eb3bae | 3118 | |
44361e8c MS |
3119 | err = fuse_writeback_range(inode, offset, endbyte); |
3120 | if (err) | |
3121 | goto out; | |
3634a632 BF |
3122 | } |
3123 | ||
0cbade02 LB |
3124 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
3125 | offset + length > i_size_read(inode)) { | |
3126 | err = inode_newsize_ok(inode, offset + length); | |
3127 | if (err) | |
35d6fcbb | 3128 | goto out; |
0cbade02 LB |
3129 | } |
3130 | ||
4a6f278d MS |
3131 | err = file_modified(file); |
3132 | if (err) | |
3133 | goto out; | |
3134 | ||
0ab08f57 MP |
3135 | if (!(mode & FALLOC_FL_KEEP_SIZE)) |
3136 | set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); | |
3137 | ||
d5b48543 MS |
3138 | args.opcode = FUSE_FALLOCATE; |
3139 | args.nodeid = ff->nodeid; | |
3140 | args.in_numargs = 1; | |
3141 | args.in_args[0].size = sizeof(inarg); | |
3142 | args.in_args[0].value = &inarg; | |
fcee216b | 3143 | err = fuse_simple_request(fm, &args); |
519c6040 | 3144 | if (err == -ENOSYS) { |
fcee216b | 3145 | fm->fc->no_fallocate = 1; |
519c6040 MS |
3146 | err = -EOPNOTSUPP; |
3147 | } | |
bee6c307 BF |
3148 | if (err) |
3149 | goto out; | |
3150 | ||
3151 | /* we could have extended the file */ | |
b0aa7606 | 3152 | if (!(mode & FALLOC_FL_KEEP_SIZE)) { |
20235b43 | 3153 | if (fuse_write_update_attr(inode, offset + length, length)) |
93d2269d | 3154 | file_update_time(file); |
b0aa7606 | 3155 | } |
bee6c307 | 3156 | |
6b1bdb56 | 3157 | if (mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) |
bee6c307 BF |
3158 | truncate_pagecache_range(inode, offset, offset + length - 1); |
3159 | ||
fa5eee57 | 3160 | fuse_invalidate_attr_mask(inode, FUSE_STATX_MODSIZE); |
bee6c307 | 3161 | |
3634a632 | 3162 | out: |
0ab08f57 MP |
3163 | if (!(mode & FALLOC_FL_KEEP_SIZE)) |
3164 | clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); | |
3165 | ||
6ae330ca | 3166 | if (block_faults) |
8bcbbe9c | 3167 | filemap_invalidate_unlock(inode->i_mapping); |
6ae330ca | 3168 | |
44361e8c | 3169 | inode_unlock(inode); |
3634a632 | 3170 | |
5c791fe1 MS |
3171 | fuse_flush_time_update(inode); |
3172 | ||
05ba1f08 AP |
3173 | return err; |
3174 | } | |
05ba1f08 | 3175 | |
64bf5ff5 DC |
3176 | static ssize_t __fuse_copy_file_range(struct file *file_in, loff_t pos_in, |
3177 | struct file *file_out, loff_t pos_out, | |
3178 | size_t len, unsigned int flags) | |
88bc7d50 NV |
3179 | { |
3180 | struct fuse_file *ff_in = file_in->private_data; | |
3181 | struct fuse_file *ff_out = file_out->private_data; | |
a2bc9236 | 3182 | struct inode *inode_in = file_inode(file_in); |
88bc7d50 NV |
3183 | struct inode *inode_out = file_inode(file_out); |
3184 | struct fuse_inode *fi_out = get_fuse_inode(inode_out); | |
fcee216b MR |
3185 | struct fuse_mount *fm = ff_in->fm; |
3186 | struct fuse_conn *fc = fm->fc; | |
88bc7d50 NV |
3187 | FUSE_ARGS(args); |
3188 | struct fuse_copy_file_range_in inarg = { | |
3189 | .fh_in = ff_in->fh, | |
3190 | .off_in = pos_in, | |
3191 | .nodeid_out = ff_out->nodeid, | |
3192 | .fh_out = ff_out->fh, | |
3193 | .off_out = pos_out, | |
3194 | .len = len, | |
3195 | .flags = flags | |
3196 | }; | |
3197 | struct fuse_write_out outarg; | |
