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 | 16 | #include <linux/sched.h> |
d911d987 | 17 | #include <linux/fsnotify.h> |
414985ae TH |
18 | |
19 | #include "kernfs-internal.h" | |
20 | ||
21 | /* | |
c525aadd | 22 | * There's one kernfs_open_file for each open file and one kernfs_open_node |
324a56e1 | 23 | * for each kernfs_node with one or more open files. |
414985ae | 24 | * |
c525aadd TH |
25 | * kernfs_node->attr.open points to kernfs_open_node. attr.open is |
26 | * protected by kernfs_open_node_lock. | |
414985ae TH |
27 | * |
28 | * filp->private_data points to seq_file whose ->private points to | |
c525aadd TH |
29 | * kernfs_open_file. kernfs_open_files are chained at |
30 | * kernfs_open_node->files, which is protected by kernfs_open_file_mutex. | |
414985ae | 31 | */ |
c525aadd TH |
32 | static DEFINE_SPINLOCK(kernfs_open_node_lock); |
33 | static DEFINE_MUTEX(kernfs_open_file_mutex); | |
414985ae | 34 | |
c525aadd | 35 | struct kernfs_open_node { |
414985ae TH |
36 | atomic_t refcnt; |
37 | atomic_t event; | |
38 | wait_queue_head_t poll; | |
c525aadd | 39 | struct list_head files; /* goes through kernfs_open_file.list */ |
414985ae TH |
40 | }; |
41 | ||
c525aadd | 42 | static struct kernfs_open_file *kernfs_of(struct file *file) |
414985ae TH |
43 | { |
44 | return ((struct seq_file *)file->private_data)->private; | |
45 | } | |
46 | ||
47 | /* | |
324a56e1 | 48 | * Determine the kernfs_ops for the given kernfs_node. This function must |
414985ae TH |
49 | * be called while holding an active reference. |
50 | */ | |
324a56e1 | 51 | static const struct kernfs_ops *kernfs_ops(struct kernfs_node *kn) |
414985ae | 52 | { |
df23fc39 | 53 | if (kn->flags & KERNFS_LOCKDEP) |
324a56e1 | 54 | lockdep_assert_held(kn); |
adc5e8b5 | 55 | return kn->attr.ops; |
414985ae TH |
56 | } |
57 | ||
bb305947 TH |
58 | /* |
59 | * As kernfs_seq_stop() is also called after kernfs_seq_start() or | |
60 | * kernfs_seq_next() failure, it needs to distinguish whether it's stopping | |
61 | * a seq_file iteration which is fully initialized with an active reference | |
62 | * or an aborted kernfs_seq_start() due to get_active failure. The | |
63 | * position pointer is the only context for each seq_file iteration and | |
64 | * thus the stop condition should be encoded in it. As the return value is | |
65 | * directly visible to userland, ERR_PTR(-ENODEV) is the only acceptable | |
66 | * choice to indicate get_active failure. | |
67 | * | |
68 | * Unfortunately, this is complicated due to the optional custom seq_file | |
69 | * operations which may return ERR_PTR(-ENODEV) too. kernfs_seq_stop() | |
70 | * can't distinguish whether ERR_PTR(-ENODEV) is from get_active failure or | |
71 | * custom seq_file operations and thus can't decide whether put_active | |
72 | * should be performed or not only on ERR_PTR(-ENODEV). | |
73 | * | |
74 | * This is worked around by factoring out the custom seq_stop() and | |
75 | * put_active part into kernfs_seq_stop_active(), skipping it from | |
76 | * kernfs_seq_stop() if ERR_PTR(-ENODEV) while invoking it directly after | |
77 | * custom seq_file operations fail with ERR_PTR(-ENODEV) - this ensures | |
78 | * that kernfs_seq_stop_active() is skipped only after get_active failure. | |
79 | */ | |
80 | static void kernfs_seq_stop_active(struct seq_file *sf, void *v) | |
81 | { | |
82 | struct kernfs_open_file *of = sf->private; | |
83 | const struct kernfs_ops *ops = kernfs_ops(of->kn); | |
84 | ||
85 | if (ops->seq_stop) | |
86 | ops->seq_stop(sf, v); | |
87 | kernfs_put_active(of->kn); | |
88 | } | |
89 | ||
414985ae TH |
90 | static void *kernfs_seq_start(struct seq_file *sf, loff_t *ppos) |
91 | { | |
c525aadd | 92 | struct kernfs_open_file *of = sf->private; |
414985ae TH |
93 | const struct kernfs_ops *ops; |
94 | ||
95 | /* | |
96 | * @of->mutex nests outside active ref and is just to ensure that | |
97 | * the ops aren't called concurrently for the same open file. | |
98 | */ | |
99 | mutex_lock(&of->mutex); | |
c637b8ac | 100 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
101 | return ERR_PTR(-ENODEV); |
102 | ||
