]>
Commit | Line | Data |
---|---|---|
b8441ed2 TH |
1 | /* |
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 | */ | |
414985ae TH |
10 | |
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> | |
414985ae TH |
16 | #include <linux/sched.h> |
17 | ||
18 | #include "kernfs-internal.h" | |
19 | ||
20 | /* | |
c525aadd | 21 | * There's one kernfs_open_file for each open file and one kernfs_open_node |
324a56e1 | 22 | * for each kernfs_node with one or more open files. |
414985ae | 23 | * |
c525aadd TH |
24 | * kernfs_node->attr.open points to kernfs_open_node. attr.open is |
25 | * protected by kernfs_open_node_lock. | |
414985ae TH |
26 | * |
27 | * filp->private_data points to seq_file whose ->private points to | |
c525aadd TH |
28 | * kernfs_open_file. kernfs_open_files are chained at |
29 | * kernfs_open_node->files, which is protected by kernfs_open_file_mutex. | |
414985ae | 30 | */ |
c525aadd TH |
31 | static DEFINE_SPINLOCK(kernfs_open_node_lock); |
32 | static DEFINE_MUTEX(kernfs_open_file_mutex); | |
414985ae | 33 | |
c525aadd | 34 | struct kernfs_open_node { |
414985ae TH |
35 | atomic_t refcnt; |
36 | atomic_t event; | |
37 | wait_queue_head_t poll; | |
c525aadd | 38 | struct list_head files; /* goes through kernfs_open_file.list */ |
414985ae TH |
39 | }; |
40 | ||
c525aadd | 41 | static struct kernfs_open_file *kernfs_of(struct file *file) |
414985ae TH |
42 | { |
43 | return ((struct seq_file *)file->private_data)->private; | |
44 | } | |
45 | ||
46 | /* | |
324a56e1 | 47 | * Determine the kernfs_ops for the given kernfs_node. This function must |
414985ae TH |
48 | * be called while holding an active reference. |
49 | */ | |
324a56e1 | 50 | static const struct kernfs_ops *kernfs_ops(struct kernfs_node *kn) |
414985ae | 51 | { |
df23fc39 | 52 | if (kn->flags & KERNFS_LOCKDEP) |
324a56e1 | 53 | lockdep_assert_held(kn); |
adc5e8b5 | 54 | return kn->attr.ops; |
414985ae TH |
55 | } |
56 | ||
bb305947 TH |
57 | /* |
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 | */ | |
79 | static 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 | ||
414985ae TH |
89 | static void *kernfs_seq_start(struct seq_file *sf, loff_t *ppos) |
90 | { | |
c525aadd | 91 | struct kernfs_open_file *of = sf->private; |
414985ae TH |
92 | 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); | |
c637b8ac | 99 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
100 | return ERR_PTR(-ENODEV); |
101 | ||
324a56e1 | 102 | ops = kernfs_ops(of->kn); |
414985ae | 103 | if (ops->seq_start) { |
bb305947 TH |
104 | void *next = ops->seq_start(sf, ppos); |
105 | /* see the comment above kernfs_seq_stop_active() */ | |
106 | if (next == ERR_PTR(-ENODEV)) | |
107 | kernfs_seq_stop_active(sf, next); | |
108 | return next; | |
414985ae TH |
109 | } 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 | ||
118 | static void *kernfs_seq_next(struct seq_file *sf, void *v, loff_t *ppos) | |
119 | { | |
c525aadd | 120 | struct kernfs_open_file *of = sf->private; |
324a56e1 | 121 | const struct kernfs_ops *ops = kernfs_ops(of->kn); |
414985ae TH |
122 | |
123 | if (ops->seq_next) { | |
bb305947 TH |
124 | void *next = ops->seq_next(sf, v, ppos); |
125 | /* see the comment above kernfs_seq_stop_active() */ | |
126 | if (next == ERR_PTR(-ENODEV)) | |
127 | kernfs_seq_stop_active(sf, next); | |
128 | return next; | |
414985ae TH |
129 | } 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 | ||
139 | static void kernfs_seq_stop(struct seq_file *sf, void *v) | |
140 | { | |
c525aadd | 141 | struct kernfs_open_file *of = sf->private; |
414985ae | 142 | |
bb305947 TH |
143 | if (v != ERR_PTR(-ENODEV)) |
144 | kernfs_seq_stop_active(sf, v); | |
