blob: 859ea5a9605fa40fe47aedcb865ab08c66a43797 [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>
18#include <linux/module.h>
19#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 Dobriyanf0f37e2f2009-09-27 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{
84 if (is_vmalloc_addr(array))
85 vfree(array);
86 else
87 kfree(array);
88}
89
Jens Axboeb86ff9812006-03-23 19:56:55 +010090/**
91 * relay_mmap_buf: - mmap channel buffer to process address space
92 * @buf: relay channel buffer
93 * @vma: vm_area_struct describing memory to be mapped
94 *
95 * Returns 0 if ok, negative on error
96 *
97 * Caller should already have grabbed mmap_sem.
98 */
Adrian Bunk01c55ed2007-07-19 01:48:32 -070099static int relay_mmap_buf(struct rchan_buf *buf, struct vm_area_struct *vma)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100100{
101 unsigned long length = vma->vm_end - vma->vm_start;
102 struct file *filp = vma->vm_file;
103
104 if (!buf)
105 return -EBADF;
106
107 if (length != (unsigned long)buf->chan->alloc_size)
108 return -EINVAL;
109
110 vma->vm_ops = &relay_file_mmap_ops;
Nick Piggin2f987352008-02-02 03:08:53 +0100111 vma->vm_flags |= VM_DONTEXPAND;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100112 vma->vm_private_data = buf;
113 buf->chan->cb->buf_mapped(buf, filp);
114
115 return 0;
116}
117
118/**
119 * relay_alloc_buf - allocate a channel buffer
120 * @buf: the buffer struct
121 * @size: total size of the buffer
122 *
Randy Dunlap4c78a662006-09-29 01:59:10 -0700123 * Returns a pointer to the resulting buffer, %NULL if unsuccessful. The
Jens Axboe221415d2006-03-23 19:57:55 +0100124 * passed in size will get page aligned, if it isn't already.
Jens Axboeb86ff9812006-03-23 19:56:55 +0100125 */
Jens Axboe221415d2006-03-23 19:57:55 +0100126static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100127{
128 void *mem;
129 unsigned int i, j, n_pages;
130
Jens Axboe221415d2006-03-23 19:57:55 +0100131 *size = PAGE_ALIGN(*size);
132 n_pages = *size >> PAGE_SHIFT;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100133
Masami Hiramatsu68ab3d82008-04-29 01:03:46 -0700134 buf->page_array = relay_alloc_page_array(n_pages);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100135 if (!buf->page_array)
136 return NULL;
137
138 for (i = 0; i < n_pages; i++) {
139 buf->page_array[i] = alloc_page(GFP_KERNEL);
140 if (unlikely(!buf->page_array[i]))
141 goto depopulate;
Tom Zanussiebf99092007-06-04 09:12:05 +0200142 set_page_private(buf->page_array[i], (unsigned long)buf);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100143 }
144 mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL);
145 if (!mem)
146 goto depopulate;
147
Jens Axboe221415d2006-03-23 19:57:55 +0100148 memset(mem, 0, *size);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100149 buf->page_count = n_pages;
150 return mem;
151
152depopulate:
153 for (j = 0; j < i; j++)
154 __free_page(buf->page_array[j]);
Masami Hiramatsu68ab3d82008-04-29 01:03:46 -0700155 relay_free_page_array(buf->page_array);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100156 return NULL;
157}
158
159/**
160 * relay_create_buf - allocate and initialize a channel buffer
Randy Dunlap4c78a662006-09-29 01:59:10 -0700161 * @chan: the relay channel
Jens Axboeb86ff9812006-03-23 19:56:55 +0100162 *
Randy Dunlap4c78a662006-09-29 01:59:10 -0700163 * Returns channel buffer if successful, %NULL otherwise.
Jens Axboeb86ff9812006-03-23 19:56:55 +0100164 */
Adrian Bunk01c55ed2007-07-19 01:48:32 -0700165static struct rchan_buf *relay_create_buf(struct rchan *chan)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100166{
Robert P. J. Daycd861282006-12-13 00:34:52 -0800167 struct rchan_buf *buf = kzalloc(sizeof(struct rchan_buf), GFP_KERNEL);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100168 if (!buf)
169 return NULL;
170
171 buf->padding = kmalloc(chan->n_subbufs * sizeof(size_t *), GFP_KERNEL);
172 if (!buf->padding)
173 goto free_buf;
174
Jens Axboe221415d2006-03-23 19:57:55 +0100175 buf->start = relay_alloc_buf(buf, &chan->alloc_size);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100176 if (!buf->start)
177 goto free_buf;
178
179 buf->chan = chan;
180 kref_get(&buf->chan->kref);
181 return buf;
182
183free_buf:
184 kfree(buf->padding);
185 kfree(buf);
186 return NULL;
187}
188
189/**
190 * relay_destroy_channel - free the channel struct
Randy Dunlap4c78a662006-09-29 01:59:10 -0700191 * @kref: target kernel reference that contains the relay channel
Jens Axboeb86ff9812006-03-23 19:56:55 +0100192 *
193 * Should only be called from kref_put().
194 */
Adrian Bunk01c55ed2007-07-19 01:48:32 -0700195static void relay_destroy_channel(struct kref *kref)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100196{
197 struct rchan *chan = container_of(kref, struct rchan, kref);
198 kfree(chan);
199}
200
201/**
202 * relay_destroy_buf - destroy an rchan_buf struct and associated buffer
203 * @buf: the buffer struct
204 */
Adrian Bunk01c55ed2007-07-19 01:48:32 -0700205static void relay_destroy_buf(struct rchan_buf *buf)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100206{
207 struct rchan *chan = buf->chan;
208 unsigned int i;
209
210 if (likely(buf->start)) {
211 vunmap(buf->start);
212 for (i = 0; i < buf->page_count; i++)
213 __free_page(buf->page_array[i]);
Masami Hiramatsu68ab3d82008-04-29 01:03:46 -0700214 relay_free_page_array(buf->page_array);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100215 }
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800216 chan->buf[buf->cpu] = NULL;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100217 kfree(buf->padding);
218 kfree(buf);
219 kref_put(&chan->kref, relay_destroy_channel);
220}
221
222/**
223 * relay_remove_buf - remove a channel buffer
Randy Dunlap4c78a662006-09-29 01:59:10 -0700224 * @kref: target kernel reference that contains the relay buffer
Jens Axboeb86ff9812006-03-23 19:56:55 +0100225 *
226 * Removes the file from the fileystem, which also frees the
227 * rchan_buf_struct and the channel buffer. Should only be called from
228 * kref_put().