324a56e1 | 103 | ops = kernfs_ops(of->kn); |
414985ae | 104 | if (ops->seq_start) { |
bb305947 TH |
105 | void *next = ops->seq_start(sf, ppos); |
106 | /* see the comment above kernfs_seq_stop_active() */ | |
107 | if (next == ERR_PTR(-ENODEV)) | |
108 | kernfs_seq_stop_active(sf, next); | |
109 | return next; | |
414985ae TH |
110 | } else { |
111 | /* | |
112 | * The same behavior and code as single_open(). Returns | |
113 | * !NULL if pos is at the beginning; otherwise, NULL. | |
114 | */ | |
115 | return NULL + !*ppos; | |
116 | } | |
117 | } | |
118 | ||
119 | static void *kernfs_seq_next(struct seq_file *sf, void *v, loff_t *ppos) | |
120 | { | |
c525aadd | 121 | struct kernfs_open_file *of = sf->private; |
324a56e1 | 122 | const struct kernfs_ops *ops = kernfs_ops(of->kn); |
414985ae TH |
123 | |
124 | if (ops->seq_next) { | |
bb305947 TH |
125 | void *next = ops->seq_next(sf, v, ppos); |
126 | /* see the comment above kernfs_seq_stop_active() */ | |
127 | if (next == ERR_PTR(-ENODEV)) | |
128 | kernfs_seq_stop_active(sf, next); | |
129 | return next; | |
414985ae TH |
130 | } else { |
131 | /* | |
132 | * The same behavior and code as single_open(), always | |
133 | * terminate after the initial read. | |
134 | */ | |
135 | ++*ppos; | |
136 | return NULL; | |
137 | } | |
138 | } | |
139 | ||
140 | static void kernfs_seq_stop(struct seq_file *sf, void *v) | |
141 | { | |
c525aadd | 142 | struct kernfs_open_file *of = sf->private; |
414985ae | 143 | |
bb305947 TH |
144 | if (v != ERR_PTR(-ENODEV)) |
145 | kernfs_seq_stop_active(sf, v); | |
414985ae TH |
146 | mutex_unlock(&of->mutex); |
147 | } | |
148 | ||
149 | static int kernfs_seq_show(struct seq_file *sf, void *v) | |
150 | { | |
c525aadd | 151 | struct kernfs_open_file *of = sf->private; |
414985ae | 152 | |
adc5e8b5 | 153 | of->event = atomic_read(&of->kn->attr.open->event); |
414985ae | 154 | |
adc5e8b5 | 155 | return of->kn->attr.ops->seq_show(sf, v); |
414985ae TH |
156 | } |
157 | ||
158 | static const struct seq_operations kernfs_seq_ops = { | |
159 | .start = kernfs_seq_start, | |
160 | .next = kernfs_seq_next, | |
161 | .stop = kernfs_seq_stop, | |
162 | .show = kernfs_seq_show, | |
163 | }; | |
164 | ||
165 | /* | |
166 | * As reading a bin file can have side-effects, the exact offset and bytes | |
167 | * specified in read(2) call should be passed to the read callback making | |
168 | * it difficult to use seq_file. Implement simplistic custom buffering for | |
169 | * bin files. | |
170 | */ | |
c525aadd | 171 | static ssize_t kernfs_file_direct_read(struct kernfs_open_file *of, |
414985ae TH |
172 | char __user *user_buf, size_t count, |
173 | loff_t *ppos) | |
174 | { | |
175 | ssize_t len = min_t(size_t, count, PAGE_SIZE); | |
176 | const struct kernfs_ops *ops; | |
177 | char *buf; | |
178 | ||
179 | buf = kmalloc(len, GFP_KERNEL); | |
180 | if (!buf) | |
181 | return -ENOMEM; | |
182 | ||
183 | /* | |
184 | * @of->mutex nests outside active ref and is just to ensure that | |
185 | * the ops aren't called concurrently for the same open file. | |
186 | */ | |
187 | mutex_lock(&of->mutex); | |
c637b8ac | 188 | if (!kernfs_get_active(of->kn)) { |
414985ae TH |
189 | len = -ENODEV; |
190 | mutex_unlock(&of->mutex); | |
191 | goto out_free; | |
192 | } | |
193 | ||
324a56e1 | 194 | ops = kernfs_ops(of->kn); |
414985ae TH |
195 | if (ops->read) |
196 | len = ops->read(of, buf, len, *ppos); | |
197 | else | |
198 | len = -EINVAL; | |
199 | ||
c637b8ac | 200 | kernfs_put_active(of->kn); |
414985ae TH |
201 | mutex_unlock(&of->mutex); |
202 | ||
203 | if (len < 0) | |
204 | goto out_free; | |
205 | ||
206 | if (copy_to_user(user_buf, buf, len)) { | |
207 | len = -EFAULT; | |
208 | goto out_free; | |
209 | } | |
210 | ||
211 | *ppos += len; | |
212 | ||
213 | out_free: | |
214 | kfree(buf); | |
215 | return len; | |
216 | } | |
217 | ||
218 | /** | |
c637b8ac | 219 | * kernfs_fop_read - kernfs vfs read callback |
414985ae TH |
220 | * @file: file pointer |
221 | * @user_buf: data to write | |
222 | * @count: number of bytes | |
223 | * @ppos: starting offset | |
224 | */ | |
c637b8ac TH |
