blob: dbf397bfdff266d16332c57ac2de90e43ef8d8eb [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>
17
18#include "kernfs-internal.h"
19
20/*
Tejun Heoc525aad2013-12-11 14:11:55 -050021 * There's one kernfs_open_file for each open file and one kernfs_open_node
Tejun Heo324a56e2013-12-11 14:11:53 -050022 * for each kernfs_node with one or more open files.
Tejun Heo414985a2013-11-28 14:54:34 -050023 *
Tejun Heoc525aad2013-12-11 14:11:55 -050024 * kernfs_node->attr.open points to kernfs_open_node. attr.open is
25 * protected by kernfs_open_node_lock.
Tejun Heo414985a2013-11-28 14:54:34 -050026 *
27 * filp->private_data points to seq_file whose ->private points to
Tejun Heoc525aad2013-12-11 14:11:55 -050028 * kernfs_open_file. kernfs_open_files are chained at
29 * kernfs_open_node->files, which is protected by kernfs_open_file_mutex.
Tejun Heo414985a2013-11-28 14:54:34 -050030 */
Tejun Heoc525aad2013-12-11 14:11:55 -050031static DEFINE_SPINLOCK(kernfs_open_node_lock);
32static DEFINE_MUTEX(kernfs_open_file_mutex);
Tejun Heo414985a2013-11-28 14:54:34 -050033
Tejun Heoc525aad2013-12-11 14:11:55 -050034struct kernfs_open_node {
Tejun Heo414985a2013-11-28 14:54:34 -050035 atomic_t refcnt;
36 atomic_t event;
37 wait_queue_head_t poll;
Tejun Heoc525aad2013-12-11 14:11:55 -050038 struct list_head files; /* goes through kernfs_open_file.list */
Tejun Heo414985a2013-11-28 14:54:34 -050039};
40
Tejun Heoc525aad2013-12-11 14:11:55 -050041static struct kernfs_open_file *kernfs_of(struct file *file)
Tejun Heo414985a2013-11-28 14:54:34 -050042{
43 return ((struct seq_file *)file->private_data)->private;
44}
45
46/*
Tejun Heo324a56e2013-12-11 14:11:53 -050047 * Determine the kernfs_ops for the given kernfs_node. This function must
Tejun Heo414985a2013-11-28 14:54:34 -050048 * be called while holding an active reference.
49 */
Tejun Heo324a56e2013-12-11 14:11:53 -050050static const struct kernfs_ops *kernfs_ops(struct kernfs_node *kn)
Tejun Heo414985a2013-11-28 14:54:34 -050051{
Tejun Heodf23fc32013-12-11 14:11:56 -050052 if (kn->flags & KERNFS_LOCKDEP)
Tejun Heo324a56e2013-12-11 14:11:53 -050053 lockdep_assert_held(kn);
Tejun Heoadc5e8b2013-12-11 14:11:54 -050054 return kn->attr.ops;
Tejun Heo414985a2013-11-28 14:54:34 -050055}
56
Tejun Heobb305942014-01-14 09:52:01 -050057/*
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.
66 *
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).
72 *
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.
78 */
79static void kernfs_seq_stop_active(struct seq_file *sf, void *v)
80{
81 struct kernfs_open_file *of = sf->private;
82 const struct kernfs_ops *ops = kernfs_ops(of->kn);
83
84 if (ops->seq_stop)
85 ops->seq_stop(sf, v);
86 kernfs_put_active(of->kn);
87}
88
Tejun Heo414985a2013-11-28 14:54:34 -050089static void *kernfs_seq_start(struct seq_file *sf, loff_t *ppos)
90{
Tejun Heoc525aad2013-12-11 14:11:55 -050091 struct kernfs_open_file *of = sf->private;
Tejun Heo414985a2013-11-28 14:54:34 -050092 const struct kernfs_ops *ops;
93
94 /*
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.
97 */
98 mutex_lock(&of->mutex);
Tejun Heoc637b8a2013-12-11 14:11:58 -050099 if (!kernfs_get_active(of->kn))
Tejun Heo414985a2013-11-28 14:54:34 -0500100 return ERR_PTR(-ENODEV);
101
Tejun Heo324a56e2013-12-11 14:11:53 -0500102 ops = kernfs_ops(of->kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500103 if (ops->seq_start) {
Tejun Heobb305942014-01-14 09:52:01 -0500104 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);
108 return next;
Tejun Heo414985a2013-11-28 14:54:34 -0500109 } else {
110 /*
111 * The same behavior and code as single_open(). Returns
112 * !NULL if pos is at the beginning; otherwise, NULL.
113 */
114 return NULL + !*ppos;
115 }
116}
117
118static void *kernfs_seq_next(struct seq_file *sf, void *v, loff_t *ppos)
119{
Tejun Heoc525aad2013-12-11 14:11:55 -0500120 struct kernfs_open_file *of = sf->private;
Tejun Heo324a56e2013-12-11 14:11:53 -0500121 const struct kernfs_ops *ops = kernfs_ops(of->kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500122
123 if (ops->seq_next) {
Tejun Heobb305942014-01-14 09:52:01 -0500124 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);
128 return next;
Tejun Heo414985a2013-11-28 14:54:34 -0500129 } else {
130 /*
131 * The same behavior and code as single_open(), always
132 * terminate after the initial read.
