Commit | Line | Data |
---|---|---|
b86ff981 JA |
1 | /* |
2 | * Public API and common code for kernel->userspace relay file support. | |
3 | * | |
c9b3febc | 4 | * See Documentation/filesystems/relay.txt for an overview. |
b86ff981 JA |
5 | * |
6 | * Copyright (C) 2002-2005 - Tom Zanussi (zanussi@us.ibm.com), IBM Corp | |
7 | * Copyright (C) 1999-2005 - Karim Yaghmour (karim@opersys.com) | |
8 | * | |
9 | * Moved to kernel/relay.c by Paul Mundt, 2006. | |
23c88752 MD |
10 | * November 2006 - CPU hotplug support by Mathieu Desnoyers |
11 | * (mathieu.desnoyers@polymtl.ca) | |
b86ff981 JA |
12 | * |
13 | * This file is released under the GPL. | |
14 | */ | |
15 | #include <linux/errno.h> | |
16 | #include <linux/stddef.h> | |
17 | #include <linux/slab.h> | |
9984de1a | 18 | #include <linux/export.h> |
b86ff981 JA |
19 | #include <linux/string.h> |
20 | #include <linux/relay.h> | |
21 | #include <linux/vmalloc.h> | |
22 | #include <linux/mm.h> | |
23c88752 | 23 | #include <linux/cpu.h> |
d6b29d7c | 24 | #include <linux/splice.h> |
23c88752 MD |
25 | |
26 | /* list of open channels, for cpu hotplug */ | |
27 | static DEFINE_MUTEX(relay_channels_mutex); | |
28 | static LIST_HEAD(relay_channels); | |
b86ff981 JA |
29 | |
30 | /* | |
31 | * close() vm_op implementation for relay file mapping. | |
32 | */ | |
33 | static void relay_file_mmap_close(struct vm_area_struct *vma) | |
34 | { | |
35 | struct rchan_buf *buf = vma->vm_private_data; | |
36 | buf->chan->cb->buf_unmapped(buf, vma->vm_file); | |
37 | } | |
38 | ||
39 | /* | |
a1e09612 | 40 | * fault() vm_op implementation for relay file mapping. |
b86ff981 | 41 | */ |
a1e09612 | 42 | static int relay_buf_fault(struct vm_area_struct *vma, struct vm_fault *vmf) |
b86ff981 JA |
43 | { |
44 | struct page *page; | |
45 | struct rchan_buf *buf = vma->vm_private_data; | |
a1e09612 | 46 | pgoff_t pgoff = vmf->pgoff; |
b86ff981 | 47 | |
b86ff981 | 48 | if (!buf) |
a1e09612 | 49 | return VM_FAULT_OOM; |
b86ff981 | 50 | |
a1e09612 | 51 | page = vmalloc_to_page(buf->start + (pgoff << PAGE_SHIFT)); |
b86ff981 | 52 | if (!page) |
a1e09612 | 53 | return VM_FAULT_SIGBUS; |
b86ff981 | 54 | get_page(page); |
a1e09612 | 55 | vmf->page = page; |
b86ff981 | 56 | |
a1e09612 | 57 | return 0; |
b86ff981 JA |
58 | } |
59 | ||
60 | /* | |
61 | * vm_ops for relay file mappings. | |
62 | */ | |
f0f37e2f | 63 | static const struct vm_operations_struct relay_file_mmap_ops = { |
a1e09612 | 64 | .fault = relay_buf_fault, |
b86ff981 JA |
65 | .close = relay_file_mmap_close, |
66 | }; | |
67 | ||
68ab3d88 MH |
68 | /* |
69 | * allocate an array of pointers of struct page | |
70 | */ | |
71 | static struct page **relay_alloc_page_array(unsigned int n_pages) | |
72 | { | |
408af87a JJ |
73 | const size_t pa_size = n_pages * sizeof(struct page *); |
74 | if (pa_size > PAGE_SIZE) | |
75 | return vzalloc(pa_size); | |
76 | return kzalloc(pa_size, GFP_KERNEL); | |
68ab3d88 MH |
77 | } |
78 | ||
79 | /* | |
80 | * free an array of pointers of struct page | |
81 | */ | |
82 | static void relay_free_page_array(struct page **array) | |
83 | { | |
84 | if (is_vmalloc_addr(array)) | |
85 | vfree(array); | |
86 | else | |
87 | kfree(array); | |
88 | } | |
89 | ||
b86ff981 JA |
90 | /** |
91 | * relay_mmap_buf: - mmap channel buffer to process address space | |
92 | * @buf: relay channel buffer | |
93 | * @vma: vm_area_struct describing memory to be mapped | |
94 | * | |
95 | * Returns 0 if ok, negative on error | |
96 | * | |
97 | * Caller should already have grabbed mmap_sem. | |
98 | */ | |
01c55ed3 | 99 | static int relay_mmap_buf(struct rchan_buf *buf, struct vm_area_struct *vma) |
b86ff981 JA |
100 | { |
101 | unsigned long length = vma->vm_end - vma->vm_start; | |
102 | struct file *filp = vma->vm_file; | |
103 | ||
104 | if (!buf) | |
105 | return -EBADF; | |
106 | ||
107 | if (length != (unsigned long)buf->chan->alloc_size) | |
108 | return -EINVAL; | |
109 | ||
110 | vma->vm_ops = &relay_file_mmap_ops; | |
2f98735c | 111 | vma->vm_flags |= VM_DONTEXPAND; |
b86ff981 JA |
112 | vma->vm_private_data = buf; |
113 | buf->chan->cb->buf_mapped(buf, filp); | |
114 | ||
115 | return 0; | |
116 | } | |
117 | ||
118 | /** | |
119 | * relay_alloc_buf - allocate a channel buffer | |
120 | * @buf: the buffer struct | |
121 | * @size: total size of the buffer | |
122 | * | |
4c78a663 | 123 | * Returns a pointer to the resulting buffer, %NULL if unsuccessful. The |
221415d7 | 124 | * passed in size will get page aligned, if it isn't already. |
b86ff981 | 125 | */ |
221415d7 | 126 | static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size) |
b86ff981 JA |
127 | { |
128 | void *mem; | |
129 | unsigned int i, j, n_pages; | |
130 | ||
221415d7 JA |
131 | *size = PAGE_ALIGN(*size); |
132 | n_pages = *size >> PAGE_SHIFT; | |
b86ff981 | 133 | |
68ab3d88 | 134 | buf->page_array = relay_alloc_page_array(n_pages); |
b86ff981 JA |
135 | if (!buf->page_array) |
136 | return NULL; | |
137 | ||
138 | for (i = 0; i < n_pages; i++) { | |
139 | buf->page_array[i] = alloc_page(GFP_KERNEL); | |
140 | if (unlikely(!buf->page_array[i])) | |
141 | goto depopulate; | |
ebf99093 | 142 | set_page_private(buf->page_array[i], (unsigned long)buf); |
b86ff981 JA |
143 | } |
144 | mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL); | |
145 | if (!mem) | |
146 | goto depopulate; | |
147 | ||
221415d7 | 148 | memset(mem, 0, *size); |
b86ff981 JA |
149 | buf->page_count = n_pages; |
150 | return mem; | |
151 | ||
152 | depopulate: | |
153 | for (j = 0; j < i; j++) | |
154 | __free_page(buf->page_array[j]); | |
68ab3d88 | 155 | relay_free_page_array(buf->page_array); |
b86ff981 JA |
156 | return NULL; |
157 | } | |
158 | ||
159 | /** | |
160 | * relay_create_buf - allocate and initialize a channel buffer | |
4c78a663 | 161 | * @chan: the relay channel |
b86ff981 | 162 | * |
4c78a663 | 163 | * Returns channel buffer if successful, %NULL otherwise. |
b86ff981 | 164 | */ |
01c55ed3 | 165 | static struct rchan_buf *relay_create_buf(struct rchan *chan) |
b86ff981 | 166 | { |
f6302f1b DC |
167 | struct rchan_buf *buf; |
168 | ||
169 | if (chan->n_subbufs > UINT_MAX / sizeof(size_t *)) | |