225 | static ssize_t kernfs_fop_read(struct file *file, char __user *user_buf, |
226 | size_t count, loff_t *ppos) | |
414985ae | 227 | { |
c525aadd | 228 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae | 229 | |
df23fc39 | 230 | if (of->kn->flags & KERNFS_HAS_SEQ_SHOW) |
414985ae TH |
231 | return seq_read(file, user_buf, count, ppos); |
232 | else | |
233 | return kernfs_file_direct_read(of, user_buf, count, ppos); | |
234 | } | |
235 | ||
236 | /** | |
c637b8ac | 237 | * kernfs_fop_write - kernfs vfs write callback |
414985ae TH |
238 | * @file: file pointer |
239 | * @user_buf: data to write | |
240 | * @count: number of bytes | |
241 | * @ppos: starting offset | |
242 | * | |
243 | * Copy data in from userland and pass it to the matching kernfs write | |
244 | * operation. | |
245 | * | |
246 | * There is no easy way for us to know if userspace is only doing a partial | |
247 | * write, so we don't support them. We expect the entire buffer to come on | |
248 | * the first write. Hint: if you're writing a value, first read the file, | |
249 | * modify only the the value you're changing, then write entire buffer | |
250 | * back. | |
251 | */ | |
c637b8ac TH |
252 | static ssize_t kernfs_fop_write(struct file *file, const char __user *user_buf, |
253 | size_t count, loff_t *ppos) | |
414985ae | 254 | { |
c525aadd | 255 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae | 256 | const struct kernfs_ops *ops; |
b7ce40cf TH |
257 | size_t len; |
258 | char *buf; | |
4d3773c4 | 259 | |
b7ce40cf | 260 | if (of->atomic_write_len) { |
4d3773c4 | 261 | len = count; |
b7ce40cf TH |
262 | if (len > of->atomic_write_len) |
263 | return -E2BIG; | |
4d3773c4 TH |
264 | } else { |
265 | len = min_t(size_t, count, PAGE_SIZE); | |
266 | } | |
267 | ||
268 | buf = kmalloc(len + 1, GFP_KERNEL); | |
b7ce40cf TH |
269 | if (!buf) |
270 | return -ENOMEM; | |
414985ae | 271 | |
4d3773c4 TH |
272 | if (copy_from_user(buf, user_buf, len)) { |
273 | len = -EFAULT; | |
b7ce40cf | 274 | goto out_free; |
4d3773c4 TH |
275 | } |
276 | buf[len] = '\0'; /* guarantee string termination */ | |
277 | ||
b7ce40cf TH |
278 | /* |
279 | * @of->mutex nests outside active ref and is just to ensure that | |
280 | * the ops aren't called concurrently for the same open file. | |
281 | */ | |
282 | mutex_lock(&of->mutex); | |
283 | if (!kernfs_get_active(of->kn)) { | |
284 | mutex_unlock(&of->mutex); | |
285 | len = -ENODEV; | |
286 | goto out_free; | |
287 | } | |
288 | ||
289 | ops = kernfs_ops(of->kn); | |
290 | if (ops->write) | |
291 | len = ops->write(of, buf, len, *ppos); | |
292 | else | |
293 | len = -EINVAL; | |
294 | ||
c637b8ac | 295 | kernfs_put_active(of->kn); |
414985ae TH |
296 | mutex_unlock(&of->mutex); |
297 | ||
298 | if (len > 0) | |
299 | *ppos += len; | |
b7ce40cf | 300 | out_free: |
414985ae TH |
301 | kfree(buf); |
302 | return len; | |
303 | } | |
304 | ||
305 | static void kernfs_vma_open(struct vm_area_struct *vma) | |
306 | { | |
307 | struct file *file = vma->vm_file; | |
c525aadd | 308 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
309 | |
310 | if (!of->vm_ops) | |
311 | return; | |
312 | ||
c637b8ac | 313 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
314 | return; |
315 | ||
316 | if (of->vm_ops->open) | |
317 | of->vm_ops->open(vma); | |
318 | ||
c637b8ac | 319 | kernfs_put_active(of->kn); |
414985ae TH |
320 | } |
321 | ||
322 | static int kernfs_vma_fault(struct vm_area_struct *vma, struct vm_fault *vmf) | |
323 | { | |
324 | struct file *file = vma->vm_file; | |
c525aadd | 325 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
326 | int ret; |
327 | ||
328 | if (!of->vm_ops) | |
329 | return VM_FAULT_SIGBUS; | |
330 | ||
c637b8ac | 331 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
332 | return VM_FAULT_SIGBUS; |
333 | ||
334 | ret = VM_FAULT_SIGBUS; | |
335 | if (of->vm_ops->fault) | |
336 | ret = of->vm_ops->fault(vma, vmf); | |
337 | ||
c637b8ac | 338 | kernfs_put_active(of->kn); |
414985ae TH |
339 | return ret; |
340 | } | |
341 | ||
342 | static int kernfs_vma_page_mkwrite(struct vm_area_struct *vma, | |
343 | struct vm_fault *vmf) | |
344 | { | |
