Line data Source code
1 : // SPDX-License-Identifier: GPL-2.0-only
2 : /*
3 : * fs/kernfs/file.c - kernfs file implementation
4 : *
5 : * Copyright (c) 2001-3 Patrick Mochel
6 : * Copyright (c) 2007 SUSE Linux Products GmbH
7 : * Copyright (c) 2007, 2013 Tejun Heo <tj@kernel.org>
8 : */
9 :
10 : #include <linux/fs.h>
11 : #include <linux/seq_file.h>
12 : #include <linux/slab.h>
13 : #include <linux/poll.h>
14 : #include <linux/pagemap.h>
15 : #include <linux/sched/mm.h>
16 : #include <linux/fsnotify.h>
17 : #include <linux/uio.h>
18 :
19 : #include "kernfs-internal.h"
20 :
21 : struct kernfs_open_node {
22 : struct rcu_head rcu_head;
23 : atomic_t event;
24 : wait_queue_head_t poll;
25 : struct list_head files; /* goes through kernfs_open_file.list */
26 : unsigned int nr_mmapped;
27 : unsigned int nr_to_release;
28 : };
29 :
30 : /*
31 : * kernfs_notify() may be called from any context and bounces notifications
32 : * through a work item. To minimize space overhead in kernfs_node, the
33 : * pending queue is implemented as a singly linked list of kernfs_nodes.
34 : * The list is terminated with the self pointer so that whether a
35 : * kernfs_node is on the list or not can be determined by testing the next
36 : * pointer for %NULL.
37 : */
38 : #define KERNFS_NOTIFY_EOL ((void *)&kernfs_notify_list)
39 :
40 : static DEFINE_SPINLOCK(kernfs_notify_lock);
41 : static struct kernfs_node *kernfs_notify_list = KERNFS_NOTIFY_EOL;
42 :
43 : static inline struct mutex *kernfs_open_file_mutex_ptr(struct kernfs_node *kn)
44 : {
45 0 : int idx = hash_ptr(kn, NR_KERNFS_LOCK_BITS);
46 :
47 0 : return &kernfs_locks->open_file_mutex[idx];
48 : }
49 :
50 : static inline struct mutex *kernfs_open_file_mutex_lock(struct kernfs_node *kn)
51 : {
52 : struct mutex *lock;
53 :
54 0 : lock = kernfs_open_file_mutex_ptr(kn);
55 :
56 0 : mutex_lock(lock);
57 :
58 : return lock;
59 : }
60 :
61 : /**
62 : * of_on - Get the kernfs_open_node of the specified kernfs_open_file
63 : * @of: target kernfs_open_file
64 : *
65 : * Return: the kernfs_open_node of the kernfs_open_file
66 : */
67 : static struct kernfs_open_node *of_on(struct kernfs_open_file *of)
68 : {
69 0 : return rcu_dereference_protected(of->kn->attr.open,
70 : !list_empty(&of->list));
71 : }
72 :
73 : /**
74 : * kernfs_deref_open_node_locked - Get kernfs_open_node corresponding to @kn
75 : *
76 : * @kn: target kernfs_node.
77 : *
78 : * Fetch and return ->attr.open of @kn when caller holds the
79 : * kernfs_open_file_mutex_ptr(kn).
80 : *
81 : * Update of ->attr.open happens under kernfs_open_file_mutex_ptr(kn). So when
82 : * the caller guarantees that this mutex is being held, other updaters can't
83 : * change ->attr.open and this means that we can safely deref ->attr.open
84 : * outside RCU read-side critical section.
85 : *
86 : * The caller needs to make sure that kernfs_open_file_mutex is held.
87 : *
88 : * Return: @kn->attr.open when kernfs_open_file_mutex is held.
89 : */
90 : static struct kernfs_open_node *
91 : kernfs_deref_open_node_locked(struct kernfs_node *kn)
92 : {
93 : return rcu_dereference_protected(kn->attr.open,
94 : lockdep_is_held(kernfs_open_file_mutex_ptr(kn)));
95 : }
96 :
97 : static struct kernfs_open_file *kernfs_of(struct file *file)
98 : {
99 0 : return ((struct seq_file *)file->private_data)->private;
100 : }
101 :
102 : /*
103 : * Determine the kernfs_ops for the given kernfs_node. This function must
104 : * be called while holding an active reference.
105 : */
106 : static const struct kernfs_ops *kernfs_ops(struct kernfs_node *kn)
107 : {
108 : if (kn->flags & KERNFS_LOCKDEP)
109 : lockdep_assert_held(kn);
110 : return kn->attr.ops;
111 : }
112 :
113 : /*
114 : * As kernfs_seq_stop() is also called after kernfs_seq_start() or
115 : * kernfs_seq_next() failure, it needs to distinguish whether it's stopping
116 : * a seq_file iteration which is fully initialized with an active reference
117 : * or an aborted kernfs_seq_start() due to get_active failure. The
118 : * position pointer is the only context for each seq_file iteration and
119 : * thus the stop condition should be encoded in it. As the return value is
120 : * directly visible to userland, ERR_PTR(-ENODEV) is the only acceptable
121 : * choice to indicate get_active failure.
