2 * fs/kernfs/file.c - kernfs file implementation
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>
8 * This file is released under the GPLv2.
12 #include <linux/seq_file.h>
13 #include <linux/slab.h>
14 #include <linux/poll.h>
15 #include <linux/pagemap.h>
16 #include <linux/sched.h>
18 #include "kernfs-internal.h"
21 * There's one kernfs_open_file for each open file and one kernfs_open_node
22 * for each kernfs_node with one or more open files.
24 * kernfs_node->attr.open points to kernfs_open_node. attr.open is
25 * protected by kernfs_open_node_lock.
27 * filp->private_data points to seq_file whose ->private points to
28 * kernfs_open_file. kernfs_open_files are chained at
29 * kernfs_open_node->files, which is protected by kernfs_open_file_mutex.
31 static DEFINE_SPINLOCK(kernfs_open_node_lock
);
32 static DEFINE_MUTEX(kernfs_open_file_mutex
);
34 struct kernfs_open_node
{
37 wait_queue_head_t poll
;
38 struct list_head files
; /* goes through kernfs_open_file.list */
41 static struct kernfs_open_file
*kernfs_of(struct file
*file
)
43 return ((struct seq_file
*)file
->private_data
)->private;
47 * Determine the kernfs_ops for the given kernfs_node. This function must
48 * be called while holding an active reference.
50 static const struct kernfs_ops
*kernfs_ops(struct kernfs_node
*kn
)
52 if (kn
->flags
& KERNFS_LOCKDEP
)
53 lockdep_assert_held(kn
);
58 * As kernfs_seq_stop() is also called after kernfs_seq_start() or
59 * kernfs_seq_next() failure, it needs to distinguish whether it's stopping
60 * a seq_file iteration which is fully initialized with an active reference
61 * or an aborted kernfs_seq_start() due to get_active failure. The
62 * position pointer is the only context for each seq_file iteration and
63 * thus the stop condition should be encoded in it. As the return value is
64 * directly visible to userland, ERR_PTR(-ENODEV) is the only acceptable
65 * choice to indicate get_active failure.
67 * Unfortunately, this is complicated due to the optional custom seq_file
68 * operations which may return ERR_PTR(-ENODEV) too. kernfs_seq_stop()
69 * can't distinguish whether ERR_PTR(-ENODEV) is from get_active failure or
70 * custom seq_file operations and thus can't decide whether put_active
71 * should be performed or not only on ERR_PTR(-ENODEV).
73 * This is worked around by factoring out the custom seq_stop() and
74 * put_active part into kernfs_seq_stop_active(), skipping it from
75 * kernfs_seq_stop() if ERR_PTR(-ENODEV) while invoking it directly after
76 * custom seq_file operations fail with ERR_PTR(-ENODEV) - this ensures
77 * that kernfs_seq_stop_active() is skipped only after get_active failure.
79 static void kernfs_seq_stop_active(struct seq_file
*sf
, void *v
)
81 struct kernfs_open_file
*of
= sf
->private;
82 const struct kernfs_ops
*ops
= kernfs_ops(of
->kn
);
86 kernfs_put_active(of
->kn
);
89 static void *kernfs_seq_start(struct seq_file
*sf
, loff_t
*ppos
)
91 struct kernfs_open_file
*of
= sf
->private;
92 const struct kernfs_ops
*ops
;
95 * @of->mutex nests outside active ref and is just to ensure that
96 * the ops aren't called concurrently for the same open file.
98 mutex_lock(&of
->mutex
);
99 if (!kernfs_get_active(of
->kn
))
100 return ERR_PTR(-ENODEV
);
102 ops
= kernfs_ops(of
->kn
);
103 if (ops
->seq_start
) {
104 void *next
= ops
->seq_start(sf
, ppos
);
105 /* see the comment above kernfs_seq_stop_active() */
106 if (next
== ERR_PTR(-ENODEV
))
107 kernfs_seq_stop_active(sf
, next
);
111 * The same behavior and code as single_open(). Returns
112 * !NULL if pos is at the beginning; otherwise, NULL.
