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 | 255 | const struct kernfs_ops *ops; |
b7ce40cf TH |
256 | size_t len; |
257 | char *buf; | |
4d3773c4 | 258 | |
b7ce40cf | 259 | if (of->atomic_write_len) { |
4d3773c4 | 260 | len = count; |
b7ce40cf TH |
261 | if (len > of->atomic_write_len) |
262 | return -E2BIG; | |
4d3773c4 TH |
263 | } else { |
264 | len = min_t(size_t, count, PAGE_SIZE); | |
265 | } | |
266 | ||
267 | buf = kmalloc(len + 1, GFP_KERNEL); | |
b7ce40cf TH |
268 | if (!buf) |
269 | return -ENOMEM; | |
414985ae | 270 | |
4d3773c4 TH |
271 | if (copy_from_user(buf, user_buf, len)) { |
272 | len = -EFAULT; | |
b7ce40cf | 273 | goto out_free; |
4d3773c4 TH |
274 | } |
275 | buf[len] = '\0'; /* guarantee string termination */ | |
276 | ||
b7ce40cf TH |
277 | /* |
278 | * @of->mutex nests outside active ref and is just to ensure that | |
279 | * the ops aren't called concurrently for the same open file. | |
280 | */ | |
281 | mutex_lock(&of->mutex); | |
282 | if (!kernfs_get_active(of->kn)) { | |
283 | mutex_unlock(&of->mutex); | |
284 | len = -ENODEV; | |
285 | goto out_free; | |
286 | } | |
287 | ||
288 | ops = kernfs_ops(of->kn); | |
289 | if (ops->write) | |
290 | len = ops->write(of, buf, len, *ppos); | |
291 | else | |
292 | len = -EINVAL; | |
293 | ||
c637b8ac | 294 | kernfs_put_active(of->kn); |
414985ae TH |
295 | mutex_unlock(&of->mutex); |
296 | ||
297 | if (len > 0) | |
298 | *ppos += len; | |
b7ce40cf | 299 | out_free: |
414985ae TH |
300 | kfree(buf); |
301 | return len; | |
302 | } | |
303 | ||
304 | static void kernfs_vma_open(struct vm_area_struct *vma) | |
305 | { | |
306 | struct file *file = vma->vm_file; | |
c525aadd | 307 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
308 | |
309 | if (!of->vm_ops) | |
310 | return; | |
311 | ||
c637b8ac | 312 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
313 | return; |
314 | ||
315 | if (of->vm_ops->open) | |
316 | of->vm_ops->open(vma); | |
317 | ||
c637b8ac | 318 | kernfs_put_active(of->kn); |
414985ae TH |
319 | } |
320 | ||
321 | static int kernfs_vma_fault(struct vm_area_struct *vma, struct vm_fault *vmf) | |
322 | { | |
323 | struct file *file = vma->vm_file; | |
c525aadd | 324 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
325 | int ret; |
326 | ||
327 | if (!of->vm_ops) | |
328 | return VM_FAULT_SIGBUS; | |
329 | ||
c637b8ac | 330 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
331 | return VM_FAULT_SIGBUS; |
332 | ||
333 | ret = VM_FAULT_SIGBUS; | |
334 | if (of->vm_ops->fault) | |
335 | ret = of->vm_ops->fault(vma, vmf); | |
336 | ||
c637b8ac | 337 | kernfs_put_active(of->kn); |
414985ae TH |
338 | return ret; |
339 | } | |
340 | ||
341 | static int kernfs_vma_page_mkwrite(struct vm_area_struct *vma, | |
342 | struct vm_fault *vmf) | |
343 | { | |
344 | struct file *file = vma->vm_file; | |
c525aadd | 345 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
346 | int ret; |
347 | ||
348 | if (!of->vm_ops) | |
349 | return VM_FAULT_SIGBUS; | |
350 | ||
c637b8ac | 351 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
352 | return VM_FAULT_SIGBUS; |
353 | ||
354 | ret = 0; | |
355 | if (of->vm_ops->page_mkwrite) | |
