blob: d797502140b9a0f1ceb36f5ccbb4b9c66382a44e [file] [log] [blame]
Jens Axboeb86ff9812006-03-23 19:56:55 +01001/*
2 * Public API and common code for kernel->userspace relay file support.
3 *
Jesper Juhlc9b3feb2007-07-31 00:39:18 -07004 * See Documentation/filesystems/relay.txt for an overview.
Jens Axboeb86ff9812006-03-23 19:56:55 +01005 *
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.
Mathieu Desnoyers23c88752007-02-10 01:45:05 -080010 * November 2006 - CPU hotplug support by Mathieu Desnoyers
11 * (mathieu.desnoyers@polymtl.ca)
Jens Axboeb86ff9812006-03-23 19:56:55 +010012 *
13 * This file is released under the GPL.
14 */
15#include <linux/errno.h>
16#include <linux/stddef.h>
17#include <linux/slab.h>
Paul Gortmaker9984de12011-05-23 14:51:41 -040018#include <linux/export.h>
Jens Axboeb86ff9812006-03-23 19:56:55 +010019#include <linux/string.h>
20#include <linux/relay.h>
21#include <linux/vmalloc.h>
22#include <linux/mm.h>
Mathieu Desnoyers23c88752007-02-10 01:45:05 -080023#include <linux/cpu.h>
Jens Axboed6b29d72007-06-04 09:59:47 +020024#include <linux/splice.h>
Mathieu Desnoyers23c88752007-02-10 01:45:05 -080025
26/* list of open channels, for cpu hotplug */
27static DEFINE_MUTEX(relay_channels_mutex);
28static LIST_HEAD(relay_channels);
Jens Axboeb86ff9812006-03-23 19:56:55 +010029
30/*
31 * close() vm_op implementation for relay file mapping.
32 */
33static 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/*
Nick Piggina1e09612008-02-06 01:37:34 -080040 * fault() vm_op implementation for relay file mapping.
Jens Axboeb86ff9812006-03-23 19:56:55 +010041 */
Nick Piggina1e09612008-02-06 01:37:34 -080042static int relay_buf_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Jens Axboeb86ff9812006-03-23 19:56:55 +010043{
44 struct page *page;
45 struct rchan_buf *buf = vma->vm_private_data;
Nick Piggina1e09612008-02-06 01:37:34 -080046 pgoff_t pgoff = vmf->pgoff;
Jens Axboeb86ff9812006-03-23 19:56:55 +010047
Jens Axboeb86ff9812006-03-23 19:56:55 +010048 if (!buf)
Nick Piggina1e09612008-02-06 01:37:34 -080049 return VM_FAULT_OOM;
Jens Axboeb86ff9812006-03-23 19:56:55 +010050
Nick Piggina1e09612008-02-06 01:37:34 -080051 page = vmalloc_to_page(buf->start + (pgoff << PAGE_SHIFT));
Jens Axboeb86ff9812006-03-23 19:56:55 +010052 if (!page)
Nick Piggina1e09612008-02-06 01:37:34 -080053 return VM_FAULT_SIGBUS;
Jens Axboeb86ff9812006-03-23 19:56:55 +010054 get_page(page);
Nick Piggina1e09612008-02-06 01:37:34 -080055 vmf->page = page;
Jens Axboeb86ff9812006-03-23 19:56:55 +010056
Nick Piggina1e09612008-02-06 01:37:34 -080057 return 0;
Jens Axboeb86ff9812006-03-23 19:56:55 +010058}
59
60/*
61 * vm_ops for relay file mappings.
62 */
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +040063static const struct vm_operations_struct relay_file_mmap_ops = {
Nick Piggina1e09612008-02-06 01:37:34 -080064 .fault = relay_buf_fault,
Jens Axboeb86ff9812006-03-23 19:56:55 +010065 .close = relay_file_mmap_close,
66};
67
Masami Hiramatsu68ab3d82008-04-29 01:03:46 -070068/*
69 * allocate an array of pointers of struct page
70 */
71static struct page **relay_alloc_page_array(unsigned int n_pages)
72{
Jesper Juhl408af872010-11-04 21:44:41 +010073 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);
Masami Hiramatsu68ab3d82008-04-29 01:03:46 -070077}
78
79/*
80 * free an array of pointers of struct page
81 */
82static void relay_free_page_array(struct page **array)
83{
Pekka Enberg200f1ce2015-06-30 14:59:06 -070084 kvfree(array);
Masami Hiramatsu68ab3d82008-04-29 01:03:46 -070085}
86
Jens Axboeb86ff9812006-03-23 19:56:55 +010087/**
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 */
Adrian Bunk01c55ed2007-07-19 01:48:32 -070096static int relay_mmap_buf(struct rchan_buf *buf, struct vm_area_struct *vma)
Jens Axboeb86ff9812006-03-23 19:56:55 +010097{
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;
Nick Piggin2f987352008-02-02 03:08:53 +0100108 vma->vm_flags |= VM_DONTEXPAND;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100109 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 *
Randy Dunlap4c78a662006-09-29 01:59:10 -0700120 * Returns a pointer to the resulting buffer, %NULL if unsuccessful. The
Jens Axboe221415d2006-03-23 19:57:55 +0100121 * passed in size will get page aligned, if it isn't already.
Jens Axboeb86ff9812006-03-23 19:56:55 +0100122 */
Jens Axboe221415d2006-03-23 19:57:55 +0100123static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100124{
125 void *mem;
126 unsigned int i, j, n_pages;
127
Jens Axboe221415d2006-03-23 19:57:55 +0100128 *size = PAGE_ALIGN(*size);
129 n_pages = *size >> PAGE_SHIFT;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100130
Masami Hiramatsu68ab3d82008-04-29 01:03:46 -0700131 buf->page_array = relay_alloc_page_array(n_pages);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100132 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;
Tom Zanussiebf99092007-06-04 09:12:05 +0200139 set_page_private(buf->page_array[i], (unsigned long)buf);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100140 }
141 mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL);
142 if (!mem)
143 goto depopulate;
144
Jens Axboe221415d2006-03-23 19:57:55 +0100145 memset(mem, 0, *size);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100146 buf->page_count = n_pages;
147 return mem;
148
149depopulate:
150 for (j = 0; j < i; j++)
151 __free_page(buf->page_array[j]);
Masami Hiramatsu68ab3d82008-04-29 01:03:46 -0700152 relay_free_page_array(buf->page_array);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100153 return NULL;
154}
155
156/**
157 * relay_create_buf - allocate and initialize a channel buffer
Randy Dunlap4c78a662006-09-29 01:59:10 -0700158 * @chan: the relay channel
Jens Axboeb86ff9812006-03-23 19:56:55 +0100159 *
Randy Dunlap4c78a662006-09-29 01:59:10 -0700160 * Returns channel buffer if successful, %NULL otherwise.
