]>
Commit | Line | Data |
---|---|---|
b8441ed2 TH |
1 | /* |
2 | * fs/kernfs/file.c - kernfs file implementation | |
3 | * | |
4 | * Copyright (c) 2001-3 Patrick Mochel | |
5 | * Copyright (c) 2007 SUSE Linux Products GmbH | |
6 | * Copyright (c) 2007, 2013 Tejun Heo <tj@kernel.org> | |
7 | * | |
8 | * This file is released under the GPLv2. | |
9 | */ | |
414985ae TH |
10 | |
11 | #include <linux/fs.h> | |
12 | #include <linux/seq_file.h> | |
13 | #include <linux/slab.h> | |
14 | #include <linux/poll.h> | |
15 | #include <linux/pagemap.h> | |
414985ae | 16 | #include <linux/sched.h> |
d911d987 | 17 | #include <linux/fsnotify.h> |
414985ae TH |
18 | |
19 | #include "kernfs-internal.h" | |
20 | ||
21 | /* | |
c525aadd | 22 | * There's one kernfs_open_file for each open file and one kernfs_open_node |
324a56e1 | 23 | * for each kernfs_node with one or more open files. |
414985ae | 24 | * |
c525aadd TH |
25 | * kernfs_node->attr.open points to kernfs_open_node. attr.open is |
26 | * protected by kernfs_open_node_lock. | |
414985ae TH |
27 | * |
28 | * filp->private_data points to seq_file whose ->private points to | |
c525aadd TH |
29 | * kernfs_open_file. kernfs_open_files are chained at |
30 | * kernfs_open_node->files, which is protected by kernfs_open_file_mutex. | |
414985ae | 31 | */ |
c525aadd TH |
32 | static DEFINE_SPINLOCK(kernfs_open_node_lock); |
33 | static DEFINE_MUTEX(kernfs_open_file_mutex); | |
414985ae | 34 | |
c525aadd | 35 | struct kernfs_open_node { |
414985ae TH |
36 | atomic_t refcnt; |
37 | atomic_t event; | |
38 | wait_queue_head_t poll; | |
c525aadd | 39 | struct list_head files; /* goes through kernfs_open_file.list */ |
414985ae TH |
40 | }; |
41 | ||
ecca47ce TH |
42 | /* |
43 | * kernfs_notify() may be called from any context and bounces notifications | |
44 | * through a work item. To minimize space overhead in kernfs_node, the | |
45 | * pending queue is implemented as a singly linked list of kernfs_nodes. | |
46 | * The list is terminated with the self pointer so that whether a | |
47 | * kernfs_node is on the list or not can be determined by testing the next | |
48 | * pointer for NULL. | |
49 | */ | |
50 | #define KERNFS_NOTIFY_EOL ((void *)&kernfs_notify_list) | |
51 | ||
52 | static DEFINE_SPINLOCK(kernfs_notify_lock); | |
53 | static struct kernfs_node *kernfs_notify_list = KERNFS_NOTIFY_EOL; | |
54 | ||
c525aadd | 55 | static struct kernfs_open_file *kernfs_of(struct file *file) |
414985ae TH |
56 | { |
57 | return ((struct seq_file *)file->private_data)->private; | |
58 | } | |
59 | ||
60 | /* | |
324a56e1 | 61 | * Determine the kernfs_ops for the given kernfs_node. This function must |
414985ae TH |
62 | * be called while holding an active reference. |
63 | */ | |
324a56e1 | 64 | static const struct kernfs_ops *kernfs_ops(struct kernfs_node *kn) |
414985ae | 65 | { |
df23fc39 | 66 | if (kn->flags & KERNFS_LOCKDEP) |
324a56e1 | 67 | lockdep_assert_held(kn); |
adc5e8b5 | 68 | return kn->attr.ops; |
414985ae TH |
69 | } |
70 | ||
bb305947 TH |
71 | /* |
72 | * As kernfs_seq_stop() is also called after kernfs_seq_start() or | |
73 | * kernfs_seq_next() failure, it needs to distinguish whether it's stopping | |
74 | * a seq_file iteration which is fully initialized with an active reference | |
75 | * or an aborted kernfs_seq_start() due to get_active failure. The | |
76 | * position pointer is the only context for each seq_file iteration and | |
77 | * thus the stop condition should be encoded in it. As the return value is | |
78 | * directly visible to userland, ERR_PTR(-ENODEV) is the only acceptable | |
79 | * choice to indicate get_active failure. | |
80 | * | |
81 | * Unfortunately, this is complicated due to the optional custom seq_file | |
82 | * operations which may return ERR_PTR(-ENODEV) too. kernfs_seq_stop() | |
83 | * can't distinguish whether ERR_PTR(-ENODEV) is from get_active failure or | |
84 | * custom seq_file operations and thus can't decide whether put_active | |
85 | * should be performed or not only on ERR_PTR(-ENODEV). | |
86 | * | |
87 | * This is worked around by factoring out the custom seq_stop() and | |
88 | * put_active part into kernfs_seq_stop_active(), skipping it from | |
89 | * kernfs_seq_stop() if ERR_PTR(-ENODEV) while invoking it directly after | |
90 | * custom seq_file operations fail with ERR_PTR(-ENODEV) - this ensures | |
91 | * that kernfs_seq_stop_active() is skipped only after get_active failure. | |
