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> |
5a8bee63 | 22 | #include <linux/file.h> |
b6aeaded | 23 | |
b9d54c6f MS |
24 | static int fuse_send_open(struct fuse_mount *fm, u64 nodeid, |
25 | unsigned int open_flags, int opcode, | |
26 | struct fuse_open_out *outargp) | |
b6aeaded | 27 | { |
b6aeaded | 28 | struct fuse_open_in inarg; |
7078187a | 29 | FUSE_ARGS(args); |
fd72faac MS |
30 | |
31 | memset(&inarg, 0, sizeof(inarg)); | |
b9d54c6f | 32 | inarg.flags = open_flags & ~(O_CREAT | O_EXCL | O_NOCTTY); |
fcee216b | 33 | if (!fm->fc->atomic_o_trunc) |
6ff958ed | 34 | inarg.flags &= ~O_TRUNC; |
643a666a VG |
35 | |
36 | if (fm->fc->handle_killpriv_v2 && | |
37 | (inarg.flags & O_TRUNC) && !capable(CAP_FSETID)) { | |
38 | inarg.open_flags |= FUSE_OPEN_KILL_SUIDGID; | |
39 | } | |
40 | ||
d5b48543 MS |
41 | args.opcode = opcode; |
42 | args.nodeid = nodeid; | |
43 | args.in_numargs = 1; | |
44 | args.in_args[0].size = sizeof(inarg); | |
45 | args.in_args[0].value = &inarg; | |
46 | args.out_numargs = 1; | |
47 | args.out_args[0].size = sizeof(*outargp); | |
48 | args.out_args[0].value = outargp; | |
fd72faac | 49 | |
fcee216b | 50 | return fuse_simple_request(fm, &args); |
fd72faac MS |
51 | } |
52 | ||
4cb54866 MS |
53 | struct fuse_release_args { |
54 | struct fuse_args args; | |
55 | struct fuse_release_in inarg; | |
56 | struct inode *inode; | |
57 | }; | |
58 | ||
fcee216b | 59 | struct fuse_file *fuse_file_alloc(struct fuse_mount *fm) |
fd72faac MS |
60 | { |
61 | struct fuse_file *ff; | |
6b2db28a | 62 | |
dc69e98c | 63 | ff = kzalloc(sizeof(struct fuse_file), GFP_KERNEL_ACCOUNT); |
6b2db28a TH |
64 | if (unlikely(!ff)) |
65 | return NULL; | |
66 | ||
fcee216b | 67 | ff->fm = fm; |
dc69e98c KK |
68 | ff->release_args = kzalloc(sizeof(*ff->release_args), |
69 | GFP_KERNEL_ACCOUNT); | |
4cb54866 | 70 | if (!ff->release_args) { |
6b2db28a TH |
71 | kfree(ff); |
72 | return NULL; | |
fd72faac | 73 | } |
6b2db28a TH |
74 | |
75 | INIT_LIST_HEAD(&ff->write_entry); | |
5d7bc7e8 | 76 | mutex_init(&ff->readdir.lock); |
4e8c2eb5 | 77 | refcount_set(&ff->count, 1); |
6b2db28a TH |
78 | RB_CLEAR_NODE(&ff->polled_node); |
79 | init_waitqueue_head(&ff->poll_wait); | |
80 | ||
fcee216b | 81 | ff->kh = atomic64_inc_return(&fm->fc->khctr); |
6b2db28a | 82 | |
fd72faac MS |
83 | return ff; |
84 | } | |
85 | ||
86 | void fuse_file_free(struct fuse_file *ff) | |
87 | { | |
4cb54866 | 88 | kfree(ff->release_args); |
5d7bc7e8 | 89 | mutex_destroy(&ff->readdir.lock); |
fd72faac MS |
90 | kfree(ff); |
91 | } | |
92 | ||
267d8444 | 93 | static struct fuse_file *fuse_file_get(struct fuse_file *ff) |
c756e0a4 | 94 | { |
4e8c2eb5 | 95 | refcount_inc(&ff->count); |
c756e0a4 MS |
96 | return ff; |
97 | } | |
98 | ||
fcee216b | 99 | static void fuse_release_end(struct fuse_mount *fm, struct fuse_args *args, |
4cb54866 | 100 | int error) |
819c4b3b | 101 | { |
4cb54866 MS |
102 | struct fuse_release_args *ra = container_of(args, typeof(*ra), args); |
103 | ||
104 | iput(ra->inode); | |
105 | kfree(ra); | |
819c4b3b MS |
106 | } |
107 | ||
2e64ff15 | 108 | static void fuse_file_put(struct fuse_file *ff, bool sync, bool isdir) |
c756e0a4 | 109 | { |
4e8c2eb5 | 110 | if (refcount_dec_and_test(&ff->count)) { |
4cb54866 | 111 | struct fuse_args *args = &ff->release_args->args; |
8b0797a4 | 112 | |
fcee216b | 113 | if (isdir ? ff->fm->fc->no_opendir : ff->fm->fc->no_open) { |
4cb54866 | 114 | /* Do nothing when client does not implement 'open' */ |
fcee216b | 115 | fuse_release_end(ff->fm, args, 0); |
7678ac50 | 116 | } else if (sync) { |
fcee216b MR |
117 | fuse_simple_request(ff->fm, args); |
118 | fuse_release_end(ff->fm, args, 0); | |
5a18ec17 | 119 | } else { |
4cb54866 | 120 | args->end = fuse_release_end; |
fcee216b | 121 | if (fuse_simple_background(ff->fm, args, |
4cb54866 | 122 | GFP_KERNEL | __GFP_NOFAIL)) |
fcee216b | 123 | fuse_release_end(ff->fm, args, -ENOTCONN); |
5a18ec17 | 124 | } |
c756e0a4 MS |
125 | kfree(ff); |
126 | } | |
127 | } | |
128 | ||
b9d54c6f MS |
129 | struct fuse_file *fuse_file_open(struct fuse_mount *fm, u64 nodeid, |
130 | unsigned int open_flags, bool isdir) | |
91fe96b4 | 131 | { |
fcee216b | 132 | struct fuse_conn *fc = fm->fc; |
91fe96b4 | 133 | struct fuse_file *ff; |
91fe96b4 MS |
134 | int opcode = isdir ? FUSE_OPENDIR : FUSE_OPEN; |
135 | ||
fcee216b | 136 | ff = fuse_file_alloc(fm); |
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); | |
d9a9ea94 | 143 | if (isdir ? !fc->no_opendir : !fc->no_open) { |
7678ac50 AG |
144 | struct fuse_open_out outarg; |
145 | int err; | |
146 | ||
b9d54c6f | 147 | err = fuse_send_open(fm, nodeid, open_flags, opcode, &outarg); |
7678ac50 AG |
148 | if (!err) { |
149 | ff->fh = outarg.fh; | |
150 | ff->open_flags = outarg.open_flags; | |
151 | ||
d9a9ea94 | 152 | } else if (err != -ENOSYS) { |
7678ac50 | 153 | fuse_file_free(ff); |
b9d54c6f | 154 | return ERR_PTR(err); |
7678ac50 | 155 | } else { |
d9a9ea94 CA |
156 | if (isdir) |
157 | fc->no_opendir = 1; | |
158 | else | |
159 | fc->no_open = 1; | |
7678ac50 | 160 | } |
91fe96b4 MS |
161 | } |
162 | ||
163 | if (isdir) | |
7678ac50 | 164 | ff->open_flags &= ~FOPEN_DIRECT_IO; |
91fe96b4 | 165 | |
91fe96b4 | 166 | ff->nodeid = nodeid; |
91fe96b4 | 167 | |
b9d54c6f MS |
168 | return ff; |
169 | } | |
170 | ||
171 | int fuse_do_open(struct fuse_mount *fm, u64 nodeid, struct file *file, | |
172 | bool isdir) | |
173 | { | |
174 | struct fuse_file *ff = fuse_file_open(fm, nodeid, file->f_flags, isdir); | |
175 | ||
176 | if (!IS_ERR(ff)) | |
177 | file->private_data = ff; | |
178 | ||
179 | return PTR_ERR_OR_ZERO(ff); | |
91fe96b4 | 180 | } |
08cbf542 | 181 | EXPORT_SYMBOL_GPL(fuse_do_open); |
91fe96b4 | 182 | |
650b22b9 PE |
183 | static void fuse_link_write_file(struct file *file) |
184 | { | |
185 | struct inode *inode = file_inode(file); | |
650b22b9 PE |
186 | struct fuse_inode *fi = get_fuse_inode(inode); |
187 | struct fuse_file *ff = file->private_data; | |
188 | /* | |
189 | * file may be written through mmap, so chain it onto the | |
190 | * inodes's write_file list | |
191 | */ | |
f15ecfef | 192 | spin_lock(&fi->lock); |
650b22b9 PE |
193 | if (list_empty(&ff->write_entry)) |
194 | list_add(&ff->write_entry, &fi->write_files); | |
f15ecfef | 195 | spin_unlock(&fi->lock); |
650b22b9 PE |
196 | } |
197 | ||
c7b7143c | 198 | void fuse_finish_open(struct inode *inode, struct file *file) |
fd72faac | 199 | { |
c7b7143c | 200 | struct fuse_file *ff = file->private_data; |
a0822c55 | 201 | struct fuse_conn *fc = get_fuse_conn(inode); |
c7b7143c | 202 | |
bbd84f33 KS |
203 | if (ff->open_flags & FOPEN_STREAM) |
204 | stream_open(inode, file); | |
205 | else if (ff->open_flags & FOPEN_NONSEEKABLE) | |
a7c1b990 | 206 | nonseekable_open(inode, file); |
76224355 | 207 | |
a0822c55 KS |
208 | if (fc->atomic_o_trunc && (file->f_flags & O_TRUNC)) { |
209 | struct fuse_inode *fi = get_fuse_inode(inode); | |
210 | ||
f15ecfef | 211 | spin_lock(&fi->lock); |
4510d86f | 212 | fi->attr_version = atomic64_inc_return(&fc->attr_version); |
a0822c55 | 213 | i_size_write(inode, 0); |
f15ecfef | 214 | spin_unlock(&fi->lock); |
20235b43 | 215 | file_update_time(file); |
fa5eee57 | 216 | fuse_invalidate_attr_mask(inode, FUSE_STATX_MODSIZE); |
a0822c55 | 217 | } |
4d99ff8f PE |
218 | if ((file->f_mode & FMODE_WRITE) && fc->writeback_cache) |
219 | fuse_link_write_file(file); | |
fd72faac MS |
220 | } |
221 | ||
91fe96b4 | 222 | int fuse_open_common(struct inode *inode, struct file *file, bool isdir) |
fd72faac | 223 | { |
fcee216b MR |
224 | struct fuse_mount *fm = get_fuse_mount(inode); |
225 | struct fuse_conn *fc = fm->fc; | |
b6aeaded | 226 | int err; |
e4648309 | 227 | bool is_wb_truncate = (file->f_flags & O_TRUNC) && |
75caeecd MP |
228 | fc->atomic_o_trunc && |
229 | fc->writeback_cache; | |
6ae330ca VG |
230 | bool dax_truncate = (file->f_flags & O_TRUNC) && |
231 | fc->atomic_o_trunc && FUSE_IS_DAX(inode); | |
b6aeaded | 232 | |
5d069dbe MS |
233 | if (fuse_is_bad(inode)) |
234 | return -EIO; | |
235 | ||
b6aeaded MS |
236 | err = generic_file_open(inode, file); |
237 | if (err) | |
238 | return err; | |
239 | ||
2fdbb8dd | 240 | if (is_wb_truncate || dax_truncate) |
5955102c | 241 | inode_lock(inode); |
75caeecd | 242 | |
6ae330ca | 243 | if (dax_truncate) { |
8bcbbe9c | 244 | filemap_invalidate_lock(inode->i_mapping); |
6ae330ca VG |
245 | err = fuse_dax_break_layouts(inode, 0, 0); |
246 | if (err) | |
2fdbb8dd | 247 | goto out_inode_unlock; |
6ae330ca | 248 | } |
b6aeaded | 249 | |
2fdbb8dd MS |
250 | if (is_wb_truncate || dax_truncate) |
251 | fuse_set_nowrite(inode); | |
252 | ||
fcee216b | 253 | err = fuse_do_open(fm, get_node_id(inode), file, isdir); |
75caeecd MP |
254 | if (!err) |
255 | fuse_finish_open(inode, file); | |
91fe96b4 | 256 | |
2fdbb8dd MS |
257 | if (is_wb_truncate || dax_truncate) |
258 | fuse_release_nowrite(inode); | |
259 | if (!err) { | |
260 | struct fuse_file *ff = file->private_data; | |
261 | ||
262 | if (fc->atomic_o_trunc && (file->f_flags & O_TRUNC)) | |
263 | truncate_pagecache(inode, 0); | |
264 | else if (!(ff->open_flags & FOPEN_KEEP_CACHE)) | |
265 | invalidate_inode_pages2(inode->i_mapping); | |
266 | } | |
6ae330ca | 267 | if (dax_truncate) |
8bcbbe9c | 268 | filemap_invalidate_unlock(inode->i_mapping); |
2fdbb8dd MS |
269 | out_inode_unlock: |
270 | if (is_wb_truncate || dax_truncate) | |
5955102c | 271 | inode_unlock(inode); |
75caeecd MP |
272 | |
273 | return err; | |
b6aeaded MS |
274 | } |
275 | ||
ebf84d0c | 276 | static void fuse_prepare_release(struct fuse_inode *fi, struct fuse_file *ff, |
54d601cb | 277 | unsigned int flags, int opcode) |
64c6d8ed | 278 | { |
fcee216b | 279 | struct fuse_conn *fc = ff->fm->fc; |
4cb54866 | 280 | struct fuse_release_args *ra = ff->release_args; |
b6aeaded | 281 | |
f15ecfef KT |
282 | /* Inode is NULL on error path of fuse_create_open() */ |
283 | if (likely(fi)) { | |
284 | spin_lock(&fi->lock); | |
285 | list_del(&ff->write_entry); | |
286 | spin_unlock(&fi->lock); | |
287 | } | |
8b0797a4 | 288 | spin_lock(&fc->lock); |
8b0797a4 MS |
289 | if (!RB_EMPTY_NODE(&ff->polled_node)) |
290 | rb_erase(&ff->polled_node, &fc->polled_files); | |
291 | spin_unlock(&fc->lock); | |
292 | ||
357ccf2b | 293 | wake_up_interruptible_all(&ff->poll_wait); |
8b0797a4 | 294 | |
4cb54866 MS |
295 | ra->inarg.fh = ff->fh; |
296 | ra->inarg.flags = flags; | |
297 | ra->args.in_numargs = 1; | |
298 | ra->args.in_args[0].size = sizeof(struct fuse_release_in); | |
299 | ra->args.in_args[0].value = &ra->inarg; | |
300 | ra->args.opcode = opcode; | |
301 | ra->args.nodeid = ff->nodeid; | |
302 | ra->args.force = true; | |
303 | ra->args.nocreds = true; | |
fd72faac MS |
304 | } |
305 | ||
b9d54c6f MS |
306 | void fuse_file_release(struct inode *inode, struct fuse_file *ff, |
307 | unsigned int open_flags, fl_owner_t id, bool isdir) | |
fd72faac | 308 | { |
b9d54c6f | 309 | struct fuse_inode *fi = get_fuse_inode(inode); |
4cb54866 | 310 | struct fuse_release_args *ra = ff->release_args; |
2e64ff15 | 311 | int opcode = isdir ? FUSE_RELEASEDIR : FUSE_RELEASE; |
6b2db28a | 312 | |
b9d54c6f | 313 | fuse_prepare_release(fi, ff, open_flags, opcode); |
6b2db28a | 314 | |
37fb3a30 | 315 | if (ff->flock) { |
4cb54866 | 316 | ra->inarg.release_flags |= FUSE_RELEASE_FLOCK_UNLOCK; |
b9d54c6f | 317 | ra->inarg.lock_owner = fuse_lock_owner_id(ff->fm->fc, id); |
37fb3a30 | 318 | } |
baebccbe | 319 | /* Hold inode until release is finished */ |
b9d54c6f | 320 | ra->inode = igrab(inode); |
6b2db28a | 321 | |
6b2db28a TH |
322 | /* |
323 | * Normally this will send the RELEASE request, however if | |
324 | * some asynchronous READ or WRITE requests are outstanding, | |
325 | * the sending will be delayed. | |
5a18ec17 MS |
326 | * |
327 | * Make the release synchronous if this is a fuseblk mount, | |
328 | * synchronous RELEASE is allowed (and desirable) in this case | |
329 | * because the server can be trusted not to screw up. | |
6b2db28a | 330 | */ |
fcee216b | 331 | fuse_file_put(ff, ff->fm->fc->destroy, isdir); |
b6aeaded MS |
332 | } |
333 | ||
b9d54c6f MS |
334 | void fuse_release_common(struct file *file, bool isdir) |
335 | { | |
336 | fuse_file_release(file_inode(file), file->private_data, file->f_flags, | |
337 | (fl_owner_t) file, isdir); | |
338 | } | |
339 | ||
04730fef MS |
340 | static int fuse_open(struct inode *inode, struct file *file) |
341 | { | |
91fe96b4 | 342 | return fuse_open_common(inode, file, false); |
04730fef MS |
343 | } |
344 | ||
345 | static int fuse_release(struct inode *inode, struct file *file) | |
346 | { | |
035ff33c MS |
347 | struct fuse_conn *fc = get_fuse_conn(inode); |
348 | ||
349 | /* | |
350 | * Dirty pages might remain despite write_inode_now() call from | |
351 | * fuse_flush() due to writes racing with the close. | |
352 | */ | |
353 | if (fc->writeback_cache) | |
354 | write_inode_now(inode, 1); | |
355 | ||
2e64ff15 | 356 | fuse_release_common(file, false); |
8b0797a4 MS |
357 | |
358 | /* return value is ignored by VFS */ | |
359 | return 0; | |
360 | } | |
361 | ||
54d601cb MS |
362 | void fuse_sync_release(struct fuse_inode *fi, struct fuse_file *ff, |
363 | unsigned int flags) | |
8b0797a4 | 364 | { |
4e8c2eb5 | 365 | WARN_ON(refcount_read(&ff->count) > 1); |
ebf84d0c | 366 | fuse_prepare_release(fi, ff, flags, FUSE_RELEASE); |
267d8444 MS |
367 | /* |
368 | * iput(NULL) is a no-op and since the refcount is 1 and everything's | |
369 | * synchronous, we are fine with not doing igrab() here" | |
370 | */ | |
2e64ff15 | 371 | fuse_file_put(ff, true, false); |
04730fef | 372 | } |
08cbf542 | 373 | EXPORT_SYMBOL_GPL(fuse_sync_release); |
04730fef | 374 | |
71421259 | 375 | /* |
9c8ef561 MS |
376 | * Scramble the ID space with XTEA, so that the value of the files_struct |
377 | * pointer is not exposed to userspace. | |
71421259 | 378 | */ |
f3332114 | 379 | u64 fuse_lock_owner_id(struct fuse_conn *fc, fl_owner_t id) |
71421259 | 380 | { |
9c8ef561 MS |
381 | u32 *k = fc->scramble_key; |
382 | u64 v = (unsigned long) id; | |
383 | u32 v0 = v; | |
384 | u32 v1 = v >> 32; | |
385 | u32 sum = 0; | |
386 | int i; | |
387 | ||
388 | for (i = 0; i < 32; i++) { | |
389 | v0 += ((v1 << 4 ^ v1 >> 5) + v1) ^ (sum + k[sum & 3]); | |
390 | sum += 0x9E3779B9; | |
391 | v1 += ((v0 << 4 ^ v0 >> 5) + v0) ^ (sum + k[sum>>11 & 3]); | |
392 | } | |
393 | ||
394 | return (u64) v0 + ((u64) v1 << 32); | |
71421259 MS |
395 | } |
396 | ||
33826ebb MS |
397 | struct fuse_writepage_args { |
398 | struct fuse_io_args ia; | |
6b2fb799 | 399 | struct rb_node writepages_entry; |
33826ebb MS |
400 | struct list_head queue_entry; |
401 | struct fuse_writepage_args *next; | |
402 | struct inode *inode; | |
660585b5 | 403 | struct fuse_sync_bucket *bucket; |
33826ebb MS |
404 | }; |
405 | ||
406 | static struct fuse_writepage_args *fuse_find_writeback(struct fuse_inode *fi, | |
2fe93bd4 MS |
407 | pgoff_t idx_from, pgoff_t idx_to) |
408 | { | |
6b2fb799 MP |
409 | struct rb_node *n; |
410 | ||
411 | n = fi->writepages.rb_node; | |
2fe93bd4 | 412 | |
6b2fb799 MP |
413 | while (n) { |
414 | struct fuse_writepage_args *wpa; | |
2fe93bd4 MS |
415 | pgoff_t curr_index; |
416 | ||
6b2fb799 | 417 | wpa = rb_entry(n, struct fuse_writepage_args, writepages_entry); |
33826ebb MS |
418 | WARN_ON(get_fuse_inode(wpa->inode) != fi); |
419 | curr_index = wpa->ia.write.in.offset >> PAGE_SHIFT; | |
6b2fb799 MP |
420 | if (idx_from >= curr_index + wpa->ia.ap.num_pages) |
421 | n = n->rb_right; | |
422 | else if (idx_to < curr_index) | |
423 | n = n->rb_left; | |
424 | else | |
33826ebb | 425 | return wpa; |
2fe93bd4 MS |
426 | } |
427 | return NULL; | |
428 | } | |
429 | ||
3be5a52b | 430 | /* |
ea8cd333 | 431 | * Check if any page in a range is under writeback |
3be5a52b MS |
432 | * |
433 | * This is currently done by walking the list of writepage requests | |
434 | * for the inode, which can be pretty inefficient. | |