Jens Axboeb86ff9812006-03-23 19:56:55 +0100161 */
Adrian Bunk01c55ed2007-07-19 01:48:32 -0700162static struct rchan_buf *relay_create_buf(struct rchan *chan)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100163{
Dan Carpenterf6302f12012-02-10 09:03:58 +0100164 struct rchan_buf *buf;
165
166 if (chan->n_subbufs > UINT_MAX / sizeof(size_t *))
Jens Axboeb86ff9812006-03-23 19:56:55 +0100167 return NULL;
168
Dan Carpenterf6302f12012-02-10 09:03:58 +0100169 buf = kzalloc(sizeof(struct rchan_buf), GFP_KERNEL);
170 if (!buf)
171 return NULL;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100172 buf->padding = kmalloc(chan->n_subbufs * sizeof(size_t *), GFP_KERNEL);
173 if (!buf->padding)
174 goto free_buf;
175
Jens Axboe221415d2006-03-23 19:57:55 +0100176 buf->start = relay_alloc_buf(buf, &chan->alloc_size);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100177 if (!buf->start)
178 goto free_buf;
179
180 buf->chan = chan;
181 kref_get(&buf->chan->kref);
182 return buf;
183
184free_buf:
185 kfree(buf->padding);
186 kfree(buf);
187 return NULL;
188}
189
190/**
191 * relay_destroy_channel - free the channel struct
Randy Dunlap4c78a662006-09-29 01:59:10 -0700192 * @kref: target kernel reference that contains the relay channel
Jens Axboeb86ff9812006-03-23 19:56:55 +0100193 *
194 * Should only be called from kref_put().
195 */
Adrian Bunk01c55ed2007-07-19 01:48:32 -0700196static void relay_destroy_channel(struct kref *kref)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100197{
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 */
Adrian Bunk01c55ed2007-07-19 01:48:32 -0700206static void relay_destroy_buf(struct rchan_buf *buf)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100207{
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]);
Masami Hiramatsu68ab3d82008-04-29 01:03:46 -0700215 relay_free_page_array(buf->page_array);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100216 }
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800217 chan->buf[buf->cpu] = NULL;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100218 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
Randy Dunlap4c78a662006-09-29 01:59:10 -0700225 * @kref: target kernel reference that contains the relay buffer
Jens Axboeb86ff9812006-03-23 19:56:55 +0100226 *
Masanari Iidae2278672014-02-18 22:54:36 +0900227 * Removes the file from the filesystem, which also frees the
Jens Axboeb86ff9812006-03-23 19:56:55 +0100228 * rchan_buf_struct and the channel buffer. Should only be called from
229 * kref_put().
230 */
Adrian Bunk01c55ed2007-07-19 01:48:32 -0700231static void relay_remove_buf(struct kref *kref)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100232{
233 struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100234 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 */
Adrian Bunk01c55ed2007-07-19 01:48:32 -0700243static int relay_buf_empty(struct rchan_buf *buf)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100244{
245 return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1;
246}
Jens Axboeb86ff9812006-03-23 19:56:55 +0100247
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 */
254int 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}
259EXPORT_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 */
273static 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 */
287static 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 */
295static 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 */
303static struct dentry *create_buf_file_default_callback(const char *filename,
304 struct dentry *parent,
Al Virof4ae40a2011-07-24 04:33:43 -0400305 umode_t mode,
Jens Axboeb86ff9812006-03-23 19:56:55 +0100306 struct rchan_buf *buf,
307 int *is_global)
308{
309 return NULL;
310}
311
312/*
313 * remove_buf_file() default callback. Does nothing.
314 */
315static int remove_buf_file_default_callback(struct dentry *dentry)
316{
317 return -EINVAL;
318}
319
320/* relay channel default callbacks */
321static 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
Linus Torvalds9a9136e22007-05-09 12:54:17 -0700331 * @data: contains the channel buffer
Jens Axboeb86ff9812006-03-23 19:56:55 +0100332 *
Tom Zanussi7c9cb382007-05-09 02:34:01 -0700333 * This is the timer function used to defer reader waking.
Jens Axboeb86ff9812006-03-23 19:56:55 +0100334 */
Tom Zanussi7c9cb382007-05-09 02:34:01 -0700335static void wakeup_readers(unsigned long data)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100336{
Tom Zanussi7c9cb382007-05-09 02:34:01 -0700337 struct rchan_buf *buf = (struct rchan_buf *)data;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100338 wake_up_interruptible(&buf->read_wait);
339}
340
341/**
342 * __relay_reset - reset a channel buffer
343 * @buf: the channel buffer
344 * @init: 1 if this is a first-time initialization
345 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800346 * See relay_reset() for description of effect.
Jens Axboeb86ff9812006-03-23 19:56:55 +0100347 */
Andrew Morton192636a2006-12-22 01:11:30 -0800348static void __relay_reset(struct rchan_buf *buf, unsigned int init)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100349{
350 size_t i;
351
352 if (init) {
353 init_waitqueue_head(&buf->read_wait);
354 kref_init(&buf->kref);
Tom Zanussi7c9cb382007-05-09 02:34:01 -0700355 setup_timer(&buf->timer, wakeup_readers, (unsigned long)buf);
356 } else
357 del_timer_sync(&buf->timer);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100358
359 buf->subbufs_produced = 0;
360 buf->subbufs_consumed = 0;
361 buf->bytes_consumed = 0;
362 buf->finalized = 0;
363 buf->data = buf->start;
364 buf->offset = 0;
365
366 for (i = 0; i < buf->chan->n_subbufs; i++)
367 buf->padding[i] = 0;
368
369 buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0);
370}
371
372/**
373 * relay_reset - reset the channel
374 * @chan: the channel
375 *
376 * This has the effect of erasing all data from all channel buffers
377 * and restarting the channel in its initial state. The buffers
378 * are not freed, so any mappings are still in effect.
379 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800380 * NOTE. Care should be taken that the channel isn't actually
Jens Axboeb86ff9812006-03-23 19:56:55 +0100381 * being used by anything when this call is made.
382 */
383void relay_reset(struct rchan *chan)
384{
385 unsigned int i;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100386
387 if (!chan)
388 return;
389
Daniel Walker1bfbc602007-10-18 03:06:05 -0700390 if (chan->is_global && chan->buf[0]) {
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800391 __relay_reset(chan->buf[0], 0);
392 return;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100393 }
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800394
395 mutex_lock(&relay_channels_mutex);
Lai Jiangshan98ba4032008-11-14 10:44:59 +0100396 for_each_possible_cpu(i)
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800397 if (chan->buf[i])
398 __relay_reset(chan->buf[i], 0);
399 mutex_unlock(&relay_channels_mutex);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100400}
401EXPORT_SYMBOL_GPL(relay_reset);
402
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700403static inline void relay_set_buf_dentry(struct rchan_buf *buf,
404 struct dentry *dentry)
405{
406 buf->dentry = dentry;
David Howells7682c912015-03-17 22:26:16 +0000407 d_inode(buf->dentry)->i_size = buf->early_bytes;
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700408}
409
410static struct dentry *relay_create_buf_file(struct rchan *chan,
411 struct rchan_buf *buf,
412 unsigned int cpu)
413{
414 struct dentry *dentry;
415 char *tmpname;
416
417 tmpname = kzalloc(NAME_MAX + 1, GFP_KERNEL);
418 if (!tmpname)
419 return NULL;
420 snprintf(tmpname, NAME_MAX, "%s%d", chan->base_filename, cpu);
421
422 /* Create file in fs */
423 dentry = chan->cb->create_buf_file(tmpname, chan->parent,
424 S_IRUSR, buf,
425 &chan->is_global);
426
427 kfree(tmpname);
428
429 return dentry;
430}
431
Randy Dunlap4c78a662006-09-29 01:59:10 -0700432/*
Jens Axboeb86ff9812006-03-23 19:56:55 +0100433 * relay_open_buf - create a new relay channel buffer
434 *
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800435 * used by relay_open() and CPU hotplug.
