1 // SPDX-License-Identifier: GPL-2.0-only
3 * fs/kernfs/file.c - kernfs file implementation
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>
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>
19 #include "kernfs-internal.h"
21 struct kernfs_open_node {
22 struct rcu_head rcu_head;
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;
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
38 #define KERNFS_NOTIFY_EOL ((void *)&kernfs_notify_list)
40 static DEFINE_SPINLOCK(kernfs_notify_lock);
41 static struct kernfs_node *kernfs_notify_list = KERNFS_NOTIFY_EOL;
43 static inline struct mutex *kernfs_open_file_mutex_ptr(struct kernfs_node *kn)
45 int idx = hash_ptr(kn, NR_KERNFS_LOCK_BITS);
47 return &kernfs_locks->open_file_mutex[idx];
50 static inline struct mutex *kernfs_open_file_mutex_lock(struct kernfs_node *kn)
54 lock = kernfs_open_file_mutex_ptr(kn);
62 * of_on - Get the kernfs_open_node of the specified kernfs_open_file
63 * @of: target kernfs_open_file
65 * Return: the kernfs_open_node of the kernfs_open_file
67 static struct kernfs_open_node *of_on(struct kernfs_open_file *of)
69 return rcu_dereference_protected(of->kn->attr.open,
70 !list_empty(&of->list));
74 * kernfs_deref_open_node_locked - Get kernfs_open_node corresponding to @kn
76 * @kn: target kernfs_node.
78 * Fetch and return ->attr.open of @kn when caller holds the
79 * kernfs_open_file_mutex_ptr(kn).
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.
86 * The caller needs to make sure that kernfs_open_file_mutex is held.
88 * Return: @kn->attr.open when kernfs_open_file_mutex is held.
90 static struct kernfs_open_node *
91 kernfs_deref_open_node_locked(struct kernfs_node *kn)
93 return rcu_dereference_protected(kn->attr.open,
94 lockdep_is_held(kernfs_open_file_mutex_ptr(kn)));
97 static struct kernfs_open_file *kernfs_of(struct file *file)
99 return ((struct seq_file *)file->private_data)->private;
103 * Determine the kernfs_ops for the given kernfs_node. This function must
104 * be called while holding an active reference.
106 static const struct kernfs_ops *kernfs_ops(struct kernfs_node *kn)
108 if (kn->flags & KERNFS_LOCKDEP)
109 lockdep_assert_held(kn);
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.
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).
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.
135 static void kernfs_seq_stop_active(struct seq_file *sf, void *v)
137 struct kernfs_open_file *of = sf->private;
138 const struct kernfs_ops *ops = kernfs_ops(of->kn);
141 ops->seq_stop(sf, v);
142 kernfs_put_active(of->kn);
145 static void *kernfs_seq_start(struct seq_file *sf, loff_t *ppos)
147 struct kernfs_open_file *of = sf->private;
148 const struct kernfs_ops *ops;
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.
154 mutex_lock(&of->mutex);
155 if (!kernfs_get_active(of->kn))
156 return ERR_PTR(-ENODEV);
158 ops = kernfs_ops(of->kn);
159 if (ops->seq_start) {
160 void *next = ops->seq_start(sf, ppos);
161 /* see the comment above kernfs_seq_stop_active() */
162 if (next == ERR_PTR(-ENODEV))
163 kernfs_seq_stop_active(sf, next);
166 return single_start(sf, ppos);
169 static void *kernfs_seq_next(struct seq_file *sf, void *v, loff_t *ppos)
171 struct kernfs_open_file *of = sf->private;
172 const struct kernfs_ops *ops = kernfs_ops(of->kn);
175 void *next = ops->seq_next(sf, v, ppos);
176 /* see the comment above kernfs_seq_stop_active() */
177 if (next == ERR_PTR(-ENODEV))
178 kernfs_seq_stop_active(sf, next);
182 * The same behavior and code as single_open(), always
183 * terminate after the initial read.
