blob: 78219d5644e90aacaf3aeb9fdfe2a234f4e31b84 [file] [log] [blame]
Tejun Heob8441ed2013-11-24 09:54:58 -05001/*
2 * fs/kernfs/file.c - kernfs file implementation
3 *
4 * Copyright (c) 2001-3 Patrick Mochel
5 * Copyright (c) 2007 SUSE Linux Products GmbH
6 * Copyright (c) 2007, 2013 Tejun Heo <tj@kernel.org>
7 *
8 * This file is released under the GPLv2.
9 */
Tejun Heo414985a2013-11-28 14:54:34 -050010
11#include <linux/fs.h>
12#include <linux/seq_file.h>
13#include <linux/slab.h>
14#include <linux/poll.h>
15#include <linux/pagemap.h>
Tejun Heo414985a2013-11-28 14:54:34 -050016#include <linux/sched.h>
Tejun Heod911d982014-04-09 11:07:31 -040017#include <linux/fsnotify.h>
Tejun Heo414985a2013-11-28 14:54:34 -050018
19#include "kernfs-internal.h"
20
21/*
Tejun Heoc525aad2013-12-11 14:11:55 -050022 * There's one kernfs_open_file for each open file and one kernfs_open_node
Tejun Heo324a56e2013-12-11 14:11:53 -050023 * for each kernfs_node with one or more open files.
Tejun Heo414985a2013-11-28 14:54:34 -050024 *
Tejun Heoc525aad2013-12-11 14:11:55 -050025 * kernfs_node->attr.open points to kernfs_open_node. attr.open is
26 * protected by kernfs_open_node_lock.
Tejun Heo414985a2013-11-28 14:54:34 -050027 *
28 * filp->private_data points to seq_file whose ->private points to
Tejun Heoc525aad2013-12-11 14:11:55 -050029 * kernfs_open_file. kernfs_open_files are chained at
30 * kernfs_open_node->files, which is protected by kernfs_open_file_mutex.
Tejun Heo414985a2013-11-28 14:54:34 -050031 */
Tejun Heoc525aad2013-12-11 14:11:55 -050032static DEFINE_SPINLOCK(kernfs_open_node_lock);
33static DEFINE_MUTEX(kernfs_open_file_mutex);
Tejun Heo414985a2013-11-28 14:54:34 -050034
Tejun Heoc525aad2013-12-11 14:11:55 -050035struct kernfs_open_node {
Tejun Heo414985a2013-11-28 14:54:34 -050036 atomic_t refcnt;
37 atomic_t event;
38 wait_queue_head_t poll;
Tejun Heoc525aad2013-12-11 14:11:55 -050039 struct list_head files; /* goes through kernfs_open_file.list */
Tejun Heo414985a2013-11-28 14:54:34 -050040};
41
Tejun Heoecca47c2014-07-01 16:41:03 -040042/*
43 * kernfs_notify() may be called from any context and bounces notifications
44 * through a work item. To minimize space overhead in kernfs_node, the
45 * pending queue is implemented as a singly linked list of kernfs_nodes.
46 * The list is terminated with the self pointer so that whether a
47 * kernfs_node is on the list or not can be determined by testing the next
48 * pointer for NULL.
49 */
50#define KERNFS_NOTIFY_EOL ((void *)&kernfs_notify_list)
51
52static DEFINE_SPINLOCK(kernfs_notify_lock);
53static struct kernfs_node *kernfs_notify_list = KERNFS_NOTIFY_EOL;
54
Tejun Heoc525aad2013-12-11 14:11:55 -050055static struct kernfs_open_file *kernfs_of(struct file *file)
Tejun Heo414985a2013-11-28 14:54:34 -050056{
57 return ((struct seq_file *)file->private_data)->private;
58}
59
60/*
Tejun Heo324a56e2013-12-11 14:11:53 -050061 * Determine the kernfs_ops for the given kernfs_node. This function must
Tejun Heo414985a2013-11-28 14:54:34 -050062 * be called while holding an active reference.
63 */
Tejun Heo324a56e2013-12-11 14:11:53 -050064static const struct kernfs_ops *kernfs_ops(struct kernfs_node *kn)
Tejun Heo414985a2013-11-28 14:54:34 -050065{
Tejun Heodf23fc32013-12-11 14:11:56 -050066 if (kn->flags & KERNFS_LOCKDEP)
Tejun Heo324a56e2013-12-11 14:11:53 -050067 lockdep_assert_held(kn);
Tejun Heoadc5e8b2013-12-11 14:11:54 -050068 return kn->attr.ops;
Tejun Heo414985a2013-11-28 14:54:34 -050069}
70
Tejun Heobb305942014-01-14 09:52:01 -050071/*
72 * As kernfs_seq_stop() is also called after kernfs_seq_start() or
73 * kernfs_seq_next() failure, it needs to distinguish whether it's stopping
74 * a seq_file iteration which is fully initialized with an active reference
75 * or an aborted kernfs_seq_start() due to get_active failure. The
76 * position pointer is the only context for each seq_file iteration and
77 * thus the stop condition should be encoded in it. As the return value is
78 * directly visible to userland, ERR_PTR(-ENODEV) is the only acceptable
79 * choice to indicate get_active failure.
80 *
81 * Unfortunately, this is complicated due to the optional custom seq_file
82 * operations which may return ERR_PTR(-ENODEV) too. kernfs_seq_stop()
83 * can't distinguish whether ERR_PTR(-ENODEV) is from get_active failure or
84 * custom seq_file operations and thus can't decide whether put_active
85 * should be performed or not only on ERR_PTR(-ENODEV).
86 *
87 * This is worked around by factoring out the custom seq_stop() and
88 * put_active part into kernfs_seq_stop_active(), skipping it from
89 * kernfs_seq_stop() if ERR_PTR(-ENODEV) while invoking it directly after
90 * custom seq_file operations fail with ERR_PTR(-ENODEV) - this ensures
91 * that kernfs_seq_stop_active() is skipped only after get_active failure.
92 */
93static void kernfs_seq_stop_active(struct seq_file *sf, void *v)
94{
95 struct kernfs_open_file *of = sf->private;
96 const struct kernfs_ops *ops = kernfs_ops(of->kn);
97
98 if (ops->seq_stop)
99 ops->seq_stop(sf, v);
100 kernfs_put_active(of->kn);
101}
102
Tejun Heo414985a2013-11-28 14:54:34 -0500103static void *kernfs_seq_start(struct seq_file *sf, loff_t *ppos)
104{
Tejun Heoc525aad2013-12-11 14:11:55 -0500105 struct kernfs_open_file *of = sf->private;
Tejun Heo414985a2013-11-28 14:54:34 -0500106 const struct kernfs_ops *ops;
107
108 /*
NeilBrown2b758692014-10-13 16:41:28 +1100109 * @of->mutex nests outside active ref and is primarily to ensure that
Tejun Heo414985a2013-11-28 14:54:34 -0500110 * the ops aren't called concurrently for the same open file.