345 | struct file *file = vma->vm_file; | |
c525aadd | 346 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
347 | int ret; |
348 | ||
349 | if (!of->vm_ops) | |
350 | return VM_FAULT_SIGBUS; | |
351 | ||
c637b8ac | 352 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
353 | return VM_FAULT_SIGBUS; |
354 | ||
355 | ret = 0; | |
356 | if (of->vm_ops->page_mkwrite) | |
357 | ret = of->vm_ops->page_mkwrite(vma, vmf); | |
358 | else | |
359 | file_update_time(file); | |
360 | ||
c637b8ac | 361 | kernfs_put_active(of->kn); |
414985ae TH |
362 | return ret; |
363 | } | |
364 | ||
365 | static int kernfs_vma_access(struct vm_area_struct *vma, unsigned long addr, | |
366 | void *buf, int len, int write) | |
367 | { | |
368 | struct file *file = vma->vm_file; | |
c525aadd | 369 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
370 | int ret; |
371 | ||
372 | if (!of->vm_ops) | |
373 | return -EINVAL; | |
374 | ||
c637b8ac | 375 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
376 | return -EINVAL; |
377 | ||
378 | ret = -EINVAL; | |
379 | if (of->vm_ops->access) | |
380 | ret = of->vm_ops->access(vma, addr, buf, len, write); | |
381 | ||
c637b8ac | 382 | kernfs_put_active(of->kn); |
414985ae TH |
383 | return ret; |
384 | } | |
385 | ||
386 | #ifdef CONFIG_NUMA | |
387 | static int kernfs_vma_set_policy(struct vm_area_struct *vma, | |
388 | struct mempolicy *new) | |
389 | { | |
390 | struct file *file = vma->vm_file; | |
c525aadd | 391 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
392 | int ret; |
393 | ||
394 | if (!of->vm_ops) | |
395 | return 0; | |
396 | ||
c637b8ac | 397 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
398 | return -EINVAL; |
399 | ||
400 | ret = 0; | |
401 | if (of->vm_ops->set_policy) | |
402 | ret = of->vm_ops->set_policy(vma, new); | |
403 | ||
c637b8ac | 404 | kernfs_put_active(of->kn); |
414985ae TH |
405 | return ret; |
406 | } | |
407 | ||
408 | static struct mempolicy *kernfs_vma_get_policy(struct vm_area_struct *vma, | |
409 | unsigned long addr) | |
410 | { | |
411 | struct file *file = vma->vm_file; | |
c525aadd | 412 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
413 | struct mempolicy *pol; |
414 | ||
415 | if (!of->vm_ops) | |
416 | return vma->vm_policy; | |
417 | ||
c637b8ac | 418 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
419 | return vma->vm_policy; |
420 | ||
421 | pol = vma->vm_policy; | |
422 | if (of->vm_ops->get_policy) | |
423 | pol = of->vm_ops->get_policy(vma, addr); | |
424 | ||
c637b8ac | 425 | kernfs_put_active(of->kn); |
414985ae TH |
426 | return pol; |
427 | } | |
428 | ||
429 | static int kernfs_vma_migrate(struct vm_area_struct *vma, | |
430 | const nodemask_t *from, const nodemask_t *to, | |
431 | unsigned long flags) | |
432 | { | |
433 | struct file *file = vma->vm_file; | |
c525aadd | 434 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
435 | int ret; |
436 | ||
437 | if (!of->vm_ops) | |
438 | return 0; | |
439 | ||
c637b8ac | 440 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
441 | return 0; |
442 | ||
443 | ret = 0; | |
444 | if (of->vm_ops->migrate) | |
445 | ret = of->vm_ops->migrate(vma, from, to, flags); | |
446 | ||
c637b8ac | 447 | kernfs_put_active(of->kn); |
414985ae TH |
448 | return ret; |
449 | } | |
450 | #endif | |
451 | ||
452 | static const struct vm_operations_struct kernfs_vm_ops = { | |
453 | .open = kernfs_vma_open, | |
454 | .fault = kernfs_vma_fault, | |
455 | .page_mkwrite = kernfs_vma_page_mkwrite, | |
456 | .access = kernfs_vma_access, | |
457 | #ifdef CONFIG_NUMA | |
458 | .set_policy = kernfs_vma_set_policy, | |
459 | .get_policy = kernfs_vma_get_policy, | |
460 | .migrate = kernfs_vma_migrate, | |
461 | #endif | |
462 | }; | |
463 | ||
c637b8ac | 464 | static int kernfs_fop_mmap(struct file *file, struct vm_area_struct *vma) |
414985ae | 465 | { |
c525aadd | 466 | struct kernfs_open_file *of = kernfs_of(file); |
414985ae TH |
467 | const struct kernfs_ops *ops; |
468 | int rc; | |
469 | ||
9b2db6e1 TH |
470 | /* |
471 | * mmap path and of->mutex are prone to triggering spurious lockdep | |
472 | * warnings and we don't want to add spurious locking dependency | |