133 */
134 ++*ppos;
135 return NULL;
136 }
137}
138
139static void kernfs_seq_stop(struct seq_file *sf, void *v)
140{
Tejun Heoc525aad2013-12-11 14:11:55 -0500141 struct kernfs_open_file *of = sf->private;
Tejun Heo414985a2013-11-28 14:54:34 -0500142
Tejun Heobb305942014-01-14 09:52:01 -0500143 if (v != ERR_PTR(-ENODEV))
144 kernfs_seq_stop_active(sf, v);
Tejun Heo414985a2013-11-28 14:54:34 -0500145 mutex_unlock(&of->mutex);
146}
147
148static int kernfs_seq_show(struct seq_file *sf, void *v)
149{
Tejun Heoc525aad2013-12-11 14:11:55 -0500150 struct kernfs_open_file *of = sf->private;
Tejun Heo414985a2013-11-28 14:54:34 -0500151
Tejun Heoadc5e8b2013-12-11 14:11:54 -0500152 of->event = atomic_read(&of->kn->attr.open->event);
Tejun Heo414985a2013-11-28 14:54:34 -0500153
Tejun Heoadc5e8b2013-12-11 14:11:54 -0500154 return of->kn->attr.ops->seq_show(sf, v);
Tejun Heo414985a2013-11-28 14:54:34 -0500155}
156
157static 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,
162};
163
164/*
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
168 * bin files.
169 */
Tejun Heoc525aad2013-12-11 14:11:55 -0500170static ssize_t kernfs_file_direct_read(struct kernfs_open_file *of,
Tejun Heo414985a2013-11-28 14:54:34 -0500171 char __user *user_buf, size_t count,
172 loff_t *ppos)
173{
174 ssize_t len = min_t(size_t, count, PAGE_SIZE);
175 const struct kernfs_ops *ops;
176 char *buf;
177
178 buf = kmalloc(len, GFP_KERNEL);
179 if (!buf)
180 return -ENOMEM;
181
182 /*
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.
185 */
186 mutex_lock(&of->mutex);
Tejun Heoc637b8a2013-12-11 14:11:58 -0500187 if (!kernfs_get_active(of->kn)) {
Tejun Heo414985a2013-11-28 14:54:34 -0500188 len = -ENODEV;
189 mutex_unlock(&of->mutex);
190 goto out_free;
191 }
192
Tejun Heo324a56e2013-12-11 14:11:53 -0500193 ops = kernfs_ops(of->kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500194 if (ops->read)
195 len = ops->read(of, buf, len, *ppos);
196 else
197 len = -EINVAL;
198
Tejun Heoc637b8a2013-12-11 14:11:58 -0500199 kernfs_put_active(of->kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500200 mutex_unlock(&of->mutex);
201
202 if (len < 0)
203 goto out_free;
204
205 if (copy_to_user(user_buf, buf, len)) {
206 len = -EFAULT;
207 goto out_free;
208 }
209
210 *ppos += len;
211
212 out_free:
213 kfree(buf);
214 return len;
215}
216
217/**
Tejun Heoc637b8a2013-12-11 14:11:58 -0500218 * kernfs_fop_read - kernfs vfs read callback
Tejun Heo414985a2013-11-28 14:54:34 -0500219 * @file: file pointer
220 * @user_buf: data to write
221 * @count: number of bytes
222 * @ppos: starting offset
223 */
Tejun Heoc637b8a2013-12-11 14:11:58 -0500224static ssize_t kernfs_fop_read(struct file *file, char __user *user_buf,
225 size_t count, loff_t *ppos)
Tejun Heo414985a2013-11-28 14:54:34 -0500226{
Tejun Heoc525aad2013-12-11 14:11:55 -0500227 struct kernfs_open_file *of = kernfs_of(file);
Tejun Heo414985a2013-11-28 14:54:34 -0500228
Tejun Heodf23fc32013-12-11 14:11:56 -0500229 if (of->kn->flags & KERNFS_HAS_SEQ_SHOW)
Tejun Heo414985a2013-11-28 14:54:34 -0500230 return seq_read(file, user_buf, count, ppos);
231 else
232 return kernfs_file_direct_read(of, user_buf, count, ppos);
233}
234
235/**
Tejun Heoc637b8a2013-12-11 14:11:58 -0500236 * kernfs_fop_write - kernfs vfs write callback
Tejun Heo414985a2013-11-28 14:54:34 -0500237 * @file: file pointer
238 * @user_buf: data to write
239 * @count: number of bytes
240 * @ppos: starting offset
241 *
242 * Copy data in from userland and pass it to the matching kernfs write
243 * operation.
244 *
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
249 * back.
250 */
Tejun Heoc637b8a2013-12-11 14:11:58 -0500251static ssize_t kernfs_fop_write(struct file *file, const char __user *user_buf,
252 size_t count, loff_t *ppos)
Tejun Heo414985a2013-11-28 14:54:34 -0500253{
Tejun Heoc525aad2013-12-11 14:11:55 -0500254 struct kernfs_open_file *of = kernfs_of(file);
Tejun Heo414985a2013-11-28 14:54:34 -0500255 ssize_t len = min_t(size_t, count, PAGE_SIZE);
256 const struct kernfs_ops *ops;
257 char *buf;
258
259 buf = kmalloc(len + 1, GFP_KERNEL);
260 if (!buf)
261 return -ENOMEM;
262
263 if (copy_from_user(buf, user_buf, len)) {
264 len = -EFAULT;
265 goto out_free;
266 }
267 buf[len] = '\0'; /* guarantee string termination */
268
269 /*
270 * @of->mutex nests outside active ref and is just to ensure that
271 * the ops aren't called concurrently for the same open file.