190 static void kernfs_seq_stop(struct seq_file *sf, void *v)
192 struct kernfs_open_file *of = sf->private;
194 if (v != ERR_PTR(-ENODEV))
195 kernfs_seq_stop_active(sf, v);
196 mutex_unlock(&of->mutex);
199 static int kernfs_seq_show(struct seq_file *sf, void *v)
201 struct kernfs_open_file *of = sf->private;
203 of->event = atomic_read(&of_on(of)->event);
205 return of->kn->attr.ops->seq_show(sf, v);
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,
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
221 static ssize_t kernfs_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
223 struct kernfs_open_file *of = kernfs_of(iocb->ki_filp);
224 ssize_t len = min_t(size_t, iov_iter_count(iter), PAGE_SIZE);
225 const struct kernfs_ops *ops;
228 buf = of->prealloc_buf;
230 mutex_lock(&of->prealloc_mutex);
232 buf = kmalloc(len, GFP_KERNEL);
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.
240 mutex_lock(&of->mutex);
241 if (!kernfs_get_active(of->kn)) {
243 mutex_unlock(&of->mutex);
247 of->event = atomic_read(&of_on(of)->event);
249 ops = kernfs_ops(of->kn);
251 len = ops->read(of, buf, len, iocb->ki_pos);
255 kernfs_put_active(of->kn);
256 mutex_unlock(&of->mutex);
261 if (copy_to_iter(buf, len, iter) != len) {
269 if (buf == of->prealloc_buf)
270 mutex_unlock(&of->prealloc_mutex);
276 static ssize_t kernfs_fop_read_iter(struct kiocb *iocb, struct iov_iter *iter)
278 if (kernfs_of(iocb->ki_filp)->kn->flags & KERNFS_HAS_SEQ_SHOW)
279 return seq_read_iter(iocb, iter);
280 return kernfs_file_read_iter(iocb, iter);
284 * Copy data in from userland and pass it to the matching kernfs write
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
293 static ssize_t kernfs_fop_write_iter(struct kiocb *iocb, struct iov_iter *iter)
295 struct kernfs_open_file *of = kernfs_of(iocb->ki_filp);
296 ssize_t len = iov_iter_count(iter);
297 const struct kernfs_ops *ops;
300 if (of->atomic_write_len) {
301 if (len > of->atomic_write_len)
304 len = min_t(size_t, len, PAGE_SIZE);
307 buf = of->prealloc_buf;
309 mutex_lock(&of->prealloc_mutex);
311 buf = kmalloc(len + 1, GFP_KERNEL);
315 if (copy_from_iter(buf, len, iter) != len) {
319 buf[len] = '\0'; /* guarantee string termination */
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.
325 mutex_lock(&of->mutex);
326 if (!kernfs_get_active(of->kn)) {
327 mutex_unlock(&of->mutex);
332 ops = kernfs_ops(of->kn);
334 len = ops->write(of, buf, len, iocb->ki_pos);
338 kernfs_put_active(of->kn);
339 mutex_unlock(&of->mutex);
345 if (buf == of->prealloc_buf)
346 mutex_unlock(&of->prealloc_mutex);
352 static void kernfs_vma_open(struct vm_area_struct *vma)
354 struct file *file = vma->vm_file;
355 struct kernfs_open_file *of = kernfs_of(file);
360 if (!kernfs_get_active(of->kn))
363 if (of->vm_ops->open)
364 of->vm_ops->open(vma);
366 kernfs_put_active(of->kn);
369 static vm_fault_t kernfs_vma_fault(struct vm_fault *vmf)
371 struct file *file = vmf->vma->vm_file;
372 struct kernfs_open_file *of = kernfs_of(file);
376 return VM_FAULT_SIGBUS;
378 if (!kernfs_get_active(of->kn))
379 return VM_FAULT_SIGBUS;
381 ret = VM_FAULT_SIGBUS;
382 if (of->vm_ops->fault)
383 ret = of->vm_ops->fault(vmf);
385 kernfs_put_active(of->kn);
389 static vm_fault_t kernfs_vma_page_mkwrite(struct vm_fault *vmf)
391 struct file *file = vmf->vma->vm_file;
392 struct kernfs_open_file *of = kernfs_of(file);
396 return VM_FAULT_SIGBUS;
398 if (!kernfs_get_active(of->kn))
399 return VM_FAULT_SIGBUS;
402 if (of->vm_ops->page_mkwrite)
403 ret = of->vm_ops->page_mkwrite(vmf);
405 file_update_time(file);
407 kernfs_put_active(of->kn);
411 static int kernfs_vma_access(struct vm_area_struct *vma, unsigned long addr,
412 void *buf, int len, int write)
414 struct file *file = vma->vm_file;
415 struct kernfs_open_file *of = kernfs_of(file);
421 if (!kernfs_get_active(of->kn))
425 if (of->vm_ops->access)
426 ret = of->vm_ops->access(vma, addr, buf, len, write);
428 kernfs_put_active(of->kn);
433 static int kernfs_vma_set_policy(struct vm_area_struct *vma,
