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 TH |
16 | #include <linux/sched.h> |
17 | ||
18 | #include "kernfs-internal.h" | |
19 | ||
20 | /* | |
21 | * There's one sysfs_open_file for each open file and one sysfs_open_dirent | |
22 | * for each sysfs_dirent with one or more open files. | |
23 | * | |
24 | * sysfs_dirent->s_attr.open points to sysfs_open_dirent. s_attr.open is | |
25 | * protected by sysfs_open_dirent_lock. | |
26 | * | |
27 | * filp->private_data points to seq_file whose ->private points to | |
28 | * sysfs_open_file. sysfs_open_files are chained at | |
29 | * sysfs_open_dirent->files, which is protected by sysfs_open_file_mutex. | |
30 | */ | |
31 | static DEFINE_SPINLOCK(sysfs_open_dirent_lock); | |
32 | static DEFINE_MUTEX(sysfs_open_file_mutex); | |
33 | ||
34 | struct sysfs_open_dirent { | |
35 | atomic_t refcnt; | |
36 | atomic_t event; | |
37 | wait_queue_head_t poll; | |
38 | struct list_head files; /* goes through sysfs_open_file.list */ | |
39 | }; | |
40 | ||
41 | static struct sysfs_open_file *sysfs_of(struct file *file) | |
42 | { | |
43 | return ((struct seq_file *)file->private_data)->private; | |
44 | } | |
45 | ||
46 | /* | |
47 | * Determine the kernfs_ops for the given sysfs_dirent. This function must | |
48 | * be called while holding an active reference. | |
49 | */ | |
50 | static const struct kernfs_ops *kernfs_ops(struct sysfs_dirent *sd) | |
51 | { | |
52 | if (sd->s_flags & SYSFS_FLAG_LOCKDEP) | |
53 | lockdep_assert_held(sd); | |
54 | return sd->s_attr.ops; | |
55 | } | |
56 | ||
57 | static void *kernfs_seq_start(struct seq_file *sf, loff_t *ppos) | |
58 | { | |
59 | struct sysfs_open_file *of = sf->private; | |
60 | const struct kernfs_ops *ops; | |
61 | ||
62 | /* | |
63 | * @of->mutex nests outside active ref and is just to ensure that | |
64 | * the ops aren't called concurrently for the same open file. | |
65 | */ | |
66 | mutex_lock(&of->mutex); | |
67 | if (!sysfs_get_active(of->sd)) | |
68 | return ERR_PTR(-ENODEV); | |
69 | ||
70 | ops = kernfs_ops(of->sd); | |
71 | if (ops->seq_start) { | |
72 | return ops->seq_start(sf, ppos); | |
73 | } else { | |
74 | /* | |
75 | * The same behavior and code as single_open(). Returns | |
76 | * !NULL if pos is at the beginning; otherwise, NULL. | |
77 | */ | |
78 | return NULL + !*ppos; | |
79 | } | |
80 | } | |
81 | ||
82 | static void *kernfs_seq_next(struct seq_file *sf, void *v, loff_t *ppos) | |
83 | { | |
84 | struct sysfs_open_file *of = sf->private; | |
85 | const struct kernfs_ops *ops = kernfs_ops(of->sd); | |
86 | ||
87 | if (ops->seq_next) { | |
88 | return ops->seq_next(sf, v, ppos); | |
89 | } else { | |
90 | /* | |
91 | * The same behavior and code as single_open(), always | |
92 | * terminate after the initial read. | |
93 | */ | |
94 | ++*ppos; | |
95 | return NULL; | |
96 | } | |
97 | } | |
98 | ||
99 | static void kernfs_seq_stop(struct seq_file *sf, void *v) | |
100 | { | |
101 | struct sysfs_open_file *of = sf->private; | |
102 | const struct kernfs_ops *ops = kernfs_ops(of->sd); | |
103 | ||
104 | if (ops->seq_stop) | |
105 | ops->seq_stop(sf, v); | |
106 | ||
107 | sysfs_put_active(of->sd); | |
108 | mutex_unlock(&of->mutex); | |
109 | } | |
110 | ||
111 | static int kernfs_seq_show(struct seq_file *sf, void *v) | |
112 | { | |