229 */
Adrian Bunk01c55ed2007-07-19 01:48:32 -0700230static void relay_remove_buf(struct kref *kref)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100231{
232 struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
233 buf->chan->cb->remove_buf_file(buf->dentry);
234 relay_destroy_buf(buf);
235}
236
237/**
238 * relay_buf_empty - boolean, is the channel buffer empty?
239 * @buf: channel buffer
240 *
241 * Returns 1 if the buffer is empty, 0 otherwise.
242 */
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,
305 int mode,
306 struct rchan_buf *buf,
307 int *is_global)
308{
309 return NULL;
310}
311
312/*
313 * remove_buf_file() default callback. Does nothing.
314 */
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;
407 buf->dentry->d_inode->i_size = buf->early_bytes;
408}
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);
454 }
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800455
456 buf->cpu = cpu;
457 __relay_reset(buf, 1);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100458
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800459 if(chan->is_global) {
460 chan->buf[0] = buf;
461 buf->cpu = 0;
462 }
463
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700464 return buf;
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800465
466free_buf:
467 relay_destroy_buf(buf);
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700468 return NULL;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100469}
470
471/**
472 * relay_close_buf - close a channel buffer
473 * @buf: channel buffer
474 *
475 * Marks the buffer finalized and restores the default callbacks.
476 * The channel buffer and channel buffer data structure are then freed
477 * automatically when the last reference is given up.
478 */
Andrew Morton192636a2006-12-22 01:11:30 -0800479static void relay_close_buf(struct rchan_buf *buf)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100480{
481 buf->finalized = 1;
Tom Zanussi7c9cb382007-05-09 02:34:01 -0700482 del_timer_sync(&buf->timer);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100483 kref_put(&buf->kref, relay_remove_buf);
484}
485
Andrew Morton192636a2006-12-22 01:11:30 -0800486static void setup_callbacks(struct rchan *chan,
Jens Axboeb86ff9812006-03-23 19:56:55 +0100487 struct rchan_callbacks *cb)
488{
489 if (!cb) {
490 chan->cb = &default_channel_callbacks;
491 return;
492 }
493
494 if (!cb->subbuf_start)
495 cb->subbuf_start = subbuf_start_default_callback;
496 if (!cb->buf_mapped)
497 cb->buf_mapped = buf_mapped_default_callback;
498 if (!cb->buf_unmapped)
499 cb->buf_unmapped = buf_unmapped_default_callback;
500 if (!cb->create_buf_file)
501 cb->create_buf_file = create_buf_file_default_callback;
502 if (!cb->remove_buf_file)
503 cb->remove_buf_file = remove_buf_file_default_callback;
504 chan->cb = cb;
505}
506
507/**
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800508 * relay_hotcpu_callback - CPU hotplug callback
509 * @nb: notifier block
510 * @action: hotplug action to take
511 * @hcpu: CPU number
512 *
Randy Dunlap05fb6bf2007-02-28 20:12:13 -0800513 * Returns the success/failure of the operation. (%NOTIFY_OK, %NOTIFY_BAD)
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800514 */
515static int __cpuinit relay_hotcpu_callback(struct notifier_block *nb,
516 unsigned long action,
517 void *hcpu)
518{
519 unsigned int hotcpu = (unsigned long)hcpu;
520 struct rchan *chan;
521
522 switch(action) {
523 case CPU_UP_PREPARE:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -0700524 case CPU_UP_PREPARE_FROZEN:
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800525 mutex_lock(&relay_channels_mutex);
526 list_for_each_entry(chan, &relay_channels, list) {
527 if (chan->buf[hotcpu])
528 continue;
529 chan->buf[hotcpu] = relay_open_buf(chan, hotcpu);
530 if(!chan->buf[hotcpu]) {
531 printk(KERN_ERR
532 "relay_hotcpu_callback: cpu %d buffer "
533 "creation failed\n", hotcpu);
534 mutex_unlock(&relay_channels_mutex);
Akinobu Mita80b51842010-05-26 14:43:32 -0700535 return notifier_from_errno(-ENOMEM);
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800536 }
537 }
538 mutex_unlock(&relay_channels_mutex);
539 break;
540 case CPU_DEAD:
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -0700541 case CPU_DEAD_FROZEN:
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800542 /* No need to flush the cpu : will be flushed upon
543 * final relay_flush() call. */
544 break;
545 }
546 return NOTIFY_OK;
547}
548
549/**
Jens Axboeb86ff9812006-03-23 19:56:55 +0100550 * relay_open - create a new relay channel
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700551 * @base_filename: base name of files to create, %NULL for buffering only
552 * @parent: dentry of parent directory, %NULL for root directory or buffer
Jens Axboeb86ff9812006-03-23 19:56:55 +0100553 * @subbuf_size: size of sub-buffers
554 * @n_subbufs: number of sub-buffers
555 * @cb: client callback functions
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800556 * @private_data: user-defined data
Jens Axboeb86ff9812006-03-23 19:56:55 +0100557 *
Randy Dunlap4c78a662006-09-29 01:59:10 -0700558 * Returns channel pointer if successful, %NULL otherwise.
Jens Axboeb86ff9812006-03-23 19:56:55 +0100559 *
560 * Creates a channel buffer for each cpu using the sizes and
561 * attributes specified. The created channel buffer files
562 * will be named base_filename0...base_filenameN-1. File
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800563 * permissions will be %S_IRUSR.