111 */
112 mutex_lock(&of->mutex);
Tejun Heoc637b8a2013-12-11 14:11:58 -0500113 if (!kernfs_get_active(of->kn))
Tejun Heo414985a2013-11-28 14:54:34 -0500114 return ERR_PTR(-ENODEV);
115
Tejun Heo324a56e2013-12-11 14:11:53 -0500116 ops = kernfs_ops(of->kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500117 if (ops->seq_start) {
Tejun Heobb305942014-01-14 09:52:01 -0500118 void *next = ops->seq_start(sf, ppos);
119 /* see the comment above kernfs_seq_stop_active() */
120 if (next == ERR_PTR(-ENODEV))
121 kernfs_seq_stop_active(sf, next);
122 return next;
Tejun Heo414985a2013-11-28 14:54:34 -0500123 } else {
124 /*
125 * The same behavior and code as single_open(). Returns
126 * !NULL if pos is at the beginning; otherwise, NULL.
127 */
128 return NULL + !*ppos;
129 }
130}
131
132static void *kernfs_seq_next(struct seq_file *sf, void *v, loff_t *ppos)
133{
Tejun Heoc525aad2013-12-11 14:11:55 -0500134 struct kernfs_open_file *of = sf->private;
Tejun Heo324a56e2013-12-11 14:11:53 -0500135 const struct kernfs_ops *ops = kernfs_ops(of->kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500136
137 if (ops->seq_next) {
Tejun Heobb305942014-01-14 09:52:01 -0500138 void *next = ops->seq_next(sf, v, ppos);
139 /* see the comment above kernfs_seq_stop_active() */
140 if (next == ERR_PTR(-ENODEV))
141 kernfs_seq_stop_active(sf, next);
142 return next;
Tejun Heo414985a2013-11-28 14:54:34 -0500143 } else {
144 /*
145 * The same behavior and code as single_open(), always
146 * terminate after the initial read.
147 */
148 ++*ppos;
149 return NULL;
150 }
151}
152
153static void kernfs_seq_stop(struct seq_file *sf, void *v)
154{
Tejun Heoc525aad2013-12-11 14:11:55 -0500155 struct kernfs_open_file *of = sf->private;
Tejun Heo414985a2013-11-28 14:54:34 -0500156
Tejun Heobb305942014-01-14 09:52:01 -0500157 if (v != ERR_PTR(-ENODEV))
158 kernfs_seq_stop_active(sf, v);
Tejun Heo414985a2013-11-28 14:54:34 -0500159 mutex_unlock(&of->mutex);
160}
161
162static int kernfs_seq_show(struct seq_file *sf, void *v)
163{
Tejun Heoc525aad2013-12-11 14:11:55 -0500164 struct kernfs_open_file *of = sf->private;
Tejun Heo414985a2013-11-28 14:54:34 -0500165
Tejun Heoadc5e8b2013-12-11 14:11:54 -0500166 of->event = atomic_read(&of->kn->attr.open->event);
Tejun Heo414985a2013-11-28 14:54:34 -0500167
Tejun Heoadc5e8b2013-12-11 14:11:54 -0500168 return of->kn->attr.ops->seq_show(sf, v);
Tejun Heo414985a2013-11-28 14:54:34 -0500169}
170
171static const struct seq_operations kernfs_seq_ops = {
172 .start = kernfs_seq_start,
173 .next = kernfs_seq_next,
174 .stop = kernfs_seq_stop,
175 .show = kernfs_seq_show,
176};
177
178/*
179 * As reading a bin file can have side-effects, the exact offset and bytes
180 * specified in read(2) call should be passed to the read callback making
181 * it difficult to use seq_file. Implement simplistic custom buffering for
182 * bin files.
183 */
Tejun Heoc525aad2013-12-11 14:11:55 -0500184static ssize_t kernfs_file_direct_read(struct kernfs_open_file *of,
Tejun Heo414985a2013-11-28 14:54:34 -0500185 char __user *user_buf, size_t count,
186 loff_t *ppos)
187{
188 ssize_t len = min_t(size_t, count, PAGE_SIZE);
189 const struct kernfs_ops *ops;
190 char *buf;
191
NeilBrown4ef67a82014-10-14 16:57:26 +1100192 buf = of->prealloc_buf;
Chris Wilsone4234a12016-03-31 11:45:06 +0100193 if (buf)
194 mutex_lock(&of->prealloc_mutex);
195 else
NeilBrown4ef67a82014-10-14 16:57:26 +1100196 buf = kmalloc(len, GFP_KERNEL);
Tejun Heo414985a2013-11-28 14:54:34 -0500197 if (!buf)
198 return -ENOMEM;
199
200 /*
NeilBrown4ef67a82014-10-14 16:57:26 +1100201 * @of->mutex nests outside active ref and is used both to ensure that
Chris Wilsone4234a12016-03-31 11:45:06 +0100202 * the ops aren't called concurrently for the same open file.
Tejun Heo414985a2013-11-28 14:54:34 -0500203 */
204 mutex_lock(&of->mutex);
Tejun Heoc637b8a2013-12-11 14:11:58 -0500205 if (!kernfs_get_active(of->kn)) {
Tejun Heo414985a2013-11-28 14:54:34 -0500206 len = -ENODEV;
207 mutex_unlock(&of->mutex);
208 goto out_free;
209 }
210
NeilBrown7cff4b12015-03-16 10:44:52 +1100211 of->event = atomic_read(&of->kn->attr.open->event);
Tejun Heo324a56e2013-12-11 14:11:53 -0500212 ops = kernfs_ops(of->kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500213 if (ops->read)
214 len = ops->read(of, buf, len, *ppos);
215 else
216 len = -EINVAL;
217
Chris Wilsone4234a12016-03-31 11:45:06 +0100218 kernfs_put_active(of->kn);
219 mutex_unlock(&of->mutex);
220
Tejun Heo414985a2013-11-28 14:54:34 -0500221 if (len < 0)
Chris Wilsone4234a12016-03-31 11:45:06 +0100222 goto out_free;
Tejun Heo414985a2013-11-28 14:54:34 -0500223
224 if (copy_to_user(user_buf, buf, len)) {
225 len = -EFAULT;
Chris Wilsone4234a12016-03-31 11:45:06 +0100226 goto out_free;
Tejun Heo414985a2013-11-28 14:54:34 -0500227 }
228
229 *ppos += len;
230
231 out_free:
Chris Wilsone4234a12016-03-31 11:45:06 +0100232 if (buf == of->prealloc_buf)
233 mutex_unlock(&of->prealloc_mutex);
234 else
NeilBrown4ef67a82014-10-14 16:57:26 +1100235 kfree(buf);
Tejun Heo414985a2013-11-28 14:54:34 -0500236 return len;
237}
238
239/**
Tejun Heoc637b8a2013-12-11 14:11:58 -0500240 * kernfs_fop_read - kernfs vfs read callback
Tejun Heo414985a2013-11-28 14:54:34 -0500241 * @file: file pointer
242 * @user_buf: data to write
243 * @count: number of bytes
244 * @ppos: starting offset
245 */
Tejun Heoc637b8a2013-12-11 14:11:58 -0500246static ssize_t kernfs_fop_read(struct file *file, char __user *user_buf,
247 size_t count, loff_t *ppos)
Tejun Heo414985a2013-11-28 14:54:34 -0500248{
Tejun Heoc525aad2013-12-11 14:11:55 -0500249 struct kernfs_open_file *of = kernfs_of(file);
Tejun Heo414985a2013-11-28 14:54:34 -0500250
Tejun Heodf23fc32013-12-11 14:11:56 -0500251 if (of->kn->flags & KERNFS_HAS_SEQ_SHOW)
Tejun Heo414985a2013-11-28 14:54:34 -0500252 return seq_read(file, user_buf, count, ppos);
253 else
254 return kernfs_file_direct_read(of, user_buf, count, ppos);
255}
256
257/**
Tejun Heoc637b8a2013-12-11 14:11:58 -0500258 * kernfs_fop_write - kernfs vfs write callback
Tejun Heo414985a2013-11-28 14:54:34 -0500259 * @file: file pointer
260 * @user_buf: data to write
261 * @count: number of bytes
262 * @ppos: starting offset
263 *
264 * Copy data in from userland and pass it to the matching kernfs write
265 * operation.