272 */
273 mutex_lock(&of->mutex);
Tejun Heoc637b8a2013-12-11 14:11:58 -0500274 if (!kernfs_get_active(of->kn)) {
Tejun Heo414985a2013-11-28 14:54:34 -0500275 mutex_unlock(&of->mutex);
276 len = -ENODEV;
277 goto out_free;
278 }
279
Tejun Heo324a56e2013-12-11 14:11:53 -0500280 ops = kernfs_ops(of->kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500281 if (ops->write)
282 len = ops->write(of, buf, len, *ppos);
283 else
284 len = -EINVAL;
285
Tejun Heoc637b8a2013-12-11 14:11:58 -0500286 kernfs_put_active(of->kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500287 mutex_unlock(&of->mutex);
288
289 if (len > 0)
290 *ppos += len;
291out_free:
292 kfree(buf);
293 return len;
294}
295
296static void kernfs_vma_open(struct vm_area_struct *vma)
297{
298 struct file *file = vma->vm_file;
Tejun Heoc525aad2013-12-11 14:11:55 -0500299 struct kernfs_open_file *of = kernfs_of(file);
Tejun Heo414985a2013-11-28 14:54:34 -0500300
301 if (!of->vm_ops)
302 return;
303
Tejun Heoc637b8a2013-12-11 14:11:58 -0500304 if (!kernfs_get_active(of->kn))
Tejun Heo414985a2013-11-28 14:54:34 -0500305 return;
306
307 if (of->vm_ops->open)
308 of->vm_ops->open(vma);
309
Tejun Heoc637b8a2013-12-11 14:11:58 -0500310 kernfs_put_active(of->kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500311}
312
313static int kernfs_vma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
314{
315 struct file *file = vma->vm_file;
Tejun Heoc525aad2013-12-11 14:11:55 -0500316 struct kernfs_open_file *of = kernfs_of(file);
Tejun Heo414985a2013-11-28 14:54:34 -0500317 int ret;
318
319 if (!of->vm_ops)
320 return VM_FAULT_SIGBUS;
321
Tejun Heoc637b8a2013-12-11 14:11:58 -0500322 if (!kernfs_get_active(of->kn))
Tejun Heo414985a2013-11-28 14:54:34 -0500323 return VM_FAULT_SIGBUS;
324
325 ret = VM_FAULT_SIGBUS;
326 if (of->vm_ops->fault)
327 ret = of->vm_ops->fault(vma, vmf);
328
Tejun Heoc637b8a2013-12-11 14:11:58 -0500329 kernfs_put_active(of->kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500330 return ret;
331}
332
333static int kernfs_vma_page_mkwrite(struct vm_area_struct *vma,
334 struct vm_fault *vmf)
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 int ret;
339
340 if (!of->vm_ops)
341 return VM_FAULT_SIGBUS;
342
Tejun Heoc637b8a2013-12-11 14:11:58 -0500343 if (!kernfs_get_active(of->kn))
Tejun Heo414985a2013-11-28 14:54:34 -0500344 return VM_FAULT_SIGBUS;
345
346 ret = 0;
347 if (of->vm_ops->page_mkwrite)
348 ret = of->vm_ops->page_mkwrite(vma, vmf);
349 else
350 file_update_time(file);
351
Tejun Heoc637b8a2013-12-11 14:11:58 -0500352 kernfs_put_active(of->kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500353 return ret;
354}
355
356static int kernfs_vma_access(struct vm_area_struct *vma, unsigned long addr,
357 void *buf, int len, int write)
358{
359 struct file *file = vma->vm_file;
Tejun Heoc525aad2013-12-11 14:11:55 -0500360 struct kernfs_open_file *of = kernfs_of(file);
Tejun Heo414985a2013-11-28 14:54:34 -0500361 int ret;
362
363 if (!of->vm_ops)
364 return -EINVAL;
365
Tejun Heoc637b8a2013-12-11 14:11:58 -0500366 if (!kernfs_get_active(of->kn))
Tejun Heo414985a2013-11-28 14:54:34 -0500367 return -EINVAL;
368
369 ret = -EINVAL;
370 if (of->vm_ops->access)
371 ret = of->vm_ops->access(vma, addr, buf, len, write);
372
Tejun Heoc637b8a2013-12-11 14:11:58 -0500373 kernfs_put_active(of->kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500374 return ret;
375}
376
377#ifdef CONFIG_NUMA
378static int kernfs_vma_set_policy(struct vm_area_struct *vma,
379 struct mempolicy *new)
380{
381 struct file *file = vma->vm_file;
Tejun Heoc525aad2013-12-11 14:11:55 -0500382 struct kernfs_open_file *of = kernfs_of(file);
Tejun Heo414985a2013-11-28 14:54:34 -0500383 int ret;
384
385 if (!of->vm_ops)
386 return 0;
387
Tejun Heoc637b8a2013-12-11 14:11:58 -0500388 if (!kernfs_get_active(of->kn))
Tejun Heo414985a2013-11-28 14:54:34 -0500389 return -EINVAL;
390
391 ret = 0;
392 if (of->vm_ops->set_policy)
393 ret = of->vm_ops->set_policy(vma, new);
394
Tejun Heoc637b8a2013-12-11 14:11:58 -0500395 kernfs_put_active(of->kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500396 return ret;
397}
398
399static struct mempolicy *kernfs_vma_get_policy(struct vm_area_struct *vma,
400 unsigned long addr)
401{
402 struct file *file = vma->vm_file;
Tejun Heoc525aad2013-12-11 14:11:55 -0500403 struct kernfs_open_file *of = kernfs_of(file);