b86ff981 JA |
170 | return NULL; |
171 | ||
f6302f1b DC |
172 | buf = kzalloc(sizeof(struct rchan_buf), GFP_KERNEL); |
173 | if (!buf) | |
174 | return NULL; | |
b86ff981 JA |
175 | buf->padding = kmalloc(chan->n_subbufs * sizeof(size_t *), GFP_KERNEL); |
176 | if (!buf->padding) | |
177 | goto free_buf; | |
178 | ||
221415d7 | 179 | buf->start = relay_alloc_buf(buf, &chan->alloc_size); |
b86ff981 JA |
180 | if (!buf->start) |
181 | goto free_buf; | |
182 | ||
183 | buf->chan = chan; | |
184 | kref_get(&buf->chan->kref); | |
185 | return buf; | |
186 | ||
187 | free_buf: | |
188 | kfree(buf->padding); | |
189 | kfree(buf); | |
190 | return NULL; | |
191 | } | |
192 | ||
193 | /** | |
194 | * relay_destroy_channel - free the channel struct | |
4c78a663 | 195 | * @kref: target kernel reference that contains the relay channel |
b86ff981 JA |
196 | * |
197 | * Should only be called from kref_put(). | |
198 | */ | |
01c55ed3 | 199 | static void relay_destroy_channel(struct kref *kref) |
b86ff981 JA |
200 | { |
201 | struct rchan *chan = container_of(kref, struct rchan, kref); | |
202 | kfree(chan); | |
203 | } | |
204 | ||
205 | /** | |
206 | * relay_destroy_buf - destroy an rchan_buf struct and associated buffer | |
207 | * @buf: the buffer struct | |
208 | */ | |
01c55ed3 | 209 | static void relay_destroy_buf(struct rchan_buf *buf) |
b86ff981 JA |
210 | { |
211 | struct rchan *chan = buf->chan; | |
212 | unsigned int i; | |
213 | ||
214 | if (likely(buf->start)) { | |
215 | vunmap(buf->start); | |
216 | for (i = 0; i < buf->page_count; i++) | |
217 | __free_page(buf->page_array[i]); | |
68ab3d88 | 218 | relay_free_page_array(buf->page_array); |
b86ff981 | 219 | } |
23c88752 | 220 | chan->buf[buf->cpu] = NULL; |
b86ff981 JA |
221 | kfree(buf->padding); |
222 | kfree(buf); | |
223 | kref_put(&chan->kref, relay_destroy_channel); | |
224 | } | |
225 | ||
226 | /** | |
227 | * relay_remove_buf - remove a channel buffer | |
4c78a663 | 228 | * @kref: target kernel reference that contains the relay buffer |
b86ff981 JA |
229 | * |
230 | * Removes the file from the fileystem, which also frees the | |
231 | * rchan_buf_struct and the channel buffer. Should only be called from | |
232 | * kref_put(). | |
233 | */ | |
01c55ed3 | 234 | static void relay_remove_buf(struct kref *kref) |
b86ff981 JA |
235 | { |
236 | struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref); | |
237 | buf->chan->cb->remove_buf_file(buf->dentry); | |
238 | relay_destroy_buf(buf); | |
239 | } | |
240 | ||
241 | /** | |
242 | * relay_buf_empty - boolean, is the channel buffer empty? | |
243 | * @buf: channel buffer | |
244 | * | |
245 | * Returns 1 if the buffer is empty, 0 otherwise. | |
246 | */ | |
01c55ed3 | 247 | static int relay_buf_empty(struct rchan_buf *buf) |
b86ff981 JA |
248 | { |
249 | return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1; | |
250 | } | |
b86ff981 JA |
251 | |
252 | /** | |
253 | * relay_buf_full - boolean, is the channel buffer full? | |
254 | * @buf: channel buffer | |
255 | * | |
256 | * Returns 1 if the buffer is full, 0 otherwise. | |
257 | */ | |
258 | int relay_buf_full(struct rchan_buf *buf) | |
259 | { | |
260 | size_t ready = buf->subbufs_produced - buf->subbufs_consumed; | |
261 | return (ready >= buf->chan->n_subbufs) ? 1 : 0; | |
262 | } | |
263 | EXPORT_SYMBOL_GPL(relay_buf_full); | |
264 | ||
265 | /* | |
266 | * High-level relay kernel API and associated functions. | |
267 | */ | |
268 | ||
269 | /* | |
270 | * rchan_callback implementations defining default channel behavior. Used | |
271 | * in place of corresponding NULL values in client callback struct. | |
272 | */ | |
273 | ||
274 | /* | |
275 | * subbuf_start() default callback. Does nothing. | |
276 | */ | |
277 | static int subbuf_start_default_callback (struct rchan_buf *buf, | |
278 | void *subbuf, | |
279 | void *prev_subbuf, | |
280 | size_t prev_padding) | |
281 | { | |
282 | if (relay_buf_full(buf)) | |
283 | return 0; | |
284 | ||
285 | return 1; | |
286 | } | |
287 | ||
288 | /* | |
289 | * buf_mapped() default callback. Does nothing. | |
290 | */ | |
291 | static void buf_mapped_default_callback(struct rchan_buf *buf, | |
292 | struct file *filp) | |
293 | { | |
294 | } | |
295 | ||
296 | /* | |
297 | * buf_unmapped() default callback. Does nothing. | |
298 | */ | |
299 | static void buf_unmapped_default_callback(struct rchan_buf *buf, | |
300 | struct file *filp) | |
301 | { | |
302 | } | |
303 | ||
304 | /* | |
305 | * create_buf_file_create() default callback. Does nothing. | |
306 | */ | |
307 | static struct dentry *create_buf_file_default_callback(const char *filename, | |
308 | struct dentry *parent, | |
f4ae40a6 | 309 | umode_t mode, |
b86ff981 JA |
310 | struct rchan_buf *buf, |
311 | int *is_global) | |
312 | { | |
313 | return NULL; | |
314 | } | |
315 | ||
316 | /* | |
317 | * remove_buf_file() default callback. Does nothing. | |
318 | */ | |
319 | static int remove_buf_file_default_callback(struct dentry *dentry) | |
320 | { | |
321 | return -EINVAL; | |
322 | } | |
323 | ||
324 | /* relay channel default callbacks */ | |
325 | static struct rchan_callbacks default_channel_callbacks = { | |
326 | .subbuf_start = subbuf_start_default_callback, | |
327 | .buf_mapped = buf_mapped_default_callback, | |
328 | .buf_unmapped = buf_unmapped_default_callback, | |
329 | .create_buf_file = create_buf_file_default_callback, | |
330 | .remove_buf_file = remove_buf_file_default_callback, | |
331 | }; | |
332 | ||
333 | /** | |
334 | * wakeup_readers - wake up readers waiting on a channel | |
9a9136e2 | 335 | * @data: contains the channel buffer |
b86ff981 | 336 | * |
7c9cb383 | 337 | * This is the timer function used to defer reader waking. |
b86ff981 | 338 | */ |
7c9cb383 | 339 | static void wakeup_readers(unsigned long data) |
b86ff981 | 340 | { |
7c9cb383 | 341 | struct rchan_buf *buf = (struct rchan_buf *)data; |
b86ff981 JA |
342 | wake_up_interruptible(&buf->read_wait); |
343 | } | |
344 | ||
345 | /** | |
346 | * __relay_reset - reset a channel buffer | |
347 | * @buf: the channel buffer | |
348 | * @init: 1 if this is a first-time initialization | |
