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 | /* | |
c525aadd | 21 | * There's one kernfs_open_file for each open file and one kernfs_open_node |
324a56e1 | 22 | * for each kernfs_node with one or more open files. |
414985ae | 23 | * |
c525aadd TH |
24 | * kernfs_node->attr.open points to kernfs_open_node. attr.open is |
25 | * protected by kernfs_open_node_lock. | |
414985ae TH |
26 | * |
27 | * filp->private_data points to seq_file whose ->private points to | |
c525aadd TH |
28 | * kernfs_open_file. kernfs_open_files are chained at |
29 | * kernfs_open_node->files, which is protected by kernfs_open_file_mutex. | |
414985ae | 30 | */ |
c525aadd TH |
31 | static DEFINE_SPINLOCK(kernfs_open_node_lock); |
32 | static DEFINE_MUTEX(kernfs_open_file_mutex); | |
414985ae | 33 | |
c525aadd | 34 | struct kernfs_open_node { |
414985ae TH |
35 | atomic_t refcnt; |
36 | atomic_t event; | |
37 | wait_queue_head_t poll; | |
c525aadd | 38 | struct list_head files; /* goes through kernfs_open_file.list */ |
414985ae TH |
39 | }; |
40 | ||
c525aadd | 41 | static struct kernfs_open_file *kernfs_of(struct file *file) |
414985ae TH |
42 | { |
43 | return ((struct seq_file *)file->private_data)->private; | |
44 | } | |
45 | ||
46 | /* | |
324a56e1 | 47 | * Determine the kernfs_ops for the given kernfs_node. This function must |
414985ae TH |
48 | * be called while holding an active reference. |
49 | */ | |
324a56e1 | 50 | static const struct kernfs_ops *kernfs_ops(struct kernfs_node *kn) |
414985ae | 51 | { |
adc5e8b5 | 52 | if (kn->flags & SYSFS_FLAG_LOCKDEP) |
324a56e1 | 53 | lockdep_assert_held(kn); |
adc5e8b5 | 54 | return kn->attr.ops; |
414985ae TH |
55 | } |
56 | ||
57 | static void *kernfs_seq_start(struct seq_file *sf, loff_t *ppos) | |
58 | { | |
c525aadd | 59 | struct kernfs_open_file *of = sf->private; |
414985ae TH |
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); | |
324a56e1 | 67 | if (!sysfs_get_active(of->kn)) |
414985ae TH |
68 | return ERR_PTR(-ENODEV); |
69 | ||
324a56e1 | 70 | ops = kernfs_ops(of->kn); |
414985ae TH |
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 | { | |
c525aadd | 84 | struct kernfs_open_file *of = sf->private; |
324a56e1 | 85 | const struct kernfs_ops *ops = kernfs_ops(of->kn); |
414985ae TH |
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 | { | |
c525aadd | 101 | struct kernfs_open_file *of = sf->private; |
324a56e1 | 102 | const struct kernfs_ops *ops = kernfs_ops(of->kn); |
414985ae TH |
103 | |
104 | if (ops->seq_stop) | |
105 | ops->seq_stop(sf, v); | |
106 | ||
324a56e1 | 107 | sysfs_put_active(of->kn); |
414985ae TH |
108 | mutex_unlock(&of->mutex); |
109 | } | |
110 | ||
111 | static int kernfs_seq_show(struct seq_file *sf, void *v) | |
112 | { | |
c525aadd | 113 | struct kernfs_open_file *of = sf->private; |
414985ae | 114 | |
adc5e8b5 | 115 | of->event = atomic_read(&of->kn->attr.open->event); |
414985ae | 116 | |
adc5e8b5 | 117 | return of->kn->attr.ops->seq_show(sf, v); |
414985ae TH |
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 | */ | |
c525aadd | 133 | static ssize_t kernfs_file_direct_read(struct kernfs_open_file *of, |
414985ae TH |
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); | |
324a56e1 | 150 | if (!sysfs_get_active(of->kn)) { |
414985ae TH |
151 | len = -ENODEV; |
152 | mutex_unlock(&of->mutex); | |
153 | goto out_free; | |
154 | } | |
155 | ||
324a56e1 | 156 | ops = kernfs_ops(of->kn); |
414985ae TH |
157 | if (ops->read) |