122 : *
123 : * Unfortunately, this is complicated due to the optional custom seq_file
124 : * operations which may return ERR_PTR(-ENODEV) too. kernfs_seq_stop()
125 : * can't distinguish whether ERR_PTR(-ENODEV) is from get_active failure or
126 : * custom seq_file operations and thus can't decide whether put_active
127 : * should be performed or not only on ERR_PTR(-ENODEV).
128 : *
129 : * This is worked around by factoring out the custom seq_stop() and
130 : * put_active part into kernfs_seq_stop_active(), skipping it from
131 : * kernfs_seq_stop() if ERR_PTR(-ENODEV) while invoking it directly after
132 : * custom seq_file operations fail with ERR_PTR(-ENODEV) - this ensures
133 : * that kernfs_seq_stop_active() is skipped only after get_active failure.
134 : */
135 0 : static void kernfs_seq_stop_active(struct seq_file *sf, void *v)
136 : {
137 0 : struct kernfs_open_file *of = sf->private;
138 0 : const struct kernfs_ops *ops = kernfs_ops(of->kn);
139 :
140 0 : if (ops->seq_stop)
141 0 : ops->seq_stop(sf, v);
142 0 : kernfs_put_active(of->kn);
143 0 : }
144 :
145 0 : static void *kernfs_seq_start(struct seq_file *sf, loff_t *ppos)
146 : {
147 0 : struct kernfs_open_file *of = sf->private;
148 : const struct kernfs_ops *ops;
149 :
150 : /*
151 : * @of->mutex nests outside active ref and is primarily to ensure that
152 : * the ops aren't called concurrently for the same open file.
153 : */
154 0 : mutex_lock(&of->mutex);
155 0 : if (!kernfs_get_active(of->kn))
156 : return ERR_PTR(-ENODEV);
157 :
158 0 : ops = kernfs_ops(of->kn);
159 0 : if (ops->seq_start) {
160 0 : void *next = ops->seq_start(sf, ppos);
161 : /* see the comment above kernfs_seq_stop_active() */
162 0 : if (next == ERR_PTR(-ENODEV))
163 0 : kernfs_seq_stop_active(sf, next);
164 : return next;
165 : }
166 0 : return single_start(sf, ppos);
167 : }
168 :
169 0 : static void *kernfs_seq_next(struct seq_file *sf, void *v, loff_t *ppos)
170 : {
171 0 : struct kernfs_open_file *of = sf->private;
172 0 : const struct kernfs_ops *ops = kernfs_ops(of->kn);
173 :
174 0 : if (ops->seq_next) {
175 0 : void *next = ops->seq_next(sf, v, ppos);
176 : /* see the comment above kernfs_seq_stop_active() */
177 0 : if (next == ERR_PTR(-ENODEV))
178 0 : kernfs_seq_stop_active(sf, next);
179 : return next;
180 : } else {
181 : /*
182 : * The same behavior and code as single_open(), always
183 : * terminate after the initial read.
184 : */
185 0 : ++*ppos;
186 0 : return NULL;
187 : }
188 : }
189 :
190 0 : static void kernfs_seq_stop(struct seq_file *sf, void *v)
191 : {
192 0 : struct kernfs_open_file *of = sf->private;
193 :
194 0 : if (v != ERR_PTR(-ENODEV))
195 0 : kernfs_seq_stop_active(sf, v);
196 0 : mutex_unlock(&of->mutex);
197 0 : }
198 :
199 0 : static int kernfs_seq_show(struct seq_file *sf, void *v)
200 : {
201 0 : struct kernfs_open_file *of = sf->private;
202 :
203 0 : of->event = atomic_read(&of_on(of)->event);
204 :
205 0 : return of->kn->attr.ops->seq_show(sf, v);
206 : }
207 :
208 : static const struct seq_operations kernfs_seq_ops = {
209 : .start = kernfs_seq_start,
210 : .next = kernfs_seq_next,
211 : .stop = kernfs_seq_stop,
212 : .show = kernfs_seq_show,
213 : };
214 :
215 : /*
216 : * As reading a bin file can have side-effects, the exact offset and bytes
217 : * specified in read(2) call should be passed to the read callback making
218 : * it difficult to use seq_file. Implement simplistic custom buffering for
219 : * bin files.
220 : */
221 0 : static ssize_t kernfs_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
222 : {
223 0 : struct kernfs_open_file *of = kernfs_of(iocb->ki_filp);
224 0 : ssize_t len = min_t(size_t, iov_iter_count(iter), PAGE_SIZE);
225 : const struct kernfs_ops *ops;
226 : char *buf;
227 :
228 0 : buf = of->prealloc_buf;
229 0 : if (buf)
230 0 : mutex_lock(&of->prealloc_mutex);
231 : else
232 0 : buf = kmalloc(len, GFP_KERNEL);
233 0 : if (!buf)
234 : return -ENOMEM;
235 :
236 : /*
237 : * @of->mutex nests outside active ref and is used both to ensure that
238 : * the ops aren't called concurrently for the same open file.