435 | */ | |
ea8cd333 PE |
436 | static bool fuse_range_is_writeback(struct inode *inode, pgoff_t idx_from, |
437 | pgoff_t idx_to) | |
3be5a52b | 438 | { |
3be5a52b | 439 | struct fuse_inode *fi = get_fuse_inode(inode); |
2fe93bd4 | 440 | bool found; |
3be5a52b | 441 | |
f15ecfef | 442 | spin_lock(&fi->lock); |
2fe93bd4 | 443 | found = fuse_find_writeback(fi, idx_from, idx_to); |
f15ecfef | 444 | spin_unlock(&fi->lock); |
3be5a52b MS |
445 | |
446 | return found; | |
447 | } | |
448 | ||
ea8cd333 PE |
449 | static inline bool fuse_page_is_writeback(struct inode *inode, pgoff_t index) |
450 | { | |
451 | return fuse_range_is_writeback(inode, index, index); | |
452 | } | |
453 | ||
3be5a52b MS |
454 | /* |
455 | * Wait for page writeback to be completed. | |
456 | * | |
457 | * Since fuse doesn't rely on the VM writeback tracking, this has to | |
458 | * use some other means. | |
459 | */ | |
17b2cbe2 | 460 | static void fuse_wait_on_page_writeback(struct inode *inode, pgoff_t index) |
3be5a52b MS |
461 | { |
462 | struct fuse_inode *fi = get_fuse_inode(inode); | |
463 | ||
464 | wait_event(fi->page_waitq, !fuse_page_is_writeback(inode, index)); | |
3be5a52b MS |
465 | } |
466 | ||
fe38d7df MP |
467 | /* |
468 | * Wait for all pending writepages on the inode to finish. | |
469 | * | |
470 | * This is currently done by blocking further writes with FUSE_NOWRITE | |
471 | * and waiting for all sent writes to complete. | |
472 | * | |
473 | * This must be called under i_mutex, otherwise the FUSE_NOWRITE usage | |
474 | * could conflict with truncation. | |
475 | */ | |
476 | static void fuse_sync_writes(struct inode *inode) | |
477 | { | |
478 | fuse_set_nowrite(inode); | |
479 | fuse_release_nowrite(inode); | |
480 | } | |
481 | ||
5a8bee63 EB |
482 | struct fuse_flush_args { |
483 | struct fuse_args args; | |
b6aeaded | 484 | struct fuse_flush_in inarg; |
5a8bee63 EB |
485 | struct work_struct work; |
486 | struct file *file; | |
487 | }; | |
248d86e8 | 488 | |
5a8bee63 EB |
489 | static int fuse_do_flush(struct fuse_flush_args *fa) |
490 | { | |
491 | int err; | |
492 | struct inode *inode = file_inode(fa->file); | |
493 | struct fuse_mount *fm = get_fuse_mount(inode); | |
a390ccb3 | 494 | |
1e18bda8 | 495 | err = write_inode_now(inode, 1); |
fe38d7df | 496 | if (err) |
5a8bee63 | 497 | goto out; |
fe38d7df | 498 | |
5955102c | 499 | inode_lock(inode); |
fe38d7df | 500 | fuse_sync_writes(inode); |
5955102c | 501 | inode_unlock(inode); |
fe38d7df | 502 | |
5a8bee63 | 503 | err = filemap_check_errors(fa->file->f_mapping); |
9ebce595 | 504 | if (err) |
5a8bee63 | 505 | goto out; |
9ebce595 | 506 | |
614c026e | 507 | err = 0; |
fcee216b | 508 | if (fm->fc->no_flush) |
614c026e MS |
509 | goto inval_attr_out; |
510 | ||
5a8bee63 | 511 | err = fuse_simple_request(fm, &fa->args); |
b6aeaded | 512 | if (err == -ENOSYS) { |
fcee216b | 513 | fm->fc->no_flush = 1; |
b6aeaded MS |
514 | err = 0; |
515 | } | |
cf576c58 EG |
516 | |
517 | inval_attr_out: | |
518 | /* | |
519 | * In memory i_blocks is not maintained by fuse, if writeback cache is | |
520 | * enabled, i_blocks from cached attr may not be accurate. | |
521 | */ | |
fcee216b | 522 | if (!err && fm->fc->writeback_cache) |
fa5eee57 | 523 | fuse_invalidate_attr_mask(inode, STATX_BLOCKS); |
5a8bee63 EB |
524 | |
525 | out: | |
526 | fput(fa->file); | |
527 | kfree(fa); | |
b6aeaded MS |
528 | return err; |
529 | } | |
530 | ||
5a8bee63 EB |
531 | static void fuse_flush_async(struct work_struct *work) |
532 | { | |
533 | struct fuse_flush_args *fa = container_of(work, typeof(*fa), work); | |
534 | ||
535 | fuse_do_flush(fa); | |
536 | } | |
537 | ||
538 | static int fuse_flush(struct file *file, fl_owner_t id) | |
539 | { | |
540 | struct fuse_flush_args *fa; | |
541 | struct inode *inode = file_inode(file); | |
542 | struct fuse_mount *fm = get_fuse_mount(inode); | |
543 | struct fuse_file *ff = file->private_data; | |
544 | ||
545 | if (fuse_is_bad(inode)) | |
546 | return -EIO; | |
547 | ||
548 | if (ff->open_flags & FOPEN_NOFLUSH && !fm->fc->writeback_cache) | |
549 | return 0; | |
550 | ||
551 | fa = kzalloc(sizeof(*fa), GFP_KERNEL); | |
552 | if (!fa) | |
553 | return -ENOMEM; | |
554 | ||
555 | fa->inarg.fh = ff->fh; | |
556 | fa->inarg.lock_owner = fuse_lock_owner_id(fm->fc, id); | |
557 | fa->args.opcode = FUSE_FLUSH; | |
558 | fa->args.nodeid = get_node_id(inode); | |
559 | fa->args.in_numargs = 1; | |
560 | fa->args.in_args[0].size = sizeof(fa->inarg); | |
561 | fa->args.in_args[0].value = &fa->inarg; | |
562 | fa->args.force = true; | |
563 | fa->file = get_file(file); | |
564 | ||
565 | /* Don't wait if the task is exiting */ | |
566 | if (current->flags & PF_EXITING) { | |
567 | INIT_WORK(&fa->work, fuse_flush_async); | |
568 | schedule_work(&fa->work); | |
569 | return 0; | |
570 | } | |
571 | ||
572 | return fuse_do_flush(fa); | |
573 | } | |
574 | ||
02c24a82 | 575 | int fuse_fsync_common(struct file *file, loff_t start, loff_t end, |
a9c2d1e8 | 576 | int datasync, int opcode) |
b6aeaded | 577 | { |
7ea80859 | 578 | struct inode *inode = file->f_mapping->host; |
fcee216b | 579 | struct fuse_mount *fm = get_fuse_mount(inode); |
b6aeaded | 580 | struct fuse_file *ff = file->private_data; |
7078187a | 581 | FUSE_ARGS(args); |
b6aeaded | 582 | struct fuse_fsync_in inarg; |
a9c2d1e8 MS |
583 | |
584 | memset(&inarg, 0, sizeof(inarg)); | |
585 | inarg.fh = ff->fh; | |
154603fe | 586 | inarg.fsync_flags = datasync ? FUSE_FSYNC_FDATASYNC : 0; |
d5b48543 MS |
587 | args.opcode = opcode; |
588 | args.nodeid = get_node_id(inode); | |
589 | args.in_numargs = 1; | |
590 | args.in_args[0].size = sizeof(inarg); | |
591 | args.in_args[0].value = &inarg; | |
fcee216b | 592 | return fuse_simple_request(fm, &args); |
a9c2d1e8 MS |
593 | } |
594 | ||
595 | static int fuse_fsync(struct file *file, loff_t start, loff_t end, | |
596 | int datasync) | |
597 | { | |
598 | struct inode *inode = file->f_mapping->host; | |
599 | struct fuse_conn *fc = get_fuse_conn(inode); | |
b6aeaded MS |
600 | int err; |
601 | ||
5d069dbe | 602 | if (fuse_is_bad(inode)) |
248d86e8 MS |
603 | return -EIO; |
604 | ||
5955102c | 605 | inode_lock(inode); |
02c24a82 | 606 | |
3be5a52b MS |
607 | /* |
608 | * Start writeback against all dirty pages of the inode, then | |
609 | * wait for all outstanding writes, before sending the FSYNC | |
610 | * request. | |
611 | */ | |
7e51fe1d | 612 | err = file_write_and_wait_range(file, start, end); |
3be5a52b | 613 | if (err) |
02c24a82 | 614 | goto out; |
3be5a52b MS |
615 | |
616 | fuse_sync_writes(inode); | |
ac7f052b AK |
617 | |
618 | /* | |
619 | * Due to implementation of fuse writeback | |
7e51fe1d | 620 | * file_write_and_wait_range() does not catch errors. |
ac7f052b AK |
621 | * We have to do this directly after fuse_sync_writes() |
622 | */ | |
7e51fe1d | 623 | err = file_check_and_advance_wb_err(file); |
ac7f052b AK |
624 | if (err) |
625 | goto out; | |
626 | ||
1e18bda8 MS |
627 | err = sync_inode_metadata(inode, 1); |
628 | if (err) | |
629 | goto out; | |
3be5a52b | 630 | |
a9c2d1e8 | 631 | if (fc->no_fsync) |
22401e7b | 632 | goto out; |
b0aa7606 | 633 | |
a9c2d1e8 | 634 | err = fuse_fsync_common(file, start, end, datasync, FUSE_FSYNC); |
b6aeaded | 635 | if (err == -ENOSYS) { |
a9c2d1e8 | 636 | fc->no_fsync = 1; |
b6aeaded MS |
637 | err = 0; |
638 | } | |
02c24a82 | 639 | out: |
5955102c | 640 | inode_unlock(inode); |
b6aeaded | 641 | |
a9c2d1e8 | 642 | return err; |
82547981 MS |
643 | } |
644 | ||
00793ca5 MS |
645 | void fuse_read_args_fill(struct fuse_io_args *ia, struct file *file, loff_t pos, |
646 | size_t count, int opcode) | |
647 | { | |
648 | struct fuse_file *ff = file->private_data; | |
649 | struct fuse_args *args = &ia->ap.args; | |
650 | ||
651 | ia->read.in.fh = ff->fh; | |
652 | ia->read.in.offset = pos; | |
653 | ia->read.in.size = count; | |
654 | ia->read.in.flags = file->f_flags; | |
655 | args->opcode = opcode; | |
656 | args->nodeid = ff->nodeid; | |
657 | args->in_numargs = 1; | |
658 | args->in_args[0].size = sizeof(ia->read.in); | |
659 | args->in_args[0].value = &ia->read.in; | |
660 | args->out_argvar = true; | |
661 | args->out_numargs = 1; | |
662 | args->out_args[0].size = count; | |
663 | } | |
664 | ||
45ac96ed MS |
665 | static void fuse_release_user_pages(struct fuse_args_pages *ap, |
666 | bool should_dirty) | |
187c5c36 | 667 | { |
45ac96ed | 668 | unsigned int i; |
187c5c36 | 669 | |
45ac96ed | 670 | for (i = 0; i < ap->num_pages; i++) { |
8fba54ae | 671 | if (should_dirty) |
45ac96ed MS |
672 | set_page_dirty_lock(ap->pages[i]); |
673 | put_page(ap->pages[i]); | |
187c5c36 MP |
674 | } |
675 | } | |
676 | ||
744742d6 SF |
677 | static void fuse_io_release(struct kref *kref) |
678 | { | |
679 | kfree(container_of(kref, struct fuse_io_priv, refcnt)); | |
680 | } | |
681 | ||
9d5722b7 CH |
682 | static ssize_t fuse_get_res_by_io(struct fuse_io_priv *io) |
683 | { | |
684 | if (io->err) | |
685 | return io->err; | |
686 | ||
687 | if (io->bytes >= 0 && io->write) | |
688 | return -EIO; | |
689 | ||
690 | return io->bytes < 0 ? io->size : io->bytes; | |
691 | } | |
692 | ||
06bbb761 | 693 | /* |
01e9d11a MP |
694 | * In case of short read, the caller sets 'pos' to the position of |
695 | * actual end of fuse request in IO request. Otherwise, if bytes_requested | |
696 | * == bytes_transferred or rw == WRITE, the caller sets 'pos' to -1. | |
697 | * | |
698 | * An example: | |
c4e0cd4e | 699 | * User requested DIO read of 64K. It was split into two 32K fuse requests, |
01e9d11a MP |
700 | * both submitted asynchronously. The first of them was ACKed by userspace as |
701 | * fully completed (req->out.args[0].size == 32K) resulting in pos == -1. The | |
702 | * second request was ACKed as short, e.g. only 1K was read, resulting in | |
703 | * pos == 33K. | |
704 | * | |
705 | * Thus, when all fuse requests are completed, the minimal non-negative 'pos' | |
706 | * will be equal to the length of the longest contiguous fragment of | |
707 | * transferred data starting from the beginning of IO request. | |
708 | */ | |
709 | static void fuse_aio_complete(struct fuse_io_priv *io, int err, ssize_t pos) | |
710 | { | |
711 | int left; | |
712 | ||
713 | spin_lock(&io->lock); | |
714 | if (err) | |
715 | io->err = io->err ? : err; | |
716 | else if (pos >= 0 && (io->bytes < 0 || pos < io->bytes)) | |
717 | io->bytes = pos; | |
718 | ||
719 | left = --io->reqs; | |
7879c4e5 | 720 | if (!left && io->blocking) |
9d5722b7 | 721 | complete(io->done); |
01e9d11a MP |
722 | spin_unlock(&io->lock); |
723 | ||
7879c4e5 | 724 | if (!left && !io->blocking) { |
9d5722b7 | 725 | ssize_t res = fuse_get_res_by_io(io); |
01e9d11a | 726 | |
9d5722b7 CH |
727 | if (res >= 0) { |
728 | struct inode *inode = file_inode(io->iocb->ki_filp); | |
729 | struct fuse_conn *fc = get_fuse_conn(inode); | |
730 | struct fuse_inode *fi = get_fuse_inode(inode); | |
01e9d11a | 731 | |
f15ecfef | 732 | spin_lock(&fi->lock); |
4510d86f | 733 | fi->attr_version = atomic64_inc_return(&fc->attr_version); |
f15ecfef | 734 | spin_unlock(&fi->lock); |
01e9d11a MP |
735 | } |
736 | ||
6b19b766 | 737 | io->iocb->ki_complete(io->iocb, res); |
01e9d11a | 738 | } |
744742d6 SF |
739 | |
740 | kref_put(&io->refcnt, fuse_io_release); | |
01e9d11a MP |
741 | } |
742 | ||
45ac96ed MS |
743 | static struct fuse_io_args *fuse_io_alloc(struct fuse_io_priv *io, |
744 | unsigned int npages) | |
745 | { | |
746 | struct fuse_io_args *ia; | |
747 | ||
748 | ia = kzalloc(sizeof(*ia), GFP_KERNEL); | |
749 | if (ia) { | |
750 | ia->io = io; | |
751 | ia->ap.pages = fuse_pages_alloc(npages, GFP_KERNEL, | |
752 | &ia->ap.descs); | |
753 | if (!ia->ap.pages) { | |
754 | kfree(ia); | |
755 | ia = NULL; | |
756 | } | |
757 | } | |
758 | return ia; | |
759 | } | |
760 | ||
761 | static void fuse_io_free(struct fuse_io_args *ia) | |
01e9d11a | 762 | { |
45ac96ed MS |
763 | kfree(ia->ap.pages); |
764 | kfree(ia); | |
765 | } | |
766 | ||
fcee216b | 767 | static void fuse_aio_complete_req(struct fuse_mount *fm, struct fuse_args *args, |
45ac96ed MS |
768 | int err) |
769 | { | |
770 | struct fuse_io_args *ia = container_of(args, typeof(*ia), ap.args); | |
771 | struct fuse_io_priv *io = ia->io; | |
01e9d11a MP |
772 | ssize_t pos = -1; |
773 | ||
45ac96ed | 774 | fuse_release_user_pages(&ia->ap, io->should_dirty); |
01e9d11a | 775 | |
45ac96ed MS |
776 | if (err) { |
777 | /* Nothing */ | |
778 | } else if (io->write) { | |
779 | if (ia->write.out.size > ia->write.in.size) { | |
780 | err = -EIO; | |
781 | } else if (ia->write.in.size != ia->write.out.size) { | |
782 | pos = ia->write.in.offset - io->offset + | |
783 | ia->write.out.size; | |
784 | } | |
01e9d11a | 785 | } else { |
45ac96ed MS |
786 | u32 outsize = args->out_args[0].size; |
787 | ||
788 | if (ia->read.in.size != outsize) | |
789 | pos = ia->read.in.offset - io->offset + outsize; | |
01e9d11a MP |
790 | } |
791 | ||
45ac96ed MS |
792 | fuse_aio_complete(io, err, pos); |
793 | fuse_io_free(ia); | |
01e9d11a MP |
794 | } |
795 | ||
fcee216b | 796 | static ssize_t fuse_async_req_send(struct fuse_mount *fm, |
45ac96ed | 797 | struct fuse_io_args *ia, size_t num_bytes) |
01e9d11a | 798 | { |
45ac96ed MS |
799 | ssize_t err; |
800 | struct fuse_io_priv *io = ia->io; | |
801 | ||
01e9d11a | 802 | spin_lock(&io->lock); |
744742d6 | 803 | kref_get(&io->refcnt); |
01e9d11a MP |
804 | io->size += num_bytes; |
805 | io->reqs++; | |
806 | spin_unlock(&io->lock); | |
807 | ||
45ac96ed | 808 | ia->ap.args.end = fuse_aio_complete_req; |
bb737bbe | 809 | ia->ap.args.may_block = io->should_dirty; |
fcee216b | 810 | err = fuse_simple_background(fm, &ia->ap.args, GFP_KERNEL); |
f1ebdeff | 811 | if (err) |
fcee216b | 812 | fuse_aio_complete_req(fm, &ia->ap.args, err); |
01e9d11a | 813 | |
f1ebdeff | 814 | return num_bytes; |
01e9d11a MP |
815 | } |
816 | ||
45ac96ed MS |
817 | static ssize_t fuse_send_read(struct fuse_io_args *ia, loff_t pos, size_t count, |
818 | fl_owner_t owner) | |
04730fef | 819 | { |
45ac96ed | 820 | struct file *file = ia->io->iocb->ki_filp; |
2106cb18 | 821 | struct fuse_file *ff = file->private_data; |
fcee216b | 822 | struct fuse_mount *fm = ff->fm; |
f3332114 | 823 | |
45ac96ed | 824 | fuse_read_args_fill(ia, file, pos, count, FUSE_READ); |
f3332114 | 825 | if (owner != NULL) { |
45ac96ed | 826 | ia->read.in.read_flags |= FUSE_READ_LOCKOWNER; |
fcee216b | 827 | ia->read.in.lock_owner = fuse_lock_owner_id(fm->fc, owner); |
f3332114 | 828 | } |
36cf66ed | 829 | |
45ac96ed | 830 | if (ia->io->async) |
fcee216b | 831 | return fuse_async_req_send(fm, ia, count); |
36cf66ed | 832 | |
fcee216b | 833 | return fuse_simple_request(fm, &ia->ap.args); |
04730fef MS |
834 | } |
835 | ||
5c5c5e51 MS |
836 | static void fuse_read_update_size(struct inode *inode, loff_t size, |
837 | u64 attr_ver) | |
838 | { | |
839 | struct fuse_conn *fc = get_fuse_conn(inode); | |
840 | struct fuse_inode *fi = get_fuse_inode(inode); | |
841 | ||
f15ecfef | 842 | spin_lock(&fi->lock); |
484ce657 | 843 | if (attr_ver >= fi->attr_version && size < inode->i_size && |
06a7c3c2 | 844 | !test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) { |
4510d86f | 845 | fi->attr_version = atomic64_inc_return(&fc->attr_version); |
5c5c5e51 MS |
846 | i_size_write(inode, size); |
847 | } | |
f15ecfef | 848 | spin_unlock(&fi->lock); |
5c5c5e51 MS |
849 | } |
850 | ||
a0d45d84 | 851 | static void fuse_short_read(struct inode *inode, u64 attr_ver, size_t num_read, |
134831e3 | 852 | struct fuse_args_pages *ap) |
a92adc82 | 853 | { |
8373200b PE |
854 | struct fuse_conn *fc = get_fuse_conn(inode); |
855 | ||
a73d47f5 MS |
856 | /* |
857 | * If writeback_cache is enabled, a short read means there's a hole in | |
858 | * the file. Some data after the hole is in page cache, but has not | |
859 | * reached the client fs yet. So the hole is not present there. | |