113 | struct sysfs_open_file *of = sf->private; | |
114 | ||
115 | of->event = atomic_read(&of->sd->s_attr.open->event); | |
116 | ||
117 | return of->sd->s_attr.ops->seq_show(sf, v); | |
118 | } | |
119 | ||
120 | static const struct seq_operations kernfs_seq_ops = { | |
121 | .start = kernfs_seq_start, | |
122 | .next = kernfs_seq_next, | |
123 | .stop = kernfs_seq_stop, | |
124 | .show = kernfs_seq_show, | |
125 | }; | |
126 | ||
127 | /* | |
128 | * As reading a bin file can have side-effects, the exact offset and bytes | |
129 | * specified in read(2) call should be passed to the read callback making | |
130 | * it difficult to use seq_file. Implement simplistic custom buffering for | |
131 | * bin files. | |
132 | */ | |
133 | static ssize_t kernfs_file_direct_read(struct sysfs_open_file *of, | |
134 | char __user *user_buf, size_t count, | |
135 | loff_t *ppos) | |
136 | { | |
137 | ssize_t len = min_t(size_t, count, PAGE_SIZE); | |
138 | const struct kernfs_ops *ops; | |
139 | char *buf; | |
140 | ||
141 | buf = kmalloc(len, GFP_KERNEL); | |
142 | if (!buf) | |
143 | return -ENOMEM; | |
144 | ||
145 | /* | |
146 | * @of->mutex nests outside active ref and is just to ensure that | |
147 | * the ops aren't called concurrently for the same open file. | |
148 | */ | |
149 | mutex_lock(&of->mutex); | |
150 | if (!sysfs_get_active(of->sd)) { | |
151 | len = -ENODEV; | |
152 | mutex_unlock(&of->mutex); | |
153 | goto out_free; | |
154 | } | |
155 | ||
156 | ops = kernfs_ops(of->sd); | |
157 | if (ops->read) | |
158 | len = ops->read(of, buf, len, *ppos); | |
159 | else | |
160 | len = -EINVAL; | |
161 | ||
162 | sysfs_put_active(of->sd); | |
163 | mutex_unlock(&of->mutex); | |
164 | ||
165 | if (len < 0) | |
166 | goto out_free; | |
167 | ||
168 | if (copy_to_user(user_buf, buf, len)) { | |
169 | len = -EFAULT; | |
170 | goto out_free; | |
171 | } | |
172 | ||
173 | *ppos += len; | |
174 | ||
175 | out_free: | |
176 | kfree(buf); | |
177 | return len; | |
178 | } | |
179 | ||
180 | /** | |
181 | * kernfs_file_read - kernfs vfs read callback | |
182 | * @file: file pointer | |
183 | * @user_buf: data to write | |
184 | * @count: number of bytes | |
185 | * @ppos: starting offset | |
186 | */ | |
187 | static ssize_t kernfs_file_read(struct file *file, char __user *user_buf, | |
188 | size_t count, loff_t *ppos) | |
189 | { | |
190 | struct sysfs_open_file *of = sysfs_of(file); | |
191 | ||
192 | if (of->sd->s_flags & SYSFS_FLAG_HAS_SEQ_SHOW) | |
193 | return seq_read(file, user_buf, count, ppos); | |
194 | else | |
195 | return kernfs_file_direct_read(of, user_buf, count, ppos); | |
196 | } | |
197 | ||
198 | /** | |
199 | * kernfs_file_write - kernfs vfs write callback | |
200 | * @file: file pointer | |
201 | * @user_buf: data to write | |
202 | * @count: number of bytes | |
203 | * @ppos: starting offset | |
204 | * | |
205 | * Copy data in from userland and pass it to the matching kernfs write | |
206 | * operation. | |
207 | * | |
208 | * There is no easy way for us to know if userspace is only doing a partial | |
209 | * write, so we don't support them. We expect the entire buffer to come on | |
210 | * the first write. Hint: if you're writing a value, first read the file, | |
211 | * modify only the the value you're changing, then write entire buffer | |
212 | * back. | |
213 | */ | |