356 | ret = of->vm_ops->page_mkwrite(vma, vmf); | |
357 | else | |
358 | file_update_time(file); | |
359 | ||
c637b8ac | 360 | kernfs_put_active(of->kn); |
414985ae TH |
361 | return ret; |
362 | } | |
363 | ||
364 | static int kernfs_vma_access(struct vm_area_struct *vma, unsigned long addr, | |
365 | void *buf, int len, int write) | |
366 | { | |
367 | struct file *file = vma->vm_file; | |
c525aadd | 368 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
369 | int ret; |
370 | ||
371 | if (!of->vm_ops) | |
372 | return -EINVAL; | |
373 | ||
c637b8ac | 374 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
375 | return -EINVAL; |
376 | ||
377 | ret = -EINVAL; | |
378 | if (of->vm_ops->access) | |
379 | ret = of->vm_ops->access(vma, addr, buf, len, write); | |
380 | ||
c637b8ac | 381 | kernfs_put_active(of->kn); |
414985ae TH |
382 | return ret; |
383 | } | |
384 | ||
385 | #ifdef CONFIG_NUMA | |
386 | static int kernfs_vma_set_policy(struct vm_area_struct *vma, | |
387 | struct mempolicy *new) | |
388 | { | |
389 | struct file *file = vma->vm_file; | |
c525aadd | 390 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
391 | int ret; |
392 | ||
393 | if (!of->vm_ops) | |
394 | return 0; | |
395 | ||
c637b8ac | 396 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
397 | return -EINVAL; |
398 | ||
399 | ret = 0; | |
400 | if (of->vm_ops->set_policy) | |
401 | ret = of->vm_ops->set_policy(vma, new); | |
402 | ||
c637b8ac | 403 | kernfs_put_active(of->kn); |
414985ae TH |
404 | return ret; |
405 | } | |
406 | ||
407 | static struct mempolicy *kernfs_vma_get_policy(struct vm_area_struct *vma, | |
408 | unsigned long addr) | |
409 | { | |
410 | struct file *file = vma->vm_file; | |
c525aadd | 411 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
412 | struct mempolicy *pol; |
413 | ||
414 | if (!of->vm_ops) | |
415 | return vma->vm_policy; | |
416 | ||
c637b8ac | 417 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
418 | return vma->vm_policy; |
419 | ||
420 | pol = vma->vm_policy; | |
421 | if (of->vm_ops->get_policy) | |
422 | pol = of->vm_ops->get_policy(vma, addr); | |
423 | ||
c637b8ac | 424 | kernfs_put_active(of->kn); |
414985ae TH |
425 | return pol; |
426 | } | |
427 | ||
428 | static int kernfs_vma_migrate(struct vm_area_struct *vma, | |
429 | const nodemask_t *from, const nodemask_t *to, | |
430 | unsigned long flags) | |
431 | { | |
432 | struct file *file = vma->vm_file; | |
c525aadd | 433 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
434 | int ret; |
435 | ||
436 | if (!of->vm_ops) | |
437 | return 0; | |
438 | ||
c637b8ac | 439 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
440 | return 0; |
441 | ||
442 | ret = 0; | |
443 | if (of->vm_ops->migrate) | |
444 | ret = of->vm_ops->migrate(vma, from, to, flags); | |
445 | ||
c637b8ac | 446 | kernfs_put_active(of->kn); |
414985ae TH |
447 | return ret; |
448 | } | |
449 | #endif | |
450 | ||
451 | static const struct vm_operations_struct kernfs_vm_ops = { | |
452 | .open = kernfs_vma_open, | |
453 | .fault = kernfs_vma_fault, | |
454 | .page_mkwrite = kernfs_vma_page_mkwrite, | |
455 | .access = kernfs_vma_access, | |
456 | #ifdef CONFIG_NUMA | |
457 | .set_policy = kernfs_vma_set_policy, | |
