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 | { | |
200f1ce3 | 84 | kvfree(array); |
68ab3d88 MH |
85 | } |
86 | ||
b86ff981 JA |
87 | /** |
88 | * relay_mmap_buf: - mmap channel buffer to process address space | |
89 | * @buf: relay channel buffer | |
90 | * @vma: vm_area_struct describing memory to be mapped | |
91 | * | |
92 | * Returns 0 if ok, negative on error | |
93 | * | |
94 | * Caller should already have grabbed mmap_sem. | |
95 | */ | |
01c55ed3 | 96 | static int relay_mmap_buf(struct rchan_buf *buf, struct vm_area_struct *vma) |
b86ff981 JA |
97 | { |
98 | unsigned long length = vma->vm_end - vma->vm_start; | |
99 | struct file *filp = vma->vm_file; | |
100 | ||
101 | if (!buf) | |
102 | return -EBADF; | |
103 | ||
104 | if (length != (unsigned long)buf->chan->alloc_size) | |
105 | return -EINVAL; | |
106 | ||
107 | vma->vm_ops = &relay_file_mmap_ops; | |
2f98735c | 108 | vma->vm_flags |= VM_DONTEXPAND; |
b86ff981 JA |
109 | vma->vm_private_data = buf; |
110 | buf->chan->cb->buf_mapped(buf, filp); | |
111 | ||
112 | return 0; | |
113 | } | |
114 | ||
115 | /** | |
116 | * relay_alloc_buf - allocate a channel buffer | |
117 | * @buf: the buffer struct | |
118 | * @size: total size of the buffer | |
119 | * | |
4c78a663 | 120 | * Returns a pointer to the resulting buffer, %NULL if unsuccessful. The |
221415d7 | 121 | * passed in size will get page aligned, if it isn't already. |
b86ff981 | 122 | */ |
221415d7 | 123 | static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size) |
b86ff981 JA |
124 | { |
125 | void *mem; | |
126 | unsigned int i, j, n_pages; | |
127 | ||
221415d7 JA |
128 | *size = PAGE_ALIGN(*size); |
129 | n_pages = *size >> PAGE_SHIFT; | |
b86ff981 | 130 | |
68ab3d88 | 131 | buf->page_array = relay_alloc_page_array(n_pages); |
b86ff981 JA |
132 | if (!buf->page_array) |
133 | return NULL; | |
134 | ||
135 | for (i = 0; i < n_pages; i++) { | |
136 | buf->page_array[i] = alloc_page(GFP_KERNEL); | |
137 | if (unlikely(!buf->page_array[i])) | |
138 | goto depopulate; | |
ebf99093 | 139 | set_page_private(buf->page_array[i], (unsigned long)buf); |
b86ff981 JA |
140 | } |
141 | mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL); | |
142 | if (!mem) | |
143 | goto depopulate; | |
144 | ||
221415d7 | 145 | memset(mem, 0, *size); |
b86ff981 JA |
146 | buf->page_count = n_pages; |
147 | return mem; | |
148 | ||
149 | depopulate: | |
150 | for (j = 0; j < i; j++) | |
151 | __free_page(buf->page_array[j]); | |
68ab3d88 | 152 | relay_free_page_array(buf->page_array); |
b86ff981 JA |
153 | return NULL; |
154 | } | |
155 | ||
156 | /** | |
157 | * relay_create_buf - allocate and initialize a channel buffer | |
4c78a663 | 158 | * @chan: the relay channel |
b86ff981 | 159 | * |
4c78a663 | 160 | * Returns channel buffer if successful, %NULL otherwise. |
b86ff981 | 161 | */ |
01c55ed3 | 162 | static struct rchan_buf *relay_create_buf(struct rchan *chan) |
b86ff981 | 163 | { |
f6302f1b DC |
164 | struct rchan_buf *buf; |
165 | ||
166 | if (chan->n_subbufs > UINT_MAX / sizeof(size_t *)) | |
b86ff981 JA |
167 | return NULL; |
168 | ||
f6302f1b DC |
169 | buf = kzalloc(sizeof(struct rchan_buf), GFP_KERNEL); |
170 | if (!buf) | |
171 | return NULL; | |
b86ff981 JA |
172 | buf->padding = kmalloc(chan->n_subbufs * sizeof(size_t *), GFP_KERNEL); |
173 | if (!buf->padding) | |
174 | goto free_buf; | |
175 | ||
221415d7 | 176 | buf->start = relay_alloc_buf(buf, &chan->alloc_size); |
b86ff981 JA |
177 | if (!buf->start) |
178 | goto free_buf; | |
179 | ||
180 | buf->chan = chan; | |
181 | kref_get(&buf->chan->kref); | |
182 | return buf; | |
183 | ||
184 | free_buf: | |
185 | kfree(buf->padding); | |
186 | kfree(buf); | |
187 | return NULL; | |
188 | } | |
189 | ||
190 | /** | |
191 | * relay_destroy_channel - free the channel struct | |
4c78a663 | 192 | * @kref: target kernel reference that contains the relay channel |
b86ff981 JA |
193 | * |
194 | * Should only be called from kref_put(). | |
195 | */ | |
01c55ed3 | 196 | static void relay_destroy_channel(struct kref *kref) |
b86ff981 JA |
197 | { |
198 | struct rchan *chan = container_of(kref, struct rchan, kref); | |
199 | kfree(chan); | |
200 | } | |
201 | ||
202 | /** | |
203 | * relay_destroy_buf - destroy an rchan_buf struct and associated buffer | |
204 | * @buf: the buffer struct | |
205 | */ | |
01c55ed3 | 206 | static void relay_destroy_buf(struct rchan_buf *buf) |
b86ff981 JA |
207 | { |
208 | struct rchan *chan = buf->chan; | |
209 | unsigned int i; | |
210 | ||
211 | if (likely(buf->start)) { | |
212 | vunmap(buf->start); | |
213 | for (i = 0; i < buf->page_count; i++) | |
214 | __free_page(buf->page_array[i]); | |
68ab3d88 | 215 | relay_free_page_array(buf->page_array); |
b86ff981 | 216 | } |
017c59c0 | 217 | *per_cpu_ptr(chan->buf, buf->cpu) = NULL; |
b86ff981 JA |
218 | kfree(buf->padding); |
219 | kfree(buf); | |
220 | kref_put(&chan->kref, relay_destroy_channel); | |
221 | } | |
222 | ||
223 | /** | |
224 | * relay_remove_buf - remove a channel buffer | |
4c78a663 | 225 | * @kref: target kernel reference that contains the relay buffer |
b86ff981 | 226 | * |
e227867f | 227 | * Removes the file from the filesystem, which also frees the |
b86ff981 JA |
228 | * rchan_buf_struct and the channel buffer. Should only be called from |
229 | * kref_put(). | |
230 | */ | |
01c55ed3 | 231 | static void relay_remove_buf(struct kref *kref) |
b86ff981 JA |
232 | { |
233 | struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref); | |
b86ff981 JA |