860 | */ | |
861 | if (!fc->writeback_cache) { | |
134831e3 | 862 | loff_t pos = page_offset(ap->pages[0]) + num_read; |
8373200b PE |
863 | fuse_read_update_size(inode, pos, attr_ver); |
864 | } | |
a92adc82 PE |
865 | } |
866 | ||
482fce55 | 867 | static int fuse_do_readpage(struct file *file, struct page *page) |
b6aeaded MS |
868 | { |
869 | struct inode *inode = page->mapping->host; | |
fcee216b | 870 | struct fuse_mount *fm = get_fuse_mount(inode); |
5c5c5e51 | 871 | loff_t pos = page_offset(page); |
00793ca5 MS |
872 | struct fuse_page_desc desc = { .length = PAGE_SIZE }; |
873 | struct fuse_io_args ia = { | |
874 | .ap.args.page_zeroing = true, | |
875 | .ap.args.out_pages = true, | |
876 | .ap.num_pages = 1, | |
877 | .ap.pages = &page, | |
878 | .ap.descs = &desc, | |
879 | }; | |
880 | ssize_t res; | |
5c5c5e51 | 881 | u64 attr_ver; |
248d86e8 | 882 | |
3be5a52b | 883 | /* |
25985edc | 884 | * Page writeback can extend beyond the lifetime of the |
3be5a52b MS |
885 | * page-cache page, so make sure we read a properly synced |
886 | * page. | |
887 | */ | |
888 | fuse_wait_on_page_writeback(inode, page->index); | |
889 | ||
fcee216b | 890 | attr_ver = fuse_get_attr_version(fm->fc); |
5c5c5e51 | 891 | |
2f139829 MS |
892 | /* Don't overflow end offset */ |
893 | if (pos + (desc.length - 1) == LLONG_MAX) | |
894 | desc.length--; | |
895 | ||
00793ca5 | 896 | fuse_read_args_fill(&ia, file, pos, desc.length, FUSE_READ); |
fcee216b | 897 | res = fuse_simple_request(fm, &ia.ap.args); |
00793ca5 MS |
898 | if (res < 0) |
899 | return res; | |
900 | /* | |
901 | * Short read means EOF. If file size is larger, truncate it | |
902 | */ | |
903 | if (res < desc.length) | |
134831e3 | 904 | fuse_short_read(inode, attr_ver, res, &ia.ap); |
5c5c5e51 | 905 | |
00793ca5 | 906 | SetPageUptodate(page); |
482fce55 | 907 | |
00793ca5 | 908 | return 0; |
482fce55 MP |
909 | } |
910 | ||
5efd00e4 | 911 | static int fuse_read_folio(struct file *file, struct folio *folio) |
482fce55 | 912 | { |
5efd00e4 | 913 | struct page *page = &folio->page; |
482fce55 MP |
914 | struct inode *inode = page->mapping->host; |
915 | int err; | |
916 | ||
917 | err = -EIO; | |
5d069dbe | 918 | if (fuse_is_bad(inode)) |
482fce55 MP |
919 | goto out; |
920 | ||
921 | err = fuse_do_readpage(file, page); | |
451418fc | 922 | fuse_invalidate_atime(inode); |
b6aeaded MS |
923 | out: |
924 | unlock_page(page); | |
925 | return err; | |
926 | } | |
927 | ||
fcee216b | 928 | static void fuse_readpages_end(struct fuse_mount *fm, struct fuse_args *args, |
134831e3 | 929 | int err) |
db50b96c | 930 | { |
c1aa96a5 | 931 | int i; |
134831e3 MS |
932 | struct fuse_io_args *ia = container_of(args, typeof(*ia), ap.args); |
933 | struct fuse_args_pages *ap = &ia->ap; | |
934 | size_t count = ia->read.in.size; | |
935 | size_t num_read = args->out_args[0].size; | |
ce534fb0 | 936 | struct address_space *mapping = NULL; |
c1aa96a5 | 937 | |
134831e3 MS |
938 | for (i = 0; mapping == NULL && i < ap->num_pages; i++) |
939 | mapping = ap->pages[i]->mapping; | |
5c5c5e51 | 940 | |
ce534fb0 MS |
941 | if (mapping) { |
942 | struct inode *inode = mapping->host; | |
943 | ||
944 | /* | |
945 | * Short read means EOF. If file size is larger, truncate it | |
946 | */ | |
134831e3 MS |
947 | if (!err && num_read < count) |
948 | fuse_short_read(inode, ia->read.attr_ver, num_read, ap); | |
ce534fb0 | 949 | |
451418fc | 950 | fuse_invalidate_atime(inode); |
ce534fb0 | 951 | } |
c1aa96a5 | 952 | |
134831e3 MS |
953 | for (i = 0; i < ap->num_pages; i++) { |
954 | struct page *page = ap->pages[i]; | |
955 | ||
956 | if (!err) | |
db50b96c | 957 | SetPageUptodate(page); |
c1aa96a5 MS |
958 | else |
959 | SetPageError(page); | |
db50b96c | 960 | unlock_page(page); |
09cbfeaf | 961 | put_page(page); |
db50b96c | 962 | } |
134831e3 MS |
963 | if (ia->ff) |
964 | fuse_file_put(ia->ff, false, false); | |
965 | ||
966 | fuse_io_free(ia); | |
c1aa96a5 MS |
967 | } |
968 | ||
134831e3 | 969 | static void fuse_send_readpages(struct fuse_io_args *ia, struct file *file) |
c1aa96a5 | 970 | { |
2106cb18 | 971 | struct fuse_file *ff = file->private_data; |
fcee216b | 972 | struct fuse_mount *fm = ff->fm; |
134831e3 MS |
973 | struct fuse_args_pages *ap = &ia->ap; |
974 | loff_t pos = page_offset(ap->pages[0]); | |
975 | size_t count = ap->num_pages << PAGE_SHIFT; | |
7df1e988 | 976 | ssize_t res; |
134831e3 MS |
977 | int err; |
978 | ||
979 | ap->args.out_pages = true; | |
980 | ap->args.page_zeroing = true; | |
981 | ap->args.page_replace = true; | |
2f139829 MS |
982 | |
983 | /* Don't overflow end offset */ | |
984 | if (pos + (count - 1) == LLONG_MAX) { | |
985 | count--; | |
986 | ap->descs[ap->num_pages - 1].length--; | |
987 | } | |
988 | WARN_ON((loff_t) (pos + count) < 0); | |
989 | ||
134831e3 | 990 | fuse_read_args_fill(ia, file, pos, count, FUSE_READ); |
fcee216b MR |
991 | ia->read.attr_ver = fuse_get_attr_version(fm->fc); |
992 | if (fm->fc->async_read) { | |
134831e3 MS |
993 | ia->ff = fuse_file_get(ff); |
994 | ap->args.end = fuse_readpages_end; | |
fcee216b | 995 | err = fuse_simple_background(fm, &ap->args, GFP_KERNEL); |
134831e3 MS |
996 | if (!err) |
997 | return; | |
9cd68455 | 998 | } else { |
fcee216b | 999 | res = fuse_simple_request(fm, &ap->args); |
7df1e988 | 1000 | err = res < 0 ? res : 0; |
9cd68455 | 1001 | } |
fcee216b | 1002 | fuse_readpages_end(fm, &ap->args, err); |
db50b96c MS |
1003 | } |
1004 | ||
76a0294e | 1005 | static void fuse_readahead(struct readahead_control *rac) |
db50b96c | 1006 | { |
76a0294e | 1007 | struct inode *inode = rac->mapping->host; |
db50b96c | 1008 | struct fuse_conn *fc = get_fuse_conn(inode); |
76a0294e | 1009 | unsigned int i, max_pages, nr_pages = 0; |
db50b96c | 1010 | |
5d069dbe | 1011 | if (fuse_is_bad(inode)) |
76a0294e | 1012 | return; |
248d86e8 | 1013 | |
76a0294e MWO |
1014 | max_pages = min_t(unsigned int, fc->max_pages, |
1015 | fc->max_read / PAGE_SIZE); | |
db50b96c | 1016 | |
76a0294e MWO |
1017 | for (;;) { |
1018 | struct fuse_io_args *ia; | |
1019 | struct fuse_args_pages *ap; | |
1020 | ||
670d21c6 N |
1021 | if (fc->num_background >= fc->congestion_threshold && |
1022 | rac->ra->async_size >= readahead_count(rac)) | |
1023 | /* | |
1024 | * Congested and only async pages left, so skip the | |
1025 | * rest. | |
1026 | */ | |
1027 | break; | |
1028 | ||
76a0294e MWO |
1029 | nr_pages = readahead_count(rac) - nr_pages; |
1030 | if (nr_pages > max_pages) | |
1031 | nr_pages = max_pages; | |
1032 | if (nr_pages == 0) | |
1033 | break; | |
1034 | ia = fuse_io_alloc(NULL, nr_pages); | |
1035 | if (!ia) | |
1036 | return; | |
1037 | ap = &ia->ap; | |
1038 | nr_pages = __readahead_batch(rac, ap->pages, nr_pages); | |
1039 | for (i = 0; i < nr_pages; i++) { | |
1040 | fuse_wait_on_page_writeback(inode, | |
1041 | readahead_index(rac) + i); | |
1042 | ap->descs[i].length = PAGE_SIZE; | |
1043 | } | |
1044 | ap->num_pages = nr_pages; | |
1045 | fuse_send_readpages(ia, rac->file); | |
d3406ffa | 1046 | } |
db50b96c MS |
1047 | } |
1048 | ||
55752a3a | 1049 | static ssize_t fuse_cache_read_iter(struct kiocb *iocb, struct iov_iter *to) |
bcb4be80 MS |
1050 | { |
1051 | struct inode *inode = iocb->ki_filp->f_mapping->host; | |
a8894274 | 1052 | struct fuse_conn *fc = get_fuse_conn(inode); |
bcb4be80 | 1053 | |
a8894274 BF |
1054 | /* |
1055 | * In auto invalidate mode, always update attributes on read. | |
1056 | * Otherwise, only update if we attempt to read past EOF (to ensure | |
1057 | * i_size is up to date). | |
1058 | */ | |
1059 | if (fc->auto_inval_data || | |
37c20f16 | 1060 | (iocb->ki_pos + iov_iter_count(to) > i_size_read(inode))) { |
bcb4be80 | 1061 | int err; |
c6c745b8 | 1062 | err = fuse_update_attributes(inode, iocb->ki_filp, STATX_SIZE); |
bcb4be80 MS |
1063 | if (err) |
1064 | return err; | |
1065 | } | |
1066 | ||
37c20f16 | 1067 | return generic_file_read_iter(iocb, to); |
bcb4be80 MS |
1068 | } |
1069 | ||
338f2e3f MS |
1070 | static void fuse_write_args_fill(struct fuse_io_args *ia, struct fuse_file *ff, |
1071 | loff_t pos, size_t count) | |
1072 | { | |
1073 | struct fuse_args *args = &ia->ap.args; | |
1074 | ||
1075 | ia->write.in.fh = ff->fh; | |
1076 | ia->write.in.offset = pos; | |
1077 | ia->write.in.size = count; | |
1078 | args->opcode = FUSE_WRITE; | |
1079 | args->nodeid = ff->nodeid; | |
1080 | args->in_numargs = 2; | |
fcee216b | 1081 | if (ff->fm->fc->minor < 9) |
338f2e3f MS |
1082 | args->in_args[0].size = FUSE_COMPAT_WRITE_IN_SIZE; |
1083 | else | |
1084 | args->in_args[0].size = sizeof(ia->write.in); | |
1085 | args->in_args[0].value = &ia->write.in; | |
1086 | args->in_args[1].size = count; | |
1087 | args->out_numargs = 1; | |
1088 | args->out_args[0].size = sizeof(ia->write.out); | |
1089 | args->out_args[0].value = &ia->write.out; | |
1090 | } | |
1091 | ||
1092 | static unsigned int fuse_write_flags(struct kiocb *iocb) | |
1093 | { | |
1094 | unsigned int flags = iocb->ki_filp->f_flags; | |
1095 | ||
91b94c5d | 1096 | if (iocb_is_dsync(iocb)) |
338f2e3f MS |
1097 | flags |= O_DSYNC; |
1098 | if (iocb->ki_flags & IOCB_SYNC) | |
1099 | flags |= O_SYNC; | |
1100 | ||
1101 | return flags; | |
1102 | } | |
1103 | ||
45ac96ed MS |
1104 | static ssize_t fuse_send_write(struct fuse_io_args *ia, loff_t pos, |
1105 | size_t count, fl_owner_t owner) | |
b25e82e5 | 1106 | { |
45ac96ed | 1107 | struct kiocb *iocb = ia->io->iocb; |
e1c0eecb | 1108 | struct file *file = iocb->ki_filp; |
2106cb18 | 1109 | struct fuse_file *ff = file->private_data; |
fcee216b | 1110 | struct fuse_mount *fm = ff->fm; |
45ac96ed MS |
1111 | struct fuse_write_in *inarg = &ia->write.in; |
1112 | ssize_t err; | |
2d698b07 | 1113 | |
45ac96ed | 1114 | fuse_write_args_fill(ia, ff, pos, count); |
338f2e3f | 1115 | inarg->flags = fuse_write_flags(iocb); |
f3332114 | 1116 | if (owner != NULL) { |
f3332114 | 1117 | inarg->write_flags |= FUSE_WRITE_LOCKOWNER; |
fcee216b | 1118 | inarg->lock_owner = fuse_lock_owner_id(fm->fc, owner); |
f3332114 | 1119 | } |
36cf66ed | 1120 | |
45ac96ed | 1121 | if (ia->io->async) |
fcee216b | 1122 | return fuse_async_req_send(fm, ia, count); |
45ac96ed | 1123 | |
fcee216b | 1124 | err = fuse_simple_request(fm, &ia->ap.args); |
45ac96ed MS |
1125 | if (!err && ia->write.out.size > count) |
1126 | err = -EIO; | |
36cf66ed | 1127 | |
45ac96ed | 1128 | return err ?: ia->write.out.size; |
b6aeaded MS |
1129 | } |
1130 | ||
d347739a | 1131 | bool fuse_write_update_attr(struct inode *inode, loff_t pos, ssize_t written) |
854512ec MS |
1132 | { |
1133 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1134 | struct fuse_inode *fi = get_fuse_inode(inode); | |
b0aa7606 | 1135 | bool ret = false; |
854512ec | 1136 | |
f15ecfef | 1137 | spin_lock(&fi->lock); |
4510d86f | 1138 | fi->attr_version = atomic64_inc_return(&fc->attr_version); |
d347739a | 1139 | if (written > 0 && pos > inode->i_size) { |
854512ec | 1140 | i_size_write(inode, pos); |
b0aa7606 MP |
1141 | ret = true; |
1142 | } | |
f15ecfef | 1143 | spin_unlock(&fi->lock); |
b0aa7606 | 1144 | |
d347739a MS |
1145 | fuse_invalidate_attr_mask(inode, FUSE_STATX_MODSIZE); |
1146 | ||
b0aa7606 | 1147 | return ret; |
854512ec MS |
1148 | } |
1149 | ||
338f2e3f MS |
1150 | static ssize_t fuse_send_write_pages(struct fuse_io_args *ia, |
1151 | struct kiocb *iocb, struct inode *inode, | |
1152 | loff_t pos, size_t count) | |
ea9b9907 | 1153 | { |
338f2e3f MS |
1154 | struct fuse_args_pages *ap = &ia->ap; |
1155 | struct file *file = iocb->ki_filp; | |
1156 | struct fuse_file *ff = file->private_data; | |
fcee216b | 1157 | struct fuse_mount *fm = ff->fm; |
338f2e3f | 1158 | unsigned int offset, i; |
4f06dd92 | 1159 | bool short_write; |
338f2e3f | 1160 | int err; |
ea9b9907 | 1161 | |
338f2e3f MS |
1162 | for (i = 0; i < ap->num_pages; i++) |
1163 | fuse_wait_on_page_writeback(inode, ap->pages[i]->index); | |
ea9b9907 | 1164 | |
338f2e3f MS |
1165 | fuse_write_args_fill(ia, ff, pos, count); |
1166 | ia->write.in.flags = fuse_write_flags(iocb); | |
b8667395 VG |
1167 | if (fm->fc->handle_killpriv_v2 && !capable(CAP_FSETID)) |
1168 | ia->write.in.write_flags |= FUSE_WRITE_KILL_SUIDGID; | |
ea9b9907 | 1169 | |
fcee216b | 1170 | err = fuse_simple_request(fm, &ap->args); |
8aab336b MS |
1171 | if (!err && ia->write.out.size > count) |
1172 | err = -EIO; | |
338f2e3f | 1173 | |
4f06dd92 | 1174 | short_write = ia->write.out.size < count; |
338f2e3f MS |
1175 | offset = ap->descs[0].offset; |
1176 | count = ia->write.out.size; | |
1177 | for (i = 0; i < ap->num_pages; i++) { | |
1178 | struct page *page = ap->pages[i]; | |
ea9b9907 | 1179 | |
4f06dd92 VG |
1180 | if (err) { |
1181 | ClearPageUptodate(page); | |
1182 | } else { | |
1183 | if (count >= PAGE_SIZE - offset) | |
1184 | count -= PAGE_SIZE - offset; | |
1185 | else { | |
1186 | if (short_write) | |
1187 | ClearPageUptodate(page); | |
1188 | count = 0; | |
1189 | } | |
1190 | offset = 0; | |
1191 | } | |
1192 | if (ia->write.page_locked && (i == ap->num_pages - 1)) | |
1193 | unlock_page(page); | |
09cbfeaf | 1194 | put_page(page); |
ea9b9907 NP |
1195 | } |
1196 | ||
338f2e3f | 1197 | return err; |
ea9b9907 NP |
1198 | } |
1199 | ||
4f06dd92 | 1200 | static ssize_t fuse_fill_write_pages(struct fuse_io_args *ia, |
338f2e3f MS |
1201 | struct address_space *mapping, |
1202 | struct iov_iter *ii, loff_t pos, | |
1203 | unsigned int max_pages) | |
ea9b9907 | 1204 | { |
4f06dd92 | 1205 | struct fuse_args_pages *ap = &ia->ap; |
ea9b9907 | 1206 | struct fuse_conn *fc = get_fuse_conn(mapping->host); |
09cbfeaf | 1207 | unsigned offset = pos & (PAGE_SIZE - 1); |
ea9b9907 NP |
1208 | size_t count = 0; |
1209 | int err; | |
1210 | ||
338f2e3f MS |
1211 | ap->args.in_pages = true; |
1212 | ap->descs[0].offset = offset; | |
ea9b9907 NP |
1213 | |
1214 | do { | |
1215 | size_t tmp; | |
1216 | struct page *page; | |
09cbfeaf KS |
1217 | pgoff_t index = pos >> PAGE_SHIFT; |
1218 | size_t bytes = min_t(size_t, PAGE_SIZE - offset, | |
ea9b9907 NP |
1219 | iov_iter_count(ii)); |
1220 | ||
1221 | bytes = min_t(size_t, bytes, fc->max_write - count); | |
1222 | ||
1223 | again: | |
1224 | err = -EFAULT; | |
a6294593 | 1225 | if (fault_in_iov_iter_readable(ii, bytes)) |
ea9b9907 NP |
1226 | break; |
1227 | ||
1228 | err = -ENOMEM; | |
b7446e7c | 1229 | page = grab_cache_page_write_begin(mapping, index); |
ea9b9907 NP |
1230 | if (!page) |
1231 | break; | |
1232 | ||
931e80e4 | 1233 | if (mapping_writably_mapped(mapping)) |
1234 | flush_dcache_page(page); | |
1235 | ||
f0b65f39 | 1236 | tmp = copy_page_from_iter_atomic(page, offset, bytes, ii); |
ea9b9907 NP |
1237 | flush_dcache_page(page); |
1238 | ||
1239 | if (!tmp) { | |
1240 | unlock_page(page); | |
09cbfeaf | 1241 | put_page(page); |
ea9b9907 NP |
1242 | goto again; |
1243 | } | |
1244 | ||
1245 | err = 0; | |
338f2e3f MS |
1246 | ap->pages[ap->num_pages] = page; |
1247 | ap->descs[ap->num_pages].length = tmp; | |
1248 | ap->num_pages++; | |
ea9b9907 | 1249 | |
ea9b9907 NP |
1250 | count += tmp; |
1251 | pos += tmp; | |
1252 | offset += tmp; | |
09cbfeaf | 1253 | if (offset == PAGE_SIZE) |
ea9b9907 NP |
1254 | offset = 0; |
1255 | ||
4f06dd92 VG |
1256 | /* If we copied full page, mark it uptodate */ |
1257 | if (tmp == PAGE_SIZE) | |
1258 | SetPageUptodate(page); | |
1259 | ||
1260 | if (PageUptodate(page)) { | |
1261 | unlock_page(page); | |
1262 | } else { | |
1263 | ia->write.page_locked = true; | |
1264 | break; | |
1265 | } | |
78bb6cb9 MS |
1266 | if (!fc->big_writes) |
1267 | break; | |
ea9b9907 | 1268 | } while (iov_iter_count(ii) && count < fc->max_write && |
338f2e3f | 1269 | ap->num_pages < max_pages && offset == 0); |
ea9b9907 NP |
1270 | |
1271 | return count > 0 ? count : err; | |
1272 | } | |
1273 | ||
5da784cc CS |
1274 | static inline unsigned int fuse_wr_pages(loff_t pos, size_t len, |
1275 | unsigned int max_pages) | |
d07f09f5 | 1276 | { |
5da784cc | 1277 | return min_t(unsigned int, |