239 : */
240 0 : mutex_lock(&of->mutex);
241 0 : if (!kernfs_get_active(of->kn)) {
242 0 : len = -ENODEV;
243 0 : mutex_unlock(&of->mutex);
244 : goto out_free;
245 : }
246 :
247 0 : of->event = atomic_read(&of_on(of)->event);
248 :
249 0 : ops = kernfs_ops(of->kn);
250 0 : if (ops->read)
251 0 : len = ops->read(of, buf, len, iocb->ki_pos);
252 : else
253 : len = -EINVAL;
254 :
255 0 : kernfs_put_active(of->kn);
256 0 : mutex_unlock(&of->mutex);
257 :
258 0 : if (len < 0)
259 : goto out_free;
260 :
261 0 : if (copy_to_iter(buf, len, iter) != len) {
262 : len = -EFAULT;
263 : goto out_free;
264 : }
265 :
266 0 : iocb->ki_pos += len;
267 :
268 : out_free:
269 0 : if (buf == of->prealloc_buf)
270 0 : mutex_unlock(&of->prealloc_mutex);
271 : else
272 0 : kfree(buf);
273 : return len;
274 : }
275 :
276 0 : static ssize_t kernfs_fop_read_iter(struct kiocb *iocb, struct iov_iter *iter)
277 : {
278 0 : if (kernfs_of(iocb->ki_filp)->kn->flags & KERNFS_HAS_SEQ_SHOW)
279 0 : return seq_read_iter(iocb, iter);
280 0 : return kernfs_file_read_iter(iocb, iter);
281 : }
282 :
283 : /*
284 : * Copy data in from userland and pass it to the matching kernfs write
285 : * operation.
286 : *
287 : * There is no easy way for us to know if userspace is only doing a partial
288 : * write, so we don't support them. We expect the entire buffer to come on
289 : * the first write. Hint: if you're writing a value, first read the file,
290 : * modify only the value you're changing, then write entire buffer
291 : * back.
292 : */
293 0 : static ssize_t kernfs_fop_write_iter(struct kiocb *iocb, struct iov_iter *iter)
294 : {
295 0 : struct kernfs_open_file *of = kernfs_of(iocb->ki_filp);
296 0 : ssize_t len = iov_iter_count(iter);
297 : const struct kernfs_ops *ops;
298 : char *buf;
299 :
300 0 : if (of->atomic_write_len) {
301 0 : if (len > of->atomic_write_len)
302 : return -E2BIG;
303 : } else {
304 0 : len = min_t(size_t, len, PAGE_SIZE);
305 : }
306 :
307 0 : buf = of->prealloc_buf;
308 0 : if (buf)
309 0 : mutex_lock(&of->prealloc_mutex);
310 : else
311 0 : buf = kmalloc(len + 1, GFP_KERNEL);
312 0 : if (!buf)
313 : return -ENOMEM;
314 :
315 0 : if (copy_from_iter(buf, len, iter) != len) {
316 : len = -EFAULT;
317 : goto out_free;
318 : }
319 0 : buf[len] = '\0'; /* guarantee string termination */
320 :
321 : /*
322 : * @of->mutex nests outside active ref and is used both to ensure that
323 : * the ops aren't called concurrently for the same open file.
324 : */
325 0 : mutex_lock(&of->mutex);
326 0 : if (!kernfs_get_active(of->kn)) {
327 0 : mutex_unlock(&of->mutex);
328 0 : len = -ENODEV;
329 0 : goto out_free;
330 : }
331 :
332 0 : ops = kernfs_ops(of->kn);
333 0 : if (ops->write)
334 0 : len = ops->write(of, buf, len, iocb->ki_pos);
335 : else
336 : len = -EINVAL;
337 :
338 0 : kernfs_put_active(of->kn);
339 0 : mutex_unlock(&of->mutex);
340 :
341 0 : if (len > 0)
342 0 : iocb->ki_pos += len;
343 :
344 : out_free:
345 0 : if (buf == of->prealloc_buf)
346 0 : mutex_unlock(&of->prealloc_mutex);
347 : else
348 0 : kfree(buf);
349 : return len;
350 : }
351 :
352 0 : static void kernfs_vma_open(struct vm_area_struct *vma)
353 : {
354 0 : struct file *file = vma->vm_file;
355 0 : struct kernfs_open_file *of = kernfs_of(file);
356 :
357 0 : if (!of->vm_ops)
358 : return;
359 :
360 0 : if (!kernfs_get_active(of->kn))
361 : return;
362 :
363 0 : if (of->vm_ops->open)
364 0 : of->vm_ops->open(vma);
365 :
366 0 : kernfs_put_active(of->kn);
367 : }
368 :
369 0 : static vm_fault_t kernfs_vma_fault(struct vm_fault *vmf)
370 : {
371 0 : struct file *file = vmf->vma->vm_file;
372 0 : struct kernfs_open_file *of = kernfs_of(file);
373 : vm_fault_t ret;
374 :
375 0 : if (!of->vm_ops)
376 : return VM_FAULT_SIGBUS;
377 :
378 0 : if (!kernfs_get_active(of->kn))
379 : return VM_FAULT_SIGBUS;
380 :
381 0 : ret = VM_FAULT_SIGBUS;
382 0 : if (of->vm_ops->fault)
383 0 : ret = of->vm_ops->fault(vmf);
384 :
385 0 : kernfs_put_active(of->kn);
386 0 : return ret;
387 : }
388 :
389 0 : static vm_fault_t kernfs_vma_page_mkwrite(struct vm_fault *vmf)
390 : {
391 0 : struct file *file = vmf->vma->vm_file;
392 0 : struct kernfs_open_file *of = kernfs_of(file);
393 : vm_fault_t ret;
394 :
395 0 : if (!of->vm_ops)
396 : return VM_FAULT_SIGBUS;
397 :
398 0 : if (!kernfs_get_active(of->kn))