234 | relay_destroy_buf(buf); |
235 | } | |
236 | ||
237 | /** | |
238 | * relay_buf_empty - boolean, is the channel buffer empty? | |
239 | * @buf: channel buffer | |
240 | * | |
241 | * Returns 1 if the buffer is empty, 0 otherwise. | |
242 | */ | |
01c55ed3 | 243 | static int relay_buf_empty(struct rchan_buf *buf) |
b86ff981 JA |
244 | { |
245 | return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1; | |
246 | } | |
b86ff981 JA |
247 | |
248 | /** | |
249 | * relay_buf_full - boolean, is the channel buffer full? | |
250 | * @buf: channel buffer | |
251 | * | |
252 | * Returns 1 if the buffer is full, 0 otherwise. | |
253 | */ | |
254 | int relay_buf_full(struct rchan_buf *buf) | |
255 | { | |
256 | size_t ready = buf->subbufs_produced - buf->subbufs_consumed; | |
257 | return (ready >= buf->chan->n_subbufs) ? 1 : 0; | |
258 | } | |
259 | EXPORT_SYMBOL_GPL(relay_buf_full); | |
260 | ||
261 | /* | |
262 | * High-level relay kernel API and associated functions. | |
263 | */ | |
264 | ||
265 | /* | |
266 | * rchan_callback implementations defining default channel behavior. Used | |
267 | * in place of corresponding NULL values in client callback struct. | |
268 | */ | |
269 | ||
270 | /* | |
271 | * subbuf_start() default callback. Does nothing. | |
272 | */ | |
273 | static int subbuf_start_default_callback (struct rchan_buf *buf, | |
274 | void *subbuf, | |
275 | void *prev_subbuf, | |
276 | size_t prev_padding) | |
277 | { | |
278 | if (relay_buf_full(buf)) | |
279 | return 0; | |
280 | ||
281 | return 1; | |
282 | } | |
283 | ||
284 | /* | |
285 | * buf_mapped() default callback. Does nothing. | |
286 | */ | |
287 | static void buf_mapped_default_callback(struct rchan_buf *buf, | |
288 | struct file *filp) | |
289 | { | |
290 | } | |
291 | ||
292 | /* | |
293 | * buf_unmapped() default callback. Does nothing. | |
294 | */ | |
295 | static void buf_unmapped_default_callback(struct rchan_buf *buf, | |
296 | struct file *filp) | |
297 | { | |
298 | } | |
299 | ||
300 | /* | |
301 | * create_buf_file_create() default callback. Does nothing. | |
302 | */ | |
303 | static struct dentry *create_buf_file_default_callback(const char *filename, | |
304 | struct dentry *parent, | |
f4ae40a6 | 305 | umode_t mode, |
b86ff981 JA |
306 | struct rchan_buf *buf, |
307 | int *is_global) | |
308 | { | |
309 | return NULL; | |
310 | } | |
311 | ||
312 | /* | |
313 | * remove_buf_file() default callback. Does nothing. | |
314 | */ | |
315 | static int remove_buf_file_default_callback(struct dentry *dentry) | |
316 | { | |
317 | return -EINVAL; | |
318 | } | |
319 | ||
320 | /* relay channel default callbacks */ | |
321 | static struct rchan_callbacks default_channel_callbacks = { | |
322 | .subbuf_start = subbuf_start_default_callback, | |
323 | .buf_mapped = buf_mapped_default_callback, | |
324 | .buf_unmapped = buf_unmapped_default_callback, | |
325 | .create_buf_file = create_buf_file_default_callback, | |
326 | .remove_buf_file = remove_buf_file_default_callback, | |
327 | }; | |
328 | ||
329 | /** | |
330 | * wakeup_readers - wake up readers waiting on a channel | |
26b5679e | 331 | * @work: contains the channel buffer |
b86ff981 | 332 | * |
26b5679e | 333 | * This is the function used to defer reader waking |
b86ff981 | 334 | */ |
26b5679e | 335 | static void wakeup_readers(struct irq_work *work) |
b86ff981 | 336 | { |
26b5679e PZ |
337 | struct rchan_buf *buf; |
338 | ||
339 | buf = container_of(work, struct rchan_buf, wakeup_work); | |
b86ff981 JA |
340 | wake_up_interruptible(&buf->read_wait); |
341 | } | |
342 | ||
343 | /** | |
344 | * __relay_reset - reset a channel buffer | |
345 | * @buf: the channel buffer | |
346 | * @init: 1 if this is a first-time initialization | |
347 | * | |
72fd4a35 | 348 | * See relay_reset() for description of effect. |
b86ff981 | 349 | */ |
192636ad | 350 | static void __relay_reset(struct rchan_buf *buf, unsigned int init) |
b86ff981 JA |
351 | { |
352 | size_t i; | |
353 | ||
354 | if (init) { | |
355 | init_waitqueue_head(&buf->read_wait); | |
356 | kref_init(&buf->kref); | |
26b5679e PZ |
357 | init_irq_work(&buf->wakeup_work, wakeup_readers); |
358 | } else { | |
359 | irq_work_sync(&buf->wakeup_work); | |
360 | } | |
b86ff981 JA |
361 | |
362 | buf->subbufs_produced = 0; | |
363 | buf->subbufs_consumed = 0; | |
364 | buf->bytes_consumed = 0; | |
365 | buf->finalized = 0; | |
366 | buf->data = buf->start; | |
367 | buf->offset = 0; | |
368 | ||
369 | for (i = 0; i < buf->chan->n_subbufs; i++) | |
370 | buf->padding[i] = 0; | |
371 | ||
372 | buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0); | |
373 | } | |
374 | ||
375 | /** | |
376 | * relay_reset - reset the channel | |
377 | * @chan: the channel | |
378 | * | |
379 | * This has the effect of erasing all data from all channel buffers | |
380 | * and restarting the channel in its initial state. The buffers | |
381 | * are not freed, so any mappings are still in effect. | |
382 | * | |
72fd4a35 | 383 | * NOTE. Care should be taken that the channel isn't actually |
b86ff981 JA |
384 | * being used by anything when this call is made. |
385 | */ | |
386 | void relay_reset(struct rchan *chan) | |
387 | { | |
017c59c0 | 388 | struct rchan_buf *buf; |
b86ff981 | 389 | unsigned int i; |
b86ff981 JA |
390 | |
391 | if (!chan) | |
392 | return; | |
393 | ||
017c59c0 AG |
394 | if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0))) { |
395 | __relay_reset(buf, 0); | |
23c88752 | 396 | return; |
b86ff981 | 397 | } |
23c88752 MD |