158 | len = ops->read(of, buf, len, *ppos); | |
159 | else | |
160 | len = -EINVAL; | |
161 | ||
324a56e1 | 162 | sysfs_put_active(of->kn); |
414985ae TH |
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 | { | |
c525aadd | 190 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae | 191 | |
adc5e8b5 | 192 | if (of->kn->flags & SYSFS_FLAG_HAS_SEQ_SHOW) |
414985ae TH |
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 | { | |
c525aadd | 217 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
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); | |
324a56e1 | 237 | if (!sysfs_get_active(of->kn)) { |
414985ae TH |
238 | mutex_unlock(&of->mutex); |
239 | len = -ENODEV; | |
240 | goto out_free; | |
241 | } | |
242 | ||
324a56e1 | 243 | ops = kernfs_ops(of->kn); |
414985ae TH |
244 | if (ops->write) |
245 | len = ops->write(of, buf, len, *ppos); | |
246 | else | |
247 | len = -EINVAL; | |
248 | ||
324a56e1 | 249 | sysfs_put_active(of->kn); |
414985ae TH |
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; | |
c525aadd | 262 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
263 | |
264 | if (!of->vm_ops) | |
265 | return; | |
266 | ||
324a56e1 | 267 | if (!sysfs_get_active(of->kn)) |
414985ae TH |
268 | return; |
269 | ||
270 | if (of->vm_ops->open) | |
271 | of->vm_ops->open(vma); | |
272 | ||
324a56e1 | 273 | sysfs_put_active(of->kn); |
414985ae TH |
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; | |
c525aadd | 279 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
280 | int ret; |
281 | ||
282 | if (!of->vm_ops) | |
283 | return VM_FAULT_SIGBUS; | |
284 | ||
324a56e1 | 285 | if (!sysfs_get_active(of->kn)) |
414985ae TH |
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 | ||
324a56e1 | 292 | sysfs_put_active(of->kn); |
414985ae TH |
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; | |
c525aadd | 300 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
301 | int ret; |
302 | ||
303 | if (!of->vm_ops) | |
304 | return VM_FAULT_SIGBUS; | |
305 | ||
324a56e1 | 306 | if (!sysfs_get_active(of->kn)) |
414985ae TH |
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 | ||
324a56e1 | 315 | sysfs_put_active(of->kn); |
414985ae TH |
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; | |
c525aadd | 323 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
324 | int ret; |
325 | ||
326 | if (!of->vm_ops) | |
327 | return -EINVAL; | |
328 | ||
324a56e1 | 329 | if (!sysfs_get_active(of->kn)) |
414985ae TH |
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 | ||
324a56e1 | 336 | sysfs_put_active(of->kn); |
414985ae TH |
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; | |
c525aadd | 345 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
346 | int ret; |
347 | ||
348 | if (!of->vm_ops) | |
349 | return 0; | |
350 | ||
324a56e1 | 351 | if (!sysfs_get_active(of->kn)) |
414985ae TH |
352 | return -EINVAL; |
353 | ||
354 | ret = 0; | |
355 | if (of->vm_ops->set_policy) | |
356 | ret = of->vm_ops->set_policy(vma, new); | |
357 | ||
324a56e1 | 358 | sysfs_put_active(of->kn); |
414985ae TH |
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; | |
c525aadd | 366 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
367 | struct mempolicy *pol; |
368 | ||
369 | if (!of->vm_ops) | |
370 | return vma->vm_policy; | |
371 | ||
324a56e1 | 372 | if (!sysfs_get_active(of->kn)) |
414985ae TH |
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 | ||
324a56e1 | 379 | sysfs_put_active(of->kn); |