399 : return VM_FAULT_SIGBUS;
400 :
401 0 : ret = 0;
402 0 : if (of->vm_ops->page_mkwrite)
403 0 : ret = of->vm_ops->page_mkwrite(vmf);
404 : else
405 0 : file_update_time(file);
406 :
407 0 : kernfs_put_active(of->kn);
408 0 : return ret;
409 : }
410 :
411 0 : static int kernfs_vma_access(struct vm_area_struct *vma, unsigned long addr,
412 : void *buf, int len, int write)
413 : {
414 0 : struct file *file = vma->vm_file;
415 0 : struct kernfs_open_file *of = kernfs_of(file);
416 : int ret;
417 :
418 0 : if (!of->vm_ops)
419 : return -EINVAL;
420 :
421 0 : if (!kernfs_get_active(of->kn))
422 : return -EINVAL;
423 :
424 0 : ret = -EINVAL;
425 0 : if (of->vm_ops->access)
426 0 : ret = of->vm_ops->access(vma, addr, buf, len, write);
427 :
428 0 : kernfs_put_active(of->kn);
429 0 : return ret;
430 : }
431 :
432 : #ifdef CONFIG_NUMA
433 : static int kernfs_vma_set_policy(struct vm_area_struct *vma,
434 : struct mempolicy *new)
435 : {
436 : struct file *file = vma->vm_file;
437 : struct kernfs_open_file *of = kernfs_of(file);
438 : int ret;
439 :
440 : if (!of->vm_ops)
441 : return 0;
442 :
443 : if (!kernfs_get_active(of->kn))
444 : return -EINVAL;
445 :
446 : ret = 0;
447 : if (of->vm_ops->set_policy)
448 : ret = of->vm_ops->set_policy(vma, new);
449 :
450 : kernfs_put_active(of->kn);
451 : return ret;
452 : }
453 :
454 : static struct mempolicy *kernfs_vma_get_policy(struct vm_area_struct *vma,
455 : unsigned long addr)
456 : {
457 : struct file *file = vma->vm_file;
458 : struct kernfs_open_file *of = kernfs_of(file);
459 : struct mempolicy *pol;
460 :
461 : if (!of->vm_ops)
462 : return vma->vm_policy;
463 :
464 : if (!kernfs_get_active(of->kn))
465 : return vma->vm_policy;
466 :
467 : pol = vma->vm_policy;
468 : if (of->vm_ops->get_policy)
469 : pol = of->vm_ops->get_policy(vma, addr);
470 :
471 : kernfs_put_active(of->kn);
472 : return pol;
473 : }
474 :
475 : #endif
476 :
477 : static const struct vm_operations_struct kernfs_vm_ops = {
478 : .open = kernfs_vma_open,
479 : .fault = kernfs_vma_fault,
480 : .page_mkwrite = kernfs_vma_page_mkwrite,
481 : .access = kernfs_vma_access,
482 : #ifdef CONFIG_NUMA
483 : .set_policy = kernfs_vma_set_policy,
484 : .get_policy = kernfs_vma_get_policy,
485 : #endif
486 : };
487 :
488 0 : static int kernfs_fop_mmap(struct file *file, struct vm_area_struct *vma)
489 : {
490 0 : struct kernfs_open_file *of = kernfs_of(file);
491 : const struct kernfs_ops *ops;
492 : int rc;
493 :
494 : /*
495 : * mmap path and of->mutex are prone to triggering spurious lockdep
496 : * warnings and we don't want to add spurious locking dependency
497 : * between the two. Check whether mmap is actually implemented
498 : * without grabbing @of->mutex by testing HAS_MMAP flag. See the
499 : * comment in kernfs_file_open() for more details.
500 : */
501 0 : if (!(of->kn->flags & KERNFS_HAS_MMAP))
502 : return -ENODEV;
503 :
504 0 : mutex_lock(&of->mutex);
505 :
506 0 : rc = -ENODEV;
507 0 : if (!kernfs_get_active(of->kn))
508 : goto out_unlock;
509 :
510 0 : ops = kernfs_ops(of->kn);
511 0 : rc = ops->mmap(of, vma);
512 0 : if (rc)
513 : goto out_put;
514 :
515 : /*
516 : * PowerPC's pci_mmap of legacy_mem uses shmem_zero_setup()
517 : * to satisfy versions of X which crash if the mmap fails: that
518 : * substitutes a new vm_file, and we don't then want bin_vm_ops.
519 : */
520 0 : if (vma->vm_file != file)
521 : goto out_put;
522 :
523 0 : rc = -EINVAL;
524 0 : if (of->mmapped && of->vm_ops != vma->vm_ops)
525 : goto out_put;
526 :
527 : /*
528 : * It is not possible to successfully wrap close.
529 : * So error if someone is trying to use close.
530 : */
531 0 : if (vma->vm_ops && vma->vm_ops->close)
532 : goto out_put;
533 :
534 0 : rc = 0;
535 0 : of->mmapped = true;
536 0 : of_on(of)->nr_mmapped++;
537 0 : of->vm_ops = vma->vm_ops;
538 0 : vma->vm_ops = &kernfs_vm_ops;
539 : out_put:
540 0 : kernfs_put_active(of->kn);
541 : out_unlock:
542 0 : mutex_unlock(&of->mutex);
543 :
544 0 : return rc;
545 : }
546 :
547 : /**
548 : * kernfs_get_open_node - get or create kernfs_open_node
549 : * @kn: target kernfs_node
550 : * @of: kernfs_open_file for this instance of open
551 : *
552 : * If @kn->attr.open exists, increment its reference count; otherwise,
553 : * create one. @of is chained to the files list.