434 struct mempolicy *new)
436 struct file *file = vma->vm_file;
437 struct kernfs_open_file *of = kernfs_of(file);
443 if (!kernfs_get_active(of->kn))
447 if (of->vm_ops->set_policy)
448 ret = of->vm_ops->set_policy(vma, new);
450 kernfs_put_active(of->kn);
454 static struct mempolicy *kernfs_vma_get_policy(struct vm_area_struct *vma,
457 struct file *file = vma->vm_file;
458 struct kernfs_open_file *of = kernfs_of(file);
459 struct mempolicy *pol;
462 return vma->vm_policy;
464 if (!kernfs_get_active(of->kn))
465 return vma->vm_policy;
467 pol = vma->vm_policy;
468 if (of->vm_ops->get_policy)
469 pol = of->vm_ops->get_policy(vma, addr);
471 kernfs_put_active(of->kn);
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,
483 .set_policy = kernfs_vma_set_policy,
484 .get_policy = kernfs_vma_get_policy,
488 static int kernfs_fop_mmap(struct file *file, struct vm_area_struct *vma)
490 struct kernfs_open_file *of = kernfs_of(file);
491 const struct kernfs_ops *ops;
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.
501 if (!(of->kn->flags & KERNFS_HAS_MMAP))
504 mutex_lock(&of->mutex);
507 if (!kernfs_get_active(of->kn))
510 ops = kernfs_ops(of->kn);
511 rc = ops->mmap(of, vma);
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.
520 if (vma->vm_file != file)
524 if (of->mmapped && of->vm_ops != vma->vm_ops)
528 * It is not possible to successfully wrap close.
529 * So error if someone is trying to use close.
531 if (vma->vm_ops && vma->vm_ops->close)
536 of_on(of)->nr_mmapped++;
537 of->vm_ops = vma->vm_ops;
538 vma->vm_ops = &kernfs_vm_ops;
540 kernfs_put_active(of->kn);
542 mutex_unlock(&of->mutex);
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
552 * If @kn->attr.open exists, increment its reference count; otherwise,
553 * create one. @of is chained to the files list.
556 * Kernel thread context (may sleep).
559 * %0 on success, -errno on failure.
561 static int kernfs_get_open_node(struct kernfs_node *kn,
562 struct kernfs_open_file *of)
564 struct kernfs_open_node *on;
567 mutex = kernfs_open_file_mutex_lock(kn);
568 on = kernfs_deref_open_node_locked(kn);
571 /* not there, initialize a new one */
572 on = kzalloc(sizeof(*on), GFP_KERNEL);
577 atomic_set(&on->event, 1);
578 init_waitqueue_head(&on->poll);
579 INIT_LIST_HEAD(&on->files);
580 rcu_assign_pointer(kn->attr.open, on);
583 list_add_tail(&of->list, &on->files);
584 if (kn->flags & KERNFS_HAS_RELEASE)
592 * kernfs_unlink_open_file - Unlink @of from @kn.
594 * @kn: target kernfs_node
595 * @of: associated kernfs_open_file
596 * @open_failed: ->open() failed, cancel ->release()
598 * Unlink @of from list of @kn's associated open files. If list of
599 * associated open files becomes empty, disassociate and free
605 static void kernfs_unlink_open_file(struct kernfs_node *kn,
606 struct kernfs_open_file *of,
609 struct kernfs_open_node *on;
612 mutex = kernfs_open_file_mutex_lock(kn);
614 on = kernfs_deref_open_node_locked(kn);
621 if (kn->flags & KERNFS_HAS_RELEASE) {
622 WARN_ON_ONCE(of->released == open_failed);
631 if (list_empty(&on->files)) {
632 rcu_assign_pointer(kn->attr.open, NULL);
633 kfree_rcu(on, rcu_head);
639 static int kernfs_fop_open(struct inode *inode, struct file *file)
641 struct kernfs_node *kn = inode->i_private;
642 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;
648 if (!kernfs_get_active(kn))
651 ops = kernfs_ops(kn);
653 has_read = ops->seq_show || ops->read || ops->mmap;
654 has_write = ops->write || ops->mmap;
655 has_mmap = ops->mmap;
657 /* see the flag definition for details */
658 if (root->flags & KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK) {
659 if ((file->f_mode & FMODE_WRITE) &&
660 (!(inode->i_mode & S_IWUGO) || !has_write))
663 if ((file->f_mode & FMODE_READ) &&
664 (!(inode->i_mode & S_IRUGO) || !has_read))
668 /* allocate a kernfs_open_file for the file */
670 of = kzalloc(sizeof(struct kernfs_open_file), GFP_KERNEL);
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.