398 | |
399 | mutex_lock(&relay_channels_mutex); | |
98ba4031 | 400 | for_each_possible_cpu(i) |
017c59c0 AG |
401 | if ((buf = *per_cpu_ptr(chan->buf, i))) |
402 | __relay_reset(buf, 0); | |
23c88752 | 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; | |
7682c918 | 411 | d_inode(buf->dentry)->i_size = buf->early_bytes; |
20d8b67c EGM |
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) |
017c59c0 | 447 | return *per_cpu_ptr(chan->buf, 0); |
b86ff981 JA |
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); | |
59dbb2a0 AG |
458 | } else { |
459 | /* Only retrieve global info, nothing more, nothing less */ | |
460 | dentry = chan->cb->create_buf_file(NULL, NULL, | |
461 | S_IRUSR, buf, | |
462 | &chan->is_global); | |
463 | if (WARN_ON(dentry)) | |
464 | goto free_buf; | |
20d8b67c | 465 | } |
23c88752 MD |
466 | |
467 | buf->cpu = cpu; | |
468 | __relay_reset(buf, 1); | |
b86ff981 | 469 | |
23c88752 | 470 | if(chan->is_global) { |
017c59c0 | 471 | *per_cpu_ptr(chan->buf, 0) = buf; |
23c88752 MD |
472 | buf->cpu = 0; |
473 | } | |
474 | ||
20d8b67c | 475 | return buf; |
23c88752 MD |
476 | |
477 | free_buf: | |
478 | relay_destroy_buf(buf); | |
20d8b67c | 479 | return NULL; |
b86ff981 JA |
480 | } |
481 | ||
482 | /** | |
483 | * relay_close_buf - close a channel buffer | |
484 | * @buf: channel buffer | |
485 | * | |
486 | * Marks the buffer finalized and restores the default callbacks. | |
487 | * The channel buffer and channel buffer data structure are then freed | |
488 | * automatically when the last reference is given up. | |
489 | */ | |
192636ad | 490 | static void relay_close_buf(struct rchan_buf *buf) |
b86ff981 JA |
491 | { |
492 | buf->finalized = 1; | |
26b5679e | 493 | irq_work_sync(&buf->wakeup_work); |
b8d4a5bf | 494 | buf->chan->cb->remove_buf_file(buf->dentry); |
b86ff981 JA |
495 | kref_put(&buf->kref, relay_remove_buf); |
496 | } | |
497 | ||
192636ad | 498 | static void setup_callbacks(struct rchan *chan, |
b86ff981 JA |
499 | struct rchan_callbacks *cb) |
500 | { | |
501 | if (!cb) { | |
502 | chan->cb = &default_channel_callbacks; | |
503 | return; | |
504 | } | |
505 | ||
506 | if (!cb->subbuf_start) | |
507 | cb->subbuf_start = subbuf_start_default_callback; | |
508 | if (!cb->buf_mapped) | |
509 | cb->buf_mapped = buf_mapped_default_callback; | |
510 | if (!cb->buf_unmapped) | |
511 | cb->buf_unmapped = buf_unmapped_default_callback; | |
512 | if (!cb->create_buf_file) | |
513 | cb->create_buf_file = create_buf_file_default_callback; | |
514 | if (!cb->remove_buf_file) | |
515 | cb->remove_buf_file = remove_buf_file_default_callback; | |
516 | chan->cb = cb; | |
517 | } | |
518 | ||
e6d4989a | 519 | int relay_prepare_cpu(unsigned int cpu) |
23c88752 | 520 | { |
23c88752 | 521 | struct rchan *chan; |
017c59c0 | 522 | struct rchan_buf *buf; |
23c88752 | 523 | |
e6d4989a RW |
524 | mutex_lock(&relay_channels_mutex); |
525 | list_for_each_entry(chan, &relay_channels, list) { | |
526 | if ((buf = *per_cpu_ptr(chan->buf, cpu))) | |
527 | continue; | |
528 | buf = relay_open_buf(chan, cpu); | |
529 | if (!buf) { | |
530 | pr_err("relay: cpu %d buffer creation failed\n", cpu); | |
531 | mutex_unlock(&relay_channels_mutex); | |
532 | return -ENOMEM; | |
23c88752 | 533 | } |
e6d4989a | 534 | *per_cpu_ptr(chan->buf, cpu) = buf; |
23c88752 | 535 | } |
e6d4989a RW |
536 | mutex_unlock(&relay_channels_mutex); |
537 | return 0; | |
23c88752 MD |
538 | } |
539 | ||
b86ff981 JA |
540 | /** |
541 | * relay_open - create a new relay channel | |
20d8b67c EGM |
542 | * @base_filename: base name of files to create, %NULL for buffering only |
543 | * @parent: dentry of parent directory, %NULL for root directory or buffer | |
b86ff981 JA |
544 | * @subbuf_size: size of sub-buffers |
545 | * @n_subbufs: number of sub-buffers | |
546 | * @cb: client callback functions | |
23c88752 | 547 | * @private_data: user-defined data |
b86ff981 | 548 | * |
4c78a663 | 549 | * Returns channel pointer if successful, %NULL otherwise. |
b86ff981 JA |
550 | * |
551 | * Creates a channel buffer for each cpu using the sizes and | |
552 | * attributes specified. The created channel buffer files | |
553 | * will be named base_filename0...base_filenameN-1. File | |
72fd4a35 | 554 | * permissions will be %S_IRUSR. |
59dbb2a0 AG |
555 | * |
556 | * If opening a buffer (@parent = NULL) that you later wish to register | |
557 | * in a filesystem, call relay_late_setup_files() once the @parent dentry | |
558 | * is available. | |
b86ff981 JA |
559 | */ |
560 | struct rchan *relay_open(const char *base_filename, | |
561 | struct dentry *parent, | |
562 | size_t subbuf_size, | |
563 | size_t n_subbufs, | |
23c88752 MD |
564 | struct rchan_callbacks *cb, |
565 | void *private_data) | |
b86ff981 JA |
566 | { |
567 | unsigned int i; | |
568 | struct rchan *chan; | |
017c59c0 | 569 | struct rchan_buf *buf; |
b86ff981 JA |
570 | |
571 | if (!(subbuf_size && n_subbufs)) | |
572 | return NULL; | |
f6302f1b DC |
573 | if (subbuf_size > UINT_MAX / n_subbufs) |
574 | return NULL; | |
b86ff981 | 575 | |
cd861280 | 576 | chan = kzalloc(sizeof(struct rchan), GFP_KERNEL); |
b86ff981 JA |
577 | if (!chan) |
578 | return NULL; | |
579 | ||
017c59c0 | 580 | chan->buf = alloc_percpu(struct rchan_buf *); |
b86ff981 JA |
581 | chan->version = RELAYFS_CHANNEL_VERSION; |
582 | chan->n_subbufs = n_subbufs; | |
583 | chan->subbuf_size = subbuf_size; | |
a05342cb | 584 | chan->alloc_size = PAGE_ALIGN(subbuf_size * n_subbufs); |
23c88752 MD |