554 : *
555 : * Locking:
556 : * Kernel thread context (may sleep).
557 : *
558 : * Return:
559 : * %0 on success, -errno on failure.
560 : */
561 0 : static int kernfs_get_open_node(struct kernfs_node *kn,
562 : struct kernfs_open_file *of)
563 : {
564 : struct kernfs_open_node *on;
565 : struct mutex *mutex;
566 :
567 0 : mutex = kernfs_open_file_mutex_lock(kn);
568 0 : on = kernfs_deref_open_node_locked(kn);
569 :
570 0 : if (!on) {
571 : /* not there, initialize a new one */
572 0 : on = kzalloc(sizeof(*on), GFP_KERNEL);
573 0 : if (!on) {
574 0 : mutex_unlock(mutex);
575 0 : return -ENOMEM;
576 : }
577 0 : atomic_set(&on->event, 1);
578 0 : init_waitqueue_head(&on->poll);
579 0 : INIT_LIST_HEAD(&on->files);
580 0 : rcu_assign_pointer(kn->attr.open, on);
581 : }
582 :
583 0 : list_add_tail(&of->list, &on->files);
584 0 : if (kn->flags & KERNFS_HAS_RELEASE)
585 0 : on->nr_to_release++;
586 :
587 0 : mutex_unlock(mutex);
588 0 : return 0;
589 : }
590 :
591 : /**
592 : * kernfs_unlink_open_file - Unlink @of from @kn.
593 : *
594 : * @kn: target kernfs_node
595 : * @of: associated kernfs_open_file
596 : * @open_failed: ->open() failed, cancel ->release()
597 : *
598 : * Unlink @of from list of @kn's associated open files. If list of
599 : * associated open files becomes empty, disassociate and free
600 : * kernfs_open_node.
601 : *
602 : * LOCKING:
603 : * None.
604 : */
605 0 : static void kernfs_unlink_open_file(struct kernfs_node *kn,
606 : struct kernfs_open_file *of,
607 : bool open_failed)
608 : {
609 : struct kernfs_open_node *on;
610 : struct mutex *mutex;
611 :
612 0 : mutex = kernfs_open_file_mutex_lock(kn);
613 :
614 0 : on = kernfs_deref_open_node_locked(kn);
615 0 : if (!on) {
616 0 : mutex_unlock(mutex);
617 0 : return;
618 : }
619 :
620 0 : if (of) {
621 0 : if (kn->flags & KERNFS_HAS_RELEASE) {
622 0 : WARN_ON_ONCE(of->released == open_failed);
623 0 : if (open_failed)
624 0 : on->nr_to_release--;
625 : }
626 0 : if (of->mmapped)
627 0 : on->nr_mmapped--;
628 0 : list_del(&of->list);
629 : }
630 :
631 0 : if (list_empty(&on->files)) {
632 0 : rcu_assign_pointer(kn->attr.open, NULL);
633 0 : kfree_rcu(on, rcu_head);
634 : }
635 :
636 0 : mutex_unlock(mutex);
637 : }
638 :
639 0 : static int kernfs_fop_open(struct inode *inode, struct file *file)
640 : {
641 0 : struct kernfs_node *kn = inode->i_private;
642 0 : struct kernfs_root *root = kernfs_root(kn);
643 : const struct kernfs_ops *ops;
644 : struct kernfs_open_file *of;
645 : bool has_read, has_write, has_mmap;
646 0 : int error = -EACCES;
647 :
648 0 : if (!kernfs_get_active(kn))
649 : return -ENODEV;
650 :
651 0 : ops = kernfs_ops(kn);
652 :
653 0 : has_read = ops->seq_show || ops->read || ops->mmap;
654 0 : has_write = ops->write || ops->mmap;
655 0 : has_mmap = ops->mmap;
656 :
657 : /* see the flag definition for details */
658 0 : if (root->flags & KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK) {
659 0 : if ((file->f_mode & FMODE_WRITE) &&
660 0 : (!(inode->i_mode & S_IWUGO) || !has_write))
661 : goto err_out;
662 :
663 0 : if ((file->f_mode & FMODE_READ) &&
664 0 : (!(inode->i_mode & S_IRUGO) || !has_read))
665 : goto err_out;
666 : }
667 :
668 : /* allocate a kernfs_open_file for the file */
669 0 : error = -ENOMEM;
670 0 : of = kzalloc(sizeof(struct kernfs_open_file), GFP_KERNEL);
671 0 : if (!of)
672 : goto err_out;
673 :
674 : /*
675 : * The following is done to give a different lockdep key to
676 : * @of->mutex for files which implement mmap. This is a rather
677 : * crude way to avoid false positive lockdep warning around
678 : * mm->mmap_lock - mmap nests @of->mutex under mm->mmap_lock and
679 : * reading /sys/block/sda/trace/act_mask grabs sr_mutex, under
680 : * which mm->mmap_lock nests, while holding @of->mutex. As each
681 : * open file has a separate mutex, it's okay as long as those don't
682 : * happen on the same file. At this point, we can't easily give
683 : * each file a separate locking class. Let's differentiate on
684 : * whether the file has mmap or not for now.