Jens Axboeb86ff9812006-03-23 19:56:55 +0100436 */
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800437static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100438{
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800439 struct rchan_buf *buf = NULL;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100440 struct dentry *dentry;
441
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800442 if (chan->is_global)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100443 return chan->buf[0];
444
445 buf = relay_create_buf(chan);
446 if (!buf)
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700447 return NULL;
448
449 if (chan->has_base_filename) {
450 dentry = relay_create_buf_file(chan, buf, cpu);
451 if (!dentry)
452 goto free_buf;
453 relay_set_buf_dentry(buf, dentry);
Akash Goel59dbb2a2016-08-02 14:07:18 -0700454 } else {
455 /* Only retrieve global info, nothing more, nothing less */
456 dentry = chan->cb->create_buf_file(NULL, NULL,
457 S_IRUSR, buf,
458 &chan->is_global);
459 if (WARN_ON(dentry))
460 goto free_buf;
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700461 }
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800462
463 buf->cpu = cpu;
464 __relay_reset(buf, 1);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100465
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800466 if(chan->is_global) {
467 chan->buf[0] = buf;
468 buf->cpu = 0;
469 }
470
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700471 return buf;
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800472
473free_buf:
474 relay_destroy_buf(buf);
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700475 return NULL;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100476}
477
478/**
479 * relay_close_buf - close a channel buffer
480 * @buf: channel buffer
481 *
482 * Marks the buffer finalized and restores the default callbacks.
483 * The channel buffer and channel buffer data structure are then freed
484 * automatically when the last reference is given up.
485 */
Andrew Morton192636a2006-12-22 01:11:30 -0800486static void relay_close_buf(struct rchan_buf *buf)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100487{
488 buf->finalized = 1;
Tom Zanussi7c9cb382007-05-09 02:34:01 -0700489 del_timer_sync(&buf->timer);
Dmitry Monakhovb8d4a5b2013-04-22 11:41:41 +0400490 buf->chan->cb->remove_buf_file(buf->dentry);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100491 kref_put(&buf->kref, relay_remove_buf);
492}
493
Andrew Morton192636a2006-12-22 01:11:30 -0800494static void setup_callbacks(struct rchan *chan,
Jens Axboeb86ff9812006-03-23 19:56:55 +0100495 struct rchan_callbacks *cb)
496{
497 if (!cb) {
498 chan->cb = &default_channel_callbacks;
499 return;
500 }
501
502 if (!cb->subbuf_start)
503 cb->subbuf_start = subbuf_start_default_callback;
504 if (!cb->buf_mapped)
505 cb->buf_mapped = buf_mapped_default_callback;
506 if (!cb->buf_unmapped)
507 cb->buf_unmapped = buf_unmapped_default_callback;
508 if (!cb->create_buf_file)
509 cb->create_buf_file = create_buf_file_default_callback;
510 if (!cb->remove_buf_file)
511 cb->remove_buf_file = remove_buf_file_default_callback;
512 chan->cb = cb;
513}
514
515/**
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800516 * relay_hotcpu_callback - CPU hotplug callback
517 * @nb: notifier block
518 * @action: hotplug action to take
519 * @hcpu: CPU number
520 *
Randy Dunlap05fb6bf2007-02-28 20:12:13 -0800521 * Returns the success/failure of the operation. (%NOTIFY_OK, %NOTIFY_BAD)
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800522 */
Paul Gortmaker0db06282013-06-19 14:53:51 -0400523static int relay_hotcpu_callback(struct notifier_block *nb,
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800524 unsigned long action,
525 void *hcpu)
526{
527 unsigned int hotcpu = (unsigned long)hcpu;
528 struct rchan *chan;
529
530 switch(action) {
531 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -0700532 case CPU_UP_PREPARE_FROZEN:
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800533 mutex_lock(&relay_channels_mutex);
534 list_for_each_entry(chan, &relay_channels, list) {
535 if (chan->buf[hotcpu])
536 continue;
537 chan->buf[hotcpu] = relay_open_buf(chan, hotcpu);
538 if(!chan->buf[hotcpu]) {
539 printk(KERN_ERR
540 "relay_hotcpu_callback: cpu %d buffer "
541 "creation failed\n", hotcpu);
542 mutex_unlock(&relay_channels_mutex);
Akinobu Mita80b51842010-05-26 14:43:32 -0700543 return notifier_from_errno(-ENOMEM);
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800544 }
545 }
546 mutex_unlock(&relay_channels_mutex);
547 break;
548 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -0700549 case CPU_DEAD_FROZEN:
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800550 /* No need to flush the cpu : will be flushed upon
551 * final relay_flush() call. */
552 break;
553 }
554 return NOTIFY_OK;
555}
556
557/**
Jens Axboeb86ff9812006-03-23 19:56:55 +0100558 * relay_open - create a new relay channel
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700559 * @base_filename: base name of files to create, %NULL for buffering only
560 * @parent: dentry of parent directory, %NULL for root directory or buffer
Jens Axboeb86ff9812006-03-23 19:56:55 +0100561 * @subbuf_size: size of sub-buffers
562 * @n_subbufs: number of sub-buffers
563 * @cb: client callback functions
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800564 * @private_data: user-defined data
Jens Axboeb86ff9812006-03-23 19:56:55 +0100565 *
Randy Dunlap4c78a662006-09-29 01:59:10 -0700566 * Returns channel pointer if successful, %NULL otherwise.
Jens Axboeb86ff9812006-03-23 19:56:55 +0100567 *
568 * Creates a channel buffer for each cpu using the sizes and
569 * attributes specified. The created channel buffer files
570 * will be named base_filename0...base_filenameN-1. File
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800571 * permissions will be %S_IRUSR.
Akash Goel59dbb2a2016-08-02 14:07:18 -0700572 *
573 * If opening a buffer (@parent = NULL) that you later wish to register
574 * in a filesystem, call relay_late_setup_files() once the @parent dentry
575 * is available.