214 | static ssize_t kernfs_file_write(struct file *file, const char __user *user_buf, | |
215 | size_t count, loff_t *ppos) | |
216 | { | |
217 | struct sysfs_open_file *of = sysfs_of(file); | |
218 | ssize_t len = min_t(size_t, count, PAGE_SIZE); | |
219 | const struct kernfs_ops *ops; | |
220 | char *buf; | |
221 | ||
222 | buf = kmalloc(len + 1, GFP_KERNEL); | |
223 | if (!buf) | |
224 | return -ENOMEM; | |
225 | ||
226 | if (copy_from_user(buf, user_buf, len)) { | |
227 | len = -EFAULT; | |
228 | goto out_free; | |
229 | } | |
230 | buf[len] = '\0'; /* guarantee string termination */ | |
231 | ||
232 | /* | |
233 | * @of->mutex nests outside active ref and is just to ensure that | |
234 | * the ops aren't called concurrently for the same open file. | |
235 | */ | |
236 | mutex_lock(&of->mutex); | |
237 | if (!sysfs_get_active(of->sd)) { | |
238 | mutex_unlock(&of->mutex); | |
239 | len = -ENODEV; | |
240 | goto out_free; | |
241 | } | |
242 | ||
243 | ops = kernfs_ops(of->sd); | |
244 | if (ops->write) | |
245 | len = ops->write(of, buf, len, *ppos); | |
246 | else | |
247 | len = -EINVAL; | |
248 | ||
249 | sysfs_put_active(of->sd); | |
250 | mutex_unlock(&of->mutex); | |
251 | ||
252 | if (len > 0) | |
253 | *ppos += len; | |
254 | out_free: | |
255 | kfree(buf); | |
256 | return len; | |
257 | } | |
258 | ||
259 | static void kernfs_vma_open(struct vm_area_struct *vma) | |
260 | { | |
261 | struct file *file = vma->vm_file; | |
262 | struct sysfs_open_file *of = sysfs_of(file); | |
263 | ||
264 | if (!of->vm_ops) | |
265 | return; | |
266 | ||
267 | if (!sysfs_get_active(of->sd)) | |
268 | return; | |
269 | ||
270 | if (of->vm_ops->open) | |
271 | of->vm_ops->open(vma); | |
272 | ||
273 | sysfs_put_active(of->sd); | |
274 | } | |
275 | ||
276 | static int kernfs_vma_fault(struct vm_area_struct *vma, struct vm_fault *vmf) | |
277 | { | |
278 | struct file *file = vma->vm_file; | |
279 | struct sysfs_open_file *of = sysfs_of(file); | |
280 | int ret; | |
281 | ||
282 | if (!of->vm_ops) | |
283 | return VM_FAULT_SIGBUS; | |
284 | ||
285 | if (!sysfs_get_active(of->sd)) | |
286 | return VM_FAULT_SIGBUS; | |
287 | ||
288 | ret = VM_FAULT_SIGBUS; | |
289 | if (of->vm_ops->fault) | |
290 | ret = of->vm_ops->fault(vma, vmf); | |
291 | ||
292 | sysfs_put_active(of->sd); | |
293 | return ret; | |
294 | } | |
295 | ||
296 | static int kernfs_vma_page_mkwrite(struct vm_area_struct *vma, | |
297 | struct vm_fault *vmf) | |
298 | { | |
299 | struct file *file = vma->vm_file; | |
300 | struct sysfs_open_file *of = sysfs_of(file); | |
301 | int ret; | |
302 | ||
303 | if (!of->vm_ops) | |
304 | return VM_FAULT_SIGBUS; | |
305 | ||
306 | if (!sysfs_get_active(of->sd)) | |
307 | return VM_FAULT_SIGBUS; | |
308 | ||
309 | ret = 0; | |
310 | if (of->vm_ops->page_mkwrite) | |
311 | ret = of->vm_ops->page_mkwrite(vma, vmf); | |
312 | else | |
313 | file_update_time(file); | |
314 | ||
315 | sysfs_put_active(of->sd); | |
316 | return ret; | |
317 | } | |
318 | ||
319 | static int kernfs_vma_access(struct vm_area_struct *vma, unsigned long addr, | |
320 | void *buf, int len, int write) | |
321 | { | |
322 | struct file *file = vma->vm_file; | |
323 | struct sysfs_open_file *of = sysfs_of(file); | |
324 | int ret; | |
325 | ||
326 | if (!of->vm_ops) | |
327 | return -EINVAL; | |