685 : *
686 : * Both paths of the branch look the same. They're supposed to
687 : * look that way and give @of->mutex different static lockdep keys.
688 : */
689 0 : if (has_mmap)
690 0 : mutex_init(&of->mutex);
691 : else
692 0 : mutex_init(&of->mutex);
693 :
694 0 : of->kn = kn;
695 0 : of->file = file;
696 :
697 : /*
698 : * Write path needs to atomic_write_len outside active reference.
699 : * Cache it in open_file. See kernfs_fop_write_iter() for details.
700 : */
701 0 : of->atomic_write_len = ops->atomic_write_len;
702 :
703 0 : error = -EINVAL;
704 : /*
705 : * ->seq_show is incompatible with ->prealloc,
706 : * as seq_read does its own allocation.
707 : * ->read must be used instead.
708 : */
709 0 : if (ops->prealloc && ops->seq_show)
710 : goto err_free;
711 0 : if (ops->prealloc) {
712 0 : int len = of->atomic_write_len ?: PAGE_SIZE;
713 0 : of->prealloc_buf = kmalloc(len + 1, GFP_KERNEL);
714 0 : error = -ENOMEM;
715 0 : if (!of->prealloc_buf)
716 : goto err_free;
717 0 : mutex_init(&of->prealloc_mutex);
718 : }
719 :
720 : /*
721 : * Always instantiate seq_file even if read access doesn't use
722 : * seq_file or is not requested. This unifies private data access
723 : * and readable regular files are the vast majority anyway.
724 : */
725 0 : if (ops->seq_show)
726 0 : error = seq_open(file, &kernfs_seq_ops);
727 : else
728 0 : error = seq_open(file, NULL);
729 0 : if (error)
730 : goto err_free;
731 :
732 0 : of->seq_file = file->private_data;
733 0 : of->seq_file->private = of;
734 :
735 : /* seq_file clears PWRITE unconditionally, restore it if WRITE */
736 0 : if (file->f_mode & FMODE_WRITE)
737 0 : file->f_mode |= FMODE_PWRITE;
738 :
739 : /* make sure we have open node struct */
740 0 : error = kernfs_get_open_node(kn, of);
741 0 : if (error)
742 : goto err_seq_release;
743 :
744 0 : if (ops->open) {
745 : /* nobody has access to @of yet, skip @of->mutex */
746 0 : error = ops->open(of);
747 0 : if (error)
748 : goto err_put_node;
749 : }
750 :
751 : /* open succeeded, put active references */
752 0 : kernfs_put_active(kn);
753 0 : return 0;
754 :
755 : err_put_node:
756 0 : kernfs_unlink_open_file(kn, of, true);
757 : err_seq_release:
758 0 : seq_release(inode, file);
759 : err_free:
760 0 : kfree(of->prealloc_buf);
761 0 : kfree(of);
762 : err_out:
763 0 : kernfs_put_active(kn);
764 0 : return error;
765 : }
766 :
767 : /* used from release/drain to ensure that ->release() is called exactly once */
768 : static void kernfs_release_file(struct kernfs_node *kn,
769 : struct kernfs_open_file *of)
770 : {
771 : /*
772 : * @of is guaranteed to have no other file operations in flight and
773 : * we just want to synchronize release and drain paths.
774 : * @kernfs_open_file_mutex_ptr(kn) is enough. @of->mutex can't be used
775 : * here because drain path may be called from places which can
776 : * cause circular dependency.
777 : */
778 0 : lockdep_assert_held(kernfs_open_file_mutex_ptr(kn));
779 :
780 0 : if (!of->released) {
781 : /*
782 : * A file is never detached without being released and we
783 : * need to be able to release files which are deactivated
784 : * and being drained. Don't use kernfs_ops().
785 : */
786 0 : kn->attr.ops->release(of);
787 0 : of->released = true;
788 0 : of_on(of)->nr_to_release--;
789 : }
790 : }
791 :
792 0 : static int kernfs_fop_release(struct inode *inode, struct file *filp)
793 : {
794 0 : struct kernfs_node *kn = inode->i_private;
795 0 : struct kernfs_open_file *of = kernfs_of(filp);
796 :
797 0 : if (kn->flags & KERNFS_HAS_RELEASE) {
798 : struct mutex *mutex;
799 :
800 0 : mutex = kernfs_open_file_mutex_lock(kn);
801 0 : kernfs_release_file(kn, of);
802 0 : mutex_unlock(mutex);
803 : }
804 :
805 0 : kernfs_unlink_open_file(kn, of, false);
806 0 : seq_release(inode, filp);
807 0 : kfree(of->prealloc_buf);
808 0 : kfree(of);
809 :
810 0 : return 0;
811 : }
812 :
813 212 : bool kernfs_should_drain_open_files(struct kernfs_node *kn)
814 : {
815 : struct kernfs_open_node *on;
816 : bool ret;
817 :
818 : /*
819 : * @kn being deactivated guarantees that @kn->attr.open can't change
820 : * beneath us making the lockless test below safe.