473 | * between the two. Check whether mmap is actually implemented | |
474 | * without grabbing @of->mutex by testing HAS_MMAP flag. See the | |
475 | * comment in kernfs_file_open() for more details. | |
476 | */ | |
df23fc39 | 477 | if (!(of->kn->flags & KERNFS_HAS_MMAP)) |
9b2db6e1 TH |
478 | return -ENODEV; |
479 | ||
414985ae TH |
480 | mutex_lock(&of->mutex); |
481 | ||
482 | rc = -ENODEV; | |
c637b8ac | 483 | if (!kernfs_get_active(of->kn)) |
414985ae TH |
484 | goto out_unlock; |
485 | ||
324a56e1 | 486 | ops = kernfs_ops(of->kn); |
9b2db6e1 | 487 | rc = ops->mmap(of, vma); |
b44b2140 TH |
488 | if (rc) |
489 | goto out_put; | |
414985ae TH |
490 | |
491 | /* | |
492 | * PowerPC's pci_mmap of legacy_mem uses shmem_zero_setup() | |
493 | * to satisfy versions of X which crash if the mmap fails: that | |
494 | * substitutes a new vm_file, and we don't then want bin_vm_ops. | |
495 | */ | |
496 | if (vma->vm_file != file) | |
497 | goto out_put; | |
498 | ||
499 | rc = -EINVAL; | |
500 | if (of->mmapped && of->vm_ops != vma->vm_ops) | |
501 | goto out_put; | |
502 | ||
503 | /* | |
504 | * It is not possible to successfully wrap close. | |
505 | * So error if someone is trying to use close. | |
506 | */ | |
507 | rc = -EINVAL; | |
508 | if (vma->vm_ops && vma->vm_ops->close) | |
509 | goto out_put; | |
510 | ||
511 | rc = 0; | |
512 | of->mmapped = 1; | |
513 | of->vm_ops = vma->vm_ops; | |
514 | vma->vm_ops = &kernfs_vm_ops; | |
515 | out_put: | |
c637b8ac | 516 | kernfs_put_active(of->kn); |
414985ae TH |
517 | out_unlock: |
518 | mutex_unlock(&of->mutex); | |
519 | ||
520 | return rc; | |
521 | } | |
522 | ||
523 | /** | |
c637b8ac | 524 | * kernfs_get_open_node - get or create kernfs_open_node |
324a56e1 | 525 | * @kn: target kernfs_node |
c525aadd | 526 | * @of: kernfs_open_file for this instance of open |
414985ae | 527 | * |
adc5e8b5 TH |
528 | * If @kn->attr.open exists, increment its reference count; otherwise, |
529 | * create one. @of is chained to the files list. | |
414985ae TH |
530 | * |
531 | * LOCKING: | |
532 | * Kernel thread context (may sleep). | |
533 | * | |
534 | * RETURNS: | |
535 | * 0 on success, -errno on failure. | |
536 | */ | |
c637b8ac TH |
537 | static int kernfs_get_open_node(struct kernfs_node *kn, |
538 | struct kernfs_open_file *of) | |
414985ae | 539 | { |
c525aadd | 540 | struct kernfs_open_node *on, *new_on = NULL; |
414985ae TH |
541 | |
542 | retry: | |
c525aadd TH |
543 | mutex_lock(&kernfs_open_file_mutex); |
544 | spin_lock_irq(&kernfs_open_node_lock); | |
414985ae | 545 | |
c525aadd TH |
546 | if (!kn->attr.open && new_on) { |
547 | kn->attr.open = new_on; | |
548 | new_on = NULL; | |
414985ae TH |
549 | } |
550 | ||
c525aadd TH |
551 | on = kn->attr.open; |
552 | if (on) { | |
553 | atomic_inc(&on->refcnt); | |
554 | list_add_tail(&of->list, &on->files); | |
414985ae TH |
555 | } |
556 | ||
c525aadd TH |
557 | spin_unlock_irq(&kernfs_open_node_lock); |
558 | mutex_unlock(&kernfs_open_file_mutex); | |
414985ae | 559 | |
c525aadd TH |
560 | if (on) { |
561 | kfree(new_on); | |
414985ae TH |
562 | return 0; |
563 | } | |
564 | ||
565 | /* not there, initialize a new one and retry */ | |
c525aadd TH |
566 | new_on = kmalloc(sizeof(*new_on), GFP_KERNEL); |
567 | if (!new_on) | |
414985ae TH |
568 | return -ENOMEM; |
569 | ||
c525aadd TH |
570 | atomic_set(&new_on->refcnt, 0); |
571 | atomic_set(&new_on->event, 1); | |
572 | init_waitqueue_head(&new_on->poll); | |
573 | INIT_LIST_HEAD(&new_on->files); | |
414985ae TH |
574 | goto retry; |
575 | } | |
576 | ||
577 | /** | |
c637b8ac | 578 | * kernfs_put_open_node - put kernfs_open_node |
324a56e1 | 579 | * @kn: target kernfs_nodet |
c525aadd | 580 | * @of: associated kernfs_open_file |
414985ae | 581 | * |
adc5e8b5 | 582 | * Put @kn->attr.open and unlink @of from the files list. If |
414985ae TH |
583 | * reference count reaches zero, disassociate and free it. |
584 | * | |
585 | * LOCKING: | |
586 | * None. | |
587 | */ | |
c637b8ac TH |
588 | static void kernfs_put_open_node(struct kernfs_node *kn, |
589 | struct kernfs_open_file *of) | |