328 | ||
329 | if (!sysfs_get_active(of->sd)) | |
330 | return -EINVAL; | |
331 | ||
332 | ret = -EINVAL; | |
333 | if (of->vm_ops->access) | |
334 | ret = of->vm_ops->access(vma, addr, buf, len, write); | |
335 | ||
336 | sysfs_put_active(of->sd); | |
337 | return ret; | |
338 | } | |
339 | ||
340 | #ifdef CONFIG_NUMA | |
341 | static int kernfs_vma_set_policy(struct vm_area_struct *vma, | |
342 | struct mempolicy *new) | |
343 | { | |
344 | struct file *file = vma->vm_file; | |
345 | struct sysfs_open_file *of = sysfs_of(file); | |
346 | int ret; | |
347 | ||
348 | if (!of->vm_ops) | |
349 | return 0; | |
350 | ||
351 | if (!sysfs_get_active(of->sd)) | |
352 | return -EINVAL; | |
353 | ||
354 | ret = 0; | |
355 | if (of->vm_ops->set_policy) | |
356 | ret = of->vm_ops->set_policy(vma, new); | |
357 | ||
358 | sysfs_put_active(of->sd); | |
359 | return ret; | |
360 | } | |
361 | ||
362 | static struct mempolicy *kernfs_vma_get_policy(struct vm_area_struct *vma, | |
363 | unsigned long addr) | |
364 | { | |
365 | struct file *file = vma->vm_file; | |
366 | struct sysfs_open_file *of = sysfs_of(file); | |
367 | struct mempolicy *pol; | |
368 | ||
369 | if (!of->vm_ops) | |
370 | return vma->vm_policy; | |
371 | ||
372 | if (!sysfs_get_active(of->sd)) | |
373 | return vma->vm_policy; | |
374 | ||
375 | pol = vma->vm_policy; | |
376 | if (of->vm_ops->get_policy) | |
377 | pol = of->vm_ops->get_policy(vma, addr); | |
378 | ||
379 | sysfs_put_active(of->sd); | |
380 | return pol; | |
381 | } | |
382 | ||
383 | static int kernfs_vma_migrate(struct vm_area_struct *vma, | |
384 | const nodemask_t *from, const nodemask_t *to, | |
385 | unsigned long flags) | |
386 | { | |
387 | struct file *file = vma->vm_file; | |
388 | struct sysfs_open_file *of = sysfs_of(file); | |
389 | int ret; | |
390 | ||
391 | if (!of->vm_ops) | |
392 | return 0; | |
393 | ||
394 | if (!sysfs_get_active(of->sd)) | |
395 | return 0; | |
396 | ||
397 | ret = 0; | |
398 | if (of->vm_ops->migrate) | |
399 | ret = of->vm_ops->migrate(vma, from, to, flags); | |
400 | ||
401 | sysfs_put_active(of->sd); | |
402 | return ret; | |
403 | } | |
404 | #endif | |
405 | ||
406 | static const struct vm_operations_struct kernfs_vm_ops = { | |
407 | .open = kernfs_vma_open, | |
408 | .fault = kernfs_vma_fault, | |
409 | .page_mkwrite = kernfs_vma_page_mkwrite, | |
410 | .access = kernfs_vma_access, | |
411 | #ifdef CONFIG_NUMA | |
412 | .set_policy = kernfs_vma_set_policy, | |
413 | .get_policy = kernfs_vma_get_policy, | |
414 | .migrate = kernfs_vma_migrate, | |
415 | #endif | |
416 | }; | |
417 | ||
418 | static int kernfs_file_mmap(struct file *file, struct vm_area_struct *vma) | |
419 | { | |
420 | struct sysfs_open_file *of = sysfs_of(file); | |
421 | const struct kernfs_ops *ops; | |
422 | int rc; | |
423 | ||
9b2db6e1 TH |
424 | /* |
425 | * mmap path and of->mutex are prone to triggering spurious lockdep | |
426 | * warnings and we don't want to add spurious locking dependency | |
427 | * between the two. Check whether mmap is actually implemented | |
428 | * without grabbing @of->mutex by testing HAS_MMAP flag. See the | |
429 | * comment in kernfs_file_open() for more details. | |
430 | */ | |
431 | if (!(of->sd->s_flags & SYSFS_FLAG_HAS_MMAP)) | |
432 | return -ENODEV; | |
433 | ||
414985ae TH |
434 | mutex_lock(&of->mutex); |