349 | * | |
72fd4a35 | 350 | * See relay_reset() for description of effect. |
b86ff981 | 351 | */ |
192636ad | 352 | static void __relay_reset(struct rchan_buf *buf, unsigned int init) |
b86ff981 JA |
353 | { |
354 | size_t i; | |
355 | ||
356 | if (init) { | |
357 | init_waitqueue_head(&buf->read_wait); | |
358 | kref_init(&buf->kref); | |
7c9cb383 TZ |
359 | setup_timer(&buf->timer, wakeup_readers, (unsigned long)buf); |
360 | } else | |
361 | del_timer_sync(&buf->timer); | |
b86ff981 JA |
362 | |
363 | buf->subbufs_produced = 0; | |
364 | buf->subbufs_consumed = 0; | |
365 | buf->bytes_consumed = 0; | |
366 | buf->finalized = 0; | |
367 | buf->data = buf->start; | |
368 | buf->offset = 0; | |
369 | ||
370 | for (i = 0; i < buf->chan->n_subbufs; i++) | |
371 | buf->padding[i] = 0; | |
372 | ||
373 | buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0); | |
374 | } | |
375 | ||
376 | /** | |
377 | * relay_reset - reset the channel | |
378 | * @chan: the channel | |
379 | * | |
380 | * This has the effect of erasing all data from all channel buffers | |
381 | * and restarting the channel in its initial state. The buffers | |
382 | * are not freed, so any mappings are still in effect. | |
383 | * | |
72fd4a35 | 384 | * NOTE. Care should be taken that the channel isn't actually |
b86ff981 JA |
385 | * being used by anything when this call is made. |
386 | */ | |
387 | void relay_reset(struct rchan *chan) | |
388 | { | |
389 | unsigned int i; | |
b86ff981 JA |
390 | |
391 | if (!chan) | |
392 | return; | |
393 | ||
1bfbc608 | 394 | if (chan->is_global && chan->buf[0]) { |
23c88752 MD |
395 | __relay_reset(chan->buf[0], 0); |
396 | return; | |
b86ff981 | 397 | } |
23c88752 MD |
398 | |
399 | mutex_lock(&relay_channels_mutex); | |
98ba4031 | 400 | for_each_possible_cpu(i) |
23c88752 MD |
401 | if (chan->buf[i]) |
402 | __relay_reset(chan->buf[i], 0); | |
403 | mutex_unlock(&relay_channels_mutex); | |
b86ff981 JA |
404 | } |
405 | EXPORT_SYMBOL_GPL(relay_reset); | |
406 | ||
20d8b67c EGM |
407 | static inline void relay_set_buf_dentry(struct rchan_buf *buf, |
408 | struct dentry *dentry) | |
409 | { | |
410 | buf->dentry = dentry; | |
411 | buf->dentry->d_inode->i_size = buf->early_bytes; | |
412 | } | |
413 | ||
414 | static struct dentry *relay_create_buf_file(struct rchan *chan, | |
415 | struct rchan_buf *buf, | |
416 | unsigned int cpu) | |
417 | { | |
418 | struct dentry *dentry; | |
419 | char *tmpname; | |
420 | ||
421 | tmpname = kzalloc(NAME_MAX + 1, GFP_KERNEL); | |
422 | if (!tmpname) | |
423 | return NULL; | |
424 | snprintf(tmpname, NAME_MAX, "%s%d", chan->base_filename, cpu); | |
425 | ||
426 | /* Create file in fs */ | |
427 | dentry = chan->cb->create_buf_file(tmpname, chan->parent, | |
428 | S_IRUSR, buf, | |
429 | &chan->is_global); | |
430 | ||
431 | kfree(tmpname); | |
432 | ||
433 | return dentry; | |
434 | } | |
435 | ||
4c78a663 | 436 | /* |
b86ff981 JA |
437 | * relay_open_buf - create a new relay channel buffer |
438 | * | |
23c88752 | 439 | * used by relay_open() and CPU hotplug. |
b86ff981 | 440 | */ |
23c88752 | 441 | static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu) |
b86ff981 | 442 | { |
23c88752 | 443 | struct rchan_buf *buf = NULL; |
b86ff981 JA |
444 | struct dentry *dentry; |
445 | ||
23c88752 | 446 | if (chan->is_global) |
b86ff981 JA |
447 | return chan->buf[0]; |
448 | ||
449 | buf = relay_create_buf(chan); | |
450 | if (!buf) | |
20d8b67c EGM |
451 | return NULL; |
452 | ||
453 | if (chan->has_base_filename) { | |
454 | dentry = relay_create_buf_file(chan, buf, cpu); | |
455 | if (!dentry) | |
456 | goto free_buf; | |
457 | relay_set_buf_dentry(buf, dentry); | |
458 | } | |
23c88752 MD |
459 | |
460 | buf->cpu = cpu; | |
461 | __relay_reset(buf, 1); | |
b86ff981 | 462 | |
23c88752 MD |
463 | if(chan->is_global) { |
464 | chan->buf[0] = buf; | |
465 | buf->cpu = 0; | |
466 | } | |
467 | ||
20d8b67c | 468 | return buf; |
23c88752 MD |
469 | |
470 | free_buf: | |
471 | relay_destroy_buf(buf); | |
20d8b67c | 472 | return NULL; |
b86ff981 JA |
473 | } |
474 | ||
475 | /** | |
476 | * relay_close_buf - close a channel buffer | |
477 | * @buf: channel buffer | |
478 | * | |
479 | * Marks the buffer finalized and restores the default callbacks. | |
480 | * The channel buffer and channel buffer data structure are then freed | |
481 | * automatically when the last reference is given up. | |
482 | */ | |
192636ad | 483 | static void relay_close_buf(struct rchan_buf *buf) |
b86ff981 JA |
484 | { |
485 | buf->finalized = 1; | |
7c9cb383 | 486 | del_timer_sync(&buf->timer); |
b86ff981 JA |
487 | kref_put(&buf->kref, relay_remove_buf); |
488 | } | |
489 | ||
192636ad | 490 | static void setup_callbacks(struct rchan *chan, |
b86ff981 JA |
491 | struct rchan_callbacks *cb) |
492 | { | |
493 | if (!cb) { | |
494 | chan->cb = &default_channel_callbacks; | |
495 | return; | |
496 | } | |
497 | ||
498 | if (!cb->subbuf_start) | |
499 | cb->subbuf_start = subbuf_start_default_callback; | |
500 | if (!cb->buf_mapped) | |
501 | cb->buf_mapped = buf_mapped_default_callback; | |
502 | if (!cb->buf_unmapped) | |
503 | cb->buf_unmapped = buf_unmapped_default_callback; | |
504 | if (!cb->create_buf_file) | |
505 | cb->create_buf_file = create_buf_file_default_callback; | |
506 | if (!cb->remove_buf_file) | |
507 | cb->remove_buf_file = remove_buf_file_default_callback; | |
508 | chan->cb = cb; | |
509 | } | |
510 | ||
23c88752 | 511 | /** |
23c88752 MD |
512 | * relay_hotcpu_callback - CPU hotplug callback |
513 | * @nb: notifier block | |
514 | * @action: hotplug action to take | |
515 | * @hcpu: CPU number | |
516 | * | |
05fb6bf0 | 517 | * Returns the success/failure of the operation. (%NOTIFY_OK, %NOTIFY_BAD) |
23c88752 MD |
518 | */ |
519 | static int __cpuinit relay_hotcpu_callback(struct notifier_block *nb, | |
520 | unsigned long action, | |
521 | void *hcpu) | |
522 | { | |
523 | unsigned int hotcpu = (unsigned long)hcpu; | |
524 | struct rchan *chan; | |
525 | ||
526 | switch(action) { | |
527 | case CPU_UP_PREPARE: | |
8bb78442 | 528 | case CPU_UP_PREPARE_FROZEN: |
23c88752 MD |
529 | mutex_lock(&relay_channels_mutex); |