414985ae TH |
145 | mutex_unlock(&of->mutex); |
146 | } | |
147 | ||
148 | static int kernfs_seq_show(struct seq_file *sf, void *v) | |
149 | { | |
c525aadd | 150 | struct kernfs_open_file *of = sf->private; |
414985ae | 151 | |
adc5e8b5 | 152 | of->event = atomic_read(&of->kn->attr.open->event); |
414985ae | 153 | |
adc5e8b5 | 154 | return of->kn->attr.ops->seq_show(sf, v); |
414985ae TH |
155 | } |
156 | ||
157 | static const struct seq_operations kernfs_seq_ops = { | |
158 | .start = kernfs_seq_start, | |
159 | .next = kernfs_seq_next, | |
160 | .stop = kernfs_seq_stop, | |
161 | .show = kernfs_seq_show, | |
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 | */ | |
c525aadd | 170 | static ssize_t kernfs_file_direct_read(struct kernfs_open_file *of, |
414985ae TH |
171 | 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); | |
c637b8ac | 187 | if (!kernfs_get_active(of->kn)) { |
414985ae TH |
188 | len = -ENODEV; |
189 | mutex_unlock(&of->mutex); | |
190 | goto out_free; | |
191 | } | |
192 | ||
324a56e1 | 193 | ops = kernfs_ops(of->kn); |
414985ae TH |
194 | if (ops->read) |
195 | len = ops->read(of, buf, len, *ppos); | |
196 | else | |
197 | len = -EINVAL; | |
198 | ||
c637b8ac | 199 | kernfs_put_active(of->kn); |
414985ae TH |
200 | 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 | /** | |
c637b8ac | 218 | * kernfs_fop_read - kernfs vfs read callback |
414985ae TH |
219 | * @file: file pointer |
220 | * @user_buf: data to write | |
221 | * @count: number of bytes | |
222 | * @ppos: starting offset | |
223 | */ | |
c637b8ac TH |
224 | static ssize_t kernfs_fop_read(struct file *file, char __user *user_buf, |
225 | size_t count, loff_t *ppos) | |
414985ae | 226 | { |
c525aadd | 227 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae | 228 | |
df23fc39 | 229 | if (of->kn->flags & KERNFS_HAS_SEQ_SHOW) |
414985ae TH |
230 | return seq_read(file, user_buf, count, ppos); |
231 | else | |
232 | return kernfs_file_direct_read(of, user_buf, count, ppos); | |
233 | } | |
234 | ||
235 | /** | |
c637b8ac | 236 | * kernfs_fop_write - kernfs vfs write callback |
414985ae TH |
237 | * @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 | */ | |
c637b8ac TH |
251 | static ssize_t kernfs_fop_write(struct file *file, const char __user *user_buf, |
252 | size_t count, loff_t *ppos) | |
414985ae | 253 | { |
c525aadd | 254 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
255 | 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); | |
c637b8ac | 274 | if (!kernfs_get_active(of->kn)) { |
414985ae TH |
275 | mutex_unlock(&of->mutex); |
276 | len = -ENODEV; | |
277 | goto out_free; | |
278 | } | |
279 | ||
324a56e1 | 280 | ops = kernfs_ops(of->kn); |
414985ae TH |
281 | if (ops->write) |
282 | len = ops->write(of, buf, len, *ppos); | |
283 | else | |
284 | len = -EINVAL; | |
285 | ||
c637b8ac | 286 | kernfs_put_active(of->kn); |
414985ae TH |
287 | mutex_unlock(&of->mutex); |
288 | ||
289 | if (len > 0) | |
290 | *ppos += len; | |
291 | out_free: | |
292 | kfree(buf); | |
293 | return len; | |
294 | } | |
295 | ||
296 | static void kernfs_vma_open(struct vm_area_struct *vma) | |
297 | { | |
298 | struct file *file = vma->vm_file; | |
c525aadd | 299 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
300 | |
301 | if (!of->vm_ops) | |
302 | return; | |
303 | ||
c637b8ac | 304 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
305 | return; |
306 | ||
307 | if (of->vm_ops->open) | |
308 | of->vm_ops->open(vma); | |
309 | ||
c637b8ac | 310 | kernfs_put_active(of->kn); |
414985ae TH |
311 | } |
312 | ||
313 | static int kernfs_vma_fault(struct vm_area_struct *vma, struct vm_fault *vmf) | |