92 | */ | |
93 | static void kernfs_seq_stop_active(struct seq_file *sf, void *v) | |
94 | { | |
95 | struct kernfs_open_file *of = sf->private; | |
96 | const struct kernfs_ops *ops = kernfs_ops(of->kn); | |
97 | ||
98 | if (ops->seq_stop) | |
99 | ops->seq_stop(sf, v); | |
100 | kernfs_put_active(of->kn); | |
101 | } | |
102 | ||
414985ae TH |
103 | static void *kernfs_seq_start(struct seq_file *sf, loff_t *ppos) |
104 | { | |
c525aadd | 105 | struct kernfs_open_file *of = sf->private; |
414985ae TH |
106 | const struct kernfs_ops *ops; |
107 | ||
108 | /* | |
109 | * @of->mutex nests outside active ref and is just to ensure that | |
110 | * the ops aren't called concurrently for the same open file. | |
111 | */ | |
112 | mutex_lock(&of->mutex); | |
c637b8ac | 113 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
114 | return ERR_PTR(-ENODEV); |
115 | ||
324a56e1 | 116 | ops = kernfs_ops(of->kn); |
414985ae | 117 | if (ops->seq_start) { |
bb305947 TH |
118 | void *next = ops->seq_start(sf, ppos); |
119 | /* see the comment above kernfs_seq_stop_active() */ | |
120 | if (next == ERR_PTR(-ENODEV)) | |
121 | kernfs_seq_stop_active(sf, next); | |
122 | return next; | |
414985ae TH |
123 | } else { |
124 | /* | |
125 | * The same behavior and code as single_open(). Returns | |
126 | * !NULL if pos is at the beginning; otherwise, NULL. | |
127 | */ | |
128 | return NULL + !*ppos; | |
129 | } | |
130 | } | |
131 | ||
132 | static void *kernfs_seq_next(struct seq_file *sf, void *v, loff_t *ppos) | |
133 | { | |
c525aadd | 134 | struct kernfs_open_file *of = sf->private; |
324a56e1 | 135 | const struct kernfs_ops *ops = kernfs_ops(of->kn); |
414985ae TH |
136 | |
137 | if (ops->seq_next) { | |
bb305947 TH |
138 | void *next = ops->seq_next(sf, v, ppos); |
139 | /* see the comment above kernfs_seq_stop_active() */ | |
140 | if (next == ERR_PTR(-ENODEV)) | |
141 | kernfs_seq_stop_active(sf, next); | |
142 | return next; | |
414985ae TH |
143 | } else { |
144 | /* | |
145 | * The same behavior and code as single_open(), always | |
146 | * terminate after the initial read. | |
147 | */ | |
148 | ++*ppos; | |
149 | return NULL; | |
150 | } | |
151 | } | |
152 | ||
153 | static void kernfs_seq_stop(struct seq_file *sf, void *v) | |
154 | { | |
c525aadd | 155 | struct kernfs_open_file *of = sf->private; |
414985ae | 156 | |
bb305947 TH |
157 | if (v != ERR_PTR(-ENODEV)) |
158 | kernfs_seq_stop_active(sf, v); | |
414985ae TH |
159 | mutex_unlock(&of->mutex); |
160 | } | |
161 | ||
162 | static int kernfs_seq_show(struct seq_file *sf, void *v) | |
163 | { | |
c525aadd | 164 | struct kernfs_open_file *of = sf->private; |
414985ae | 165 | |
adc5e8b5 | 166 | of->event = atomic_read(&of->kn->attr.open->event); |
414985ae | 167 | |
adc5e8b5 | 168 | return of->kn->attr.ops->seq_show(sf, v); |
414985ae TH |
169 | } |
170 | ||
171 | static const struct seq_operations kernfs_seq_ops = { | |
172 | .start = kernfs_seq_start, | |
173 | .next = kernfs_seq_next, | |
174 | .stop = kernfs_seq_stop, | |
175 | .show = kernfs_seq_show, | |
176 | }; | |
177 | ||
178 | /* | |
179 | * As reading a bin file can have side-effects, the exact offset and bytes | |
180 | * specified in read(2) call should be passed to the read callback making | |
181 | * it difficult to use seq_file. Implement simplistic custom buffering for | |
182 | * bin files. | |
183 | */ | |
c525aadd | 184 | static ssize_t kernfs_file_direct_read(struct kernfs_open_file *of, |
414985ae TH |
185 | char __user *user_buf, size_t count, |
186 | loff_t *ppos) | |
187 | { | |
188 | ssize_t len = min_t(size_t, count, PAGE_SIZE); | |
189 | const struct kernfs_ops *ops; | |
190 | char *buf; | |
191 | ||
192 | buf = kmalloc(len, GFP_KERNEL); | |
193 | if (!buf) | |
194 | return -ENOMEM; | |
195 | ||
196 | /* | |
197 | * @of->mutex nests outside active ref and is just to ensure that | |
198 | * the ops aren't called concurrently for the same open file. | |
199 | */ | |
200 | mutex_lock(&of->mutex); | |
c637b8ac | 201 | if (!kernfs_get_active(of->kn)) { |
414985ae TH |
202 | len = -ENODEV; |
203 | mutex_unlock(&of->mutex); | |
204 | goto out_free; | |
205 | } | |
206 | ||
324a56e1 | 207 | ops = kernfs_ops(of->kn); |
414985ae TH |
208 | if (ops->read) |
209 | len = ops->read(of, buf, len, *ppos); | |
210 | else | |
211 | len = -EINVAL; | |
212 | ||
c637b8ac | 213 | kernfs_put_active(of->kn); |