Tejun Heo414985a2013-11-28 14:54:34 -0500404 struct mempolicy *pol;
405
406 if (!of->vm_ops)
407 return vma->vm_policy;
408
Tejun Heoc637b8a2013-12-11 14:11:58 -0500409 if (!kernfs_get_active(of->kn))
Tejun Heo414985a2013-11-28 14:54:34 -0500410 return vma->vm_policy;
411
412 pol = vma->vm_policy;
413 if (of->vm_ops->get_policy)
414 pol = of->vm_ops->get_policy(vma, addr);
415
Tejun Heoc637b8a2013-12-11 14:11:58 -0500416 kernfs_put_active(of->kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500417 return pol;
418}
419
420static int kernfs_vma_migrate(struct vm_area_struct *vma,
421 const nodemask_t *from, const nodemask_t *to,
422 unsigned long flags)
423{
424 struct file *file = vma->vm_file;
Tejun Heoc525aad2013-12-11 14:11:55 -0500425 struct kernfs_open_file *of = kernfs_of(file);
Tejun Heo414985a2013-11-28 14:54:34 -0500426 int ret;
427
428 if (!of->vm_ops)
429 return 0;
430
Tejun Heoc637b8a2013-12-11 14:11:58 -0500431 if (!kernfs_get_active(of->kn))
Tejun Heo414985a2013-11-28 14:54:34 -0500432 return 0;
433
434 ret = 0;
435 if (of->vm_ops->migrate)
436 ret = of->vm_ops->migrate(vma, from, to, flags);
437
Tejun Heoc637b8a2013-12-11 14:11:58 -0500438 kernfs_put_active(of->kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500439 return ret;
440}
441#endif
442
443static const struct vm_operations_struct kernfs_vm_ops = {
444 .open = kernfs_vma_open,
445 .fault = kernfs_vma_fault,
446 .page_mkwrite = kernfs_vma_page_mkwrite,
447 .access = kernfs_vma_access,
448#ifdef CONFIG_NUMA
449 .set_policy = kernfs_vma_set_policy,
450 .get_policy = kernfs_vma_get_policy,
451 .migrate = kernfs_vma_migrate,
452#endif
453};
454
Tejun Heoc637b8a2013-12-11 14:11:58 -0500455static int kernfs_fop_mmap(struct file *file, struct vm_area_struct *vma)
Tejun Heo414985a2013-11-28 14:54:34 -0500456{
Tejun Heoc525aad2013-12-11 14:11:55 -0500457 struct kernfs_open_file *of = kernfs_of(file);
Tejun Heo414985a2013-11-28 14:54:34 -0500458 const struct kernfs_ops *ops;
459 int rc;
460
Tejun Heo9b2db6e2013-12-10 09:29:17 -0500461 /*
462 * mmap path and of->mutex are prone to triggering spurious lockdep
463 * warnings and we don't want to add spurious locking dependency
464 * between the two. Check whether mmap is actually implemented
465 * without grabbing @of->mutex by testing HAS_MMAP flag. See the
466 * comment in kernfs_file_open() for more details.
467 */
Tejun Heodf23fc32013-12-11 14:11:56 -0500468 if (!(of->kn->flags & KERNFS_HAS_MMAP))
Tejun Heo9b2db6e2013-12-10 09:29:17 -0500469 return -ENODEV;
470
Tejun Heo414985a2013-11-28 14:54:34 -0500471 mutex_lock(&of->mutex);
472
473 rc = -ENODEV;
Tejun Heoc637b8a2013-12-11 14:11:58 -0500474 if (!kernfs_get_active(of->kn))
Tejun Heo414985a2013-11-28 14:54:34 -0500475 goto out_unlock;
476
Tejun Heo324a56e2013-12-11 14:11:53 -0500477 ops = kernfs_ops(of->kn);
Tejun Heo9b2db6e2013-12-10 09:29:17 -0500478 rc = ops->mmap(of, vma);
Tejun Heo414985a2013-11-28 14:54:34 -0500479
480 /*
481 * PowerPC's pci_mmap of legacy_mem uses shmem_zero_setup()
482 * to satisfy versions of X which crash if the mmap fails: that
483 * substitutes a new vm_file, and we don't then want bin_vm_ops.
484 */
485 if (vma->vm_file != file)
486 goto out_put;
487
488 rc = -EINVAL;
489 if (of->mmapped && of->vm_ops != vma->vm_ops)
490 goto out_put;
491
492 /*
493 * It is not possible to successfully wrap close.
494 * So error if someone is trying to use close.
495 */
496 rc = -EINVAL;
497 if (vma->vm_ops && vma->vm_ops->close)
498 goto out_put;
499
500 rc = 0;
501 of->mmapped = 1;
502 of->vm_ops = vma->vm_ops;
503 vma->vm_ops = &kernfs_vm_ops;
504out_put:
Tejun Heoc637b8a2013-12-11 14:11:58 -0500505 kernfs_put_active(of->kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500506out_unlock:
507 mutex_unlock(&of->mutex);
508
509 return rc;
510}
511
512/**
Tejun Heoc637b8a2013-12-11 14:11:58 -0500513 * kernfs_get_open_node - get or create kernfs_open_node
Tejun Heo324a56e2013-12-11 14:11:53 -0500514 * @kn: target kernfs_node
Tejun Heoc525aad2013-12-11 14:11:55 -0500515 * @of: kernfs_open_file for this instance of open
Tejun Heo414985a2013-11-28 14:54:34 -0500516 *
Tejun Heoadc5e8b2013-12-11 14:11:54 -0500517 * If @kn->attr.open exists, increment its reference count; otherwise,
518 * create one. @of is chained to the files list.
