Commit | Line | Data |
---|---|---|
b86ff981 JA |
1 | /* |
2 | * Public API and common code for kernel->userspace relay file support. | |
3 | * | |
0c1bc6b8 | 4 | * See Documentation/filesystems/relay.rst 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 | 29 | |
b86ff981 | 30 | /* |
a1e09612 | 31 | * fault() vm_op implementation for relay file mapping. |
b86ff981 | 32 | */ |
3fb3894b | 33 | static vm_fault_t relay_buf_fault(struct vm_fault *vmf) |
b86ff981 JA |
34 | { |
35 | struct page *page; | |
11bac800 | 36 | struct rchan_buf *buf = vmf->vma->vm_private_data; |
a1e09612 | 37 | pgoff_t pgoff = vmf->pgoff; |
b86ff981 | 38 | |
b86ff981 | 39 | if (!buf) |
a1e09612 | 40 | return VM_FAULT_OOM; |
b86ff981 | 41 | |
a1e09612 | 42 | page = vmalloc_to_page(buf->start + (pgoff << PAGE_SHIFT)); |
b86ff981 | 43 | if (!page) |
a1e09612 | 44 | return VM_FAULT_SIGBUS; |
b86ff981 | 45 | get_page(page); |
a1e09612 | 46 | vmf->page = page; |
b86ff981 | 47 | |
a1e09612 | 48 | return 0; |
b86ff981 JA |
49 | } |
50 | ||
51 | /* | |
52 | * vm_ops for relay file mappings. | |
53 | */ | |
f0f37e2f | 54 | static const struct vm_operations_struct relay_file_mmap_ops = { |
a1e09612 | 55 | .fault = relay_buf_fault, |
b86ff981 JA |
56 | }; |
57 | ||
68ab3d88 MH |
58 | /* |
59 | * allocate an array of pointers of struct page | |
60 | */ | |
61 | static struct page **relay_alloc_page_array(unsigned int n_pages) | |
62 | { | |
408af87a JJ |
63 | const size_t pa_size = n_pages * sizeof(struct page *); |
64 | if (pa_size > PAGE_SIZE) | |
65 | return vzalloc(pa_size); | |
66 | return kzalloc(pa_size, GFP_KERNEL); | |
68ab3d88 MH |
67 | } |
68 | ||
69 | /* | |
70 | * free an array of pointers of struct page | |
71 | */ | |
72 | static void relay_free_page_array(struct page **array) | |
73 | { | |
200f1ce3 | 74 | kvfree(array); |
68ab3d88 MH |
75 | } |
76 | ||
b86ff981 JA |
77 | /** |
78 | * relay_mmap_buf: - mmap channel buffer to process address space | |
79 | * @buf: relay channel buffer | |
80 | * @vma: vm_area_struct describing memory to be mapped | |
81 | * | |
82 | * Returns 0 if ok, negative on error | |
83 | * | |
c1e8d7c6 | 84 | * Caller should already have grabbed mmap_lock. |
b86ff981 | 85 | */ |
01c55ed3 | 86 | static int relay_mmap_buf(struct rchan_buf *buf, struct vm_area_struct *vma) |
b86ff981 JA |
87 | { |
88 | unsigned long length = vma->vm_end - vma->vm_start; | |
b86ff981 JA |
89 | |
90 | if (!buf) | |
91 | return -EBADF; | |
92 | ||
93 | if (length != (unsigned long)buf->chan->alloc_size) | |
94 | return -EINVAL; | |
95 | ||
96 | vma->vm_ops = &relay_file_mmap_ops; | |
2f98735c | 97 | vma->vm_flags |= VM_DONTEXPAND; |
b86ff981 | 98 | vma->vm_private_data = buf; |
b86ff981 JA |
99 | |
100 | return 0; | |
101 | } | |
102 | ||
103 | /** | |
104 | * relay_alloc_buf - allocate a channel buffer | |
105 | * @buf: the buffer struct | |
106 | * @size: total size of the buffer | |
107 | * | |
4c78a663 | 108 | * Returns a pointer to the resulting buffer, %NULL if unsuccessful. The |
221415d7 | 109 | * passed in size will get page aligned, if it isn't already. |
b86ff981 | 110 | */ |
221415d7 | 111 | static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size) |
b86ff981 JA |
112 | { |
113 | void *mem; | |
114 | unsigned int i, j, n_pages; | |
115 | ||
221415d7 JA |
116 | *size = PAGE_ALIGN(*size); |
117 | n_pages = *size >> PAGE_SHIFT; | |
b86ff981 | 118 | |
68ab3d88 | 119 | buf->page_array = relay_alloc_page_array(n_pages); |
b86ff981 JA |
120 | if (!buf->page_array) |
121 | return NULL; | |
122 | ||
123 | for (i = 0; i < n_pages; i++) { | |
124 | buf->page_array[i] = alloc_page(GFP_KERNEL); | |
125 | if (unlikely(!buf->page_array[i])) | |
126 | goto depopulate; | |
ebf99093 | 127 | set_page_private(buf->page_array[i], (unsigned long)buf); |
b86ff981 JA |
128 | } |
129 | mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL); | |
130 | if (!mem) | |
131 | goto depopulate; | |
132 | ||
221415d7 | 133 | memset(mem, 0, *size); |
b86ff981 JA |
134 | buf->page_count = n_pages; |
135 | return mem; | |
136 | ||
137 | depopulate: | |
138 | for (j = 0; j < i; j++) | |
139 | __free_page(buf->page_array[j]); | |
68ab3d88 | 140 | relay_free_page_array(buf->page_array); |
b86ff981 JA |
141 | return NULL; |
142 | } | |
143 | ||
144 | /** | |
145 | * relay_create_buf - allocate and initialize a channel buffer | |
4c78a663 | 146 | * @chan: the relay channel |
b86ff981 | 147 | * |
4c78a663 | 148 | * Returns channel buffer if successful, %NULL otherwise. |
b86ff981 | 149 | */ |
01c55ed3 | 150 | static struct rchan_buf *relay_create_buf(struct rchan *chan) |
b86ff981 | 151 | { |
f6302f1b DC |
152 | struct rchan_buf *buf; |
153 | ||
88913bd8 | 154 | if (chan->n_subbufs > KMALLOC_MAX_SIZE / sizeof(size_t *)) |
b86ff981 JA |
155 | return NULL; |
156 | ||
f6302f1b DC |
157 | buf = kzalloc(sizeof(struct rchan_buf), GFP_KERNEL); |
158 | if (!buf) | |
159 | return NULL; | |
6da2ec56 KC |
160 | buf->padding = kmalloc_array(chan->n_subbufs, sizeof(size_t *), |
161 | GFP_KERNEL); | |
b86ff981 JA |
162 | if (!buf->padding) |
163 | goto free_buf; | |
164 | ||
221415d7 | 165 | buf->start = relay_alloc_buf(buf, &chan->alloc_size); |
b86ff981 JA |
166 | if (!buf->start) |
167 | goto free_buf; | |
168 | ||
169 | buf->chan = chan; | |
170 | kref_get(&buf->chan->kref); | |
171 | return buf; | |
172 | ||
173 | free_buf: | |
174 | kfree(buf->padding); | |
175 | kfree(buf); | |
176 | return NULL; | |
177 | } | |
178 | ||
179 | /** | |
180 | * relay_destroy_channel - free the channel struct | |
4c78a663 | 181 | * @kref: target kernel reference that contains the relay channel |
b86ff981 JA |
182 | * |
183 | * Should only be called from kref_put(). | |
184 | */ | |
01c55ed3 | 185 | static void relay_destroy_channel(struct kref *kref) |
b86ff981 JA |
186 | { |
187 | struct rchan *chan = container_of(kref, struct rchan, kref); | |