530 | list_for_each_entry(chan, &relay_channels, list) { | |
531 | if (chan->buf[hotcpu]) | |
532 | continue; | |
533 | chan->buf[hotcpu] = relay_open_buf(chan, hotcpu); | |
534 | if(!chan->buf[hotcpu]) { | |
535 | printk(KERN_ERR | |
536 | "relay_hotcpu_callback: cpu %d buffer " | |
537 | "creation failed\n", hotcpu); | |
538 | mutex_unlock(&relay_channels_mutex); | |
80b5184c | 539 | return notifier_from_errno(-ENOMEM); |
23c88752 MD |
540 | } |
541 | } | |
542 | mutex_unlock(&relay_channels_mutex); | |
543 | break; | |
544 | case CPU_DEAD: | |
8bb78442 | 545 | case CPU_DEAD_FROZEN: |
23c88752 MD |
546 | /* No need to flush the cpu : will be flushed upon |
547 | * final relay_flush() call. */ | |
548 | break; | |
549 | } | |
550 | return NOTIFY_OK; | |
551 | } | |
552 | ||
536b39ec J |
553 | /* Needs a _much_ better name... */ |
554 | #define FIX_SIZE(x) ((((x) - 1) & PAGE_MASK) + PAGE_SIZE) | |
555 | ||
b86ff981 JA |
556 | /** |
557 | * relay_open - create a new relay channel | |
20d8b67c EGM |
558 | * @base_filename: base name of files to create, %NULL for buffering only |
559 | * @parent: dentry of parent directory, %NULL for root directory or buffer | |
b86ff981 JA |
560 | * @subbuf_size: size of sub-buffers |
561 | * @n_subbufs: number of sub-buffers | |
562 | * @cb: client callback functions | |
23c88752 | 563 | * @private_data: user-defined data |
b86ff981 | 564 | * |
4c78a663 | 565 | * Returns channel pointer if successful, %NULL otherwise. |
b86ff981 JA |
566 | * |
567 | * Creates a channel buffer for each cpu using the sizes and | |
568 | * attributes specified. The created channel buffer files | |
569 | * will be named base_filename0...base_filenameN-1. File | |
72fd4a35 | 570 | * permissions will be %S_IRUSR. |
b86ff981 JA |
571 | */ |
572 | struct rchan *relay_open(const char *base_filename, | |
573 | struct dentry *parent, | |
574 | size_t subbuf_size, | |
575 | size_t n_subbufs, | |
23c88752 MD |
576 | struct rchan_callbacks *cb, |
577 | void *private_data) | |
b86ff981 JA |
578 | { |
579 | unsigned int i; | |
580 | struct rchan *chan; | |
b86ff981 JA |
581 | |
582 | if (!(subbuf_size && n_subbufs)) | |
583 | return NULL; | |
f6302f1b DC |
584 | if (subbuf_size > UINT_MAX / n_subbufs) |
585 | return NULL; | |
b86ff981 | 586 | |
cd861280 | 587 | chan = kzalloc(sizeof(struct rchan), GFP_KERNEL); |
b86ff981 JA |
588 | if (!chan) |
589 | return NULL; | |
590 | ||
591 | chan->version = RELAYFS_CHANNEL_VERSION; | |
592 | chan->n_subbufs = n_subbufs; | |
593 | chan->subbuf_size = subbuf_size; | |
594 | chan->alloc_size = FIX_SIZE(subbuf_size * n_subbufs); | |
23c88752 MD |
595 | chan->parent = parent; |
596 | chan->private_data = private_data; | |
20d8b67c EGM |
597 | if (base_filename) { |
598 | chan->has_base_filename = 1; | |
599 | strlcpy(chan->base_filename, base_filename, NAME_MAX); | |
600 | } | |
b86ff981 JA |
601 | setup_callbacks(chan, cb); |
602 | kref_init(&chan->kref); | |
603 | ||
23c88752 | 604 | mutex_lock(&relay_channels_mutex); |
b86ff981 | 605 | for_each_online_cpu(i) { |
23c88752 | 606 | chan->buf[i] = relay_open_buf(chan, i); |
b86ff981 JA |
607 | if (!chan->buf[i]) |
608 | goto free_bufs; | |
b86ff981 | 609 | } |
23c88752 MD |
610 | list_add(&chan->list, &relay_channels); |
611 | mutex_unlock(&relay_channels_mutex); | |
b86ff981 | 612 | |
b86ff981 JA |
613 | return chan; |
614 | ||
615 | free_bufs: | |
98ba4031 LJ |
616 | for_each_possible_cpu(i) { |
617 | if (chan->buf[i]) | |
618 | relay_close_buf(chan->buf[i]); | |
b86ff981 | 619 | } |
b86ff981 | 620 | |
b86ff981 | 621 | kref_put(&chan->kref, relay_destroy_channel); |
23c88752 | 622 | mutex_unlock(&relay_channels_mutex); |
b86ff981 JA |
623 | return NULL; |
624 | } | |
625 | EXPORT_SYMBOL_GPL(relay_open); | |
626 | ||
20d8b67c EGM |
627 | struct rchan_percpu_buf_dispatcher { |
628 | struct rchan_buf *buf; | |
629 | struct dentry *dentry; | |
630 | }; | |
631 | ||
632 | /* Called in atomic context. */ | |
633 | static void __relay_set_buf_dentry(void *info) | |
634 | { | |
635 | struct rchan_percpu_buf_dispatcher *p = info; | |
636 | ||
637 | relay_set_buf_dentry(p->buf, p->dentry); | |
638 | } | |
639 | ||
640 | /** | |
641 | * relay_late_setup_files - triggers file creation | |
642 | * @chan: channel to operate on | |
643 | * @base_filename: base name of files to create | |
644 | * @parent: dentry of parent directory, %NULL for root directory | |
645 | * | |
646 | * Returns 0 if successful, non-zero otherwise. | |
647 | * | |
648 | * Use to setup files for a previously buffer-only channel. | |
649 | * Useful to do early tracing in kernel, before VFS is up, for example. | |
650 | */ | |
651 | int relay_late_setup_files(struct rchan *chan, | |
652 | const char *base_filename, | |
653 | struct dentry *parent) | |
654 | { | |
655 | int err = 0; | |
656 | unsigned int i, curr_cpu; | |
657 | unsigned long flags; | |
658 | struct dentry *dentry; | |
659 | struct rchan_percpu_buf_dispatcher disp; | |
660 | ||
661 | if (!chan || !base_filename) | |
662 | return -EINVAL; | |
663 | ||
664 | strlcpy(chan->base_filename, base_filename, NAME_MAX); | |
665 | ||
666 | mutex_lock(&relay_channels_mutex); | |
667 | /* Is chan already set up? */ | |
b786c6a9 JS |
668 | if (unlikely(chan->has_base_filename)) { |
669 | mutex_unlock(&relay_channels_mutex); | |
20d8b67c | 670 | return -EEXIST; |
b786c6a9 | 671 | } |
20d8b67c EGM |
672 | chan->has_base_filename = 1; |
673 | chan->parent = parent; | |
674 | curr_cpu = get_cpu(); | |
675 | /* | |
676 | * The CPU hotplug notifier ran before us and created buffers with | |
677 | * no files associated. So it's safe to call relay_setup_buf_file() | |
678 | * on all currently online CPUs. | |
679 | */ | |
680 | for_each_online_cpu(i) { | |
681 | if (unlikely(!chan->buf[i])) { | |
7a51cffb | 682 | WARN_ONCE(1, KERN_ERR "CPU has no buffer!\n"); |
20d8b67c EGM |
683 | err = -EINVAL; |
684 | break; | |
685 | } | |
686 | ||
687 | dentry = relay_create_buf_file(chan, chan->buf[i], i); | |
688 | if (unlikely(!dentry)) { | |
689 | err = -EINVAL; | |
690 | break; | |
691 | } | |
692 | ||
693 | if (curr_cpu == i) { | |
