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 sysfs_open_file for each open file and one sysfs_open_dirent
22 * for each sysfs_dirent with one or more open files.
24 * sysfs_dirent->s_attr.open points to sysfs_open_dirent. s_attr.open is
25 * protected by sysfs_open_dirent_lock.
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.
31 static DEFINE_SPINLOCK(sysfs_open_dirent_lock);
32 static DEFINE_MUTEX(sysfs_open_file_mutex);
34 struct sysfs_open_dirent {
37 wait_queue_head_t poll;
38 struct list_head files; /* goes through sysfs_open_file.list */
41 static struct sysfs_open_file *sysfs_of(struct file *file)
43 return ((struct seq_file *)file->private_data)->private;
47 * Determine the kernfs_ops for the given sysfs_dirent. This function must
48 * be called while holding an active reference.
50 static const struct kernfs_ops *kernfs_ops(struct sysfs_dirent *sd)
52 if (sd->s_flags & SYSFS_FLAG_LOCKDEP)
53 lockdep_assert_held(sd);
54 return sd->s_attr.ops;
57 static void *kernfs_seq_start(struct seq_file *sf, loff_t *ppos)
59 struct sysfs_open_file *of = sf->private;
60 const struct kernfs_ops *ops;
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.
66 mutex_lock(&of->mutex);
67 if (!sysfs_get_active(of->sd))
68 return ERR_PTR(-ENODEV);
70 ops = kernfs_ops(of->sd);
72 return ops->seq_start(sf, ppos);
75 * The same behavior and code as single_open(). Returns
76 * !NULL if pos is at the beginning; otherwise, NULL.
82 static void *kernfs_seq_next(struct seq_file *sf, void *v, loff_t *ppos)
84 struct sysfs_open_file *of = sf->private;
85 const struct kernfs_ops *ops = kernfs_ops(of->sd);
88 return ops->seq_next(sf, v, ppos);
91 * The same behavior and code as single_open(), always
92 * terminate after the initial read.
99 static void kernfs_seq_stop(struct seq_file *sf, void *v)
101 struct sysfs_open_file *of = sf->private;
102 const struct kernfs_ops *ops = kernfs_ops(of->sd);
105 ops->seq_stop(sf, v);
107 sysfs_put_active(of->sd);
108 mutex_unlock(&of->mutex);
111 static int kernfs_seq_show(struct seq_file *sf, void *v)
113 struct sysfs_open_file *of = sf->private;
115 of->event = atomic_read(&of->sd->s_attr.open->event);
117 return of->sd->s_attr.ops->seq_show(sf, v);
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,
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
133 static ssize_t kernfs_file_direct_read(struct sysfs_open_file *of,
134 char __user *user_buf, size_t count,
137 ssize_t len = min_t(size_t, count, PAGE_SIZE);
138 const struct kernfs_ops *ops;
141 buf = kmalloc(len, GFP_KERNEL);
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.
149 mutex_lock(&of->mutex);
150 if (!sysfs_get_active(of->sd)) {
152 mutex_unlock(&of->mutex);
156 ops = kernfs_ops(of->sd);
158 len = ops->read(of, buf, len, *ppos);
162 sysfs_put_active(of->sd);
163 mutex_unlock(&of->mutex);
168 if (copy_to_user(user_buf, buf, len)) {
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
187 static ssize_t kernfs_file_read(struct file *file, char __user *user_buf,
188 size_t count, loff_t *ppos)
190 struct sysfs_open_file *of = sysfs_of(file);
192 if (of->sd->s_flags & SYSFS_FLAG_HAS_SEQ_SHOW)
193 return seq_read(file, user_buf, count, ppos);
195 return kernfs_file_direct_read(of, user_buf, count, ppos);
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
205 * Copy data in from userland and pass it to the matching kernfs write
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
214 static ssize_t kernfs_file_write(struct file *file, const char __user *user_buf,
215 size_t count, loff_t *ppos)
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;
222 buf = kmalloc(len + 1, GFP_KERNEL);
226 if (copy_from_user(buf, user_buf, len)) {
230 buf[len] = '\0'; /* guarantee string termination */
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.