71e84329 | 188 | free_percpu(chan->buf); |
b86ff981 JA |
189 | kfree(chan); |
190 | } | |
191 | ||
192 | /** | |
193 | * relay_destroy_buf - destroy an rchan_buf struct and associated buffer | |
194 | * @buf: the buffer struct | |
195 | */ | |
01c55ed3 | 196 | static void relay_destroy_buf(struct rchan_buf *buf) |
b86ff981 JA |
197 | { |
198 | struct rchan *chan = buf->chan; | |
199 | unsigned int i; | |
200 | ||
201 | if (likely(buf->start)) { | |
202 | vunmap(buf->start); | |
203 | for (i = 0; i < buf->page_count; i++) | |
204 | __free_page(buf->page_array[i]); | |
68ab3d88 | 205 | relay_free_page_array(buf->page_array); |
b86ff981 | 206 | } |
017c59c0 | 207 | *per_cpu_ptr(chan->buf, buf->cpu) = NULL; |
b86ff981 JA |
208 | kfree(buf->padding); |
209 | kfree(buf); | |
210 | kref_put(&chan->kref, relay_destroy_channel); | |
211 | } | |
212 | ||
213 | /** | |
214 | * relay_remove_buf - remove a channel buffer | |
4c78a663 | 215 | * @kref: target kernel reference that contains the relay buffer |
b86ff981 | 216 | * |
e227867f | 217 | * Removes the file from the filesystem, which also frees the |
b86ff981 JA |
218 | * rchan_buf_struct and the channel buffer. Should only be called from |
219 | * kref_put(). | |
220 | */ | |
01c55ed3 | 221 | static void relay_remove_buf(struct kref *kref) |
b86ff981 JA |
222 | { |
223 | struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref); | |
b86ff981 JA |
224 | relay_destroy_buf(buf); |
225 | } | |
226 | ||
227 | /** | |
228 | * relay_buf_empty - boolean, is the channel buffer empty? | |
229 | * @buf: channel buffer | |
230 | * | |
231 | * Returns 1 if the buffer is empty, 0 otherwise. | |
232 | */ | |
01c55ed3 | 233 | static int relay_buf_empty(struct rchan_buf *buf) |
b86ff981 JA |
234 | { |
235 | return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1; | |
236 | } | |
b86ff981 JA |
237 | |
238 | /** | |
239 | * relay_buf_full - boolean, is the channel buffer full? | |
240 | * @buf: channel buffer | |
241 | * | |
242 | * Returns 1 if the buffer is full, 0 otherwise. | |
243 | */ | |
244 | int relay_buf_full(struct rchan_buf *buf) | |
245 | { | |
246 | size_t ready = buf->subbufs_produced - buf->subbufs_consumed; | |
247 | return (ready >= buf->chan->n_subbufs) ? 1 : 0; | |
248 | } | |
249 | EXPORT_SYMBOL_GPL(relay_buf_full); | |
250 | ||
251 | /* | |
252 | * High-level relay kernel API and associated functions. | |
253 | */ | |
254 | ||
255 | /* | |
256 | * rchan_callback implementations defining default channel behavior. Used | |
257 | * in place of corresponding NULL values in client callback struct. | |
258 | */ | |
259 | ||
260 | /* | |
261 | * subbuf_start() default callback. Does nothing. | |
262 | */ | |
263 | static int subbuf_start_default_callback (struct rchan_buf *buf, | |
264 | void *subbuf, | |
265 | void *prev_subbuf, | |
266 | size_t prev_padding) | |
267 | { | |
268 | if (relay_buf_full(buf)) | |
269 | return 0; | |
270 | ||
271 | return 1; | |
272 | } | |
273 | ||
b86ff981 JA |
274 | /* |
275 | * create_buf_file_create() default callback. Does nothing. | |
276 | */ | |
277 | static struct dentry *create_buf_file_default_callback(const char *filename, | |
278 | struct dentry *parent, | |
f4ae40a6 | 279 | umode_t mode, |
b86ff981 JA |
280 | struct rchan_buf *buf, |
281 | int *is_global) | |
282 | { | |
283 | return NULL; | |
284 | } | |
285 | ||
286 | /* | |
287 | * remove_buf_file() default callback. Does nothing. | |
288 | */ | |
289 | static int remove_buf_file_default_callback(struct dentry *dentry) | |
290 | { | |
291 | return -EINVAL; | |
292 | } | |
293 | ||
294 | /* relay channel default callbacks */ | |
295 | static struct rchan_callbacks default_channel_callbacks = { | |
296 | .subbuf_start = subbuf_start_default_callback, | |
b86ff981 JA |
297 | .create_buf_file = create_buf_file_default_callback, |
298 | .remove_buf_file = remove_buf_file_default_callback, | |
299 | }; | |
300 | ||
301 | /** | |
302 | * wakeup_readers - wake up readers waiting on a channel | |
26b5679e | 303 | * @work: contains the channel buffer |
b86ff981 | 304 | * |
26b5679e | 305 | * This is the function used to defer reader waking |
b86ff981 | 306 | */ |
26b5679e | 307 | static void wakeup_readers(struct irq_work *work) |
b86ff981 | 308 | { |
26b5679e PZ |
309 | struct rchan_buf *buf; |
310 | ||
311 | buf = container_of(work, struct rchan_buf, wakeup_work); | |
b86ff981 JA |
312 | wake_up_interruptible(&buf->read_wait); |
313 | } | |
314 | ||
315 | /** | |
316 | * __relay_reset - reset a channel buffer | |
317 | * @buf: the channel buffer | |
318 | * @init: 1 if this is a first-time initialization | |
319 | * | |
72fd4a35 | 320 | * See relay_reset() for description of effect. |
b86ff981 | 321 | */ |
192636ad | 322 | static void __relay_reset(struct rchan_buf *buf, unsigned int init) |
b86ff981 JA |
323 | { |
324 | size_t i; | |
325 | ||
326 | if (init) { | |
327 | init_waitqueue_head(&buf->read_wait); | |
328 | kref_init(&buf->kref); | |
26b5679e PZ |
329 | init_irq_work(&buf->wakeup_work, wakeup_readers); |
330 | } else { | |
331 | irq_work_sync(&buf->wakeup_work); | |
332 | } | |
b86ff981 JA |
333 | |
334 | buf->subbufs_produced = 0; | |
335 | buf->subbufs_consumed = 0; | |
336 | buf->bytes_consumed = 0; | |
337 | buf->finalized = 0; | |
338 | buf->data = buf->start; | |
339 | buf->offset = 0; | |
340 | ||
341 | for (i = 0; i < buf->chan->n_subbufs; i++) | |
342 | buf->padding[i] = 0; | |
343 | ||
344 | buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0); | |
345 | } | |
346 | ||
347 | /** | |
348 | * relay_reset - reset the channel | |
349 | * @chan: the channel | |
350 | * | |
351 | * This has the effect of erasing all data from all channel buffers | |
352 | * and restarting the channel in its initial state. The buffers | |
353 | * are not freed, so any mappings are still in effect. | |