821 : */
822 424 : WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
823 :
824 : rcu_read_lock();
825 212 : on = rcu_dereference(kn->attr.open);
826 212 : ret = on && (on->nr_mmapped || on->nr_to_release);
827 : rcu_read_unlock();
828 :
829 212 : return ret;
830 : }
831 :
832 0 : void kernfs_drain_open_files(struct kernfs_node *kn)
833 : {
834 : struct kernfs_open_node *on;
835 : struct kernfs_open_file *of;
836 : struct mutex *mutex;
837 :
838 0 : mutex = kernfs_open_file_mutex_lock(kn);
839 0 : on = kernfs_deref_open_node_locked(kn);
840 0 : if (!on) {
841 0 : mutex_unlock(mutex);
842 0 : return;
843 : }
844 :
845 0 : list_for_each_entry(of, &on->files, list) {
846 0 : struct inode *inode = file_inode(of->file);
847 :
848 0 : if (of->mmapped) {
849 0 : unmap_mapping_range(inode->i_mapping, 0, 0, 1);
850 0 : of->mmapped = false;
851 0 : on->nr_mmapped--;
852 : }
853 :
854 0 : if (kn->flags & KERNFS_HAS_RELEASE)
855 0 : kernfs_release_file(kn, of);
856 : }
857 :
858 0 : WARN_ON_ONCE(on->nr_mmapped || on->nr_to_release);
859 0 : mutex_unlock(mutex);
860 : }
861 :
862 : /*
863 : * Kernfs attribute files are pollable. The idea is that you read
864 : * the content and then you use 'poll' or 'select' to wait for
865 : * the content to change. When the content changes (assuming the
866 : * manager for the kobject supports notification), poll will
867 : * return EPOLLERR|EPOLLPRI, and select will return the fd whether
868 : * it is waiting for read, write, or exceptions.
869 : * Once poll/select indicates that the value has changed, you
870 : * need to close and re-open the file, or seek to 0 and read again.
871 : * Reminder: this only works for attributes which actively support
872 : * it, and it is not possible to test an attribute from userspace
873 : * to see if it supports poll (Neither 'poll' nor 'select' return
874 : * an appropriate error code). When in doubt, set a suitable timeout value.
875 : */
876 0 : __poll_t kernfs_generic_poll(struct kernfs_open_file *of, poll_table *wait)
877 : {
878 0 : struct kernfs_open_node *on = of_on(of);
879 :
880 0 : poll_wait(of->file, &on->poll, wait);
881 :
882 0 : if (of->event != atomic_read(&on->event))
883 : return DEFAULT_POLLMASK|EPOLLERR|EPOLLPRI;
884 :
885 0 : return DEFAULT_POLLMASK;
886 : }
887 :
888 0 : static __poll_t kernfs_fop_poll(struct file *filp, poll_table *wait)
889 : {
890 0 : struct kernfs_open_file *of = kernfs_of(filp);
891 0 : struct kernfs_node *kn = kernfs_dentry_node(filp->f_path.dentry);
892 : __poll_t ret;
893 :
894 0 : if (!kernfs_get_active(kn))
895 : return DEFAULT_POLLMASK|EPOLLERR|EPOLLPRI;
896 :
897 0 : if (kn->attr.ops->poll)
898 0 : ret = kn->attr.ops->poll(of, wait);
899 : else
900 : ret = kernfs_generic_poll(of, wait);
901 :
902 0 : kernfs_put_active(kn);
903 0 : return ret;
904 : }
905 :
906 1 : static void kernfs_notify_workfn(struct work_struct *work)
907 : {
908 : struct kernfs_node *kn;
909 : struct kernfs_super_info *info;
910 : struct kernfs_root *root;
911 : repeat:
912 : /* pop one off the notify_list */
913 2 : spin_lock_irq(&kernfs_notify_lock);
914 2 : kn = kernfs_notify_list;
915 2 : if (kn == KERNFS_NOTIFY_EOL) {
916 1 : spin_unlock_irq(&kernfs_notify_lock);
917 1 : return;
918 : }
919 1 : kernfs_notify_list = kn->attr.notify_next;
920 1 : kn->attr.notify_next = NULL;
921 1 : spin_unlock_irq(&kernfs_notify_lock);
922 :
923 1 : root = kernfs_root(kn);
924 : /* kick fsnotify */
925 :
926 1 : down_read(&root->kernfs_supers_rwsem);
927 2 : list_for_each_entry(info, &kernfs_root(kn)->supers, node) {
928 : struct kernfs_node *parent;
929 0 : struct inode *p_inode = NULL;
930 : struct inode *inode;
931 : struct qstr name;
932 :
933 : /*
934 : * We want fsnotify_modify() on @kn but as the
935 : * modifications aren't originating from userland don't
936 : * have the matching @file available. Look up the inodes
937 : * and generate the events manually.