Jens Axboeb86ff9812006-03-23 19:56:55 +0100564 */
565struct rchan *relay_open(const char *base_filename,
566 struct dentry *parent,
567 size_t subbuf_size,
568 size_t n_subbufs,
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800569 struct rchan_callbacks *cb,
570 void *private_data)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100571{
572 unsigned int i;
573 struct rchan *chan;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100574
575 if (!(subbuf_size && n_subbufs))
576 return NULL;
577
Robert P. J. Daycd861282006-12-13 00:34:52 -0800578 chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100579 if (!chan)
580 return NULL;
581
582 chan->version = RELAYFS_CHANNEL_VERSION;
583 chan->n_subbufs = n_subbufs;
584 chan->subbuf_size = subbuf_size;
585 chan->alloc_size = FIX_SIZE(subbuf_size * n_subbufs);
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800586 chan->parent = parent;
587 chan->private_data = private_data;
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700588 if (base_filename) {
589 chan->has_base_filename = 1;
590 strlcpy(chan->base_filename, base_filename, NAME_MAX);
591 }
Jens Axboeb86ff9812006-03-23 19:56:55 +0100592 setup_callbacks(chan, cb);
593 kref_init(&chan->kref);
594
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800595 mutex_lock(&relay_channels_mutex);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100596 for_each_online_cpu(i) {
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800597 chan->buf[i] = relay_open_buf(chan, i);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100598 if (!chan->buf[i])
599 goto free_bufs;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100600 }
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800601 list_add(&chan->list, &relay_channels);
602 mutex_unlock(&relay_channels_mutex);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100603
Jens Axboeb86ff9812006-03-23 19:56:55 +0100604 return chan;
605
606free_bufs:
Lai Jiangshan98ba4032008-11-14 10:44:59 +0100607 for_each_possible_cpu(i) {
608 if (chan->buf[i])
609 relay_close_buf(chan->buf[i]);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100610 }
Jens Axboeb86ff9812006-03-23 19:56:55 +0100611
Jens Axboeb86ff9812006-03-23 19:56:55 +0100612 kref_put(&chan->kref, relay_destroy_channel);
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800613 mutex_unlock(&relay_channels_mutex);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100614 return NULL;
615}
616EXPORT_SYMBOL_GPL(relay_open);
617
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700618struct rchan_percpu_buf_dispatcher {
619 struct rchan_buf *buf;
620 struct dentry *dentry;
621};
622
623/* Called in atomic context. */
624static void __relay_set_buf_dentry(void *info)
625{
626 struct rchan_percpu_buf_dispatcher *p = info;
627
628 relay_set_buf_dentry(p->buf, p->dentry);
629}
630
631/**
632 * relay_late_setup_files - triggers file creation
633 * @chan: channel to operate on
634 * @base_filename: base name of files to create
635 * @parent: dentry of parent directory, %NULL for root directory
636 *
637 * Returns 0 if successful, non-zero otherwise.
638 *
639 * Use to setup files for a previously buffer-only channel.
640 * Useful to do early tracing in kernel, before VFS is up, for example.
641 */
642int relay_late_setup_files(struct rchan *chan,
643 const char *base_filename,
644 struct dentry *parent)
645{
646 int err = 0;
647 unsigned int i, curr_cpu;
648 unsigned long flags;
649 struct dentry *dentry;
650 struct rchan_percpu_buf_dispatcher disp;
651
652 if (!chan || !base_filename)
653 return -EINVAL;
654
655 strlcpy(chan->base_filename, base_filename, NAME_MAX);
656
657 mutex_lock(&relay_channels_mutex);
658 /* Is chan already set up? */
Jiri Slabyb786c6a2009-01-17 12:04:36 +0100659 if (unlikely(chan->has_base_filename)) {
660 mutex_unlock(&relay_channels_mutex);
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700661 return -EEXIST;
Jiri Slabyb786c6a2009-01-17 12:04:36 +0100662 }
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700663 chan->has_base_filename = 1;
664 chan->parent = parent;
665 curr_cpu = get_cpu();
666 /*
667 * The CPU hotplug notifier ran before us and created buffers with
668 * no files associated. So it's safe to call relay_setup_buf_file()
669 * on all currently online CPUs.
670 */
671 for_each_online_cpu(i) {
672 if (unlikely(!chan->buf[i])) {
Ingo Molnar7a51cff2008-12-29 16:03:40 +0100673 WARN_ONCE(1, KERN_ERR "CPU has no buffer!\n");
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700674 err = -EINVAL;
675 break;
676 }
677
678 dentry = relay_create_buf_file(chan, chan->buf[i], i);
679 if (unlikely(!dentry)) {
680 err = -EINVAL;
681 break;
682 }
683
684 if (curr_cpu == i) {
685 local_irq_save(flags);
686 relay_set_buf_dentry(chan->buf[i], dentry);
687 local_irq_restore(flags);
688 } else {
689 disp.buf = chan->buf[i];
690 disp.dentry = dentry;
691 smp_mb();
692 /* relay_channels_mutex must be held, so wait. */
693 err = smp_call_function_single(i,
694 __relay_set_buf_dentry,
695 &disp, 1);
696 }
697 if (unlikely(err))
698 break;
699 }
700 put_cpu();
701 mutex_unlock(&relay_channels_mutex);
702
703 return err;
704}
705
Jens Axboeb86ff9812006-03-23 19:56:55 +0100706/**
707 * relay_switch_subbuf - switch to a new sub-buffer
708 * @buf: channel buffer
709 * @length: size of current event
710 *
711 * Returns either the length passed in or 0 if full.
712 *
713 * Performs sub-buffer-switch tasks such as invoking callbacks,
714 * updating padding counts, waking up readers, etc.