114 return NULL
+ !*ppos
;
118 static void *kernfs_seq_next(struct seq_file
*sf
, void *v
, loff_t
*ppos
)
120 struct kernfs_open_file
*of
= sf
->private;
121 const struct kernfs_ops
*ops
= kernfs_ops(of
->kn
);
124 void *next
= ops
->seq_next(sf
, v
, ppos
);
125 /* see the comment above kernfs_seq_stop_active() */
126 if (next
== ERR_PTR(-ENODEV
))
127 kernfs_seq_stop_active(sf
, next
);
131 * The same behavior and code as single_open(), always
132 * terminate after the initial read.
139 static void kernfs_seq_stop(struct seq_file
*sf
, void *v
)
141 struct kernfs_open_file
*of
= sf
->private;
143 if (v
!= ERR_PTR(-ENODEV
))
144 kernfs_seq_stop_active(sf
, v
);
145 mutex_unlock(&of
->mutex
);
148 static int kernfs_seq_show(struct seq_file
*sf
, void *v
)
150 struct kernfs_open_file
*of
= sf
->private;
152 of
->event
= atomic_read(&of
->kn
->attr
.open
->event
);
154 return of
->kn
->attr
.ops
->seq_show(sf
, v
);
157 static const struct seq_operations kernfs_seq_ops
= {
158 .start
= kernfs_seq_start
,
159 .next
= kernfs_seq_next
,
160 .stop
= kernfs_seq_stop
,
161 .show
= kernfs_seq_show
,
165 * As reading a bin file can have side-effects, the exact offset and bytes
166 * specified in read(2) call should be passed to the read callback making
167 * it difficult to use seq_file. Implement simplistic custom buffering for
170 static ssize_t
kernfs_file_direct_read(struct kernfs_open_file
*of
,
171 char __user
*user_buf
, size_t count
,
174 ssize_t len
= min_t(size_t, count
, PAGE_SIZE
);
175 const struct kernfs_ops
*ops
;
178 buf
= kmalloc(len
, GFP_KERNEL
);
183 * @of->mutex nests outside active ref and is just to ensure that
184 * the ops aren't called concurrently for the same open file.
186 mutex_lock(&of
->mutex
);
187 if (!kernfs_get_active(of
->kn
)) {
189 mutex_unlock(&of
->mutex
);
193 ops
= kernfs_ops(of
->kn
);
195 len
= ops
->read(of
, buf
, len
, *ppos
);
199 kernfs_put_active(of
->kn
);
200 mutex_unlock(&of
->mutex
);
205 if (copy_to_user(user_buf
, buf
, len
)) {
218 * kernfs_fop_read - kernfs vfs read callback
219 * @file: file pointer
220 * @user_buf: data to write
221 * @count: number of bytes
222 * @ppos: starting offset
224 static ssize_t
kernfs_fop_read(struct file
*file
, char __user
*user_buf
,
225 size_t count
, loff_t
*ppos
)
227 struct kernfs_open_file
*of
= kernfs_of(file
);
229 if (of
->kn
->flags
& KERNFS_HAS_SEQ_SHOW
)
230 return seq_read(file
, user_buf
, count
, ppos
);
232 return kernfs_file_direct_read(of
, user_buf
, count
, ppos
);
236 * kernfs_fop_write - kernfs vfs write callback
237 * @file: file pointer
238 * @user_buf: data to write
239 * @count: number of bytes
240 * @ppos: starting offset
242 * Copy data in from userland and pass it to the matching kernfs write
245 * There is no easy way for us to know if userspace is only doing a partial
246 * write, so we don't support them. We expect the entire buffer to come on
247 * the first write. Hint: if you're writing a value, first read the file,
248 * modify only the the value you're changing, then write entire buffer
251 static ssize_t
kernfs_fop_write(struct file
*file
, const char __user
*user_buf
,
252 size_t count
, loff_t
*ppos
)
254 struct kernfs_open_file
*of
= kernfs_of(file
);
255 const struct kernfs_ops
*ops
;
259 if (of
->atomic_write_len
) {
261 if (len
> of
->atomic_write_len
)
264 len
= min_t(size_t, count
, PAGE_SIZE
);
267 buf
= kmalloc(len
+ 1, GFP_KERNEL
);
271 if (copy_from_user(buf
, user_buf
, len
)) {
275 buf
[len
] = '\0'; /* guarantee string termination */
278 * @of->mutex nests outside active ref and is just to ensure that
279 * the ops aren't called concurrently for the same open file.