694 | local_irq_save(flags); | |
695 | relay_set_buf_dentry(chan->buf[i], dentry); | |
696 | local_irq_restore(flags); | |
697 | } else { | |
698 | disp.buf = chan->buf[i]; | |
699 | disp.dentry = dentry; | |
700 | smp_mb(); | |
701 | /* relay_channels_mutex must be held, so wait. */ | |
702 | err = smp_call_function_single(i, | |
703 | __relay_set_buf_dentry, | |
704 | &disp, 1); | |
705 | } | |
706 | if (unlikely(err)) | |
707 | break; | |
708 | } | |
709 | put_cpu(); | |
710 | mutex_unlock(&relay_channels_mutex); | |
711 | ||
712 | return err; | |
713 | } | |
714 | ||
b86ff981 JA |
715 | /** |
716 | * relay_switch_subbuf - switch to a new sub-buffer | |
717 | * @buf: channel buffer | |
718 | * @length: size of current event | |
719 | * | |
720 | * Returns either the length passed in or 0 if full. | |
721 | * | |
722 | * Performs sub-buffer-switch tasks such as invoking callbacks, | |
723 | * updating padding counts, waking up readers, etc. | |
724 | */ | |
725 | size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length) | |
726 | { | |
727 | void *old, *new; | |
728 | size_t old_subbuf, new_subbuf; | |
729 | ||
730 | if (unlikely(length > buf->chan->subbuf_size)) | |
731 | goto toobig; | |
732 | ||
733 | if (buf->offset != buf->chan->subbuf_size + 1) { | |
734 | buf->prev_padding = buf->chan->subbuf_size - buf->offset; | |
735 | old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs; | |
736 | buf->padding[old_subbuf] = buf->prev_padding; | |
737 | buf->subbufs_produced++; | |
20d8b67c EGM |
738 | if (buf->dentry) |
739 | buf->dentry->d_inode->i_size += | |
740 | buf->chan->subbuf_size - | |
741 | buf->padding[old_subbuf]; | |
742 | else | |
743 | buf->early_bytes += buf->chan->subbuf_size - | |
744 | buf->padding[old_subbuf]; | |
221415d7 | 745 | smp_mb(); |
7c9cb383 TZ |
746 | if (waitqueue_active(&buf->read_wait)) |
747 | /* | |
748 | * Calling wake_up_interruptible() from here | |
749 | * will deadlock if we happen to be logging | |
750 | * from the scheduler (trying to re-grab | |
751 | * rq->lock), so defer it. | |
752 | */ | |
74019224 | 753 | mod_timer(&buf->timer, jiffies + 1); |
b86ff981 JA |
754 | } |
755 | ||
756 | old = buf->data; | |
757 | new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs; | |
758 | new = buf->start + new_subbuf * buf->chan->subbuf_size; | |
759 | buf->offset = 0; | |
760 | if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) { | |
761 | buf->offset = buf->chan->subbuf_size + 1; | |
762 | return 0; | |
763 | } | |
764 | buf->data = new; | |
765 | buf->padding[new_subbuf] = 0; | |
766 | ||
767 | if (unlikely(length + buf->offset > buf->chan->subbuf_size)) | |
768 | goto toobig; | |
769 | ||
770 | return length; | |
771 | ||
772 | toobig: | |
773 | buf->chan->last_toobig = length; | |
774 | return 0; | |
775 | } | |
776 | EXPORT_SYMBOL_GPL(relay_switch_subbuf); | |
777 | ||
778 | /** | |
779 | * relay_subbufs_consumed - update the buffer's sub-buffers-consumed count | |
780 | * @chan: the channel | |
781 | * @cpu: the cpu associated with the channel buffer to update | |
782 | * @subbufs_consumed: number of sub-buffers to add to current buf's count | |
783 | * | |
784 | * Adds to the channel buffer's consumed sub-buffer count. | |
785 | * subbufs_consumed should be the number of sub-buffers newly consumed, | |
786 | * not the total consumed. | |
787 | * | |
72fd4a35 | 788 | * NOTE. Kernel clients don't need to call this function if the channel |
b86ff981 JA |
789 | * mode is 'overwrite'. |
790 | */ | |
791 | void relay_subbufs_consumed(struct rchan *chan, | |
792 | unsigned int cpu, | |
793 | size_t subbufs_consumed) | |
794 | { | |
795 | struct rchan_buf *buf; | |
796 | ||
797 | if (!chan) | |
798 | return; | |
799 | ||
2c53d910 AS |
800 | if (cpu >= NR_CPUS || !chan->buf[cpu] || |
801 | subbufs_consumed > chan->n_subbufs) | |
b86ff981 JA |
802 | return; |
803 | ||
804 | buf = chan->buf[cpu]; | |
2c53d910 | 805 | if (subbufs_consumed > buf->subbufs_produced - buf->subbufs_consumed) |
b86ff981 | 806 | buf->subbufs_consumed = buf->subbufs_produced; |
2c53d910 AS |
807 | else |
808 | buf->subbufs_consumed += subbufs_consumed; | |
b86ff981 JA |
809 | } |
810 | EXPORT_SYMBOL_GPL(relay_subbufs_consumed); | |
811 | ||
812 | /** | |
813 | * relay_close - close the channel | |
814 | * @chan: the channel | |
815 | * | |
816 | * Closes all channel buffers and frees the channel. | |
817 | */ | |
818 | void relay_close(struct rchan *chan) | |
819 | { | |
820 | unsigned int i; | |
b86ff981 JA |
821 | |
822 | if (!chan) | |
823 | return; | |
824 | ||
23c88752 MD |
825 | mutex_lock(&relay_channels_mutex); |
826 | if (chan->is_global && chan->buf[0]) | |
827 | relay_close_buf(chan->buf[0]); | |
828 | else | |
829 | for_each_possible_cpu(i) | |
830 | if (chan->buf[i]) | |
831 | relay_close_buf(chan->buf[i]); | |
b86ff981 JA |
832 | |
833 | if (chan->last_toobig) | |
834 | printk(KERN_WARNING "relay: one or more items not logged " | |
835 | "[item size (%Zd) > sub-buffer size (%Zd)]\n", | |
836 | chan->last_toobig, chan->subbuf_size); | |
837 | ||
23c88752 | 838 | list_del(&chan->list); |
b86ff981 | 839 | kref_put(&chan->kref, relay_destroy_channel); |
23c88752 | 840 | mutex_unlock(&relay_channels_mutex); |
b86ff981 JA |
841 | } |
842 | EXPORT_SYMBOL_GPL(relay_close); | |
843 | ||
844 | /** | |
845 | * relay_flush - close the channel | |
846 | * @chan: the channel | |
847 | * | |
4c78a663 | 848 | * Flushes all channel buffers, i.e. forces buffer switch. |
b86ff981 JA |
849 | */ |
850 | void relay_flush(struct rchan *chan) | |
851 | { | |
852 | unsigned int i; | |
b86ff981 JA |
853 | |
854 | if (!chan) | |
855 | return; | |
856 | ||
23c88752 MD |
857 | if (chan->is_global && chan->buf[0]) { |
858 | relay_switch_subbuf(chan->buf[0], 0); | |
859 | return; | |
b86ff981 | 860 | } |
23c88752 MD |
861 | |
862 | mutex_lock(&relay_channels_mutex); | |
863 | for_each_possible_cpu(i) | |
864 | if (chan->buf[i]) | |
865 | relay_switch_subbuf(chan->buf[i], 0); | |
866 | mutex_unlock(&relay_channels_mutex); | |
b86ff981 JA |
867 | } |
868 | EXPORT_SYMBOL_GPL(relay_flush); | |
869 | ||
870 | /** | |
871 | * relay_file_open - open file op for relay files | |