414985ae | 590 | { |
c525aadd | 591 | struct kernfs_open_node *on = kn->attr.open; |
414985ae TH |
592 | unsigned long flags; |
593 | ||
c525aadd TH |
594 | mutex_lock(&kernfs_open_file_mutex); |
595 | spin_lock_irqsave(&kernfs_open_node_lock, flags); | |
414985ae TH |
596 | |
597 | if (of) | |
598 | list_del(&of->list); | |
599 | ||
c525aadd | 600 | if (atomic_dec_and_test(&on->refcnt)) |
adc5e8b5 | 601 | kn->attr.open = NULL; |
414985ae | 602 | else |
c525aadd | 603 | on = NULL; |
414985ae | 604 | |
c525aadd TH |
605 | spin_unlock_irqrestore(&kernfs_open_node_lock, flags); |
606 | mutex_unlock(&kernfs_open_file_mutex); | |
414985ae | 607 | |
c525aadd | 608 | kfree(on); |
414985ae TH |
609 | } |
610 | ||
c637b8ac | 611 | static int kernfs_fop_open(struct inode *inode, struct file *file) |
414985ae | 612 | { |
324a56e1 | 613 | struct kernfs_node *kn = file->f_path.dentry->d_fsdata; |
555724a8 | 614 | struct kernfs_root *root = kernfs_root(kn); |
414985ae | 615 | const struct kernfs_ops *ops; |
c525aadd | 616 | struct kernfs_open_file *of; |
414985ae TH |
617 | bool has_read, has_write, has_mmap; |
618 | int error = -EACCES; | |
619 | ||
c637b8ac | 620 | if (!kernfs_get_active(kn)) |
414985ae TH |
621 | return -ENODEV; |
622 | ||
324a56e1 | 623 | ops = kernfs_ops(kn); |
414985ae TH |
624 | |
625 | has_read = ops->seq_show || ops->read || ops->mmap; | |
626 | has_write = ops->write || ops->mmap; | |
627 | has_mmap = ops->mmap; | |
628 | ||
555724a8 TH |
629 | /* see the flag definition for details */ |
630 | if (root->flags & KERNFS_ROOT_EXTRA_OPEN_PERM_CHECK) { | |
631 | if ((file->f_mode & FMODE_WRITE) && | |
632 | (!(inode->i_mode & S_IWUGO) || !has_write)) | |
633 | goto err_out; | |
414985ae | 634 | |
555724a8 TH |
635 | if ((file->f_mode & FMODE_READ) && |
636 | (!(inode->i_mode & S_IRUGO) || !has_read)) | |
637 | goto err_out; | |
638 | } | |
414985ae | 639 | |
c525aadd | 640 | /* allocate a kernfs_open_file for the file */ |
414985ae | 641 | error = -ENOMEM; |
c525aadd | 642 | of = kzalloc(sizeof(struct kernfs_open_file), GFP_KERNEL); |
414985ae TH |
643 | if (!of) |
644 | goto err_out; | |
645 | ||
646 | /* | |
647 | * The following is done to give a different lockdep key to | |
648 | * @of->mutex for files which implement mmap. This is a rather | |
649 | * crude way to avoid false positive lockdep warning around | |
650 | * mm->mmap_sem - mmap nests @of->mutex under mm->mmap_sem and | |
651 | * reading /sys/block/sda/trace/act_mask grabs sr_mutex, under | |
652 | * which mm->mmap_sem nests, while holding @of->mutex. As each | |
653 | * open file has a separate mutex, it's okay as long as those don't | |
654 | * happen on the same file. At this point, we can't easily give | |
655 | * each file a separate locking class. Let's differentiate on | |
656 | * whether the file has mmap or not for now. | |
9b2db6e1 TH |
657 | * |
658 | * Both paths of the branch look the same. They're supposed to | |
659 | * look that way and give @of->mutex different static lockdep keys. | |
414985ae TH |
660 | */ |
661 | if (has_mmap) | |
662 | mutex_init(&of->mutex); | |
663 | else | |
664 | mutex_init(&of->mutex); | |
665 | ||
324a56e1 | 666 | of->kn = kn; |
414985ae TH |
667 | of->file = file; |
668 | ||
b7ce40cf TH |
669 | /* |
670 | * Write path needs to atomic_write_len outside active reference. | |
671 | * Cache it in open_file. See kernfs_fop_write() for details. | |
672 | */ | |
673 | of->atomic_write_len = ops->atomic_write_len; | |
674 | ||
414985ae TH |
675 | /* |
676 | * Always instantiate seq_file even if read access doesn't use | |
677 | * seq_file or is not requested. This unifies private data access | |
678 | * and readable regular files are the vast majority anyway. | |
679 | */ | |
680 | if (ops->seq_show) | |
681 | error = seq_open(file, &kernfs_seq_ops); | |
682 | else | |
683 | error = seq_open(file, NULL); | |
684 | if (error) | |
685 | goto err_free; | |
686 | ||
687 | ((struct seq_file *)file->private_data)->private = of; | |
688 | ||
689 | /* seq_file clears PWRITE unconditionally, restore it if WRITE */ | |
690 | if (file->f_mode & FMODE_WRITE) | |