354 | * | |
72fd4a35 | 355 | * NOTE. Care should be taken that the channel isn't actually |
b86ff981 JA |
356 | * being used by anything when this call is made. |
357 | */ | |
358 | void relay_reset(struct rchan *chan) | |
359 | { | |
017c59c0 | 360 | struct rchan_buf *buf; |
b86ff981 | 361 | unsigned int i; |
b86ff981 JA |
362 | |
363 | if (!chan) | |
364 | return; | |
365 | ||
017c59c0 AG |
366 | if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0))) { |
367 | __relay_reset(buf, 0); | |
23c88752 | 368 | return; |
b86ff981 | 369 | } |
23c88752 MD |
370 | |
371 | mutex_lock(&relay_channels_mutex); | |
98ba4031 | 372 | for_each_possible_cpu(i) |
017c59c0 AG |
373 | if ((buf = *per_cpu_ptr(chan->buf, i))) |
374 | __relay_reset(buf, 0); | |
23c88752 | 375 | mutex_unlock(&relay_channels_mutex); |
b86ff981 JA |
376 | } |
377 | EXPORT_SYMBOL_GPL(relay_reset); | |
378 | ||
20d8b67c EGM |
379 | static inline void relay_set_buf_dentry(struct rchan_buf *buf, |
380 | struct dentry *dentry) | |
381 | { | |
382 | buf->dentry = dentry; | |
7682c918 | 383 | d_inode(buf->dentry)->i_size = buf->early_bytes; |
20d8b67c EGM |
384 | } |
385 | ||
386 | static struct dentry *relay_create_buf_file(struct rchan *chan, | |
387 | struct rchan_buf *buf, | |
388 | unsigned int cpu) | |
389 | { | |
390 | struct dentry *dentry; | |
391 | char *tmpname; | |
392 | ||
393 | tmpname = kzalloc(NAME_MAX + 1, GFP_KERNEL); | |
394 | if (!tmpname) | |
395 | return NULL; | |
396 | snprintf(tmpname, NAME_MAX, "%s%d", chan->base_filename, cpu); | |
397 | ||
398 | /* Create file in fs */ | |
399 | dentry = chan->cb->create_buf_file(tmpname, chan->parent, | |
400 | S_IRUSR, buf, | |
401 | &chan->is_global); | |
2c1cf00e GKH |
402 | if (IS_ERR(dentry)) |
403 | dentry = NULL; | |
20d8b67c EGM |
404 | |
405 | kfree(tmpname); | |
406 | ||
407 | return dentry; | |
408 | } | |
409 | ||
4c78a663 | 410 | /* |
b86ff981 JA |
411 | * relay_open_buf - create a new relay channel buffer |
412 | * | |
23c88752 | 413 | * used by relay_open() and CPU hotplug. |
b86ff981 | 414 | */ |
23c88752 | 415 | static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu) |
b86ff981 | 416 | { |
23c88752 | 417 | struct rchan_buf *buf = NULL; |
b86ff981 JA |
418 | struct dentry *dentry; |
419 | ||
23c88752 | 420 | if (chan->is_global) |
017c59c0 | 421 | return *per_cpu_ptr(chan->buf, 0); |
b86ff981 JA |
422 | |
423 | buf = relay_create_buf(chan); | |
424 | if (!buf) | |
20d8b67c EGM |
425 | return NULL; |
426 | ||
427 | if (chan->has_base_filename) { | |
428 | dentry = relay_create_buf_file(chan, buf, cpu); | |
429 | if (!dentry) | |
430 | goto free_buf; | |
431 | relay_set_buf_dentry(buf, dentry); | |
59dbb2a0 AG |
432 | } else { |
433 | /* Only retrieve global info, nothing more, nothing less */ | |
434 | dentry = chan->cb->create_buf_file(NULL, NULL, | |
435 | S_IRUSR, buf, | |
436 | &chan->is_global); | |
2c1cf00e | 437 | if (IS_ERR_OR_NULL(dentry)) |
59dbb2a0 | 438 | goto free_buf; |
20d8b67c | 439 | } |
23c88752 MD |
440 | |
441 | buf->cpu = cpu; | |
442 | __relay_reset(buf, 1); | |
b86ff981 | 443 | |
23c88752 | 444 | if(chan->is_global) { |
017c59c0 | 445 | *per_cpu_ptr(chan->buf, 0) = buf; |
23c88752 MD |
446 | buf->cpu = 0; |
447 | } | |
448 | ||
20d8b67c | 449 | return buf; |
23c88752 MD |
450 | |
451 | free_buf: | |
452 | relay_destroy_buf(buf); | |
20d8b67c | 453 | return NULL; |
b86ff981 JA |
454 | } |
455 | ||
456 | /** | |
457 | * relay_close_buf - close a channel buffer | |
458 | * @buf: channel buffer | |
459 | * | |
460 | * Marks the buffer finalized and restores the default callbacks. | |
461 | * The channel buffer and channel buffer data structure are then freed | |
462 | * automatically when the last reference is given up. | |
463 | */ | |
192636ad | 464 | static void relay_close_buf(struct rchan_buf *buf) |
b86ff981 JA |
465 | { |
466 | buf->finalized = 1; | |
26b5679e | 467 | irq_work_sync(&buf->wakeup_work); |
b8d4a5bf | 468 | buf->chan->cb->remove_buf_file(buf->dentry); |
b86ff981 JA |
469 | kref_put(&buf->kref, relay_remove_buf); |
470 | } | |
471 | ||
192636ad | 472 | static void setup_callbacks(struct rchan *chan, |
b86ff981 JA |
473 | struct rchan_callbacks *cb) |
474 | { | |
475 | if (!cb) { | |
476 | chan->cb = &default_channel_callbacks; | |
477 | return; | |
478 | } | |
479 | ||
480 | if (!cb->subbuf_start) | |
481 | cb->subbuf_start = subbuf_start_default_callback; | |
b86ff981 JA |
482 | if (!cb->create_buf_file) |
483 | cb->create_buf_file = create_buf_file_default_callback; | |
484 | if (!cb->remove_buf_file) | |
485 | cb->remove_buf_file = remove_buf_file_default_callback; | |
486 | chan->cb = cb; | |
487 | } | |
488 | ||
e6d4989a | 489 | int relay_prepare_cpu(unsigned int cpu) |
23c88752 | 490 | { |
23c88752 | 491 | struct rchan *chan; |
017c59c0 | 492 | struct rchan_buf *buf; |
23c88752 | 493 | |
e6d4989a RW |
494 | mutex_lock(&relay_channels_mutex); |
495 | list_for_each_entry(chan, &relay_channels, list) { | |
496 | if ((buf = *per_cpu_ptr(chan->buf, cpu))) | |
497 | continue; | |
498 | buf = relay_open_buf(chan, cpu); | |
499 | if (!buf) { | |
500 | pr_err("relay: cpu %d buffer creation failed\n", cpu); | |
501 | mutex_unlock(&relay_channels_mutex); | |
502 | return -ENOMEM; | |
23c88752 | 503 | } |
e6d4989a | 504 | *per_cpu_ptr(chan->buf, cpu) = buf; |
23c88752 | 505 | } |
e6d4989a RW |
506 | mutex_unlock(&relay_channels_mutex); |
507 | return 0; | |
23c88752 MD |
508 | } |
509 | ||
b86ff981 JA |
510 | /** |
511 | * relay_open - create a new relay channel | |
20d8b67c EGM |
512 | * @base_filename: base name of files to create, %NULL for buffering only |
513 | * @parent: dentry of parent directory, %NULL for root directory or buffer | |
b86ff981 JA |
514 | * @subbuf_size: size of sub-buffers |
515 | * @n_subbufs: number of sub-buffers | |