281 mutex_lock(&of
->mutex
);
282 if (!kernfs_get_active(of
->kn
)) {
283 mutex_unlock(&of
->mutex
);
288 ops
= kernfs_ops(of
->kn
);
290 len
= ops
->write(of
, buf
, len
, *ppos
);
294 kernfs_put_active(of
->kn
);
295 mutex_unlock(&of
->mutex
);
304 static void kernfs_vma_open(struct vm_area_struct
*vma
)
306 struct file
*file
= vma
->vm_file
;
307 struct kernfs_open_file
*of
= kernfs_of(file
);
312 if (!kernfs_get_active(of
->kn
))
315 if (of
->vm_ops
->open
)
316 of
->vm_ops
->open(vma
);
318 kernfs_put_active(of
->kn
);
321 static int kernfs_vma_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
323 struct file
*file
= vma
->vm_file
;
324 struct kernfs_open_file
*of
= kernfs_of(file
);
328 return VM_FAULT_SIGBUS
;
330 if (!kernfs_get_active(of
->kn
))
331 return VM_FAULT_SIGBUS
;
333 ret
= VM_FAULT_SIGBUS
;
334 if (of
->vm_ops
->fault
)
335 ret
= of
->vm_ops
->fault(vma
, vmf
);
337 kernfs_put_active(of
->kn
);
341 static int kernfs_vma_page_mkwrite(struct vm_area_struct
*vma
,
342 struct vm_fault
*vmf
)
344 struct file
*file
= vma
->vm_file
;
345 struct kernfs_open_file
*of
= kernfs_of(file
);
349 return VM_FAULT_SIGBUS
;
351 if (!kernfs_get_active(of
->kn
))
352 return VM_FAULT_SIGBUS
;
355 if (of
->vm_ops
->page_mkwrite
)
356 ret
= of
->vm_ops
->page_mkwrite(vma
, vmf
);
358 file_update_time(file
);
360 kernfs_put_active(of
->kn
);
364 static int kernfs_vma_access(struct vm_area_struct
*vma
, unsigned long addr
,
365 void *buf
, int len
, int write
)
367 struct file
*file
= vma
->vm_file
;
368 struct kernfs_open_file
*of
= kernfs_of(file
);
374 if (!kernfs_get_active(of
->kn
))
378 if (of
->vm_ops
->access
)
379 ret
= of
->vm_ops
->access(vma
, addr
, buf
, len
, write
);
381 kernfs_put_active(of
->kn
);
386 static int kernfs_vma_set_policy(struct vm_area_struct
*vma
,
387 struct mempolicy
*new)
389 struct file
*file
= vma
->vm_file
;
390 struct kernfs_open_file
*of
= kernfs_of(file
);
396 if (!kernfs_get_active(of
->kn
))
400 if (of
->vm_ops
->set_policy
)
401 ret
= of
->vm_ops
->set_policy(vma
, new);
403 kernfs_put_active(of
->kn
);
407 static struct mempolicy
*kernfs_vma_get_policy(struct vm_area_struct
*vma
,
410 struct file
*file
= vma
->vm_file
;
411 struct kernfs_open_file
*of
= kernfs_of(file
);
412 struct mempolicy
*pol
;
415 return vma
->vm_policy
;
417 if (!kernfs_get_active(of
->kn
))
418 return vma
->vm_policy
;
420 pol
= vma
->vm_policy
;
421 if (of
->vm_ops
->get_policy
)
422 pol
= of
->vm_ops
->get_policy(vma
, addr
);
424 kernfs_put_active(of
->kn
);
428 static int kernfs_vma_migrate(struct vm_area_struct
*vma
,
429 const nodemask_t
*from
, const nodemask_t
*to
,
432 struct file
*file
= vma
->vm_file
;
433 struct kernfs_open_file
*of
= kernfs_of(file
);
439 if (!kernfs_get_active(of
->kn
))
443 if (of
->vm_ops
->migrate
)
444 ret
= of
->vm_ops
->migrate(vma
, from
, to
, flags
);
446 kernfs_put_active(of
->kn
);
451 static const struct vm_operations_struct kernfs_vm_ops
= {
452 .open
= kernfs_vma_open
,
453 .fault
= kernfs_vma_fault
,
454 .page_mkwrite
= kernfs_vma_page_mkwrite
,
455 .access
= kernfs_vma_access
,
457 .set_policy
= kernfs_vma_set_policy
,
458 .get_policy
= kernfs_vma_get_policy
,
459 .migrate
= kernfs_vma_migrate
,
463 static int kernfs_fop_mmap(struct file
*file
, struct vm_area_struct
*vma
)
465 struct kernfs_open_file
*of
= kernfs_of(file
);
466 const struct kernfs_ops
*ops
;
470 * mmap path and of->mutex are prone to triggering spurious lockdep
471 * warnings and we don't want to add spurious locking dependency
472 * between the two. Check whether mmap is actually implemented
473 * without grabbing @of->mutex by testing HAS_MMAP flag. See the
474 * comment in kernfs_file_open() for more details.