09cbfeaf KS |
1278 | ((pos + len - 1) >> PAGE_SHIFT) - |
1279 | (pos >> PAGE_SHIFT) + 1, | |
5da784cc | 1280 | max_pages); |
d07f09f5 MP |
1281 | } |
1282 | ||
e1c0eecb | 1283 | static ssize_t fuse_perform_write(struct kiocb *iocb, |
ea9b9907 NP |
1284 | struct address_space *mapping, |
1285 | struct iov_iter *ii, loff_t pos) | |
1286 | { | |
1287 | struct inode *inode = mapping->host; | |
1288 | struct fuse_conn *fc = get_fuse_conn(inode); | |
06a7c3c2 | 1289 | struct fuse_inode *fi = get_fuse_inode(inode); |
ea9b9907 NP |
1290 | int err = 0; |
1291 | ssize_t res = 0; | |
1292 | ||
06a7c3c2 MP |
1293 | if (inode->i_size < pos + iov_iter_count(ii)) |
1294 | set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); | |
1295 | ||
ea9b9907 | 1296 | do { |
ea9b9907 | 1297 | ssize_t count; |
338f2e3f MS |
1298 | struct fuse_io_args ia = {}; |
1299 | struct fuse_args_pages *ap = &ia.ap; | |
5da784cc CS |
1300 | unsigned int nr_pages = fuse_wr_pages(pos, iov_iter_count(ii), |
1301 | fc->max_pages); | |
ea9b9907 | 1302 | |
338f2e3f MS |
1303 | ap->pages = fuse_pages_alloc(nr_pages, GFP_KERNEL, &ap->descs); |
1304 | if (!ap->pages) { | |
1305 | err = -ENOMEM; | |
ea9b9907 NP |
1306 | break; |
1307 | } | |
1308 | ||
4f06dd92 | 1309 | count = fuse_fill_write_pages(&ia, mapping, ii, pos, nr_pages); |
ea9b9907 NP |
1310 | if (count <= 0) { |
1311 | err = count; | |
1312 | } else { | |
338f2e3f MS |
1313 | err = fuse_send_write_pages(&ia, iocb, inode, |
1314 | pos, count); | |
ea9b9907 | 1315 | if (!err) { |
338f2e3f MS |
1316 | size_t num_written = ia.write.out.size; |
1317 | ||
ea9b9907 NP |
1318 | res += num_written; |
1319 | pos += num_written; | |
1320 | ||
1321 | /* break out of the loop on short write */ | |
1322 | if (num_written != count) | |
1323 | err = -EIO; | |
1324 | } | |
1325 | } | |
338f2e3f | 1326 | kfree(ap->pages); |
ea9b9907 NP |
1327 | } while (!err && iov_iter_count(ii)); |
1328 | ||
d347739a | 1329 | fuse_write_update_attr(inode, pos, res); |
06a7c3c2 | 1330 | clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); |
ea9b9907 NP |
1331 | |
1332 | return res > 0 ? res : err; | |
1333 | } | |
1334 | ||
55752a3a | 1335 | static ssize_t fuse_cache_write_iter(struct kiocb *iocb, struct iov_iter *from) |
ea9b9907 NP |
1336 | { |
1337 | struct file *file = iocb->ki_filp; | |
1338 | struct address_space *mapping = file->f_mapping; | |
ea9b9907 | 1339 | ssize_t written = 0; |
4273b793 | 1340 | ssize_t written_buffered = 0; |
ea9b9907 NP |
1341 | struct inode *inode = mapping->host; |
1342 | ssize_t err; | |
8981bdfd | 1343 | struct fuse_conn *fc = get_fuse_conn(inode); |
4273b793 | 1344 | loff_t endbyte = 0; |
ea9b9907 | 1345 | |
8981bdfd | 1346 | if (fc->writeback_cache) { |
4d99ff8f | 1347 | /* Update size (EOF optimization) and mode (SUID clearing) */ |
c6c745b8 MS |
1348 | err = fuse_update_attributes(mapping->host, file, |
1349 | STATX_SIZE | STATX_MODE); | |
4d99ff8f PE |
1350 | if (err) |
1351 | return err; | |
1352 | ||
8981bdfd | 1353 | if (fc->handle_killpriv_v2 && |
9452e93e CB |
1354 | setattr_should_drop_suidgid(&nop_mnt_idmap, |
1355 | file_inode(file))) { | |
8981bdfd VG |
1356 | goto writethrough; |
1357 | } | |
1358 | ||
84c3d55c | 1359 | return generic_file_write_iter(iocb, from); |
4d99ff8f PE |
1360 | } |
1361 | ||
8981bdfd | 1362 | writethrough: |
5955102c | 1363 | inode_lock(inode); |
ea9b9907 NP |
1364 | |
1365 | /* We can write back this queue in page reclaim */ | |
de1414a6 | 1366 | current->backing_dev_info = inode_to_bdi(inode); |
ea9b9907 | 1367 | |
3309dd04 AV |
1368 | err = generic_write_checks(iocb, from); |
1369 | if (err <= 0) | |
ea9b9907 NP |
1370 | goto out; |
1371 | ||
5fa8e0a1 | 1372 | err = file_remove_privs(file); |
ea9b9907 NP |
1373 | if (err) |
1374 | goto out; | |
1375 | ||
c3b2da31 JB |
1376 | err = file_update_time(file); |
1377 | if (err) | |
1378 | goto out; | |
ea9b9907 | 1379 | |
2ba48ce5 | 1380 | if (iocb->ki_flags & IOCB_DIRECT) { |
3309dd04 | 1381 | loff_t pos = iocb->ki_pos; |
1af5bb49 | 1382 | written = generic_file_direct_write(iocb, from); |
84c3d55c | 1383 | if (written < 0 || !iov_iter_count(from)) |
4273b793 AA |
1384 | goto out; |
1385 | ||
1386 | pos += written; | |
ea9b9907 | 1387 | |
e1c0eecb | 1388 | written_buffered = fuse_perform_write(iocb, mapping, from, pos); |
4273b793 AA |
1389 | if (written_buffered < 0) { |
1390 | err = written_buffered; | |
1391 | goto out; | |
1392 | } | |
1393 | endbyte = pos + written_buffered - 1; | |
1394 | ||
1395 | err = filemap_write_and_wait_range(file->f_mapping, pos, | |
1396 | endbyte); | |
1397 | if (err) | |
1398 | goto out; | |
1399 | ||
1400 | invalidate_mapping_pages(file->f_mapping, | |
09cbfeaf KS |
1401 | pos >> PAGE_SHIFT, |
1402 | endbyte >> PAGE_SHIFT); | |
4273b793 AA |
1403 | |
1404 | written += written_buffered; | |
1405 | iocb->ki_pos = pos + written_buffered; | |
1406 | } else { | |
e1c0eecb | 1407 | written = fuse_perform_write(iocb, mapping, from, iocb->ki_pos); |
4273b793 | 1408 | if (written >= 0) |
3309dd04 | 1409 | iocb->ki_pos += written; |
4273b793 | 1410 | } |
ea9b9907 NP |
1411 | out: |
1412 | current->backing_dev_info = NULL; | |
5955102c | 1413 | inode_unlock(inode); |
e1c0eecb MS |
1414 | if (written > 0) |
1415 | written = generic_write_sync(iocb, written); | |
ea9b9907 NP |
1416 | |
1417 | return written ? written : err; | |
1418 | } | |
1419 | ||
7c190c8b MP |
1420 | static inline unsigned long fuse_get_user_addr(const struct iov_iter *ii) |
1421 | { | |
de4f5fed | 1422 | return (unsigned long)iter_iov(ii)->iov_base + ii->iov_offset; |
7c190c8b MP |
1423 | } |
1424 | ||
1425 | static inline size_t fuse_get_frag_size(const struct iov_iter *ii, | |
1426 | size_t max_size) | |
1427 | { | |
1428 | return min(iov_iter_single_seg_count(ii), max_size); | |
1429 | } | |
1430 | ||
45ac96ed MS |
1431 | static int fuse_get_user_pages(struct fuse_args_pages *ap, struct iov_iter *ii, |
1432 | size_t *nbytesp, int write, | |
1433 | unsigned int max_pages) | |
413ef8cb | 1434 | { |
7c190c8b | 1435 | size_t nbytes = 0; /* # bytes already packed in req */ |
742f9927 | 1436 | ssize_t ret = 0; |
b98d023a | 1437 | |
f4975c67 | 1438 | /* Special case for kernel I/O: can copy directly into the buffer */ |
00e23707 | 1439 | if (iov_iter_is_kvec(ii)) { |
7c190c8b MP |
1440 | unsigned long user_addr = fuse_get_user_addr(ii); |
1441 | size_t frag_size = fuse_get_frag_size(ii, *nbytesp); | |
1442 | ||
f4975c67 | 1443 | if (write) |
45ac96ed | 1444 | ap->args.in_args[1].value = (void *) user_addr; |
f4975c67 | 1445 | else |
45ac96ed | 1446 | ap->args.out_args[0].value = (void *) user_addr; |
f4975c67 | 1447 | |
b98d023a MP |
1448 | iov_iter_advance(ii, frag_size); |
1449 | *nbytesp = frag_size; | |
f4975c67 MS |
1450 | return 0; |
1451 | } | |
413ef8cb | 1452 | |
45ac96ed | 1453 | while (nbytes < *nbytesp && ap->num_pages < max_pages) { |
7c190c8b | 1454 | unsigned npages; |
f67da30c | 1455 | size_t start; |
1ef255e2 | 1456 | ret = iov_iter_get_pages2(ii, &ap->pages[ap->num_pages], |
2c80929c | 1457 | *nbytesp - nbytes, |
45ac96ed | 1458 | max_pages - ap->num_pages, |
c7f3888a | 1459 | &start); |
7c190c8b | 1460 | if (ret < 0) |
742f9927 | 1461 | break; |
7c190c8b | 1462 | |
c9c37e2e | 1463 | nbytes += ret; |
7c190c8b | 1464 | |
c9c37e2e | 1465 | ret += start; |
6c88632b | 1466 | npages = DIV_ROUND_UP(ret, PAGE_SIZE); |
7c190c8b | 1467 | |
45ac96ed MS |
1468 | ap->descs[ap->num_pages].offset = start; |
1469 | fuse_page_descs_length_init(ap->descs, ap->num_pages, npages); | |
7c190c8b | 1470 | |
45ac96ed MS |
1471 | ap->num_pages += npages; |
1472 | ap->descs[ap->num_pages - 1].length -= | |
c9c37e2e | 1473 | (PAGE_SIZE - ret) & (PAGE_SIZE - 1); |
7c190c8b | 1474 | } |
f4975c67 | 1475 | |
0c4bcfde | 1476 | ap->args.user_pages = true; |
f4975c67 | 1477 | if (write) |
cabdb4fa | 1478 | ap->args.in_pages = true; |
f4975c67 | 1479 | else |
cabdb4fa | 1480 | ap->args.out_pages = true; |
f4975c67 | 1481 | |
7c190c8b | 1482 | *nbytesp = nbytes; |
f4975c67 | 1483 | |
2c932d4c | 1484 | return ret < 0 ? ret : 0; |
413ef8cb MS |
1485 | } |
1486 | ||
d22a943f AV |
1487 | ssize_t fuse_direct_io(struct fuse_io_priv *io, struct iov_iter *iter, |
1488 | loff_t *ppos, int flags) | |
413ef8cb | 1489 | { |
ea8cd333 PE |
1490 | int write = flags & FUSE_DIO_WRITE; |
1491 | int cuse = flags & FUSE_DIO_CUSE; | |
e1c0eecb | 1492 | struct file *file = io->iocb->ki_filp; |
ea8cd333 | 1493 | struct inode *inode = file->f_mapping->host; |
2106cb18 | 1494 | struct fuse_file *ff = file->private_data; |
fcee216b | 1495 | struct fuse_conn *fc = ff->fm->fc; |
413ef8cb MS |
1496 | size_t nmax = write ? fc->max_write : fc->max_read; |
1497 | loff_t pos = *ppos; | |
d22a943f | 1498 | size_t count = iov_iter_count(iter); |
09cbfeaf KS |
1499 | pgoff_t idx_from = pos >> PAGE_SHIFT; |
1500 | pgoff_t idx_to = (pos + count - 1) >> PAGE_SHIFT; | |
413ef8cb | 1501 | ssize_t res = 0; |
742f9927 | 1502 | int err = 0; |
45ac96ed MS |
1503 | struct fuse_io_args *ia; |
1504 | unsigned int max_pages; | |
248d86e8 | 1505 | |
45ac96ed MS |
1506 | max_pages = iov_iter_npages(iter, fc->max_pages); |
1507 | ia = fuse_io_alloc(io, max_pages); | |
1508 | if (!ia) | |
1509 | return -ENOMEM; | |
413ef8cb | 1510 | |
ea8cd333 PE |
1511 | if (!cuse && fuse_range_is_writeback(inode, idx_from, idx_to)) { |
1512 | if (!write) | |
5955102c | 1513 | inode_lock(inode); |
ea8cd333 PE |
1514 | fuse_sync_writes(inode); |
1515 | if (!write) | |
5955102c | 1516 | inode_unlock(inode); |
ea8cd333 PE |
1517 | } |
1518 | ||
fcb14cb1 | 1519 | io->should_dirty = !write && user_backed_iter(iter); |
413ef8cb | 1520 | while (count) { |
45ac96ed | 1521 | ssize_t nres; |
2106cb18 | 1522 | fl_owner_t owner = current->files; |
f4975c67 | 1523 | size_t nbytes = min(count, nmax); |
45ac96ed MS |
1524 | |
1525 | err = fuse_get_user_pages(&ia->ap, iter, &nbytes, write, | |
1526 | max_pages); | |
742f9927 | 1527 | if (err && !nbytes) |
413ef8cb | 1528 | break; |
f4975c67 | 1529 | |
4a2abf99 | 1530 | if (write) { |
45ac96ed | 1531 | if (!capable(CAP_FSETID)) |
10c52c84 | 1532 | ia->write.in.write_flags |= FUSE_WRITE_KILL_SUIDGID; |
4a2abf99 | 1533 | |
45ac96ed | 1534 | nres = fuse_send_write(ia, pos, nbytes, owner); |
4a2abf99 | 1535 | } else { |
45ac96ed | 1536 | nres = fuse_send_read(ia, pos, nbytes, owner); |
4a2abf99 | 1537 | } |
2106cb18 | 1538 | |
45ac96ed MS |
1539 | if (!io->async || nres < 0) { |
1540 | fuse_release_user_pages(&ia->ap, io->should_dirty); | |
1541 | fuse_io_free(ia); | |
1542 | } | |
1543 | ia = NULL; | |
1544 | if (nres < 0) { | |
f658adee | 1545 | iov_iter_revert(iter, nbytes); |
45ac96ed | 1546 | err = nres; |
413ef8cb MS |
1547 | break; |
1548 | } | |
45ac96ed MS |
1549 | WARN_ON(nres > nbytes); |
1550 | ||
413ef8cb MS |
1551 | count -= nres; |
1552 | res += nres; | |
1553 | pos += nres; | |
f658adee MS |
1554 | if (nres != nbytes) { |
1555 | iov_iter_revert(iter, nbytes - nres); | |
413ef8cb | 1556 | break; |
f658adee | 1557 | } |
56cf34ff | 1558 | if (count) { |
45ac96ed MS |
1559 | max_pages = iov_iter_npages(iter, fc->max_pages); |
1560 | ia = fuse_io_alloc(io, max_pages); | |
1561 | if (!ia) | |
56cf34ff MS |
1562 | break; |
1563 | } | |
413ef8cb | 1564 | } |
45ac96ed MS |
1565 | if (ia) |
1566 | fuse_io_free(ia); | |
d09cb9d7 | 1567 | if (res > 0) |
413ef8cb | 1568 | *ppos = pos; |
413ef8cb | 1569 | |
742f9927 | 1570 | return res > 0 ? res : err; |
413ef8cb | 1571 | } |
08cbf542 | 1572 | EXPORT_SYMBOL_GPL(fuse_direct_io); |
413ef8cb | 1573 | |
36cf66ed | 1574 | static ssize_t __fuse_direct_read(struct fuse_io_priv *io, |
d22a943f AV |
1575 | struct iov_iter *iter, |
1576 | loff_t *ppos) | |
413ef8cb | 1577 | { |
d09cb9d7 | 1578 | ssize_t res; |
e1c0eecb | 1579 | struct inode *inode = file_inode(io->iocb->ki_filp); |
d09cb9d7 | 1580 | |
d22a943f | 1581 | res = fuse_direct_io(io, iter, ppos, 0); |
d09cb9d7 | 1582 | |
9a2eb24d | 1583 | fuse_invalidate_atime(inode); |
d09cb9d7 MS |
1584 | |
1585 | return res; | |
413ef8cb MS |
1586 | } |
1587 | ||
23c94e1c MR |
1588 | static ssize_t fuse_direct_IO(struct kiocb *iocb, struct iov_iter *iter); |
1589 | ||
15316263 | 1590 | static ssize_t fuse_direct_read_iter(struct kiocb *iocb, struct iov_iter *to) |
b98d023a | 1591 | { |
23c94e1c MR |
1592 | ssize_t res; |
1593 | ||
1594 | if (!is_sync_kiocb(iocb) && iocb->ki_flags & IOCB_DIRECT) { | |
23c94e1c MR |
1595 | res = fuse_direct_IO(iocb, to); |
1596 | } else { | |
1597 | struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb); | |
1598 | ||
1599 | res = __fuse_direct_read(&io, to, &iocb->ki_pos); | |
1600 | } | |
1601 | ||
1602 | return res; | |
b98d023a MP |
1603 | } |
1604 | ||
15352405 DS |
1605 | static bool fuse_direct_write_extending_i_size(struct kiocb *iocb, |
1606 | struct iov_iter *iter) | |
1607 | { | |
1608 | struct inode *inode = file_inode(iocb->ki_filp); | |
1609 | ||
1610 | return iocb->ki_pos + iov_iter_count(iter) > i_size_read(inode); | |
1611 | } | |
1612 | ||
15316263 | 1613 | static ssize_t fuse_direct_write_iter(struct kiocb *iocb, struct iov_iter *from) |
4273b793 | 1614 | { |
e1c0eecb | 1615 | struct inode *inode = file_inode(iocb->ki_filp); |
15352405 DS |
1616 | struct file *file = iocb->ki_filp; |
1617 | struct fuse_file *ff = file->private_data; | |
e1c0eecb | 1618 | struct fuse_io_priv io = FUSE_IO_PRIV_SYNC(iocb); |
15316263 | 1619 | ssize_t res; |
15352405 DS |
1620 | bool exclusive_lock = |
1621 | !(ff->open_flags & FOPEN_PARALLEL_DIRECT_WRITES) || | |
1622 | iocb->ki_flags & IOCB_APPEND || | |
1623 | fuse_direct_write_extending_i_size(iocb, from); | |
1624 | ||
1625 | /* | |
1626 | * Take exclusive lock if | |
1627 | * - Parallel direct writes are disabled - a user space decision | |
1628 | * - Parallel direct writes are enabled and i_size is being extended. | |
1629 | * This might not be needed at all, but needs further investigation. | |
1630 | */ | |
1631 | if (exclusive_lock) | |
1632 | inode_lock(inode); | |
1633 | else { | |
1634 | inode_lock_shared(inode); | |
1635 | ||
1636 | /* A race with truncate might have come up as the decision for | |
1637 | * the lock type was done without holding the lock, check again. | |
1638 | */ | |
1639 | if (fuse_direct_write_extending_i_size(iocb, from)) { | |
1640 | inode_unlock_shared(inode); | |
1641 | inode_lock(inode); | |
1642 | exclusive_lock = true; | |
1643 | } | |
1644 | } | |
4273b793 | 1645 | |
3309dd04 | 1646 | res = generic_write_checks(iocb, from); |
23c94e1c MR |
1647 | if (res > 0) { |
1648 | if (!is_sync_kiocb(iocb) && iocb->ki_flags & IOCB_DIRECT) { | |
1649 | res = fuse_direct_IO(iocb, from); | |
1650 | } else { | |
1651 | res = fuse_direct_io(&io, from, &iocb->ki_pos, | |
1652 | FUSE_DIO_WRITE); | |
d347739a | 1653 | fuse_write_update_attr(inode, iocb->ki_pos, res); |
23c94e1c MR |
1654 | } |
1655 | } | |
15352405 DS |
1656 | if (exclusive_lock) |
1657 | inode_unlock(inode); | |
1658 | else | |
1659 | inode_unlock_shared(inode); | |
4273b793 AA |
1660 | |
1661 | return res; | |
1662 | } | |
1663 | ||
55752a3a MS |
1664 | static ssize_t fuse_file_read_iter(struct kiocb *iocb, struct iov_iter *to) |
1665 | { | |
2f7b6f5b MS |
1666 | struct file *file = iocb->ki_filp; |
1667 | struct fuse_file *ff = file->private_data; | |
c2d0ad00 | 1668 | struct inode *inode = file_inode(file); |
2f7b6f5b | 1669 | |
5d069dbe | 1670 | if (fuse_is_bad(inode)) |
2f7b6f5b | 1671 | return -EIO; |
55752a3a | 1672 | |
c2d0ad00 VG |
1673 | if (FUSE_IS_DAX(inode)) |
1674 | return fuse_dax_read_iter(iocb, to); | |
1675 | ||
55752a3a MS |
1676 | if (!(ff->open_flags & FOPEN_DIRECT_IO)) |
1677 | return fuse_cache_read_iter(iocb, to); | |
1678 | else | |
1679 | return fuse_direct_read_iter(iocb, to); | |
1680 | } | |
1681 | ||
1682 | static ssize_t fuse_file_write_iter(struct kiocb *iocb, struct iov_iter *from) | |
1683 | { | |
2f7b6f5b MS |
1684 | struct file *file = iocb->ki_filp; |