314 | { | |
315 | struct file *file = vma->vm_file; | |
c525aadd | 316 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
317 | int ret; |
318 | ||
319 | if (!of->vm_ops) | |
320 | return VM_FAULT_SIGBUS; | |
321 | ||
c637b8ac | 322 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
323 | 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 | ||
c637b8ac | 329 | kernfs_put_active(of->kn); |
414985ae TH |
330 | return ret; |
331 | } | |
332 | ||
333 | static int kernfs_vma_page_mkwrite(struct vm_area_struct *vma, | |
334 | struct vm_fault *vmf) | |
335 | { | |
336 | struct file *file = vma->vm_file; | |
c525aadd | 337 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
338 | int ret; |
339 | ||
340 | if (!of->vm_ops) | |
341 | return VM_FAULT_SIGBUS; | |
342 | ||
c637b8ac | 343 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
344 | 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 | ||
c637b8ac | 352 | kernfs_put_active(of->kn); |
414985ae TH |
353 | return ret; |
354 | } | |
355 | ||
356 | static 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; | |
c525aadd | 360 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
361 | int ret; |
362 | ||
363 | if (!of->vm_ops) | |
364 | return -EINVAL; | |
365 | ||
c637b8ac | 366 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
367 | return -EINVAL; |
368 | ||
369 | ret = -EINVAL; | |
370 | if (of->vm_ops->access) | |
371 | ret = of->vm_ops->access(vma, addr, buf, len, write); | |
372 | ||
c637b8ac | 373 | kernfs_put_active(of->kn); |
414985ae TH |
374 | return ret; |
375 | } | |
376 | ||
377 | #ifdef CONFIG_NUMA | |
378 | static int kernfs_vma_set_policy(struct vm_area_struct *vma, | |
379 | struct mempolicy *new) | |
380 | { | |
381 | struct file *file = vma->vm_file; | |
c525aadd | 382 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
383 | int ret; |
384 | ||
385 | if (!of->vm_ops) | |
386 | return 0; | |
387 | ||
c637b8ac | 388 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
389 | return -EINVAL; |
390 | ||
391 | ret = 0; | |
392 | if (of->vm_ops->set_policy) | |
393 | ret = of->vm_ops->set_policy(vma, new); | |
394 | ||
c637b8ac | 395 | kernfs_put_active(of->kn); |
414985ae TH |
396 | return ret; |
397 | } | |
398 | ||
399 | static struct mempolicy *kernfs_vma_get_policy(struct vm_area_struct *vma, | |
400 | unsigned long addr) | |
401 | { | |
402 | struct file *file = vma->vm_file; | |
c525aadd | 403 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
404 | struct mempolicy *pol; |
405 | ||
406 | if (!of->vm_ops) | |
407 | return vma->vm_policy; | |
408 | ||
c637b8ac | 409 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
410 | 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 | ||
c637b8ac | 416 | kernfs_put_active(of->kn); |
414985ae TH |
417 | return pol; |
418 | } | |
419 | ||
420 | static 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; | |
c525aadd | 425 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
426 | int ret; |
427 | ||
428 | if (!of->vm_ops) | |
429 | return 0; | |
430 | ||
c637b8ac | 431 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
432 | return 0; |
433 | ||
434 | ret = 0; | |
435 | if (of->vm_ops->migrate) | |
436 | ret = of->vm_ops->migrate(vma, from, to, flags); | |
437 | ||
c637b8ac | 438 | kernfs_put_active(of->kn); |
414985ae TH |
439 | return ret; |
440 | } | |
441 | #endif | |
442 | ||
443 | static 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 | ||
c637b8ac | 455 | static int kernfs_fop_mmap(struct file *file, struct vm_area_struct *vma) |
414985ae | 456 | { |
c525aadd | 457 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
458 | const struct kernfs_ops *ops; |