476 if (!(of
->kn
->flags
& KERNFS_HAS_MMAP
))
479 mutex_lock(&of
->mutex
);
482 if (!kernfs_get_active(of
->kn
))
485 ops
= kernfs_ops(of
->kn
);
486 rc
= ops
->mmap(of
, vma
);
491 * PowerPC's pci_mmap of legacy_mem uses shmem_zero_setup()
492 * to satisfy versions of X which crash if the mmap fails: that
493 * substitutes a new vm_file, and we don't then want bin_vm_ops.
495 if (vma
->vm_file
!= file
)
499 if (of
->mmapped
&& of
->vm_ops
!= vma
->vm_ops
)
503 * It is not possible to successfully wrap close.
504 * So error if someone is trying to use close.
507 if (vma
->vm_ops
&& vma
->vm_ops
->close
)
512 of
->vm_ops
= vma
->vm_ops
;
513 vma
->vm_ops
= &kernfs_vm_ops
;
515 kernfs_put_active(of
->kn
);
517 mutex_unlock(&of
->mutex
);
523 * kernfs_get_open_node - get or create kernfs_open_node
524 * @kn: target kernfs_node
525 * @of: kernfs_open_file for this instance of open
527 * If @kn->attr.open exists, increment its reference count; otherwise,
528 * create one. @of is chained to the files list.
531 * Kernel thread context (may sleep).
534 * 0 on success, -errno on failure.
536 static int kernfs_get_open_node(struct kernfs_node
*kn
,
537 struct kernfs_open_file
*of
)
539 struct kernfs_open_node
*on
, *new_on
= NULL
;
542 mutex_lock(&kernfs_open_file_mutex
);
543 spin_lock_irq(&kernfs_open_node_lock
);
545 if (!kn
->attr
.open
&& new_on
) {
546 kn
->attr
.open
= new_on
;
552 atomic_inc(&on
->refcnt
);
553 list_add_tail(&of
->list
, &on
->files
);
556 spin_unlock_irq(&kernfs_open_node_lock
);
557 mutex_unlock(&kernfs_open_file_mutex
);
564 /* not there, initialize a new one and retry */
565 new_on
= kmalloc(sizeof(*new_on
), GFP_KERNEL
);
569 atomic_set(&new_on
->refcnt
, 0);
570 atomic_set(&new_on
->event
, 1);
571 init_waitqueue_head(&new_on
->poll
);
572 INIT_LIST_HEAD(&new_on
->files
);
577 * kernfs_put_open_node - put kernfs_open_node
578 * @kn: target kernfs_nodet
579 * @of: associated kernfs_open_file
581 * Put @kn->attr.open and unlink @of from the files list. If
582 * reference count reaches zero, disassociate and free it.