414985ae TH |
214 | mutex_unlock(&of->mutex); |
215 | ||
216 | if (len < 0) | |
217 | goto out_free; | |
218 | ||
219 | if (copy_to_user(user_buf, buf, len)) { | |
220 | len = -EFAULT; | |
221 | goto out_free; | |
222 | } | |
223 | ||
224 | *ppos += len; | |
225 | ||
226 | out_free: | |
227 | kfree(buf); | |
228 | return len; | |
229 | } | |
230 | ||
231 | /** | |
c637b8ac | 232 | * kernfs_fop_read - kernfs vfs read callback |
414985ae TH |
233 | * @file: file pointer |
234 | * @user_buf: data to write | |
235 | * @count: number of bytes | |
236 | * @ppos: starting offset | |
237 | */ | |
c637b8ac TH |
238 | static ssize_t kernfs_fop_read(struct file *file, char __user *user_buf, |
239 | size_t count, loff_t *ppos) | |
414985ae | 240 | { |
c525aadd | 241 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae | 242 | |
df23fc39 | 243 | if (of->kn->flags & KERNFS_HAS_SEQ_SHOW) |
414985ae TH |
244 | return seq_read(file, user_buf, count, ppos); |
245 | else | |
246 | return kernfs_file_direct_read(of, user_buf, count, ppos); | |
247 | } | |
248 | ||
249 | /** | |
c637b8ac | 250 | * kernfs_fop_write - kernfs vfs write callback |
414985ae TH |
251 | * @file: file pointer |
252 | * @user_buf: data to write | |
253 | * @count: number of bytes | |
254 | * @ppos: starting offset | |
255 | * | |
256 | * Copy data in from userland and pass it to the matching kernfs write | |
257 | * operation. | |
258 | * | |
259 | * There is no easy way for us to know if userspace is only doing a partial | |
260 | * write, so we don't support them. We expect the entire buffer to come on | |
261 | * the first write. Hint: if you're writing a value, first read the file, | |
262 | * modify only the the value you're changing, then write entire buffer | |
263 | * back. | |
264 | */ | |
c637b8ac TH |
265 | static ssize_t kernfs_fop_write(struct file *file, const char __user *user_buf, |
266 | size_t count, loff_t *ppos) | |
414985ae | 267 | { |
c525aadd | 268 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae | 269 | const struct kernfs_ops *ops; |
b7ce40cf TH |
270 | size_t len; |
271 | char *buf; | |
4d3773c4 | 272 | |
b7ce40cf | 273 | if (of->atomic_write_len) { |
4d3773c4 | 274 | len = count; |
b7ce40cf TH |
275 | if (len > of->atomic_write_len) |
276 | return -E2BIG; | |
4d3773c4 TH |
277 | } else { |
278 | len = min_t(size_t, count, PAGE_SIZE); | |
279 | } | |
280 | ||
281 | buf = kmalloc(len + 1, GFP_KERNEL); | |
b7ce40cf TH |
282 | if (!buf) |
283 | return -ENOMEM; | |
414985ae | 284 | |
4d3773c4 TH |
285 | if (copy_from_user(buf, user_buf, len)) { |
286 | len = -EFAULT; | |
b7ce40cf | 287 | goto out_free; |
4d3773c4 TH |
288 | } |
289 | buf[len] = '\0'; /* guarantee string termination */ | |
290 | ||
b7ce40cf TH |
291 | /* |
292 | * @of->mutex nests outside active ref and is just to ensure that | |
293 | * the ops aren't called concurrently for the same open file. | |
294 | */ | |
295 | mutex_lock(&of->mutex); | |
296 | if (!kernfs_get_active(of->kn)) { | |
297 | mutex_unlock(&of->mutex); | |
298 | len = -ENODEV; | |
299 | goto out_free; | |
300 | } | |
301 | ||
302 | ops = kernfs_ops(of->kn); | |
303 | if (ops->write) | |
304 | len = ops->write(of, buf, len, *ppos); | |
305 | else | |
306 | len = -EINVAL; | |
307 | ||
c637b8ac | 308 | kernfs_put_active(of->kn); |
414985ae TH |
309 | mutex_unlock(&of->mutex); |
310 | ||
311 | if (len > 0) | |
312 | *ppos += len; | |
b7ce40cf | 313 | out_free: |
414985ae TH |
314 | kfree(buf); |
315 | return len; | |
316 | } | |
317 | ||
318 | static void kernfs_vma_open(struct vm_area_struct *vma) | |
319 | { | |
320 | struct file *file = vma->vm_file; | |
c525aadd | 321 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
322 | |
323 | if (!of->vm_ops) | |
324 | return; | |
325 | ||
c637b8ac | 326 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
327 | return; |
328 | ||
329 | if (of->vm_ops->open) | |
330 | of->vm_ops->open(vma); | |
331 | ||
c637b8ac | 332 | kernfs_put_active(of->kn); |
414985ae TH |
333 | } |
334 | ||
335 | static int kernfs_vma_fault(struct vm_area_struct *vma, struct vm_fault *vmf) | |
336 | { | |
337 | struct file *file = vma->vm_file; | |
c525aadd | 338 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
339 | int ret; |
340 | ||
341 | if (!of->vm_ops) | |
342 | return VM_FAULT_SIGBUS; | |
343 | ||
c637b8ac | 344 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