266 *
267 * There is no easy way for us to know if userspace is only doing a partial
268 * write, so we don't support them. We expect the entire buffer to come on
269 * the first write. Hint: if you're writing a value, first read the file,
270 * modify only the the value you're changing, then write entire buffer
271 * back.
272 */
Tejun Heoc637b8a2013-12-11 14:11:58 -0500273static ssize_t kernfs_fop_write(struct file *file, const char __user *user_buf,
274 size_t count, loff_t *ppos)
Tejun Heo414985a2013-11-28 14:54:34 -0500275{
Tejun Heoc525aad2013-12-11 14:11:55 -0500276 struct kernfs_open_file *of = kernfs_of(file);
Tejun Heo414985a2013-11-28 14:54:34 -0500277 const struct kernfs_ops *ops;
Tejun Heob7ce40c2014-03-04 15:38:46 -0500278 size_t len;
279 char *buf;
280
281 if (of->atomic_write_len) {
282 len = count;
283 if (len > of->atomic_write_len)
284 return -E2BIG;
285 } else {
286 len = min_t(size_t, count, PAGE_SIZE);
287 }
288
NeilBrown2b758692014-10-13 16:41:28 +1100289 buf = of->prealloc_buf;
Chris Wilsone4234a12016-03-31 11:45:06 +0100290 if (buf)
291 mutex_lock(&of->prealloc_mutex);
292 else
NeilBrown2b758692014-10-13 16:41:28 +1100293 buf = kmalloc(len + 1, GFP_KERNEL);
Tejun Heob7ce40c2014-03-04 15:38:46 -0500294 if (!buf)
295 return -ENOMEM;
296
Chris Wilsone4234a12016-03-31 11:45:06 +0100297 if (copy_from_user(buf, user_buf, len)) {
298 len = -EFAULT;
299 goto out_free;
300 }
301 buf[len] = '\0'; /* guarantee string termination */
302
Tejun Heo414985a2013-11-28 14:54:34 -0500303 /*
NeilBrown2b758692014-10-13 16:41:28 +1100304 * @of->mutex nests outside active ref and is used both to ensure that
Chris Wilsone4234a12016-03-31 11:45:06 +0100305 * the ops aren't called concurrently for the same open file.
Tejun Heo414985a2013-11-28 14:54:34 -0500306 */
307 mutex_lock(&of->mutex);
Tejun Heoc637b8a2013-12-11 14:11:58 -0500308 if (!kernfs_get_active(of->kn)) {
Tejun Heo414985a2013-11-28 14:54:34 -0500309 mutex_unlock(&of->mutex);
Tejun Heob7ce40c2014-03-04 15:38:46 -0500310 len = -ENODEV;
311 goto out_free;
Tejun Heo414985a2013-11-28 14:54:34 -0500312 }
313
Tejun Heo324a56e2013-12-11 14:11:53 -0500314 ops = kernfs_ops(of->kn);
Tejun Heob7ce40c2014-03-04 15:38:46 -0500315 if (ops->write)
316 len = ops->write(of, buf, len, *ppos);
317 else
Tejun Heo414985a2013-11-28 14:54:34 -0500318 len = -EINVAL;
319
Chris Wilsone4234a12016-03-31 11:45:06 +0100320 kernfs_put_active(of->kn);
321 mutex_unlock(&of->mutex);
322
Tejun Heo414985a2013-11-28 14:54:34 -0500323 if (len > 0)
324 *ppos += len;
NeilBrown2b758692014-10-13 16:41:28 +1100325
Tejun Heob7ce40c2014-03-04 15:38:46 -0500326out_free:
Chris Wilsone4234a12016-03-31 11:45:06 +0100327 if (buf == of->prealloc_buf)
328 mutex_unlock(&of->prealloc_mutex);
329 else
NeilBrown2b758692014-10-13 16:41:28 +1100330 kfree(buf);
Tejun Heo414985a2013-11-28 14:54:34 -0500331 return len;
332}
333
334static void kernfs_vma_open(struct vm_area_struct *vma)
335{
336 struct file *file = vma->vm_file;
Tejun Heoc525aad2013-12-11 14:11:55 -0500337 struct kernfs_open_file *of = kernfs_of(file);
Tejun Heo414985a2013-11-28 14:54:34 -0500338
339 if (!of->vm_ops)
340 return;
341
Tejun Heoc637b8a2013-12-11 14:11:58 -0500342 if (!kernfs_get_active(of->kn))
Tejun Heo414985a2013-11-28 14:54:34 -0500343 return;
344
345 if (of->vm_ops->open)
346 of->vm_ops->open(vma);
347
Tejun Heoc637b8a2013-12-11 14:11:58 -0500348 kernfs_put_active(of->kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500349}
350
351static int kernfs_vma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
352{
353 struct file *file = vma->vm_file;
Tejun Heoc525aad2013-12-11 14:11:55 -0500354 struct kernfs_open_file *of = kernfs_of(file);
Tejun Heo414985a2013-11-28 14:54:34 -0500355 int ret;
356
357 if (!of->vm_ops)
358 return VM_FAULT_SIGBUS;
359
Tejun Heoc637b8a2013-12-11 14:11:58 -0500360 if (!kernfs_get_active(of->kn))
Tejun Heo414985a2013-11-28 14:54:34 -0500361 return VM_FAULT_SIGBUS;
362
363 ret = VM_FAULT_SIGBUS;
364 if (of->vm_ops->fault)
365 ret = of->vm_ops->fault(vma, vmf);
366
Tejun Heoc637b8a2013-12-11 14:11:58 -0500367 kernfs_put_active(of->kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500368 return ret;
369}
370
371static int kernfs_vma_page_mkwrite(struct vm_area_struct *vma,
372 struct vm_fault *vmf)
373{
374 struct file *file = vma->vm_file;
Tejun Heoc525aad2013-12-11 14:11:55 -0500375 struct kernfs_open_file *of = kernfs_of(file);
Tejun Heo414985a2013-11-28 14:54:34 -0500376 int ret;
377
378 if (!of->vm_ops)
379 return VM_FAULT_SIGBUS;
380
Tejun Heoc637b8a2013-12-11 14:11:58 -0500381 if (!kernfs_get_active(of->kn))
Tejun Heo414985a2013-11-28 14:54:34 -0500382 return VM_FAULT_SIGBUS;
383
384 ret = 0;