459 | int rc; | |
460 | ||
9b2db6e1 TH |
461 | /* |
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 | */ | |
df23fc39 | 468 | if (!(of->kn->flags & KERNFS_HAS_MMAP)) |
9b2db6e1 TH |
469 | return -ENODEV; |
470 | ||
414985ae TH |
471 | mutex_lock(&of->mutex); |
472 | ||
473 | rc = -ENODEV; | |
c637b8ac | 474 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
475 | goto out_unlock; |
476 | ||
324a56e1 | 477 | ops = kernfs_ops(of->kn); |
9b2db6e1 | 478 | rc = ops->mmap(of, vma); |
414985ae TH |
479 | |
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; | |
504 | out_put: | |
c637b8ac | 505 | kernfs_put_active(of->kn); |
414985ae TH |
506 | out_unlock: |
507 | mutex_unlock(&of->mutex); | |
508 | ||
509 | return rc; | |
510 | } | |
511 | ||
512 | /** | |
c637b8ac | 513 | * kernfs_get_open_node - get or create kernfs_open_node |
324a56e1 | 514 | * @kn: target kernfs_node |
c525aadd | 515 | * @of: kernfs_open_file for this instance of open |
414985ae | 516 | * |
adc5e8b5 TH |
517 | * If @kn->attr.open exists, increment its reference count; otherwise, |
518 | * create one. @of is chained to the files list. | |
414985ae TH |
519 | * |
520 | * LOCKING: | |
521 | * Kernel thread context (may sleep). | |
522 | * | |
523 | * RETURNS: | |
524 | * 0 on success, -errno on failure. | |
525 | */ | |
c637b8ac TH |
526 | static int kernfs_get_open_node(struct kernfs_node *kn, |
527 | struct kernfs_open_file *of) | |
414985ae | 528 | { |
c525aadd | 529 | struct kernfs_open_node *on, *new_on = NULL; |
414985ae TH |
530 | |
531 | retry: | |
c525aadd TH |
532 | mutex_lock(&kernfs_open_file_mutex); |
533 | spin_lock_irq(&kernfs_open_node_lock); | |
414985ae | 534 | |
c525aadd TH |
535 | if (!kn->attr.open && new_on) { |
536 | kn->attr.open = new_on; | |
537 | new_on = NULL; | |
414985ae TH |
538 | } |
539 | ||
c525aadd TH |
540 | on = kn->attr.open; |
541 | if (on) { | |
542 | atomic_inc(&on->refcnt); | |
543 | list_add_tail(&of->list, &on->files); | |
414985ae TH |
544 | } |
545 | ||
c525aadd TH |
546 | spin_unlock_irq(&kernfs_open_node_lock); |
547 | mutex_unlock(&kernfs_open_file_mutex); | |
414985ae | 548 | |
c525aadd TH |
549 | if (on) { |
550 | kfree(new_on); | |
414985ae TH |
551 | return 0; |
552 | } | |
553 | ||
554 | /* not there, initialize a new one and retry */ | |
c525aadd TH |
555 | new_on = kmalloc(sizeof(*new_on), GFP_KERNEL); |
556 | if (!new_on) | |
414985ae TH |
557 | return -ENOMEM; |
558 | ||
c525aadd TH |
559 | 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); | |
414985ae TH |
563 | goto retry; |
564 | } | |
565 | ||
566 | /** | |
c637b8ac | 567 | * kernfs_put_open_node - put kernfs_open_node |
324a56e1 | 568 | * @kn: target kernfs_nodet |
c525aadd | 569 | * @of: associated kernfs_open_file |
414985ae | 570 | * |
adc5e8b5 | 571 | * Put @kn->attr.open and unlink @of from the files list. If |
414985ae TH |
572 | * reference count reaches zero, disassociate and free it. |
573 | * | |
574 | * LOCKING: | |
575 | * None. | |
576 | */ | |
c637b8ac TH |
577 | static void kernfs_put_open_node(struct kernfs_node *kn, |
578 | struct kernfs_open_file *of) | |
414985ae | 579 | { |
c525aadd | 580 | struct kernfs_open_node *on = kn->attr.open; |
414985ae TH |
581 | unsigned long flags; |
582 | ||
c525aadd TH |
583 | mutex_lock(&kernfs_open_file_mutex); |
584 | spin_lock_irqsave(&kernfs_open_node_lock, flags); | |
414985ae TH |
585 | |
586 | if (of) | |
587 | list_del(&of->list); | |
588 | ||
c525aadd | 589 | if (atomic_dec_and_test(&on->refcnt)) |
adc5e8b5 | 590 | kn->attr.open = NULL; |
414985ae | 591 | else |
c525aadd | 592 | on = NULL; |
414985ae | 593 | |
c525aadd TH |
594 | spin_unlock_irqrestore(&kernfs_open_node_lock, flags); |