345 | return VM_FAULT_SIGBUS; |
346 | ||
347 | ret = VM_FAULT_SIGBUS; | |
348 | if (of->vm_ops->fault) | |
349 | ret = of->vm_ops->fault(vma, vmf); | |
350 | ||
c637b8ac | 351 | kernfs_put_active(of->kn); |
414985ae TH |
352 | return ret; |
353 | } | |
354 | ||
355 | static int kernfs_vma_page_mkwrite(struct vm_area_struct *vma, | |
356 | struct vm_fault *vmf) | |
357 | { | |
358 | struct file *file = vma->vm_file; | |
c525aadd | 359 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
360 | int ret; |
361 | ||
362 | if (!of->vm_ops) | |
363 | return VM_FAULT_SIGBUS; | |
364 | ||
c637b8ac | 365 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
366 | return VM_FAULT_SIGBUS; |
367 | ||
368 | ret = 0; | |
369 | if (of->vm_ops->page_mkwrite) | |
370 | ret = of->vm_ops->page_mkwrite(vma, vmf); | |
371 | else | |
372 | file_update_time(file); | |
373 | ||
c637b8ac | 374 | kernfs_put_active(of->kn); |
414985ae TH |
375 | return ret; |
376 | } | |
377 | ||
378 | static int kernfs_vma_access(struct vm_area_struct *vma, unsigned long addr, | |
379 | void *buf, int len, int write) | |
380 | { | |
381 | struct file *file = vma->vm_file; | |
c525aadd | 382 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
383 | int ret; |
384 | ||
385 | if (!of->vm_ops) | |
386 | return -EINVAL; | |
387 | ||
c637b8ac | 388 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
389 | return -EINVAL; |
390 | ||
391 | ret = -EINVAL; | |
392 | if (of->vm_ops->access) | |
393 | ret = of->vm_ops->access(vma, addr, buf, len, write); | |
394 | ||
c637b8ac | 395 | kernfs_put_active(of->kn); |
414985ae TH |
396 | return ret; |
397 | } | |
398 | ||
399 | #ifdef CONFIG_NUMA | |
400 | static int kernfs_vma_set_policy(struct vm_area_struct *vma, | |
401 | struct mempolicy *new) | |
402 | { | |
403 | struct file *file = vma->vm_file; | |
c525aadd | 404 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
405 | int ret; |
406 | ||
407 | if (!of->vm_ops) | |
408 | return 0; | |
409 | ||
c637b8ac | 410 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
411 | return -EINVAL; |
412 | ||
413 | ret = 0; | |
414 | if (of->vm_ops->set_policy) | |
415 | ret = of->vm_ops->set_policy(vma, new); | |
416 | ||
c637b8ac | 417 | kernfs_put_active(of->kn); |
414985ae TH |
418 | return ret; |
419 | } | |
420 | ||
421 | static struct mempolicy *kernfs_vma_get_policy(struct vm_area_struct *vma, | |
422 | unsigned long addr) | |
423 | { | |
424 | struct file *file = vma->vm_file; | |
c525aadd | 425 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
426 | struct mempolicy *pol; |
427 | ||
428 | if (!of->vm_ops) | |
429 | return vma->vm_policy; | |
430 | ||
c637b8ac | 431 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
432 | return vma->vm_policy; |
433 | ||
434 | pol = vma->vm_policy; | |
435 | if (of->vm_ops->get_policy) | |
436 | pol = of->vm_ops->get_policy(vma, addr); | |
437 | ||
c637b8ac | 438 | kernfs_put_active(of->kn); |
414985ae TH |
439 | return pol; |
440 | } | |
441 | ||
442 | static int kernfs_vma_migrate(struct vm_area_struct *vma, | |
443 | const nodemask_t *from, const nodemask_t *to, | |
444 | unsigned long flags) | |
445 | { | |
446 | struct file *file = vma->vm_file; | |
c525aadd | 447 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
448 | int ret; |
449 | ||
450 | if (!of->vm_ops) | |
451 | return 0; | |
452 | ||
c637b8ac | 453 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
454 | return 0; |
455 | ||
456 | ret = 0; | |
457 | if (of->vm_ops->migrate) | |
458 | ret = of->vm_ops->migrate(vma, from, to, flags); | |
459 | ||
c637b8ac | 460 | kernfs_put_active(of->kn); |
414985ae TH |
461 | return ret; |
462 | } | |
463 | #endif | |
464 | ||
465 | static const struct vm_operations_struct kernfs_vm_ops = { | |
466 | .open = kernfs_vma_open, | |
467 | .fault = kernfs_vma_fault, | |
468 | .page_mkwrite = kernfs_vma_page_mkwrite, | |
469 | .access = kernfs_vma_access, | |
470 | #ifdef CONFIG_NUMA | |
471 | .set_policy = kernfs_vma_set_policy, | |
472 | .get_policy = kernfs_vma_get_policy, | |
473 | .migrate = kernfs_vma_migrate, | |
474 | #endif | |
475 | }; | |
476 | ||
c637b8ac | 477 | static int kernfs_fop_mmap(struct file *file, struct vm_area_struct *vma) |
414985ae | 478 | { |
c525aadd | 479 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
480 | const struct kernfs_ops *ops; |
481 | int rc; | |
482 | ||
9b2db6e1 TH |
483 | /* |