872 | * @inode: the inode | |
873 | * @filp: the file | |
874 | * | |
875 | * Increments the channel buffer refcount. | |
876 | */ | |
877 | static int relay_file_open(struct inode *inode, struct file *filp) | |
878 | { | |
8e18e294 | 879 | struct rchan_buf *buf = inode->i_private; |
b86ff981 JA |
880 | kref_get(&buf->kref); |
881 | filp->private_data = buf; | |
882 | ||
37529fe9 | 883 | return nonseekable_open(inode, filp); |
b86ff981 JA |
884 | } |
885 | ||
886 | /** | |
887 | * relay_file_mmap - mmap file op for relay files | |
888 | * @filp: the file | |
889 | * @vma: the vma describing what to map | |
890 | * | |
72fd4a35 | 891 | * Calls upon relay_mmap_buf() to map the file into user space. |
b86ff981 JA |
892 | */ |
893 | static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma) | |
894 | { | |
895 | struct rchan_buf *buf = filp->private_data; | |
896 | return relay_mmap_buf(buf, vma); | |
897 | } | |
898 | ||
899 | /** | |
900 | * relay_file_poll - poll file op for relay files | |
901 | * @filp: the file | |
902 | * @wait: poll table | |
903 | * | |
904 | * Poll implemention. | |
905 | */ | |
906 | static unsigned int relay_file_poll(struct file *filp, poll_table *wait) | |
907 | { | |
908 | unsigned int mask = 0; | |
909 | struct rchan_buf *buf = filp->private_data; | |
910 | ||
911 | if (buf->finalized) | |
912 | return POLLERR; | |
913 | ||
914 | if (filp->f_mode & FMODE_READ) { | |
915 | poll_wait(filp, &buf->read_wait, wait); | |
916 | if (!relay_buf_empty(buf)) | |
917 | mask |= POLLIN | POLLRDNORM; | |
918 | } | |
919 | ||
920 | return mask; | |
921 | } | |
922 | ||
923 | /** | |
924 | * relay_file_release - release file op for relay files | |
925 | * @inode: the inode | |
926 | * @filp: the file | |
927 | * | |
928 | * Decrements the channel refcount, as the filesystem is | |
929 | * no longer using it. | |
930 | */ | |
931 | static int relay_file_release(struct inode *inode, struct file *filp) | |
932 | { | |
933 | struct rchan_buf *buf = filp->private_data; | |
934 | kref_put(&buf->kref, relay_remove_buf); | |
935 | ||
936 | return 0; | |
937 | } | |
938 | ||
4c78a663 | 939 | /* |
b86ff981 JA |
940 | * relay_file_read_consume - update the consumed count for the buffer |
941 | */ | |
942 | static void relay_file_read_consume(struct rchan_buf *buf, | |
943 | size_t read_pos, | |
944 | size_t bytes_consumed) | |
945 | { | |
946 | size_t subbuf_size = buf->chan->subbuf_size; | |
947 | size_t n_subbufs = buf->chan->n_subbufs; | |
948 | size_t read_subbuf; | |
949 | ||
32194450 TZ |
950 | if (buf->subbufs_produced == buf->subbufs_consumed && |
951 | buf->offset == buf->bytes_consumed) | |
952 | return; | |
953 | ||
b86ff981 JA |
954 | if (buf->bytes_consumed + bytes_consumed > subbuf_size) { |
955 | relay_subbufs_consumed(buf->chan, buf->cpu, 1); | |
956 | buf->bytes_consumed = 0; | |
957 | } | |
958 | ||
959 | buf->bytes_consumed += bytes_consumed; | |
a66e356c MH |
960 | if (!read_pos) |
961 | read_subbuf = buf->subbufs_consumed % n_subbufs; | |
962 | else | |
963 | read_subbuf = read_pos / buf->chan->subbuf_size; | |
b86ff981 JA |
964 | if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) { |
965 | if ((read_subbuf == buf->subbufs_produced % n_subbufs) && | |
966 | (buf->offset == subbuf_size)) | |
967 | return; | |
968 | relay_subbufs_consumed(buf->chan, buf->cpu, 1); | |
969 | buf->bytes_consumed = 0; | |
970 | } | |
971 | } | |
972 | ||
4c78a663 | 973 | /* |
b86ff981 JA |
974 | * relay_file_read_avail - boolean, are there unconsumed bytes available? |
975 | */ | |
976 | static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos) | |
977 | { | |
b86ff981 JA |
978 | size_t subbuf_size = buf->chan->subbuf_size; |
979 | size_t n_subbufs = buf->chan->n_subbufs; | |
221415d7 JA |
980 | size_t produced = buf->subbufs_produced; |
981 | size_t consumed = buf->subbufs_consumed; | |
b86ff981 | 982 | |
221415d7 | 983 | relay_file_read_consume(buf, read_pos, 0); |
b86ff981 | 984 | |
32194450 TZ |
985 | consumed = buf->subbufs_consumed; |
986 | ||
221415d7 JA |
987 | if (unlikely(buf->offset > subbuf_size)) { |
988 | if (produced == consumed) | |
989 | return 0; | |
990 | return 1; | |
b86ff981 JA |
991 | } |
992 | ||
221415d7 | 993 | if (unlikely(produced - consumed >= n_subbufs)) { |
a66e356c | 994 | consumed = produced - n_subbufs + 1; |
221415d7 | 995 | buf->subbufs_consumed = consumed; |
a66e356c | 996 | buf->bytes_consumed = 0; |
221415d7 | 997 | } |
1bfbc608 | 998 | |
221415d7 JA |
999 | produced = (produced % n_subbufs) * subbuf_size + buf->offset; |
1000 | consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed; | |
1001 | ||
1002 | if (consumed > produced) | |
1003 | produced += n_subbufs * subbuf_size; | |
1bfbc608 | 1004 | |
32194450 TZ |
1005 | if (consumed == produced) { |
1006 | if (buf->offset == subbuf_size && | |
1007 | buf->subbufs_produced > buf->subbufs_consumed) | |
1008 | return 1; | |
b86ff981 | 1009 | return 0; |
32194450 | 1010 | } |
b86ff981 | 1011 | |
b86ff981 JA |
1012 | return 1; |
1013 | } | |
1014 | ||
1015 | /** | |
1016 | * relay_file_read_subbuf_avail - return bytes available in sub-buffer | |
4c78a663 RD |
1017 | * @read_pos: file read position |
1018 | * @buf: relay channel buffer | |
b86ff981 JA |
1019 | */ |
1020 | static size_t relay_file_read_subbuf_avail(size_t read_pos, | |
1021 | struct rchan_buf *buf) | |
1022 | { | |
1023 | size_t padding, avail = 0; | |
1024 | size_t read_subbuf, read_offset, write_subbuf, write_offset; | |
1025 | size_t subbuf_size = buf->chan->subbuf_size; | |
1026 | ||
1027 | write_subbuf = (buf->data - buf->start) / subbuf_size; | |
1028 | write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset; | |
1029 | read_subbuf = read_pos / subbuf_size; | |
1030 | read_offset = read_pos % subbuf_size; | |
1031 | padding = buf->padding[read_subbuf]; | |
1032 | ||
1033 | if (read_subbuf == write_subbuf) { | |
1034 | if (read_offset + padding < write_offset) | |
1035 | avail = write_offset - (read_offset + padding); | |
1036 | } else | |
1037 | avail = (subbuf_size - padding) - read_offset; | |
1038 | ||
1039 | return avail; | |
1040 | } | |