458 | .get_policy = kernfs_vma_get_policy, | |
459 | .migrate = kernfs_vma_migrate, | |
460 | #endif | |
461 | }; | |
462 | ||
c637b8ac | 463 | static int kernfs_fop_mmap(struct file *file, struct vm_area_struct *vma) |
414985ae | 464 | { |
c525aadd | 465 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
466 | const struct kernfs_ops *ops; |
467 | int rc; | |
468 | ||
9b2db6e1 TH |
469 | /* |
470 | * mmap path and of->mutex are prone to triggering spurious lockdep | |
471 | * warnings and we don't want to add spurious locking dependency | |
472 | * between the two. Check whether mmap is actually implemented | |
473 | * without grabbing @of->mutex by testing HAS_MMAP flag. See the | |
474 | * comment in kernfs_file_open() for more details. | |
475 | */ | |
df23fc39 | 476 | if (!(of->kn->flags & KERNFS_HAS_MMAP)) |
9b2db6e1 TH |
477 | return -ENODEV; |
478 | ||
414985ae TH |
479 | mutex_lock(&of->mutex); |
480 | ||
481 | rc = -ENODEV; | |
c637b8ac | 482 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
483 | goto out_unlock; |
484 | ||
324a56e1 | 485 | ops = kernfs_ops(of->kn); |
9b2db6e1 | 486 | rc = ops->mmap(of, vma); |
414985ae TH |
487 | |
488 | /* | |
489 | * PowerPC's pci_mmap of legacy_mem uses shmem_zero_setup() | |
490 | * to satisfy versions of X which crash if the mmap fails: that | |
491 | * substitutes a new vm_file, and we don't then want bin_vm_ops. | |
492 | */ | |
493 | if (vma->vm_file != file) | |
494 | goto out_put; | |
495 | ||
496 | rc = -EINVAL; | |
497 | if (of->mmapped && of->vm_ops != vma->vm_ops) | |
498 | goto out_put; | |
499 | ||
500 | /* | |
501 | * It is not possible to successfully wrap close. | |
502 | * So error if someone is trying to use close. | |
503 | */ | |
504 | rc = -EINVAL; | |
505 | if (vma->vm_ops && vma->vm_ops->close) | |
506 | goto out_put; | |
507 | ||
508 | rc = 0; | |
509 | of->mmapped = 1; | |
510 | of->vm_ops = vma->vm_ops; | |
511 | vma->vm_ops = &kernfs_vm_ops; | |
512 | out_put: | |
c637b8ac | 513 | kernfs_put_active(of->kn); |
414985ae TH |
514 | out_unlock: |
515 | mutex_unlock(&of->mutex); | |
516 | ||
517 | return rc; | |
518 | } | |
519 | ||
520 | /** | |
c637b8ac | 521 | * kernfs_get_open_node - get or create kernfs_open_node |
324a56e1 | 522 | * @kn: target kernfs_node |
c525aadd | 523 | * @of: kernfs_open_file for this instance of open |
414985ae | 524 | * |
adc5e8b5 TH |
525 | * If @kn->attr.open exists, increment its reference count; otherwise, |
526 | * create one. @of is chained to the files list. | |
414985ae TH |
527 | * |
528 | * LOCKING: | |
529 | * Kernel thread context (may sleep). | |
530 | * | |
531 | * RETURNS: | |
532 | * 0 on success, -errno on failure. | |
533 | */ | |
c637b8ac TH |
534 | static int kernfs_get_open_node(struct kernfs_node *kn, |
535 | struct kernfs_open_file *of) | |
414985ae | 536 | { |
c525aadd | 537 | struct kernfs_open_node *on, *new_on = NULL; |
414985ae TH |
538 | |
539 | retry: | |
c525aadd TH |
540 | mutex_lock(&kernfs_open_file_mutex); |
541 | spin_lock_irq(&kernfs_open_node_lock); | |
414985ae | 542 | |
c525aadd TH |
543 | if (!kn->attr.open && new_on) { |
544 | kn->attr.open = new_on; | |
545 | new_on = NULL; | |
414985ae TH |
546 | } |