484 | * mmap path and of->mutex are prone to triggering spurious lockdep | |
485 | * warnings and we don't want to add spurious locking dependency | |
486 | * between the two. Check whether mmap is actually implemented | |
487 | * without grabbing @of->mutex by testing HAS_MMAP flag. See the | |
488 | * comment in kernfs_file_open() for more details. | |
489 | */ | |
df23fc39 | 490 | if (!(of->kn->flags & KERNFS_HAS_MMAP)) |
9b2db6e1 TH |
491 | return -ENODEV; |
492 | ||
414985ae TH |
493 | mutex_lock(&of->mutex); |
494 | ||
495 | rc = -ENODEV; | |
c637b8ac | 496 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
497 | goto out_unlock; |
498 | ||
324a56e1 | 499 | ops = kernfs_ops(of->kn); |
9b2db6e1 | 500 | rc = ops->mmap(of, vma); |
b44b2140 TH |
501 | if (rc) |
502 | goto out_put; | |
414985ae TH |
503 | |
504 | /* | |
505 | * PowerPC's pci_mmap of legacy_mem uses shmem_zero_setup() | |
506 | * to satisfy versions of X which crash if the mmap fails: that | |
507 | * substitutes a new vm_file, and we don't then want bin_vm_ops. | |
508 | */ | |
509 | if (vma->vm_file != file) | |
510 | goto out_put; | |
511 | ||
512 | rc = -EINVAL; | |
513 | if (of->mmapped && of->vm_ops != vma->vm_ops) | |
514 | goto out_put; | |
515 | ||
516 | /* | |
517 | * It is not possible to successfully wrap close. | |
518 | * So error if someone is trying to use close. | |
519 | */ | |
520 | rc = -EINVAL; | |
521 | if (vma->vm_ops && vma->vm_ops->close) | |
522 | goto out_put; | |
523 | ||
524 | rc = 0; | |
525 | of->mmapped = 1; | |
526 | of->vm_ops = vma->vm_ops; | |
527 | vma->vm_ops = &kernfs_vm_ops; | |
528 | out_put: | |
c637b8ac | 529 | kernfs_put_active(of->kn); |
414985ae TH |
530 | out_unlock: |
531 | mutex_unlock(&of->mutex); | |
532 | ||
533 | return rc; | |
534 | } | |
535 | ||
536 | /** | |
c637b8ac | 537 | * kernfs_get_open_node - get or create kernfs_open_node |
324a56e1 | 538 | * @kn: target kernfs_node |
c525aadd | 539 | * @of: kernfs_open_file for this instance of open |
414985ae | 540 | * |
adc5e8b5 TH |
541 | * If @kn->attr.open exists, increment its reference count; otherwise, |
542 | * create one. @of is chained to the files list. | |
414985ae TH |
543 | * |
544 | * LOCKING: | |
545 | * Kernel thread context (may sleep). | |
546 | * | |
547 | * RETURNS: | |
548 | * 0 on success, -errno on failure. | |
549 | */ | |
c637b8ac TH |
550 | static int kernfs_get_open_node(struct kernfs_node *kn, |
551 | struct kernfs_open_file *of) | |
414985ae | 552 | { |
c525aadd | 553 | struct kernfs_open_node *on, *new_on = NULL; |
414985ae TH |
554 | |
555 | retry: | |
c525aadd TH |
556 | mutex_lock(&kernfs_open_file_mutex); |
557 | spin_lock_irq(&kernfs_open_node_lock); | |
414985ae | 558 | |
c525aadd TH |
559 | if (!kn->attr.open && new_on) { |
560 | kn->attr.open = new_on; | |
561 | new_on = NULL; | |
414985ae TH |
562 | } |
563 | ||
c525aadd TH |
564 | on = kn->attr.open; |
565 | if (on) { | |
566 | atomic_inc(&on->refcnt); | |
567 | list_add_tail(&of->list, &on->files); | |
414985ae TH |
568 | } |
569 | ||
c525aadd TH |
570 | spin_unlock_irq(&kernfs_open_node_lock); |
571 | mutex_unlock(&kernfs_open_file_mutex); | |
414985ae | 572 | |
c525aadd TH |
573 | if (on) { |
574 | kfree(new_on); | |
414985ae TH |
575 | return 0; |
576 | } | |
577 | ||
578 | /* not there, initialize a new one and retry */ | |
c525aadd TH |
579 | new_on = kmalloc(sizeof(*new_on), GFP_KERNEL); |
580 | if (!new_on) | |
414985ae TH |
581 | return -ENOMEM; |
582 | ||
c525aadd TH |
583 | atomic_set(&new_on->refcnt, 0); |
584 | atomic_set(&new_on->event, 1); | |
585 | init_waitqueue_head(&new_on->poll); | |
586 | INIT_LIST_HEAD(&new_on->files); | |
414985ae TH |
587 | goto retry; |
588 | } | |
589 | ||
590 | /** | |
c637b8ac | 591 | * kernfs_put_open_node - put kernfs_open_node |
324a56e1 | 592 | * @kn: target kernfs_nodet |
c525aadd | 593 | * @of: associated kernfs_open_file |
414985ae | 594 | * |
adc5e8b5 | 595 | * Put @kn->attr.open and unlink @of from the files list. If |
414985ae TH |
596 | * reference count reaches zero, disassociate and free it. |
597 | * | |
598 | * LOCKING: | |
599 | * None. | |
600 | */ | |
c637b8ac TH |
601 | static void kernfs_put_open_node(struct kernfs_node *kn, |
602 | struct kernfs_open_file *of) | |
414985ae | 603 | { |
c525aadd | 604 | struct kernfs_open_node *on = kn->attr.open; |
414985ae TH |
605 | unsigned long flags; |
606 | ||
c525aadd TH |
607 | mutex_lock(&kernfs_open_file_mutex); |