414985ae TH |
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; | |
c525aadd | 388 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
389 | int ret; |
390 | ||
391 | if (!of->vm_ops) | |
392 | return 0; | |
393 | ||
324a56e1 | 394 | if (!sysfs_get_active(of->kn)) |
414985ae TH |
395 | return 0; |
396 | ||
397 | ret = 0; | |
398 | if (of->vm_ops->migrate) | |
399 | ret = of->vm_ops->migrate(vma, from, to, flags); | |
400 | ||
324a56e1 | 401 | sysfs_put_active(of->kn); |
414985ae TH |
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 | { | |
c525aadd | 420 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
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 | */ | |
adc5e8b5 | 431 | if (!(of->kn->flags & SYSFS_FLAG_HAS_MMAP)) |
9b2db6e1 TH |
432 | return -ENODEV; |
433 | ||
414985ae TH |
434 | mutex_lock(&of->mutex); |
435 | ||
436 | rc = -ENODEV; | |
324a56e1 | 437 | if (!sysfs_get_active(of->kn)) |
414985ae TH |
438 | goto out_unlock; |
439 | ||
324a56e1 | 440 | ops = kernfs_ops(of->kn); |
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: | |
324a56e1 | 468 | sysfs_put_active(of->kn); |
414985ae TH |
469 | out_unlock: |
470 | mutex_unlock(&of->mutex); | |
471 | ||
472 | return rc; | |
473 | } | |
474 | ||
475 | /** | |
c525aadd | 476 | * sysfs_get_open_dirent - get or create kernfs_open_node |
324a56e1 | 477 | * @kn: target kernfs_node |
c525aadd | 478 | * @of: kernfs_open_file for this instance of open |
414985ae | 479 | * |
adc5e8b5 TH |
480 | * If @kn->attr.open exists, increment its reference count; otherwise, |
481 | * create one. @of is chained to the files list. | |
414985ae TH |
482 | * |
483 | * LOCKING: | |
484 | * Kernel thread context (may sleep). | |
485 | * | |
486 | * RETURNS: | |
487 | * 0 on success, -errno on failure. | |
488 | */ | |
324a56e1 | 489 | static int sysfs_get_open_dirent(struct kernfs_node *kn, |
c525aadd | 490 | struct kernfs_open_file *of) |
414985ae | 491 | { |
c525aadd | 492 | struct kernfs_open_node *on, *new_on = NULL; |
414985ae TH |
493 | |
494 | retry: | |
c525aadd TH |
495 | mutex_lock(&kernfs_open_file_mutex); |
496 | spin_lock_irq(&kernfs_open_node_lock); | |
414985ae | 497 | |
c525aadd TH |
498 | if (!kn->attr.open && new_on) { |
499 | kn->attr.open = new_on; | |
500 | new_on = NULL; | |
414985ae TH |
501 | } |
502 | ||
c525aadd TH |
503 | on = kn->attr.open; |
504 | if (on) { | |
505 | atomic_inc(&on->refcnt); | |
506 | list_add_tail(&of->list, &on->files); | |
414985ae TH |
507 | } |
508 | ||
c525aadd TH |
509 | spin_unlock_irq(&kernfs_open_node_lock); |
510 | mutex_unlock(&kernfs_open_file_mutex); | |
414985ae | 511 | |
c525aadd TH |
512 | if (on) { |
513 | kfree(new_on); | |
414985ae TH |
514 | return 0; |
515 | } | |
516 | ||
517 | /* not there, initialize a new one and retry */ | |
c525aadd TH |
518 | new_on = kmalloc(sizeof(*new_on), GFP_KERNEL); |
519 | if (!new_on) | |
414985ae TH |
520 | return -ENOMEM; |
521 | ||
c525aadd TH |
522 | atomic_set(&new_on->refcnt, 0); |
523 | atomic_set(&new_on->event, 1); | |
524 | init_waitqueue_head(&new_on->poll); | |
525 | INIT_LIST_HEAD(&new_on->files); | |
414985ae TH |
526 | goto retry; |
527 | } | |
528 | ||
529 | /** | |
c525aadd | 530 | * sysfs_put_open_dirent - put kernfs_open_node |
324a56e1 | 531 | * @kn: target kernfs_nodet |
c525aadd | 532 | * @of: associated kernfs_open_file |
414985ae | 533 | * |
adc5e8b5 | 534 | * Put @kn->attr.open and unlink @of from the files list. If |