Tejun Heo414985a2013-11-28 14:54:34 -0500519 *
520 * LOCKING:
521 * Kernel thread context (may sleep).
522 *
523 * RETURNS:
524 * 0 on success, -errno on failure.
525 */
Tejun Heoc637b8a2013-12-11 14:11:58 -0500526static int kernfs_get_open_node(struct kernfs_node *kn,
527 struct kernfs_open_file *of)
Tejun Heo414985a2013-11-28 14:54:34 -0500528{
Tejun Heoc525aad2013-12-11 14:11:55 -0500529 struct kernfs_open_node *on, *new_on = NULL;
Tejun Heo414985a2013-11-28 14:54:34 -0500530
531 retry:
Tejun Heoc525aad2013-12-11 14:11:55 -0500532 mutex_lock(&kernfs_open_file_mutex);
533 spin_lock_irq(&kernfs_open_node_lock);
Tejun Heo414985a2013-11-28 14:54:34 -0500534
Tejun Heoc525aad2013-12-11 14:11:55 -0500535 if (!kn->attr.open && new_on) {
536 kn->attr.open = new_on;
537 new_on = NULL;
Tejun Heo414985a2013-11-28 14:54:34 -0500538 }
539
Tejun Heoc525aad2013-12-11 14:11:55 -0500540 on = kn->attr.open;
541 if (on) {
542 atomic_inc(&on->refcnt);
543 list_add_tail(&of->list, &on->files);
Tejun Heo414985a2013-11-28 14:54:34 -0500544 }
545
Tejun Heoc525aad2013-12-11 14:11:55 -0500546 spin_unlock_irq(&kernfs_open_node_lock);
547 mutex_unlock(&kernfs_open_file_mutex);
Tejun Heo414985a2013-11-28 14:54:34 -0500548
Tejun Heoc525aad2013-12-11 14:11:55 -0500549 if (on) {
550 kfree(new_on);
Tejun Heo414985a2013-11-28 14:54:34 -0500551 return 0;
552 }
553
554 /* not there, initialize a new one and retry */
Tejun Heoc525aad2013-12-11 14:11:55 -0500555 new_on = kmalloc(sizeof(*new_on), GFP_KERNEL);
556 if (!new_on)
Tejun Heo414985a2013-11-28 14:54:34 -0500557 return -ENOMEM;
558
Tejun Heoc525aad2013-12-11 14:11:55 -0500559 atomic_set(&new_on->refcnt, 0);
560 atomic_set(&new_on->event, 1);
561 init_waitqueue_head(&new_on->poll);
562 INIT_LIST_HEAD(&new_on->files);
Tejun Heo414985a2013-11-28 14:54:34 -0500563 goto retry;
564}
565
566/**
Tejun Heoc637b8a2013-12-11 14:11:58 -0500567 * kernfs_put_open_node - put kernfs_open_node
Tejun Heo324a56e2013-12-11 14:11:53 -0500568 * @kn: target kernfs_nodet
Tejun Heoc525aad2013-12-11 14:11:55 -0500569 * @of: associated kernfs_open_file
Tejun Heo414985a2013-11-28 14:54:34 -0500570 *
Tejun Heoadc5e8b2013-12-11 14:11:54 -0500571 * Put @kn->attr.open and unlink @of from the files list. If
Tejun Heo414985a2013-11-28 14:54:34 -0500572 * reference count reaches zero, disassociate and free it.
573 *
574 * LOCKING:
575 * None.
576 */
Tejun Heoc637b8a2013-12-11 14:11:58 -0500577static void kernfs_put_open_node(struct kernfs_node *kn,
578 struct kernfs_open_file *of)
Tejun Heo414985a2013-11-28 14:54:34 -0500579{
Tejun Heoc525aad2013-12-11 14:11:55 -0500580 struct kernfs_open_node *on = kn->attr.open;
Tejun Heo414985a2013-11-28 14:54:34 -0500581 unsigned long flags;
582
Tejun Heoc525aad2013-12-11 14:11:55 -0500583 mutex_lock(&kernfs_open_file_mutex);
584 spin_lock_irqsave(&kernfs_open_node_lock, flags);
Tejun Heo414985a2013-11-28 14:54:34 -0500585
586 if (of)
587 list_del(&of->list);
588
Tejun Heoc525aad2013-12-11 14:11:55 -0500589 if (atomic_dec_and_test(&on->refcnt))
Tejun Heoadc5e8b2013-12-11 14:11:54 -0500590 kn->attr.open = NULL;
Tejun Heo414985a2013-11-28 14:54:34 -0500591 else
Tejun Heoc525aad2013-12-11 14:11:55 -0500592 on = NULL;
Tejun Heo414985a2013-11-28 14:54:34 -0500593
Tejun Heoc525aad2013-12-11 14:11:55 -0500594 spin_unlock_irqrestore(&kernfs_open_node_lock, flags);
595 mutex_unlock(&kernfs_open_file_mutex);
Tejun Heo414985a2013-11-28 14:54:34 -0500596
Tejun Heoc525aad2013-12-11 14:11:55 -0500597 kfree(on);
Tejun Heo414985a2013-11-28 14:54:34 -0500598}
599
Tejun Heoc637b8a2013-12-11 14:11:58 -0500600static int kernfs_fop_open(struct inode *inode, struct file *file)
Tejun Heo414985a2013-11-28 14:54:34 -0500601{
Tejun Heo324a56e2013-12-11 14:11:53 -0500602 struct kernfs_node *kn = file->f_path.dentry->d_fsdata;