595 | mutex_unlock(&kernfs_open_file_mutex); | |
414985ae | 596 | |
c525aadd | 597 | kfree(on); |
414985ae TH |
598 | } |
599 | ||
c637b8ac | 600 | static int kernfs_fop_open(struct inode *inode, struct file *file) |
414985ae | 601 | { |
324a56e1 | 602 | struct kernfs_node *kn = file->f_path.dentry->d_fsdata; |
414985ae | 603 | const struct kernfs_ops *ops; |
c525aadd | 604 | struct kernfs_open_file *of; |
414985ae TH |
605 | bool has_read, has_write, has_mmap; |
606 | int error = -EACCES; | |
607 | ||
c637b8ac | 608 | if (!kernfs_get_active(kn)) |
414985ae TH |
609 | return -ENODEV; |
610 | ||
324a56e1 | 611 | ops = kernfs_ops(kn); |
414985ae TH |
612 | |
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 | ||
c525aadd | 626 | /* allocate a kernfs_open_file for the file */ |
414985ae | 627 | error = -ENOMEM; |
c525aadd | 628 | of = kzalloc(sizeof(struct kernfs_open_file), GFP_KERNEL); |
414985ae TH |
629 | 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. | |
9b2db6e1 TH |
643 | * |
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. | |
414985ae TH |
646 | */ |
647 | if (has_mmap) | |
648 | mutex_init(&of->mutex); | |
649 | else | |
650 | mutex_init(&of->mutex); | |
651 | ||
324a56e1 | 652 | of->kn = kn; |
414985ae TH |
653 | 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 | ||
c637b8ac TH |
673 | /* make sure we have open node struct */ |
674 | error = kernfs_get_open_node(kn, of); | |
414985ae TH |
675 | if (error) |
676 | goto err_close; | |
677 | ||
678 | /* open succeeded, put active references */ | |
c637b8ac | 679 | kernfs_put_active(kn); |
414985ae TH |
680 | return 0; |
681 | ||
682 | err_close: | |
683 | seq_release(inode, file); | |
684 | err_free: | |
685 | kfree(of); | |
686 | err_out: | |
c637b8ac | 687 | kernfs_put_active(kn); |
414985ae TH |
688 | return error; |
689 | } | |
690 | ||
c637b8ac | 691 | static int kernfs_fop_release(struct inode *inode, struct file *filp) |
414985ae | 692 | { |
324a56e1 | 693 | struct kernfs_node *kn = filp->f_path.dentry->d_fsdata; |
c525aadd | 694 | struct kernfs_open_file *of = kernfs_of(filp); |
414985ae | 695 | |
c637b8ac | 696 | kernfs_put_open_node(kn, of); |
414985ae TH |
697 | seq_release(inode, filp); |
698 | kfree(of); | |
699 | ||
700 | return 0; | |
701 | } | |
702 | ||
55f6e30d | 703 | void kernfs_unmap_bin_file(struct kernfs_node *kn) |
414985ae | 704 | { |
c525aadd TH |
705 | struct kernfs_open_node *on; |
706 | struct kernfs_open_file *of; | |
414985ae | 707 | |
55f6e30d GKH |
708 | if (!(kn->flags & KERNFS_HAS_MMAP)) |
709 | return; | |
710 | ||
c525aadd TH |
711 | 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) | |
414985ae TH |
717 | return; |
718 | ||
c525aadd TH |
719 | mutex_lock(&kernfs_open_file_mutex); |
720 | list_for_each_entry(of, &on->files, list) { | |
414985ae TH |
721 | struct inode *inode = file_inode(of->file); |
722 | unmap_mapping_range(inode->i_mapping, 0, 0, 1); | |
723 | } | |
c525aadd | 724 | mutex_unlock(&kernfs_open_file_mutex); |
414985ae | 725 | |
c637b8ac | 726 | kernfs_put_open_node(kn, NULL); |
414985ae TH |
727 | } |
728 | ||
c637b8ac TH |
729 | /* |
730 | * Kernfs attribute files are pollable. The idea is that you read | |
414985ae TH |
731 | * 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 | */ | |
c637b8ac | 743 | static unsigned int kernfs_fop_poll(struct file *filp, poll_table *wait) |
414985ae | 744 | { |
c525aadd | 745 | struct kernfs_open_file *of = kernfs_of(filp); |
324a56e1 | 746 | struct kernfs_node *kn = filp->f_path.dentry->d_fsdata; |
c525aadd | 747 | struct kernfs_open_node *on = kn->attr.open; |
414985ae TH |
748 | |
749 | /* need parent for the kobj, grab both */ | |