938 : */
939 0 : inode = ilookup(info->sb, kernfs_ino(kn));
940 0 : if (!inode)
941 0 : continue;
942 :
943 0 : name = (struct qstr)QSTR_INIT(kn->name, strlen(kn->name));
944 0 : parent = kernfs_get_parent(kn);
945 0 : if (parent) {
946 0 : p_inode = ilookup(info->sb, kernfs_ino(parent));
947 0 : if (p_inode) {
948 0 : fsnotify(FS_MODIFY | FS_EVENT_ON_CHILD,
949 : inode, FSNOTIFY_EVENT_INODE,
950 : p_inode, &name, inode, 0);
951 0 : iput(p_inode);
952 : }
953 :
954 0 : kernfs_put(parent);
955 : }
956 :
957 0 : if (!p_inode)
958 0 : fsnotify_inode(inode, FS_MODIFY);
959 :
960 0 : iput(inode);
961 : }
962 :
963 1 : up_read(&root->kernfs_supers_rwsem);
964 1 : kernfs_put(kn);
965 1 : goto repeat;
966 : }
967 :
968 : /**
969 : * kernfs_notify - notify a kernfs file
970 : * @kn: file to notify
971 : *
972 : * Notify @kn such that poll(2) on @kn wakes up. Maybe be called from any
973 : * context.
974 : */
975 2 : void kernfs_notify(struct kernfs_node *kn)
976 : {
977 : static DECLARE_WORK(kernfs_notify_work, kernfs_notify_workfn);
978 : unsigned long flags;
979 : struct kernfs_open_node *on;
980 :
981 4 : if (WARN_ON(kernfs_type(kn) != KERNFS_FILE))
982 : return;
983 :
984 : /* kick poll immediately */
985 : rcu_read_lock();
986 2 : on = rcu_dereference(kn->attr.open);
987 2 : if (on) {
988 0 : atomic_inc(&on->event);
989 0 : wake_up_interruptible(&on->poll);
990 : }
991 : rcu_read_unlock();
992 :
993 : /* schedule work to kick fsnotify */
994 2 : spin_lock_irqsave(&kernfs_notify_lock, flags);
995 2 : if (!kn->attr.notify_next) {
996 1 : kernfs_get(kn);
997 1 : kn->attr.notify_next = kernfs_notify_list;
998 1 : kernfs_notify_list = kn;
999 : schedule_work(&kernfs_notify_work);
1000 : }
1001 : spin_unlock_irqrestore(&kernfs_notify_lock, flags);
1002 : }
1003 : EXPORT_SYMBOL_GPL(kernfs_notify);
1004 :
1005 : const struct file_operations kernfs_file_fops = {
1006 : .read_iter = kernfs_fop_read_iter,
1007 : .write_iter = kernfs_fop_write_iter,
1008 : .llseek = generic_file_llseek,
1009 : .mmap = kernfs_fop_mmap,
1010 : .open = kernfs_fop_open,
1011 : .release = kernfs_fop_release,
1012 : .poll = kernfs_fop_poll,
1013 : .fsync = noop_fsync,
1014 : .splice_read = copy_splice_read,
1015 : .splice_write = iter_file_splice_write,
1016 : };
1017 :
1018 : /**
1019 : * __kernfs_create_file - kernfs internal function to create a file
1020 : * @parent: directory to create the file in
1021 : * @name: name of the file
1022 : * @mode: mode of the file
1023 : * @uid: uid of the file
1024 : * @gid: gid of the file
1025 : * @size: size of the file
1026 : * @ops: kernfs operations for the file
1027 : * @priv: private data for the file
1028 : * @ns: optional namespace tag of the file
1029 : * @key: lockdep key for the file's active_ref, %NULL to disable lockdep
1030 : *
1031 : * Return: the created node on success, ERR_PTR() value on error.
1032 : */
1033 5467 : struct kernfs_node *__kernfs_create_file(struct kernfs_node *parent,
1034 : const char *name,
1035 : umode_t mode, kuid_t uid, kgid_t gid,
1036 : loff_t size,
1037 : const struct kernfs_ops *ops,
1038 : void *priv, const void *ns,
1039 : struct lock_class_key *key)
1040 : {
1041 : struct kernfs_node *kn;
1042 : unsigned flags;
1043 : int rc;
1044 :
1045 5467 : flags = KERNFS_FILE;
1046 :
1047 5467 : kn = kernfs_new_node(parent, name, (mode & S_IALLUGO) | S_IFREG,
1048 : uid, gid, flags);
1049 5467 : if (!kn)
1050 : return ERR_PTR(-ENOMEM);
1051 :
1052 5467 : kn->attr.ops = ops;
1053 5467 : kn->attr.size = size;
1054 5467 : kn->ns = ns;
1055 5467 : kn->priv = priv;
1056 :
1057 : #ifdef CONFIG_DEBUG_LOCK_ALLOC
1058 : if (key) {
1059 : lockdep_init_map(&kn->dep_map, "kn->active", key, 0);
1060 : kn->flags |= KERNFS_LOCKDEP;
1061 : }
1062 : #endif
1063 :
1064 : /*
1065 : * kn->attr.ops is accessible only while holding active ref. We
1066 : * need to know whether some ops are implemented outside active
1067 : * ref. Cache their existence in flags.
1068 : */
1069 5467 : if (ops->seq_show)
1070 5458 : kn->flags |= KERNFS_HAS_SEQ_SHOW;
1071 5467 : if (ops->mmap)
1072 0 : kn->flags |= KERNFS_HAS_MMAP;
1073 5467 : if (ops->release)
1074 0 : kn->flags |= KERNFS_HAS_RELEASE;
1075 :
1076 5467 : rc = kernfs_add_one(kn);
1077 5467 : if (rc) {
1078 0 : kernfs_put(kn);
1079 0 : return ERR_PTR(rc);
1080 : }
1081 : return kn;
1082 : }
|