385 if (of->vm_ops->page_mkwrite)
386 ret = of->vm_ops->page_mkwrite(vma, vmf);
387 else
388 file_update_time(file);
389
Tejun Heoc637b8a2013-12-11 14:11:58 -0500390 kernfs_put_active(of->kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500391 return ret;
392}
393
394static int kernfs_vma_access(struct vm_area_struct *vma, unsigned long addr,
395 void *buf, int len, int write)
396{
397 struct file *file = vma->vm_file;
Tejun Heoc525aad2013-12-11 14:11:55 -0500398 struct kernfs_open_file *of = kernfs_of(file);
Tejun Heo414985a2013-11-28 14:54:34 -0500399 int ret;
400
401 if (!of->vm_ops)
402 return -EINVAL;
403
Tejun Heoc637b8a2013-12-11 14:11:58 -0500404 if (!kernfs_get_active(of->kn))
Tejun Heo414985a2013-11-28 14:54:34 -0500405 return -EINVAL;
406
407 ret = -EINVAL;
408 if (of->vm_ops->access)
409 ret = of->vm_ops->access(vma, addr, buf, len, write);
410
Tejun Heoc637b8a2013-12-11 14:11:58 -0500411 kernfs_put_active(of->kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500412 return ret;
413}
414
415#ifdef CONFIG_NUMA
416static int kernfs_vma_set_policy(struct vm_area_struct *vma,
417 struct mempolicy *new)
418{
419 struct file *file = vma->vm_file;
Tejun Heoc525aad2013-12-11 14:11:55 -0500420 struct kernfs_open_file *of = kernfs_of(file);
Tejun Heo414985a2013-11-28 14:54:34 -0500421 int ret;
422
423 if (!of->vm_ops)
424 return 0;
425
Tejun Heoc637b8a2013-12-11 14:11:58 -0500426 if (!kernfs_get_active(of->kn))
Tejun Heo414985a2013-11-28 14:54:34 -0500427 return -EINVAL;
428
429 ret = 0;
430 if (of->vm_ops->set_policy)
431 ret = of->vm_ops->set_policy(vma, new);
432
Tejun Heoc637b8a2013-12-11 14:11:58 -0500433 kernfs_put_active(of->kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500434 return ret;
435}
436
437static struct mempolicy *kernfs_vma_get_policy(struct vm_area_struct *vma,
438 unsigned long addr)
439{
440 struct file *file = vma->vm_file;
Tejun Heoc525aad2013-12-11 14:11:55 -0500441 struct kernfs_open_file *of = kernfs_of(file);
Tejun Heo414985a2013-11-28 14:54:34 -0500442 struct mempolicy *pol;
443
444 if (!of->vm_ops)
445 return vma->vm_policy;
446
Tejun Heoc637b8a2013-12-11 14:11:58 -0500447 if (!kernfs_get_active(of->kn))
Tejun Heo414985a2013-11-28 14:54:34 -0500448 return vma->vm_policy;
449
450 pol = vma->vm_policy;
451 if (of->vm_ops->get_policy)
452 pol = of->vm_ops->get_policy(vma, addr);
453
Tejun Heoc637b8a2013-12-11 14:11:58 -0500454 kernfs_put_active(of->kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500455 return pol;
456}
457
Tejun Heo414985a2013-11-28 14:54:34 -0500458#endif
459
460static const struct vm_operations_struct kernfs_vm_ops = {
461 .open = kernfs_vma_open,
462 .fault = kernfs_vma_fault,
463 .page_mkwrite = kernfs_vma_page_mkwrite,
464 .access = kernfs_vma_access,
465#ifdef CONFIG_NUMA
466 .set_policy = kernfs_vma_set_policy,
467 .get_policy = kernfs_vma_get_policy,
Tejun Heo414985a2013-11-28 14:54:34 -0500468#endif
469};
470
Tejun Heoc637b8a2013-12-11 14:11:58 -0500471static int kernfs_fop_mmap(struct file *file, struct vm_area_struct *vma)
Tejun Heo414985a2013-11-28 14:54:34 -0500472{
Tejun Heoc525aad2013-12-11 14:11:55 -0500473 struct kernfs_open_file *of = kernfs_of(file);
Tejun Heo414985a2013-11-28 14:54:34 -0500474 const struct kernfs_ops *ops;
475 int rc;
476
Tejun Heo9b2db6e2013-12-10 09:29:17 -0500477 /*
478 * mmap path and of->mutex are prone to triggering spurious lockdep
479 * warnings and we don't want to add spurious locking dependency
480 * between the two. Check whether mmap is actually implemented
481 * without grabbing @of->mutex by testing HAS_MMAP flag. See the
482 * comment in kernfs_file_open() for more details.
483 */
Tejun Heodf23fc32013-12-11 14:11:56 -0500484 if (!(of->kn->flags & KERNFS_HAS_MMAP))
Tejun Heo9b2db6e2013-12-10 09:29:17 -0500485 return -ENODEV;
486
Tejun Heo414985a2013-11-28 14:54:34 -0500487 mutex_lock(&of->mutex);
488
489 rc = -ENODEV;
Tejun Heoc637b8a2013-12-11 14:11:58 -0500490 if (!kernfs_get_active(of->kn))
Tejun Heo414985a2013-11-28 14:54:34 -0500491 goto out_unlock;
492
Tejun Heo324a56e2013-12-11 14:11:53 -0500493 ops = kernfs_ops(of->kn);
Tejun Heo9b2db6e2013-12-10 09:29:17 -0500494 rc = ops->mmap(of, vma);
Tejun Heob44b2142014-04-20 08:29:21 -0400495 if (rc)
496 goto out_put;
Tejun Heo414985a2013-11-28 14:54:34 -0500497
498 /*
499 * PowerPC's pci_mmap of legacy_mem uses shmem_zero_setup()
500 * to satisfy versions of X which crash if the mmap fails: that
501 * substitutes a new vm_file, and we don't then want bin_vm_ops.
502 */
503 if (vma->vm_file != file)
504 goto out_put;
505
506 rc = -EINVAL;
507 if (of->mmapped && of->vm_ops != vma->vm_ops)
508 goto out_put;
509
510 /*
511 * It is not possible to successfully wrap close.