608 | spin_lock_irqsave(&kernfs_open_node_lock, flags); | |
414985ae TH |
609 | |
610 | if (of) | |
611 | list_del(&of->list); | |
612 | ||
c525aadd | 613 | if (atomic_dec_and_test(&on->refcnt)) |
adc5e8b5 | 614 | kn->attr.open = NULL; |
414985ae | 615 | else |
c525aadd | 616 | on = NULL; |
414985ae | 617 | |
c525aadd TH |
618 | spin_unlock_irqrestore(&kernfs_open_node_lock, flags); |
619 | mutex_unlock(&kernfs_open_file_mutex); | |
414985ae | 620 | |
c525aadd | 621 | kfree(on); |
414985ae TH |
622 | } |
623 | ||
c637b8ac | 624 | static int kernfs_fop_open(struct inode *inode, struct file *file) |
414985ae | 625 | { |
324a56e1 | 626 | struct kernfs_node *kn = file->f_path.dentry->d_fsdata; |
555724a8 | 627 | struct kernfs_root *root = kernfs_root(kn); |
414985ae | 628 | const struct kernfs_ops *ops; |
c525aadd | 629 | struct kernfs_open_file *of; |
414985ae TH |
630 | bool has_read, has_write, has_mmap; |
631 | int error = -EACCES; | |
632 | ||
c637b8ac | 633 | if (!kernfs_get_active(kn)) |
414985ae TH |
634 | return -ENODEV; |
635 | ||
324a56e1 | 636 | ops = kernfs_ops(kn); |
414985ae TH |
637 | |
638 | has_read = ops->seq_show || ops->read || ops->mmap; | |
639 | has_write = ops->write || ops->mmap; | |
640 | has_mmap = ops->mmap; | |
641 | ||
555724a8 TH |
642 | /* see the flag definition for details */ |
643 | if (root->flags & KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK) { | |
644 | if ((file->f_mode & FMODE_WRITE) && | |
645 | (!(inode->i_mode & S_IWUGO) || !has_write)) | |
646 | goto err_out; | |
414985ae | 647 | |
555724a8 TH |
648 | if ((file->f_mode & FMODE_READ) && |
649 | (!(inode->i_mode & S_IRUGO) || !has_read)) | |
650 | goto err_out; | |
651 | } | |
414985ae | 652 | |
c525aadd | 653 | /* allocate a kernfs_open_file for the file */ |
414985ae | 654 | error = -ENOMEM; |
c525aadd | 655 | of = kzalloc(sizeof(struct kernfs_open_file), GFP_KERNEL); |
414985ae TH |
656 | if (!of) |
657 | goto err_out; | |
658 | ||
659 | /* | |
660 | * The following is done to give a different lockdep key to | |
661 | * @of->mutex for files which implement mmap. This is a rather | |
662 | * crude way to avoid false positive lockdep warning around | |
663 | * mm->mmap_sem - mmap nests @of->mutex under mm->mmap_sem and | |
664 | * reading /sys/block/sda/trace/act_mask grabs sr_mutex, under | |
665 | * which mm->mmap_sem nests, while holding @of->mutex. As each | |
666 | * open file has a separate mutex, it's okay as long as those don't | |
667 | * happen on the same file. At this point, we can't easily give | |
668 | * each file a separate locking class. Let's differentiate on | |
669 | * whether the file has mmap or not for now. | |
9b2db6e1 TH |
670 | * |
671 | * Both paths of the branch look the same. They're supposed to | |
672 | * look that way and give @of->mutex different static lockdep keys. | |
414985ae TH |
673 | */ |
674 | if (has_mmap) | |
675 | mutex_init(&of->mutex); | |
676 | else | |
677 | mutex_init(&of->mutex); | |
678 | ||
324a56e1 | 679 | of->kn = kn; |
414985ae TH |
680 | of->file = file; |
681 | ||
b7ce40cf TH |
682 | /* |
683 | * Write path needs to atomic_write_len outside active reference. | |
684 | * Cache it in open_file. See kernfs_fop_write() for details. | |
685 | */ | |
686 | of->atomic_write_len = ops->atomic_write_len; | |
687 | ||
414985ae TH |
688 | /* |
689 | * Always instantiate seq_file even if read access doesn't use | |
690 | * seq_file or is not requested. This unifies private data access | |
691 | * and readable regular files are the vast majority anyway. | |
692 | */ | |
693 | if (ops->seq_show) | |
694 | error = seq_open(file, &kernfs_seq_ops); | |
695 | else | |
696 | error = seq_open(file, NULL); | |
697 | if (error) | |
698 | goto err_free; | |
699 | ||
700 | ((struct seq_file *)file->private_data)->private = of; | |
701 | ||
702 | /* seq_file clears PWRITE unconditionally, restore it if WRITE */ | |
703 | if (file->f_mode & FMODE_WRITE) | |
704 | file->f_mode |= FMODE_PWRITE; | |
705 | ||
c637b8ac TH |
706 | /* make sure we have open node struct */ |
707 | error = kernfs_get_open_node(kn, of); | |
414985ae TH |
708 | if (error) |
709 | goto err_close; | |
710 | ||
711 | /* open succeeded, put active references */ | |
c637b8ac | 712 | kernfs_put_active(kn); |
414985ae TH |
713 | return 0; |
714 | ||
715 | err_close: | |
716 | seq_release(inode, file); | |
717 | err_free: | |
718 | kfree(of); | |