715 */
716size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
717{
718 void *old, *new;
719 size_t old_subbuf, new_subbuf;
720
721 if (unlikely(length > buf->chan->subbuf_size))
722 goto toobig;
723
724 if (buf->offset != buf->chan->subbuf_size + 1) {
725 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
726 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
727 buf->padding[old_subbuf] = buf->prev_padding;
728 buf->subbufs_produced++;
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700729 if (buf->dentry)
730 buf->dentry->d_inode->i_size +=
731 buf->chan->subbuf_size -
732 buf->padding[old_subbuf];
733 else
734 buf->early_bytes += buf->chan->subbuf_size -
735 buf->padding[old_subbuf];
Jens Axboe221415d2006-03-23 19:57:55 +0100736 smp_mb();
Tom Zanussi7c9cb382007-05-09 02:34:01 -0700737 if (waitqueue_active(&buf->read_wait))
738 /*
739 * Calling wake_up_interruptible() from here
740 * will deadlock if we happen to be logging
741 * from the scheduler (trying to re-grab
742 * rq->lock), so defer it.
743 */
Ingo Molnar74019222009-02-18 12:23:29 +0100744 mod_timer(&buf->timer, jiffies + 1);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100745 }
746
747 old = buf->data;
748 new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
749 new = buf->start + new_subbuf * buf->chan->subbuf_size;
750 buf->offset = 0;
751 if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
752 buf->offset = buf->chan->subbuf_size + 1;
753 return 0;
754 }
755 buf->data = new;
756 buf->padding[new_subbuf] = 0;
757
758 if (unlikely(length + buf->offset > buf->chan->subbuf_size))
759 goto toobig;
760
761 return length;
762
763toobig:
764 buf->chan->last_toobig = length;
765 return 0;
766}
767EXPORT_SYMBOL_GPL(relay_switch_subbuf);
768
769/**
770 * relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
771 * @chan: the channel
772 * @cpu: the cpu associated with the channel buffer to update
773 * @subbufs_consumed: number of sub-buffers to add to current buf's count
774 *
775 * Adds to the channel buffer's consumed sub-buffer count.
776 * subbufs_consumed should be the number of sub-buffers newly consumed,
777 * not the total consumed.
778 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800779 * NOTE. Kernel clients don't need to call this function if the channel
Jens Axboeb86ff9812006-03-23 19:56:55 +0100780 * mode is 'overwrite'.
781 */
782void relay_subbufs_consumed(struct rchan *chan,
783 unsigned int cpu,
784 size_t subbufs_consumed)
785{
786 struct rchan_buf *buf;
787
788 if (!chan)
789 return;
790
Aravind Srinivasan2c53d912009-04-02 16:58:59 -0700791 if (cpu >= NR_CPUS || !chan->buf[cpu] ||
792 subbufs_consumed > chan->n_subbufs)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100793 return;
794
795 buf = chan->buf[cpu];
Aravind Srinivasan2c53d912009-04-02 16:58:59 -0700796 if (subbufs_consumed > buf->subbufs_produced - buf->subbufs_consumed)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100797 buf->subbufs_consumed = buf->subbufs_produced;
Aravind Srinivasan2c53d912009-04-02 16:58:59 -0700798 else
799 buf->subbufs_consumed += subbufs_consumed;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100800}
801EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
802
803/**
804 * relay_close - close the channel
805 * @chan: the channel
806 *
807 * Closes all channel buffers and frees the channel.
808 */
809void relay_close(struct rchan *chan)
810{
811 unsigned int i;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100812
813 if (!chan)
814 return;
815
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800816 mutex_lock(&relay_channels_mutex);
817 if (chan->is_global && chan->buf[0])
818 relay_close_buf(chan->buf[0]);
819 else
820 for_each_possible_cpu(i)
821 if (chan->buf[i])
822 relay_close_buf(chan->buf[i]);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100823
824 if (chan->last_toobig)
825 printk(KERN_WARNING "relay: one or more items not logged "
826 "[item size (%Zd) > sub-buffer size (%Zd)]\n",
827 chan->last_toobig, chan->subbuf_size);
828
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800829 list_del(&chan->list);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100830 kref_put(&chan->kref, relay_destroy_channel);
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800831 mutex_unlock(&relay_channels_mutex);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100832}
833EXPORT_SYMBOL_GPL(relay_close);
834
835/**
836 * relay_flush - close the channel
837 * @chan: the channel
838 *
Randy Dunlap4c78a662006-09-29 01:59:10 -0700839 * Flushes all channel buffers, i.e. forces buffer switch.
Jens Axboeb86ff9812006-03-23 19:56:55 +0100840 */
841void relay_flush(struct rchan *chan)
842{
843 unsigned int i;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100844
845 if (!chan)
846 return;
847
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800848 if (chan->is_global && chan->buf[0]) {
849 relay_switch_subbuf(chan->buf[0], 0);
850 return;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100851 }
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800852
853 mutex_lock(&relay_channels_mutex);
854 for_each_possible_cpu(i)
855 if (chan->buf[i])
856 relay_switch_subbuf(chan->buf[i], 0);
857 mutex_unlock(&relay_channels_mutex);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100858}
859EXPORT_SYMBOL_GPL(relay_flush);
860
861/**
862 * relay_file_open - open file op for relay files
863 * @inode: the inode
864 * @filp: the file
865 *
866 * Increments the channel buffer refcount.
867 */
868static int relay_file_open(struct inode *inode, struct file *filp)
869{
Theodore Ts'o8e18e292006-09-27 01:50:46 -0700870 struct rchan_buf *buf = inode->i_private;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100871 kref_get(&buf->kref);
872 filp->private_data = buf;
873
Lai Jiangshan37529fe2008-03-26 12:01:28 +0100874 return nonseekable_open(inode, filp);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100875}
876
877/**
878 * relay_file_mmap - mmap file op for relay files
879 * @filp: the file
880 * @vma: the vma describing what to map
881 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800882 * Calls upon relay_mmap_buf() to map the file into user space.