1041 | ||
1042 | /** | |
1043 | * relay_file_read_start_pos - find the first available byte to read | |
4c78a663 RD |
1044 | * @read_pos: file read position |
1045 | * @buf: relay channel buffer | |
b86ff981 | 1046 | * |
72fd4a35 | 1047 | * If the @read_pos is in the middle of padding, return the |
b86ff981 JA |
1048 | * position of the first actually available byte, otherwise |
1049 | * return the original value. | |
1050 | */ | |
1051 | static size_t relay_file_read_start_pos(size_t read_pos, | |
1052 | struct rchan_buf *buf) | |
1053 | { | |
1054 | size_t read_subbuf, padding, padding_start, padding_end; | |
1055 | size_t subbuf_size = buf->chan->subbuf_size; | |
1056 | size_t n_subbufs = buf->chan->n_subbufs; | |
8d62fdeb | 1057 | size_t consumed = buf->subbufs_consumed % n_subbufs; |
b86ff981 | 1058 | |
8d62fdeb DW |
1059 | if (!read_pos) |
1060 | read_pos = consumed * subbuf_size + buf->bytes_consumed; | |
b86ff981 JA |
1061 | read_subbuf = read_pos / subbuf_size; |
1062 | padding = buf->padding[read_subbuf]; | |
1063 | padding_start = (read_subbuf + 1) * subbuf_size - padding; | |
1064 | padding_end = (read_subbuf + 1) * subbuf_size; | |
1065 | if (read_pos >= padding_start && read_pos < padding_end) { | |
1066 | read_subbuf = (read_subbuf + 1) % n_subbufs; | |
1067 | read_pos = read_subbuf * subbuf_size; | |
1068 | } | |
1069 | ||
1070 | return read_pos; | |
1071 | } | |
1072 | ||
1073 | /** | |
1074 | * relay_file_read_end_pos - return the new read position | |
4c78a663 RD |
1075 | * @read_pos: file read position |
1076 | * @buf: relay channel buffer | |
1077 | * @count: number of bytes to be read | |
b86ff981 JA |
1078 | */ |
1079 | static size_t relay_file_read_end_pos(struct rchan_buf *buf, | |
1080 | size_t read_pos, | |
1081 | size_t count) | |
1082 | { | |
1083 | size_t read_subbuf, padding, end_pos; | |
1084 | size_t subbuf_size = buf->chan->subbuf_size; | |
1085 | size_t n_subbufs = buf->chan->n_subbufs; | |
1086 | ||
1087 | read_subbuf = read_pos / subbuf_size; | |
1088 | padding = buf->padding[read_subbuf]; | |
1089 | if (read_pos % subbuf_size + count + padding == subbuf_size) | |
1090 | end_pos = (read_subbuf + 1) * subbuf_size; | |
1091 | else | |
1092 | end_pos = read_pos + count; | |
1093 | if (end_pos >= subbuf_size * n_subbufs) | |
1094 | end_pos = 0; | |
1095 | ||
1096 | return end_pos; | |
1097 | } | |
1098 | ||
4c78a663 | 1099 | /* |
6dac40a7 | 1100 | * subbuf_read_actor - read up to one subbuf's worth of data |
b86ff981 | 1101 | */ |
6dac40a7 TZ |
1102 | static int subbuf_read_actor(size_t read_start, |
1103 | struct rchan_buf *buf, | |
1104 | size_t avail, | |
8359f689 | 1105 | read_descriptor_t *desc) |
b86ff981 | 1106 | { |
b86ff981 | 1107 | void *from; |
6dac40a7 | 1108 | int ret = 0; |
b86ff981 JA |
1109 | |
1110 | from = buf->start + read_start; | |
6dac40a7 | 1111 | ret = avail; |
ba2397ef | 1112 | if (copy_to_user(desc->arg.buf, from, avail)) { |
6dac40a7 TZ |
1113 | desc->error = -EFAULT; |
1114 | ret = 0; | |
b86ff981 | 1115 | } |
6dac40a7 TZ |
1116 | desc->arg.data += ret; |
1117 | desc->written += ret; | |
1118 | desc->count -= ret; | |
1119 | ||
b86ff981 JA |
1120 | return ret; |
1121 | } | |
1122 | ||
6dac40a7 TZ |
1123 | typedef int (*subbuf_actor_t) (size_t read_start, |
1124 | struct rchan_buf *buf, | |
1125 | size_t avail, | |
8359f689 | 1126 | read_descriptor_t *desc); |
6dac40a7 | 1127 | |
4c78a663 | 1128 | /* |
6dac40a7 TZ |
1129 | * relay_file_read_subbufs - read count bytes, bridging subbuf boundaries |
1130 | */ | |
192636ad AM |
1131 | static ssize_t relay_file_read_subbufs(struct file *filp, loff_t *ppos, |
1132 | subbuf_actor_t subbuf_actor, | |
192636ad | 1133 | read_descriptor_t *desc) |
221415d7 | 1134 | { |
6dac40a7 TZ |
1135 | struct rchan_buf *buf = filp->private_data; |
1136 | size_t read_start, avail; | |
6dac40a7 | 1137 | int ret; |
221415d7 | 1138 | |
ba2397ef | 1139 | if (!desc->count) |
221415d7 JA |
1140 | return 0; |
1141 | ||
496ad9aa | 1142 | mutex_lock(&file_inode(filp)->i_mutex); |
221415d7 | 1143 | do { |
6dac40a7 TZ |
1144 | if (!relay_file_read_avail(buf, *ppos)) |
1145 | break; | |
1146 | ||
1147 | read_start = relay_file_read_start_pos(*ppos, buf); | |
1148 | avail = relay_file_read_subbuf_avail(read_start, buf); | |
1149 | if (!avail) | |
221415d7 | 1150 | break; |
221415d7 | 1151 | |
ba2397ef | 1152 | avail = min(desc->count, avail); |
8359f689 | 1153 | ret = subbuf_actor(read_start, buf, avail, desc); |
ba2397ef | 1154 | if (desc->error < 0) |
6dac40a7 TZ |
1155 | break; |
1156 | ||
1157 | if (ret) { | |
1158 | relay_file_read_consume(buf, read_start, ret); | |
1159 | *ppos = relay_file_read_end_pos(buf, read_start, ret); | |
1160 | } | |
ba2397ef | 1161 | } while (desc->count && ret); |
496ad9aa | 1162 | mutex_unlock(&file_inode(filp)->i_mutex); |
6dac40a7 | 1163 | |
ba2397ef | 1164 | return desc->written; |
6dac40a7 TZ |
1165 | } |
1166 | ||
1167 | static ssize_t relay_file_read(struct file *filp, | |
1168 | char __user *buffer, | |
1169 | size_t count, | |
1170 | loff_t *ppos) | |
1171 | { | |
ba2397ef AV |
1172 | read_descriptor_t desc; |
1173 | desc.written = 0; | |
1174 | desc.count = count; | |
1175 | desc.arg.buf = buffer; | |
1176 | desc.error = 0; | |
8359f689 | 1177 | return relay_file_read_subbufs(filp, ppos, subbuf_read_actor, &desc); |
6dac40a7 TZ |
1178 | } |
1179 | ||
1db60cf2 JA |
1180 | static void relay_consume_bytes(struct rchan_buf *rbuf, int bytes_consumed) |
1181 | { | |
1182 | rbuf->bytes_consumed += bytes_consumed; | |
1183 | ||
1184 | if (rbuf->bytes_consumed >= rbuf->chan->subbuf_size) { | |
1185 | relay_subbufs_consumed(rbuf->chan, rbuf->cpu, 1); | |
1186 | rbuf->bytes_consumed %= rbuf->chan->subbuf_size; | |
1187 | } | |
1188 | } | |
1189 | ||
ebf99093 TZ |
1190 | static void relay_pipe_buf_release(struct pipe_inode_info *pipe, |
1191 | struct pipe_buffer *buf) | |
6dac40a7 | 1192 | { |
ebf99093 TZ |
1193 | struct rchan_buf *rbuf; |
1194 | ||
1195 | rbuf = (struct rchan_buf *)page_private(buf->page); | |
1db60cf2 | 1196 | relay_consume_bytes(rbuf, buf->private); |
ebf99093 TZ |
1197 | } |
1198 | ||