719 | err_out: | |
c637b8ac | 720 | kernfs_put_active(kn); |
414985ae TH |
721 | return error; |
722 | } | |
723 | ||
c637b8ac | 724 | static int kernfs_fop_release(struct inode *inode, struct file *filp) |
414985ae | 725 | { |
324a56e1 | 726 | struct kernfs_node *kn = filp->f_path.dentry->d_fsdata; |
c525aadd | 727 | struct kernfs_open_file *of = kernfs_of(filp); |
414985ae | 728 | |
c637b8ac | 729 | kernfs_put_open_node(kn, of); |
414985ae TH |
730 | seq_release(inode, filp); |
731 | kfree(of); | |
732 | ||
733 | return 0; | |
734 | } | |
735 | ||
55f6e30d | 736 | void kernfs_unmap_bin_file(struct kernfs_node *kn) |
414985ae | 737 | { |
c525aadd TH |
738 | struct kernfs_open_node *on; |
739 | struct kernfs_open_file *of; | |
414985ae | 740 | |
55f6e30d GKH |
741 | if (!(kn->flags & KERNFS_HAS_MMAP)) |
742 | return; | |
743 | ||
c525aadd TH |
744 | spin_lock_irq(&kernfs_open_node_lock); |
745 | on = kn->attr.open; | |
746 | if (on) | |
747 | atomic_inc(&on->refcnt); | |
748 | spin_unlock_irq(&kernfs_open_node_lock); | |
749 | if (!on) | |
414985ae TH |
750 | return; |
751 | ||
c525aadd TH |
752 | mutex_lock(&kernfs_open_file_mutex); |
753 | list_for_each_entry(of, &on->files, list) { | |
414985ae TH |
754 | struct inode *inode = file_inode(of->file); |
755 | unmap_mapping_range(inode->i_mapping, 0, 0, 1); | |
756 | } | |
c525aadd | 757 | mutex_unlock(&kernfs_open_file_mutex); |
414985ae | 758 | |
c637b8ac | 759 | kernfs_put_open_node(kn, NULL); |
414985ae TH |
760 | } |
761 | ||
c637b8ac TH |
762 | /* |
763 | * Kernfs attribute files are pollable. The idea is that you read | |
414985ae TH |
764 | * the content and then you use 'poll' or 'select' to wait for |
765 | * the content to change. When the content changes (assuming the | |
766 | * manager for the kobject supports notification), poll will | |
767 | * return POLLERR|POLLPRI, and select will return the fd whether | |
768 | * it is waiting for read, write, or exceptions. | |
769 | * Once poll/select indicates that the value has changed, you | |
770 | * need to close and re-open the file, or seek to 0 and read again. | |
771 | * Reminder: this only works for attributes which actively support | |
772 | * it, and it is not possible to test an attribute from userspace | |
773 | * to see if it supports poll (Neither 'poll' nor 'select' return | |
774 | * an appropriate error code). When in doubt, set a suitable timeout value. | |
775 | */ | |
c637b8ac | 776 | static unsigned int kernfs_fop_poll(struct file *filp, poll_table *wait) |
414985ae | 777 | { |
c525aadd | 778 | struct kernfs_open_file *of = kernfs_of(filp); |
324a56e1 | 779 | struct kernfs_node *kn = filp->f_path.dentry->d_fsdata; |
c525aadd | 780 | struct kernfs_open_node *on = kn->attr.open; |
414985ae TH |
781 | |
782 | /* need parent for the kobj, grab both */ | |
c637b8ac | 783 | if (!kernfs_get_active(kn)) |
414985ae TH |
784 | goto trigger; |
785 | ||
c525aadd | 786 | poll_wait(filp, &on->poll, wait); |
414985ae | 787 | |
c637b8ac | 788 | kernfs_put_active(kn); |
414985ae | 789 | |
c525aadd | 790 | if (of->event != atomic_read(&on->event)) |
414985ae TH |
791 | goto trigger; |
792 | ||
793 | return DEFAULT_POLLMASK; | |
794 | ||
795 | trigger: | |
796 | return DEFAULT_POLLMASK|POLLERR|POLLPRI; | |
797 | } | |
798 | ||
ecca47ce | 799 | static void kernfs_notify_workfn(struct work_struct *work) |
414985ae | 800 | { |
ecca47ce | 801 | struct kernfs_node *kn; |
c525aadd | 802 | struct kernfs_open_node *on; |
d911d987 | 803 | struct kernfs_super_info *info; |
ecca47ce TH |
804 | repeat: |
805 | /* pop one off the notify_list */ | |
806 | spin_lock_irq(&kernfs_notify_lock); | |
807 | kn = kernfs_notify_list; | |
808 | if (kn == KERNFS_NOTIFY_EOL) { | |
809 | spin_unlock_irq(&kernfs_notify_lock); | |
d911d987 | 810 | return; |
ecca47ce TH |
811 | } |
812 | kernfs_notify_list = kn->attr.notify_next; | |
813 | kn->attr.notify_next = NULL; | |
814 | spin_unlock_irq(&kernfs_notify_lock); | |
d911d987 TH |
815 | |
816 | /* kick poll */ | |
ecca47ce | 817 | spin_lock_irq(&kernfs_open_node_lock); |
414985ae | 818 | |
d911d987 TH |
819 | on = kn->attr.open; |
820 | if (on) { | |
821 | atomic_inc(&on->event); | |
822 | wake_up_interruptible(&on->poll); | |
414985ae TH |
823 | } |
824 | ||
ecca47ce | 825 | spin_unlock_irq(&kernfs_open_node_lock); |
d911d987 TH |
826 | |
827 | /* kick fsnotify */ | |
828 | mutex_lock(&kernfs_mutex); | |