236 mutex_lock(&of->mutex);
237 if (!sysfs_get_active(of->sd)) {
238 mutex_unlock(&of->mutex);
243 ops = kernfs_ops(of->sd);
245 len = ops->write(of, buf, len, *ppos);
249 sysfs_put_active(of->sd);
250 mutex_unlock(&of->mutex);
259 static void kernfs_vma_open(struct vm_area_struct *vma)
261 struct file *file = vma->vm_file;
262 struct sysfs_open_file *of = sysfs_of(file);
267 if (!sysfs_get_active(of->sd))
270 if (of->vm_ops->open)
271 of->vm_ops->open(vma);
273 sysfs_put_active(of->sd);
276 static int kernfs_vma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
278 struct file *file = vma->vm_file;
279 struct sysfs_open_file *of = sysfs_of(file);
283 return VM_FAULT_SIGBUS;
285 if (!sysfs_get_active(of->sd))
286 return VM_FAULT_SIGBUS;
288 ret = VM_FAULT_SIGBUS;
289 if (of->vm_ops->fault)
290 ret = of->vm_ops->fault(vma, vmf);
292 sysfs_put_active(of->sd);
296 static int kernfs_vma_page_mkwrite(struct vm_area_struct *vma,
297 struct vm_fault *vmf)
299 struct file *file = vma->vm_file;
300 struct sysfs_open_file *of = sysfs_of(file);
304 return VM_FAULT_SIGBUS;
306 if (!sysfs_get_active(of->sd))
307 return VM_FAULT_SIGBUS;
310 if (of->vm_ops->page_mkwrite)
311 ret = of->vm_ops->page_mkwrite(vma, vmf);
313 file_update_time(file);
315 sysfs_put_active(of->sd);
319 static int kernfs_vma_access(struct vm_area_struct *vma, unsigned long addr,
320 void *buf, int len, int write)
322 struct file *file = vma->vm_file;
323 struct sysfs_open_file *of = sysfs_of(file);
329 if (!sysfs_get_active(of->sd))
333 if (of->vm_ops->access)
334 ret = of->vm_ops->access(vma, addr, buf, len, write);
336 sysfs_put_active(of->sd);
341 static int kernfs_vma_set_policy(struct vm_area_struct *vma,
342 struct mempolicy *new)
344 struct file *file = vma->vm_file;
345 struct sysfs_open_file *of = sysfs_of(file);
351 if (!sysfs_get_active(of->sd))
355 if (of->vm_ops->set_policy)
356 ret = of->vm_ops->set_policy(vma, new);
358 sysfs_put_active(of->sd);
362 static struct mempolicy *kernfs_vma_get_policy(struct vm_area_struct *vma,
365 struct file *file = vma->vm_file;
366 struct sysfs_open_file *of = sysfs_of(file);
367 struct mempolicy *pol;
370 return vma->vm_policy;
372 if (!sysfs_get_active(of->sd))
373 return vma->vm_policy;
375 pol = vma->vm_policy;
376 if (of->vm_ops->get_policy)
377 pol = of->vm_ops->get_policy(vma, addr);
379 sysfs_put_active(of->sd);
383 static int kernfs_vma_migrate(struct vm_area_struct *vma,
384 const nodemask_t *from, const nodemask_t *to,
387 struct file *file = vma->vm_file;
388 struct sysfs_open_file *of = sysfs_of(file);
394 if (!sysfs_get_active(of->sd))
398 if (of->vm_ops->migrate)
399 ret = of->vm_ops->migrate(vma, from, to, flags);
401 sysfs_put_active(of->sd);
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,
412 .set_policy = kernfs_vma_set_policy,
413 .get_policy = kernfs_vma_get_policy,
414 .migrate = kernfs_vma_migrate,
418 static int kernfs_file_mmap(struct file *file, struct vm_area_struct *vma)
420 struct sysfs_open_file *of = sysfs_of(file);
421 const struct kernfs_ops *ops;
424 mutex_lock(&of->mutex);
427 if (!sysfs_get_active(of->sd))
430 ops = kernfs_ops(of->sd);
432 rc = ops->mmap(of, vma);
437 * PowerPC's pci_mmap of legacy_mem uses shmem_zero_setup()
438 * to satisfy versions of X which crash if the mmap fails: that
439 * substitutes a new vm_file, and we don't then want bin_vm_ops.
441 if (vma->vm_file != file)
445 if (of->mmapped && of->vm_ops != vma->vm_ops)
449 * It is not possible to successfully wrap close.
450 * So error if someone is trying to use close.
453 if (vma->vm_ops && vma->vm_ops->close)
458 of->vm_ops = vma->vm_ops;
459 vma->vm_ops = &kernfs_vm_ops;
461 sysfs_put_active(of->sd);
463 mutex_unlock(&of->mutex);
469 * sysfs_get_open_dirent - get or create sysfs_open_dirent
470 * @sd: target sysfs_dirent
471 * @of: sysfs_open_file for this instance of open
473 * If @sd->s_attr.open exists, increment its reference count;
474 * otherwise, create one. @of is chained to the files list.
477 * Kernel thread context (may sleep).
480 * 0 on success, -errno on failure.