Jens Axboeb86ff9812006-03-23 19:56:55 +0100576 */
577struct rchan *relay_open(const char *base_filename,
578 struct dentry *parent,
579 size_t subbuf_size,
580 size_t n_subbufs,
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800581 struct rchan_callbacks *cb,
582 void *private_data)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100583{
584 unsigned int i;
585 struct rchan *chan;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100586
587 if (!(subbuf_size && n_subbufs))
588 return NULL;
Dan Carpenterf6302f12012-02-10 09:03:58 +0100589 if (subbuf_size > UINT_MAX / n_subbufs)
590 return NULL;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100591
Robert P. J. Daycd861282006-12-13 00:34:52 -0800592 chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100593 if (!chan)
594 return NULL;
595
596 chan->version = RELAYFS_CHANNEL_VERSION;
597 chan->n_subbufs = n_subbufs;
598 chan->subbuf_size = subbuf_size;
zhangwei(Jovi)a05342cb2013-04-30 15:28:41 -0700599 chan->alloc_size = PAGE_ALIGN(subbuf_size * n_subbufs);
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800600 chan->parent = parent;
601 chan->private_data = private_data;
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700602 if (base_filename) {
603 chan->has_base_filename = 1;
604 strlcpy(chan->base_filename, base_filename, NAME_MAX);
605 }
Jens Axboeb86ff9812006-03-23 19:56:55 +0100606 setup_callbacks(chan, cb);
607 kref_init(&chan->kref);
608
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800609 mutex_lock(&relay_channels_mutex);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100610 for_each_online_cpu(i) {
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800611 chan->buf[i] = relay_open_buf(chan, i);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100612 if (!chan->buf[i])
613 goto free_bufs;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100614 }
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800615 list_add(&chan->list, &relay_channels);
616 mutex_unlock(&relay_channels_mutex);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100617
Jens Axboeb86ff9812006-03-23 19:56:55 +0100618 return chan;
619
620free_bufs:
Lai Jiangshan98ba4032008-11-14 10:44:59 +0100621 for_each_possible_cpu(i) {
622 if (chan->buf[i])
623 relay_close_buf(chan->buf[i]);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100624 }
Jens Axboeb86ff9812006-03-23 19:56:55 +0100625
Jens Axboeb86ff9812006-03-23 19:56:55 +0100626 kref_put(&chan->kref, relay_destroy_channel);
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800627 mutex_unlock(&relay_channels_mutex);
Zhouyi Zhouba62baf2016-06-08 15:33:53 -0700628 kfree(chan);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100629 return NULL;
630}
631EXPORT_SYMBOL_GPL(relay_open);
632
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700633struct rchan_percpu_buf_dispatcher {
634 struct rchan_buf *buf;
635 struct dentry *dentry;
636};
637
638/* Called in atomic context. */
639static void __relay_set_buf_dentry(void *info)
640{
641 struct rchan_percpu_buf_dispatcher *p = info;
642
643 relay_set_buf_dentry(p->buf, p->dentry);
644}
645
646/**
647 * relay_late_setup_files - triggers file creation
648 * @chan: channel to operate on
649 * @base_filename: base name of files to create
650 * @parent: dentry of parent directory, %NULL for root directory
651 *
652 * Returns 0 if successful, non-zero otherwise.
653 *
Akash Goel59dbb2a2016-08-02 14:07:18 -0700654 * Use to setup files for a previously buffer-only channel created
655 * by relay_open() with a NULL parent dentry.
656 *
657 * For example, this is useful for perfomring early tracing in kernel,
658 * before VFS is up and then exposing the early results once the dentry
659 * is available.
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700660 */
661int relay_late_setup_files(struct rchan *chan,
662 const char *base_filename,
663 struct dentry *parent)
664{
665 int err = 0;
666 unsigned int i, curr_cpu;
667 unsigned long flags;
668 struct dentry *dentry;
669 struct rchan_percpu_buf_dispatcher disp;
670
671 if (!chan || !base_filename)
672 return -EINVAL;
673
674 strlcpy(chan->base_filename, base_filename, NAME_MAX);
675
676 mutex_lock(&relay_channels_mutex);
677 /* Is chan already set up? */
Jiri Slabyb786c6a2009-01-17 12:04:36 +0100678 if (unlikely(chan->has_base_filename)) {
679 mutex_unlock(&relay_channels_mutex);
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700680 return -EEXIST;
Jiri Slabyb786c6a2009-01-17 12:04:36 +0100681 }
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700682 chan->has_base_filename = 1;
683 chan->parent = parent;
Akash Goel59dbb2a2016-08-02 14:07:18 -0700684
685 if (chan->is_global) {
686 err = -EINVAL;
687 if (!WARN_ON_ONCE(!chan->buf[0])) {
688 dentry = relay_create_buf_file(chan, chan->buf[0], 0);
689 if (dentry && !WARN_ON_ONCE(!chan->is_global)) {
690 relay_set_buf_dentry(chan->buf[0], dentry);
691 err = 0;
692 }
693 }
694 mutex_unlock(&relay_channels_mutex);
695 return err;
696 }
697
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700698 curr_cpu = get_cpu();
699 /*
700 * The CPU hotplug notifier ran before us and created buffers with
701 * no files associated. So it's safe to call relay_setup_buf_file()
702 * on all currently online CPUs.
703 */
704 for_each_online_cpu(i) {
705 if (unlikely(!chan->buf[i])) {
Ingo Molnar7a51cff2008-12-29 16:03:40 +0100706 WARN_ONCE(1, KERN_ERR "CPU has no buffer!\n");
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700707 err = -EINVAL;
708 break;
709 }
710
711 dentry = relay_create_buf_file(chan, chan->buf[i], i);
712 if (unlikely(!dentry)) {
713 err = -EINVAL;
714 break;
715 }
716
717 if (curr_cpu == i) {
718 local_irq_save(flags);
719 relay_set_buf_dentry(chan->buf[i], dentry);
720 local_irq_restore(flags);
721 } else {
722 disp.buf = chan->buf[i];
723 disp.dentry = dentry;
724 smp_mb();
725 /* relay_channels_mutex must be held, so wait. */
726 err = smp_call_function_single(i,
727 __relay_set_buf_dentry,
728 &disp, 1);
729 }
730 if (unlikely(err))
731 break;
732 }
733 put_cpu();
734 mutex_unlock(&relay_channels_mutex);
735
736 return err;
737}
Akash Goel59dbb2a2016-08-02 14:07:18 -0700738EXPORT_SYMBOL_GPL(relay_late_setup_files);
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700739
Jens Axboeb86ff9812006-03-23 19:56:55 +0100740/**
741 * relay_switch_subbuf - switch to a new sub-buffer
742 * @buf: channel buffer
743 * @length: size of current event
744 *
745 * Returns either the length passed in or 0 if full.
746 *
747 * Performs sub-buffer-switch tasks such as invoking callbacks,
748 * updating padding counts, waking up readers, etc.
749 */
750size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
751{
752 void *old, *new;
753 size_t old_subbuf, new_subbuf;
754
755 if (unlikely(length > buf->chan->subbuf_size))
756 goto toobig;
757
758 if (buf->offset != buf->chan->subbuf_size + 1) {
759 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
760 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
761 buf->padding[old_subbuf] = buf->prev_padding;
762 buf->subbufs_produced++;
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700763 if (buf->dentry)
David Howells7682c912015-03-17 22:26:16 +0000764 d_inode(buf->dentry)->i_size +=
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700765 buf->chan->subbuf_size -
766 buf->padding[old_subbuf];
767 else
768 buf->early_bytes += buf->chan->subbuf_size -
769 buf->padding[old_subbuf];
Jens Axboe221415d2006-03-23 19:57:55 +0100770 smp_mb();
Tom Zanussi7c9cb382007-05-09 02:34:01 -0700771 if (waitqueue_active(&buf->read_wait))
772 /*
773 * Calling wake_up_interruptible() from here
774 * will deadlock if we happen to be logging
775 * from the scheduler (trying to re-grab
776 * rq->lock), so defer it.