414985ae TH |
535 | * reference count reaches zero, disassociate and free it. |
536 | * | |
537 | * LOCKING: | |
538 | * None. | |
539 | */ | |
324a56e1 | 540 | static void sysfs_put_open_dirent(struct kernfs_node *kn, |
c525aadd | 541 | struct kernfs_open_file *of) |
414985ae | 542 | { |
c525aadd | 543 | struct kernfs_open_node *on = kn->attr.open; |
414985ae TH |
544 | unsigned long flags; |
545 | ||
c525aadd TH |
546 | mutex_lock(&kernfs_open_file_mutex); |
547 | spin_lock_irqsave(&kernfs_open_node_lock, flags); | |
414985ae TH |
548 | |
549 | if (of) | |
550 | list_del(&of->list); | |
551 | ||
c525aadd | 552 | if (atomic_dec_and_test(&on->refcnt)) |
adc5e8b5 | 553 | kn->attr.open = NULL; |
414985ae | 554 | else |
c525aadd | 555 | on = NULL; |
414985ae | 556 | |
c525aadd TH |
557 | spin_unlock_irqrestore(&kernfs_open_node_lock, flags); |
558 | mutex_unlock(&kernfs_open_file_mutex); | |
414985ae | 559 | |
c525aadd | 560 | kfree(on); |
414985ae TH |
561 | } |
562 | ||
563 | static int kernfs_file_open(struct inode *inode, struct file *file) | |
564 | { | |
324a56e1 | 565 | struct kernfs_node *kn = file->f_path.dentry->d_fsdata; |
414985ae | 566 | const struct kernfs_ops *ops; |
c525aadd | 567 | struct kernfs_open_file *of; |
414985ae TH |
568 | bool has_read, has_write, has_mmap; |
569 | int error = -EACCES; | |
570 | ||
324a56e1 | 571 | if (!sysfs_get_active(kn)) |
414985ae TH |
572 | return -ENODEV; |
573 | ||
324a56e1 | 574 | ops = kernfs_ops(kn); |
414985ae TH |
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 | ||
c525aadd | 589 | /* allocate a kernfs_open_file for the file */ |
414985ae | 590 | error = -ENOMEM; |
c525aadd | 591 | of = kzalloc(sizeof(struct kernfs_open_file), GFP_KERNEL); |
414985ae TH |
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 | ||
324a56e1 | 615 | of->kn = kn; |
414985ae TH |
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 */ | |
324a56e1 | 637 | error = sysfs_get_open_dirent(kn, of); |
414985ae TH |
638 | if (error) |
639 | goto err_close; | |
640 | ||
641 | /* open succeeded, put active references */ | |
324a56e1 | 642 | sysfs_put_active(kn); |
414985ae TH |
643 | return 0; |
644 | ||
645 | err_close: | |
646 | seq_release(inode, file); | |
647 | err_free: | |
648 | kfree(of); | |
649 | err_out: | |
324a56e1 | 650 | sysfs_put_active(kn); |
414985ae TH |
651 | return error; |
652 | } | |
653 | ||
654 | static int kernfs_file_release(struct inode *inode, struct file *filp) | |
655 | { | |
324a56e1 | 656 | struct kernfs_node *kn = filp->f_path.dentry->d_fsdata; |
c525aadd | 657 | struct kernfs_open_file *of = kernfs_of(filp); |
414985ae | 658 | |
324a56e1 | 659 | sysfs_put_open_dirent(kn, of); |
414985ae TH |
660 | seq_release(inode, filp); |
661 | kfree(of); | |
662 | ||
663 | return 0; | |
664 | } | |
665 | ||
324a56e1 | 666 | void sysfs_unmap_bin_file(struct kernfs_node *kn) |
414985ae | 667 | { |
c525aadd TH |
668 | struct kernfs_open_node *on; |
669 | struct kernfs_open_file *of; | |
414985ae | 670 | |
adc5e8b5 | 671 | if (!(kn->flags & SYSFS_FLAG_HAS_MMAP)) |
414985ae TH |
672 | return; |
673 | ||
c525aadd TH |
674 | spin_lock_irq(&kernfs_open_node_lock); |
675 | on = kn->attr.open; | |
676 | if (on) | |
677 | atomic_inc(&on->refcnt); | |
678 | spin_unlock_irq(&kernfs_open_node_lock); | |
679 | if (!on) | |
414985ae TH |
680 | return; |
681 | ||
c525aadd TH |
682 | mutex_lock(&kernfs_open_file_mutex); |
683 | list_for_each_entry(of, &on->files, list) { | |