585 | chan->parent = parent; |
586 | chan->private_data = private_data; | |
20d8b67c EGM |
587 | if (base_filename) { |
588 | chan->has_base_filename = 1; | |
589 | strlcpy(chan->base_filename, base_filename, NAME_MAX); | |
590 | } | |
b86ff981 JA |
591 | setup_callbacks(chan, cb); |
592 | kref_init(&chan->kref); | |
593 | ||
23c88752 | 594 | mutex_lock(&relay_channels_mutex); |
b86ff981 | 595 | for_each_online_cpu(i) { |
017c59c0 AG |
596 | buf = relay_open_buf(chan, i); |
597 | if (!buf) | |
b86ff981 | 598 | goto free_bufs; |
017c59c0 | 599 | *per_cpu_ptr(chan->buf, i) = buf; |
b86ff981 | 600 | } |
23c88752 MD |
601 | list_add(&chan->list, &relay_channels); |
602 | mutex_unlock(&relay_channels_mutex); | |
b86ff981 | 603 | |
b86ff981 JA |
604 | return chan; |
605 | ||
606 | free_bufs: | |
98ba4031 | 607 | for_each_possible_cpu(i) { |
017c59c0 AG |
608 | if ((buf = *per_cpu_ptr(chan->buf, i))) |
609 | relay_close_buf(buf); | |
b86ff981 | 610 | } |
b86ff981 | 611 | |
b86ff981 | 612 | kref_put(&chan->kref, relay_destroy_channel); |
23c88752 | 613 | mutex_unlock(&relay_channels_mutex); |
ba62bafe | 614 | kfree(chan); |
b86ff981 JA |
615 | return NULL; |
616 | } | |
617 | EXPORT_SYMBOL_GPL(relay_open); | |
618 | ||
20d8b67c EGM |
619 | struct rchan_percpu_buf_dispatcher { |
620 | struct rchan_buf *buf; | |
621 | struct dentry *dentry; | |
622 | }; | |
623 | ||
624 | /* Called in atomic context. */ | |
625 | static void __relay_set_buf_dentry(void *info) | |
626 | { | |
627 | struct rchan_percpu_buf_dispatcher *p = info; | |
628 | ||
629 | relay_set_buf_dentry(p->buf, p->dentry); | |
630 | } | |
631 | ||
632 | /** | |
633 | * relay_late_setup_files - triggers file creation | |
634 | * @chan: channel to operate on | |
635 | * @base_filename: base name of files to create | |
636 | * @parent: dentry of parent directory, %NULL for root directory | |
637 | * | |
638 | * Returns 0 if successful, non-zero otherwise. | |
639 | * | |
59dbb2a0 AG |
640 | * Use to setup files for a previously buffer-only channel created |
641 | * by relay_open() with a NULL parent dentry. | |
642 | * | |
643 | * For example, this is useful for perfomring early tracing in kernel, | |
644 | * before VFS is up and then exposing the early results once the dentry | |
645 | * is available. | |
20d8b67c EGM |
646 | */ |
647 | int relay_late_setup_files(struct rchan *chan, | |
648 | const char *base_filename, | |
649 | struct dentry *parent) | |
650 | { | |
651 | int err = 0; | |
652 | unsigned int i, curr_cpu; | |
653 | unsigned long flags; | |
654 | struct dentry *dentry; | |
017c59c0 | 655 | struct rchan_buf *buf; |
20d8b67c EGM |
656 | struct rchan_percpu_buf_dispatcher disp; |
657 | ||
658 | if (!chan || !base_filename) | |
659 | return -EINVAL; | |
660 | ||
661 | strlcpy(chan->base_filename, base_filename, NAME_MAX); | |
662 | ||
663 | mutex_lock(&relay_channels_mutex); | |
664 | /* Is chan already set up? */ | |
b786c6a9 JS |
665 | if (unlikely(chan->has_base_filename)) { |
666 | mutex_unlock(&relay_channels_mutex); | |
20d8b67c | 667 | return -EEXIST; |
b786c6a9 | 668 | } |
20d8b67c EGM |
669 | chan->has_base_filename = 1; |
670 | chan->parent = parent; | |
59dbb2a0 AG |
671 | |
672 | if (chan->is_global) { | |
673 | err = -EINVAL; | |
017c59c0 AG |
674 | buf = *per_cpu_ptr(chan->buf, 0); |
675 | if (!WARN_ON_ONCE(!buf)) { | |
676 | dentry = relay_create_buf_file(chan, buf, 0); | |
59dbb2a0 | 677 | if (dentry && !WARN_ON_ONCE(!chan->is_global)) { |
017c59c0 | 678 | relay_set_buf_dentry(buf, dentry); |
59dbb2a0 AG |
679 | err = 0; |
680 | } | |
681 | } | |
682 | mutex_unlock(&relay_channels_mutex); | |
683 | return err; | |
684 | } | |
685 | ||
20d8b67c EGM |
686 | curr_cpu = get_cpu(); |
687 | /* | |
688 | * The CPU hotplug notifier ran before us and created buffers with | |
689 | * no files associated. So it's safe to call relay_setup_buf_file() | |
690 | * on all currently online CPUs. | |
691 | */ | |
692 | for_each_online_cpu(i) { | |
017c59c0 AG |
693 | buf = *per_cpu_ptr(chan->buf, i); |
694 | if (unlikely(!buf)) { | |
7a51cffb | 695 | WARN_ONCE(1, KERN_ERR "CPU has no buffer!\n"); |
20d8b67c EGM |
696 | err = -EINVAL; |
697 | break; | |
698 | } | |
699 | ||
017c59c0 | 700 | dentry = relay_create_buf_file(chan, buf, i); |
20d8b67c EGM |
701 | if (unlikely(!dentry)) { |
702 | err = -EINVAL; | |
703 | break; | |
704 | } | |
705 | ||
706 | if (curr_cpu == i) { | |
707 | local_irq_save(flags); | |
017c59c0 | 708 | relay_set_buf_dentry(buf, dentry); |
20d8b67c EGM |
709 | local_irq_restore(flags); |
710 | } else { | |
017c59c0 | 711 | disp.buf = buf; |
20d8b67c EGM |
712 | disp.dentry = dentry; |
713 | smp_mb(); | |
714 | /* relay_channels_mutex must be held, so wait. */ | |
715 | err = smp_call_function_single(i, | |
716 | __relay_set_buf_dentry, | |
717 | &disp, 1); | |
718 | } | |
719 | if (unlikely(err)) | |
720 | break; | |
721 | } | |
722 | put_cpu(); | |
723 | mutex_unlock(&relay_channels_mutex); | |
724 | ||
725 | return err; | |
726 | } | |
59dbb2a0 | 727 | EXPORT_SYMBOL_GPL(relay_late_setup_files); |
20d8b67c | 728 | |
b86ff981 JA |
729 | /** |
730 | * relay_switch_subbuf - switch to a new sub-buffer | |
731 | * @buf: channel buffer | |
732 | * @length: size of current event | |
733 | * | |
734 | * Returns either the length passed in or 0 if full. | |
735 | * | |
736 | * Performs sub-buffer-switch tasks such as invoking callbacks, | |
737 | * updating padding counts, waking up readers, etc. | |
738 | */ | |
739 | size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length) | |