777 */
Ingo Molnar74019222009-02-18 12:23:29 +0100778 mod_timer(&buf->timer, jiffies + 1);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100779 }
780
781 old = buf->data;
782 new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
783 new = buf->start + new_subbuf * buf->chan->subbuf_size;
784 buf->offset = 0;
785 if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
786 buf->offset = buf->chan->subbuf_size + 1;
787 return 0;
788 }
789 buf->data = new;
790 buf->padding[new_subbuf] = 0;
791
792 if (unlikely(length + buf->offset > buf->chan->subbuf_size))
793 goto toobig;
794
795 return length;
796
797toobig:
798 buf->chan->last_toobig = length;
799 return 0;
800}
801EXPORT_SYMBOL_GPL(relay_switch_subbuf);
802
803/**
804 * relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
805 * @chan: the channel
806 * @cpu: the cpu associated with the channel buffer to update
807 * @subbufs_consumed: number of sub-buffers to add to current buf's count
808 *
809 * Adds to the channel buffer's consumed sub-buffer count.
810 * subbufs_consumed should be the number of sub-buffers newly consumed,
811 * not the total consumed.
812 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800813 * NOTE. Kernel clients don't need to call this function if the channel
Jens Axboeb86ff9812006-03-23 19:56:55 +0100814 * mode is 'overwrite'.
815 */
816void relay_subbufs_consumed(struct rchan *chan,
817 unsigned int cpu,
818 size_t subbufs_consumed)
819{
820 struct rchan_buf *buf;
821
822 if (!chan)
823 return;
824
Aravind Srinivasan2c53d912009-04-02 16:58:59 -0700825 if (cpu >= NR_CPUS || !chan->buf[cpu] ||
826 subbufs_consumed > chan->n_subbufs)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100827 return;
828
829 buf = chan->buf[cpu];
Aravind Srinivasan2c53d912009-04-02 16:58:59 -0700830 if (subbufs_consumed > buf->subbufs_produced - buf->subbufs_consumed)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100831 buf->subbufs_consumed = buf->subbufs_produced;
Aravind Srinivasan2c53d912009-04-02 16:58:59 -0700832 else
833 buf->subbufs_consumed += subbufs_consumed;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100834}
835EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
836
837/**
838 * relay_close - close the channel
839 * @chan: the channel
840 *
841 * Closes all channel buffers and frees the channel.
842 */
843void relay_close(struct rchan *chan)
844{
845 unsigned int i;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100846
847 if (!chan)
848 return;
849
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800850 mutex_lock(&relay_channels_mutex);
851 if (chan->is_global && chan->buf[0])
852 relay_close_buf(chan->buf[0]);
853 else
854 for_each_possible_cpu(i)
855 if (chan->buf[i])
856 relay_close_buf(chan->buf[i]);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100857
858 if (chan->last_toobig)
859 printk(KERN_WARNING "relay: one or more items not logged "
860 "[item size (%Zd) > sub-buffer size (%Zd)]\n",
861 chan->last_toobig, chan->subbuf_size);
862
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800863 list_del(&chan->list);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100864 kref_put(&chan->kref, relay_destroy_channel);
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800865 mutex_unlock(&relay_channels_mutex);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100866}
867EXPORT_SYMBOL_GPL(relay_close);
868
869/**
870 * relay_flush - close the channel
871 * @chan: the channel
872 *
Randy Dunlap4c78a662006-09-29 01:59:10 -0700873 * Flushes all channel buffers, i.e. forces buffer switch.
Jens Axboeb86ff9812006-03-23 19:56:55 +0100874 */
875void relay_flush(struct rchan *chan)
876{
877 unsigned int i;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100878
879 if (!chan)
880 return;
881
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800882 if (chan->is_global && chan->buf[0]) {
883 relay_switch_subbuf(chan->buf[0], 0);
884 return;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100885 }
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800886
887 mutex_lock(&relay_channels_mutex);
888 for_each_possible_cpu(i)
889 if (chan->buf[i])
890 relay_switch_subbuf(chan->buf[i], 0);
891 mutex_unlock(&relay_channels_mutex);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100892}
893EXPORT_SYMBOL_GPL(relay_flush);
894
895/**
896 * relay_file_open - open file op for relay files
897 * @inode: the inode
898 * @filp: the file
899 *
900 * Increments the channel buffer refcount.
901 */
902static int relay_file_open(struct inode *inode, struct file *filp)
903{
Theodore Ts'o8e18e292006-09-27 01:50:46 -0700904 struct rchan_buf *buf = inode->i_private;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100905 kref_get(&buf->kref);
906 filp->private_data = buf;
907
Lai Jiangshan37529fe2008-03-26 12:01:28 +0100908 return nonseekable_open(inode, filp);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100909}
910
911/**
912 * relay_file_mmap - mmap file op for relay files
913 * @filp: the file
914 * @vma: the vma describing what to map
915 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800916 * Calls upon relay_mmap_buf() to map the file into user space.
Jens Axboeb86ff9812006-03-23 19:56:55 +0100917 */
918static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
919{
920 struct rchan_buf *buf = filp->private_data;
921 return relay_mmap_buf(buf, vma);
922}
923
924/**
925 * relay_file_poll - poll file op for relay files
926 * @filp: the file
927 * @wait: poll table
928 *
929 * Poll implemention.
930 */
931static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
932{
933 unsigned int mask = 0;
934 struct rchan_buf *buf = filp->private_data;
935
936 if (buf->finalized)
937 return POLLERR;
938
939 if (filp->f_mode & FMODE_READ) {
940 poll_wait(filp, &buf->read_wait, wait);
941 if (!relay_buf_empty(buf))
942 mask |= POLLIN | POLLRDNORM;
943 }
944
945 return mask;
946}
947
948/**
949 * relay_file_release - release file op for relay files
950 * @inode: the inode
951 * @filp: the file
952 *
953 * Decrements the channel refcount, as the filesystem is
954 * no longer using it.
955 */
956static int relay_file_release(struct inode *inode, struct file *filp)
957{
958 struct rchan_buf *buf = filp->private_data;
959 kref_put(&buf->kref, relay_remove_buf);
960
961 return 0;
962}
963
Randy Dunlap4c78a662006-09-29 01:59:10 -0700964/*
Jens Axboeb86ff9812006-03-23 19:56:55 +0100965 * relay_file_read_consume - update the consumed count for the buffer
966 */
967static void relay_file_read_consume(struct rchan_buf *buf,
968 size_t read_pos,
969 size_t bytes_consumed)
970{
971 size_t subbuf_size = buf->chan->subbuf_size;
972 size_t n_subbufs = buf->chan->n_subbufs;
973 size_t read_subbuf;
974
Tom Zanussi32194452008-08-05 13:01:10 -0700975 if (buf->subbufs_produced == buf->subbufs_consumed &&
976 buf->offset == buf->bytes_consumed)
977 return;
978
Jens Axboeb86ff9812006-03-23 19:56:55 +0100979 if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
980 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
981 buf->bytes_consumed = 0;
982 }
983
984 buf->bytes_consumed += bytes_consumed;
Masami Hiramatsua66e3562007-06-27 14:10:04 -0700985 if (!read_pos)
986 read_subbuf = buf->subbufs_consumed % n_subbufs;
987 else
988 read_subbuf = read_pos / buf->chan->subbuf_size;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100989 if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
990 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
991 (buf->offset == subbuf_size))
992 return;
993 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
994 buf->bytes_consumed = 0;
995 }
996}
997
Randy Dunlap4c78a662006-09-29 01:59:10 -0700998/*
Jens Axboeb86ff9812006-03-23 19:56:55 +0100999 * relay_file_read_avail - boolean, are there unconsumed bytes available?