1685 | struct fuse_file *ff = file->private_data; | |
c2d0ad00 | 1686 | struct inode *inode = file_inode(file); |
2f7b6f5b | 1687 | |
5d069dbe | 1688 | if (fuse_is_bad(inode)) |
2f7b6f5b | 1689 | return -EIO; |
55752a3a | 1690 | |
c2d0ad00 VG |
1691 | if (FUSE_IS_DAX(inode)) |
1692 | return fuse_dax_write_iter(iocb, from); | |
1693 | ||
55752a3a MS |
1694 | if (!(ff->open_flags & FOPEN_DIRECT_IO)) |
1695 | return fuse_cache_write_iter(iocb, from); | |
1696 | else | |
1697 | return fuse_direct_write_iter(iocb, from); | |
1698 | } | |
1699 | ||
33826ebb | 1700 | static void fuse_writepage_free(struct fuse_writepage_args *wpa) |
b6aeaded | 1701 | { |
33826ebb | 1702 | struct fuse_args_pages *ap = &wpa->ia.ap; |
385b1268 PE |
1703 | int i; |
1704 | ||
660585b5 MS |
1705 | if (wpa->bucket) |
1706 | fuse_sync_bucket_dec(wpa->bucket); | |
1707 | ||
33826ebb MS |
1708 | for (i = 0; i < ap->num_pages; i++) |
1709 | __free_page(ap->pages[i]); | |
1710 | ||
1711 | if (wpa->ia.ff) | |
1712 | fuse_file_put(wpa->ia.ff, false, false); | |
8b284dc4 | 1713 | |
33826ebb MS |
1714 | kfree(ap->pages); |
1715 | kfree(wpa); | |
3be5a52b MS |
1716 | } |
1717 | ||
fcee216b | 1718 | static void fuse_writepage_finish(struct fuse_mount *fm, |
33826ebb | 1719 | struct fuse_writepage_args *wpa) |
3be5a52b | 1720 | { |
33826ebb MS |
1721 | struct fuse_args_pages *ap = &wpa->ia.ap; |
1722 | struct inode *inode = wpa->inode; | |
3be5a52b | 1723 | struct fuse_inode *fi = get_fuse_inode(inode); |
de1414a6 | 1724 | struct backing_dev_info *bdi = inode_to_bdi(inode); |
385b1268 | 1725 | int i; |
3be5a52b | 1726 | |
33826ebb | 1727 | for (i = 0; i < ap->num_pages; i++) { |
93f78d88 | 1728 | dec_wb_stat(&bdi->wb, WB_WRITEBACK); |
33826ebb | 1729 | dec_node_page_state(ap->pages[i], NR_WRITEBACK_TEMP); |
93f78d88 | 1730 | wb_writeout_inc(&bdi->wb); |
385b1268 | 1731 | } |
3be5a52b MS |
1732 | wake_up(&fi->page_waitq); |
1733 | } | |
1734 | ||
f15ecfef | 1735 | /* Called under fi->lock, may release and reacquire it */ |
fcee216b | 1736 | static void fuse_send_writepage(struct fuse_mount *fm, |
33826ebb | 1737 | struct fuse_writepage_args *wpa, loff_t size) |
f15ecfef KT |
1738 | __releases(fi->lock) |
1739 | __acquires(fi->lock) | |
3be5a52b | 1740 | { |
33826ebb MS |
1741 | struct fuse_writepage_args *aux, *next; |
1742 | struct fuse_inode *fi = get_fuse_inode(wpa->inode); | |
1743 | struct fuse_write_in *inarg = &wpa->ia.write.in; | |
1744 | struct fuse_args *args = &wpa->ia.ap.args; | |
1745 | __u64 data_size = wpa->ia.ap.num_pages * PAGE_SIZE; | |
1746 | int err; | |
3be5a52b | 1747 | |
33826ebb | 1748 | fi->writectr++; |
385b1268 PE |
1749 | if (inarg->offset + data_size <= size) { |
1750 | inarg->size = data_size; | |
3be5a52b | 1751 | } else if (inarg->offset < size) { |
385b1268 | 1752 | inarg->size = size - inarg->offset; |
3be5a52b MS |
1753 | } else { |
1754 | /* Got truncated off completely */ | |
1755 | goto out_free; | |
b6aeaded | 1756 | } |
3be5a52b | 1757 | |
33826ebb MS |
1758 | args->in_args[1].size = inarg->size; |
1759 | args->force = true; | |
1760 | args->nocreds = true; | |
1761 | ||
fcee216b | 1762 | err = fuse_simple_background(fm, args, GFP_ATOMIC); |
33826ebb MS |
1763 | if (err == -ENOMEM) { |
1764 | spin_unlock(&fi->lock); | |
fcee216b | 1765 | err = fuse_simple_background(fm, args, GFP_NOFS | __GFP_NOFAIL); |
33826ebb MS |
1766 | spin_lock(&fi->lock); |
1767 | } | |
1768 | ||
f15ecfef | 1769 | /* Fails on broken connection only */ |
33826ebb | 1770 | if (unlikely(err)) |
f15ecfef KT |
1771 | goto out_free; |
1772 | ||
3be5a52b MS |
1773 | return; |
1774 | ||
1775 | out_free: | |
33826ebb | 1776 | fi->writectr--; |
69a6487a | 1777 | rb_erase(&wpa->writepages_entry, &fi->writepages); |
fcee216b | 1778 | fuse_writepage_finish(fm, wpa); |
f15ecfef | 1779 | spin_unlock(&fi->lock); |
e2653bd5 MS |
1780 | |
1781 | /* After fuse_writepage_finish() aux request list is private */ | |
33826ebb MS |
1782 | for (aux = wpa->next; aux; aux = next) { |
1783 | next = aux->next; | |
1784 | aux->next = NULL; | |
1785 | fuse_writepage_free(aux); | |
e2653bd5 MS |
1786 | } |
1787 | ||
33826ebb | 1788 | fuse_writepage_free(wpa); |
f15ecfef | 1789 | spin_lock(&fi->lock); |
b6aeaded MS |
1790 | } |
1791 | ||
3be5a52b MS |
1792 | /* |
1793 | * If fi->writectr is positive (no truncate or fsync going on) send | |
1794 | * all queued writepage requests. | |
1795 | * | |
f15ecfef | 1796 | * Called with fi->lock |
3be5a52b MS |
1797 | */ |
1798 | void fuse_flush_writepages(struct inode *inode) | |
f15ecfef KT |
1799 | __releases(fi->lock) |
1800 | __acquires(fi->lock) | |
b6aeaded | 1801 | { |
fcee216b | 1802 | struct fuse_mount *fm = get_fuse_mount(inode); |
3be5a52b | 1803 | struct fuse_inode *fi = get_fuse_inode(inode); |
9de5be06 | 1804 | loff_t crop = i_size_read(inode); |
33826ebb | 1805 | struct fuse_writepage_args *wpa; |
3be5a52b MS |
1806 | |
1807 | while (fi->writectr >= 0 && !list_empty(&fi->queued_writes)) { | |
33826ebb MS |
1808 | wpa = list_entry(fi->queued_writes.next, |
1809 | struct fuse_writepage_args, queue_entry); | |
1810 | list_del_init(&wpa->queue_entry); | |
fcee216b | 1811 | fuse_send_writepage(fm, wpa, crop); |
3be5a52b MS |
1812 | } |
1813 | } | |
1814 | ||
c146024e MS |
1815 | static struct fuse_writepage_args *fuse_insert_writeback(struct rb_root *root, |
1816 | struct fuse_writepage_args *wpa) | |
6b2fb799 MP |
1817 | { |
1818 | pgoff_t idx_from = wpa->ia.write.in.offset >> PAGE_SHIFT; | |
1819 | pgoff_t idx_to = idx_from + wpa->ia.ap.num_pages - 1; | |
1820 | struct rb_node **p = &root->rb_node; | |
1821 | struct rb_node *parent = NULL; | |
1822 | ||
1823 | WARN_ON(!wpa->ia.ap.num_pages); | |
1824 | while (*p) { | |
1825 | struct fuse_writepage_args *curr; | |
1826 | pgoff_t curr_index; | |
1827 | ||
1828 | parent = *p; | |
1829 | curr = rb_entry(parent, struct fuse_writepage_args, | |
1830 | writepages_entry); | |
1831 | WARN_ON(curr->inode != wpa->inode); | |
1832 | curr_index = curr->ia.write.in.offset >> PAGE_SHIFT; | |
1833 | ||
1834 | if (idx_from >= curr_index + curr->ia.ap.num_pages) | |
1835 | p = &(*p)->rb_right; | |
1836 | else if (idx_to < curr_index) | |
1837 | p = &(*p)->rb_left; | |
1838 | else | |
c146024e | 1839 | return curr; |
6b2fb799 MP |
1840 | } |
1841 | ||
1842 | rb_link_node(&wpa->writepages_entry, parent, p); | |
1843 | rb_insert_color(&wpa->writepages_entry, root); | |
c146024e MS |
1844 | return NULL; |
1845 | } | |
1846 | ||
1847 | static void tree_insert(struct rb_root *root, struct fuse_writepage_args *wpa) | |
1848 | { | |
1849 | WARN_ON(fuse_insert_writeback(root, wpa)); | |
6b2fb799 MP |
1850 | } |
1851 | ||
fcee216b | 1852 | static void fuse_writepage_end(struct fuse_mount *fm, struct fuse_args *args, |
33826ebb | 1853 | int error) |
3be5a52b | 1854 | { |
33826ebb MS |
1855 | struct fuse_writepage_args *wpa = |
1856 | container_of(args, typeof(*wpa), ia.ap.args); | |
1857 | struct inode *inode = wpa->inode; | |
3be5a52b | 1858 | struct fuse_inode *fi = get_fuse_inode(inode); |
3466958b | 1859 | struct fuse_conn *fc = get_fuse_conn(inode); |
3be5a52b | 1860 | |
33826ebb | 1861 | mapping_set_error(inode->i_mapping, error); |
3466958b VG |
1862 | /* |
1863 | * A writeback finished and this might have updated mtime/ctime on | |
1864 | * server making local mtime/ctime stale. Hence invalidate attrs. | |
1865 | * Do this only if writeback_cache is not enabled. If writeback_cache | |
1866 | * is enabled, we trust local ctime/mtime. | |
1867 | */ | |
1868 | if (!fc->writeback_cache) | |
fa5eee57 | 1869 | fuse_invalidate_attr_mask(inode, FUSE_STATX_MODIFY); |
f15ecfef | 1870 | spin_lock(&fi->lock); |
69a6487a | 1871 | rb_erase(&wpa->writepages_entry, &fi->writepages); |
33826ebb | 1872 | while (wpa->next) { |
fcee216b | 1873 | struct fuse_mount *fm = get_fuse_mount(inode); |
33826ebb MS |
1874 | struct fuse_write_in *inarg = &wpa->ia.write.in; |
1875 | struct fuse_writepage_args *next = wpa->next; | |
1876 | ||
1877 | wpa->next = next->next; | |
1878 | next->next = NULL; | |
1879 | next->ia.ff = fuse_file_get(wpa->ia.ff); | |
6b2fb799 | 1880 | tree_insert(&fi->writepages, next); |
6eaf4782 MP |
1881 | |
1882 | /* | |
1883 | * Skip fuse_flush_writepages() to make it easy to crop requests | |
1884 | * based on primary request size. | |
1885 | * | |
1886 | * 1st case (trivial): there are no concurrent activities using | |
1887 | * fuse_set/release_nowrite. Then we're on safe side because | |
1888 | * fuse_flush_writepages() would call fuse_send_writepage() | |
1889 | * anyway. | |
1890 | * | |
1891 | * 2nd case: someone called fuse_set_nowrite and it is waiting | |
1892 | * now for completion of all in-flight requests. This happens | |
1893 | * rarely and no more than once per page, so this should be | |
1894 | * okay. | |
1895 | * | |
1896 | * 3rd case: someone (e.g. fuse_do_setattr()) is in the middle | |
1897 | * of fuse_set_nowrite..fuse_release_nowrite section. The fact | |
1898 | * that fuse_set_nowrite returned implies that all in-flight | |
1899 | * requests were completed along with all of their secondary | |
1900 | * requests. Further primary requests are blocked by negative | |
1901 | * writectr. Hence there cannot be any in-flight requests and | |
1902 | * no invocations of fuse_writepage_end() while we're in | |
1903 | * fuse_set_nowrite..fuse_release_nowrite section. | |
1904 | */ | |
fcee216b | 1905 | fuse_send_writepage(fm, next, inarg->offset + inarg->size); |
8b284dc4 | 1906 | } |
3be5a52b | 1907 | fi->writectr--; |
fcee216b | 1908 | fuse_writepage_finish(fm, wpa); |
f15ecfef | 1909 | spin_unlock(&fi->lock); |
33826ebb | 1910 | fuse_writepage_free(wpa); |
3be5a52b MS |
1911 | } |
1912 | ||
a9667ac8 | 1913 | static struct fuse_file *__fuse_write_file_get(struct fuse_inode *fi) |
adcadfa8 | 1914 | { |
84840efc | 1915 | struct fuse_file *ff; |
adcadfa8 | 1916 | |
f15ecfef | 1917 | spin_lock(&fi->lock); |
84840efc MS |
1918 | ff = list_first_entry_or_null(&fi->write_files, struct fuse_file, |
1919 | write_entry); | |
1920 | if (ff) | |
72523425 | 1921 | fuse_file_get(ff); |
f15ecfef | 1922 | spin_unlock(&fi->lock); |
adcadfa8 PE |
1923 | |
1924 | return ff; | |
1925 | } | |
1926 | ||
a9667ac8 | 1927 | static struct fuse_file *fuse_write_file_get(struct fuse_inode *fi) |
1e18bda8 | 1928 | { |
a9667ac8 | 1929 | struct fuse_file *ff = __fuse_write_file_get(fi); |
1e18bda8 MS |
1930 | WARN_ON(!ff); |
1931 | return ff; | |
1932 | } | |
1933 | ||
1934 | int fuse_write_inode(struct inode *inode, struct writeback_control *wbc) | |
1935 | { | |
1e18bda8 MS |
1936 | struct fuse_inode *fi = get_fuse_inode(inode); |
1937 | struct fuse_file *ff; | |
1938 | int err; | |
1939 | ||
5c791fe1 MS |
1940 | /* |
1941 | * Inode is always written before the last reference is dropped and | |
1942 | * hence this should not be reached from reclaim. | |
1943 | * | |
1944 | * Writing back the inode from reclaim can deadlock if the request | |
1945 | * processing itself needs an allocation. Allocations triggering | |
1946 | * reclaim while serving a request can't be prevented, because it can | |
1947 | * involve any number of unrelated userspace processes. | |
1948 | */ | |
1949 | WARN_ON(wbc->for_reclaim); | |
1950 | ||
a9667ac8 | 1951 | ff = __fuse_write_file_get(fi); |
ab9e13f7 | 1952 | err = fuse_flush_times(inode, ff); |
1e18bda8 | 1953 | if (ff) |
2e64ff15 | 1954 | fuse_file_put(ff, false, false); |
1e18bda8 MS |
1955 | |
1956 | return err; | |
1957 | } | |
1958 | ||
33826ebb MS |
1959 | static struct fuse_writepage_args *fuse_writepage_args_alloc(void) |
1960 | { | |
1961 | struct fuse_writepage_args *wpa; | |
1962 | struct fuse_args_pages *ap; | |
1963 | ||
1964 | wpa = kzalloc(sizeof(*wpa), GFP_NOFS); | |
1965 | if (wpa) { | |
1966 | ap = &wpa->ia.ap; | |
1967 | ap->num_pages = 0; | |
1968 | ap->pages = fuse_pages_alloc(1, GFP_NOFS, &ap->descs); | |
1969 | if (!ap->pages) { | |
1970 | kfree(wpa); | |
1971 | wpa = NULL; | |
1972 | } | |
1973 | } | |
1974 | return wpa; | |
1975 | ||
1976 | } | |
1977 | ||
660585b5 MS |
1978 | static void fuse_writepage_add_to_bucket(struct fuse_conn *fc, |
1979 | struct fuse_writepage_args *wpa) | |
1980 | { | |
1981 | if (!fc->sync_fs) | |
1982 | return; | |
1983 | ||
1984 | rcu_read_lock(); | |
1985 | /* Prevent resurrection of dead bucket in unlikely race with syncfs */ | |
1986 | do { | |
1987 | wpa->bucket = rcu_dereference(fc->curr_bucket); | |
1988 | } while (unlikely(!atomic_inc_not_zero(&wpa->bucket->count))); | |
1989 | rcu_read_unlock(); | |
1990 | } | |
1991 | ||
3be5a52b MS |
1992 | static int fuse_writepage_locked(struct page *page) |
1993 | { | |
1994 | struct address_space *mapping = page->mapping; | |
1995 | struct inode *inode = mapping->host; | |
1996 | struct fuse_conn *fc = get_fuse_conn(inode); | |
1997 | struct fuse_inode *fi = get_fuse_inode(inode); | |
33826ebb MS |
1998 | struct fuse_writepage_args *wpa; |
1999 | struct fuse_args_pages *ap; | |
3be5a52b | 2000 | struct page *tmp_page; |
72523425 | 2001 | int error = -ENOMEM; |
3be5a52b MS |
2002 | |
2003 | set_page_writeback(page); | |
2004 | ||
33826ebb MS |
2005 | wpa = fuse_writepage_args_alloc(); |
2006 | if (!wpa) | |
3be5a52b | 2007 | goto err; |
33826ebb | 2008 | ap = &wpa->ia.ap; |
3be5a52b MS |
2009 | |
2010 | tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
2011 | if (!tmp_page) | |
2012 | goto err_free; | |
2013 | ||
72523425 | 2014 | error = -EIO; |
a9667ac8 | 2015 | wpa->ia.ff = fuse_write_file_get(fi); |
33826ebb | 2016 | if (!wpa->ia.ff) |
27f1b363 | 2017 | goto err_nofile; |
72523425 | 2018 | |
660585b5 | 2019 | fuse_writepage_add_to_bucket(fc, wpa); |
33826ebb | 2020 | fuse_write_args_fill(&wpa->ia, wpa->ia.ff, page_offset(page), 0); |
3be5a52b MS |
2021 | |
2022 | copy_highpage(tmp_page, page); | |
33826ebb MS |
2023 | wpa->ia.write.in.write_flags |= FUSE_WRITE_CACHE; |
2024 | wpa->next = NULL; | |
2025 | ap->args.in_pages = true; | |
2026 | ap->num_pages = 1; | |
2027 | ap->pages[0] = tmp_page; | |
2028 | ap->descs[0].offset = 0; | |
2029 | ap->descs[0].length = PAGE_SIZE; | |
2030 | ap->args.end = fuse_writepage_end; | |
2031 | wpa->inode = inode; | |
3be5a52b | 2032 | |
93f78d88 | 2033 | inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK); |
11fb9989 | 2034 | inc_node_page_state(tmp_page, NR_WRITEBACK_TEMP); |
3be5a52b | 2035 | |
f15ecfef | 2036 | spin_lock(&fi->lock); |
6b2fb799 | 2037 | tree_insert(&fi->writepages, wpa); |
33826ebb | 2038 | list_add_tail(&wpa->queue_entry, &fi->queued_writes); |
3be5a52b | 2039 | fuse_flush_writepages(inode); |
f15ecfef | 2040 | spin_unlock(&fi->lock); |
3be5a52b | 2041 | |
4a4ac4eb MP |
2042 | end_page_writeback(page); |
2043 | ||
3be5a52b MS |
2044 | return 0; |
2045 | ||
27f1b363 MP |
2046 | err_nofile: |
2047 | __free_page(tmp_page); | |
3be5a52b | 2048 | err_free: |
33826ebb | 2049 | kfree(wpa); |
3be5a52b | 2050 | err: |
9183976e | 2051 | mapping_set_error(page->mapping, error); |
3be5a52b | 2052 | end_page_writeback(page); |
72523425 | 2053 | return error; |
3be5a52b MS |
2054 | } |
2055 | ||
2056 | static int fuse_writepage(struct page *page, struct writeback_control *wbc) | |
2057 | { | |
670d21c6 | 2058 | struct fuse_conn *fc = get_fuse_conn(page->mapping->host); |
3be5a52b MS |
2059 | int err; |
2060 | ||
ff17be08 MS |
2061 | if (fuse_page_is_writeback(page->mapping->host, page->index)) { |
2062 | /* | |
2063 | * ->writepages() should be called for sync() and friends. We | |
2064 | * should only get here on direct reclaim and then we are | |
2065 | * allowed to skip a page which is already in flight | |
2066 | */ | |
2067 | WARN_ON(wbc->sync_mode == WB_SYNC_ALL); | |
2068 | ||
2069 | redirty_page_for_writepage(wbc, page); | |
d5880c7a | 2070 | unlock_page(page); |
ff17be08 MS |
2071 | return 0; |
2072 | } | |
2073 | ||
670d21c6 N |
2074 | if (wbc->sync_mode == WB_SYNC_NONE && |
2075 | fc->num_background >= fc->congestion_threshold) | |
2076 | return AOP_WRITEPAGE_ACTIVATE; | |