740 | { | |
741 | void *old, *new; | |
742 | size_t old_subbuf, new_subbuf; | |
743 | ||
744 | if (unlikely(length > buf->chan->subbuf_size)) | |
745 | goto toobig; | |
746 | ||
747 | if (buf->offset != buf->chan->subbuf_size + 1) { | |
748 | buf->prev_padding = buf->chan->subbuf_size - buf->offset; | |
749 | old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs; | |
750 | buf->padding[old_subbuf] = buf->prev_padding; | |
751 | buf->subbufs_produced++; | |
20d8b67c | 752 | if (buf->dentry) |
7682c918 | 753 | d_inode(buf->dentry)->i_size += |
20d8b67c EGM |
754 | buf->chan->subbuf_size - |
755 | buf->padding[old_subbuf]; | |
756 | else | |
757 | buf->early_bytes += buf->chan->subbuf_size - | |
758 | buf->padding[old_subbuf]; | |
221415d7 | 759 | smp_mb(); |
26b5679e | 760 | if (waitqueue_active(&buf->read_wait)) { |
7c9cb383 TZ |
761 | /* |
762 | * Calling wake_up_interruptible() from here | |
763 | * will deadlock if we happen to be logging | |
764 | * from the scheduler (trying to re-grab | |
765 | * rq->lock), so defer it. | |
766 | */ | |
26b5679e PZ |
767 | irq_work_queue(&buf->wakeup_work); |
768 | } | |
b86ff981 JA |
769 | } |
770 | ||
771 | old = buf->data; | |
772 | new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs; | |
773 | new = buf->start + new_subbuf * buf->chan->subbuf_size; | |
774 | buf->offset = 0; | |
775 | if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) { | |
776 | buf->offset = buf->chan->subbuf_size + 1; | |
777 | return 0; | |
778 | } | |
779 | buf->data = new; | |
780 | buf->padding[new_subbuf] = 0; | |
781 | ||
782 | if (unlikely(length + buf->offset > buf->chan->subbuf_size)) | |
783 | goto toobig; | |
784 | ||
785 | return length; | |
786 | ||
787 | toobig: | |
788 | buf->chan->last_toobig = length; | |
789 | return 0; | |
790 | } | |
791 | EXPORT_SYMBOL_GPL(relay_switch_subbuf); | |
792 | ||
793 | /** | |
794 | * relay_subbufs_consumed - update the buffer's sub-buffers-consumed count | |
795 | * @chan: the channel | |
796 | * @cpu: the cpu associated with the channel buffer to update | |
797 | * @subbufs_consumed: number of sub-buffers to add to current buf's count | |
798 | * | |
799 | * Adds to the channel buffer's consumed sub-buffer count. | |
800 | * subbufs_consumed should be the number of sub-buffers newly consumed, | |
801 | * not the total consumed. | |
802 | * | |
72fd4a35 | 803 | * NOTE. Kernel clients don't need to call this function if the channel |
b86ff981 JA |
804 | * mode is 'overwrite'. |
805 | */ | |
806 | void relay_subbufs_consumed(struct rchan *chan, | |
807 | unsigned int cpu, | |
808 | size_t subbufs_consumed) | |
809 | { | |
810 | struct rchan_buf *buf; | |
811 | ||
9a29d0fb | 812 | if (!chan || cpu >= NR_CPUS) |
b86ff981 JA |
813 | return; |
814 | ||
017c59c0 | 815 | buf = *per_cpu_ptr(chan->buf, cpu); |
9a29d0fb | 816 | if (!buf || subbufs_consumed > chan->n_subbufs) |
b86ff981 JA |
817 | return; |
818 | ||
2c53d910 | 819 | if (subbufs_consumed > buf->subbufs_produced - buf->subbufs_consumed) |
b86ff981 | 820 | buf->subbufs_consumed = buf->subbufs_produced; |
2c53d910 AS |
821 | else |
822 | buf->subbufs_consumed += subbufs_consumed; | |
b86ff981 JA |
823 | } |
824 | EXPORT_SYMBOL_GPL(relay_subbufs_consumed); | |
825 | ||
826 | /** | |
827 | * relay_close - close the channel | |
828 | * @chan: the channel | |
829 | * | |
830 | * Closes all channel buffers and frees the channel. | |
831 | */ | |
832 | void relay_close(struct rchan *chan) | |
833 | { | |
017c59c0 | 834 | struct rchan_buf *buf; |
b86ff981 | 835 | unsigned int i; |
b86ff981 JA |
836 | |
837 | if (!chan) | |
838 | return; | |
839 | ||
23c88752 | 840 | mutex_lock(&relay_channels_mutex); |
017c59c0 AG |
841 | if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0))) |
842 | relay_close_buf(buf); | |
23c88752 MD |
843 | else |
844 | for_each_possible_cpu(i) | |
017c59c0 AG |
845 | if ((buf = *per_cpu_ptr(chan->buf, i))) |
846 | relay_close_buf(buf); | |
b86ff981 JA |
847 | |
848 | if (chan->last_toobig) | |
849 | printk(KERN_WARNING "relay: one or more items not logged " | |
850 | "[item size (%Zd) > sub-buffer size (%Zd)]\n", | |
851 | chan->last_toobig, chan->subbuf_size); | |
852 | ||
23c88752 | 853 | list_del(&chan->list); |
b86ff981 | 854 | kref_put(&chan->kref, relay_destroy_channel); |
23c88752 | 855 | mutex_unlock(&relay_channels_mutex); |
b86ff981 JA |
856 | } |
857 | EXPORT_SYMBOL_GPL(relay_close); | |
858 | ||
859 | /** | |
860 | * relay_flush - close the channel | |
861 | * @chan: the channel | |
862 | * | |
4c78a663 | 863 | * Flushes all channel buffers, i.e. forces buffer switch. |
b86ff981 JA |
864 | */ |
865 | void relay_flush(struct rchan *chan) | |
866 | { | |
017c59c0 | 867 | struct rchan_buf *buf; |
b86ff981 | 868 | unsigned int i; |
b86ff981 JA |
869 | |
870 | if (!chan) | |
871 | return; | |
872 | ||
017c59c0 AG |
873 | if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0))) { |
874 | relay_switch_subbuf(buf, 0); | |
23c88752 | 875 | return; |
b86ff981 | 876 | } |
23c88752 MD |
877 | |
878 | mutex_lock(&relay_channels_mutex); | |
879 | for_each_possible_cpu(i) | |
017c59c0 AG |
880 | if ((buf = *per_cpu_ptr(chan->buf, i))) |
881 | relay_switch_subbuf(buf, 0); | |
23c88752 | 882 | mutex_unlock(&relay_channels_mutex); |
b86ff981 JA |
883 | } |
884 | EXPORT_SYMBOL_GPL(relay_flush); | |
885 | ||
886 | /** | |
887 | * relay_file_open - open file op for relay files | |
888 | * @inode: the inode | |
889 | * @filp: the file | |
890 | * | |