28dfef8f | 1199 | static const struct pipe_buf_operations relay_pipe_buf_ops = { |
ebf99093 TZ |
1200 | .can_merge = 0, |
1201 | .map = generic_pipe_buf_map, | |
1202 | .unmap = generic_pipe_buf_unmap, | |
cac36bb0 | 1203 | .confirm = generic_pipe_buf_confirm, |
ebf99093 TZ |
1204 | .release = relay_pipe_buf_release, |
1205 | .steal = generic_pipe_buf_steal, | |
1206 | .get = generic_pipe_buf_get, | |
1207 | }; | |
1208 | ||
5eb7f9fa JA |
1209 | static void relay_page_release(struct splice_pipe_desc *spd, unsigned int i) |
1210 | { | |
1211 | } | |
1212 | ||
d3f35d98 | 1213 | /* |
ebf99093 TZ |
1214 | * subbuf_splice_actor - splice up to one subbuf's worth of data |
1215 | */ | |
5f1664f9 | 1216 | static ssize_t subbuf_splice_actor(struct file *in, |
ebf99093 TZ |
1217 | loff_t *ppos, |
1218 | struct pipe_inode_info *pipe, | |
1219 | size_t len, | |
1220 | unsigned int flags, | |
1221 | int *nonpad_ret) | |
1222 | { | |
5f1664f9 | 1223 | unsigned int pidx, poff, total_len, subbuf_pages, nr_pages; |
ebf99093 TZ |
1224 | struct rchan_buf *rbuf = in->private_data; |
1225 | unsigned int subbuf_size = rbuf->chan->subbuf_size; | |
24da24de TZ |
1226 | uint64_t pos = (uint64_t) *ppos; |
1227 | uint32_t alloc_size = (uint32_t) rbuf->chan->alloc_size; | |
1228 | size_t read_start = (size_t) do_div(pos, alloc_size); | |
ebf99093 TZ |
1229 | size_t read_subbuf = read_start / subbuf_size; |
1230 | size_t padding = rbuf->padding[read_subbuf]; | |
1231 | size_t nonpad_end = read_subbuf * subbuf_size + subbuf_size - padding; | |
35f3d14d JA |
1232 | struct page *pages[PIPE_DEF_BUFFERS]; |
1233 | struct partial_page partial[PIPE_DEF_BUFFERS]; | |
1db60cf2 JA |
1234 | struct splice_pipe_desc spd = { |
1235 | .pages = pages, | |
1236 | .nr_pages = 0, | |
047fe360 | 1237 | .nr_pages_max = PIPE_DEF_BUFFERS, |
1db60cf2 JA |
1238 | .partial = partial, |
1239 | .flags = flags, | |
1240 | .ops = &relay_pipe_buf_ops, | |
5eb7f9fa | 1241 | .spd_release = relay_page_release, |
1db60cf2 | 1242 | }; |
5f1664f9 | 1243 | ssize_t ret; |
ebf99093 TZ |
1244 | |
1245 | if (rbuf->subbufs_produced == rbuf->subbufs_consumed) | |
1246 | return 0; | |
35f3d14d JA |
1247 | if (splice_grow_spd(pipe, &spd)) |
1248 | return -ENOMEM; | |
ebf99093 | 1249 | |
1db60cf2 JA |
1250 | /* |
1251 | * Adjust read len, if longer than what is available | |
1252 | */ | |
1253 | if (len > (subbuf_size - read_start % subbuf_size)) | |
1254 | len = subbuf_size - read_start % subbuf_size; | |
ebf99093 TZ |
1255 | |
1256 | subbuf_pages = rbuf->chan->alloc_size >> PAGE_SHIFT; | |
1257 | pidx = (read_start / PAGE_SIZE) % subbuf_pages; | |
1258 | poff = read_start & ~PAGE_MASK; | |
35f3d14d | 1259 | nr_pages = min_t(unsigned int, subbuf_pages, pipe->buffers); |
ebf99093 | 1260 | |
16d54669 | 1261 | for (total_len = 0; spd.nr_pages < nr_pages; spd.nr_pages++) { |
1db60cf2 JA |
1262 | unsigned int this_len, this_end, private; |
1263 | unsigned int cur_pos = read_start + total_len; | |
ebf99093 | 1264 | |
1db60cf2 | 1265 | if (!len) |
ebf99093 | 1266 | break; |
ebf99093 | 1267 | |
1db60cf2 JA |
1268 | this_len = min_t(unsigned long, len, PAGE_SIZE - poff); |
1269 | private = this_len; | |
ebf99093 | 1270 | |
1db60cf2 JA |
1271 | spd.pages[spd.nr_pages] = rbuf->page_array[pidx]; |
1272 | spd.partial[spd.nr_pages].offset = poff; | |
ebf99093 | 1273 | |
1db60cf2 JA |
1274 | this_end = cur_pos + this_len; |
1275 | if (this_end >= nonpad_end) { | |
1276 | this_len = nonpad_end - cur_pos; | |
1277 | private = this_len + padding; | |
ebf99093 | 1278 | } |
1db60cf2 JA |
1279 | spd.partial[spd.nr_pages].len = this_len; |
1280 | spd.partial[spd.nr_pages].private = private; | |
ebf99093 | 1281 | |
1db60cf2 JA |
1282 | len -= this_len; |
1283 | total_len += this_len; | |
1284 | poff = 0; | |
1285 | pidx = (pidx + 1) % subbuf_pages; | |
ebf99093 | 1286 | |
1db60cf2 JA |
1287 | if (this_end >= nonpad_end) { |
1288 | spd.nr_pages++; | |
ebf99093 TZ |
1289 | break; |
1290 | } | |
ebf99093 TZ |
1291 | } |
1292 | ||
35f3d14d | 1293 | ret = 0; |
1db60cf2 | 1294 | if (!spd.nr_pages) |
35f3d14d | 1295 | goto out; |
ebf99093 | 1296 | |
1db60cf2 JA |
1297 | ret = *nonpad_ret = splice_to_pipe(pipe, &spd); |
1298 | if (ret < 0 || ret < total_len) | |
35f3d14d | 1299 | goto out; |
ebf99093 | 1300 | |
1db60cf2 JA |
1301 | if (read_start + ret == nonpad_end) |
1302 | ret += padding; | |
1303 | ||
35f3d14d | 1304 | out: |
047fe360 ED |
1305 | splice_shrink_spd(&spd); |
1306 | return ret; | |
ebf99093 TZ |
1307 | } |
1308 | ||
1309 | static ssize_t relay_file_splice_read(struct file *in, | |
1310 | loff_t *ppos, | |
1311 | struct pipe_inode_info *pipe, | |
1312 | size_t len, | |
1313 | unsigned int flags) | |
1314 | { | |
1315 | ssize_t spliced; | |
1316 | int ret; | |
1317 | int nonpad_ret = 0; | |
1318 | ||
1319 | ret = 0; | |
1320 | spliced = 0; | |
1321 | ||
a82c53a0 | 1322 | while (len && !spliced) { |
ebf99093 TZ |
1323 | ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret); |
1324 | if (ret < 0) | |
1325 | break; | |
1326 | else if (!ret) { | |
fbb5b7ae | 1327 | if (flags & SPLICE_F_NONBLOCK) |
ebf99093 | 1328 | ret = -EAGAIN; |
fbb5b7ae | 1329 | break; |
ebf99093 TZ |
1330 | } |
1331 | ||
1332 | *ppos += ret; | |
1333 | if (ret > len) | |
1334 | len = 0; | |
1335 | else | |
1336 | len -= ret; | |
1337 | spliced += nonpad_ret; | |
1338 | nonpad_ret = 0; | |
1339 | } | |
1340 | ||
1341 | if (spliced) | |
1342 | return spliced; | |
1343 | ||
1344 | return ret; | |
221415d7 JA |
1345 | } |
1346 | ||
15ad7cdc | 1347 | const struct file_operations relay_file_operations = { |
b86ff981 JA |
1348 | .open = relay_file_open, |
1349 | .poll = relay_file_poll, | |
1350 | .mmap = relay_file_mmap, | |
1351 | .read = relay_file_read, | |
1352 | .llseek = no_llseek, | |
1353 | .release = relay_file_release, | |
ebf99093 | 1354 | .splice_read = relay_file_splice_read, |
b86ff981 JA |
1355 | }; |
1356 | EXPORT_SYMBOL_GPL(relay_file_operations); | |
23c88752 MD |
1357 | |
1358 | static __init int relay_init(void) | |
1359 | { | |
1360 | ||
1361 | hotcpu_notifier(relay_hotcpu_callback, 0); | |
1362 | return 0; | |
1363 | } | |
1364 | ||
20d8b67c | 1365 | early_initcall(relay_init); |