2077 | ||
3be5a52b MS |
2078 | err = fuse_writepage_locked(page); |
2079 | unlock_page(page); | |
2080 | ||
2081 | return err; | |
2082 | } | |
2083 | ||
26d614df | 2084 | struct fuse_fill_wb_data { |
33826ebb | 2085 | struct fuse_writepage_args *wpa; |
26d614df PE |
2086 | struct fuse_file *ff; |
2087 | struct inode *inode; | |
2d033eaa | 2088 | struct page **orig_pages; |
33826ebb | 2089 | unsigned int max_pages; |
26d614df PE |
2090 | }; |
2091 | ||
33826ebb MS |
2092 | static bool fuse_pages_realloc(struct fuse_fill_wb_data *data) |
2093 | { | |
2094 | struct fuse_args_pages *ap = &data->wpa->ia.ap; | |
2095 | struct fuse_conn *fc = get_fuse_conn(data->inode); | |
2096 | struct page **pages; | |
2097 | struct fuse_page_desc *descs; | |
2098 | unsigned int npages = min_t(unsigned int, | |
2099 | max_t(unsigned int, data->max_pages * 2, | |
2100 | FUSE_DEFAULT_MAX_PAGES_PER_REQ), | |
2101 | fc->max_pages); | |
2102 | WARN_ON(npages <= data->max_pages); | |
2103 | ||
2104 | pages = fuse_pages_alloc(npages, GFP_NOFS, &descs); | |
2105 | if (!pages) | |
2106 | return false; | |
2107 | ||
2108 | memcpy(pages, ap->pages, sizeof(struct page *) * ap->num_pages); | |
2109 | memcpy(descs, ap->descs, sizeof(struct fuse_page_desc) * ap->num_pages); | |
2110 | kfree(ap->pages); | |
2111 | ap->pages = pages; | |
2112 | ap->descs = descs; | |
2113 | data->max_pages = npages; | |
2114 | ||
2115 | return true; | |
2116 | } | |
2117 | ||
26d614df PE |
2118 | static void fuse_writepages_send(struct fuse_fill_wb_data *data) |
2119 | { | |
33826ebb | 2120 | struct fuse_writepage_args *wpa = data->wpa; |
26d614df | 2121 | struct inode *inode = data->inode; |
26d614df | 2122 | struct fuse_inode *fi = get_fuse_inode(inode); |
33826ebb | 2123 | int num_pages = wpa->ia.ap.num_pages; |
2d033eaa | 2124 | int i; |
26d614df | 2125 | |
33826ebb | 2126 | wpa->ia.ff = fuse_file_get(data->ff); |
f15ecfef | 2127 | spin_lock(&fi->lock); |
33826ebb | 2128 | list_add_tail(&wpa->queue_entry, &fi->queued_writes); |
26d614df | 2129 | fuse_flush_writepages(inode); |
f15ecfef | 2130 | spin_unlock(&fi->lock); |
2d033eaa MP |
2131 | |
2132 | for (i = 0; i < num_pages; i++) | |
2133 | end_page_writeback(data->orig_pages[i]); | |
26d614df PE |
2134 | } |
2135 | ||
7f305ca1 | 2136 | /* |
c146024e MS |
2137 | * Check under fi->lock if the page is under writeback, and insert it onto the |
2138 | * rb_tree if not. Otherwise iterate auxiliary write requests, to see if there's | |
419234d5 MS |
2139 | * one already added for a page at this offset. If there's none, then insert |
2140 | * this new request onto the auxiliary list, otherwise reuse the existing one by | |
c146024e | 2141 | * swapping the new temp page with the old one. |
7f305ca1 | 2142 | */ |
c146024e MS |
2143 | static bool fuse_writepage_add(struct fuse_writepage_args *new_wpa, |
2144 | struct page *page) | |
8b284dc4 | 2145 | { |
33826ebb MS |
2146 | struct fuse_inode *fi = get_fuse_inode(new_wpa->inode); |
2147 | struct fuse_writepage_args *tmp; | |
2148 | struct fuse_writepage_args *old_wpa; | |
2149 | struct fuse_args_pages *new_ap = &new_wpa->ia.ap; | |
8b284dc4 | 2150 | |
33826ebb | 2151 | WARN_ON(new_ap->num_pages != 0); |
c146024e | 2152 | new_ap->num_pages = 1; |
8b284dc4 | 2153 | |
f15ecfef | 2154 | spin_lock(&fi->lock); |
c146024e | 2155 | old_wpa = fuse_insert_writeback(&fi->writepages, new_wpa); |
33826ebb | 2156 | if (!old_wpa) { |
f15ecfef | 2157 | spin_unlock(&fi->lock); |
c146024e | 2158 | return true; |
f6011081 | 2159 | } |
8b284dc4 | 2160 | |
33826ebb | 2161 | for (tmp = old_wpa->next; tmp; tmp = tmp->next) { |
7f305ca1 MS |
2162 | pgoff_t curr_index; |
2163 | ||
33826ebb MS |
2164 | WARN_ON(tmp->inode != new_wpa->inode); |
2165 | curr_index = tmp->ia.write.in.offset >> PAGE_SHIFT; | |
419234d5 | 2166 | if (curr_index == page->index) { |
33826ebb MS |
2167 | WARN_ON(tmp->ia.ap.num_pages != 1); |
2168 | swap(tmp->ia.ap.pages[0], new_ap->pages[0]); | |
7f305ca1 | 2169 | break; |
8b284dc4 MS |
2170 | } |
2171 | } | |
2172 | ||
7f305ca1 | 2173 | if (!tmp) { |
33826ebb MS |
2174 | new_wpa->next = old_wpa->next; |
2175 | old_wpa->next = new_wpa; | |
7f305ca1 | 2176 | } |
41b6e41f | 2177 | |
f15ecfef | 2178 | spin_unlock(&fi->lock); |
7f305ca1 MS |
2179 | |
2180 | if (tmp) { | |
33826ebb | 2181 | struct backing_dev_info *bdi = inode_to_bdi(new_wpa->inode); |
8b284dc4 | 2182 | |
93f78d88 | 2183 | dec_wb_stat(&bdi->wb, WB_WRITEBACK); |
33826ebb | 2184 | dec_node_page_state(new_ap->pages[0], NR_WRITEBACK_TEMP); |
93f78d88 | 2185 | wb_writeout_inc(&bdi->wb); |
33826ebb | 2186 | fuse_writepage_free(new_wpa); |
8b284dc4 | 2187 | } |
7f305ca1 | 2188 | |
c146024e | 2189 | return false; |
8b284dc4 MS |
2190 | } |
2191 | ||
6ddf3af9 MS |
2192 | static bool fuse_writepage_need_send(struct fuse_conn *fc, struct page *page, |
2193 | struct fuse_args_pages *ap, | |
2194 | struct fuse_fill_wb_data *data) | |
2195 | { | |
2196 | WARN_ON(!ap->num_pages); | |
2197 | ||
2198 | /* | |
2199 | * Being under writeback is unlikely but possible. For example direct | |
2200 | * read to an mmaped fuse file will set the page dirty twice; once when | |
2201 | * the pages are faulted with get_user_pages(), and then after the read | |
2202 | * completed. | |
2203 | */ | |
2204 | if (fuse_page_is_writeback(data->inode, page->index)) | |
2205 | return true; | |
2206 | ||
2207 | /* Reached max pages */ | |
2208 | if (ap->num_pages == fc->max_pages) | |
2209 | return true; | |
2210 | ||
2211 | /* Reached max write bytes */ | |
2212 | if ((ap->num_pages + 1) * PAGE_SIZE > fc->max_write) | |
2213 | return true; | |
2214 | ||
2215 | /* Discontinuity */ | |
2216 | if (data->orig_pages[ap->num_pages - 1]->index + 1 != page->index) | |
2217 | return true; | |
2218 | ||
2219 | /* Need to grow the pages array? If so, did the expansion fail? */ | |
2220 | if (ap->num_pages == data->max_pages && !fuse_pages_realloc(data)) | |
2221 | return true; | |
2222 | ||
2223 | return false; | |
2224 | } | |
2225 | ||
d585bdbe | 2226 | static int fuse_writepages_fill(struct folio *folio, |
26d614df PE |
2227 | struct writeback_control *wbc, void *_data) |
2228 | { | |
2229 | struct fuse_fill_wb_data *data = _data; | |
33826ebb MS |
2230 | struct fuse_writepage_args *wpa = data->wpa; |
2231 | struct fuse_args_pages *ap = &wpa->ia.ap; | |
26d614df | 2232 | struct inode *inode = data->inode; |
f15ecfef | 2233 | struct fuse_inode *fi = get_fuse_inode(inode); |
26d614df PE |
2234 | struct fuse_conn *fc = get_fuse_conn(inode); |
2235 | struct page *tmp_page; | |
2236 | int err; | |
2237 | ||
2238 | if (!data->ff) { | |
2239 | err = -EIO; | |
a9667ac8 | 2240 | data->ff = fuse_write_file_get(fi); |
26d614df PE |
2241 | if (!data->ff) |
2242 | goto out_unlock; | |
2243 | } | |
2244 | ||
d585bdbe | 2245 | if (wpa && fuse_writepage_need_send(fc, &folio->page, ap, data)) { |
8b284dc4 | 2246 | fuse_writepages_send(data); |
33826ebb | 2247 | data->wpa = NULL; |
26d614df | 2248 | } |
e52a8250 | 2249 | |
26d614df PE |
2250 | err = -ENOMEM; |
2251 | tmp_page = alloc_page(GFP_NOFS | __GFP_HIGHMEM); | |
2252 | if (!tmp_page) | |
2253 | goto out_unlock; | |
2254 | ||
2255 | /* | |
2256 | * The page must not be redirtied until the writeout is completed | |
2257 | * (i.e. userspace has sent a reply to the write request). Otherwise | |
2258 | * there could be more than one temporary page instance for each real | |
2259 | * page. | |
2260 | * | |
2261 | * This is ensured by holding the page lock in page_mkwrite() while | |
2262 | * checking fuse_page_is_writeback(). We already hold the page lock | |
2263 | * since clear_page_dirty_for_io() and keep it held until we add the | |
33826ebb | 2264 | * request to the fi->writepages list and increment ap->num_pages. |
26d614df PE |
2265 | * After this fuse_page_is_writeback() will indicate that the page is |
2266 | * under writeback, so we can release the page lock. | |
2267 | */ | |
33826ebb | 2268 | if (data->wpa == NULL) { |
26d614df | 2269 | err = -ENOMEM; |
33826ebb MS |
2270 | wpa = fuse_writepage_args_alloc(); |
2271 | if (!wpa) { | |
26d614df PE |
2272 | __free_page(tmp_page); |
2273 | goto out_unlock; | |
2274 | } | |
660585b5 MS |
2275 | fuse_writepage_add_to_bucket(fc, wpa); |
2276 | ||
33826ebb | 2277 | data->max_pages = 1; |
26d614df | 2278 | |
33826ebb | 2279 | ap = &wpa->ia.ap; |
d585bdbe | 2280 | fuse_write_args_fill(&wpa->ia, data->ff, folio_pos(folio), 0); |
33826ebb MS |
2281 | wpa->ia.write.in.write_flags |= FUSE_WRITE_CACHE; |
2282 | wpa->next = NULL; | |
2283 | ap->args.in_pages = true; | |
2284 | ap->args.end = fuse_writepage_end; | |
2285 | ap->num_pages = 0; | |
2286 | wpa->inode = inode; | |
26d614df | 2287 | } |
d585bdbe | 2288 | folio_start_writeback(folio); |
26d614df | 2289 | |
d585bdbe | 2290 | copy_highpage(tmp_page, &folio->page); |
33826ebb MS |
2291 | ap->pages[ap->num_pages] = tmp_page; |
2292 | ap->descs[ap->num_pages].offset = 0; | |
2293 | ap->descs[ap->num_pages].length = PAGE_SIZE; | |
d585bdbe | 2294 | data->orig_pages[ap->num_pages] = &folio->page; |
26d614df | 2295 | |
93f78d88 | 2296 | inc_wb_stat(&inode_to_bdi(inode)->wb, WB_WRITEBACK); |
11fb9989 | 2297 | inc_node_page_state(tmp_page, NR_WRITEBACK_TEMP); |
8b284dc4 MS |
2298 | |
2299 | err = 0; | |
c146024e MS |
2300 | if (data->wpa) { |
2301 | /* | |
2302 | * Protected by fi->lock against concurrent access by | |
2303 | * fuse_page_is_writeback(). | |
2304 | */ | |
2305 | spin_lock(&fi->lock); | |
2306 | ap->num_pages++; | |
2307 | spin_unlock(&fi->lock); | |
d585bdbe | 2308 | } else if (fuse_writepage_add(wpa, &folio->page)) { |
c146024e MS |
2309 | data->wpa = wpa; |
2310 | } else { | |
d585bdbe | 2311 | folio_end_writeback(folio); |
8b284dc4 | 2312 | } |
26d614df | 2313 | out_unlock: |
d585bdbe | 2314 | folio_unlock(folio); |
26d614df PE |
2315 | |
2316 | return err; | |
2317 | } | |
2318 | ||
2319 | static int fuse_writepages(struct address_space *mapping, | |
2320 | struct writeback_control *wbc) | |
2321 | { | |
2322 | struct inode *inode = mapping->host; | |
5da784cc | 2323 | struct fuse_conn *fc = get_fuse_conn(inode); |
26d614df PE |
2324 | struct fuse_fill_wb_data data; |
2325 | int err; | |
2326 | ||
2327 | err = -EIO; | |
5d069dbe | 2328 | if (fuse_is_bad(inode)) |
26d614df PE |
2329 | goto out; |
2330 | ||
670d21c6 N |
2331 | if (wbc->sync_mode == WB_SYNC_NONE && |
2332 | fc->num_background >= fc->congestion_threshold) | |
2333 | return 0; | |
2334 | ||
26d614df | 2335 | data.inode = inode; |
33826ebb | 2336 | data.wpa = NULL; |
26d614df PE |
2337 | data.ff = NULL; |
2338 | ||
2d033eaa | 2339 | err = -ENOMEM; |
5da784cc | 2340 | data.orig_pages = kcalloc(fc->max_pages, |
f2b3455e | 2341 | sizeof(struct page *), |
2d033eaa MP |
2342 | GFP_NOFS); |
2343 | if (!data.orig_pages) | |
2344 | goto out; | |
2345 | ||
26d614df | 2346 | err = write_cache_pages(mapping, wbc, fuse_writepages_fill, &data); |
33826ebb | 2347 | if (data.wpa) { |
33826ebb | 2348 | WARN_ON(!data.wpa->ia.ap.num_pages); |
26d614df | 2349 | fuse_writepages_send(&data); |
26d614df PE |
2350 | } |
2351 | if (data.ff) | |
2e64ff15 | 2352 | fuse_file_put(data.ff, false, false); |
2d033eaa MP |
2353 | |
2354 | kfree(data.orig_pages); | |
26d614df PE |
2355 | out: |
2356 | return err; | |
2357 | } | |
2358 | ||
6b12c1b3 PE |
2359 | /* |
2360 | * It's worthy to make sure that space is reserved on disk for the write, | |
2361 | * but how to implement it without killing performance need more thinking. | |
2362 | */ | |
2363 | static int fuse_write_begin(struct file *file, struct address_space *mapping, | |
9d6b0cd7 | 2364 | loff_t pos, unsigned len, struct page **pagep, void **fsdata) |
6b12c1b3 | 2365 | { |
09cbfeaf | 2366 | pgoff_t index = pos >> PAGE_SHIFT; |
a455589f | 2367 | struct fuse_conn *fc = get_fuse_conn(file_inode(file)); |
6b12c1b3 PE |
2368 | struct page *page; |
2369 | loff_t fsize; | |
2370 | int err = -ENOMEM; | |
2371 | ||
2372 | WARN_ON(!fc->writeback_cache); | |
2373 | ||
b7446e7c | 2374 | page = grab_cache_page_write_begin(mapping, index); |
6b12c1b3 PE |
2375 | if (!page) |
2376 | goto error; | |
2377 | ||
2378 | fuse_wait_on_page_writeback(mapping->host, page->index); | |
2379 | ||
09cbfeaf | 2380 | if (PageUptodate(page) || len == PAGE_SIZE) |
6b12c1b3 PE |
2381 | goto success; |
2382 | /* | |
2383 | * Check if the start this page comes after the end of file, in which | |
2384 | * case the readpage can be optimized away. | |
2385 | */ | |
2386 | fsize = i_size_read(mapping->host); | |
09cbfeaf KS |
2387 | if (fsize <= (pos & PAGE_MASK)) { |
2388 | size_t off = pos & ~PAGE_MASK; | |
6b12c1b3 PE |
2389 | if (off) |
2390 | zero_user_segment(page, 0, off); | |
2391 | goto success; | |
2392 | } | |
2393 | err = fuse_do_readpage(file, page); | |
2394 | if (err) | |
2395 | goto cleanup; | |
2396 | success: | |
2397 | *pagep = page; | |
2398 | return 0; | |
2399 | ||
2400 | cleanup: | |
2401 | unlock_page(page); | |
09cbfeaf | 2402 | put_page(page); |
6b12c1b3 PE |
2403 | error: |
2404 | return err; | |
2405 | } | |
2406 | ||
2407 | static int fuse_write_end(struct file *file, struct address_space *mapping, | |
2408 | loff_t pos, unsigned len, unsigned copied, | |
2409 | struct page *page, void *fsdata) | |
2410 | { | |
2411 | struct inode *inode = page->mapping->host; | |
2412 | ||
59c3b76c MS |
2413 | /* Haven't copied anything? Skip zeroing, size extending, dirtying. */ |
2414 | if (!copied) | |
2415 | goto unlock; | |
2416 | ||
8c56e03d | 2417 | pos += copied; |
6b12c1b3 PE |
2418 | if (!PageUptodate(page)) { |
2419 | /* Zero any unwritten bytes at the end of the page */ | |
8c56e03d | 2420 | size_t endoff = pos & ~PAGE_MASK; |
6b12c1b3 | 2421 | if (endoff) |
09cbfeaf | 2422 | zero_user_segment(page, endoff, PAGE_SIZE); |
6b12c1b3 PE |
2423 | SetPageUptodate(page); |
2424 | } | |
2425 | ||
8c56e03d MS |
2426 | if (pos > inode->i_size) |
2427 | i_size_write(inode, pos); | |
2428 | ||
6b12c1b3 | 2429 | set_page_dirty(page); |
59c3b76c MS |
2430 | |
2431 | unlock: | |
6b12c1b3 | 2432 | unlock_page(page); |
09cbfeaf | 2433 | put_page(page); |
6b12c1b3 PE |
2434 | |
2435 | return copied; | |
2436 | } | |
2437 | ||
2bf06b8e | 2438 | static int fuse_launder_folio(struct folio *folio) |
3be5a52b MS |
2439 | { |
2440 | int err = 0; | |
2bf06b8e MWO |
2441 | if (folio_clear_dirty_for_io(folio)) { |
2442 | struct inode *inode = folio->mapping->host; | |
3993382b MS |
2443 | |
2444 | /* Serialize with pending writeback for the same page */ | |
2bf06b8e MWO |
2445 | fuse_wait_on_page_writeback(inode, folio->index); |
2446 | err = fuse_writepage_locked(&folio->page); | |
3be5a52b | 2447 | if (!err) |
2bf06b8e | 2448 | fuse_wait_on_page_writeback(inode, folio->index); |
3be5a52b MS |
2449 | } |
2450 | return err; | |
2451 | } | |
2452 | ||
2453 | /* | |
36ea2337 MS |
2454 | * Write back dirty data/metadata now (there may not be any suitable |
2455 | * open files later for data) | |
3be5a52b MS |
2456 | */ |
2457 | static void fuse_vma_close(struct vm_area_struct *vma) | |
2458 | { | |
36ea2337 MS |
2459 | int err; |
2460 | ||
2461 | err = write_inode_now(vma->vm_file->f_mapping->host, 1); | |
2462 | mapping_set_error(vma->vm_file->f_mapping, err); | |
3be5a52b MS |
2463 | } |
2464 | ||
2465 | /* | |
2466 | * Wait for writeback against this page to complete before allowing it | |
2467 | * to be marked dirty again, and hence written back again, possibly | |
2468 | * before the previous writepage completed. | |
2469 | * | |
2470 | * Block here, instead of in ->writepage(), so that the userspace fs | |
2471 | * can only block processes actually operating on the filesystem. | |
2472 | * | |
2473 | * Otherwise unprivileged userspace fs would be able to block | |
2474 | * unrelated: | |
2475 | * | |
2476 | * - page migration | |
2477 | * - sync(2) | |
2478 | * - try_to_free_pages() with order > PAGE_ALLOC_COSTLY_ORDER | |
2479 | */ | |