547 | ||
c525aadd TH |
548 | on = kn->attr.open; |
549 | if (on) { | |
550 | atomic_inc(&on->refcnt); | |
551 | list_add_tail(&of->list, &on->files); | |
414985ae TH |
552 | } |
553 | ||
c525aadd TH |
554 | spin_unlock_irq(&kernfs_open_node_lock); |
555 | mutex_unlock(&kernfs_open_file_mutex); | |
414985ae | 556 | |
c525aadd TH |
557 | if (on) { |
558 | kfree(new_on); | |
414985ae TH |
559 | return 0; |
560 | } | |
561 | ||
562 | /* not there, initialize a new one and retry */ | |
c525aadd TH |
563 | new_on = kmalloc(sizeof(*new_on), GFP_KERNEL); |
564 | if (!new_on) | |
414985ae TH |
565 | return -ENOMEM; |
566 | ||
c525aadd TH |
567 | atomic_set(&new_on->refcnt, 0); |
568 | atomic_set(&new_on->event, 1); | |
569 | init_waitqueue_head(&new_on->poll); | |
570 | INIT_LIST_HEAD(&new_on->files); | |
414985ae TH |
571 | goto retry; |
572 | } | |
573 | ||
574 | /** | |
c637b8ac | 575 | * kernfs_put_open_node - put kernfs_open_node |
324a56e1 | 576 | * @kn: target kernfs_nodet |
c525aadd | 577 | * @of: associated kernfs_open_file |
414985ae | 578 | * |
adc5e8b5 | 579 | * Put @kn->attr.open and unlink @of from the files list. If |
414985ae TH |
580 | * reference count reaches zero, disassociate and free it. |
581 | * | |
582 | * LOCKING: | |
583 | * None. | |
584 | */ | |
c637b8ac TH |
585 | static void kernfs_put_open_node(struct kernfs_node *kn, |
586 | struct kernfs_open_file *of) | |
414985ae | 587 | { |
c525aadd | 588 | struct kernfs_open_node *on = kn->attr.open; |
414985ae TH |
589 | unsigned long flags; |
590 | ||
c525aadd TH |
591 | mutex_lock(&kernfs_open_file_mutex); |
592 | spin_lock_irqsave(&kernfs_open_node_lock, flags); | |
414985ae TH |
593 | |
594 | if (of) | |
595 | list_del(&of->list); | |
596 | ||
c525aadd | 597 | if (atomic_dec_and_test(&on->refcnt)) |
adc5e8b5 | 598 | kn->attr.open = NULL; |
414985ae | 599 | else |
c525aadd | 600 | on = NULL; |
414985ae | 601 | |
c525aadd TH |
602 | spin_unlock_irqrestore(&kernfs_open_node_lock, flags); |
603 | mutex_unlock(&kernfs_open_file_mutex); | |
414985ae | 604 | |
c525aadd | 605 | kfree(on); |
414985ae TH |
606 | } |
607 | ||
c637b8ac | 608 | static int kernfs_fop_open(struct inode *inode, struct file *file) |
414985ae | 609 | { |
324a56e1 | 610 | struct kernfs_node *kn = file->f_path.dentry->d_fsdata; |
414985ae | 611 | const struct kernfs_ops *ops; |
c525aadd | 612 | struct kernfs_open_file *of; |
414985ae TH |
613 | bool has_read, has_write, has_mmap; |
614 | int error = -EACCES; | |
615 | ||
c637b8ac | 616 | if (!kernfs_get_active(kn)) |
414985ae TH |
617 | return -ENODEV; |
618 | ||
324a56e1 | 619 | ops = kernfs_ops(kn); |
414985ae TH |
620 | |
621 | has_read = ops->seq_show || ops->read || ops->mmap; | |
622 | has_write = ops->write || ops->mmap; | |
623 | has_mmap = ops->mmap; | |
624 | ||
625 | /* check perms and supported operations */ | |
626 | if ((file->f_mode & FMODE_WRITE) && | |
627 | (!(inode->i_mode & S_IWUGO) || !has_write)) | |
628 | goto err_out; | |
629 | ||
630 | if ((file->f_mode & FMODE_READ) && | |
631 | (!(inode->i_mode & S_IRUGO) || !has_read)) | |
632 | goto err_out; | |
633 | ||
c525aadd | 634 | /* allocate a kernfs_open_file for the file */ |
414985ae | 635 | error = -ENOMEM; |