414985ae TH |
684 | struct inode *inode = file_inode(of->file); |
685 | unmap_mapping_range(inode->i_mapping, 0, 0, 1); | |
686 | } | |
c525aadd | 687 | mutex_unlock(&kernfs_open_file_mutex); |
414985ae | 688 | |
324a56e1 | 689 | sysfs_put_open_dirent(kn, NULL); |
414985ae TH |
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 | { | |
c525aadd | 707 | struct kernfs_open_file *of = kernfs_of(filp); |
324a56e1 | 708 | struct kernfs_node *kn = filp->f_path.dentry->d_fsdata; |
c525aadd | 709 | struct kernfs_open_node *on = kn->attr.open; |
414985ae TH |
710 | |
711 | /* need parent for the kobj, grab both */ | |
324a56e1 | 712 | if (!sysfs_get_active(kn)) |
414985ae TH |
713 | goto trigger; |
714 | ||
c525aadd | 715 | poll_wait(filp, &on->poll, wait); |
414985ae | 716 | |
324a56e1 | 717 | sysfs_put_active(kn); |
414985ae | 718 | |
c525aadd | 719 | if (of->event != atomic_read(&on->event)) |
414985ae TH |
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 | |
324a56e1 | 730 | * @kn: file to notify |
414985ae | 731 | * |
324a56e1 | 732 | * Notify @kn such that poll(2) on @kn wakes up. |
414985ae | 733 | */ |
324a56e1 | 734 | void kernfs_notify(struct kernfs_node *kn) |
414985ae | 735 | { |
c525aadd | 736 | struct kernfs_open_node *on; |
414985ae TH |
737 | unsigned long flags; |
738 | ||
c525aadd | 739 | spin_lock_irqsave(&kernfs_open_node_lock, flags); |
414985ae | 740 | |
324a56e1 | 741 | if (!WARN_ON(sysfs_type(kn) != SYSFS_KOBJ_ATTR)) { |
c525aadd TH |
742 | on = kn->attr.open; |
743 | if (on) { | |
744 | atomic_inc(&on->event); | |
745 | wake_up_interruptible(&on->poll); | |
414985ae TH |
746 | } |
747 | } | |
748 | ||
c525aadd | 749 | spin_unlock_irqrestore(&kernfs_open_node_lock, flags); |
414985ae TH |
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 | */ | |
324a56e1 TH |
776 | struct kernfs_node *kernfs_create_file_ns_key(struct kernfs_node *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) | |
414985ae | 782 | { |
c525aadd | 783 | struct kernfs_addrm_cxt acxt; |
324a56e1 | 784 | struct kernfs_node *kn; |
414985ae TH |
785 | int rc; |
786 | ||
324a56e1 | 787 | kn = sysfs_new_dirent(kernfs_root(parent), name, |
bc755553 | 788 | (mode & S_IALLUGO) | S_IFREG, SYSFS_KOBJ_ATTR); |
324a56e1 | 789 | if (!kn) |
414985ae TH |
790 | return ERR_PTR(-ENOMEM); |
791 | ||
adc5e8b5 TH |
792 | kn->attr.ops = ops; |
793 | kn->attr.size = size; | |
794 | kn->ns = ns; | |
324a56e1 | 795 | kn->priv = priv; |
414985ae TH |
796 | |
797 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
798 | if (key) { | |
324a56e1 | 799 | lockdep_init_map(&kn->dep_map, "s_active", key, 0); |
adc5e8b5 | 800 | kn->flags |= SYSFS_FLAG_LOCKDEP; |
414985ae TH |
801 | } |
802 | #endif | |
803 | ||
804 | /* | |
adc5e8b5 | 805 | * kn->attr.ops is accesible only while holding active ref. We |
414985ae TH |
806 | * need to know whether some ops are implemented outside active |
807 | * ref. Cache their existence in flags. | |
808 | */ | |
809 | if (ops->seq_show) | |
adc5e8b5 | 810 | kn->flags |= SYSFS_FLAG_HAS_SEQ_SHOW; |
414985ae | 811 | if (ops->mmap) |
adc5e8b5 | 812 | kn->flags |= SYSFS_FLAG_HAS_MMAP; |
414985ae TH |
813 | |
814 | sysfs_addrm_start(&acxt); | |
324a56e1 | 815 | rc = sysfs_add_one(&acxt, kn, parent); |
414985ae TH |
816 | sysfs_addrm_finish(&acxt); |
817 | ||
818 | if (rc) { | |
324a56e1 | 819 | kernfs_put(kn); |
414985ae TH |
820 | return ERR_PTR(rc); |
821 | } | |
324a56e1 | 822 | return kn; |
414985ae | 823 | } |