Tejun Heo414985a2013-11-28 14:54:34 -0500603 const struct kernfs_ops *ops;
Tejun Heoc525aad2013-12-11 14:11:55 -0500604 struct kernfs_open_file *of;
Tejun Heo414985a2013-11-28 14:54:34 -0500605 bool has_read, has_write, has_mmap;
606 int error = -EACCES;
607
Tejun Heoc637b8a2013-12-11 14:11:58 -0500608 if (!kernfs_get_active(kn))
Tejun Heo414985a2013-11-28 14:54:34 -0500609 return -ENODEV;
610
Tejun Heo324a56e2013-12-11 14:11:53 -0500611 ops = kernfs_ops(kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500612
613 has_read = ops->seq_show || ops->read || ops->mmap;
614 has_write = ops->write || ops->mmap;
615 has_mmap = ops->mmap;
616
617 /* check perms and supported operations */
618 if ((file->f_mode & FMODE_WRITE) &&
619 (!(inode->i_mode & S_IWUGO) || !has_write))
620 goto err_out;
621
622 if ((file->f_mode & FMODE_READ) &&
623 (!(inode->i_mode & S_IRUGO) || !has_read))
624 goto err_out;
625
Tejun Heoc525aad2013-12-11 14:11:55 -0500626 /* allocate a kernfs_open_file for the file */
Tejun Heo414985a2013-11-28 14:54:34 -0500627 error = -ENOMEM;
Tejun Heoc525aad2013-12-11 14:11:55 -0500628 of = kzalloc(sizeof(struct kernfs_open_file), GFP_KERNEL);
Tejun Heo414985a2013-11-28 14:54:34 -0500629 if (!of)
630 goto err_out;
631
632 /*
633 * The following is done to give a different lockdep key to
634 * @of->mutex for files which implement mmap. This is a rather
635 * crude way to avoid false positive lockdep warning around
636 * mm->mmap_sem - mmap nests @of->mutex under mm->mmap_sem and
637 * reading /sys/block/sda/trace/act_mask grabs sr_mutex, under
638 * which mm->mmap_sem nests, while holding @of->mutex. As each
639 * open file has a separate mutex, it's okay as long as those don't
640 * happen on the same file. At this point, we can't easily give
641 * each file a separate locking class. Let's differentiate on
642 * whether the file has mmap or not for now.
Tejun Heo9b2db6e2013-12-10 09:29:17 -0500643 *
644 * Both paths of the branch look the same. They're supposed to
645 * look that way and give @of->mutex different static lockdep keys.
Tejun Heo414985a2013-11-28 14:54:34 -0500646 */
647 if (has_mmap)
648 mutex_init(&of->mutex);
649 else
650 mutex_init(&of->mutex);
651
Tejun Heo324a56e2013-12-11 14:11:53 -0500652 of->kn = kn;
Tejun Heo414985a2013-11-28 14:54:34 -0500653 of->file = file;
654
655 /*
656 * Always instantiate seq_file even if read access doesn't use
657 * seq_file or is not requested. This unifies private data access
658 * and readable regular files are the vast majority anyway.
659 */
660 if (ops->seq_show)
661 error = seq_open(file, &kernfs_seq_ops);
662 else
663 error = seq_open(file, NULL);
664 if (error)
665 goto err_free;
666
667 ((struct seq_file *)file->private_data)->private = of;
668
669 /* seq_file clears PWRITE unconditionally, restore it if WRITE */
670 if (file->f_mode & FMODE_WRITE)
671 file->f_mode |= FMODE_PWRITE;
672
Tejun Heoc637b8a2013-12-11 14:11:58 -0500673 /* make sure we have open node struct */
674 error = kernfs_get_open_node(kn, of);
Tejun Heo414985a2013-11-28 14:54:34 -0500675 if (error)
676 goto err_close;
677
678 /* open succeeded, put active references */
Tejun Heoc637b8a2013-12-11 14:11:58 -0500679 kernfs_put_active(kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500680 return 0;
681
682err_close:
683 seq_release(inode, file);
684err_free:
685 kfree(of);
686err_out:
Tejun Heoc637b8a2013-12-11 14:11:58 -0500687 kernfs_put_active(kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500688 return error;
689}
690
Tejun Heoc637b8a2013-12-11 14:11:58 -0500691static int kernfs_fop_release(struct inode *inode, struct file *filp)
Tejun Heo414985a2013-11-28 14:54:34 -0500692{
Tejun Heo324a56e2013-12-11 14:11:53 -0500693 struct kernfs_node *kn = filp->f_path.dentry->d_fsdata;
Tejun Heoc525aad2013-12-11 14:11:55 -0500694 struct kernfs_open_file *of = kernfs_of(filp);