516 | * @cb: client callback functions | |
23c88752 | 517 | * @private_data: user-defined data |
b86ff981 | 518 | * |
4c78a663 | 519 | * Returns channel pointer if successful, %NULL otherwise. |
b86ff981 JA |
520 | * |
521 | * Creates a channel buffer for each cpu using the sizes and | |
522 | * attributes specified. The created channel buffer files | |
523 | * will be named base_filename0...base_filenameN-1. File | |
72fd4a35 | 524 | * permissions will be %S_IRUSR. |
59dbb2a0 AG |
525 | * |
526 | * If opening a buffer (@parent = NULL) that you later wish to register | |
527 | * in a filesystem, call relay_late_setup_files() once the @parent dentry | |
528 | * is available. | |
b86ff981 JA |
529 | */ |
530 | struct rchan *relay_open(const char *base_filename, | |
531 | struct dentry *parent, | |
532 | size_t subbuf_size, | |
533 | size_t n_subbufs, | |
23c88752 MD |
534 | struct rchan_callbacks *cb, |
535 | void *private_data) | |
b86ff981 JA |
536 | { |
537 | unsigned int i; | |
538 | struct rchan *chan; | |
017c59c0 | 539 | struct rchan_buf *buf; |
b86ff981 JA |
540 | |
541 | if (!(subbuf_size && n_subbufs)) | |
542 | return NULL; | |
f6302f1b DC |
543 | if (subbuf_size > UINT_MAX / n_subbufs) |
544 | return NULL; | |
b86ff981 | 545 | |
cd861280 | 546 | chan = kzalloc(sizeof(struct rchan), GFP_KERNEL); |
b86ff981 JA |
547 | if (!chan) |
548 | return NULL; | |
549 | ||
017c59c0 | 550 | chan->buf = alloc_percpu(struct rchan_buf *); |
54e200ab DA |
551 | if (!chan->buf) { |
552 | kfree(chan); | |
553 | return NULL; | |
554 | } | |
555 | ||
b86ff981 JA |
556 | chan->version = RELAYFS_CHANNEL_VERSION; |
557 | chan->n_subbufs = n_subbufs; | |
558 | chan->subbuf_size = subbuf_size; | |
a05342cb | 559 | chan->alloc_size = PAGE_ALIGN(subbuf_size * n_subbufs); |
23c88752 MD |
560 | chan->parent = parent; |
561 | chan->private_data = private_data; | |
20d8b67c EGM |
562 | if (base_filename) { |
563 | chan->has_base_filename = 1; | |
564 | strlcpy(chan->base_filename, base_filename, NAME_MAX); | |
565 | } | |
b86ff981 JA |
566 | setup_callbacks(chan, cb); |
567 | kref_init(&chan->kref); | |
568 | ||
23c88752 | 569 | mutex_lock(&relay_channels_mutex); |
b86ff981 | 570 | for_each_online_cpu(i) { |
017c59c0 AG |
571 | buf = relay_open_buf(chan, i); |
572 | if (!buf) | |
b86ff981 | 573 | goto free_bufs; |
017c59c0 | 574 | *per_cpu_ptr(chan->buf, i) = buf; |
b86ff981 | 575 | } |
23c88752 MD |
576 | list_add(&chan->list, &relay_channels); |
577 | mutex_unlock(&relay_channels_mutex); | |
b86ff981 | 578 | |
b86ff981 JA |
579 | return chan; |
580 | ||
581 | free_bufs: | |
98ba4031 | 582 | for_each_possible_cpu(i) { |
017c59c0 AG |
583 | if ((buf = *per_cpu_ptr(chan->buf, i))) |
584 | relay_close_buf(buf); | |
b86ff981 | 585 | } |
b86ff981 | 586 | |
b86ff981 | 587 | kref_put(&chan->kref, relay_destroy_channel); |
23c88752 | 588 | mutex_unlock(&relay_channels_mutex); |
b86ff981 JA |
589 | return NULL; |
590 | } | |
591 | EXPORT_SYMBOL_GPL(relay_open); | |
592 | ||
20d8b67c EGM |
593 | struct rchan_percpu_buf_dispatcher { |
594 | struct rchan_buf *buf; | |
595 | struct dentry *dentry; | |
596 | }; | |
597 | ||
598 | /* Called in atomic context. */ | |
599 | static void __relay_set_buf_dentry(void *info) | |
600 | { | |
601 | struct rchan_percpu_buf_dispatcher *p = info; | |
602 | ||
603 | relay_set_buf_dentry(p->buf, p->dentry); | |
604 | } | |
605 | ||
606 | /** | |
607 | * relay_late_setup_files - triggers file creation | |
608 | * @chan: channel to operate on | |
609 | * @base_filename: base name of files to create | |
610 | * @parent: dentry of parent directory, %NULL for root directory | |
611 | * | |
612 | * Returns 0 if successful, non-zero otherwise. | |
613 | * | |
59dbb2a0 AG |
614 | * Use to setup files for a previously buffer-only channel created |
615 | * by relay_open() with a NULL parent dentry. | |
616 | * | |
617 | * For example, this is useful for perfomring early tracing in kernel, | |
618 | * before VFS is up and then exposing the early results once the dentry | |
619 | * is available. | |
20d8b67c EGM |
620 | */ |
621 | int relay_late_setup_files(struct rchan *chan, | |
622 | const char *base_filename, | |
623 | struct dentry *parent) | |
624 | { | |
625 | int err = 0; | |
626 | unsigned int i, curr_cpu; | |
627 | unsigned long flags; | |
628 | struct dentry *dentry; | |
017c59c0 | 629 | struct rchan_buf *buf; |
20d8b67c EGM |
630 | struct rchan_percpu_buf_dispatcher disp; |
631 | ||
632 | if (!chan || !base_filename) | |
633 | return -EINVAL; | |
634 | ||
635 | strlcpy(chan->base_filename, base_filename, NAME_MAX); | |
636 | ||
637 | mutex_lock(&relay_channels_mutex); | |
638 | /* Is chan already set up? */ | |
b786c6a9 JS |
639 | if (unlikely(chan->has_base_filename)) { |
640 | mutex_unlock(&relay_channels_mutex); | |
20d8b67c | 641 | return -EEXIST; |
b786c6a9 | 642 | } |
20d8b67c EGM |
643 | chan->has_base_filename = 1; |
644 | chan->parent = parent; | |
59dbb2a0 AG |
645 | |
646 | if (chan->is_global) { | |
647 | err = -EINVAL; | |
017c59c0 AG |
648 | buf = *per_cpu_ptr(chan->buf, 0); |
649 | if (!WARN_ON_ONCE(!buf)) { | |
650 | dentry = relay_create_buf_file(chan, buf, 0); | |
59dbb2a0 | 651 | if (dentry && !WARN_ON_ONCE(!chan->is_global)) { |
017c59c0 | 652 | relay_set_buf_dentry(buf, dentry); |
59dbb2a0 AG |
653 | err = 0; |
654 | } | |
655 | } | |
656 | mutex_unlock(&relay_channels_mutex); | |
657 | return err; | |
658 | } | |
659 | ||
20d8b67c EGM |
660 | curr_cpu = get_cpu(); |
661 | /* | |
662 | * The CPU hotplug notifier ran before us and created buffers with | |
663 | * no files associated. So it's safe to call relay_setup_buf_file() | |
664 | * on all currently online CPUs. | |
665 | */ | |
666 | for_each_online_cpu(i) { | |
017c59c0 AG |
667 | buf = *per_cpu_ptr(chan->buf, i); |
668 | if (unlikely(!buf)) { | |