c525aadd | 636 | of = kzalloc(sizeof(struct kernfs_open_file), GFP_KERNEL); |
414985ae TH |
637 | if (!of) |
638 | goto err_out; | |
639 | ||
640 | /* | |
641 | * The following is done to give a different lockdep key to | |
642 | * @of->mutex for files which implement mmap. This is a rather | |
643 | * crude way to avoid false positive lockdep warning around | |
644 | * mm->mmap_sem - mmap nests @of->mutex under mm->mmap_sem and | |
645 | * reading /sys/block/sda/trace/act_mask grabs sr_mutex, under | |
646 | * which mm->mmap_sem nests, while holding @of->mutex. As each | |
647 | * open file has a separate mutex, it's okay as long as those don't | |
648 | * happen on the same file. At this point, we can't easily give | |
649 | * each file a separate locking class. Let's differentiate on | |
650 | * whether the file has mmap or not for now. | |
9b2db6e1 TH |
651 | * |
652 | * Both paths of the branch look the same. They're supposed to | |
653 | * look that way and give @of->mutex different static lockdep keys. | |
414985ae TH |
654 | */ |
655 | if (has_mmap) | |
656 | mutex_init(&of->mutex); | |
657 | else | |
658 | mutex_init(&of->mutex); | |
659 | ||
324a56e1 | 660 | of->kn = kn; |
414985ae TH |
661 | of->file = file; |
662 | ||
b7ce40cf TH |
663 | /* |
664 | * Write path needs to atomic_write_len outside active reference. | |
665 | * Cache it in open_file. See kernfs_fop_write() for details. | |
666 | */ | |
667 | of->atomic_write_len = ops->atomic_write_len; | |
668 | ||
414985ae TH |
669 | /* |
670 | * Always instantiate seq_file even if read access doesn't use | |
671 | * seq_file or is not requested. This unifies private data access | |
672 | * and readable regular files are the vast majority anyway. | |
673 | */ | |
674 | if (ops->seq_show) | |
675 | error = seq_open(file, &kernfs_seq_ops); | |
676 | else | |
677 | error = seq_open(file, NULL); | |
678 | if (error) | |
679 | goto err_free; | |
680 | ||
681 | ((struct seq_file *)file->private_data)->private = of; | |
682 | ||
683 | /* seq_file clears PWRITE unconditionally, restore it if WRITE */ | |
684 | if (file->f_mode & FMODE_WRITE) | |
685 | file->f_mode |= FMODE_PWRITE; | |
686 | ||
c637b8ac TH |
687 | /* make sure we have open node struct */ |
688 | error = kernfs_get_open_node(kn, of); | |
414985ae TH |
689 | if (error) |
690 | goto err_close; | |
691 | ||
692 | /* open succeeded, put active references */ | |
c637b8ac | 693 | kernfs_put_active(kn); |
414985ae TH |
694 | return 0; |
695 | ||
696 | err_close: | |
697 | seq_release(inode, file); | |
698 | err_free: | |
699 | kfree(of); | |
700 | err_out: | |
c637b8ac | 701 | kernfs_put_active(kn); |
414985ae TH |
702 | return error; |
703 | } | |
704 | ||
c637b8ac | 705 | static int kernfs_fop_release(struct inode *inode, struct file *filp) |
414985ae | 706 | { |
324a56e1 | 707 | struct kernfs_node *kn = filp->f_path.dentry->d_fsdata; |
c525aadd | 708 | struct kernfs_open_file *of = kernfs_of(filp); |
414985ae | 709 | |
c637b8ac | 710 | kernfs_put_open_node(kn, of); |
414985ae TH |
711 | seq_release(inode, filp); |
712 | kfree(of); | |
713 | ||
714 | return 0; | |
715 | } | |
716 | ||
55f6e30d | 717 | void kernfs_unmap_bin_file(struct kernfs_node *kn) |
