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