Jens Axboeb86ff9812006-03-23 19:56:55 +0100883 */
884static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
885{
886 struct rchan_buf *buf = filp->private_data;
887 return relay_mmap_buf(buf, vma);
888}
889
890/**
891 * relay_file_poll - poll file op for relay files
892 * @filp: the file
893 * @wait: poll table
894 *
895 * Poll implemention.
896 */
897static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
898{
899 unsigned int mask = 0;
900 struct rchan_buf *buf = filp->private_data;
901
902 if (buf->finalized)
903 return POLLERR;
904
905 if (filp->f_mode & FMODE_READ) {
906 poll_wait(filp, &buf->read_wait, wait);
907 if (!relay_buf_empty(buf))
908 mask |= POLLIN | POLLRDNORM;
909 }
910
911 return mask;
912}
913
914/**
915 * relay_file_release - release file op for relay files
916 * @inode: the inode
917 * @filp: the file
918 *
919 * Decrements the channel refcount, as the filesystem is
920 * no longer using it.
921 */
922static int relay_file_release(struct inode *inode, struct file *filp)
923{
924 struct rchan_buf *buf = filp->private_data;
925 kref_put(&buf->kref, relay_remove_buf);
926
927 return 0;
928}
929
Randy Dunlap4c78a662006-09-29 01:59:10 -0700930/*
Jens Axboeb86ff9812006-03-23 19:56:55 +0100931 * relay_file_read_consume - update the consumed count for the buffer
932 */
933static void relay_file_read_consume(struct rchan_buf *buf,
934 size_t read_pos,
935 size_t bytes_consumed)
936{
937 size_t subbuf_size = buf->chan->subbuf_size;
938 size_t n_subbufs = buf->chan->n_subbufs;
939 size_t read_subbuf;
940
Tom Zanussi32194452008-08-05 13:01:10 -0700941 if (buf->subbufs_produced == buf->subbufs_consumed &&
942 buf->offset == buf->bytes_consumed)
943 return;
944
Jens Axboeb86ff9812006-03-23 19:56:55 +0100945 if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
946 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
947 buf->bytes_consumed = 0;
948 }
949
950 buf->bytes_consumed += bytes_consumed;
Masami Hiramatsua66e3562007-06-27 14:10:04 -0700951 if (!read_pos)
952 read_subbuf = buf->subbufs_consumed % n_subbufs;
953 else
954 read_subbuf = read_pos / buf->chan->subbuf_size;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100955 if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
956 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
957 (buf->offset == subbuf_size))
958 return;
959 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
960 buf->bytes_consumed = 0;
961 }
962}
963
Randy Dunlap4c78a662006-09-29 01:59:10 -0700964/*
Jens Axboeb86ff9812006-03-23 19:56:55 +0100965 * relay_file_read_avail - boolean, are there unconsumed bytes available?
966 */
967static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
968{
Jens Axboeb86ff9812006-03-23 19:56:55 +0100969 size_t subbuf_size = buf->chan->subbuf_size;
970 size_t n_subbufs = buf->chan->n_subbufs;
Jens Axboe221415d2006-03-23 19:57:55 +0100971 size_t produced = buf->subbufs_produced;
972 size_t consumed = buf->subbufs_consumed;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100973
974 relay_file_read_consume(buf, read_pos, 0);
975
Tom Zanussi32194452008-08-05 13:01:10 -0700976 consumed = buf->subbufs_consumed;
977
Jens Axboe221415d2006-03-23 19:57:55 +0100978 if (unlikely(buf->offset > subbuf_size)) {
979 if (produced == consumed)
980 return 0;
981 return 1;
982 }
983
984 if (unlikely(produced - consumed >= n_subbufs)) {
Masami Hiramatsua66e3562007-06-27 14:10:04 -0700985 consumed = produced - n_subbufs + 1;
Jens Axboe221415d2006-03-23 19:57:55 +0100986 buf->subbufs_consumed = consumed;
Masami Hiramatsua66e3562007-06-27 14:10:04 -0700987 buf->bytes_consumed = 0;
Jens Axboe221415d2006-03-23 19:57:55 +0100988 }
Daniel Walker1bfbc602007-10-18 03:06:05 -0700989
Jens Axboe221415d2006-03-23 19:57:55 +0100990 produced = (produced % n_subbufs) * subbuf_size + buf->offset;
991 consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
992
993 if (consumed > produced)
994 produced += n_subbufs * subbuf_size;
Daniel Walker1bfbc602007-10-18 03:06:05 -0700995
Tom Zanussi32194452008-08-05 13:01:10 -0700996 if (consumed == produced) {
997 if (buf->offset == subbuf_size &&
998 buf->subbufs_produced > buf->subbufs_consumed)
999 return 1;
Jens Axboe221415d2006-03-23 19:57:55 +01001000 return 0;
Tom Zanussi32194452008-08-05 13:01:10 -07001001 }
Jens Axboe221415d2006-03-23 19:57:55 +01001002
Jens Axboeb86ff9812006-03-23 19:56:55 +01001003 return 1;
1004}
1005
1006/**
1007 * relay_file_read_subbuf_avail - return bytes available in sub-buffer
Randy Dunlap4c78a662006-09-29 01:59:10 -07001008 * @read_pos: file read position
1009 * @buf: relay channel buffer
Jens Axboeb86ff9812006-03-23 19:56:55 +01001010 */
1011static size_t relay_file_read_subbuf_avail(size_t read_pos,
1012 struct rchan_buf *buf)
1013{
1014 size_t padding, avail = 0;
1015 size_t read_subbuf, read_offset, write_subbuf, write_offset;
1016 size_t subbuf_size = buf->chan->subbuf_size;
1017
1018 write_subbuf = (buf->data - buf->start) / subbuf_size;
1019 write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
1020 read_subbuf = read_pos / subbuf_size;
1021 read_offset = read_pos % subbuf_size;
1022 padding = buf->padding[read_subbuf];
1023
1024 if (read_subbuf == write_subbuf) {
1025 if (read_offset + padding < write_offset)
1026 avail = write_offset - (read_offset + padding);
1027 } else
1028 avail = (subbuf_size - padding) - read_offset;
1029
1030 return avail;
1031}
1032
1033/**
1034 * relay_file_read_start_pos - find the first available byte to read
Randy Dunlap4c78a662006-09-29 01:59:10 -07001035 * @read_pos: file read position
1036 * @buf: relay channel buffer
Jens Axboeb86ff9812006-03-23 19:56:55 +01001037 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08001038 * If the @read_pos is in the middle of padding, return the
Jens Axboeb86ff9812006-03-23 19:56:55 +01001039 * position of the first actually available byte, otherwise
1040 * return the original value.