c637b8ac | 750 | if (!kernfs_get_active(kn)) |
414985ae TH |
751 | goto trigger; |
752 | ||
c525aadd | 753 | poll_wait(filp, &on->poll, wait); |
414985ae | 754 | |
c637b8ac | 755 | kernfs_put_active(kn); |
414985ae | 756 | |
c525aadd | 757 | if (of->event != atomic_read(&on->event)) |
414985ae TH |
758 | 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 | |
324a56e1 | 768 | * @kn: file to notify |
414985ae | 769 | * |
324a56e1 | 770 | * Notify @kn such that poll(2) on @kn wakes up. |
414985ae | 771 | */ |
324a56e1 | 772 | void kernfs_notify(struct kernfs_node *kn) |
414985ae | 773 | { |
c525aadd | 774 | struct kernfs_open_node *on; |
414985ae TH |
775 | unsigned long flags; |
776 | ||
c525aadd | 777 | spin_lock_irqsave(&kernfs_open_node_lock, flags); |
414985ae | 778 | |
df23fc39 | 779 | if (!WARN_ON(kernfs_type(kn) != KERNFS_FILE)) { |
c525aadd TH |
780 | on = kn->attr.open; |
781 | if (on) { | |
782 | atomic_inc(&on->event); | |
783 | wake_up_interruptible(&on->poll); | |
414985ae TH |
784 | } |
785 | } | |
786 | ||
c525aadd | 787 | spin_unlock_irqrestore(&kernfs_open_node_lock, flags); |
414985ae TH |
788 | } |
789 | EXPORT_SYMBOL_GPL(kernfs_notify); | |
790 | ||
a797bfc3 | 791 | const struct file_operations kernfs_file_fops = { |
c637b8ac TH |
792 | .read = kernfs_fop_read, |
793 | .write = kernfs_fop_write, | |
414985ae | 794 | .llseek = generic_file_llseek, |
c637b8ac TH |
795 | .mmap = kernfs_fop_mmap, |
796 | .open = kernfs_fop_open, | |
797 | .release = kernfs_fop_release, | |
798 | .poll = kernfs_fop_poll, | |
414985ae TH |
799 | }; |
800 | ||
801 | /** | |
2063d608 | 802 | * __kernfs_create_file - kernfs internal function to create a file |
414985ae TH |
803 | * @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 | |
2063d608 | 810 | * @static_name: don't copy file name |
414985ae TH |
811 | * @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 | */ | |
2063d608 TH |
815 | struct 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) | |
414985ae | 822 | { |
7653fe9d | 823 | struct kernfs_addrm_cxt acxt; |
324a56e1 | 824 | struct kernfs_node *kn; |
2063d608 | 825 | unsigned flags; |
414985ae TH |
826 | int rc; |
827 | ||
2063d608 TH |
828 | flags = KERNFS_FILE; |
829 | if (name_is_static) | |
830 | flags |= KERNFS_STATIC_NAME; | |
831 | ||
c637b8ac | 832 | kn = kernfs_new_node(kernfs_root(parent), name, |
2063d608 | 833 | (mode & S_IALLUGO) | S_IFREG, flags); |
324a56e1 | 834 | if (!kn) |
414985ae TH |
835 | return ERR_PTR(-ENOMEM); |
836 | ||
adc5e8b5 TH |
837 | kn->attr.ops = ops; |
838 | kn->attr.size = size; | |
839 | kn->ns = ns; | |
324a56e1 | 840 | kn->priv = priv; |
414985ae TH |
841 | |
842 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
843 | if (key) { | |
324a56e1 | 844 | lockdep_init_map(&kn->dep_map, "s_active", key, 0); |
df23fc39 | 845 | kn->flags |= KERNFS_LOCKDEP; |
414985ae TH |
846 | } |
847 | #endif | |
848 | ||
849 | /* | |
adc5e8b5 | 850 | * kn->attr.ops is accesible only while holding active ref. We |
414985ae TH |
851 | * need to know whether some ops are implemented outside active |
852 | * ref. Cache their existence in flags. | |
853 | */ | |
854 | if (ops->seq_show) | |
df23fc39 | 855 | kn->flags |= KERNFS_HAS_SEQ_SHOW; |
414985ae | 856 | if (ops->mmap) |
df23fc39 | 857 | kn->flags |= KERNFS_HAS_MMAP; |
414985ae | 858 | |
798c75a0 GKH |
859 | kernfs_addrm_start(&acxt); |
860 | rc = kernfs_add_one(&acxt, kn, parent); | |
861 | kernfs_addrm_finish(&acxt); | |
7653fe9d | 862 | |
414985ae | 863 | if (rc) { |
324a56e1 | 864 | kernfs_put(kn); |
414985ae TH |
865 | return ERR_PTR(rc); |
866 | } | |
324a56e1 | 867 | return kn; |
414985ae | 868 | } |