435 | ||
436 | rc = -ENODEV; | |
437 | if (!sysfs_get_active(of->sd)) | |
438 | goto out_unlock; | |
439 | ||
440 | ops = kernfs_ops(of->sd); | |
9b2db6e1 | 441 | rc = ops->mmap(of, vma); |
414985ae TH |
442 | |
443 | /* | |
444 | * PowerPC's pci_mmap of legacy_mem uses shmem_zero_setup() | |
445 | * to satisfy versions of X which crash if the mmap fails: that | |
446 | * substitutes a new vm_file, and we don't then want bin_vm_ops. | |
447 | */ | |
448 | if (vma->vm_file != file) | |
449 | goto out_put; | |
450 | ||
451 | rc = -EINVAL; | |
452 | if (of->mmapped && of->vm_ops != vma->vm_ops) | |
453 | goto out_put; | |
454 | ||
455 | /* | |
456 | * It is not possible to successfully wrap close. | |
457 | * So error if someone is trying to use close. | |
458 | */ | |
459 | rc = -EINVAL; | |
460 | if (vma->vm_ops && vma->vm_ops->close) | |
461 | goto out_put; | |
462 | ||
463 | rc = 0; | |
464 | of->mmapped = 1; | |
465 | of->vm_ops = vma->vm_ops; | |
466 | vma->vm_ops = &kernfs_vm_ops; | |
467 | out_put: | |
468 | sysfs_put_active(of->sd); | |
469 | out_unlock: | |
470 | mutex_unlock(&of->mutex); | |
471 | ||
472 | return rc; | |
473 | } | |
474 | ||
475 | /** | |
476 | * sysfs_get_open_dirent - get or create sysfs_open_dirent | |
477 | * @sd: target sysfs_dirent | |
478 | * @of: sysfs_open_file for this instance of open | |
479 | * | |
480 | * If @sd->s_attr.open exists, increment its reference count; | |
481 | * otherwise, create one. @of is chained to the files list. | |
482 | * | |
483 | * LOCKING: | |
484 | * Kernel thread context (may sleep). | |
485 | * | |
486 | * RETURNS: | |
487 | * 0 on success, -errno on failure. | |
488 | */ | |
489 | static int sysfs_get_open_dirent(struct sysfs_dirent *sd, | |
490 | struct sysfs_open_file *of) | |
491 | { | |
492 | struct sysfs_open_dirent *od, *new_od = NULL; | |
493 | ||
494 | retry: | |
495 | mutex_lock(&sysfs_open_file_mutex); | |
496 | spin_lock_irq(&sysfs_open_dirent_lock); | |
497 | ||
498 | if (!sd->s_attr.open && new_od) { | |
499 | sd->s_attr.open = new_od; | |
500 | new_od = NULL; | |
501 | } | |
502 | ||
503 | od = sd->s_attr.open; | |
504 | if (od) { | |
505 | atomic_inc(&od->refcnt); | |
506 | list_add_tail(&of->list, &od->files); | |
507 | } | |
508 | ||
509 | spin_unlock_irq(&sysfs_open_dirent_lock); | |
510 | mutex_unlock(&sysfs_open_file_mutex); | |
511 | ||
512 | if (od) { | |
513 | kfree(new_od); | |
514 | return 0; | |
515 | } | |
516 | ||
517 | /* not there, initialize a new one and retry */ | |
518 | new_od = kmalloc(sizeof(*new_od), GFP_KERNEL); | |
519 | if (!new_od) | |
520 | return -ENOMEM; | |
521 | ||
522 | atomic_set(&new_od->refcnt, 0); | |
523 | atomic_set(&new_od->event, 1); | |
524 | init_waitqueue_head(&new_od->poll); | |
525 | INIT_LIST_HEAD(&new_od->files); | |
526 | goto retry; | |
527 | } | |
528 | ||
529 | /** | |
530 | * sysfs_put_open_dirent - put sysfs_open_dirent | |
531 | * @sd: target sysfs_dirent | |
532 | * @of: associated sysfs_open_file | |
533 | * | |
534 | * Put @sd->s_attr.open and unlink @of from the files list. If | |
535 | * reference count reaches zero, disassociate and free it. | |
536 | * | |
537 | * LOCKING: | |
538 | * None. | |
539 | */ | |
540 | static void sysfs_put_open_dirent(struct sysfs_dirent *sd, | |
541 | struct sysfs_open_file *of) | |