482 static int sysfs_get_open_dirent(struct sysfs_dirent *sd,
483 struct sysfs_open_file *of)
485 struct sysfs_open_dirent *od, *new_od = NULL;
488 mutex_lock(&sysfs_open_file_mutex);
489 spin_lock_irq(&sysfs_open_dirent_lock);
491 if (!sd->s_attr.open && new_od) {
492 sd->s_attr.open = new_od;
496 od = sd->s_attr.open;
498 atomic_inc(&od->refcnt);
499 list_add_tail(&of->list, &od->files);
502 spin_unlock_irq(&sysfs_open_dirent_lock);
503 mutex_unlock(&sysfs_open_file_mutex);
510 /* not there, initialize a new one and retry */
511 new_od = kmalloc(sizeof(*new_od), GFP_KERNEL);
515 atomic_set(&new_od->refcnt, 0);
516 atomic_set(&new_od->event, 1);
517 init_waitqueue_head(&new_od->poll);
518 INIT_LIST_HEAD(&new_od->files);
523 * sysfs_put_open_dirent - put sysfs_open_dirent
524 * @sd: target sysfs_dirent
525 * @of: associated sysfs_open_file
527 * Put @sd->s_attr.open and unlink @of from the files list. If
528 * reference count reaches zero, disassociate and free it.
533 static void sysfs_put_open_dirent(struct sysfs_dirent *sd,
534 struct sysfs_open_file *of)
536 struct sysfs_open_dirent *od = sd->s_attr.open;
539 mutex_lock(&sysfs_open_file_mutex);
540 spin_lock_irqsave(&sysfs_open_dirent_lock, flags);
545 if (atomic_dec_and_test(&od->refcnt))
546 sd->s_attr.open = NULL;
550 spin_unlock_irqrestore(&sysfs_open_dirent_lock, flags);
551 mutex_unlock(&sysfs_open_file_mutex);
556 static int kernfs_file_open(struct inode *inode, struct file *file)
558 struct sysfs_dirent *attr_sd = file->f_path.dentry->d_fsdata;
559 const struct kernfs_ops *ops;
560 struct sysfs_open_file *of;
561 bool has_read, has_write, has_mmap;
564 if (!sysfs_get_active(attr_sd))
567 ops = kernfs_ops(attr_sd);
569 has_read = ops->seq_show || ops->read || ops->mmap;
570 has_write = ops->write || ops->mmap;
571 has_mmap = ops->mmap;
573 /* check perms and supported operations */
574 if ((file->f_mode & FMODE_WRITE) &&
575 (!(inode->i_mode & S_IWUGO) || !has_write))
578 if ((file->f_mode & FMODE_READ) &&
579 (!(inode->i_mode & S_IRUGO) || !has_read))
582 /* allocate a sysfs_open_file for the file */
584 of = kzalloc(sizeof(struct sysfs_open_file), GFP_KERNEL);
589 * The following is done to give a different lockdep key to
590 * @of->mutex for files which implement mmap. This is a rather
591 * crude way to avoid false positive lockdep warning around
592 * mm->mmap_sem - mmap nests @of->mutex under mm->mmap_sem and
593 * reading /sys/block/sda/trace/act_mask grabs sr_mutex, under
594 * which mm->mmap_sem nests, while holding @of->mutex. As each
595 * open file has a separate mutex, it's okay as long as those don't
596 * happen on the same file. At this point, we can't easily give
597 * each file a separate locking class. Let's differentiate on
598 * whether the file has mmap or not for now.
601 mutex_init(&of->mutex);
603 mutex_init(&of->mutex);
609 * Always instantiate seq_file even if read access doesn't use
610 * seq_file or is not requested. This unifies private data access
611 * and readable regular files are the vast majority anyway.
614 error = seq_open(file, &kernfs_seq_ops);
616 error = seq_open(file, NULL);
620 ((struct seq_file *)file->private_data)->private = of;
622 /* seq_file clears PWRITE unconditionally, restore it if WRITE */
623 if (file->f_mode & FMODE_WRITE)
624 file->f_mode |= FMODE_PWRITE;
626 /* make sure we have open dirent struct */
627 error = sysfs_get_open_dirent(attr_sd, of);
631 /* open succeeded, put active references */
632 sysfs_put_active(attr_sd);
636 seq_release(inode, file);
640 sysfs_put_active(attr_sd);
644 static int kernfs_file_release(struct inode *inode, struct file *filp)
646 struct sysfs_dirent *sd = filp->f_path.dentry->d_fsdata;
647 struct sysfs_open_file *of = sysfs_of(filp);
649 sysfs_put_open_dirent(sd, of);
650 seq_release(inode, filp);
656 void sysfs_unmap_bin_file(struct sysfs_dirent *sd)
658 struct sysfs_open_dirent *od;
659 struct sysfs_open_file *of;
661 if (!(sd->s_flags & SYSFS_FLAG_HAS_MMAP))
664 spin_lock_irq(&sysfs_open_dirent_lock);
665 od = sd->s_attr.open;
667 atomic_inc(&od->refcnt);
668 spin_unlock_irq(&sysfs_open_dirent_lock);
672 mutex_lock(&sysfs_open_file_mutex);
673 list_for_each_entry(of, &od->files, list) {
674 struct inode *inode = file_inode(of->file);
675 unmap_mapping_range(inode->i_mapping, 0, 0, 1);
677 mutex_unlock(&sysfs_open_file_mutex);
679 sysfs_put_open_dirent(sd, NULL);
682 /* Sysfs attribute files are pollable. The idea is that you read
683 * the content and then you use 'poll' or 'select' to wait for
684 * the content to change. When the content changes (assuming the
685 * manager for the kobject supports notification), poll will
686 * return POLLERR|POLLPRI, and select will return the fd whether
687 * it is waiting for read, write, or exceptions.