891 | * Increments the channel buffer refcount. | |
892 | */ | |
893 | static int relay_file_open(struct inode *inode, struct file *filp) | |
894 | { | |
8e18e294 | 895 | struct rchan_buf *buf = inode->i_private; |
b86ff981 JA |
896 | kref_get(&buf->kref); |
897 | filp->private_data = buf; | |
898 | ||
37529fe9 | 899 | return nonseekable_open(inode, filp); |
b86ff981 JA |
900 | } |
901 | ||
902 | /** | |
903 | * relay_file_mmap - mmap file op for relay files | |
904 | * @filp: the file | |
905 | * @vma: the vma describing what to map | |
906 | * | |
72fd4a35 | 907 | * Calls upon relay_mmap_buf() to map the file into user space. |
b86ff981 JA |
908 | */ |
909 | static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma) | |
910 | { | |
911 | struct rchan_buf *buf = filp->private_data; | |
912 | return relay_mmap_buf(buf, vma); | |
913 | } | |
914 | ||
915 | /** | |
916 | * relay_file_poll - poll file op for relay files | |
917 | * @filp: the file | |
918 | * @wait: poll table | |
919 | * | |
920 | * Poll implemention. | |
921 | */ | |
922 | static unsigned int relay_file_poll(struct file *filp, poll_table *wait) | |
923 | { | |
924 | unsigned int mask = 0; | |
925 | struct rchan_buf *buf = filp->private_data; | |
926 | ||
927 | if (buf->finalized) | |
928 | return POLLERR; | |
929 | ||
930 | if (filp->f_mode & FMODE_READ) { | |
931 | poll_wait(filp, &buf->read_wait, wait); | |
932 | if (!relay_buf_empty(buf)) | |
933 | mask |= POLLIN | POLLRDNORM; | |
934 | } | |
935 | ||
936 | return mask; | |
937 | } | |
938 | ||
939 | /** | |
940 | * relay_file_release - release file op for relay files | |
941 | * @inode: the inode | |
942 | * @filp: the file | |
943 | * | |
944 | * Decrements the channel refcount, as the filesystem is | |
945 | * no longer using it. | |
946 | */ | |
947 | static int relay_file_release(struct inode *inode, struct file *filp) | |
948 | { | |
949 | struct rchan_buf *buf = filp->private_data; | |
950 | kref_put(&buf->kref, relay_remove_buf); | |
951 | ||
952 | return 0; | |
953 | } | |
954 | ||
4c78a663 | 955 | /* |
b86ff981 JA |
956 | * relay_file_read_consume - update the consumed count for the buffer |
957 | */ | |
958 | static void relay_file_read_consume(struct rchan_buf *buf, | |
959 | size_t read_pos, | |
960 | size_t bytes_consumed) | |
961 | { | |
962 | size_t subbuf_size = buf->chan->subbuf_size; | |
963 | size_t n_subbufs = buf->chan->n_subbufs; | |
964 | size_t read_subbuf; | |
965 | ||
32194450 TZ |
966 | if (buf->subbufs_produced == buf->subbufs_consumed && |
967 | buf->offset == buf->bytes_consumed) | |
968 | return; | |
969 | ||
b86ff981 JA |
970 | if (buf->bytes_consumed + bytes_consumed > subbuf_size) { |
971 | relay_subbufs_consumed(buf->chan, buf->cpu, 1); | |
972 | buf->bytes_consumed = 0; | |
973 | } | |
974 | ||
975 | buf->bytes_consumed += bytes_consumed; | |
a66e356c MH |
976 | if (!read_pos) |
977 | read_subbuf = buf->subbufs_consumed % n_subbufs; | |
978 | else | |
979 | read_subbuf = read_pos / buf->chan->subbuf_size; | |
b86ff981 JA |
980 | if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) { |
981 | if ((read_subbuf == buf->subbufs_produced % n_subbufs) && | |
982 | (buf->offset == subbuf_size)) | |
983 | return; | |
984 | relay_subbufs_consumed(buf->chan, buf->cpu, 1); | |
985 | buf->bytes_consumed = 0; | |
986 | } | |
987 | } | |
988 | ||
4c78a663 | 989 | /* |
b86ff981 JA |
990 | * relay_file_read_avail - boolean, are there unconsumed bytes available? |
991 | */ | |
992 | static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos) | |
993 | { | |
b86ff981 JA |
994 | size_t subbuf_size = buf->chan->subbuf_size; |
995 | size_t n_subbufs = buf->chan->n_subbufs; | |
221415d7 JA |
996 | size_t produced = buf->subbufs_produced; |
997 | size_t consumed = buf->subbufs_consumed; | |
b86ff981 | 998 | |
221415d7 | 999 | relay_file_read_consume(buf, read_pos, 0); |
b86ff981 | 1000 | |
32194450 TZ |
1001 | consumed = buf->subbufs_consumed; |
1002 | ||
221415d7 JA |
1003 | if (unlikely(buf->offset > subbuf_size)) { |
1004 | if (produced == consumed) | |
1005 | return 0; | |
1006 | return 1; | |
b86ff981 JA |
1007 | } |
1008 | ||
221415d7 | 1009 | if (unlikely(produced - consumed >= n_subbufs)) { |
a66e356c | 1010 | consumed = produced - n_subbufs + 1; |
221415d7 | 1011 | buf->subbufs_consumed = consumed; |
a66e356c | 1012 | buf->bytes_consumed = 0; |
221415d7 | 1013 | } |
1bfbc608 | 1014 | |
221415d7 JA |
1015 | produced = (produced % n_subbufs) * subbuf_size + buf->offset; |
1016 | consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed; | |
1017 | ||
1018 | if (consumed > produced) | |
1019 | produced += n_subbufs * subbuf_size; | |
1bfbc608 | 1020 | |
32194450 TZ |
1021 | if (consumed == produced) { |
1022 | if (buf->offset == subbuf_size && | |
1023 | buf->subbufs_produced > buf->subbufs_consumed) | |
1024 | return 1; | |
b86ff981 | 1025 | return 0; |
32194450 | 1026 | } |
b86ff981 | 1027 | |
b86ff981 JA |
1028 | return 1; |
1029 | } | |
1030 | ||
1031 | /** | |
1032 | * relay_file_read_subbuf_avail - return bytes available in sub-buffer | |
4c78a663 RD |
1033 | * @read_pos: file read position |
1034 | * @buf: relay channel buffer | |
b86ff981 JA |
1035 | */ |
1036 | static size_t relay_file_read_subbuf_avail(size_t read_pos, | |
1037 | struct rchan_buf *buf) | |
1038 | { | |
1039 | size_t padding, avail = 0; | |
1040 | size_t read_subbuf, read_offset, write_subbuf, write_offset; | |
1041 | size_t subbuf_size = buf->chan->subbuf_size; | |
1042 | ||