1000 */
1001static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
1002{
Jens Axboeb86ff9812006-03-23 19:56:55 +01001003 size_t subbuf_size = buf->chan->subbuf_size;
1004 size_t n_subbufs = buf->chan->n_subbufs;
Jens Axboe221415d2006-03-23 19:57:55 +01001005 size_t produced = buf->subbufs_produced;
1006 size_t consumed = buf->subbufs_consumed;
Jens Axboeb86ff9812006-03-23 19:56:55 +01001007
1008 relay_file_read_consume(buf, read_pos, 0);
1009
Tom Zanussi32194452008-08-05 13:01:10 -07001010 consumed = buf->subbufs_consumed;
1011
Jens Axboe221415d2006-03-23 19:57:55 +01001012 if (unlikely(buf->offset > subbuf_size)) {
1013 if (produced == consumed)
1014 return 0;
1015 return 1;
1016 }
1017
1018 if (unlikely(produced - consumed >= n_subbufs)) {
Masami Hiramatsua66e3562007-06-27 14:10:04 -07001019 consumed = produced - n_subbufs + 1;
Jens Axboe221415d2006-03-23 19:57:55 +01001020 buf->subbufs_consumed = consumed;
Masami Hiramatsua66e3562007-06-27 14:10:04 -07001021 buf->bytes_consumed = 0;
Jens Axboe221415d2006-03-23 19:57:55 +01001022 }
Daniel Walker1bfbc602007-10-18 03:06:05 -07001023
Jens Axboe221415d2006-03-23 19:57:55 +01001024 produced = (produced % n_subbufs) * subbuf_size + buf->offset;
1025 consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
1026
1027 if (consumed > produced)
1028 produced += n_subbufs * subbuf_size;
Daniel Walker1bfbc602007-10-18 03:06:05 -07001029
Tom Zanussi32194452008-08-05 13:01:10 -07001030 if (consumed == produced) {
1031 if (buf->offset == subbuf_size &&
1032 buf->subbufs_produced > buf->subbufs_consumed)
1033 return 1;
Jens Axboe221415d2006-03-23 19:57:55 +01001034 return 0;
Tom Zanussi32194452008-08-05 13:01:10 -07001035 }
Jens Axboe221415d2006-03-23 19:57:55 +01001036
Jens Axboeb86ff9812006-03-23 19:56:55 +01001037 return 1;
1038}
1039
1040/**
1041 * relay_file_read_subbuf_avail - return bytes available in sub-buffer
Randy Dunlap4c78a662006-09-29 01:59:10 -07001042 * @read_pos: file read position
1043 * @buf: relay channel buffer
Jens Axboeb86ff9812006-03-23 19:56:55 +01001044 */
1045static size_t relay_file_read_subbuf_avail(size_t read_pos,
1046 struct rchan_buf *buf)
1047{
1048 size_t padding, avail = 0;
1049 size_t read_subbuf, read_offset, write_subbuf, write_offset;
1050 size_t subbuf_size = buf->chan->subbuf_size;
1051
1052 write_subbuf = (buf->data - buf->start) / subbuf_size;
1053 write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
1054 read_subbuf = read_pos / subbuf_size;
1055 read_offset = read_pos % subbuf_size;
1056 padding = buf->padding[read_subbuf];
1057
1058 if (read_subbuf == write_subbuf) {
1059 if (read_offset + padding < write_offset)
1060 avail = write_offset - (read_offset + padding);
1061 } else
1062 avail = (subbuf_size - padding) - read_offset;
1063
1064 return avail;
1065}
1066
1067/**
1068 * relay_file_read_start_pos - find the first available byte to read
Randy Dunlap4c78a662006-09-29 01:59:10 -07001069 * @read_pos: file read position
1070 * @buf: relay channel buffer
Jens Axboeb86ff9812006-03-23 19:56:55 +01001071 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08001072 * If the @read_pos is in the middle of padding, return the
Jens Axboeb86ff9812006-03-23 19:56:55 +01001073 * position of the first actually available byte, otherwise
1074 * return the original value.
1075 */
1076static size_t relay_file_read_start_pos(size_t read_pos,
1077 struct rchan_buf *buf)
1078{
1079 size_t read_subbuf, padding, padding_start, padding_end;
1080 size_t subbuf_size = buf->chan->subbuf_size;
1081 size_t n_subbufs = buf->chan->n_subbufs;
David Wilder8d62fde2007-06-27 14:10:03 -07001082 size_t consumed = buf->subbufs_consumed % n_subbufs;
Jens Axboeb86ff9812006-03-23 19:56:55 +01001083
David Wilder8d62fde2007-06-27 14:10:03 -07001084 if (!read_pos)
1085 read_pos = consumed * subbuf_size + buf->bytes_consumed;
Jens Axboeb86ff9812006-03-23 19:56:55 +01001086 read_subbuf = read_pos / subbuf_size;
1087 padding = buf->padding[read_subbuf];
1088 padding_start = (read_subbuf + 1) * subbuf_size - padding;
1089 padding_end = (read_subbuf + 1) * subbuf_size;
1090 if (read_pos >= padding_start && read_pos < padding_end) {
1091 read_subbuf = (read_subbuf + 1) % n_subbufs;
1092 read_pos = read_subbuf * subbuf_size;
1093 }
1094
1095 return read_pos;
1096}
1097
1098/**
1099 * relay_file_read_end_pos - return the new read position
Randy Dunlap4c78a662006-09-29 01:59:10 -07001100 * @read_pos: file read position
1101 * @buf: relay channel buffer
1102 * @count: number of bytes to be read
Jens Axboeb86ff9812006-03-23 19:56:55 +01001103 */
1104static size_t relay_file_read_end_pos(struct rchan_buf *buf,
1105 size_t read_pos,
1106 size_t count)
1107{
1108 size_t read_subbuf, padding, end_pos;
1109 size_t subbuf_size = buf->chan->subbuf_size;
1110 size_t n_subbufs = buf->chan->n_subbufs;
1111
1112 read_subbuf = read_pos / subbuf_size;
1113 padding = buf->padding[read_subbuf];
1114 if (read_pos % subbuf_size + count + padding == subbuf_size)
1115 end_pos = (read_subbuf + 1) * subbuf_size;
1116 else
1117 end_pos = read_pos + count;
1118 if (end_pos >= subbuf_size * n_subbufs)
1119 end_pos = 0;
1120
1121 return end_pos;
1122}
1123
Randy Dunlap4c78a662006-09-29 01:59:10 -07001124/*
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001125 * subbuf_read_actor - read up to one subbuf's worth of data
Jens Axboeb86ff9812006-03-23 19:56:55 +01001126 */
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001127static int subbuf_read_actor(size_t read_start,
1128 struct rchan_buf *buf,
1129 size_t avail,
zhangwei(Jovi)8359f682013-04-30 15:28:39 -07001130 read_descriptor_t *desc)