7a51cffb | 669 | WARN_ONCE(1, KERN_ERR "CPU has no buffer!\n"); |
20d8b67c EGM |
670 | err = -EINVAL; |
671 | break; | |
672 | } | |
673 | ||
017c59c0 | 674 | dentry = relay_create_buf_file(chan, buf, i); |
20d8b67c EGM |
675 | if (unlikely(!dentry)) { |
676 | err = -EINVAL; | |
677 | break; | |
678 | } | |
679 | ||
680 | if (curr_cpu == i) { | |
681 | local_irq_save(flags); | |
017c59c0 | 682 | relay_set_buf_dentry(buf, dentry); |
20d8b67c EGM |
683 | local_irq_restore(flags); |
684 | } else { | |
017c59c0 | 685 | disp.buf = buf; |
20d8b67c EGM |
686 | disp.dentry = dentry; |
687 | smp_mb(); | |
688 | /* relay_channels_mutex must be held, so wait. */ | |
689 | err = smp_call_function_single(i, | |
690 | __relay_set_buf_dentry, | |
691 | &disp, 1); | |
692 | } | |
693 | if (unlikely(err)) | |
694 | break; | |
695 | } | |
696 | put_cpu(); | |
697 | mutex_unlock(&relay_channels_mutex); | |
698 | ||
699 | return err; | |
700 | } | |
59dbb2a0 | 701 | EXPORT_SYMBOL_GPL(relay_late_setup_files); |
20d8b67c | 702 | |
b86ff981 JA |
703 | /** |
704 | * relay_switch_subbuf - switch to a new sub-buffer | |
705 | * @buf: channel buffer | |
706 | * @length: size of current event | |
707 | * | |
708 | * Returns either the length passed in or 0 if full. | |
709 | * | |
710 | * Performs sub-buffer-switch tasks such as invoking callbacks, | |
711 | * updating padding counts, waking up readers, etc. | |
712 | */ | |
713 | size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length) | |
714 | { | |
715 | void *old, *new; | |
716 | size_t old_subbuf, new_subbuf; | |
717 | ||
718 | if (unlikely(length > buf->chan->subbuf_size)) | |
719 | goto toobig; | |
720 | ||
721 | if (buf->offset != buf->chan->subbuf_size + 1) { | |
722 | buf->prev_padding = buf->chan->subbuf_size - buf->offset; | |
723 | old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs; | |
724 | buf->padding[old_subbuf] = buf->prev_padding; | |
725 | buf->subbufs_produced++; | |
20d8b67c | 726 | if (buf->dentry) |
7682c918 | 727 | d_inode(buf->dentry)->i_size += |
20d8b67c EGM |
728 | buf->chan->subbuf_size - |
729 | buf->padding[old_subbuf]; | |
730 | else | |
731 | buf->early_bytes += buf->chan->subbuf_size - | |
732 | buf->padding[old_subbuf]; | |
221415d7 | 733 | smp_mb(); |
26b5679e | 734 | if (waitqueue_active(&buf->read_wait)) { |
7c9cb383 TZ |
735 | /* |
736 | * Calling wake_up_interruptible() from here | |
737 | * will deadlock if we happen to be logging | |
738 | * from the scheduler (trying to re-grab | |
739 | * rq->lock), so defer it. | |
740 | */ | |
26b5679e PZ |
741 | irq_work_queue(&buf->wakeup_work); |
742 | } | |
b86ff981 JA |
743 | } |
744 | ||
745 | old = buf->data; | |
746 | new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs; | |
747 | new = buf->start + new_subbuf * buf->chan->subbuf_size; | |
748 | buf->offset = 0; | |
749 | if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) { | |
750 | buf->offset = buf->chan->subbuf_size + 1; | |
751 | return 0; | |
752 | } | |
753 | buf->data = new; | |
754 | buf->padding[new_subbuf] = 0; | |
755 | ||
756 | if (unlikely(length + buf->offset > buf->chan->subbuf_size)) | |
757 | goto toobig; | |
758 | ||
759 | return length; | |
760 | ||
761 | toobig: | |
762 | buf->chan->last_toobig = length; | |
763 | return 0; | |
764 | } | |
765 | EXPORT_SYMBOL_GPL(relay_switch_subbuf); | |
766 | ||
767 | /** | |
768 | * relay_subbufs_consumed - update the buffer's sub-buffers-consumed count | |
769 | * @chan: the channel | |
770 | * @cpu: the cpu associated with the channel buffer to update | |
771 | * @subbufs_consumed: number of sub-buffers to add to current buf's count | |
772 | * | |
773 | * Adds to the channel buffer's consumed sub-buffer count. | |
774 | * subbufs_consumed should be the number of sub-buffers newly consumed, | |
775 | * not the total consumed. | |
776 | * | |
72fd4a35 | 777 | * NOTE. Kernel clients don't need to call this function if the channel |
b86ff981 JA |
778 | * mode is 'overwrite'. |
779 | */ | |
780 | void relay_subbufs_consumed(struct rchan *chan, | |
781 | unsigned int cpu, | |
782 | size_t subbufs_consumed) | |
783 | { | |
784 | struct rchan_buf *buf; | |
785 | ||
9a29d0fb | 786 | if (!chan || cpu >= NR_CPUS) |
b86ff981 JA |
787 | return; |
788 | ||
017c59c0 | 789 | buf = *per_cpu_ptr(chan->buf, cpu); |
9a29d0fb | 790 | if (!buf || subbufs_consumed > chan->n_subbufs) |
b86ff981 JA |
791 | return; |
792 | ||
2c53d910 | 793 | if (subbufs_consumed > buf->subbufs_produced - buf->subbufs_consumed) |
b86ff981 | 794 | buf->subbufs_consumed = buf->subbufs_produced; |
2c53d910 AS |
795 | else |
796 | buf->subbufs_consumed += subbufs_consumed; | |
b86ff981 JA |
797 | } |
798 | EXPORT_SYMBOL_GPL(relay_subbufs_consumed); | |
799 | ||
800 | /** | |
801 | * relay_close - close the channel | |
802 | * @chan: the channel | |
803 | * | |
804 | * Closes all channel buffers and frees the channel. | |
805 | */ | |
806 | void relay_close(struct rchan *chan) | |
807 | { | |
017c59c0 | 808 | struct rchan_buf *buf; |
b86ff981 | 809 | unsigned int i; |
b86ff981 JA |
810 | |
811 | if (!chan) | |
812 | return; | |
813 | ||
23c88752 | 814 | mutex_lock(&relay_channels_mutex); |
017c59c0 AG |
815 | if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0))) |
816 | relay_close_buf(buf); | |
23c88752 MD |
817 | else |
818 | for_each_possible_cpu(i) | |
017c59c0 AG |
819 | if ((buf = *per_cpu_ptr(chan->buf, i))) |
820 | relay_close_buf(buf); | |
b86ff981 JA |
821 | |
822 | if (chan->last_toobig) | |
823 | printk(KERN_WARNING "relay: one or more items not logged " | |
5b5e0928 | 824 | "[item size (%zd) > sub-buffer size (%zd)]\n", |
b86ff981 JA |
825 | chan->last_toobig, chan->subbuf_size); |
826 | ||
23c88752 | 827 | list_del(&chan->list); |
b86ff981 | 828 | kref_put(&chan->kref, relay_destroy_channel); |
23c88752 | 829 | mutex_unlock(&relay_channels_mutex); |