3198 | ssize_t err; | |
3199 | /* mark unstable when write-back is not used, and file_out gets | |
3200 | * extended */ | |
3201 | bool is_unstable = (!fc->writeback_cache) && | |
3202 | ((pos_out + len) > inode_out->i_size); | |
3203 | ||
3204 | if (fc->no_copy_file_range) | |
3205 | return -EOPNOTSUPP; | |
3206 | ||
5dae222a AG |
3207 | if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb) |
3208 | return -EXDEV; | |
3209 | ||
2c4656df MS |
3210 | inode_lock(inode_in); |
3211 | err = fuse_writeback_range(inode_in, pos_in, pos_in + len - 1); | |
3212 | inode_unlock(inode_in); | |
3213 | if (err) | |
3214 | return err; | |
a2bc9236 | 3215 | |
88bc7d50 NV |
3216 | inode_lock(inode_out); |
3217 | ||
fe0da9c0 AG |
3218 | err = file_modified(file_out); |
3219 | if (err) | |
3220 | goto out; | |
3221 | ||
9b46418c MS |
3222 | /* |
3223 | * Write out dirty pages in the destination file before sending the COPY | |
3224 | * request to userspace. After the request is completed, truncate off | |
3225 | * pages (including partial ones) from the cache that have been copied, | |
3226 | * since these contain stale data at that point. | |
3227 | * | |
3228 | * This should be mostly correct, but if the COPY writes to partial | |
3229 | * pages (at the start or end) and the parts not covered by the COPY are | |
3230 | * written through a memory map after calling fuse_writeback_range(), | |
3231 | * then these partial page modifications will be lost on truncation. | |
3232 | * | |
3233 | * It is unlikely that someone would rely on such mixed style | |
3234 | * modifications. Yet this does give less guarantees than if the | |
3235 | * copying was performed with write(2). | |
3236 | * | |
8bcbbe9c | 3237 | * To fix this a mapping->invalidate_lock could be used to prevent new |
9b46418c MS |
3238 | * faults while the copy is ongoing. |
3239 | */ | |
2c4656df MS |
3240 | err = fuse_writeback_range(inode_out, pos_out, pos_out + len - 1); |
3241 | if (err) | |
3242 | goto out; | |
88bc7d50 NV |
3243 | |
3244 | if (is_unstable) | |
3245 | set_bit(FUSE_I_SIZE_UNSTABLE, &fi_out->state); | |
3246 | ||
d5b48543 MS |
3247 | args.opcode = FUSE_COPY_FILE_RANGE; |
3248 | args.nodeid = ff_in->nodeid; | |
3249 | args.in_numargs = 1; | |
3250 | args.in_args[0].size = sizeof(inarg); | |
3251 | args.in_args[0].value = &inarg; | |
3252 | args.out_numargs = 1; | |
3253 | args.out_args[0].size = sizeof(outarg); | |
3254 | args.out_args[0].value = &outarg; | |
fcee216b | 3255 | err = fuse_simple_request(fm, &args); |
88bc7d50 NV |
3256 | if (err == -ENOSYS) { |
3257 | fc->no_copy_file_range = 1; | |
3258 | err = -EOPNOTSUPP; | |
3259 | } | |
3260 | if (err) | |
3261 | goto out; | |
3262 | ||
9b46418c MS |
3263 | truncate_inode_pages_range(inode_out->i_mapping, |
3264 | ALIGN_DOWN(pos_out, PAGE_SIZE), | |
3265 | ALIGN(pos_out + outarg.size, PAGE_SIZE) - 1); | |
3266 | ||
20235b43 MS |
3267 | file_update_time(file_out); |
3268 | fuse_write_update_attr(inode_out, pos_out + outarg.size, outarg.size); | |
88bc7d50 NV |
3269 | |
3270 | err = outarg.size; | |
3271 | out: | |
3272 | if (is_unstable) | |
3273 | clear_bit(FUSE_I_SIZE_UNSTABLE, &fi_out->state); | |
3274 | ||