Jens Axboeb86ff9812006-03-23 19:56:55 +01001131{
Jens Axboeb86ff9812006-03-23 19:56:55 +01001132 void *from;
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001133 int ret = 0;
Jens Axboeb86ff9812006-03-23 19:56:55 +01001134
1135 from = buf->start + read_start;
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001136 ret = avail;
Al Viroba2397e2006-10-10 22:46:57 +01001137 if (copy_to_user(desc->arg.buf, from, avail)) {
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001138 desc->error = -EFAULT;
1139 ret = 0;
Jens Axboeb86ff9812006-03-23 19:56:55 +01001140 }
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001141 desc->arg.data += ret;
1142 desc->written += ret;
1143 desc->count -= ret;
1144
Jens Axboeb86ff9812006-03-23 19:56:55 +01001145 return ret;
1146}
1147
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001148typedef int (*subbuf_actor_t) (size_t read_start,
1149 struct rchan_buf *buf,
1150 size_t avail,
zhangwei(Jovi)8359f682013-04-30 15:28:39 -07001151 read_descriptor_t *desc);
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001152
Randy Dunlap4c78a662006-09-29 01:59:10 -07001153/*
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001154 * relay_file_read_subbufs - read count bytes, bridging subbuf boundaries
1155 */
Andrew Morton192636a2006-12-22 01:11:30 -08001156static ssize_t relay_file_read_subbufs(struct file *filp, loff_t *ppos,
1157 subbuf_actor_t subbuf_actor,
Andrew Morton192636a2006-12-22 01:11:30 -08001158 read_descriptor_t *desc)
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001159{
1160 struct rchan_buf *buf = filp->private_data;
1161 size_t read_start, avail;
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001162 int ret;
1163
Al Viroba2397e2006-10-10 22:46:57 +01001164 if (!desc->count)
Jens Axboe221415d2006-03-23 19:57:55 +01001165 return 0;
1166
Al Viro59551022016-01-22 15:40:57 -05001167 inode_lock(file_inode(filp));
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001168 do {
1169 if (!relay_file_read_avail(buf, *ppos))
Jens Axboe221415d2006-03-23 19:57:55 +01001170 break;
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001171
1172 read_start = relay_file_read_start_pos(*ppos, buf);
1173 avail = relay_file_read_subbuf_avail(read_start, buf);
1174 if (!avail)
1175 break;
1176
Al Viroba2397e2006-10-10 22:46:57 +01001177 avail = min(desc->count, avail);
zhangwei(Jovi)8359f682013-04-30 15:28:39 -07001178 ret = subbuf_actor(read_start, buf, avail, desc);
Al Viroba2397e2006-10-10 22:46:57 +01001179 if (desc->error < 0)
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001180 break;
1181
1182 if (ret) {
1183 relay_file_read_consume(buf, read_start, ret);
1184 *ppos = relay_file_read_end_pos(buf, read_start, ret);
Jens Axboe221415d2006-03-23 19:57:55 +01001185 }
Al Viroba2397e2006-10-10 22:46:57 +01001186 } while (desc->count && ret);
Al Viro59551022016-01-22 15:40:57 -05001187 inode_unlock(file_inode(filp));
Jens Axboe221415d2006-03-23 19:57:55 +01001188
Al Viroba2397e2006-10-10 22:46:57 +01001189 return desc->written;
Jens Axboe221415d2006-03-23 19:57:55 +01001190}
1191
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001192static ssize_t relay_file_read(struct file *filp,
1193 char __user *buffer,
1194 size_t count,
1195 loff_t *ppos)
1196{
Al Viroba2397e2006-10-10 22:46:57 +01001197 read_descriptor_t desc;
1198 desc.written = 0;
1199 desc.count = count;
1200 desc.arg.buf = buffer;
1201 desc.error = 0;
zhangwei(Jovi)8359f682013-04-30 15:28:39 -07001202 return relay_file_read_subbufs(filp, ppos, subbuf_read_actor, &desc);
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001203}
1204
Jens Axboe1db60cf2007-06-12 08:43:46 +02001205static void relay_consume_bytes(struct rchan_buf *rbuf, int bytes_consumed)
1206{
1207 rbuf->bytes_consumed += bytes_consumed;
1208
1209 if (rbuf->bytes_consumed >= rbuf->chan->subbuf_size) {
1210 relay_subbufs_consumed(rbuf->chan, rbuf->cpu, 1);
1211 rbuf->bytes_consumed %= rbuf->chan->subbuf_size;
1212 }
1213}
1214
Tom Zanussiebf99092007-06-04 09:12:05 +02001215static void relay_pipe_buf_release(struct pipe_inode_info *pipe,
1216 struct pipe_buffer *buf)
Jens Axboe221415d2006-03-23 19:57:55 +01001217{
Tom Zanussiebf99092007-06-04 09:12:05 +02001218 struct rchan_buf *rbuf;
1219
1220 rbuf = (struct rchan_buf *)page_private(buf->page);
Jens Axboe1db60cf2007-06-12 08:43:46 +02001221 relay_consume_bytes(rbuf, buf->private);
Tom Zanussiebf99092007-06-04 09:12:05 +02001222}
1223
Alexey Dobriyan28dfef82009-12-15 16:46:48 -08001224static const struct pipe_buf_operations relay_pipe_buf_ops = {
Tom Zanussiebf99092007-06-04 09:12:05 +02001225 .can_merge = 0,
Jens Axboecac36bb02007-06-14 13:10:48 +02001226 .confirm = generic_pipe_buf_confirm,
Tom Zanussiebf99092007-06-04 09:12:05 +02001227 .release = relay_pipe_buf_release,
1228 .steal = generic_pipe_buf_steal,
1229 .get = generic_pipe_buf_get,
1230};
1231
Jens Axboe5eb7f9f2008-03-26 12:04:09 +01001232static void relay_page_release(struct splice_pipe_desc *spd, unsigned int i)
1233{
1234}
1235
Tom Zanussid3f35d92007-07-12 08:12:05 +02001236/*
Tom Zanussiebf99092007-06-04 09:12:05 +02001237 * subbuf_splice_actor - splice up to one subbuf's worth of data
1238 */
Dan Carpenter5f1664f2010-03-05 13:42:51 -08001239static ssize_t subbuf_splice_actor(struct file *in,
Tom Zanussiebf99092007-06-04 09:12:05 +02001240 loff_t *ppos,
1241 struct pipe_inode_info *pipe,
1242 size_t len,
1243 unsigned int flags,
1244 int *nonpad_ret)
1245{
Dan Carpenter5f1664f2010-03-05 13:42:51 -08001246 unsigned int pidx, poff, total_len, subbuf_pages, nr_pages;
Tom Zanussiebf99092007-06-04 09:12:05 +02001247 struct rchan_buf *rbuf = in->private_data;