46fb504a | 2480 | static vm_fault_t fuse_page_mkwrite(struct vm_fault *vmf) |
3be5a52b | 2481 | { |
c2ec175c | 2482 | struct page *page = vmf->page; |
11bac800 | 2483 | struct inode *inode = file_inode(vmf->vma->vm_file); |
cca24370 | 2484 | |
11bac800 | 2485 | file_update_time(vmf->vma->vm_file); |
cca24370 MS |
2486 | lock_page(page); |
2487 | if (page->mapping != inode->i_mapping) { | |
2488 | unlock_page(page); | |
2489 | return VM_FAULT_NOPAGE; | |
2490 | } | |
3be5a52b MS |
2491 | |
2492 | fuse_wait_on_page_writeback(inode, page->index); | |
cca24370 | 2493 | return VM_FAULT_LOCKED; |
3be5a52b MS |
2494 | } |
2495 | ||
f0f37e2f | 2496 | static const struct vm_operations_struct fuse_file_vm_ops = { |
3be5a52b MS |
2497 | .close = fuse_vma_close, |
2498 | .fault = filemap_fault, | |
f1820361 | 2499 | .map_pages = filemap_map_pages, |
3be5a52b MS |
2500 | .page_mkwrite = fuse_page_mkwrite, |
2501 | }; | |
2502 | ||
2503 | static int fuse_file_mmap(struct file *file, struct vm_area_struct *vma) | |
2504 | { | |
55752a3a MS |
2505 | struct fuse_file *ff = file->private_data; |
2506 | ||
2a9a609a SH |
2507 | /* DAX mmap is superior to direct_io mmap */ |
2508 | if (FUSE_IS_DAX(file_inode(file))) | |
2509 | return fuse_dax_mmap(file, vma); | |
2510 | ||
55752a3a MS |
2511 | if (ff->open_flags & FOPEN_DIRECT_IO) { |
2512 | /* Can't provide the coherency needed for MAP_SHARED */ | |
2513 | if (vma->vm_flags & VM_MAYSHARE) | |
2514 | return -ENODEV; | |
2515 | ||
2516 | invalidate_inode_pages2(file->f_mapping); | |
2517 | ||
2518 | return generic_file_mmap(file, vma); | |
2519 | } | |
2520 | ||
650b22b9 PE |
2521 | if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) |
2522 | fuse_link_write_file(file); | |
2523 | ||
3be5a52b MS |
2524 | file_accessed(file); |
2525 | vma->vm_ops = &fuse_file_vm_ops; | |
b6aeaded MS |
2526 | return 0; |
2527 | } | |
2528 | ||
0b6e9ea0 SF |
2529 | static int convert_fuse_file_lock(struct fuse_conn *fc, |
2530 | const struct fuse_file_lock *ffl, | |
71421259 MS |
2531 | struct file_lock *fl) |
2532 | { | |
2533 | switch (ffl->type) { | |
2534 | case F_UNLCK: | |
2535 | break; | |
2536 | ||
2537 | case F_RDLCK: | |
2538 | case F_WRLCK: | |
2539 | if (ffl->start > OFFSET_MAX || ffl->end > OFFSET_MAX || | |
2540 | ffl->end < ffl->start) | |
2541 | return -EIO; | |
2542 | ||
2543 | fl->fl_start = ffl->start; | |
2544 | fl->fl_end = ffl->end; | |
0b6e9ea0 SF |
2545 | |
2546 | /* | |
9d5b86ac BC |
2547 | * Convert pid into init's pid namespace. The locks API will |
2548 | * translate it into the caller's pid namespace. | |
0b6e9ea0 SF |
2549 | */ |
2550 | rcu_read_lock(); | |
9d5b86ac | 2551 | fl->fl_pid = pid_nr_ns(find_pid_ns(ffl->pid, fc->pid_ns), &init_pid_ns); |
0b6e9ea0 | 2552 | rcu_read_unlock(); |
71421259 MS |
2553 | break; |
2554 | ||
2555 | default: | |
2556 | return -EIO; | |
2557 | } | |
2558 | fl->fl_type = ffl->type; | |
2559 | return 0; | |
2560 | } | |
2561 | ||
7078187a | 2562 | static void fuse_lk_fill(struct fuse_args *args, struct file *file, |
a9ff4f87 | 2563 | const struct file_lock *fl, int opcode, pid_t pid, |
7078187a | 2564 | int flock, struct fuse_lk_in *inarg) |
71421259 | 2565 | { |
6131ffaa | 2566 | struct inode *inode = file_inode(file); |
9c8ef561 | 2567 | struct fuse_conn *fc = get_fuse_conn(inode); |
71421259 | 2568 | struct fuse_file *ff = file->private_data; |
7078187a MS |
2569 | |
2570 | memset(inarg, 0, sizeof(*inarg)); | |
2571 | inarg->fh = ff->fh; | |
2572 | inarg->owner = fuse_lock_owner_id(fc, fl->fl_owner); | |
2573 | inarg->lk.start = fl->fl_start; | |
2574 | inarg->lk.end = fl->fl_end; | |
2575 | inarg->lk.type = fl->fl_type; | |
2576 | inarg->lk.pid = pid; | |
a9ff4f87 | 2577 | if (flock) |
7078187a | 2578 | inarg->lk_flags |= FUSE_LK_FLOCK; |
d5b48543 MS |
2579 | args->opcode = opcode; |
2580 | args->nodeid = get_node_id(inode); | |
2581 | args->in_numargs = 1; | |
2582 | args->in_args[0].size = sizeof(*inarg); | |
2583 | args->in_args[0].value = inarg; | |
71421259 MS |
2584 | } |
2585 | ||
2586 | static int fuse_getlk(struct file *file, struct file_lock *fl) | |
2587 | { | |
6131ffaa | 2588 | struct inode *inode = file_inode(file); |
fcee216b | 2589 | struct fuse_mount *fm = get_fuse_mount(inode); |
7078187a MS |
2590 | FUSE_ARGS(args); |
2591 | struct fuse_lk_in inarg; | |
71421259 MS |
2592 | struct fuse_lk_out outarg; |
2593 | int err; | |
2594 | ||
7078187a | 2595 | fuse_lk_fill(&args, file, fl, FUSE_GETLK, 0, 0, &inarg); |
d5b48543 MS |
2596 | args.out_numargs = 1; |
2597 | args.out_args[0].size = sizeof(outarg); | |
2598 | args.out_args[0].value = &outarg; | |
fcee216b | 2599 | err = fuse_simple_request(fm, &args); |
71421259 | 2600 | if (!err) |
fcee216b | 2601 | err = convert_fuse_file_lock(fm->fc, &outarg.lk, fl); |
71421259 MS |
2602 | |
2603 | return err; | |
2604 | } | |
2605 | ||
a9ff4f87 | 2606 | static int fuse_setlk(struct file *file, struct file_lock *fl, int flock) |
71421259 | 2607 | { |
6131ffaa | 2608 | struct inode *inode = file_inode(file); |
fcee216b | 2609 | struct fuse_mount *fm = get_fuse_mount(inode); |
7078187a MS |
2610 | FUSE_ARGS(args); |
2611 | struct fuse_lk_in inarg; | |
71421259 | 2612 | int opcode = (fl->fl_flags & FL_SLEEP) ? FUSE_SETLKW : FUSE_SETLK; |
0b6e9ea0 | 2613 | struct pid *pid = fl->fl_type != F_UNLCK ? task_tgid(current) : NULL; |
fcee216b | 2614 | pid_t pid_nr = pid_nr_ns(pid, fm->fc->pid_ns); |
71421259 MS |
2615 | int err; |
2616 | ||
8fb47a4f | 2617 | if (fl->fl_lmops && fl->fl_lmops->lm_grant) { |
48e90761 MS |
2618 | /* NLM needs asynchronous locks, which we don't support yet */ |
2619 | return -ENOLCK; | |
2620 | } | |
2621 | ||
71421259 | 2622 | /* Unlock on close is handled by the flush method */ |
50f2112c | 2623 | if ((fl->fl_flags & FL_CLOSE_POSIX) == FL_CLOSE_POSIX) |
71421259 MS |
2624 | return 0; |
2625 | ||
0b6e9ea0 | 2626 | fuse_lk_fill(&args, file, fl, opcode, pid_nr, flock, &inarg); |
fcee216b | 2627 | err = fuse_simple_request(fm, &args); |
71421259 | 2628 | |
a4d27e75 MS |
2629 | /* locking is restartable */ |
2630 | if (err == -EINTR) | |
2631 | err = -ERESTARTSYS; | |
7078187a | 2632 | |
71421259 MS |
2633 | return err; |
2634 | } | |
2635 | ||
2636 | static int fuse_file_lock(struct file *file, int cmd, struct file_lock *fl) | |
2637 | { | |
6131ffaa | 2638 | struct inode *inode = file_inode(file); |
71421259 MS |
2639 | struct fuse_conn *fc = get_fuse_conn(inode); |
2640 | int err; | |
2641 | ||
48e90761 MS |
2642 | if (cmd == F_CANCELLK) { |
2643 | err = 0; | |
2644 | } else if (cmd == F_GETLK) { | |
71421259 | 2645 | if (fc->no_lock) { |
9d6a8c5c | 2646 | posix_test_lock(file, fl); |
71421259 MS |
2647 | err = 0; |
2648 | } else | |
2649 | err = fuse_getlk(file, fl); | |
2650 | } else { | |
2651 | if (fc->no_lock) | |
48e90761 | 2652 | err = posix_lock_file(file, fl, NULL); |
71421259 | 2653 | else |
a9ff4f87 | 2654 | err = fuse_setlk(file, fl, 0); |
71421259 MS |
2655 | } |
2656 | return err; | |
2657 | } | |
2658 | ||
a9ff4f87 MS |
2659 | static int fuse_file_flock(struct file *file, int cmd, struct file_lock *fl) |
2660 | { | |
6131ffaa | 2661 | struct inode *inode = file_inode(file); |
a9ff4f87 MS |
2662 | struct fuse_conn *fc = get_fuse_conn(inode); |
2663 | int err; | |
2664 | ||
37fb3a30 | 2665 | if (fc->no_flock) { |
4f656367 | 2666 | err = locks_lock_file_wait(file, fl); |
a9ff4f87 | 2667 | } else { |
37fb3a30 MS |
2668 | struct fuse_file *ff = file->private_data; |
2669 | ||
a9ff4f87 | 2670 | /* emulate flock with POSIX locks */ |
37fb3a30 | 2671 | ff->flock = true; |
a9ff4f87 MS |
2672 | err = fuse_setlk(file, fl, 1); |
2673 | } | |
2674 | ||
2675 | return err; | |
2676 | } | |
2677 | ||
b2d2272f MS |
2678 | static sector_t fuse_bmap(struct address_space *mapping, sector_t block) |
2679 | { | |
2680 | struct inode *inode = mapping->host; | |
fcee216b | 2681 | struct fuse_mount *fm = get_fuse_mount(inode); |
7078187a | 2682 | FUSE_ARGS(args); |
b2d2272f MS |
2683 | struct fuse_bmap_in inarg; |
2684 | struct fuse_bmap_out outarg; | |
2685 | int err; | |
2686 | ||
fcee216b | 2687 | if (!inode->i_sb->s_bdev || fm->fc->no_bmap) |
b2d2272f MS |
2688 | return 0; |
2689 | ||
b2d2272f MS |
2690 | memset(&inarg, 0, sizeof(inarg)); |
2691 | inarg.block = block; | |
2692 | inarg.blocksize = inode->i_sb->s_blocksize; | |
d5b48543 MS |
2693 | args.opcode = FUSE_BMAP; |
2694 | args.nodeid = get_node_id(inode); | |
2695 | args.in_numargs = 1; | |
2696 | args.in_args[0].size = sizeof(inarg); | |
2697 | args.in_args[0].value = &inarg; | |
2698 | args.out_numargs = 1; | |
2699 | args.out_args[0].size = sizeof(outarg); | |
2700 | args.out_args[0].value = &outarg; | |
fcee216b | 2701 | err = fuse_simple_request(fm, &args); |
b2d2272f | 2702 | if (err == -ENOSYS) |
fcee216b | 2703 | fm->fc->no_bmap = 1; |
b2d2272f MS |
2704 | |
2705 | return err ? 0 : outarg.block; | |
2706 | } | |
2707 | ||
0b5da8db R |
2708 | static loff_t fuse_lseek(struct file *file, loff_t offset, int whence) |
2709 | { | |
2710 | struct inode *inode = file->f_mapping->host; | |
fcee216b | 2711 | struct fuse_mount *fm = get_fuse_mount(inode); |
0b5da8db R |
2712 | struct fuse_file *ff = file->private_data; |
2713 | FUSE_ARGS(args); | |
2714 | struct fuse_lseek_in inarg = { | |
2715 | .fh = ff->fh, | |
2716 | .offset = offset, | |
2717 | .whence = whence | |
2718 | }; | |
2719 | struct fuse_lseek_out outarg; | |
2720 | int err; | |
2721 | ||
fcee216b | 2722 | if (fm->fc->no_lseek) |
0b5da8db R |
2723 | goto fallback; |
2724 | ||
d5b48543 MS |
2725 | args.opcode = FUSE_LSEEK; |
2726 | args.nodeid = ff->nodeid; | |
2727 | args.in_numargs = 1; | |
2728 | args.in_args[0].size = sizeof(inarg); | |
2729 | args.in_args[0].value = &inarg; | |
2730 | args.out_numargs = 1; | |
2731 | args.out_args[0].size = sizeof(outarg); | |
2732 | args.out_args[0].value = &outarg; | |
fcee216b | 2733 | err = fuse_simple_request(fm, &args); |
0b5da8db R |
2734 | if (err) { |
2735 | if (err == -ENOSYS) { | |
fcee216b | 2736 | fm->fc->no_lseek = 1; |
0b5da8db R |
2737 | goto fallback; |
2738 | } | |
2739 | return err; | |
2740 | } | |
2741 | ||
2742 | return vfs_setpos(file, outarg.offset, inode->i_sb->s_maxbytes); | |
2743 | ||
2744 | fallback: | |
c6c745b8 | 2745 | err = fuse_update_attributes(inode, file, STATX_SIZE); |
0b5da8db R |
2746 | if (!err) |
2747 | return generic_file_llseek(file, offset, whence); | |
2748 | else | |
2749 | return err; | |
2750 | } | |
2751 | ||
965c8e59 | 2752 | static loff_t fuse_file_llseek(struct file *file, loff_t offset, int whence) |
5559b8f4 MS |
2753 | { |
2754 | loff_t retval; | |
6131ffaa | 2755 | struct inode *inode = file_inode(file); |
5559b8f4 | 2756 | |
0b5da8db R |
2757 | switch (whence) { |
2758 | case SEEK_SET: | |
2759 | case SEEK_CUR: | |
2760 | /* No i_mutex protection necessary for SEEK_CUR and SEEK_SET */ | |
965c8e59 | 2761 | retval = generic_file_llseek(file, offset, whence); |
0b5da8db R |
2762 | break; |
2763 | case SEEK_END: | |
5955102c | 2764 | inode_lock(inode); |
c6c745b8 | 2765 | retval = fuse_update_attributes(inode, file, STATX_SIZE); |
0b5da8db R |
2766 | if (!retval) |
2767 | retval = generic_file_llseek(file, offset, whence); | |
5955102c | 2768 | inode_unlock(inode); |
0b5da8db R |
2769 | break; |
2770 | case SEEK_HOLE: | |
2771 | case SEEK_DATA: | |
5955102c | 2772 | inode_lock(inode); |
0b5da8db | 2773 | retval = fuse_lseek(file, offset, whence); |
5955102c | 2774 | inode_unlock(inode); |
0b5da8db R |
2775 | break; |
2776 | default: | |
2777 | retval = -EINVAL; | |
2778 | } | |
c07c3d19 | 2779 | |
5559b8f4 MS |
2780 | return retval; |
2781 | } | |
2782 | ||
95668a69 TH |
2783 | /* |
2784 | * All files which have been polled are linked to RB tree | |
2785 | * fuse_conn->polled_files which is indexed by kh. Walk the tree and | |
2786 | * find the matching one. | |
2787 | */ | |
2788 | static struct rb_node **fuse_find_polled_node(struct fuse_conn *fc, u64 kh, | |
2789 | struct rb_node **parent_out) | |
2790 | { | |
2791 | struct rb_node **link = &fc->polled_files.rb_node; | |
2792 | struct rb_node *last = NULL; | |
2793 | ||
2794 | while (*link) { | |
2795 | struct fuse_file *ff; | |
2796 | ||
2797 | last = *link; | |
2798 | ff = rb_entry(last, struct fuse_file, polled_node); | |
2799 | ||
2800 | if (kh < ff->kh) | |
2801 | link = &last->rb_left; | |
2802 | else if (kh > ff->kh) | |
2803 | link = &last->rb_right; | |
2804 | else | |
2805 | return link; | |
2806 | } | |
2807 | ||
2808 | if (parent_out) | |
2809 | *parent_out = last; | |
2810 | return link; | |
2811 | } | |
2812 | ||
2813 | /* | |
2814 | * The file is about to be polled. Make sure it's on the polled_files | |
2815 | * RB tree. Note that files once added to the polled_files tree are | |
2816 | * not removed before the file is released. This is because a file | |
2817 | * polled once is likely to be polled again. | |
2818 | */ | |
2819 | static void fuse_register_polled_file(struct fuse_conn *fc, | |
2820 | struct fuse_file *ff) | |
2821 | { | |
2822 | spin_lock(&fc->lock); | |
2823 | if (RB_EMPTY_NODE(&ff->polled_node)) { | |
3f649ab7 | 2824 | struct rb_node **link, *parent; |
95668a69 TH |
2825 | |
2826 | link = fuse_find_polled_node(fc, ff->kh, &parent); | |
2827 | BUG_ON(*link); | |
2828 | rb_link_node(&ff->polled_node, parent, link); | |
2829 | rb_insert_color(&ff->polled_node, &fc->polled_files); | |
2830 | } | |
2831 | spin_unlock(&fc->lock); | |
2832 | } | |
2833 | ||
076ccb76 | 2834 | __poll_t fuse_file_poll(struct file *file, poll_table *wait) |
95668a69 | 2835 | { |
95668a69 | 2836 | struct fuse_file *ff = file->private_data; |
fcee216b | 2837 | struct fuse_mount *fm = ff->fm; |
95668a69 TH |
2838 | struct fuse_poll_in inarg = { .fh = ff->fh, .kh = ff->kh }; |
2839 | struct fuse_poll_out outarg; | |
7078187a | 2840 | FUSE_ARGS(args); |
95668a69 TH |
2841 | int err; |
2842 | ||
fcee216b | 2843 | if (fm->fc->no_poll) |
95668a69 TH |
2844 | return DEFAULT_POLLMASK; |
2845 | ||
2846 | poll_wait(file, &ff->poll_wait, wait); | |
c71d227f | 2847 | inarg.events = mangle_poll(poll_requested_events(wait)); |
95668a69 TH |
2848 | |
2849 | /* | |
2850 | * Ask for notification iff there's someone waiting for it. | |
2851 | * The client may ignore the flag and always notify. | |
2852 | */ | |
2853 | if (waitqueue_active(&ff->poll_wait)) { | |
2854 | inarg.flags |= FUSE_POLL_SCHEDULE_NOTIFY; | |
fcee216b | 2855 | fuse_register_polled_file(fm->fc, ff); |
95668a69 TH |
2856 | } |
2857 | ||
d5b48543 MS |
2858 | args.opcode = FUSE_POLL; |
2859 | args.nodeid = ff->nodeid; | |
2860 | args.in_numargs = 1; | |
2861 | args.in_args[0].size = sizeof(inarg); | |
2862 | args.in_args[0].value = &inarg; | |
2863 | args.out_numargs = 1; | |
2864 | args.out_args[0].size = sizeof(outarg); | |
2865 | args.out_args[0].value = &outarg; | |
fcee216b | 2866 | err = fuse_simple_request(fm, &args); |
95668a69 TH |
2867 | |
2868 | if (!err) | |
c71d227f | 2869 | return demangle_poll(outarg.revents); |
95668a69 | 2870 | if (err == -ENOSYS) { |
fcee216b | 2871 | fm->fc->no_poll = 1; |
95668a69 TH |
2872 | return DEFAULT_POLLMASK; |
2873 | } | |
a9a08845 | 2874 | return EPOLLERR; |
95668a69 | 2875 | } |
08cbf542 | 2876 | EXPORT_SYMBOL_GPL(fuse_file_poll); |
95668a69 TH |
2877 | |
2878 | /* | |
2879 | * This is called from fuse_handle_notify() on FUSE_NOTIFY_POLL and | |
2880 | * wakes up the poll waiters. | |
2881 | */ | |
2882 | int fuse_notify_poll_wakeup(struct fuse_conn *fc, | |
2883 | struct fuse_notify_poll_wakeup_out *outarg) | |
2884 | { | |
2885 | u64 kh = outarg->kh; | |
2886 | struct rb_node **link; | |
2887 | ||
2888 | spin_lock(&fc->lock); | |
2889 | ||
2890 | link = fuse_find_polled_node(fc, kh, NULL); | |
2891 | if (*link) { | |
2892 | struct fuse_file *ff; | |
2893 | ||
2894 | ff = rb_entry(*link, struct fuse_file, polled_node); | |
2895 | wake_up_interruptible_sync(&ff->poll_wait); | |
2896 | } | |
2897 | ||
2898 | spin_unlock(&fc->lock); | |
2899 | return 0; | |
2900 | } | |