542 | { | |
543 | struct sysfs_open_dirent *od = sd->s_attr.open; | |
544 | unsigned long flags; | |
545 | ||
546 | mutex_lock(&sysfs_open_file_mutex); | |
547 | spin_lock_irqsave(&sysfs_open_dirent_lock, flags); | |
548 | ||
549 | if (of) | |
550 | list_del(&of->list); | |
551 | ||
552 | if (atomic_dec_and_test(&od->refcnt)) | |
553 | sd->s_attr.open = NULL; | |
554 | else | |
555 | od = NULL; | |
556 | ||
557 | spin_unlock_irqrestore(&sysfs_open_dirent_lock, flags); | |
558 | mutex_unlock(&sysfs_open_file_mutex); | |
559 | ||
560 | kfree(od); | |
561 | } | |
562 | ||
563 | static int kernfs_file_open(struct inode *inode, struct file *file) | |
564 | { | |
565 | struct sysfs_dirent *attr_sd = file->f_path.dentry->d_fsdata; | |
566 | const struct kernfs_ops *ops; | |
567 | struct sysfs_open_file *of; | |
568 | bool has_read, has_write, has_mmap; | |
569 | int error = -EACCES; | |
570 | ||
571 | if (!sysfs_get_active(attr_sd)) | |
572 | return -ENODEV; | |
573 | ||
574 | ops = kernfs_ops(attr_sd); | |
575 | ||
576 | has_read = ops->seq_show || ops->read || ops->mmap; | |
577 | has_write = ops->write || ops->mmap; | |
578 | has_mmap = ops->mmap; | |
579 | ||
580 | /* check perms and supported operations */ | |
581 | if ((file->f_mode & FMODE_WRITE) && | |
582 | (!(inode->i_mode & S_IWUGO) || !has_write)) | |
583 | goto err_out; | |
584 | ||
585 | if ((file->f_mode & FMODE_READ) && | |
586 | (!(inode->i_mode & S_IRUGO) || !has_read)) | |
587 | goto err_out; | |
588 | ||
589 | /* allocate a sysfs_open_file for the file */ | |
590 | error = -ENOMEM; | |
591 | of = kzalloc(sizeof(struct sysfs_open_file), GFP_KERNEL); | |
592 | if (!of) | |
593 | goto err_out; | |
594 | ||
595 | /* | |
596 | * The following is done to give a different lockdep key to | |
597 | * @of->mutex for files which implement mmap. This is a rather | |
598 | * crude way to avoid false positive lockdep warning around | |
599 | * mm->mmap_sem - mmap nests @of->mutex under mm->mmap_sem and | |
600 | * reading /sys/block/sda/trace/act_mask grabs sr_mutex, under | |
601 | * which mm->mmap_sem nests, while holding @of->mutex. As each | |
602 | * open file has a separate mutex, it's okay as long as those don't | |
603 | * happen on the same file. At this point, we can't easily give | |
604 | * each file a separate locking class. Let's differentiate on | |
605 | * whether the file has mmap or not for now. | |
9b2db6e1 TH |
606 | * |
607 | * Both paths of the branch look the same. They're supposed to | |
608 | * look that way and give @of->mutex different static lockdep keys. | |
414985ae TH |
609 | */ |
610 | if (has_mmap) | |
611 | mutex_init(&of->mutex); | |
612 | else | |
613 | mutex_init(&of->mutex); | |
614 | ||
615 | of->sd = attr_sd; | |
616 | of->file = file; | |
617 | ||
618 | /* | |
619 | * Always instantiate seq_file even if read access doesn't use | |
620 | * seq_file or is not requested. This unifies private data access | |
621 | * and readable regular files are the vast majority anyway. | |
622 | */ | |
623 | if (ops->seq_show) | |
624 | error = seq_open(file, &kernfs_seq_ops); | |
625 | else | |
626 | error = seq_open(file, NULL); | |
627 | if (error) | |
628 | goto err_free; | |
629 | ||
630 | ((struct seq_file *)file->private_data)->private = of; | |
631 | ||