1248 unsigned int subbuf_size = rbuf->chan->subbuf_size;
Tom Zanussi24da24d2007-07-12 08:12:04 +02001249 uint64_t pos = (uint64_t) *ppos;
1250 uint32_t alloc_size = (uint32_t) rbuf->chan->alloc_size;
1251 size_t read_start = (size_t) do_div(pos, alloc_size);
Tom Zanussiebf99092007-06-04 09:12:05 +02001252 size_t read_subbuf = read_start / subbuf_size;
1253 size_t padding = rbuf->padding[read_subbuf];
1254 size_t nonpad_end = read_subbuf * subbuf_size + subbuf_size - padding;
Jens Axboe35f3d142010-05-20 10:43:18 +02001255 struct page *pages[PIPE_DEF_BUFFERS];
1256 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe1db60cf2007-06-12 08:43:46 +02001257 struct splice_pipe_desc spd = {
1258 .pages = pages,
1259 .nr_pages = 0,
Eric Dumazet047fe362012-06-12 15:24:40 +02001260 .nr_pages_max = PIPE_DEF_BUFFERS,
Jens Axboe1db60cf2007-06-12 08:43:46 +02001261 .partial = partial,
1262 .flags = flags,
1263 .ops = &relay_pipe_buf_ops,
Jens Axboe5eb7f9f2008-03-26 12:04:09 +01001264 .spd_release = relay_page_release,
Jens Axboe1db60cf2007-06-12 08:43:46 +02001265 };
Dan Carpenter5f1664f2010-03-05 13:42:51 -08001266 ssize_t ret;
Tom Zanussiebf99092007-06-04 09:12:05 +02001267
1268 if (rbuf->subbufs_produced == rbuf->subbufs_consumed)
1269 return 0;
Jens Axboe35f3d142010-05-20 10:43:18 +02001270 if (splice_grow_spd(pipe, &spd))
1271 return -ENOMEM;
Tom Zanussiebf99092007-06-04 09:12:05 +02001272
Jens Axboe1db60cf2007-06-12 08:43:46 +02001273 /*
1274 * Adjust read len, if longer than what is available
1275 */
1276 if (len > (subbuf_size - read_start % subbuf_size))
1277 len = subbuf_size - read_start % subbuf_size;
Tom Zanussiebf99092007-06-04 09:12:05 +02001278
1279 subbuf_pages = rbuf->chan->alloc_size >> PAGE_SHIFT;
1280 pidx = (read_start / PAGE_SIZE) % subbuf_pages;
1281 poff = read_start & ~PAGE_MASK;
Al Viroa786c062014-04-11 12:01:03 -04001282 nr_pages = min_t(unsigned int, subbuf_pages, spd.nr_pages_max);
Tom Zanussiebf99092007-06-04 09:12:05 +02001283
Jens Axboe16d54662008-03-17 09:04:59 +01001284 for (total_len = 0; spd.nr_pages < nr_pages; spd.nr_pages++) {
Jens Axboe1db60cf2007-06-12 08:43:46 +02001285 unsigned int this_len, this_end, private;
1286 unsigned int cur_pos = read_start + total_len;
Tom Zanussiebf99092007-06-04 09:12:05 +02001287
Jens Axboe1db60cf2007-06-12 08:43:46 +02001288 if (!len)
1289 break;
1290
1291 this_len = min_t(unsigned long, len, PAGE_SIZE - poff);
1292 private = this_len;
1293
1294 spd.pages[spd.nr_pages] = rbuf->page_array[pidx];
1295 spd.partial[spd.nr_pages].offset = poff;
1296
1297 this_end = cur_pos + this_len;
1298 if (this_end >= nonpad_end) {
1299 this_len = nonpad_end - cur_pos;
1300 private = this_len + padding;
1301 }
1302 spd.partial[spd.nr_pages].len = this_len;
1303 spd.partial[spd.nr_pages].private = private;
1304
1305 len -= this_len;
1306 total_len += this_len;
1307 poff = 0;
1308 pidx = (pidx + 1) % subbuf_pages;
1309
1310 if (this_end >= nonpad_end) {
1311 spd.nr_pages++;
Tom Zanussiebf99092007-06-04 09:12:05 +02001312 break;
1313 }
Tom Zanussiebf99092007-06-04 09:12:05 +02001314 }
1315
Jens Axboe35f3d142010-05-20 10:43:18 +02001316 ret = 0;
Jens Axboe1db60cf2007-06-12 08:43:46 +02001317 if (!spd.nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +02001318 goto out;
Tom Zanussiebf99092007-06-04 09:12:05 +02001319
Jens Axboe1db60cf2007-06-12 08:43:46 +02001320 ret = *nonpad_ret = splice_to_pipe(pipe, &spd);
1321 if (ret < 0 || ret < total_len)
Jens Axboe35f3d142010-05-20 10:43:18 +02001322 goto out;
Tom Zanussiebf99092007-06-04 09:12:05 +02001323
Jens Axboe1db60cf2007-06-12 08:43:46 +02001324 if (read_start + ret == nonpad_end)
1325 ret += padding;
1326
Jens Axboe35f3d142010-05-20 10:43:18 +02001327out:
Eric Dumazet047fe362012-06-12 15:24:40 +02001328 splice_shrink_spd(&spd);
1329 return ret;
Tom Zanussiebf99092007-06-04 09:12:05 +02001330}
1331
1332static ssize_t relay_file_splice_read(struct file *in,
1333 loff_t *ppos,
1334 struct pipe_inode_info *pipe,
1335 size_t len,
1336 unsigned int flags)
1337{
1338 ssize_t spliced;
1339 int ret;
1340 int nonpad_ret = 0;
1341
1342 ret = 0;
1343 spliced = 0;
1344
Tom Zanussia82c53a2008-05-09 13:28:36 +02001345 while (len && !spliced) {
Tom Zanussiebf99092007-06-04 09:12:05 +02001346 ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret);
1347 if (ret < 0)
1348 break;
1349 else if (!ret) {
Tom Zanussifbb5b7a2008-12-09 13:14:10 -08001350 if (flags & SPLICE_F_NONBLOCK)
Tom Zanussiebf99092007-06-04 09:12:05 +02001351 ret = -EAGAIN;
Tom Zanussifbb5b7a2008-12-09 13:14:10 -08001352 break;
Tom Zanussiebf99092007-06-04 09:12:05 +02001353 }
1354
1355 *ppos += ret;
1356 if (ret > len)
1357 len = 0;
1358 else
1359 len -= ret;
1360 spliced += nonpad_ret;
1361 nonpad_ret = 0;
1362 }
1363
1364 if (spliced)
1365 return spliced;
1366
1367 return ret;
Jens Axboe221415d2006-03-23 19:57:55 +01001368}
1369
Helge Deller15ad7cd2006-12-06 20:40:36 -08001370const struct file_operations relay_file_operations = {
Jens Axboeb86ff9812006-03-23 19:56:55 +01001371 .open = relay_file_open,
1372 .poll = relay_file_poll,
1373 .mmap = relay_file_mmap,
1374 .read = relay_file_read,
1375 .llseek = no_llseek,
1376 .release = relay_file_release,
Tom Zanussiebf99092007-06-04 09:12:05 +02001377 .splice_read = relay_file_splice_read,
Jens Axboeb86ff9812006-03-23 19:56:55 +01001378};
1379EXPORT_SYMBOL_GPL(relay_file_operations);
Mathieu Desnoyers23c88752007-02-10 01:45:05 -08001380
1381static __init int relay_init(void)
1382{
1383
1384 hotcpu_notifier(relay_hotcpu_callback, 0);
1385 return 0;
1386}
1387
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -07001388early_initcall(relay_init);