Tejun Heo414985a2013-11-28 14:54:34 -0500695
Tejun Heoc637b8a2013-12-11 14:11:58 -0500696 kernfs_put_open_node(kn, of);
Tejun Heo414985a2013-11-28 14:54:34 -0500697 seq_release(inode, filp);
698 kfree(of);
699
700 return 0;
701}
702
Greg Kroah-Hartman55f6e302014-01-13 14:27:16 -0800703void kernfs_unmap_bin_file(struct kernfs_node *kn)
Tejun Heo414985a2013-11-28 14:54:34 -0500704{
Tejun Heoc525aad2013-12-11 14:11:55 -0500705 struct kernfs_open_node *on;
706 struct kernfs_open_file *of;
Tejun Heo414985a2013-11-28 14:54:34 -0500707
Greg Kroah-Hartman55f6e302014-01-13 14:27:16 -0800708 if (!(kn->flags & KERNFS_HAS_MMAP))
709 return;
710
Tejun Heoc525aad2013-12-11 14:11:55 -0500711 spin_lock_irq(&kernfs_open_node_lock);
712 on = kn->attr.open;
713 if (on)
714 atomic_inc(&on->refcnt);
715 spin_unlock_irq(&kernfs_open_node_lock);
716 if (!on)
Tejun Heo414985a2013-11-28 14:54:34 -0500717 return;
718
Tejun Heoc525aad2013-12-11 14:11:55 -0500719 mutex_lock(&kernfs_open_file_mutex);
720 list_for_each_entry(of, &on->files, list) {
Tejun Heo414985a2013-11-28 14:54:34 -0500721 struct inode *inode = file_inode(of->file);
722 unmap_mapping_range(inode->i_mapping, 0, 0, 1);
723 }
Tejun Heoc525aad2013-12-11 14:11:55 -0500724 mutex_unlock(&kernfs_open_file_mutex);
Tejun Heo414985a2013-11-28 14:54:34 -0500725
Tejun Heoc637b8a2013-12-11 14:11:58 -0500726 kernfs_put_open_node(kn, NULL);
Tejun Heo414985a2013-11-28 14:54:34 -0500727}
728
Tejun Heoc637b8a2013-12-11 14:11:58 -0500729/*
730 * Kernfs attribute files are pollable. The idea is that you read
Tejun Heo414985a2013-11-28 14:54:34 -0500731 * the content and then you use 'poll' or 'select' to wait for
732 * the content to change. When the content changes (assuming the
733 * manager for the kobject supports notification), poll will
734 * return POLLERR|POLLPRI, and select will return the fd whether
735 * it is waiting for read, write, or exceptions.
736 * Once poll/select indicates that the value has changed, you
737 * need to close and re-open the file, or seek to 0 and read again.
738 * Reminder: this only works for attributes which actively support
739 * it, and it is not possible to test an attribute from userspace
740 * to see if it supports poll (Neither 'poll' nor 'select' return
741 * an appropriate error code). When in doubt, set a suitable timeout value.
742 */
Tejun Heoc637b8a2013-12-11 14:11:58 -0500743static unsigned int kernfs_fop_poll(struct file *filp, poll_table *wait)
Tejun Heo414985a2013-11-28 14:54:34 -0500744{
Tejun Heoc525aad2013-12-11 14:11:55 -0500745 struct kernfs_open_file *of = kernfs_of(filp);
Tejun Heo324a56e2013-12-11 14:11:53 -0500746 struct kernfs_node *kn = filp->f_path.dentry->d_fsdata;
Tejun Heoc525aad2013-12-11 14:11:55 -0500747 struct kernfs_open_node *on = kn->attr.open;
Tejun Heo414985a2013-11-28 14:54:34 -0500748
749 /* need parent for the kobj, grab both */
Tejun Heoc637b8a2013-12-11 14:11:58 -0500750 if (!kernfs_get_active(kn))
Tejun Heo414985a2013-11-28 14:54:34 -0500751 goto trigger;
752
Tejun Heoc525aad2013-12-11 14:11:55 -0500753 poll_wait(filp, &on->poll, wait);
Tejun Heo414985a2013-11-28 14:54:34 -0500754
Tejun Heoc637b8a2013-12-11 14:11:58 -0500755 kernfs_put_active(kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500756
Tejun Heoc525aad2013-12-11 14:11:55 -0500757 if (of->event != atomic_read(&on->event))
Tejun Heo414985a2013-11-28 14:54:34 -0500758 goto trigger;
759
760 return DEFAULT_POLLMASK;
761
762 trigger:
763 return DEFAULT_POLLMASK|POLLERR|POLLPRI;
764}
765
766/**
767 * kernfs_notify - notify a kernfs file
Tejun Heo324a56e2013-12-11 14:11:53 -0500768 * @kn: file to notify
Tejun Heo414985a2013-11-28 14:54:34 -0500769 *
Tejun Heo324a56e2013-12-11 14:11:53 -0500770 * Notify @kn such that poll(2) on @kn wakes up.