b86ff981 JA |
830 | } |
831 | EXPORT_SYMBOL_GPL(relay_close); | |
832 | ||
833 | /** | |
834 | * relay_flush - close the channel | |
835 | * @chan: the channel | |
836 | * | |
4c78a663 | 837 | * Flushes all channel buffers, i.e. forces buffer switch. |
b86ff981 JA |
838 | */ |
839 | void relay_flush(struct rchan *chan) | |
840 | { | |
017c59c0 | 841 | struct rchan_buf *buf; |
b86ff981 | 842 | unsigned int i; |
b86ff981 JA |
843 | |
844 | if (!chan) | |
845 | return; | |
846 | ||
017c59c0 AG |
847 | if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0))) { |
848 | relay_switch_subbuf(buf, 0); | |
23c88752 | 849 | return; |
b86ff981 | 850 | } |
23c88752 MD |
851 | |
852 | mutex_lock(&relay_channels_mutex); | |
853 | for_each_possible_cpu(i) | |
017c59c0 AG |
854 | if ((buf = *per_cpu_ptr(chan->buf, i))) |
855 | relay_switch_subbuf(buf, 0); | |
23c88752 | 856 | mutex_unlock(&relay_channels_mutex); |
b86ff981 JA |
857 | } |
858 | EXPORT_SYMBOL_GPL(relay_flush); | |
859 | ||
860 | /** | |
861 | * relay_file_open - open file op for relay files | |
862 | * @inode: the inode | |
863 | * @filp: the file | |
864 | * | |
865 | * Increments the channel buffer refcount. | |
866 | */ | |
867 | static int relay_file_open(struct inode *inode, struct file *filp) | |
868 | { | |
8e18e294 | 869 | struct rchan_buf *buf = inode->i_private; |
b86ff981 JA |
870 | kref_get(&buf->kref); |
871 | filp->private_data = buf; | |
872 | ||
37529fe9 | 873 | return nonseekable_open(inode, filp); |
b86ff981 JA |
874 | } |
875 | ||
876 | /** | |
877 | * relay_file_mmap - mmap file op for relay files | |
878 | * @filp: the file | |
879 | * @vma: the vma describing what to map | |
880 | * | |
72fd4a35 | 881 | * Calls upon relay_mmap_buf() to map the file into user space. |
b86ff981 JA |
882 | */ |
883 | static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma) | |
884 | { | |
885 | struct rchan_buf *buf = filp->private_data; | |
886 | return relay_mmap_buf(buf, vma); | |
887 | } | |
888 | ||
889 | /** | |
890 | * relay_file_poll - poll file op for relay files | |
891 | * @filp: the file | |
892 | * @wait: poll table | |
893 | * | |
894 | * Poll implemention. | |
895 | */ | |
9dd95748 | 896 | static __poll_t relay_file_poll(struct file *filp, poll_table *wait) |
b86ff981 | 897 | { |
9dd95748 | 898 | __poll_t mask = 0; |
b86ff981 JA |
899 | struct rchan_buf *buf = filp->private_data; |
900 | ||
901 | if (buf->finalized) | |
a9a08845 | 902 | return EPOLLERR; |
b86ff981 JA |
903 | |
904 | if (filp->f_mode & FMODE_READ) { | |
905 | poll_wait(filp, &buf->read_wait, wait); | |
906 | if (!relay_buf_empty(buf)) | |
a9a08845 | 907 | mask |= EPOLLIN | EPOLLRDNORM; |
b86ff981 JA |
908 | } |
909 | ||
910 | return mask; | |
911 | } | |
912 | ||
913 | /** | |
914 | * relay_file_release - release file op for relay files | |
915 | * @inode: the inode | |
916 | * @filp: the file | |
917 | * | |
918 | * Decrements the channel refcount, as the filesystem is | |
919 | * no longer using it. | |
920 | */ | |
921 | static int relay_file_release(struct inode *inode, struct file *filp) | |
922 | { | |
923 | struct rchan_buf *buf = filp->private_data; | |
924 | kref_put(&buf->kref, relay_remove_buf); | |
925 | ||
926 | return 0; | |
927 | } | |
928 | ||
4c78a663 | 929 | /* |
b86ff981 JA |
930 | * relay_file_read_consume - update the consumed count for the buffer |
931 | */ | |
932 | static void relay_file_read_consume(struct rchan_buf *buf, | |
933 | size_t read_pos, | |
934 | size_t bytes_consumed) | |
935 | { | |
936 | size_t subbuf_size = buf->chan->subbuf_size; | |
937 | size_t n_subbufs = buf->chan->n_subbufs; | |
938 | size_t read_subbuf; | |
939 | ||
32194450 TZ |
940 | if (buf->subbufs_produced == buf->subbufs_consumed && |
941 | buf->offset == buf->bytes_consumed) | |
942 | return; | |
943 | ||
b86ff981 JA |
944 | if (buf->bytes_consumed + bytes_consumed > subbuf_size) { |
945 | relay_subbufs_consumed(buf->chan, buf->cpu, 1); | |
946 | buf->bytes_consumed = 0; | |
947 | } | |
948 | ||
949 | buf->bytes_consumed += bytes_consumed; | |
a66e356c MH |
950 | if (!read_pos) |
951 | read_subbuf = buf->subbufs_consumed % n_subbufs; | |
952 | else | |
953 | read_subbuf = read_pos / buf->chan->subbuf_size; | |
b86ff981 JA |
954 | if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) { |
955 | if ((read_subbuf == buf->subbufs_produced % n_subbufs) && | |
956 | (buf->offset == subbuf_size)) | |
957 | return; | |
958 | relay_subbufs_consumed(buf->chan, buf->cpu, 1); | |
959 | buf->bytes_consumed = 0; | |
960 | } | |
961 | } | |
962 | ||
4c78a663 | 963 | /* |
b86ff981 JA |
964 | * relay_file_read_avail - boolean, are there unconsumed bytes available? |
965 | */ | |
341a7213 | 966 | static int relay_file_read_avail(struct rchan_buf *buf) |
b86ff981 | 967 | { |
b86ff981 JA |
968 | size_t subbuf_size = buf->chan->subbuf_size; |
969 | size_t n_subbufs = buf->chan->n_subbufs; | |
221415d7 | 970 | size_t produced = buf->subbufs_produced; |
ac05b7a1 | 971 | size_t consumed; |
b86ff981 | 972 | |
341a7213 | 973 | relay_file_read_consume(buf, 0, 0); |
b86ff981 | 974 | |
32194450 TZ |
975 | consumed = buf->subbufs_consumed; |
976 | ||
221415d7 JA |
977 | if (unlikely(buf->offset > subbuf_size)) { |
978 | if (produced == consumed) | |
979 | return 0; | |
980 | return 1; | |
b86ff981 JA |
981 | } |
982 | ||
221415d7 | 983 | if (unlikely(produced - consumed >= n_subbufs)) { |
a66e356c | 984 | consumed = produced - n_subbufs + 1; |
221415d7 | 985 | buf->subbufs_consumed = consumed; |
a66e356c | 986 | buf->bytes_consumed = 0; |
221415d7 | 987 | } |
1bfbc608 | 988 | |
221415d7 JA |
989 | produced = (produced % n_subbufs) * subbuf_size + buf->offset; |
990 | consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed; | |
991 | ||
992 | if (consumed > produced) | |
993 | produced += n_subbufs * subbuf_size; | |
1bfbc608 | 994 | |