512 * So error if someone is trying to use close.
513 */
514 rc = -EINVAL;
515 if (vma->vm_ops && vma->vm_ops->close)
516 goto out_put;
517
518 rc = 0;
519 of->mmapped = 1;
520 of->vm_ops = vma->vm_ops;
521 vma->vm_ops = &kernfs_vm_ops;
522out_put:
Tejun Heoc637b8a2013-12-11 14:11:58 -0500523 kernfs_put_active(of->kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500524out_unlock:
525 mutex_unlock(&of->mutex);
526
527 return rc;
528}
529
530/**
Tejun Heoc637b8a2013-12-11 14:11:58 -0500531 * kernfs_get_open_node - get or create kernfs_open_node
Tejun Heo324a56e2013-12-11 14:11:53 -0500532 * @kn: target kernfs_node
Tejun Heoc525aad2013-12-11 14:11:55 -0500533 * @of: kernfs_open_file for this instance of open
Tejun Heo414985a2013-11-28 14:54:34 -0500534 *
Tejun Heoadc5e8b2013-12-11 14:11:54 -0500535 * If @kn->attr.open exists, increment its reference count; otherwise,
536 * create one. @of is chained to the files list.
Tejun Heo414985a2013-11-28 14:54:34 -0500537 *
538 * LOCKING:
539 * Kernel thread context (may sleep).
540 *
541 * RETURNS:
542 * 0 on success, -errno on failure.
543 */
Tejun Heoc637b8a2013-12-11 14:11:58 -0500544static int kernfs_get_open_node(struct kernfs_node *kn,
545 struct kernfs_open_file *of)
Tejun Heo414985a2013-11-28 14:54:34 -0500546{
Tejun Heoc525aad2013-12-11 14:11:55 -0500547 struct kernfs_open_node *on, *new_on = NULL;
Tejun Heo414985a2013-11-28 14:54:34 -0500548
549 retry:
Tejun Heoc525aad2013-12-11 14:11:55 -0500550 mutex_lock(&kernfs_open_file_mutex);
551 spin_lock_irq(&kernfs_open_node_lock);
Tejun Heo414985a2013-11-28 14:54:34 -0500552
Tejun Heoc525aad2013-12-11 14:11:55 -0500553 if (!kn->attr.open && new_on) {
554 kn->attr.open = new_on;
555 new_on = NULL;
Tejun Heo414985a2013-11-28 14:54:34 -0500556 }
557
Tejun Heoc525aad2013-12-11 14:11:55 -0500558 on = kn->attr.open;
559 if (on) {
560 atomic_inc(&on->refcnt);
561 list_add_tail(&of->list, &on->files);
Tejun Heo414985a2013-11-28 14:54:34 -0500562 }
563
Tejun Heoc525aad2013-12-11 14:11:55 -0500564 spin_unlock_irq(&kernfs_open_node_lock);
565 mutex_unlock(&kernfs_open_file_mutex);
Tejun Heo414985a2013-11-28 14:54:34 -0500566
Tejun Heoc525aad2013-12-11 14:11:55 -0500567 if (on) {
568 kfree(new_on);
Tejun Heo414985a2013-11-28 14:54:34 -0500569 return 0;
570 }
571
572 /* not there, initialize a new one and retry */
Tejun Heoc525aad2013-12-11 14:11:55 -0500573 new_on = kmalloc(sizeof(*new_on), GFP_KERNEL);
574 if (!new_on)
Tejun Heo414985a2013-11-28 14:54:34 -0500575 return -ENOMEM;
576
Tejun Heoc525aad2013-12-11 14:11:55 -0500577 atomic_set(&new_on->refcnt, 0);
578 atomic_set(&new_on->event, 1);
579 init_waitqueue_head(&new_on->poll);
580 INIT_LIST_HEAD(&new_on->files);
Tejun Heo414985a2013-11-28 14:54:34 -0500581 goto retry;
582}
583
584/**
Tejun Heoc637b8a2013-12-11 14:11:58 -0500585 * kernfs_put_open_node - put kernfs_open_node
Tejun Heo324a56e2013-12-11 14:11:53 -0500586 * @kn: target kernfs_nodet
Tejun Heoc525aad2013-12-11 14:11:55 -0500587 * @of: associated kernfs_open_file
Tejun Heo414985a2013-11-28 14:54:34 -0500588 *
Tejun Heoadc5e8b2013-12-11 14:11:54 -0500589 * Put @kn->attr.open and unlink @of from the files list. If
Tejun Heo414985a2013-11-28 14:54:34 -0500590 * reference count reaches zero, disassociate and free it.
591 *
592 * LOCKING:
593 * None.
594 */
Tejun Heoc637b8a2013-12-11 14:11:58 -0500595static void kernfs_put_open_node(struct kernfs_node *kn,
596 struct kernfs_open_file *of)
Tejun Heo414985a2013-11-28 14:54:34 -0500597{
Tejun Heoc525aad2013-12-11 14:11:55 -0500598 struct kernfs_open_node *on = kn->attr.open;
Tejun Heo414985a2013-11-28 14:54:34 -0500599 unsigned long flags;
600
Tejun Heoc525aad2013-12-11 14:11:55 -0500601 mutex_lock(&kernfs_open_file_mutex);
602 spin_lock_irqsave(&kernfs_open_node_lock, flags);
Tejun Heo414985a2013-11-28 14:54:34 -0500603
604 if (of)
605 list_del(&of->list);
606
Tejun Heoc525aad2013-12-11 14:11:55 -0500607 if (atomic_dec_and_test(&on->refcnt))
Tejun Heoadc5e8b2013-12-11 14:11:54 -0500608 kn->attr.open = NULL;
Tejun Heo414985a2013-11-28 14:54:34 -0500609 else
Tejun Heoc525aad2013-12-11 14:11:55 -0500610 on = NULL;
Tejun Heo414985a2013-11-28 14:54:34 -0500611
Tejun Heoc525aad2013-12-11 14:11:55 -0500612 spin_unlock_irqrestore(&kernfs_open_node_lock, flags);
613 mutex_unlock(&kernfs_open_file_mutex);
Tejun Heo414985a2013-11-28 14:54:34 -0500614
Tejun Heoc525aad2013-12-11 14:11:55 -0500615 kfree(on);
Tejun Heo414985a2013-11-28 14:54:34 -0500616}
617
Tejun Heoc637b8a2013-12-11 14:11:58 -0500618static int kernfs_fop_open(struct inode *inode, struct file *file)
Tejun Heo414985a2013-11-28 14:54:34 -0500619{
Tejun Heo324a56e2013-12-11 14:11:53 -0500620 struct kernfs_node *kn = file->f_path.dentry->d_fsdata;
Tejun Heo555724a2014-05-12 13:56:27 -0400621 struct kernfs_root *root = kernfs_root(kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500622 const struct kernfs_ops *ops;
Tejun Heoc525aad2013-12-11 14:11:55 -0500623 struct kernfs_open_file *of;
Tejun Heo414985a2013-11-28 14:54:34 -0500624 bool has_read, has_write, has_mmap;
625 int error = -EACCES;
626
Tejun Heoc637b8a2013-12-11 14:11:58 -0500627 if (!kernfs_get_active(kn))