Tejun Heo414985a2013-11-28 14:54:34 -0500771 */
Tejun Heo324a56e2013-12-11 14:11:53 -0500772void kernfs_notify(struct kernfs_node *kn)
Tejun Heo414985a2013-11-28 14:54:34 -0500773{
Tejun Heoc525aad2013-12-11 14:11:55 -0500774 struct kernfs_open_node *on;
Tejun Heo414985a2013-11-28 14:54:34 -0500775 unsigned long flags;
776
Tejun Heoc525aad2013-12-11 14:11:55 -0500777 spin_lock_irqsave(&kernfs_open_node_lock, flags);
Tejun Heo414985a2013-11-28 14:54:34 -0500778
Tejun Heodf23fc32013-12-11 14:11:56 -0500779 if (!WARN_ON(kernfs_type(kn) != KERNFS_FILE)) {
Tejun Heoc525aad2013-12-11 14:11:55 -0500780 on = kn->attr.open;
781 if (on) {
782 atomic_inc(&on->event);
783 wake_up_interruptible(&on->poll);
Tejun Heo414985a2013-11-28 14:54:34 -0500784 }
785 }
786
Tejun Heoc525aad2013-12-11 14:11:55 -0500787 spin_unlock_irqrestore(&kernfs_open_node_lock, flags);
Tejun Heo414985a2013-11-28 14:54:34 -0500788}
789EXPORT_SYMBOL_GPL(kernfs_notify);
790
Tejun Heoa797bfc2013-12-11 14:11:57 -0500791const struct file_operations kernfs_file_fops = {
Tejun Heoc637b8a2013-12-11 14:11:58 -0500792 .read = kernfs_fop_read,
793 .write = kernfs_fop_write,
Tejun Heo414985a2013-11-28 14:54:34 -0500794 .llseek = generic_file_llseek,
Tejun Heoc637b8a2013-12-11 14:11:58 -0500795 .mmap = kernfs_fop_mmap,
796 .open = kernfs_fop_open,
797 .release = kernfs_fop_release,
798 .poll = kernfs_fop_poll,
Tejun Heo414985a2013-11-28 14:54:34 -0500799};
800
801/**
Tejun Heo2063d602013-12-11 16:02:57 -0500802 * __kernfs_create_file - kernfs internal function to create a file
Tejun Heo414985a2013-11-28 14:54:34 -0500803 * @parent: directory to create the file in
804 * @name: name of the file
805 * @mode: mode of the file
806 * @size: size of the file
807 * @ops: kernfs operations for the file
808 * @priv: private data for the file
809 * @ns: optional namespace tag of the file
Tejun Heo2063d602013-12-11 16:02:57 -0500810 * @static_name: don't copy file name
Tejun Heo414985a2013-11-28 14:54:34 -0500811 * @key: lockdep key for the file's active_ref, %NULL to disable lockdep
812 *
813 * Returns the created node on success, ERR_PTR() value on error.
814 */
Tejun Heo2063d602013-12-11 16:02:57 -0500815struct kernfs_node *__kernfs_create_file(struct kernfs_node *parent,
816 const char *name,
817 umode_t mode, loff_t size,
818 const struct kernfs_ops *ops,
819 void *priv, const void *ns,
820 bool name_is_static,
821 struct lock_class_key *key)
Tejun Heo414985a2013-11-28 14:54:34 -0500822{
Greg Kroah-Hartman7653fe92014-01-13 14:20:56 -0800823 struct kernfs_addrm_cxt acxt;
Tejun Heo324a56e2013-12-11 14:11:53 -0500824 struct kernfs_node *kn;
Tejun Heo2063d602013-12-11 16:02:57 -0500825 unsigned flags;
Tejun Heo414985a2013-11-28 14:54:34 -0500826 int rc;
827
Tejun Heo2063d602013-12-11 16:02:57 -0500828 flags = KERNFS_FILE;
829 if (name_is_static)
830 flags |= KERNFS_STATIC_NAME;
831
Tejun Heodb4aad22014-01-17 09:58:25 -0500832 kn = kernfs_new_node(parent, name, (mode & S_IALLUGO) | S_IFREG, flags);
Tejun Heo324a56e2013-12-11 14:11:53 -0500833 if (!kn)
Tejun Heo414985a2013-11-28 14:54:34 -0500834 return ERR_PTR(-ENOMEM);
835
Tejun Heoadc5e8b2013-12-11 14:11:54 -0500836 kn->attr.ops = ops;
837 kn->attr.size = size;
838 kn->ns = ns;
Tejun Heo324a56e2013-12-11 14:11:53 -0500839 kn->priv = priv;
Tejun Heo414985a2013-11-28 14:54:34 -0500840
841#ifdef CONFIG_DEBUG_LOCK_ALLOC
842 if (key) {
Tejun Heo324a56e2013-12-11 14:11:53 -0500843 lockdep_init_map(&kn->dep_map, "s_active", key, 0);
Tejun Heodf23fc32013-12-11 14:11:56 -0500844 kn->flags |= KERNFS_LOCKDEP;
Tejun Heo414985a2013-11-28 14:54:34 -0500845 }
846#endif
847
848 /*
Tejun Heoadc5e8b2013-12-11 14:11:54 -0500849 * kn->attr.ops is accesible only while holding active ref. We
Tejun Heo414985a2013-11-28 14:54:34 -0500850 * need to know whether some ops are implemented outside active
851 * ref. Cache their existence in flags.
852 */
853 if (ops->seq_show)
Tejun Heodf23fc32013-12-11 14:11:56 -0500854 kn->flags |= KERNFS_HAS_SEQ_SHOW;
Tejun Heo414985a2013-11-28 14:54:34 -0500855 if (ops->mmap)
Tejun Heodf23fc32013-12-11 14:11:56 -0500856 kn->flags |= KERNFS_HAS_MMAP;
Tejun Heo414985a2013-11-28 14:54:34 -0500857
Greg Kroah-Hartman798c75a2014-01-13 14:36:03 -0800858 kernfs_addrm_start(&acxt);
Tejun Heodb4aad22014-01-17 09:58:25 -0500859 rc = kernfs_add_one(&acxt, kn);
Greg Kroah-Hartman798c75a2014-01-13 14:36:03 -0800860 kernfs_addrm_finish(&acxt);
Greg Kroah-Hartman7653fe92014-01-13 14:20:56 -0800861
Tejun Heo414985a2013-11-28 14:54:34 -0500862 if (rc) {
Tejun Heo324a56e2013-12-11 14:11:53 -0500863 kernfs_put(kn);
Tejun Heo414985a2013-11-28 14:54:34 -0500864 return ERR_PTR(rc);
865 }
Tejun Heo324a56e2013-12-11 14:11:53 -0500866 return kn;
Tejun Heo414985a2013-11-28 14:54:34 -0500867}