632 | /* seq_file clears PWRITE unconditionally, restore it if WRITE */ | |
633 | if (file->f_mode & FMODE_WRITE) | |
634 | file->f_mode |= FMODE_PWRITE; | |
635 | ||
636 | /* make sure we have open dirent struct */ | |
637 | error = sysfs_get_open_dirent(attr_sd, of); | |
638 | if (error) | |
639 | goto err_close; | |
640 | ||
641 | /* open succeeded, put active references */ | |
642 | sysfs_put_active(attr_sd); | |
643 | return 0; | |
644 | ||
645 | err_close: | |
646 | seq_release(inode, file); | |
647 | err_free: | |
648 | kfree(of); | |
649 | err_out: | |
650 | sysfs_put_active(attr_sd); | |
651 | return error; | |
652 | } | |
653 | ||
654 | static int kernfs_file_release(struct inode *inode, struct file *filp) | |
655 | { | |
656 | struct sysfs_dirent *sd = filp->f_path.dentry->d_fsdata; | |
657 | struct sysfs_open_file *of = sysfs_of(filp); | |
658 | ||
659 | sysfs_put_open_dirent(sd, of); | |
660 | seq_release(inode, filp); | |
661 | kfree(of); | |
662 | ||
663 | return 0; | |
664 | } | |
665 | ||
666 | void sysfs_unmap_bin_file(struct sysfs_dirent *sd) | |
667 | { | |
668 | struct sysfs_open_dirent *od; | |
669 | struct sysfs_open_file *of; | |
670 | ||
671 | if (!(sd->s_flags & SYSFS_FLAG_HAS_MMAP)) | |
672 | return; | |
673 | ||
674 | spin_lock_irq(&sysfs_open_dirent_lock); | |
675 | od = sd->s_attr.open; | |
676 | if (od) | |
677 | atomic_inc(&od->refcnt); | |
678 | spin_unlock_irq(&sysfs_open_dirent_lock); | |
679 | if (!od) | |
680 | return; | |
681 | ||
682 | mutex_lock(&sysfs_open_file_mutex); | |
683 | list_for_each_entry(of, &od->files, list) { | |
684 | struct inode *inode = file_inode(of->file); | |
685 | unmap_mapping_range(inode->i_mapping, 0, 0, 1); | |
686 | } | |
687 | mutex_unlock(&sysfs_open_file_mutex); | |
688 | ||
689 | sysfs_put_open_dirent(sd, NULL); | |
690 | } | |
691 | ||
692 | /* Sysfs attribute files are pollable. The idea is that you read | |
693 | * the content and then you use 'poll' or 'select' to wait for | |
694 | * the content to change. When the content changes (assuming the | |
695 | * manager for the kobject supports notification), poll will | |
696 | * return POLLERR|POLLPRI, and select will return the fd whether | |
697 | * it is waiting for read, write, or exceptions. | |
698 | * Once poll/select indicates that the value has changed, you | |
699 | * need to close and re-open the file, or seek to 0 and read again. | |
700 | * Reminder: this only works for attributes which actively support | |
701 | * it, and it is not possible to test an attribute from userspace | |
702 | * to see if it supports poll (Neither 'poll' nor 'select' return | |
703 | * an appropriate error code). When in doubt, set a suitable timeout value. | |
704 | */ | |
705 | static unsigned int kernfs_file_poll(struct file *filp, poll_table *wait) | |
706 | { | |
707 | struct sysfs_open_file *of = sysfs_of(filp); | |
708 | struct sysfs_dirent *attr_sd = filp->f_path.dentry->d_fsdata; | |
709 | struct sysfs_open_dirent *od = attr_sd->s_attr.open; | |
710 | ||
711 | /* need parent for the kobj, grab both */ | |
712 | if (!sysfs_get_active(attr_sd)) | |
713 | goto trigger; | |
714 | ||
715 | poll_wait(filp, &od->poll, wait); | |
716 | ||
717 | sysfs_put_active(attr_sd); | |
718 | ||
719 | if (of->event != atomic_read(&od->event)) | |
720 | goto trigger; | |
721 | ||
722 | return DEFAULT_POLLMASK; | |