2901 | ||
efb9fa9e MP |
2902 | static void fuse_do_truncate(struct file *file) |
2903 | { | |
2904 | struct inode *inode = file->f_mapping->host; | |
2905 | struct iattr attr; | |
2906 | ||
2907 | attr.ia_valid = ATTR_SIZE; | |
2908 | attr.ia_size = i_size_read(inode); | |
2909 | ||
2910 | attr.ia_file = file; | |
2911 | attr.ia_valid |= ATTR_FILE; | |
2912 | ||
62490330 | 2913 | fuse_do_setattr(file_dentry(file), &attr, file); |
efb9fa9e MP |
2914 | } |
2915 | ||
5da784cc | 2916 | static inline loff_t fuse_round_up(struct fuse_conn *fc, loff_t off) |
e5c5f05d | 2917 | { |
5da784cc | 2918 | return round_up(off, fc->max_pages << PAGE_SHIFT); |
e5c5f05d MP |
2919 | } |
2920 | ||
4273b793 | 2921 | static ssize_t |
c8b8e32d | 2922 | fuse_direct_IO(struct kiocb *iocb, struct iov_iter *iter) |
4273b793 | 2923 | { |
9d5722b7 | 2924 | DECLARE_COMPLETION_ONSTACK(wait); |
4273b793 | 2925 | ssize_t ret = 0; |
60b9df7a MS |
2926 | struct file *file = iocb->ki_filp; |
2927 | struct fuse_file *ff = file->private_data; | |
4273b793 | 2928 | loff_t pos = 0; |
bcba24cc MP |
2929 | struct inode *inode; |
2930 | loff_t i_size; | |
933a3752 | 2931 | size_t count = iov_iter_count(iter), shortened = 0; |
c8b8e32d | 2932 | loff_t offset = iocb->ki_pos; |
36cf66ed | 2933 | struct fuse_io_priv *io; |
4273b793 | 2934 | |
4273b793 | 2935 | pos = offset; |
bcba24cc MP |
2936 | inode = file->f_mapping->host; |
2937 | i_size = i_size_read(inode); | |
4273b793 | 2938 | |
933a3752 | 2939 | if ((iov_iter_rw(iter) == READ) && (offset >= i_size)) |
9fe55eea SW |
2940 | return 0; |
2941 | ||
bcba24cc | 2942 | io = kmalloc(sizeof(struct fuse_io_priv), GFP_KERNEL); |
36cf66ed MP |
2943 | if (!io) |
2944 | return -ENOMEM; | |
bcba24cc | 2945 | spin_lock_init(&io->lock); |
744742d6 | 2946 | kref_init(&io->refcnt); |
bcba24cc MP |
2947 | io->reqs = 1; |
2948 | io->bytes = -1; | |
2949 | io->size = 0; | |
2950 | io->offset = offset; | |
6f673763 | 2951 | io->write = (iov_iter_rw(iter) == WRITE); |
bcba24cc | 2952 | io->err = 0; |
bcba24cc MP |
2953 | /* |
2954 | * By default, we want to optimize all I/Os with async request | |
60b9df7a | 2955 | * submission to the client filesystem if supported. |
bcba24cc | 2956 | */ |
69456535 | 2957 | io->async = ff->fm->fc->async_dio; |
bcba24cc | 2958 | io->iocb = iocb; |
7879c4e5 | 2959 | io->blocking = is_sync_kiocb(iocb); |
bcba24cc | 2960 | |
933a3752 AV |
2961 | /* optimization for short read */ |
2962 | if (io->async && !io->write && offset + count > i_size) { | |
69456535 | 2963 | iov_iter_truncate(iter, fuse_round_up(ff->fm->fc, i_size - offset)); |
933a3752 AV |
2964 | shortened = count - iov_iter_count(iter); |
2965 | count -= shortened; | |
2966 | } | |
2967 | ||
bcba24cc | 2968 | /* |
7879c4e5 AS |
2969 | * We cannot asynchronously extend the size of a file. |
2970 | * In such case the aio will behave exactly like sync io. | |
bcba24cc | 2971 | */ |
933a3752 | 2972 | if ((offset + count > i_size) && io->write) |
7879c4e5 | 2973 | io->blocking = true; |
4273b793 | 2974 | |
7879c4e5 | 2975 | if (io->async && io->blocking) { |
744742d6 SF |
2976 | /* |
2977 | * Additional reference to keep io around after | |
2978 | * calling fuse_aio_complete() | |
2979 | */ | |
2980 | kref_get(&io->refcnt); | |
9d5722b7 | 2981 | io->done = &wait; |
744742d6 | 2982 | } |
9d5722b7 | 2983 | |
6f673763 | 2984 | if (iov_iter_rw(iter) == WRITE) { |
6b775b18 | 2985 | ret = fuse_direct_io(io, iter, &pos, FUSE_DIO_WRITE); |
fa5eee57 | 2986 | fuse_invalidate_attr_mask(inode, FUSE_STATX_MODSIZE); |
812408fb | 2987 | } else { |
d22a943f | 2988 | ret = __fuse_direct_read(io, iter, &pos); |
812408fb | 2989 | } |
933a3752 | 2990 | iov_iter_reexpand(iter, iov_iter_count(iter) + shortened); |
36cf66ed | 2991 | |
bcba24cc | 2992 | if (io->async) { |
ebacb812 LC |
2993 | bool blocking = io->blocking; |
2994 | ||
bcba24cc MP |
2995 | fuse_aio_complete(io, ret < 0 ? ret : 0, -1); |
2996 | ||
2997 | /* we have a non-extending, async request, so return */ | |
ebacb812 | 2998 | if (!blocking) |
bcba24cc MP |
2999 | return -EIOCBQUEUED; |
3000 | ||
9d5722b7 CH |
3001 | wait_for_completion(&wait); |
3002 | ret = fuse_get_res_by_io(io); | |
bcba24cc MP |
3003 | } |
3004 | ||
744742d6 | 3005 | kref_put(&io->refcnt, fuse_io_release); |
9d5722b7 | 3006 | |
6f673763 | 3007 | if (iov_iter_rw(iter) == WRITE) { |
d347739a | 3008 | fuse_write_update_attr(inode, pos, ret); |
15352405 | 3009 | /* For extending writes we already hold exclusive lock */ |
d347739a | 3010 | if (ret < 0 && offset + count > i_size) |
efb9fa9e MP |
3011 | fuse_do_truncate(file); |
3012 | } | |
4273b793 AA |
3013 | |
3014 | return ret; | |
3015 | } | |
3016 | ||
26eb3bae MS |
3017 | static int fuse_writeback_range(struct inode *inode, loff_t start, loff_t end) |
3018 | { | |
e388164e | 3019 | int err = filemap_write_and_wait_range(inode->i_mapping, start, LLONG_MAX); |
26eb3bae MS |
3020 | |
3021 | if (!err) | |
3022 | fuse_sync_writes(inode); | |
3023 | ||
3024 | return err; | |
3025 | } | |
3026 | ||
cdadb11c MS |
3027 | static long fuse_file_fallocate(struct file *file, int mode, loff_t offset, |
3028 | loff_t length) | |
05ba1f08 AP |
3029 | { |
3030 | struct fuse_file *ff = file->private_data; | |
1c68271c | 3031 | struct inode *inode = file_inode(file); |
0ab08f57 | 3032 | struct fuse_inode *fi = get_fuse_inode(inode); |
fcee216b | 3033 | struct fuse_mount *fm = ff->fm; |
7078187a | 3034 | FUSE_ARGS(args); |
05ba1f08 AP |
3035 | struct fuse_fallocate_in inarg = { |
3036 | .fh = ff->fh, | |
3037 | .offset = offset, | |
3038 | .length = length, | |
3039 | .mode = mode | |
3040 | }; | |
3041 | int err; | |
44361e8c MS |
3042 | bool block_faults = FUSE_IS_DAX(inode) && |
3043 | (!(mode & FALLOC_FL_KEEP_SIZE) || | |
3044 | (mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE))); | |
6ae330ca | 3045 | |
6b1bdb56 RJ |
3046 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE | |
3047 | FALLOC_FL_ZERO_RANGE)) | |
4adb8302 MS |
3048 | return -EOPNOTSUPP; |
3049 | ||
fcee216b | 3050 | if (fm->fc->no_fallocate) |
519c6040 MS |
3051 | return -EOPNOTSUPP; |
3052 | ||
44361e8c MS |
3053 | inode_lock(inode); |
3054 | if (block_faults) { | |
3055 | filemap_invalidate_lock(inode->i_mapping); | |
3056 | err = fuse_dax_break_layouts(inode, 0, 0); | |
3057 | if (err) | |
3058 | goto out; | |
3059 | } | |
6ae330ca | 3060 | |
44361e8c MS |
3061 | if (mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) { |
3062 | loff_t endbyte = offset + length - 1; | |
26eb3bae | 3063 | |
44361e8c MS |
3064 | err = fuse_writeback_range(inode, offset, endbyte); |
3065 | if (err) | |
3066 | goto out; | |
3634a632 BF |
3067 | } |
3068 | ||
0cbade02 LB |
3069 | if (!(mode & FALLOC_FL_KEEP_SIZE) && |
3070 | offset + length > i_size_read(inode)) { | |
3071 | err = inode_newsize_ok(inode, offset + length); | |
3072 | if (err) | |
35d6fcbb | 3073 | goto out; |
0cbade02 LB |
3074 | } |
3075 | ||
4a6f278d MS |
3076 | err = file_modified(file); |
3077 | if (err) | |
3078 | goto out; | |
3079 | ||
0ab08f57 MP |
3080 | if (!(mode & FALLOC_FL_KEEP_SIZE)) |
3081 | set_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); | |
3082 | ||
d5b48543 MS |
3083 | args.opcode = FUSE_FALLOCATE; |
3084 | args.nodeid = ff->nodeid; | |
3085 | args.in_numargs = 1; | |
3086 | args.in_args[0].size = sizeof(inarg); | |
3087 | args.in_args[0].value = &inarg; | |
fcee216b | 3088 | err = fuse_simple_request(fm, &args); |
519c6040 | 3089 | if (err == -ENOSYS) { |
fcee216b | 3090 | fm->fc->no_fallocate = 1; |
519c6040 MS |
3091 | err = -EOPNOTSUPP; |
3092 | } | |
bee6c307 BF |
3093 | if (err) |
3094 | goto out; | |
3095 | ||
3096 | /* we could have extended the file */ | |
b0aa7606 | 3097 | if (!(mode & FALLOC_FL_KEEP_SIZE)) { |
20235b43 | 3098 | if (fuse_write_update_attr(inode, offset + length, length)) |
93d2269d | 3099 | file_update_time(file); |
b0aa7606 | 3100 | } |
bee6c307 | 3101 | |
6b1bdb56 | 3102 | if (mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_ZERO_RANGE)) |
bee6c307 BF |
3103 | truncate_pagecache_range(inode, offset, offset + length - 1); |
3104 | ||
fa5eee57 | 3105 | fuse_invalidate_attr_mask(inode, FUSE_STATX_MODSIZE); |
bee6c307 | 3106 | |
3634a632 | 3107 | out: |
0ab08f57 MP |
3108 | if (!(mode & FALLOC_FL_KEEP_SIZE)) |
3109 | clear_bit(FUSE_I_SIZE_UNSTABLE, &fi->state); | |
3110 | ||
6ae330ca | 3111 | if (block_faults) |
8bcbbe9c | 3112 | filemap_invalidate_unlock(inode->i_mapping); |
6ae330ca | 3113 | |
44361e8c | 3114 | inode_unlock(inode); |
3634a632 | 3115 | |
5c791fe1 MS |
3116 | fuse_flush_time_update(inode); |
3117 | ||
05ba1f08 AP |
3118 | return err; |
3119 | } | |
05ba1f08 | 3120 | |
64bf5ff5 DC |
3121 | static ssize_t __fuse_copy_file_range(struct file *file_in, loff_t pos_in, |
3122 | struct file *file_out, loff_t pos_out, | |
3123 | size_t len, unsigned int flags) | |
88bc7d50 NV |
3124 | { |
3125 | struct fuse_file *ff_in = file_in->private_data; | |
3126 | struct fuse_file *ff_out = file_out->private_data; | |
a2bc9236 | 3127 | struct inode *inode_in = file_inode(file_in); |
88bc7d50 NV |
3128 | struct inode *inode_out = file_inode(file_out); |
3129 | struct fuse_inode *fi_out = get_fuse_inode(inode_out); | |
fcee216b MR |
3130 | struct fuse_mount *fm = ff_in->fm; |
3131 | struct fuse_conn *fc = fm->fc; | |
88bc7d50 NV |
3132 | FUSE_ARGS(args); |
3133 | struct fuse_copy_file_range_in inarg = { | |
3134 | .fh_in = ff_in->fh, | |
3135 | .off_in = pos_in, | |
3136 | .nodeid_out = ff_out->nodeid, | |
3137 | .fh_out = ff_out->fh, | |
3138 | .off_out = pos_out, | |
3139 | .len = len, | |
3140 | .flags = flags | |
3141 | }; | |
3142 | struct fuse_write_out outarg; | |
3143 | ssize_t err; | |
3144 | /* mark unstable when write-back is not used, and file_out gets | |
3145 | * extended */ | |
3146 | bool is_unstable = (!fc->writeback_cache) && | |
3147 | ((pos_out + len) > inode_out->i_size); | |
3148 | ||
3149 | if (fc->no_copy_file_range) | |
3150 | return -EOPNOTSUPP; | |
3151 | ||
5dae222a AG |
3152 | if (file_inode(file_in)->i_sb != file_inode(file_out)->i_sb) |
3153 | return -EXDEV; | |
3154 | ||
2c4656df MS |
3155 | inode_lock(inode_in); |
3156 | err = fuse_writeback_range(inode_in, pos_in, pos_in + len - 1); | |
3157 | inode_unlock(inode_in); | |
3158 | if (err) | |
3159 | return err; | |
a2bc9236 | 3160 | |
88bc7d50 NV |
3161 | inode_lock(inode_out); |
3162 | ||
fe0da9c0 AG |
3163 | err = file_modified(file_out); |
3164 | if (err) | |
3165 | goto out; | |
3166 | ||
9b46418c MS |
3167 | /* |
3168 | * Write out dirty pages in the destination file before sending the COPY | |
3169 | * request to userspace. After the request is completed, truncate off | |
3170 | * pages (including partial ones) from the cache that have been copied, | |
3171 | * since these contain stale data at that point. | |
3172 | * | |
3173 | * This should be mostly correct, but if the COPY writes to partial | |
3174 | * pages (at the start or end) and the parts not covered by the COPY are | |
3175 | * written through a memory map after calling fuse_writeback_range(), | |
3176 | * then these partial page modifications will be lost on truncation. | |
3177 | * | |
3178 | * It is unlikely that someone would rely on such mixed style | |
3179 | * modifications. Yet this does give less guarantees than if the | |
3180 | * copying was performed with write(2). | |
3181 | * | |
8bcbbe9c | 3182 | * To fix this a mapping->invalidate_lock could be used to prevent new |
9b46418c MS |
3183 | * faults while the copy is ongoing. |
3184 | */ | |
2c4656df MS |
3185 | err = fuse_writeback_range(inode_out, pos_out, pos_out + len - 1); |
3186 | if (err) | |
3187 | goto out; | |
88bc7d50 NV |
3188 | |
3189 | if (is_unstable) | |
3190 | set_bit(FUSE_I_SIZE_UNSTABLE, &fi_out->state); | |
3191 | ||
d5b48543 MS |
3192 | args.opcode = FUSE_COPY_FILE_RANGE; |
3193 | args.nodeid = ff_in->nodeid; | |
3194 | args.in_numargs = 1; | |
3195 | args.in_args[0].size = sizeof(inarg); | |
3196 | args.in_args[0].value = &inarg; | |
3197 | args.out_numargs = 1; | |
3198 | args.out_args[0].size = sizeof(outarg); | |
3199 | args.out_args[0].value = &outarg; | |
fcee216b | 3200 | err = fuse_simple_request(fm, &args); |
88bc7d50 NV |
3201 | if (err == -ENOSYS) { |
3202 | fc->no_copy_file_range = 1; | |
3203 | err = -EOPNOTSUPP; | |
3204 | } | |
3205 | if (err) | |
3206 | goto out; | |
3207 | ||
9b46418c MS |
3208 | truncate_inode_pages_range(inode_out->i_mapping, |
3209 | ALIGN_DOWN(pos_out, PAGE_SIZE), | |
3210 | ALIGN(pos_out + outarg.size, PAGE_SIZE) - 1); | |
3211 | ||
20235b43 MS |
3212 | file_update_time(file_out); |
3213 | fuse_write_update_attr(inode_out, pos_out + outarg.size, outarg.size); | |
88bc7d50 NV |
3214 | |
3215 | err = outarg.size; | |
3216 | out: | |
3217 | if (is_unstable) | |
3218 | clear_bit(FUSE_I_SIZE_UNSTABLE, &fi_out->state); | |
3219 | ||
3220 | inode_unlock(inode_out); | |
fe0da9c0 | 3221 | file_accessed(file_in); |
88bc7d50 | 3222 | |
5c791fe1 MS |
3223 | fuse_flush_time_update(inode_out); |
3224 | ||
88bc7d50 NV |
3225 | return err; |
3226 | } | |
3227 | ||
64bf5ff5 DC |
3228 | static ssize_t fuse_copy_file_range(struct file *src_file, loff_t src_off, |
3229 | struct file *dst_file, loff_t dst_off, | |
3230 | size_t len, unsigned int flags) | |
3231 | { | |
3232 | ssize_t ret; | |
3233 | ||
3234 | ret = __fuse_copy_file_range(src_file, src_off, dst_file, dst_off, | |
3235 | len, flags); | |
3236 | ||
5dae222a | 3237 | if (ret == -EOPNOTSUPP || ret == -EXDEV) |
64bf5ff5 DC |
3238 | ret = generic_copy_file_range(src_file, src_off, dst_file, |
3239 | dst_off, len, flags); | |
3240 | return ret; | |
3241 | } | |
3242 | ||
4b6f5d20 | 3243 | static const struct file_operations fuse_file_operations = { |
5559b8f4 | 3244 | .llseek = fuse_file_llseek, |
37c20f16 | 3245 | .read_iter = fuse_file_read_iter, |
84c3d55c | 3246 | .write_iter = fuse_file_write_iter, |
b6aeaded MS |
3247 | .mmap = fuse_file_mmap, |
3248 | .open = fuse_open, | |
3249 | .flush = fuse_flush, | |
3250 | .release = fuse_release, | |
3251 | .fsync = fuse_fsync, | |
71421259 | 3252 | .lock = fuse_file_lock, |
2a9a609a | 3253 | .get_unmapped_area = thp_get_unmapped_area, |
a9ff4f87 | 3254 | .flock = fuse_file_flock, |
2cb1e089 | 3255 | .splice_read = filemap_splice_read, |
3c3db095 | 3256 | .splice_write = iter_file_splice_write, |
59efec7b TH |
3257 | .unlocked_ioctl = fuse_file_ioctl, |
3258 | .compat_ioctl = fuse_file_compat_ioctl, | |
95668a69 | 3259 | .poll = fuse_file_poll, |
05ba1f08 | 3260 | .fallocate = fuse_file_fallocate, |
d4136d60 | 3261 | .copy_file_range = fuse_copy_file_range, |
413ef8cb MS |
3262 | }; |
3263 | ||
f5e54d6e | 3264 | static const struct address_space_operations fuse_file_aops = { |
5efd00e4 | 3265 | .read_folio = fuse_read_folio, |
76a0294e | 3266 | .readahead = fuse_readahead, |
3be5a52b | 3267 | .writepage = fuse_writepage, |
26d614df | 3268 | .writepages = fuse_writepages, |
2bf06b8e | 3269 | .launder_folio = fuse_launder_folio, |
187c82cb | 3270 | .dirty_folio = filemap_dirty_folio, |
b2d2272f | 3271 | .bmap = fuse_bmap, |
4273b793 | 3272 | .direct_IO = fuse_direct_IO, |
6b12c1b3 PE |
3273 | .write_begin = fuse_write_begin, |
3274 | .write_end = fuse_write_end, | |
b6aeaded MS |
3275 | }; |
3276 | ||
93a497b9 | 3277 | void fuse_init_file_inode(struct inode *inode, unsigned int flags) |
b6aeaded | 3278 | { |
ab2257e9 MS |
3279 | struct fuse_inode *fi = get_fuse_inode(inode); |
3280 | ||
45323fb7 MS |
3281 | inode->i_fop = &fuse_file_operations; |
3282 | inode->i_data.a_ops = &fuse_file_aops; | |
ab2257e9 MS |
3283 | |
3284 | INIT_LIST_HEAD(&fi->write_files); | |
3285 | INIT_LIST_HEAD(&fi->queued_writes); | |
3286 | fi->writectr = 0; | |
3287 | init_waitqueue_head(&fi->page_waitq); | |
6b2fb799 | 3288 | fi->writepages = RB_ROOT; |
c2d0ad00 VG |
3289 | |
3290 | if (IS_ENABLED(CONFIG_FUSE_DAX)) | |
93a497b9 | 3291 | fuse_dax_inode_init(inode, flags); |
b6aeaded | 3292 | } |