587 static void kernfs_put_open_node(struct kernfs_node
*kn
,
588 struct kernfs_open_file
*of
)
590 struct kernfs_open_node
*on
= kn
->attr
.open
;
593 mutex_lock(&kernfs_open_file_mutex
);
594 spin_lock_irqsave(&kernfs_open_node_lock
, flags
);
599 if (atomic_dec_and_test(&on
->refcnt
))
600 kn
->attr
.open
= NULL
;
604 spin_unlock_irqrestore(&kernfs_open_node_lock
, flags
);
605 mutex_unlock(&kernfs_open_file_mutex
);
610 static int kernfs_fop_open(struct inode
*inode
, struct file
*file
)
612 struct kernfs_node
*kn
= file
->f_path
.dentry
->d_fsdata
;
613 struct kernfs_root
*root
= kernfs_root(kn
);
614 const struct kernfs_ops
*ops
;
615 struct kernfs_open_file
*of
;
616 bool has_read
, has_write
, has_mmap
;
619 if (!kernfs_get_active(kn
))
622 ops
= kernfs_ops(kn
);
624 has_read
= ops
->seq_show
|| ops
->read
|| ops
->mmap
;
625 has_write
= ops
->write
|| ops
->mmap
;
626 has_mmap
= ops
->mmap
;
628 /* see the flag definition for details */
629 if (root
->flags
& KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK
) {
630 if ((file
->f_mode
& FMODE_WRITE
) &&
631 (!(inode
->i_mode
& S_IWUGO
) || !has_write
))
634 if ((file
->f_mode
& FMODE_READ
) &&
635 (!(inode
->i_mode
& S_IRUGO
) || !has_read
))
639 /* allocate a kernfs_open_file for the file */
641 of
= kzalloc(sizeof(struct kernfs_open_file
), GFP_KERNEL
);
646 * The following is done to give a different lockdep key to
647 * @of->mutex for files which implement mmap. This is a rather
648 * crude way to avoid false positive lockdep warning around
649 * mm->mmap_sem - mmap nests @of->mutex under mm->mmap_sem and
650 * reading /sys/block/sda/trace/act_mask grabs sr_mutex, under
651 * which mm->mmap_sem nests, while holding @of->mutex. As each
652 * open file has a separate mutex, it's okay as long as those don't
653 * happen on the same file. At this point, we can't easily give
654 * each file a separate locking class. Let's differentiate on
655 * whether the file has mmap or not for now.
657 * Both paths of the branch look the same. They're supposed to
658 * look that way and give @of->mutex different static lockdep keys.
661 mutex_init(&of
->mutex
);
663 mutex_init(&of
->mutex
);
669 * Write path needs to atomic_write_len outside active reference.
670 * Cache it in open_file. See kernfs_fop_write() for details.
672 of
->atomic_write_len
= ops
->atomic_write_len
;
675 * Always instantiate seq_file even if read access doesn't use
676 * seq_file or is not requested. This unifies private data access
677 * and readable regular files are the vast majority anyway.
680 error
= seq_open(file
, &kernfs_seq_ops
);
682 error
= seq_open(file
, NULL
);
686 ((struct seq_file
*)file
->private_data
)->private = of
;
688 /* seq_file clears PWRITE unconditionally, restore it if WRITE */
689 if (file
->f_mode
& FMODE_WRITE
)
690 file
->f_mode
|= FMODE_PWRITE
;
692 /* make sure we have open node struct */
693 error
= kernfs_get_open_node(kn
, of
);
697 /* open succeeded, put active references */
698 kernfs_put_active(kn
);
702 seq_release(inode
, file
);
706 kernfs_put_active(kn
);
710 static int kernfs_fop_release(struct inode
*inode
, struct file
*filp
)
712 struct kernfs_node
*kn
= filp
->f_path
.dentry
->d_fsdata
;
713 struct kernfs_open_file
*of
= kernfs_of(filp
);
715 kernfs_put_open_node(kn
, of
);
716 seq_release(inode
, filp
);
722 void kernfs_unmap_bin_file(struct kernfs_node
*kn
)
724 struct kernfs_open_node
*on
;
725 struct kernfs_open_file
*of
;
727 if (!(kn
->flags
& KERNFS_HAS_MMAP
))
730 spin_lock_irq(&kernfs_open_node_lock
);
733 atomic_inc(&on
->refcnt
);
734 spin_unlock_irq(&kernfs_open_node_lock
);
738 mutex_lock(&kernfs_open_file_mutex
);
739 list_for_each_entry(of
, &on
->files
, list
) {
740 struct inode
*inode
= file_inode(of
->file
);
741 unmap_mapping_range(inode
->i_mapping
, 0, 0, 1);
743 mutex_unlock(&kernfs_open_file_mutex
);
745 kernfs_put_open_node(kn
, NULL
);
749 * Kernfs attribute files are pollable. The idea is that you read
750 * the content and then you use 'poll' or 'select' to wait for
751 * the content to change. When the content changes (assuming the
752 * manager for the kobject supports notification), poll will
753 * return POLLERR|POLLPRI, and select will return the fd whether
754 * it is waiting for read, write, or exceptions.