688 * Once poll/select indicates that the value has changed, you
689 * need to close and re-open the file, or seek to 0 and read again.
690 * Reminder: this only works for attributes which actively support
691 * it, and it is not possible to test an attribute from userspace
692 * to see if it supports poll (Neither 'poll' nor 'select' return
693 * an appropriate error code). When in doubt, set a suitable timeout value.
695 static unsigned int kernfs_file_poll(struct file *filp, poll_table *wait)
697 struct sysfs_open_file *of = sysfs_of(filp);
698 struct sysfs_dirent *attr_sd = filp->f_path.dentry->d_fsdata;
699 struct sysfs_open_dirent *od = attr_sd->s_attr.open;
701 /* need parent for the kobj, grab both */
702 if (!sysfs_get_active(attr_sd))
705 poll_wait(filp, &od->poll, wait);
707 sysfs_put_active(attr_sd);
709 if (of->event != atomic_read(&od->event))
712 return DEFAULT_POLLMASK;
715 return DEFAULT_POLLMASK|POLLERR|POLLPRI;
719 * kernfs_notify - notify a kernfs file
720 * @sd: file to notify
722 * Notify @sd such that poll(2) on @sd wakes up.
724 void kernfs_notify(struct sysfs_dirent *sd)
726 struct sysfs_open_dirent *od;
729 spin_lock_irqsave(&sysfs_open_dirent_lock, flags);
731 if (!WARN_ON(sysfs_type(sd) != SYSFS_KOBJ_ATTR)) {
732 od = sd->s_attr.open;
734 atomic_inc(&od->event);
735 wake_up_interruptible(&od->poll);
739 spin_unlock_irqrestore(&sysfs_open_dirent_lock, flags);
741 EXPORT_SYMBOL_GPL(kernfs_notify);
743 const struct file_operations kernfs_file_operations = {
744 .read = kernfs_file_read,
745 .write = kernfs_file_write,
746 .llseek = generic_file_llseek,
747 .mmap = kernfs_file_mmap,
748 .open = kernfs_file_open,
749 .release = kernfs_file_release,
750 .poll = kernfs_file_poll,
754 * kernfs_create_file_ns_key - create a file
755 * @parent: directory to create the file in
756 * @name: name of the file
757 * @mode: mode of the file
758 * @size: size of the file
759 * @ops: kernfs operations for the file
760 * @priv: private data for the file
761 * @ns: optional namespace tag of the file
762 * @key: lockdep key for the file's active_ref, %NULL to disable lockdep
764 * Returns the created node on success, ERR_PTR() value on error.
766 struct sysfs_dirent *kernfs_create_file_ns_key(struct sysfs_dirent *parent,
768 umode_t mode, loff_t size,
769 const struct kernfs_ops *ops,
770 void *priv, const void *ns,
771 struct lock_class_key *key)
773 struct sysfs_addrm_cxt acxt;
774 struct sysfs_dirent *sd;
777 sd = sysfs_new_dirent(kernfs_root(parent), name,
778 (mode & S_IALLUGO) | S_IFREG, SYSFS_KOBJ_ATTR);
780 return ERR_PTR(-ENOMEM);
782 sd->s_attr.ops = ops;
783 sd->s_attr.size = size;
787 #ifdef CONFIG_DEBUG_LOCK_ALLOC
789 lockdep_init_map(&sd->dep_map, "s_active", key, 0);
790 sd->s_flags |= SYSFS_FLAG_LOCKDEP;
795 * sd->s_attr.ops is accesible only while holding active ref. We
796 * need to know whether some ops are implemented outside active
797 * ref. Cache their existence in flags.
800 sd->s_flags |= SYSFS_FLAG_HAS_SEQ_SHOW;
802 sd->s_flags |= SYSFS_FLAG_HAS_MMAP;
804 sysfs_addrm_start(&acxt);
805 rc = sysfs_add_one(&acxt, sd, parent);
806 sysfs_addrm_finish(&acxt);