1043 | write_subbuf = (buf->data - buf->start) / subbuf_size; | |
1044 | write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset; | |
1045 | read_subbuf = read_pos / subbuf_size; | |
1046 | read_offset = read_pos % subbuf_size; | |
1047 | padding = buf->padding[read_subbuf]; | |
1048 | ||
1049 | if (read_subbuf == write_subbuf) { | |
1050 | if (read_offset + padding < write_offset) | |
1051 | avail = write_offset - (read_offset + padding); | |
1052 | } else | |
1053 | avail = (subbuf_size - padding) - read_offset; | |
1054 | ||
1055 | return avail; | |
1056 | } | |
1057 | ||
1058 | /** | |
1059 | * relay_file_read_start_pos - find the first available byte to read | |
4c78a663 RD |
1060 | * @read_pos: file read position |
1061 | * @buf: relay channel buffer | |
b86ff981 | 1062 | * |
72fd4a35 | 1063 | * If the @read_pos is in the middle of padding, return the |
b86ff981 JA |
1064 | * position of the first actually available byte, otherwise |
1065 | * return the original value. | |
1066 | */ | |
1067 | static size_t relay_file_read_start_pos(size_t read_pos, | |
1068 | struct rchan_buf *buf) | |
1069 | { | |
1070 | size_t read_subbuf, padding, padding_start, padding_end; | |
1071 | size_t subbuf_size = buf->chan->subbuf_size; | |
1072 | size_t n_subbufs = buf->chan->n_subbufs; | |
8d62fdeb | 1073 | size_t consumed = buf->subbufs_consumed % n_subbufs; |
b86ff981 | 1074 | |
8d62fdeb DW |
1075 | if (!read_pos) |
1076 | read_pos = consumed * subbuf_size + buf->bytes_consumed; | |
b86ff981 JA |
1077 | read_subbuf = read_pos / subbuf_size; |
1078 | padding = buf->padding[read_subbuf]; | |
1079 | padding_start = (read_subbuf + 1) * subbuf_size - padding; | |
1080 | padding_end = (read_subbuf + 1) * subbuf_size; | |
1081 | if (read_pos >= padding_start && read_pos < padding_end) { | |
1082 | read_subbuf = (read_subbuf + 1) % n_subbufs; | |
1083 | read_pos = read_subbuf * subbuf_size; | |
1084 | } | |
1085 | ||
1086 | return read_pos; | |
1087 | } | |
1088 | ||
1089 | /** | |
1090 | * relay_file_read_end_pos - return the new read position | |
4c78a663 RD |
1091 | * @read_pos: file read position |
1092 | * @buf: relay channel buffer | |
1093 | * @count: number of bytes to be read | |
b86ff981 JA |
1094 | */ |
1095 | static size_t relay_file_read_end_pos(struct rchan_buf *buf, | |
1096 | size_t read_pos, | |
1097 | size_t count) | |
1098 | { | |
1099 | size_t read_subbuf, padding, end_pos; | |
1100 | size_t subbuf_size = buf->chan->subbuf_size; | |
1101 | size_t n_subbufs = buf->chan->n_subbufs; | |
1102 | ||
1103 | read_subbuf = read_pos / subbuf_size; | |
1104 | padding = buf->padding[read_subbuf]; | |
1105 | if (read_pos % subbuf_size + count + padding == subbuf_size) | |
1106 | end_pos = (read_subbuf + 1) * subbuf_size; | |
1107 | else | |
1108 | end_pos = read_pos + count; | |
1109 | if (end_pos >= subbuf_size * n_subbufs) | |
1110 | end_pos = 0; | |
1111 | ||
1112 | return end_pos; | |
1113 | } | |
1114 | ||
a7c22421 AV |
1115 | static ssize_t relay_file_read(struct file *filp, |
1116 | char __user *buffer, | |
1117 | size_t count, | |
1118 | loff_t *ppos) | |
221415d7 | 1119 | { |
6dac40a7 TZ |
1120 | struct rchan_buf *buf = filp->private_data; |
1121 | size_t read_start, avail; | |
a7c22421 | 1122 | size_t written = 0; |
6dac40a7 | 1123 | int ret; |
221415d7 | 1124 | |
a7c22421 | 1125 | if (!count) |
221415d7 JA |
1126 | return 0; |
1127 | ||
5955102c | 1128 | inode_lock(file_inode(filp)); |
221415d7 | 1129 | do { |
a7c22421 AV |
1130 | void *from; |
1131 | ||
6dac40a7 TZ |
1132 | if (!relay_file_read_avail(buf, *ppos)) |
1133 | break; | |
1134 | ||
1135 | read_start = relay_file_read_start_pos(*ppos, buf); | |
1136 | avail = relay_file_read_subbuf_avail(read_start, buf); | |
1137 | if (!avail) | |
221415d7 | 1138 | break; |
221415d7 | 1139 | |
a7c22421 AV |
1140 | avail = min(count, avail); |
1141 | from = buf->start + read_start; | |
1142 | ret = avail; | |
1143 | if (copy_to_user(buffer, from, avail)) | |
6dac40a7 TZ |
1144 | break; |
1145 | ||
a7c22421 AV |
1146 | buffer += ret; |
1147 | written += ret; | |
1148 | count -= ret; | |
6dac40a7 | 1149 | |
a7c22421 AV |
1150 | relay_file_read_consume(buf, read_start, ret); |
1151 | *ppos = relay_file_read_end_pos(buf, read_start, ret); | |
1152 | } while (count); | |
1153 | inode_unlock(file_inode(filp)); | |
6dac40a7 | 1154 | |
a7c22421 | 1155 | return written; |
6dac40a7 TZ |
1156 | } |
1157 | ||
1db60cf2 JA |
1158 | static void relay_consume_bytes(struct rchan_buf *rbuf, int bytes_consumed) |
1159 | { | |
1160 | rbuf->bytes_consumed += bytes_consumed; | |
1161 | ||
1162 | if (rbuf->bytes_consumed >= rbuf->chan->subbuf_size) { | |
1163 | relay_subbufs_consumed(rbuf->chan, rbuf->cpu, 1); | |
1164 | rbuf->bytes_consumed %= rbuf->chan->subbuf_size; | |
1165 | } | |
1166 | } | |
1167 | ||
ebf99093 TZ |
1168 | static void relay_pipe_buf_release(struct pipe_inode_info *pipe, |
1169 | struct pipe_buffer *buf) | |
6dac40a7 | 1170 | { |
ebf99093 TZ |
1171 | struct rchan_buf *rbuf; |
1172 | ||
1173 | rbuf = (struct rchan_buf *)page_private(buf->page); | |
1db60cf2 | 1174 | relay_consume_bytes(rbuf, buf->private); |
ebf99093 TZ |
1175 | } |
1176 | ||
28dfef8f | 1177 | static const struct pipe_buf_operations relay_pipe_buf_ops = { |
ebf99093 | 1178 | .can_merge = 0, |
cac36bb0 | 1179 | .confirm = generic_pipe_buf_confirm, |
ebf99093 TZ |
1180 | .release = relay_pipe_buf_release, |
1181 | .steal = generic_pipe_buf_steal, | |
1182 | .get = generic_pipe_buf_get, | |
1183 | }; | |
1184 | ||