Tejun Heo414985a2013-11-28 14:54:34 -0500628 return -ENODEV;
629
Tejun Heo324a56e2013-12-11 14:11:53 -0500630 ops = kernfs_ops(kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500631
632 has_read = ops->seq_show || ops->read || ops->mmap;
633 has_write = ops->write || ops->mmap;
634 has_mmap = ops->mmap;
635
Tejun Heo555724a2014-05-12 13:56:27 -0400636 /* see the flag definition for details */
637 if (root->flags & KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK) {
638 if ((file->f_mode & FMODE_WRITE) &&
639 (!(inode->i_mode & S_IWUGO) || !has_write))
640 goto err_out;
Tejun Heo414985a2013-11-28 14:54:34 -0500641
Tejun Heo555724a2014-05-12 13:56:27 -0400642 if ((file->f_mode & FMODE_READ) &&
643 (!(inode->i_mode & S_IRUGO) || !has_read))
644 goto err_out;
645 }
Tejun Heo414985a2013-11-28 14:54:34 -0500646
Tejun Heoc525aad2013-12-11 14:11:55 -0500647 /* allocate a kernfs_open_file for the file */
Tejun Heo414985a2013-11-28 14:54:34 -0500648 error = -ENOMEM;
Tejun Heoc525aad2013-12-11 14:11:55 -0500649 of = kzalloc(sizeof(struct kernfs_open_file), GFP_KERNEL);
Tejun Heo414985a2013-11-28 14:54:34 -0500650 if (!of)
651 goto err_out;
652
653 /*
654 * The following is done to give a different lockdep key to
655 * @of->mutex for files which implement mmap. This is a rather
656 * crude way to avoid false positive lockdep warning around
657 * mm->mmap_sem - mmap nests @of->mutex under mm->mmap_sem and
658 * reading /sys/block/sda/trace/act_mask grabs sr_mutex, under
659 * which mm->mmap_sem nests, while holding @of->mutex. As each
660 * open file has a separate mutex, it's okay as long as those don't
661 * happen on the same file. At this point, we can't easily give
662 * each file a separate locking class. Let's differentiate on
663 * whether the file has mmap or not for now.
Tejun Heo9b2db6e2013-12-10 09:29:17 -0500664 *
665 * Both paths of the branch look the same. They're supposed to
666 * look that way and give @of->mutex different static lockdep keys.
Tejun Heo414985a2013-11-28 14:54:34 -0500667 */
668 if (has_mmap)
669 mutex_init(&of->mutex);
670 else
671 mutex_init(&of->mutex);
672
Tejun Heo324a56e2013-12-11 14:11:53 -0500673 of->kn = kn;
Tejun Heo414985a2013-11-28 14:54:34 -0500674 of->file = file;
675
676 /*
Tejun Heob7ce40c2014-03-04 15:38:46 -0500677 * Write path needs to atomic_write_len outside active reference.
678 * Cache it in open_file. See kernfs_fop_write() for details.
679 */
680 of->atomic_write_len = ops->atomic_write_len;
681
NeilBrown4ef67a82014-10-14 16:57:26 +1100682 error = -EINVAL;
683 /*
684 * ->seq_show is incompatible with ->prealloc,
685 * as seq_read does its own allocation.
686 * ->read must be used instead.
687 */
688 if (ops->prealloc && ops->seq_show)
689 goto err_free;
NeilBrown2b758692014-10-13 16:41:28 +1100690 if (ops->prealloc) {
691 int len = of->atomic_write_len ?: PAGE_SIZE;
692 of->prealloc_buf = kmalloc(len + 1, GFP_KERNEL);
693 error = -ENOMEM;
694 if (!of->prealloc_buf)
695 goto err_free;
Chris Wilsone4234a12016-03-31 11:45:06 +0100696 mutex_init(&of->prealloc_mutex);
NeilBrown2b758692014-10-13 16:41:28 +1100697 }
698
Tejun Heob7ce40c2014-03-04 15:38:46 -0500699 /*
Tejun Heo414985a2013-11-28 14:54:34 -0500700 * Always instantiate seq_file even if read access doesn't use
701 * seq_file or is not requested. This unifies private data access
702 * and readable regular files are the vast majority anyway.
703 */
704 if (ops->seq_show)
705 error = seq_open(file, &kernfs_seq_ops);
706 else
707 error = seq_open(file, NULL);
708 if (error)
709 goto err_free;
710
711 ((struct seq_file *)file->private_data)->private = of;
712
713 /* seq_file clears PWRITE unconditionally, restore it if WRITE */
714 if (file->f_mode & FMODE_WRITE)
715 file->f_mode |= FMODE_PWRITE;
716
Tejun Heoc637b8a2013-12-11 14:11:58 -0500717 /* make sure we have open node struct */
718 error = kernfs_get_open_node(kn, of);
Tejun Heo414985a2013-11-28 14:54:34 -0500719 if (error)
720 goto err_close;
721
722 /* open succeeded, put active references */
Tejun Heoc637b8a2013-12-11 14:11:58 -0500723 kernfs_put_active(kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500724 return 0;
725
726err_close:
727 seq_release(inode, file);
728err_free:
NeilBrown2b758692014-10-13 16:41:28 +1100729 kfree(of->prealloc_buf);
Tejun Heo414985a2013-11-28 14:54:34 -0500730 kfree(of);
731err_out:
Tejun Heoc637b8a2013-12-11 14:11:58 -0500732 kernfs_put_active(kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500733 return error;
734}
735
Tejun Heoc637b8a2013-12-11 14:11:58 -0500736static int kernfs_fop_release(struct inode *inode, struct file *filp)
Tejun Heo414985a2013-11-28 14:54:34 -0500737{
Tejun Heo324a56e2013-12-11 14:11:53 -0500738 struct kernfs_node *kn = filp->f_path.dentry->d_fsdata;
Tejun Heoc525aad2013-12-11 14:11:55 -0500739 struct kernfs_open_file *of = kernfs_of(filp);
Tejun Heo414985a2013-11-28 14:54:34 -0500740
Tejun Heoc637b8a2013-12-11 14:11:58 -0500741 kernfs_put_open_node(kn, of);
Tejun Heo414985a2013-11-28 14:54:34 -0500742 seq_release(inode, filp);
NeilBrown2b758692014-10-13 16:41:28 +1100743 kfree(of->prealloc_buf);
Tejun Heo414985a2013-11-28 14:54:34 -0500744 kfree(of);
745
746 return 0;
747}
748