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.
690 mutex_init(&of->mutex);
692 mutex_init(&of->mutex);
698 * Write path needs to atomic_write_len outside active reference.
699 * Cache it in open_file. See kernfs_fop_write_iter() for details.
701 of->atomic_write_len = ops->atomic_write_len;
705 * ->seq_show is incompatible with ->prealloc,
706 * as seq_read does its own allocation.
707 * ->read must be used instead.
709 if (ops->prealloc && ops->seq_show)
712 int len = of->atomic_write_len ?: PAGE_SIZE;
713 of->prealloc_buf = kmalloc(len + 1, GFP_KERNEL);
715 if (!of->prealloc_buf)
717 mutex_init(&of->prealloc_mutex);
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.
726 error = seq_open(file, &kernfs_seq_ops);
728 error = seq_open(file, NULL);
732 of->seq_file = file->private_data;
733 of->seq_file->private = of;
735 /* seq_file clears PWRITE unconditionally, restore it if WRITE */
736 if (file->f_mode & FMODE_WRITE)
737 file->f_mode |= FMODE_PWRITE;
739 /* make sure we have open node struct */
740 error = kernfs_get_open_node(kn, of);
742 goto err_seq_release;
745 /* nobody has access to @of yet, skip @of->mutex */
746 error = ops->open(of);
751 /* open succeeded, put active references */
752 kernfs_put_active(kn);
756 kernfs_unlink_open_file(kn, of, true);
758 seq_release(inode, file);
760 kfree(of->prealloc_buf);
763 kernfs_put_active(kn);
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)
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.
778 lockdep_assert_held(kernfs_open_file_mutex_ptr(kn));
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().
786 kn->attr.ops->release(of);
788 of_on(of)->nr_to_release--;
792 static int kernfs_fop_release(struct inode *inode, struct file *filp)
794 struct kernfs_node *kn = inode->i_private;
795 struct kernfs_open_file *of = kernfs_of(filp);
797 if (kn->flags & KERNFS_HAS_RELEASE) {
800 mutex = kernfs_open_file_mutex_lock(kn);
801 kernfs_release_file(kn, of);
805 kernfs_unlink_open_file(kn, of, false);
806 seq_release(inode, filp);
807 kfree(of->prealloc_buf);
813 bool kernfs_should_drain_open_files(struct kernfs_node *kn)
815 struct kernfs_open_node *on;
819 * @kn being deactivated guarantees that @kn->attr.open can't change
820 * beneath us making the lockless test below safe.
822 WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS);
825 on = rcu_dereference(kn->attr.open);
826 ret = on && (on->nr_mmapped || on->nr_to_release);
832 void kernfs_drain_open_files(struct kernfs_node *kn)
834 struct kernfs_open_node *on;
835 struct kernfs_open_file *of;
838 mutex = kernfs_open_file_mutex_lock(kn);
839 on = kernfs_deref_open_node_locked(kn);
845 list_for_each_entry(of, &on->files, list) {
846 struct inode *inode = file_inode(of->file);
849 unmap_mapping_range(inode->i_mapping, 0, 0, 1);
854 if (kn->flags & KERNFS_HAS_RELEASE)
855 kernfs_release_file(kn, of);
858 WARN_ON_ONCE(on->nr_mmapped || on->nr_to_release);
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.