414985ae | 718 | { |
c525aadd TH |
719 | struct kernfs_open_node *on; |
720 | struct kernfs_open_file *of; | |
414985ae | 721 | |
55f6e30d GKH |
722 | if (!(kn->flags & KERNFS_HAS_MMAP)) |
723 | return; | |
724 | ||
c525aadd TH |
725 | spin_lock_irq(&kernfs_open_node_lock); |
726 | on = kn->attr.open; | |
727 | if (on) | |
728 | atomic_inc(&on->refcnt); | |
729 | spin_unlock_irq(&kernfs_open_node_lock); | |
730 | if (!on) | |
414985ae TH |
731 | return; |
732 | ||
c525aadd TH |
733 | mutex_lock(&kernfs_open_file_mutex); |
734 | list_for_each_entry(of, &on->files, list) { | |
414985ae TH |
735 | struct inode *inode = file_inode(of->file); |
736 | unmap_mapping_range(inode->i_mapping, 0, 0, 1); | |
737 | } | |
c525aadd | 738 | mutex_unlock(&kernfs_open_file_mutex); |
414985ae | 739 | |
c637b8ac | 740 | kernfs_put_open_node(kn, NULL); |
414985ae TH |
741 | } |
742 | ||
c637b8ac TH |
743 | /* |
744 | * Kernfs attribute files are pollable. The idea is that you read | |
414985ae TH |
745 | * the content and then you use 'poll' or 'select' to wait for |
746 | * the content to change. When the content changes (assuming the | |
747 | * manager for the kobject supports notification), poll will | |
748 | * return POLLERR|POLLPRI, and select will return the fd whether | |
749 | * it is waiting for read, write, or exceptions. | |
750 | * Once poll/select indicates that the value has changed, you | |
751 | * need to close and re-open the file, or seek to 0 and read again. | |
752 | * Reminder: this only works for attributes which actively support | |
753 | * it, and it is not possible to test an attribute from userspace | |
754 | * to see if it supports poll (Neither 'poll' nor 'select' return | |
755 | * an appropriate error code). When in doubt, set a suitable timeout value. | |
756 | */ | |
c637b8ac | 757 | static unsigned int kernfs_fop_poll(struct file *filp, poll_table *wait) |
414985ae | 758 | { |
c525aadd | 759 | struct kernfs_open_file *of = kernfs_of(filp); |
324a56e1 | 760 | struct kernfs_node *kn = filp->f_path.dentry->d_fsdata; |
c525aadd | 761 | struct kernfs_open_node *on = kn->attr.open; |
414985ae TH |
762 | |
763 | /* need parent for the kobj, grab both */ | |
c637b8ac | 764 | if (!kernfs_get_active(kn)) |
414985ae TH |
765 | goto trigger; |
766 | ||
c525aadd | 767 | poll_wait(filp, &on->poll, wait); |
414985ae | 768 | |
c637b8ac | 769 | kernfs_put_active(kn); |
414985ae | 770 | |
c525aadd | 771 | if (of->event != atomic_read(&on->event)) |
414985ae TH |
772 | goto trigger; |
773 | ||
774 | return DEFAULT_POLLMASK; | |
775 | ||
776 | trigger: | |
777 | return DEFAULT_POLLMASK|POLLERR|POLLPRI; | |
778 | } | |
779 | ||
780 | /** | |
781 | * kernfs_notify - notify a kernfs file | |
324a56e1 | 782 | * @kn: file to notify |
414985ae | 783 | * |
324a56e1 | 784 | * Notify @kn such that poll(2) on @kn wakes up. |
414985ae | 785 | */ |
324a56e1 | 786 | void kernfs_notify(struct kernfs_node *kn) |
414985ae | 787 | { |
c525aadd | 788 | struct kernfs_open_node *on; |
414985ae TH |
789 | unsigned long flags; |
790 | ||
c525aadd | 791 | spin_lock_irqsave(&kernfs_open_node_lock, flags); |
414985ae | 792 | |