755 * Once poll/select indicates that the value has changed, you
756 * need to close and re-open the file, or seek to 0 and read again.
757 * Reminder: this only works for attributes which actively support
758 * it, and it is not possible to test an attribute from userspace
759 * to see if it supports poll (Neither 'poll' nor 'select' return
760 * an appropriate error code). When in doubt, set a suitable timeout value.
762 static unsigned int kernfs_fop_poll(struct file
*filp
, poll_table
*wait
)
764 struct kernfs_open_file
*of
= kernfs_of(filp
);
765 struct kernfs_node
*kn
= filp
->f_path
.dentry
->d_fsdata
;
766 struct kernfs_open_node
*on
= kn
->attr
.open
;
768 /* need parent for the kobj, grab both */
769 if (!kernfs_get_active(kn
))
772 poll_wait(filp
, &on
->poll
, wait
);
774 kernfs_put_active(kn
);
776 if (of
->event
!= atomic_read(&on
->event
))
779 return DEFAULT_POLLMASK
;
782 return DEFAULT_POLLMASK
|POLLERR
|POLLPRI
;
786 * kernfs_notify - notify a kernfs file
787 * @kn: file to notify
789 * Notify @kn such that poll(2) on @kn wakes up.
791 void kernfs_notify(struct kernfs_node
*kn
)
793 struct kernfs_open_node
*on
;
796 spin_lock_irqsave(&kernfs_open_node_lock
, flags
);
798 if (!WARN_ON(kernfs_type(kn
) != KERNFS_FILE
)) {
801 atomic_inc(&on
->event
);
802 wake_up_interruptible(&on
->poll
);
806 spin_unlock_irqrestore(&kernfs_open_node_lock
, flags
);
808 EXPORT_SYMBOL_GPL(kernfs_notify
);
810 const struct file_operations kernfs_file_fops
= {
811 .read
= kernfs_fop_read
,
812 .write
= kernfs_fop_write
,
813 .llseek
= generic_file_llseek
,
814 .mmap
= kernfs_fop_mmap
,
815 .open
= kernfs_fop_open
,
816 .release
= kernfs_fop_release
,
817 .poll
= kernfs_fop_poll
,
821 * __kernfs_create_file - kernfs internal function to create a file
822 * @parent: directory to create the file in
823 * @name: name of the file
824 * @mode: mode of the file
825 * @size: size of the file
826 * @ops: kernfs operations for the file
827 * @priv: private data for the file
828 * @ns: optional namespace tag of the file
829 * @static_name: don't copy file name
830 * @key: lockdep key for the file's active_ref, %NULL to disable lockdep
832 * Returns the created node on success, ERR_PTR() value on error.
834 struct kernfs_node
*__kernfs_create_file(struct kernfs_node
*parent
,
836 umode_t mode
, loff_t size
,
837 const struct kernfs_ops
*ops
,
838 void *priv
, const void *ns
,
840 struct lock_class_key
*key
)
842 struct kernfs_node
*kn
;
848 flags
|= KERNFS_STATIC_NAME
;
850 kn
= kernfs_new_node(parent
, name
, (mode
& S_IALLUGO
) | S_IFREG
, flags
);
852 return ERR_PTR(-ENOMEM
);
855 kn
->attr
.size
= size
;
859 #ifdef CONFIG_DEBUG_LOCK_ALLOC
861 lockdep_init_map(&kn
->dep_map
, "s_active", key
, 0);
862 kn
->flags
|= KERNFS_LOCKDEP
;
867 * kn->attr.ops is accesible only while holding active ref. We
868 * need to know whether some ops are implemented outside active
869 * ref. Cache their existence in flags.
872 kn
->flags
|= KERNFS_HAS_SEQ_SHOW
;
874 kn
->flags
|= KERNFS_HAS_MMAP
;
876 rc
= kernfs_add_one(kn
);