5eb7f9fa JA |
1185 | static void relay_page_release(struct splice_pipe_desc *spd, unsigned int i) |
1186 | { | |
1187 | } | |
1188 | ||
d3f35d98 | 1189 | /* |
ebf99093 TZ |
1190 | * subbuf_splice_actor - splice up to one subbuf's worth of data |
1191 | */ | |
5f1664f9 | 1192 | static ssize_t subbuf_splice_actor(struct file *in, |
ebf99093 TZ |
1193 | loff_t *ppos, |
1194 | struct pipe_inode_info *pipe, | |
1195 | size_t len, | |
1196 | unsigned int flags, | |
1197 | int *nonpad_ret) | |
1198 | { | |
5f1664f9 | 1199 | unsigned int pidx, poff, total_len, subbuf_pages, nr_pages; |
ebf99093 TZ |
1200 | struct rchan_buf *rbuf = in->private_data; |
1201 | unsigned int subbuf_size = rbuf->chan->subbuf_size; | |
24da24de TZ |
1202 | uint64_t pos = (uint64_t) *ppos; |
1203 | uint32_t alloc_size = (uint32_t) rbuf->chan->alloc_size; | |
1204 | size_t read_start = (size_t) do_div(pos, alloc_size); | |
ebf99093 TZ |
1205 | size_t read_subbuf = read_start / subbuf_size; |
1206 | size_t padding = rbuf->padding[read_subbuf]; | |
1207 | size_t nonpad_end = read_subbuf * subbuf_size + subbuf_size - padding; | |
35f3d14d JA |
1208 | struct page *pages[PIPE_DEF_BUFFERS]; |
1209 | struct partial_page partial[PIPE_DEF_BUFFERS]; | |
1db60cf2 JA |
1210 | struct splice_pipe_desc spd = { |
1211 | .pages = pages, | |
1212 | .nr_pages = 0, | |
047fe360 | 1213 | .nr_pages_max = PIPE_DEF_BUFFERS, |
1db60cf2 JA |
1214 | .partial = partial, |
1215 | .flags = flags, | |
1216 | .ops = &relay_pipe_buf_ops, | |
5eb7f9fa | 1217 | .spd_release = relay_page_release, |
1db60cf2 | 1218 | }; |
5f1664f9 | 1219 | ssize_t ret; |
ebf99093 TZ |
1220 | |
1221 | if (rbuf->subbufs_produced == rbuf->subbufs_consumed) | |
1222 | return 0; | |
35f3d14d JA |
1223 | if (splice_grow_spd(pipe, &spd)) |
1224 | return -ENOMEM; | |
ebf99093 | 1225 | |
1db60cf2 JA |
1226 | /* |
1227 | * Adjust read len, if longer than what is available | |
1228 | */ | |
1229 | if (len > (subbuf_size - read_start % subbuf_size)) | |
1230 | len = subbuf_size - read_start % subbuf_size; | |
ebf99093 TZ |
1231 | |
1232 | subbuf_pages = rbuf->chan->alloc_size >> PAGE_SHIFT; | |
1233 | pidx = (read_start / PAGE_SIZE) % subbuf_pages; | |
1234 | poff = read_start & ~PAGE_MASK; | |
a786c06d | 1235 | nr_pages = min_t(unsigned int, subbuf_pages, spd.nr_pages_max); |
ebf99093 | 1236 | |
16d54669 | 1237 | for (total_len = 0; spd.nr_pages < nr_pages; spd.nr_pages++) { |
1db60cf2 JA |
1238 | unsigned int this_len, this_end, private; |
1239 | unsigned int cur_pos = read_start + total_len; | |
ebf99093 | 1240 | |
1db60cf2 | 1241 | if (!len) |
ebf99093 | 1242 | break; |
ebf99093 | 1243 | |
1db60cf2 JA |
1244 | this_len = min_t(unsigned long, len, PAGE_SIZE - poff); |
1245 | private = this_len; | |
ebf99093 | 1246 | |
1db60cf2 JA |
1247 | spd.pages[spd.nr_pages] = rbuf->page_array[pidx]; |
1248 | spd.partial[spd.nr_pages].offset = poff; | |
ebf99093 | 1249 | |
1db60cf2 JA |
1250 | this_end = cur_pos + this_len; |
1251 | if (this_end >= nonpad_end) { | |
1252 | this_len = nonpad_end - cur_pos; | |
1253 | private = this_len + padding; | |
ebf99093 | 1254 | } |
1db60cf2 JA |
1255 | spd.partial[spd.nr_pages].len = this_len; |
1256 | spd.partial[spd.nr_pages].private = private; | |
ebf99093 | 1257 | |
1db60cf2 JA |
1258 | len -= this_len; |
1259 | total_len += this_len; | |
1260 | poff = 0; | |
1261 | pidx = (pidx + 1) % subbuf_pages; | |
ebf99093 | 1262 | |
1db60cf2 JA |
1263 | if (this_end >= nonpad_end) { |
1264 | spd.nr_pages++; | |
ebf99093 TZ |
1265 | break; |
1266 | } | |
ebf99093 TZ |
1267 | } |
1268 | ||
35f3d14d | 1269 | ret = 0; |
1db60cf2 | 1270 | if (!spd.nr_pages) |
35f3d14d | 1271 | goto out; |
ebf99093 | 1272 | |
1db60cf2 JA |
1273 | ret = *nonpad_ret = splice_to_pipe(pipe, &spd); |
1274 | if (ret < 0 || ret < total_len) | |
35f3d14d | 1275 | goto out; |
ebf99093 | 1276 | |
1db60cf2 JA |
1277 | if (read_start + ret == nonpad_end) |
1278 | ret += padding; | |
1279 | ||
35f3d14d | 1280 | out: |
047fe360 ED |
1281 | splice_shrink_spd(&spd); |
1282 | return ret; | |
ebf99093 TZ |
1283 | } |
1284 | ||
1285 | static ssize_t relay_file_splice_read(struct file *in, | |
1286 | loff_t *ppos, | |
1287 | struct pipe_inode_info *pipe, | |
1288 | size_t len, | |
1289 | unsigned int flags) | |
1290 | { | |
1291 | ssize_t spliced; | |
1292 | int ret; | |
1293 | int nonpad_ret = 0; | |
1294 | ||
1295 | ret = 0; | |
1296 | spliced = 0; | |
1297 | ||
a82c53a0 | 1298 | while (len && !spliced) { |
ebf99093 TZ |
1299 | ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret); |
1300 | if (ret < 0) | |
1301 | break; | |
1302 | else if (!ret) { | |
fbb5b7ae | 1303 | if (flags & SPLICE_F_NONBLOCK) |
ebf99093 | 1304 | ret = -EAGAIN; |
fbb5b7ae | 1305 | break; |
ebf99093 TZ |
1306 | } |
1307 | ||
1308 | *ppos += ret; | |
1309 | if (ret > len) | |
1310 | len = 0; | |
1311 | else | |
1312 | len -= ret; | |
1313 | spliced += nonpad_ret; | |
1314 | nonpad_ret = 0; | |
1315 | } | |
1316 | ||
1317 | if (spliced) | |
1318 | return spliced; | |
1319 | ||
1320 | return ret; | |
221415d7 JA |
1321 | } |
1322 | ||
15ad7cdc | 1323 | const struct file_operations relay_file_operations = { |
b86ff981 JA |
1324 | .open = relay_file_open, |
1325 | .poll = relay_file_poll, | |
1326 | .mmap = relay_file_mmap, | |
1327 | .read = relay_file_read, | |
1328 | .llseek = no_llseek, | |
1329 | .release = relay_file_release, | |
ebf99093 | 1330 | .splice_read = relay_file_splice_read, |
b86ff981 JA |
1331 | }; |
1332 | EXPORT_SYMBOL_GPL(relay_file_operations); |