df23fc39 | 793 | if (!WARN_ON(kernfs_type(kn) != KERNFS_FILE)) { |
c525aadd TH |
794 | on = kn->attr.open; |
795 | if (on) { | |
796 | atomic_inc(&on->event); | |
797 | wake_up_interruptible(&on->poll); | |
414985ae TH |
798 | } |
799 | } | |
800 | ||
c525aadd | 801 | spin_unlock_irqrestore(&kernfs_open_node_lock, flags); |
414985ae TH |
802 | } |
803 | EXPORT_SYMBOL_GPL(kernfs_notify); | |
804 | ||
a797bfc3 | 805 | const struct file_operations kernfs_file_fops = { |
c637b8ac TH |
806 | .read = kernfs_fop_read, |
807 | .write = kernfs_fop_write, | |
414985ae | 808 | .llseek = generic_file_llseek, |
c637b8ac TH |
809 | .mmap = kernfs_fop_mmap, |
810 | .open = kernfs_fop_open, | |
811 | .release = kernfs_fop_release, | |
812 | .poll = kernfs_fop_poll, | |
414985ae TH |
813 | }; |
814 | ||
815 | /** | |
2063d608 | 816 | * __kernfs_create_file - kernfs internal function to create a file |
414985ae TH |
817 | * @parent: directory to create the file in |
818 | * @name: name of the file | |
819 | * @mode: mode of the file | |
820 | * @size: size of the file | |
821 | * @ops: kernfs operations for the file | |
822 | * @priv: private data for the file | |
823 | * @ns: optional namespace tag of the file | |
2063d608 | 824 | * @static_name: don't copy file name |
414985ae TH |
825 | * @key: lockdep key for the file's active_ref, %NULL to disable lockdep |
826 | * | |
827 | * Returns the created node on success, ERR_PTR() value on error. | |
828 | */ | |
2063d608 TH |
829 | struct kernfs_node *__kernfs_create_file(struct kernfs_node *parent, |
830 | const char *name, | |
831 | umode_t mode, loff_t size, | |
832 | const struct kernfs_ops *ops, | |
833 | void *priv, const void *ns, | |
834 | bool name_is_static, | |
835 | struct lock_class_key *key) | |
414985ae | 836 | { |
324a56e1 | 837 | struct kernfs_node *kn; |
2063d608 | 838 | unsigned flags; |
414985ae TH |
839 | int rc; |
840 | ||
2063d608 TH |
841 | flags = KERNFS_FILE; |
842 | if (name_is_static) | |
843 | flags |= KERNFS_STATIC_NAME; | |
844 | ||
db4aad20 | 845 | kn = kernfs_new_node(parent, name, (mode & S_IALLUGO) | S_IFREG, flags); |
324a56e1 | 846 | if (!kn) |
414985ae TH |
847 | return ERR_PTR(-ENOMEM); |
848 | ||
adc5e8b5 TH |
849 | kn->attr.ops = ops; |
850 | kn->attr.size = size; | |
851 | kn->ns = ns; | |
324a56e1 | 852 | kn->priv = priv; |
414985ae TH |
853 | |
854 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
855 | if (key) { | |
324a56e1 | 856 | lockdep_init_map(&kn->dep_map, "s_active", key, 0); |
df23fc39 | 857 | kn->flags |= KERNFS_LOCKDEP; |
414985ae TH |
858 | } |
859 | #endif | |
860 | ||
861 | /* | |
adc5e8b5 | 862 | * kn->attr.ops is accesible only while holding active ref. We |
414985ae TH |
863 | * need to know whether some ops are implemented outside active |
864 | * ref. Cache their existence in flags. | |
865 | */ | |
866 | if (ops->seq_show) | |
df23fc39 | 867 | kn->flags |= KERNFS_HAS_SEQ_SHOW; |
414985ae | 868 | if (ops->mmap) |
df23fc39 | 869 | kn->flags |= KERNFS_HAS_MMAP; |
414985ae | 870 | |
988cd7af | 871 | rc = kernfs_add_one(kn); |
414985ae | 872 | if (rc) { |
324a56e1 | 873 | kernfs_put(kn); |
414985ae TH |
874 | return ERR_PTR(rc); |
875 | } | |
324a56e1 | 876 | return kn; |
414985ae | 877 | } |