829 | ||
ecca47ce | 830 | list_for_each_entry(info, &kernfs_root(kn)->supers, node) { |
d911d987 TH |
831 | struct inode *inode; |
832 | struct dentry *dentry; | |
833 | ||
834 | inode = ilookup(info->sb, kn->ino); | |
835 | if (!inode) | |
836 | continue; | |
837 | ||
838 | dentry = d_find_any_alias(inode); | |
839 | if (dentry) { | |
840 | fsnotify_parent(NULL, dentry, FS_MODIFY); | |
841 | fsnotify(inode, FS_MODIFY, inode, FSNOTIFY_EVENT_INODE, | |
842 | NULL, 0); | |
843 | dput(dentry); | |
844 | } | |
845 | ||
846 | iput(inode); | |
847 | } | |
848 | ||
849 | mutex_unlock(&kernfs_mutex); | |
ecca47ce TH |
850 | kernfs_put(kn); |
851 | goto repeat; | |
852 | } | |
853 | ||
854 | /** | |
855 | * kernfs_notify - notify a kernfs file | |
856 | * @kn: file to notify | |
857 | * | |
858 | * Notify @kn such that poll(2) on @kn wakes up. Maybe be called from any | |
859 | * context. | |
860 | */ | |
861 | void kernfs_notify(struct kernfs_node *kn) | |
862 | { | |
863 | static DECLARE_WORK(kernfs_notify_work, kernfs_notify_workfn); | |
864 | unsigned long flags; | |
865 | ||
866 | if (WARN_ON(kernfs_type(kn) != KERNFS_FILE)) | |
867 | return; | |
868 | ||
869 | spin_lock_irqsave(&kernfs_notify_lock, flags); | |
870 | if (!kn->attr.notify_next) { | |
871 | kernfs_get(kn); | |
872 | kn->attr.notify_next = kernfs_notify_list; | |
873 | kernfs_notify_list = kn; | |
874 | schedule_work(&kernfs_notify_work); | |
875 | } | |
876 | spin_unlock_irqrestore(&kernfs_notify_lock, flags); | |
414985ae TH |
877 | } |
878 | EXPORT_SYMBOL_GPL(kernfs_notify); | |
879 | ||
a797bfc3 | 880 | const struct file_operations kernfs_file_fops = { |
c637b8ac TH |
881 | .read = kernfs_fop_read, |
882 | .write = kernfs_fop_write, | |
414985ae | 883 | .llseek = generic_file_llseek, |
c637b8ac TH |
884 | .mmap = kernfs_fop_mmap, |
885 | .open = kernfs_fop_open, | |
886 | .release = kernfs_fop_release, | |
887 | .poll = kernfs_fop_poll, | |
414985ae TH |
888 | }; |
889 | ||
890 | /** | |
2063d608 | 891 | * __kernfs_create_file - kernfs internal function to create a file |
414985ae TH |
892 | * @parent: directory to create the file in |
893 | * @name: name of the file | |
894 | * @mode: mode of the file | |
895 | * @size: size of the file | |
896 | * @ops: kernfs operations for the file | |
897 | * @priv: private data for the file | |
898 | * @ns: optional namespace tag of the file | |
2063d608 | 899 | * @static_name: don't copy file name |
414985ae TH |
900 | * @key: lockdep key for the file's active_ref, %NULL to disable lockdep |
901 | * | |
902 | * Returns the created node on success, ERR_PTR() value on error. | |
903 | */ | |
2063d608 TH |
904 | struct kernfs_node *__kernfs_create_file(struct kernfs_node *parent, |
905 | const char *name, | |
906 | umode_t mode, loff_t size, | |
907 | const struct kernfs_ops *ops, | |
908 | void *priv, const void *ns, | |
909 | bool name_is_static, | |
910 | struct lock_class_key *key) | |
414985ae | 911 | { |
324a56e1 | 912 | struct kernfs_node *kn; |
2063d608 | 913 | unsigned flags; |
414985ae TH |
914 | int rc; |
915 | ||
2063d608 TH |
916 | flags = KERNFS_FILE; |
917 | if (name_is_static) | |
918 | flags |= KERNFS_STATIC_NAME; | |
919 | ||
db4aad20 | 920 | kn = kernfs_new_node(parent, name, (mode & S_IALLUGO) | S_IFREG, flags); |
324a56e1 | 921 | if (!kn) |
414985ae TH |
922 | return ERR_PTR(-ENOMEM); |
923 | ||
adc5e8b5 TH |
924 | kn->attr.ops = ops; |
925 | kn->attr.size = size; | |
926 | kn->ns = ns; | |
324a56e1 | 927 | kn->priv = priv; |
414985ae TH |
928 | |
929 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
930 | if (key) { | |
324a56e1 | 931 | lockdep_init_map(&kn->dep_map, "s_active", key, 0); |
df23fc39 | 932 | kn->flags |= KERNFS_LOCKDEP; |
414985ae TH |
933 | } |
934 | #endif | |
935 | ||
936 | /* | |
adc5e8b5 | 937 | * kn->attr.ops is accesible only while holding active ref. We |
414985ae TH |
938 | * need to know whether some ops are implemented outside active |
939 | * ref. Cache their existence in flags. | |
940 | */ | |
941 | if (ops->seq_show) | |
df23fc39 | 942 | kn->flags |= KERNFS_HAS_SEQ_SHOW; |
414985ae | 943 | if (ops->mmap) |
df23fc39 | 944 | kn->flags |= KERNFS_HAS_MMAP; |
414985ae | 945 | |
988cd7af | 946 | rc = kernfs_add_one(kn); |
414985ae | 947 | if (rc) { |
324a56e1 | 948 | kernfs_put(kn); |
414985ae TH |
949 | return ERR_PTR(rc); |
950 | } | |
324a56e1 | 951 | return kn; |
414985ae | 952 | } |