Greg Kroah-Hartman55f6e302014-01-13 14:27:16 -0800749void kernfs_unmap_bin_file(struct kernfs_node *kn)
Tejun Heo414985a2013-11-28 14:54:34 -0500750{
Tejun Heoc525aad2013-12-11 14:11:55 -0500751 struct kernfs_open_node *on;
752 struct kernfs_open_file *of;
Tejun Heo414985a2013-11-28 14:54:34 -0500753
Greg Kroah-Hartman55f6e302014-01-13 14:27:16 -0800754 if (!(kn->flags & KERNFS_HAS_MMAP))
755 return;
756
Tejun Heoc525aad2013-12-11 14:11:55 -0500757 spin_lock_irq(&kernfs_open_node_lock);
758 on = kn->attr.open;
759 if (on)
760 atomic_inc(&on->refcnt);
761 spin_unlock_irq(&kernfs_open_node_lock);
762 if (!on)
Tejun Heo414985a2013-11-28 14:54:34 -0500763 return;
764
Tejun Heoc525aad2013-12-11 14:11:55 -0500765 mutex_lock(&kernfs_open_file_mutex);
766 list_for_each_entry(of, &on->files, list) {
Tejun Heo414985a2013-11-28 14:54:34 -0500767 struct inode *inode = file_inode(of->file);
768 unmap_mapping_range(inode->i_mapping, 0, 0, 1);
769 }
Tejun Heoc525aad2013-12-11 14:11:55 -0500770 mutex_unlock(&kernfs_open_file_mutex);
Tejun Heo414985a2013-11-28 14:54:34 -0500771
Tejun Heoc637b8a2013-12-11 14:11:58 -0500772 kernfs_put_open_node(kn, NULL);
Tejun Heo414985a2013-11-28 14:54:34 -0500773}
774
Tejun Heoc637b8a2013-12-11 14:11:58 -0500775/*
776 * Kernfs attribute files are pollable. The idea is that you read
Tejun Heo414985a2013-11-28 14:54:34 -0500777 * the content and then you use 'poll' or 'select' to wait for
778 * the content to change. When the content changes (assuming the
779 * manager for the kobject supports notification), poll will
780 * return POLLERR|POLLPRI, and select will return the fd whether
781 * it is waiting for read, write, or exceptions.
782 * Once poll/select indicates that the value has changed, you
783 * need to close and re-open the file, or seek to 0 and read again.
784 * Reminder: this only works for attributes which actively support
785 * it, and it is not possible to test an attribute from userspace
786 * to see if it supports poll (Neither 'poll' nor 'select' return
787 * an appropriate error code). When in doubt, set a suitable timeout value.
788 */
Tejun Heoc637b8a2013-12-11 14:11:58 -0500789static unsigned int kernfs_fop_poll(struct file *filp, poll_table *wait)
Tejun Heo414985a2013-11-28 14:54:34 -0500790{
Tejun Heoc525aad2013-12-11 14:11:55 -0500791 struct kernfs_open_file *of = kernfs_of(filp);
Tejun Heo324a56e2013-12-11 14:11:53 -0500792 struct kernfs_node *kn = filp->f_path.dentry->d_fsdata;
Tejun Heoc525aad2013-12-11 14:11:55 -0500793 struct kernfs_open_node *on = kn->attr.open;
Tejun Heo414985a2013-11-28 14:54:34 -0500794
Tejun Heoc637b8a2013-12-11 14:11:58 -0500795 if (!kernfs_get_active(kn))
Tejun Heo414985a2013-11-28 14:54:34 -0500796 goto trigger;
797
Tejun Heoc525aad2013-12-11 14:11:55 -0500798 poll_wait(filp, &on->poll, wait);
Tejun Heo414985a2013-11-28 14:54:34 -0500799
Tejun Heoc637b8a2013-12-11 14:11:58 -0500800 kernfs_put_active(kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500801
Tejun Heoc525aad2013-12-11 14:11:55 -0500802 if (of->event != atomic_read(&on->event))
Tejun Heo414985a2013-11-28 14:54:34 -0500803 goto trigger;
804
805 return DEFAULT_POLLMASK;
806
807 trigger:
808 return DEFAULT_POLLMASK|POLLERR|POLLPRI;
809}
810
Tejun Heoecca47c2014-07-01 16:41:03 -0400811static void kernfs_notify_workfn(struct work_struct *work)
Tejun Heo414985a2013-11-28 14:54:34 -0500812{
Tejun Heoecca47c2014-07-01 16:41:03 -0400813 struct kernfs_node *kn;
Tejun Heoc525aad2013-12-11 14:11:55 -0500814 struct kernfs_open_node *on;
Tejun Heod911d982014-04-09 11:07:31 -0400815 struct kernfs_super_info *info;
Tejun Heoecca47c2014-07-01 16:41:03 -0400816repeat:
817 /* pop one off the notify_list */
818 spin_lock_irq(&kernfs_notify_lock);
819 kn = kernfs_notify_list;
820 if (kn == KERNFS_NOTIFY_EOL) {
821 spin_unlock_irq(&kernfs_notify_lock);
Tejun Heod911d982014-04-09 11:07:31 -0400822 return;
Tejun Heoecca47c2014-07-01 16:41:03 -0400823 }
824 kernfs_notify_list = kn->attr.notify_next;
825 kn->attr.notify_next = NULL;
826 spin_unlock_irq(&kernfs_notify_lock);
Tejun Heod911d982014-04-09 11:07:31 -0400827
828 /* kick poll */
Tejun Heoecca47c2014-07-01 16:41:03 -0400829 spin_lock_irq(&kernfs_open_node_lock);
Tejun Heo414985a2013-11-28 14:54:34 -0500830
Tejun Heod911d982014-04-09 11:07:31 -0400831 on = kn->attr.open;
832 if (on) {
833 atomic_inc(&on->event);
834 wake_up_interruptible(&on->poll);
Tejun Heo414985a2013-11-28 14:54:34 -0500835 }
836
Tejun Heoecca47c2014-07-01 16:41:03 -0400837 spin_unlock_irq(&kernfs_open_node_lock);
Tejun Heod911d982014-04-09 11:07:31 -0400838
839 /* kick fsnotify */
840 mutex_lock(&kernfs_mutex);
841
Tejun Heoecca47c2014-07-01 16:41:03 -0400842 list_for_each_entry(info, &kernfs_root(kn)->supers, node) {
Tejun Heodf6a58c2016-06-17 17:51:17 -0400843 struct kernfs_node *parent;
Tejun Heod911d982014-04-09 11:07:31 -0400844 struct inode *inode;
Tejun Heod911d982014-04-09 11:07:31 -0400845
Tejun Heodf6a58c2016-06-17 17:51:17 -0400846 /*
847 * We want fsnotify_modify() on @kn but as the
848 * modifications aren't originating from userland don't
849 * have the matching @file available. Look up the inodes
850 * and generate the events manually.