32194450 TZ |
995 | if (consumed == produced) { |
996 | if (buf->offset == subbuf_size && | |
997 | buf->subbufs_produced > buf->subbufs_consumed) | |
998 | return 1; | |
b86ff981 | 999 | return 0; |
32194450 | 1000 | } |
b86ff981 | 1001 | |
b86ff981 JA |
1002 | return 1; |
1003 | } | |
1004 | ||
1005 | /** | |
1006 | * relay_file_read_subbuf_avail - return bytes available in sub-buffer | |
4c78a663 RD |
1007 | * @read_pos: file read position |
1008 | * @buf: relay channel buffer | |
b86ff981 JA |
1009 | */ |
1010 | static size_t relay_file_read_subbuf_avail(size_t read_pos, | |
1011 | struct rchan_buf *buf) | |
1012 | { | |
1013 | size_t padding, avail = 0; | |
1014 | size_t read_subbuf, read_offset, write_subbuf, write_offset; | |
1015 | size_t subbuf_size = buf->chan->subbuf_size; | |
1016 | ||
1017 | write_subbuf = (buf->data - buf->start) / subbuf_size; | |
1018 | write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset; | |
1019 | read_subbuf = read_pos / subbuf_size; | |
1020 | read_offset = read_pos % subbuf_size; | |
1021 | padding = buf->padding[read_subbuf]; | |
1022 | ||
1023 | if (read_subbuf == write_subbuf) { | |
1024 | if (read_offset + padding < write_offset) | |
1025 | avail = write_offset - (read_offset + padding); | |
1026 | } else | |
1027 | avail = (subbuf_size - padding) - read_offset; | |
1028 | ||
1029 | return avail; | |
1030 | } | |
1031 | ||
1032 | /** | |
1033 | * relay_file_read_start_pos - find the first available byte to read | |
4c78a663 | 1034 | * @buf: relay channel buffer |
b86ff981 | 1035 | * |
341a7213 | 1036 | * If the read_pos is in the middle of padding, return the |
b86ff981 JA |
1037 | * position of the first actually available byte, otherwise |
1038 | * return the original value. | |
1039 | */ | |
341a7213 | 1040 | static size_t relay_file_read_start_pos(struct rchan_buf *buf) |
b86ff981 JA |
1041 | { |
1042 | size_t read_subbuf, padding, padding_start, padding_end; | |
1043 | size_t subbuf_size = buf->chan->subbuf_size; | |
1044 | size_t n_subbufs = buf->chan->n_subbufs; | |
8d62fdeb | 1045 | size_t consumed = buf->subbufs_consumed % n_subbufs; |
341a7213 | 1046 | size_t read_pos = consumed * subbuf_size + buf->bytes_consumed; |
b86ff981 JA |
1047 | |
1048 | read_subbuf = read_pos / subbuf_size; | |
1049 | padding = buf->padding[read_subbuf]; | |
1050 | padding_start = (read_subbuf + 1) * subbuf_size - padding; | |
1051 | padding_end = (read_subbuf + 1) * subbuf_size; | |
1052 | if (read_pos >= padding_start && read_pos < padding_end) { | |
1053 | read_subbuf = (read_subbuf + 1) % n_subbufs; | |
1054 | read_pos = read_subbuf * subbuf_size; | |
1055 | } | |
1056 | ||
1057 | return read_pos; | |
1058 | } | |
1059 | ||
1060 | /** | |
1061 | * relay_file_read_end_pos - return the new read position | |
4c78a663 RD |
1062 | * @read_pos: file read position |
1063 | * @buf: relay channel buffer | |
1064 | * @count: number of bytes to be read | |
b86ff981 JA |
1065 | */ |
1066 | static size_t relay_file_read_end_pos(struct rchan_buf *buf, | |
1067 | size_t read_pos, | |
1068 | size_t count) | |
1069 | { | |
1070 | size_t read_subbuf, padding, end_pos; | |
1071 | size_t subbuf_size = buf->chan->subbuf_size; | |
1072 | size_t n_subbufs = buf->chan->n_subbufs; | |
1073 | ||
1074 | read_subbuf = read_pos / subbuf_size; | |
1075 | padding = buf->padding[read_subbuf]; | |
1076 | if (read_pos % subbuf_size + count + padding == subbuf_size) | |
1077 | end_pos = (read_subbuf + 1) * subbuf_size; | |
1078 | else | |
1079 | end_pos = read_pos + count; | |
1080 | if (end_pos >= subbuf_size * n_subbufs) | |
1081 | end_pos = 0; | |
1082 | ||
1083 | return end_pos; | |
1084 | } | |
1085 | ||
a7c22421 AV |
1086 | static ssize_t relay_file_read(struct file *filp, |
1087 | char __user *buffer, | |
1088 | size_t count, | |
1089 | loff_t *ppos) | |
221415d7 | 1090 | { |
6dac40a7 TZ |
1091 | struct rchan_buf *buf = filp->private_data; |
1092 | size_t read_start, avail; | |
a7c22421 | 1093 | size_t written = 0; |
6dac40a7 | 1094 | int ret; |
221415d7 | 1095 | |
a7c22421 | 1096 | if (!count) |
221415d7 JA |
1097 | return 0; |
1098 | ||
5955102c | 1099 | inode_lock(file_inode(filp)); |
221415d7 | 1100 | do { |
a7c22421 AV |
1101 | void *from; |
1102 | ||
341a7213 | 1103 | if (!relay_file_read_avail(buf)) |
6dac40a7 TZ |
1104 | break; |
1105 | ||
341a7213 | 1106 | read_start = relay_file_read_start_pos(buf); |
6dac40a7 TZ |
1107 | avail = relay_file_read_subbuf_avail(read_start, buf); |
1108 | if (!avail) | |
221415d7 | 1109 | break; |
221415d7 | 1110 | |
a7c22421 AV |
1111 | avail = min(count, avail); |
1112 | from = buf->start + read_start; | |
1113 | ret = avail; | |
1114 | if (copy_to_user(buffer, from, avail)) | |
6dac40a7 TZ |
1115 | break; |
1116 | ||
a7c22421 AV |
1117 | buffer += ret; |
1118 | written += ret; | |
1119 | count -= ret; | |
6dac40a7 | 1120 | |
a7c22421 AV |
1121 | relay_file_read_consume(buf, read_start, ret); |
1122 | *ppos = relay_file_read_end_pos(buf, read_start, ret); | |
1123 | } while (count); | |
1124 | inode_unlock(file_inode(filp)); | |
6dac40a7 | 1125 | |
a7c22421 | 1126 | return written; |
6dac40a7 TZ |
1127 | } |
1128 | ||
1db60cf2 JA |
1129 | static void relay_consume_bytes(struct rchan_buf *rbuf, int bytes_consumed) |
1130 | { | |
1131 | rbuf->bytes_consumed += bytes_consumed; | |
1132 | ||
1133 | if (rbuf->bytes_consumed >= rbuf->chan->subbuf_size) { | |
1134 | relay_subbufs_consumed(rbuf->chan, rbuf->cpu, 1); | |
1135 | rbuf->bytes_consumed %= rbuf->chan->subbuf_size; | |
1136 | } | |
1137 | } | |
1138 | ||
ebf99093 TZ |
1139 | static void relay_pipe_buf_release(struct pipe_inode_info *pipe, |
1140 | struct pipe_buffer *buf) | |
6dac40a7 | 1141 | { |
ebf99093 TZ |
1142 | struct rchan_buf *rbuf; |
1143 | ||
1144 | rbuf = (struct rchan_buf *)page_private(buf->page); | |