691 | file->f_mode |= FMODE_PWRITE; | |
692 | ||
c637b8ac TH |
693 | /* make sure we have open node struct */ |
694 | error = kernfs_get_open_node(kn, of); | |
414985ae TH |
695 | if (error) |
696 | goto err_close; | |
697 | ||
698 | /* open succeeded, put active references */ | |
c637b8ac | 699 | kernfs_put_active(kn); |
414985ae TH |
700 | return 0; |
701 | ||
702 | err_close: | |
703 | seq_release(inode, file); | |
704 | err_free: | |
705 | kfree(of); | |
706 | err_out: | |
c637b8ac | 707 | kernfs_put_active(kn); |
414985ae TH |
708 | return error; |
709 | } | |
710 | ||
c637b8ac | 711 | static int kernfs_fop_release(struct inode *inode, struct file *filp) |
414985ae | 712 | { |
324a56e1 | 713 | struct kernfs_node *kn = filp->f_path.dentry->d_fsdata; |
c525aadd | 714 | struct kernfs_open_file *of = kernfs_of(filp); |
414985ae | 715 | |
c637b8ac | 716 | kernfs_put_open_node(kn, of); |
414985ae TH |
717 | seq_release(inode, filp); |
718 | kfree(of); | |
719 | ||
720 | return 0; | |
721 | } | |
722 | ||
55f6e30d | 723 | void kernfs_unmap_bin_file(struct kernfs_node *kn) |
414985ae | 724 | { |
c525aadd TH |
725 | struct kernfs_open_node *on; |
726 | struct kernfs_open_file *of; | |
414985ae | 727 | |
55f6e30d GKH |
728 | if (!(kn->flags & KERNFS_HAS_MMAP)) |
729 | return; | |
730 | ||
c525aadd TH |
731 | spin_lock_irq(&kernfs_open_node_lock); |
732 | on = kn->attr.open; | |
733 | if (on) | |
734 | atomic_inc(&on->refcnt); | |
735 | spin_unlock_irq(&kernfs_open_node_lock); | |
736 | if (!on) | |
414985ae TH |
737 | return; |
738 | ||
c525aadd TH |
739 | mutex_lock(&kernfs_open_file_mutex); |
740 | list_for_each_entry(of, &on->files, list) { | |
414985ae TH |
741 | struct inode *inode = file_inode(of->file); |
742 | unmap_mapping_range(inode->i_mapping, 0, 0, 1); | |
743 | } | |
c525aadd | 744 | mutex_unlock(&kernfs_open_file_mutex); |
414985ae | 745 | |
c637b8ac | 746 | kernfs_put_open_node(kn, NULL); |
414985ae TH |
747 | } |
748 | ||
c637b8ac TH |
749 | /* |
750 | * Kernfs attribute files are pollable. The idea is that you read | |
414985ae TH |
751 | * the content and then you use 'poll' or 'select' to wait for |
752 | * the content to change. When the content changes (assuming the | |
753 | * manager for the kobject supports notification), poll will | |
754 | * return POLLERR|POLLPRI, and select will return the fd whether | |
755 | * it is waiting for read, write, or exceptions. | |
756 | * Once poll/select indicates that the value has changed, you | |
757 | * need to close and re-open the file, or seek to 0 and read again. | |
758 | * Reminder: this only works for attributes which actively support | |
759 | * it, and it is not possible to test an attribute from userspace | |
760 | * to see if it supports poll (Neither 'poll' nor 'select' return | |
761 | * an appropriate error code). When in doubt, set a suitable timeout value. | |
762 | */ | |
c637b8ac | 763 | static unsigned int kernfs_fop_poll(struct file *filp, poll_table *wait) |
414985ae | 764 | { |
c525aadd | 765 | struct kernfs_open_file *of = kernfs_of(filp); |
324a56e1 | 766 | struct kernfs_node *kn = filp->f_path.dentry->d_fsdata; |
c525aadd | 767 | struct kernfs_open_node *on = kn->attr.open; |
414985ae TH |
768 | |
769 | /* need parent for the kobj, grab both */ | |
c637b8ac | 770 | if (!kernfs_get_active(kn)) |
414985ae TH |
771 | goto trigger; |
772 | ||
c525aadd | 773 | poll_wait(filp, &on->poll, wait); |
414985ae | 774 | |
c637b8ac | 775 | kernfs_put_active(kn); |
414985ae | 776 | |
c525aadd | 777 | if (of->event != atomic_read(&on->event)) |
414985ae TH |
778 | goto trigger; |
779 | ||
780 | return DEFAULT_POLLMASK; | |
781 | ||
782 | trigger: | |
783 | return DEFAULT_POLLMASK|POLLERR|POLLPRI; | |
784 | } | |
785 | ||
786 | /** | |
787 | * kernfs_notify - notify a kernfs file | |
324a56e1 | 788 | * @kn: file to notify |
414985ae | 789 | * |
324a56e1 | 790 | * Notify @kn such that poll(2) on @kn wakes up. |
414985ae | 791 | */ |
324a56e1 | 792 | void kernfs_notify(struct kernfs_node *kn) |
414985ae | 793 | { |
d911d987 | 794 | struct kernfs_root *root = kernfs_root(kn); |
c525aadd | 795 | struct kernfs_open_node *on; |