851 */
Tejun Heod911d982014-04-09 11:07:31 -0400852 inode = ilookup(info->sb, kn->ino);
853 if (!inode)
854 continue;
855
Tejun Heodf6a58c2016-06-17 17:51:17 -0400856 parent = kernfs_get_parent(kn);
857 if (parent) {
858 struct inode *p_inode;
859
860 p_inode = ilookup(info->sb, parent->ino);
861 if (p_inode) {
862 fsnotify(p_inode, FS_MODIFY | FS_EVENT_ON_CHILD,
863 inode, FSNOTIFY_EVENT_INODE, kn->name, 0);
864 iput(p_inode);
865 }
866
867 kernfs_put(parent);
Tejun Heod911d982014-04-09 11:07:31 -0400868 }
869
Tejun Heodf6a58c2016-06-17 17:51:17 -0400870 fsnotify(inode, FS_MODIFY, inode, FSNOTIFY_EVENT_INODE,
871 kn->name, 0);
Tejun Heod911d982014-04-09 11:07:31 -0400872 iput(inode);
873 }
874
875 mutex_unlock(&kernfs_mutex);
Tejun Heoecca47c2014-07-01 16:41:03 -0400876 kernfs_put(kn);
877 goto repeat;
878}
879
880/**
881 * kernfs_notify - notify a kernfs file
882 * @kn: file to notify
883 *
884 * Notify @kn such that poll(2) on @kn wakes up. Maybe be called from any
885 * context.
886 */
887void kernfs_notify(struct kernfs_node *kn)
888{
889 static DECLARE_WORK(kernfs_notify_work, kernfs_notify_workfn);
890 unsigned long flags;
891
892 if (WARN_ON(kernfs_type(kn) != KERNFS_FILE))
893 return;
894
895 spin_lock_irqsave(&kernfs_notify_lock, flags);
896 if (!kn->attr.notify_next) {
897 kernfs_get(kn);
898 kn->attr.notify_next = kernfs_notify_list;
899 kernfs_notify_list = kn;
900 schedule_work(&kernfs_notify_work);
901 }
902 spin_unlock_irqrestore(&kernfs_notify_lock, flags);
Tejun Heo414985a2013-11-28 14:54:34 -0500903}
904EXPORT_SYMBOL_GPL(kernfs_notify);
905
Tejun Heoa797bfc2013-12-11 14:11:57 -0500906const struct file_operations kernfs_file_fops = {
Tejun Heoc637b8a2013-12-11 14:11:58 -0500907 .read = kernfs_fop_read,
908 .write = kernfs_fop_write,
Tejun Heo414985a2013-11-28 14:54:34 -0500909 .llseek = generic_file_llseek,
Tejun Heoc637b8a2013-12-11 14:11:58 -0500910 .mmap = kernfs_fop_mmap,
911 .open = kernfs_fop_open,
912 .release = kernfs_fop_release,
913 .poll = kernfs_fop_poll,
Tony Luck2a9becd2016-10-14 10:56:42 -0700914 .fsync = noop_fsync,
Tejun Heo414985a2013-11-28 14:54:34 -0500915};
916
917/**
Tejun Heo2063d602013-12-11 16:02:57 -0500918 * __kernfs_create_file - kernfs internal function to create a file
Tejun Heo414985a2013-11-28 14:54:34 -0500919 * @parent: directory to create the file in
920 * @name: name of the file
921 * @mode: mode of the file
922 * @size: size of the file
923 * @ops: kernfs operations for the file
924 * @priv: private data for the file
925 * @ns: optional namespace tag of the file
926 * @key: lockdep key for the file's active_ref, %NULL to disable lockdep
927 *
928 * Returns the created node on success, ERR_PTR() value on error.
929 */
Tejun Heo2063d602013-12-11 16:02:57 -0500930struct kernfs_node *__kernfs_create_file(struct kernfs_node *parent,
931 const char *name,
932 umode_t mode, loff_t size,
933 const struct kernfs_ops *ops,
934 void *priv, const void *ns,
Tejun Heo2063d602013-12-11 16:02:57 -0500935 struct lock_class_key *key)
Tejun Heo414985a2013-11-28 14:54:34 -0500936{
Tejun Heo324a56e2013-12-11 14:11:53 -0500937 struct kernfs_node *kn;
Tejun Heo2063d602013-12-11 16:02:57 -0500938 unsigned flags;
Tejun Heo414985a2013-11-28 14:54:34 -0500939 int rc;
940
Tejun Heo2063d602013-12-11 16:02:57 -0500941 flags = KERNFS_FILE;
Tejun Heo2063d602013-12-11 16:02:57 -0500942
Tejun Heodb4aad22014-01-17 09:58:25 -0500943 kn = kernfs_new_node(parent, name, (mode & S_IALLUGO) | S_IFREG, flags);
Tejun Heo324a56e2013-12-11 14:11:53 -0500944 if (!kn)
Tejun Heo414985a2013-11-28 14:54:34 -0500945 return ERR_PTR(-ENOMEM);
946
Tejun Heoadc5e8b2013-12-11 14:11:54 -0500947 kn->attr.ops = ops;
948 kn->attr.size = size;
949 kn->ns = ns;
Tejun Heo324a56e2013-12-11 14:11:53 -0500950 kn->priv = priv;
Tejun Heo414985a2013-11-28 14:54:34 -0500951
952#ifdef CONFIG_DEBUG_LOCK_ALLOC
953 if (key) {
Tejun Heo324a56e2013-12-11 14:11:53 -0500954 lockdep_init_map(&kn->dep_map, "s_active", key, 0);
Tejun Heodf23fc32013-12-11 14:11:56 -0500955 kn->flags |= KERNFS_LOCKDEP;
Tejun Heo414985a2013-11-28 14:54:34 -0500956 }
957#endif
958
959 /*
Tejun Heoadc5e8b2013-12-11 14:11:54 -0500960 * kn->attr.ops is accesible only while holding active ref. We
Tejun Heo414985a2013-11-28 14:54:34 -0500961 * need to know whether some ops are implemented outside active
962 * ref. Cache their existence in flags.
963 */
964 if (ops->seq_show)
Tejun Heodf23fc32013-12-11 14:11:56 -0500965 kn->flags |= KERNFS_HAS_SEQ_SHOW;
Tejun Heo414985a2013-11-28 14:54:34 -0500966 if (ops->mmap)
Tejun Heodf23fc32013-12-11 14:11:56 -0500967 kn->flags |= KERNFS_HAS_MMAP;
Tejun Heo414985a2013-11-28 14:54:34 -0500968
Tejun Heo988cd7a2014-02-03 14:02:58 -0500969 rc = kernfs_add_one(kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500970 if (rc) {
Tejun Heo324a56e2013-12-11 14:11:53 -0500971 kernfs_put(kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500972 return ERR_PTR(rc);
973 }
Tejun Heo324a56e2013-12-11 14:11:53 -0500974 return kn;
Tejun Heo414985a2013-11-28 14:54:34 -0500975}