1041 */
1042static size_t relay_file_read_start_pos(size_t read_pos,
1043 struct rchan_buf *buf)
1044{
1045 size_t read_subbuf, padding, padding_start, padding_end;
1046 size_t subbuf_size = buf->chan->subbuf_size;
1047 size_t n_subbufs = buf->chan->n_subbufs;
David Wilder8d62fde2007-06-27 14:10:03 -07001048 size_t consumed = buf->subbufs_consumed % n_subbufs;
Jens Axboeb86ff9812006-03-23 19:56:55 +01001049
David Wilder8d62fde2007-06-27 14:10:03 -07001050 if (!read_pos)
1051 read_pos = consumed * subbuf_size + buf->bytes_consumed;
Jens Axboeb86ff9812006-03-23 19:56:55 +01001052 read_subbuf = read_pos / subbuf_size;
1053 padding = buf->padding[read_subbuf];
1054 padding_start = (read_subbuf + 1) * subbuf_size - padding;
1055 padding_end = (read_subbuf + 1) * subbuf_size;
1056 if (read_pos >= padding_start && read_pos < padding_end) {
1057 read_subbuf = (read_subbuf + 1) % n_subbufs;
1058 read_pos = read_subbuf * subbuf_size;
1059 }
1060
1061 return read_pos;
1062}
1063
1064/**
1065 * relay_file_read_end_pos - return the new read position
Randy Dunlap4c78a662006-09-29 01:59:10 -07001066 * @read_pos: file read position
1067 * @buf: relay channel buffer
1068 * @count: number of bytes to be read
Jens Axboeb86ff9812006-03-23 19:56:55 +01001069 */
1070static size_t relay_file_read_end_pos(struct rchan_buf *buf,
1071 size_t read_pos,
1072 size_t count)
1073{
1074 size_t read_subbuf, padding, end_pos;
1075 size_t subbuf_size = buf->chan->subbuf_size;
1076 size_t n_subbufs = buf->chan->n_subbufs;
1077
1078 read_subbuf = read_pos / subbuf_size;
1079 padding = buf->padding[read_subbuf];
1080 if (read_pos % subbuf_size + count + padding == subbuf_size)
1081 end_pos = (read_subbuf + 1) * subbuf_size;
1082 else
1083 end_pos = read_pos + count;
1084 if (end_pos >= subbuf_size * n_subbufs)
1085 end_pos = 0;
1086
1087 return end_pos;
1088}
1089
Randy Dunlap4c78a662006-09-29 01:59:10 -07001090/*
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001091 * subbuf_read_actor - read up to one subbuf's worth of data
Jens Axboeb86ff9812006-03-23 19:56:55 +01001092 */
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001093static int subbuf_read_actor(size_t read_start,
1094 struct rchan_buf *buf,
1095 size_t avail,
1096 read_descriptor_t *desc,
1097 read_actor_t actor)
Jens Axboeb86ff9812006-03-23 19:56:55 +01001098{
Jens Axboeb86ff9812006-03-23 19:56:55 +01001099 void *from;
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001100 int ret = 0;
Jens Axboeb86ff9812006-03-23 19:56:55 +01001101
1102 from = buf->start + read_start;
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001103 ret = avail;
Al Viroba2397e2006-10-10 22:46:57 +01001104 if (copy_to_user(desc->arg.buf, from, avail)) {
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001105 desc->error = -EFAULT;
1106 ret = 0;
Jens Axboeb86ff9812006-03-23 19:56:55 +01001107 }
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001108 desc->arg.data += ret;
1109 desc->written += ret;
1110 desc->count -= ret;
1111
Jens Axboeb86ff9812006-03-23 19:56:55 +01001112 return ret;
1113}
1114
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001115typedef int (*subbuf_actor_t) (size_t read_start,
1116 struct rchan_buf *buf,
1117 size_t avail,
1118 read_descriptor_t *desc,
1119 read_actor_t actor);
1120
Randy Dunlap4c78a662006-09-29 01:59:10 -07001121/*
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001122 * relay_file_read_subbufs - read count bytes, bridging subbuf boundaries
1123 */
Andrew Morton192636a2006-12-22 01:11:30 -08001124static ssize_t relay_file_read_subbufs(struct file *filp, loff_t *ppos,
1125 subbuf_actor_t subbuf_actor,
1126 read_actor_t actor,
1127 read_descriptor_t *desc)
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001128{
1129 struct rchan_buf *buf = filp->private_data;
1130 size_t read_start, avail;
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001131 int ret;
1132
Al Viroba2397e2006-10-10 22:46:57 +01001133 if (!desc->count)
Jens Axboe221415d2006-03-23 19:57:55 +01001134 return 0;
1135
Josef "Jeff" Sipekf3a43f32006-12-08 02:36:43 -08001136 mutex_lock(&filp->f_path.dentry->d_inode->i_mutex);
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001137 do {
1138 if (!relay_file_read_avail(buf, *ppos))
Jens Axboe221415d2006-03-23 19:57:55 +01001139 break;
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001140
1141 read_start = relay_file_read_start_pos(*ppos, buf);
1142 avail = relay_file_read_subbuf_avail(read_start, buf);