1db60cf2 | 1145 | relay_consume_bytes(rbuf, buf->private); |
ebf99093 TZ |
1146 | } |
1147 | ||
28dfef8f | 1148 | static const struct pipe_buf_operations relay_pipe_buf_ops = { |
c928f642 CH |
1149 | .release = relay_pipe_buf_release, |
1150 | .try_steal = generic_pipe_buf_try_steal, | |
1151 | .get = generic_pipe_buf_get, | |
ebf99093 TZ |
1152 | }; |
1153 | ||
5eb7f9fa JA |
1154 | static void relay_page_release(struct splice_pipe_desc *spd, unsigned int i) |
1155 | { | |
1156 | } | |
1157 | ||
d3f35d98 | 1158 | /* |
ebf99093 TZ |
1159 | * subbuf_splice_actor - splice up to one subbuf's worth of data |
1160 | */ | |
5f1664f9 | 1161 | static ssize_t subbuf_splice_actor(struct file *in, |
ebf99093 TZ |
1162 | loff_t *ppos, |
1163 | struct pipe_inode_info *pipe, | |
1164 | size_t len, | |
1165 | unsigned int flags, | |
1166 | int *nonpad_ret) | |
1167 | { | |
5f1664f9 | 1168 | unsigned int pidx, poff, total_len, subbuf_pages, nr_pages; |
ebf99093 TZ |
1169 | struct rchan_buf *rbuf = in->private_data; |
1170 | unsigned int subbuf_size = rbuf->chan->subbuf_size; | |
24da24de TZ |
1171 | uint64_t pos = (uint64_t) *ppos; |
1172 | uint32_t alloc_size = (uint32_t) rbuf->chan->alloc_size; | |
1173 | size_t read_start = (size_t) do_div(pos, alloc_size); | |
ebf99093 TZ |
1174 | size_t read_subbuf = read_start / subbuf_size; |
1175 | size_t padding = rbuf->padding[read_subbuf]; | |
1176 | size_t nonpad_end = read_subbuf * subbuf_size + subbuf_size - padding; | |
35f3d14d JA |
1177 | struct page *pages[PIPE_DEF_BUFFERS]; |
1178 | struct partial_page partial[PIPE_DEF_BUFFERS]; | |
1db60cf2 JA |
1179 | struct splice_pipe_desc spd = { |
1180 | .pages = pages, | |
1181 | .nr_pages = 0, | |
047fe360 | 1182 | .nr_pages_max = PIPE_DEF_BUFFERS, |
1db60cf2 | 1183 | .partial = partial, |
1db60cf2 | 1184 | .ops = &relay_pipe_buf_ops, |
5eb7f9fa | 1185 | .spd_release = relay_page_release, |
1db60cf2 | 1186 | }; |
5f1664f9 | 1187 | ssize_t ret; |
ebf99093 TZ |
1188 | |
1189 | if (rbuf->subbufs_produced == rbuf->subbufs_consumed) | |
1190 | return 0; | |
35f3d14d JA |
1191 | if (splice_grow_spd(pipe, &spd)) |
1192 | return -ENOMEM; | |
ebf99093 | 1193 | |
1db60cf2 JA |
1194 | /* |
1195 | * Adjust read len, if longer than what is available | |
1196 | */ | |
1197 | if (len > (subbuf_size - read_start % subbuf_size)) | |
1198 | len = subbuf_size - read_start % subbuf_size; | |
ebf99093 TZ |
1199 | |
1200 | subbuf_pages = rbuf->chan->alloc_size >> PAGE_SHIFT; | |
1201 | pidx = (read_start / PAGE_SIZE) % subbuf_pages; | |
1202 | poff = read_start & ~PAGE_MASK; | |
a786c06d | 1203 | nr_pages = min_t(unsigned int, subbuf_pages, spd.nr_pages_max); |
ebf99093 | 1204 | |
16d54669 | 1205 | for (total_len = 0; spd.nr_pages < nr_pages; spd.nr_pages++) { |
1db60cf2 JA |
1206 | unsigned int this_len, this_end, private; |
1207 | unsigned int cur_pos = read_start + total_len; | |
ebf99093 | 1208 | |
1db60cf2 | 1209 | if (!len) |
ebf99093 | 1210 | break; |
ebf99093 | 1211 | |
1db60cf2 JA |
1212 | this_len = min_t(unsigned long, len, PAGE_SIZE - poff); |
1213 | private = this_len; | |
ebf99093 | 1214 | |
1db60cf2 JA |
1215 | spd.pages[spd.nr_pages] = rbuf->page_array[pidx]; |
1216 | spd.partial[spd.nr_pages].offset = poff; | |
ebf99093 | 1217 | |
1db60cf2 JA |
1218 | this_end = cur_pos + this_len; |
1219 | if (this_end >= nonpad_end) { | |
1220 | this_len = nonpad_end - cur_pos; | |
1221 | private = this_len + padding; | |
ebf99093 | 1222 | } |
1db60cf2 JA |
1223 | spd.partial[spd.nr_pages].len = this_len; |
1224 | spd.partial[spd.nr_pages].private = private; | |
ebf99093 | 1225 | |
1db60cf2 JA |
1226 | len -= this_len; |
1227 | total_len += this_len; | |
1228 | poff = 0; | |
1229 | pidx = (pidx + 1) % subbuf_pages; | |
ebf99093 | 1230 | |
1db60cf2 JA |
1231 | if (this_end >= nonpad_end) { |
1232 | spd.nr_pages++; | |
ebf99093 TZ |
1233 | break; |
1234 | } | |
ebf99093 TZ |
1235 | } |
1236 | ||
35f3d14d | 1237 | ret = 0; |
1db60cf2 | 1238 | if (!spd.nr_pages) |
35f3d14d | 1239 | goto out; |
ebf99093 | 1240 | |
1db60cf2 JA |
1241 | ret = *nonpad_ret = splice_to_pipe(pipe, &spd); |
1242 | if (ret < 0 || ret < total_len) | |
35f3d14d | 1243 | goto out; |
ebf99093 | 1244 | |
1db60cf2 JA |
1245 | if (read_start + ret == nonpad_end) |
1246 | ret += padding; | |
1247 | ||
35f3d14d | 1248 | out: |
047fe360 ED |
1249 | splice_shrink_spd(&spd); |
1250 | return ret; | |
ebf99093 TZ |
1251 | } |
1252 | ||
1253 | static ssize_t relay_file_splice_read(struct file *in, | |
1254 | loff_t *ppos, | |
1255 | struct pipe_inode_info *pipe, | |
1256 | size_t len, | |
1257 | unsigned int flags) | |
1258 | { | |
1259 | ssize_t spliced; | |
1260 | int ret; | |
1261 | int nonpad_ret = 0; | |
1262 | ||
1263 | ret = 0; | |
1264 | spliced = 0; | |
1265 | ||
a82c53a0 | 1266 | while (len && !spliced) { |
ebf99093 TZ |
1267 | ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret); |
1268 | if (ret < 0) | |
1269 | break; | |
1270 | else if (!ret) { | |
fbb5b7ae | 1271 | if (flags & SPLICE_F_NONBLOCK) |
ebf99093 | 1272 | ret = -EAGAIN; |
fbb5b7ae | 1273 | break; |
ebf99093 TZ |
1274 | } |
1275 | ||
1276 | *ppos += ret; | |
1277 | if (ret > len) | |
1278 | len = 0; | |
1279 | else | |
1280 | len -= ret; | |
1281 | spliced += nonpad_ret; | |
1282 | nonpad_ret = 0; | |
1283 | } | |
1284 | ||
1285 | if (spliced) | |
1286 | return spliced; | |
1287 | ||
1288 | return ret; | |
221415d7 JA |
1289 | } |
1290 | ||
15ad7cdc | 1291 | const struct file_operations relay_file_operations = { |
b86ff981 JA |
1292 | .open = relay_file_open, |
1293 | .poll = relay_file_poll, | |
1294 | .mmap = relay_file_mmap, | |
1295 | .read = relay_file_read, | |
1296 | .llseek = no_llseek, | |
1297 | .release = relay_file_release, | |
ebf99093 | 1298 | .splice_read = relay_file_splice_read, |
b86ff981 JA |
1299 | }; |
1300 | EXPORT_SYMBOL_GPL(relay_file_operations); |