3275 | inode_unlock(inode_out); | |
fe0da9c0 | 3276 | file_accessed(file_in); |
88bc7d50 | 3277 | |
5c791fe1 MS |
3278 | fuse_flush_time_update(inode_out); |
3279 | ||
88bc7d50 NV |
3280 | return err; |
3281 | } | |
3282 | ||
64bf5ff5 DC |
3283 | static ssize_t fuse_copy_file_range(struct file *src_file, loff_t src_off, |
3284 | struct file *dst_file, loff_t dst_off, | |
3285 | size_t len, unsigned int flags) | |
3286 | { | |
3287 | ssize_t ret; | |
3288 | ||
3289 | ret = __fuse_copy_file_range(src_file, src_off, dst_file, dst_off, | |
3290 | len, flags); | |
3291 | ||
5dae222a | 3292 | if (ret == -EOPNOTSUPP || ret == -EXDEV) |
705bcfcb AG |
3293 | ret = splice_copy_file_range(src_file, src_off, dst_file, |
3294 | dst_off, len); | |
64bf5ff5 DC |
3295 | return ret; |
3296 | } | |
3297 | ||
4b6f5d20 | 3298 | static const struct file_operations fuse_file_operations = { |
5559b8f4 | 3299 | .llseek = fuse_file_llseek, |
37c20f16 | 3300 | .read_iter = fuse_file_read_iter, |
84c3d55c | 3301 | .write_iter = fuse_file_write_iter, |
b6aeaded MS |
3302 | .mmap = fuse_file_mmap, |
3303 | .open = fuse_open, | |
3304 | .flush = fuse_flush, | |
3305 | .release = fuse_release, | |
3306 | .fsync = fuse_fsync, | |
71421259 | 3307 | .lock = fuse_file_lock, |
2a9a609a | 3308 | .get_unmapped_area = thp_get_unmapped_area, |
a9ff4f87 | 3309 | .flock = fuse_file_flock, |
5ca73468 AG |
3310 | .splice_read = fuse_splice_read, |
3311 | .splice_write = fuse_splice_write, | |
59efec7b TH |
3312 | .unlocked_ioctl = fuse_file_ioctl, |
3313 | .compat_ioctl = fuse_file_compat_ioctl, | |
95668a69 | 3314 | .poll = fuse_file_poll, |
05ba1f08 | 3315 | .fallocate = fuse_file_fallocate, |
d4136d60 | 3316 | .copy_file_range = fuse_copy_file_range, |
413ef8cb MS |
3317 | }; |
3318 | ||
f5e54d6e | 3319 | static const struct address_space_operations fuse_file_aops = { |
5efd00e4 | 3320 | .read_folio = fuse_read_folio, |
76a0294e | 3321 | .readahead = fuse_readahead, |
26d614df | 3322 | .writepages = fuse_writepages, |
2bf06b8e | 3323 | .launder_folio = fuse_launder_folio, |
187c82cb | 3324 | .dirty_folio = filemap_dirty_folio, |
e1c420ac | 3325 | .migrate_folio = filemap_migrate_folio, |
b2d2272f | 3326 | .bmap = fuse_bmap, |
4273b793 | 3327 | .direct_IO = fuse_direct_IO, |
6b12c1b3 PE |
3328 | .write_begin = fuse_write_begin, |
3329 | .write_end = fuse_write_end, | |
b6aeaded MS |
3330 | }; |
3331 | ||
93a497b9 | 3332 | void fuse_init_file_inode(struct inode *inode, unsigned int flags) |
b6aeaded | 3333 | { |
ab2257e9 MS |
3334 | struct fuse_inode *fi = get_fuse_inode(inode); |
3335 | ||
45323fb7 MS |
3336 | inode->i_fop = &fuse_file_operations; |
3337 | inode->i_data.a_ops = &fuse_file_aops; | |
ab2257e9 MS |
3338 | |
3339 | INIT_LIST_HEAD(&fi->write_files); | |
3340 | INIT_LIST_HEAD(&fi->queued_writes); | |
3341 | fi->writectr = 0; | |
cb098dd2 | 3342 | fi->iocachectr = 0; |
ab2257e9 | 3343 | init_waitqueue_head(&fi->page_waitq); |
205c1d80 | 3344 | init_waitqueue_head(&fi->direct_io_waitq); |
6b2fb799 | 3345 | fi->writepages = RB_ROOT; |
c2d0ad00 VG |
3346 | |
3347 | if (IS_ENABLED(CONFIG_FUSE_DAX)) | |
93a497b9 | 3348 | fuse_dax_inode_init(inode, flags); |
b6aeaded | 3349 | } |