876 __poll_t kernfs_generic_poll(struct kernfs_open_file *of, poll_table *wait)
878 struct kernfs_open_node *on = of_on(of);
880 poll_wait(of->file, &on->poll, wait);
882 if (of->event != atomic_read(&on->event))
883 return DEFAULT_POLLMASK|EPOLLERR|EPOLLPRI;
885 return DEFAULT_POLLMASK;
888 static __poll_t kernfs_fop_poll(struct file *filp, poll_table *wait)
890 struct kernfs_open_file *of = kernfs_of(filp);
891 struct kernfs_node *kn = kernfs_dentry_node(filp->f_path.dentry);
894 if (!kernfs_get_active(kn))
895 return DEFAULT_POLLMASK|EPOLLERR|EPOLLPRI;
897 if (kn->attr.ops->poll)
898 ret = kn->attr.ops->poll(of, wait);
900 ret = kernfs_generic_poll(of, wait);
902 kernfs_put_active(kn);
906 static void kernfs_notify_workfn(struct work_struct *work)
908 struct kernfs_node *kn;
909 struct kernfs_super_info *info;
910 struct kernfs_root *root;
912 /* pop one off the notify_list */
913 spin_lock_irq(&kernfs_notify_lock);
914 kn = kernfs_notify_list;
915 if (kn == KERNFS_NOTIFY_EOL) {
916 spin_unlock_irq(&kernfs_notify_lock);
919 kernfs_notify_list = kn->attr.notify_next;
920 kn->attr.notify_next = NULL;
921 spin_unlock_irq(&kernfs_notify_lock);
923 root = kernfs_root(kn);
926 down_read(&root->kernfs_supers_rwsem);
927 list_for_each_entry(info, &kernfs_root(kn)->supers, node) {
928 struct kernfs_node *parent;
929 struct inode *p_inode = NULL;
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.
939 inode = ilookup(info->sb, kernfs_ino(kn));
943 name = (struct qstr)QSTR_INIT(kn->name, strlen(kn->name));
944 parent = kernfs_get_parent(kn);
946 p_inode = ilookup(info->sb, kernfs_ino(parent));
948 fsnotify(FS_MODIFY | FS_EVENT_ON_CHILD,
949 inode, FSNOTIFY_EVENT_INODE,
950 p_inode, &name, inode, 0);
958 fsnotify_inode(inode, FS_MODIFY);
963 up_read(&root->kernfs_supers_rwsem);
969 * kernfs_notify - notify a kernfs file
970 * @kn: file to notify
972 * Notify @kn such that poll(2) on @kn wakes up. Maybe be called from any
975 void kernfs_notify(struct kernfs_node *kn)
977 static DECLARE_WORK(kernfs_notify_work, kernfs_notify_workfn);
979 struct kernfs_open_node *on;
981 if (WARN_ON(kernfs_type(kn) != KERNFS_FILE))
984 /* kick poll immediately */
986 on = rcu_dereference(kn->attr.open);
988 atomic_inc(&on->event);
989 wake_up_interruptible(&on->poll);
993 /* schedule work to kick fsnotify */
994 spin_lock_irqsave(&kernfs_notify_lock, flags);
995 if (!kn->attr.notify_next) {
997 kn->attr.notify_next = kernfs_notify_list;
998 kernfs_notify_list = kn;
999 schedule_work(&kernfs_notify_work);
1001 spin_unlock_irqrestore(&kernfs_notify_lock, flags);
1003 EXPORT_SYMBOL_GPL(kernfs_notify);
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,
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
1031 * Return: the created node on success, ERR_PTR() value on error.
1033 struct kernfs_node *__kernfs_create_file(struct kernfs_node *parent,
1035 umode_t mode, kuid_t uid, kgid_t gid,
1037 const struct kernfs_ops *ops,
1038 void *priv, const void *ns,
1039 struct lock_class_key *key)
1041 struct kernfs_node *kn;
1045 flags = KERNFS_FILE;
1047 kn = kernfs_new_node(parent, name, (mode & S_IALLUGO) | S_IFREG,
1050 return ERR_PTR(-ENOMEM);
1053 kn->attr.size = size;
1057 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1059 lockdep_init_map(&kn->dep_map, "kn->active", key, 0);
1060 kn->flags |= KERNFS_LOCKDEP;
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.
1070 kn->flags |= KERNFS_HAS_SEQ_SHOW;
1072 kn->flags |= KERNFS_HAS_MMAP;
1074 kn->flags |= KERNFS_HAS_RELEASE;
1076 rc = kernfs_add_one(kn);