d911d987 | 796 | struct kernfs_super_info *info; |
414985ae TH |
797 | unsigned long flags; |
798 | ||
d911d987 TH |
799 | if (WARN_ON(kernfs_type(kn) != KERNFS_FILE)) |
800 | return; | |
801 | ||
802 | /* kick poll */ | |
c525aadd | 803 | spin_lock_irqsave(&kernfs_open_node_lock, flags); |
414985ae | 804 | |
d911d987 TH |
805 | on = kn->attr.open; |
806 | if (on) { | |
807 | atomic_inc(&on->event); | |
808 | wake_up_interruptible(&on->poll); | |
414985ae TH |
809 | } |
810 | ||
c525aadd | 811 | spin_unlock_irqrestore(&kernfs_open_node_lock, flags); |
d911d987 TH |
812 | |
813 | /* kick fsnotify */ | |
814 | mutex_lock(&kernfs_mutex); | |
815 | ||
816 | list_for_each_entry(info, &root->supers, node) { | |
817 | struct inode *inode; | |
818 | struct dentry *dentry; | |
819 | ||
820 | inode = ilookup(info->sb, kn->ino); | |
821 | if (!inode) | |
822 | continue; | |
823 | ||
824 | dentry = d_find_any_alias(inode); | |
825 | if (dentry) { | |
826 | fsnotify_parent(NULL, dentry, FS_MODIFY); | |
827 | fsnotify(inode, FS_MODIFY, inode, FSNOTIFY_EVENT_INODE, | |
828 | NULL, 0); | |
829 | dput(dentry); | |
830 | } | |
831 | ||
832 | iput(inode); | |
833 | } | |
834 | ||
835 | mutex_unlock(&kernfs_mutex); | |
414985ae TH |
836 | } |
837 | EXPORT_SYMBOL_GPL(kernfs_notify); | |
838 | ||
a797bfc3 | 839 | const struct file_operations kernfs_file_fops = { |
c637b8ac TH |
840 | .read = kernfs_fop_read, |
841 | .write = kernfs_fop_write, | |
414985ae | 842 | .llseek = generic_file_llseek, |
c637b8ac TH |
843 | .mmap = kernfs_fop_mmap, |
844 | .open = kernfs_fop_open, | |
845 | .release = kernfs_fop_release, | |
846 | .poll = kernfs_fop_poll, | |
414985ae TH |
847 | }; |
848 | ||
849 | /** | |
2063d608 | 850 | * __kernfs_create_file - kernfs internal function to create a file |
414985ae TH |
851 | * @parent: directory to create the file in |
852 | * @name: name of the file | |
853 | * @mode: mode of the file | |
854 | * @size: size of the file | |
855 | * @ops: kernfs operations for the file | |
856 | * @priv: private data for the file | |
857 | * @ns: optional namespace tag of the file | |
2063d608 | 858 | * @static_name: don't copy file name |
414985ae TH |
859 | * @key: lockdep key for the file's active_ref, %NULL to disable lockdep |
860 | * | |
861 | * Returns the created node on success, ERR_PTR() value on error. | |
862 | */ | |
2063d608 TH |
863 | struct kernfs_node *__kernfs_create_file(struct kernfs_node *parent, |
864 | const char *name, | |
865 | umode_t mode, loff_t size, | |
866 | const struct kernfs_ops *ops, | |
867 | void *priv, const void *ns, | |
868 | bool name_is_static, | |
869 | struct lock_class_key *key) | |
414985ae | 870 | { |
324a56e1 | 871 | struct kernfs_node *kn; |
2063d608 | 872 | unsigned flags; |
414985ae TH |
873 | int rc; |
874 | ||
2063d608 TH |
875 | flags = KERNFS_FILE; |
876 | if (name_is_static) | |
877 | flags |= KERNFS_STATIC_NAME; | |
878 | ||
db4aad20 | 879 | kn = kernfs_new_node(parent, name, (mode & S_IALLUGO) | S_IFREG, flags); |
324a56e1 | 880 | if (!kn) |
414985ae TH |
881 | return ERR_PTR(-ENOMEM); |
882 | ||
adc5e8b5 TH |
883 | kn->attr.ops = ops; |
884 | kn->attr.size = size; | |
885 | kn->ns = ns; | |
324a56e1 | 886 | kn->priv = priv; |
414985ae TH |
887 | |
888 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
889 | if (key) { | |
324a56e1 | 890 | lockdep_init_map(&kn->dep_map, "s_active", key, 0); |
df23fc39 | 891 | kn->flags |= KERNFS_LOCKDEP; |
414985ae TH |
892 | } |
893 | #endif | |
894 | ||
895 | /* | |
adc5e8b5 | 896 | * kn->attr.ops is accesible only while holding active ref. We |
414985ae TH |
897 | * need to know whether some ops are implemented outside active |
898 | * ref. Cache their existence in flags. | |
899 | */ | |
900 | if (ops->seq_show) | |
df23fc39 | 901 | kn->flags |= KERNFS_HAS_SEQ_SHOW; |
414985ae | 902 | if (ops->mmap) |
df23fc39 | 903 | kn->flags |= KERNFS_HAS_MMAP; |
414985ae | 904 | |
988cd7af | 905 | rc = kernfs_add_one(kn); |
414985ae | 906 | if (rc) { |
324a56e1 | 907 | kernfs_put(kn); |
414985ae TH |
908 | return ERR_PTR(rc); |
909 | } | |
324a56e1 | 910 | return kn; |
414985ae | 911 | } |