1143 if (!avail)
1144 break;
1145
Al Viroba2397e2006-10-10 22:46:57 +01001146 avail = min(desc->count, avail);
1147 ret = subbuf_actor(read_start, buf, avail, desc, actor);
1148 if (desc->error < 0)
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001149 break;
1150
1151 if (ret) {
1152 relay_file_read_consume(buf, read_start, ret);
1153 *ppos = relay_file_read_end_pos(buf, read_start, ret);
Jens Axboe221415d2006-03-23 19:57:55 +01001154 }
Al Viroba2397e2006-10-10 22:46:57 +01001155 } while (desc->count && ret);
Josef "Jeff" Sipekf3a43f32006-12-08 02:36:43 -08001156 mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex);
Jens Axboe221415d2006-03-23 19:57:55 +01001157
Al Viroba2397e2006-10-10 22:46:57 +01001158 return desc->written;
Jens Axboe221415d2006-03-23 19:57:55 +01001159}
1160
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001161static ssize_t relay_file_read(struct file *filp,
1162 char __user *buffer,
1163 size_t count,
1164 loff_t *ppos)
1165{
Al Viroba2397e2006-10-10 22:46:57 +01001166 read_descriptor_t desc;
1167 desc.written = 0;
1168 desc.count = count;
1169 desc.arg.buf = buffer;
1170 desc.error = 0;
1171 return relay_file_read_subbufs(filp, ppos, subbuf_read_actor,
1172 NULL, &desc);
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001173}
1174
Jens Axboe1db60cf2007-06-12 08:43:46 +02001175static void relay_consume_bytes(struct rchan_buf *rbuf, int bytes_consumed)
1176{
1177 rbuf->bytes_consumed += bytes_consumed;
1178
1179 if (rbuf->bytes_consumed >= rbuf->chan->subbuf_size) {
1180 relay_subbufs_consumed(rbuf->chan, rbuf->cpu, 1);
1181 rbuf->bytes_consumed %= rbuf->chan->subbuf_size;
1182 }
1183}
1184
Tom Zanussiebf99092007-06-04 09:12:05 +02001185static void relay_pipe_buf_release(struct pipe_inode_info *pipe,
1186 struct pipe_buffer *buf)
Jens Axboe221415d2006-03-23 19:57:55 +01001187{
Tom Zanussiebf99092007-06-04 09:12:05 +02001188 struct rchan_buf *rbuf;
1189
1190 rbuf = (struct rchan_buf *)page_private(buf->page);
Jens Axboe1db60cf2007-06-12 08:43:46 +02001191 relay_consume_bytes(rbuf, buf->private);
Tom Zanussiebf99092007-06-04 09:12:05 +02001192}
1193
Alexey Dobriyan28dfef82009-12-15 16:46:48 -08001194static const struct pipe_buf_operations relay_pipe_buf_ops = {
Tom Zanussiebf99092007-06-04 09:12:05 +02001195 .can_merge = 0,
1196 .map = generic_pipe_buf_map,
1197 .unmap = generic_pipe_buf_unmap,
Jens Axboecac36bb02007-06-14 13:10:48 +02001198 .confirm = generic_pipe_buf_confirm,
Tom Zanussiebf99092007-06-04 09:12:05 +02001199 .release = relay_pipe_buf_release,
1200 .steal = generic_pipe_buf_steal,
1201 .get = generic_pipe_buf_get,
1202};
1203
Jens Axboe5eb7f9f2008-03-26 12:04:09 +01001204static void relay_page_release(struct splice_pipe_desc *spd, unsigned int i)
1205{
1206}
1207
Tom Zanussid3f35d982007-07-12 08:12:05 +02001208/*
Tom Zanussiebf99092007-06-04 09:12:05 +02001209 * subbuf_splice_actor - splice up to one subbuf's worth of data
1210 */
Dan Carpenter5f1664f2010-03-05 13:42:51 -08001211static ssize_t subbuf_splice_actor(struct file *in,
Tom Zanussiebf99092007-06-04 09:12:05 +02001212 loff_t *ppos,
1213 struct pipe_inode_info *pipe,
1214 size_t len,
1215 unsigned int flags,
1216 int *nonpad_ret)
1217{
Dan Carpenter5f1664f2010-03-05 13:42:51 -08001218 unsigned int pidx, poff, total_len, subbuf_pages, nr_pages;
Tom Zanussiebf99092007-06-04 09:12:05 +02001219 struct rchan_buf *rbuf = in->private_data;
1220 unsigned int subbuf_size = rbuf->chan->subbuf_size;
Tom Zanussi24da24d2007-07-12 08:12:04 +02001221 uint64_t pos = (uint64_t) *ppos;
1222 uint32_t alloc_size = (uint32_t) rbuf->chan->alloc_size;
1223 size_t read_start = (size_t) do_div(pos, alloc_size);
Tom Zanussiebf99092007-06-04 09:12:05 +02001224 size_t read_subbuf = read_start / subbuf_size;
1225 size_t padding = rbuf->padding[read_subbuf];
1226 size_t nonpad_end = read_subbuf * subbuf_size + subbuf_size - padding;
Jens Axboe35f3d142010-05-20 10:43:18 +02001227 struct page *pages[PIPE_DEF_BUFFERS];
1228 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe1db60cf2007-06-12 08:43:46 +02001229 struct splice_pipe_desc spd = {
1230 .pages = pages,
1231 .nr_pages = 0,
1232 .partial = partial,
1233 .flags = flags,
1234 .ops = &relay_pipe_buf_ops,
Jens Axboe5eb7f9f2008-03-26 12:04:09 +01001235 .spd_release = relay_page_release,
Jens Axboe1db60cf2007-06-12 08:43:46 +02001236 };
Dan Carpenter5f1664f2010-03-05 13:42:51 -08001237 ssize_t ret;
Tom Zanussiebf99092007-06-04 09:12:05 +02001238