723 | ||
724 | trigger: | |
725 | return DEFAULT_POLLMASK|POLLERR|POLLPRI; | |
726 | } | |
727 | ||
728 | /** | |
729 | * kernfs_notify - notify a kernfs file | |
730 | * @sd: file to notify | |
731 | * | |
732 | * Notify @sd such that poll(2) on @sd wakes up. | |
733 | */ | |
734 | void kernfs_notify(struct sysfs_dirent *sd) | |
735 | { | |
736 | struct sysfs_open_dirent *od; | |
737 | unsigned long flags; | |
738 | ||
739 | spin_lock_irqsave(&sysfs_open_dirent_lock, flags); | |
740 | ||
741 | if (!WARN_ON(sysfs_type(sd) != SYSFS_KOBJ_ATTR)) { | |
742 | od = sd->s_attr.open; | |
743 | if (od) { | |
744 | atomic_inc(&od->event); | |
745 | wake_up_interruptible(&od->poll); | |
746 | } | |
747 | } | |
748 | ||
749 | spin_unlock_irqrestore(&sysfs_open_dirent_lock, flags); | |
750 | } | |
751 | EXPORT_SYMBOL_GPL(kernfs_notify); | |
752 | ||
753 | const struct file_operations kernfs_file_operations = { | |
754 | .read = kernfs_file_read, | |
755 | .write = kernfs_file_write, | |
756 | .llseek = generic_file_llseek, | |
757 | .mmap = kernfs_file_mmap, | |
758 | .open = kernfs_file_open, | |
759 | .release = kernfs_file_release, | |
760 | .poll = kernfs_file_poll, | |
761 | }; | |
762 | ||
763 | /** | |
764 | * kernfs_create_file_ns_key - create a file | |
765 | * @parent: directory to create the file in | |
766 | * @name: name of the file | |
767 | * @mode: mode of the file | |
768 | * @size: size of the file | |
769 | * @ops: kernfs operations for the file | |
770 | * @priv: private data for the file | |
771 | * @ns: optional namespace tag of the file | |
772 | * @key: lockdep key for the file's active_ref, %NULL to disable lockdep | |
773 | * | |
774 | * Returns the created node on success, ERR_PTR() value on error. | |
775 | */ | |
776 | struct sysfs_dirent *kernfs_create_file_ns_key(struct sysfs_dirent *parent, | |
777 | const char *name, | |
778 | umode_t mode, loff_t size, | |
779 | const struct kernfs_ops *ops, | |
780 | void *priv, const void *ns, | |
781 | struct lock_class_key *key) | |
782 | { | |
783 | struct sysfs_addrm_cxt acxt; | |
784 | struct sysfs_dirent *sd; | |
785 | int rc; | |
786 | ||
bc755553 TH |
787 | sd = sysfs_new_dirent(kernfs_root(parent), name, |
788 | (mode & S_IALLUGO) | S_IFREG, SYSFS_KOBJ_ATTR); | |
414985ae TH |
789 | if (!sd) |
790 | return ERR_PTR(-ENOMEM); | |
791 | ||
792 | sd->s_attr.ops = ops; | |
793 | sd->s_attr.size = size; | |
794 | sd->s_ns = ns; | |
795 | sd->priv = priv; | |
796 | ||
797 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
798 | if (key) { | |
799 | lockdep_init_map(&sd->dep_map, "s_active", key, 0); | |
800 | sd->s_flags |= SYSFS_FLAG_LOCKDEP; | |
801 | } | |
802 | #endif | |
803 | ||
804 | /* | |
805 | * sd->s_attr.ops is accesible only while holding active ref. We | |
806 | * need to know whether some ops are implemented outside active | |
807 | * ref. Cache their existence in flags. | |
808 | */ | |
809 | if (ops->seq_show) | |
810 | sd->s_flags |= SYSFS_FLAG_HAS_SEQ_SHOW; | |
811 | if (ops->mmap) | |
812 | sd->s_flags |= SYSFS_FLAG_HAS_MMAP; | |
813 | ||
814 | sysfs_addrm_start(&acxt); | |
815 | rc = sysfs_add_one(&acxt, sd, parent); | |
816 | sysfs_addrm_finish(&acxt); | |
817 | ||
818 | if (rc) { | |
819 | kernfs_put(sd); | |
820 | return ERR_PTR(rc); | |
821 | } | |
822 | return sd; | |
823 | } |