1239 if (rbuf->subbufs_produced == rbuf->subbufs_consumed)
1240 return 0;
Jens Axboe35f3d142010-05-20 10:43:18 +02001241 if (splice_grow_spd(pipe, &spd))
1242 return -ENOMEM;
Tom Zanussiebf99092007-06-04 09:12:05 +02001243
Jens Axboe1db60cf2007-06-12 08:43:46 +02001244 /*
1245 * Adjust read len, if longer than what is available
1246 */
1247 if (len > (subbuf_size - read_start % subbuf_size))
1248 len = subbuf_size - read_start % subbuf_size;
Tom Zanussiebf99092007-06-04 09:12:05 +02001249
1250 subbuf_pages = rbuf->chan->alloc_size >> PAGE_SHIFT;
1251 pidx = (read_start / PAGE_SIZE) % subbuf_pages;
1252 poff = read_start & ~PAGE_MASK;
Jens Axboe35f3d142010-05-20 10:43:18 +02001253 nr_pages = min_t(unsigned int, subbuf_pages, pipe->buffers);
Tom Zanussiebf99092007-06-04 09:12:05 +02001254
Jens Axboe16d546692008-03-17 09:04:59 +01001255 for (total_len = 0; spd.nr_pages < nr_pages; spd.nr_pages++) {
Jens Axboe1db60cf2007-06-12 08:43:46 +02001256 unsigned int this_len, this_end, private;
1257 unsigned int cur_pos = read_start + total_len;
Tom Zanussiebf99092007-06-04 09:12:05 +02001258
Jens Axboe1db60cf2007-06-12 08:43:46 +02001259 if (!len)
1260 break;
1261
1262 this_len = min_t(unsigned long, len, PAGE_SIZE - poff);
1263 private = this_len;
1264
1265 spd.pages[spd.nr_pages] = rbuf->page_array[pidx];
1266 spd.partial[spd.nr_pages].offset = poff;
1267
1268 this_end = cur_pos + this_len;
1269 if (this_end >= nonpad_end) {
1270 this_len = nonpad_end - cur_pos;
1271 private = this_len + padding;
1272 }
1273 spd.partial[spd.nr_pages].len = this_len;
1274 spd.partial[spd.nr_pages].private = private;
1275
1276 len -= this_len;
1277 total_len += this_len;
1278 poff = 0;
1279 pidx = (pidx + 1) % subbuf_pages;
1280
1281 if (this_end >= nonpad_end) {
1282 spd.nr_pages++;
Tom Zanussiebf99092007-06-04 09:12:05 +02001283 break;
1284 }
Tom Zanussiebf99092007-06-04 09:12:05 +02001285 }
1286
Jens Axboe35f3d142010-05-20 10:43:18 +02001287 ret = 0;
Jens Axboe1db60cf2007-06-12 08:43:46 +02001288 if (!spd.nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +02001289 goto out;
Tom Zanussiebf99092007-06-04 09:12:05 +02001290
Jens Axboe1db60cf2007-06-12 08:43:46 +02001291 ret = *nonpad_ret = splice_to_pipe(pipe, &spd);
1292 if (ret < 0 || ret < total_len)
Jens Axboe35f3d142010-05-20 10:43:18 +02001293 goto out;
Tom Zanussiebf99092007-06-04 09:12:05 +02001294
Jens Axboe1db60cf2007-06-12 08:43:46 +02001295 if (read_start + ret == nonpad_end)
1296 ret += padding;
1297
Jens Axboe35f3d142010-05-20 10:43:18 +02001298out:
1299 splice_shrink_spd(pipe, &spd);
Jens Axboe1db60cf2007-06-12 08:43:46 +02001300 return ret;
Tom Zanussiebf99092007-06-04 09:12:05 +02001301}
1302
1303static ssize_t relay_file_splice_read(struct file *in,
1304 loff_t *ppos,
1305 struct pipe_inode_info *pipe,
1306 size_t len,
1307 unsigned int flags)
1308{
1309 ssize_t spliced;
1310 int ret;
1311 int nonpad_ret = 0;
1312
1313 ret = 0;
1314 spliced = 0;
1315
Tom Zanussia82c53a2008-05-09 13:28:36 +02001316 while (len && !spliced) {
Tom Zanussiebf99092007-06-04 09:12:05 +02001317 ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret);
1318 if (ret < 0)
1319 break;
1320 else if (!ret) {
Tom Zanussifbb5b7a2008-12-09 13:14:10 -08001321 if (flags & SPLICE_F_NONBLOCK)
Tom Zanussiebf99092007-06-04 09:12:05 +02001322 ret = -EAGAIN;
Tom Zanussifbb5b7a2008-12-09 13:14:10 -08001323 break;
Tom Zanussiebf99092007-06-04 09:12:05 +02001324 }
1325
1326 *ppos += ret;
1327 if (ret > len)
1328 len = 0;
1329 else
1330 len -= ret;
1331 spliced += nonpad_ret;
1332 nonpad_ret = 0;
1333 }
1334
1335 if (spliced)
1336 return spliced;
1337
1338 return ret;
Jens Axboe221415d2006-03-23 19:57:55 +01001339}
1340
Helge Deller15ad7cd2006-12-06 20:40:36 -08001341const struct file_operations relay_file_operations = {
Jens Axboeb86ff9812006-03-23 19:56:55 +01001342 .open = relay_file_open,
1343 .poll = relay_file_poll,
1344 .mmap = relay_file_mmap,
1345 .read = relay_file_read,
1346 .llseek = no_llseek,
1347 .release = relay_file_release,
Tom Zanussiebf99092007-06-04 09:12:05 +02001348 .splice_read = relay_file_splice_read,
Jens Axboeb86ff9812006-03-23 19:56:55 +01001349};
1350EXPORT_SYMBOL_GPL(relay_file_operations);
Mathieu Desnoyers23c88752007-02-10 01:45:05 -08001351
1352static __init int relay_init(void)
1353{
1354
1355 hotcpu_notifier(relay_hotcpu_callback, 0);
1356 return 0;
1357}
1358
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -07001359early_initcall(relay_init);