blob: 5034cb3a339fb145c96921afd748404256316e64 [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
David Rientjes79dc8f32018-02-21 14:45:32 -0800166 if (chan->n_subbufs > KMALLOC_MAX_SIZE / 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 }
Akash Goel017c59c2016-09-02 21:47:38 +0200217 *per_cpu_ptr(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
Peter Zijlstra26b56792016-10-11 13:54:33 -0700331 * @work: contains the channel buffer
Jens Axboeb86ff9812006-03-23 19:56:55 +0100332 *
Peter Zijlstra26b56792016-10-11 13:54:33 -0700333 * This is the function used to defer reader waking
Jens Axboeb86ff9812006-03-23 19:56:55 +0100334 */
Peter Zijlstra26b56792016-10-11 13:54:33 -0700335static void wakeup_readers(struct irq_work *work)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100336{
Peter Zijlstra26b56792016-10-11 13:54:33 -0700337 struct rchan_buf *buf;
338
339 buf = container_of(work, struct rchan_buf, wakeup_work);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100340 wake_up_interruptible(&buf->read_wait);
341}
342
343/**
344 * __relay_reset - reset a channel buffer
345 * @buf: the channel buffer
346 * @init: 1 if this is a first-time initialization
347 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800348 * See relay_reset() for description of effect.
Jens Axboeb86ff9812006-03-23 19:56:55 +0100349 */
Andrew Morton192636a2006-12-22 01:11:30 -0800350static void __relay_reset(struct rchan_buf *buf, unsigned int init)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100351{
352 size_t i;
353
354 if (init) {
355 init_waitqueue_head(&buf->read_wait);
356 kref_init(&buf->kref);
Peter Zijlstra26b56792016-10-11 13:54:33 -0700357 init_irq_work(&buf->wakeup_work, wakeup_readers);
358 } else {
359 irq_work_sync(&buf->wakeup_work);
360 }
Jens Axboeb86ff9812006-03-23 19:56:55 +0100361
362 buf->subbufs_produced = 0;
363 buf->subbufs_consumed = 0;
364 buf->bytes_consumed = 0;
365 buf->finalized = 0;
366 buf->data = buf->start;
367 buf->offset = 0;
368
369 for (i = 0; i < buf->chan->n_subbufs; i++)
370 buf->padding[i] = 0;
371
372 buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0);
373}
374
375/**
376 * relay_reset - reset the channel
377 * @chan: the channel
378 *
379 * This has the effect of erasing all data from all channel buffers
380 * and restarting the channel in its initial state. The buffers
381 * are not freed, so any mappings are still in effect.
382 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800383 * NOTE. Care should be taken that the channel isn't actually
Jens Axboeb86ff9812006-03-23 19:56:55 +0100384 * being used by anything when this call is made.
385 */
386void relay_reset(struct rchan *chan)
387{
Akash Goel017c59c2016-09-02 21:47:38 +0200388 struct rchan_buf *buf;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100389 unsigned int i;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100390
391 if (!chan)
392 return;
393
Akash Goel017c59c2016-09-02 21:47:38 +0200394 if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0))) {
395 __relay_reset(buf, 0);
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800396 return;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100397 }
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800398
399 mutex_lock(&relay_channels_mutex);
Lai Jiangshan98ba4032008-11-14 10:44:59 +0100400 for_each_possible_cpu(i)
Akash Goel017c59c2016-09-02 21:47:38 +0200401 if ((buf = *per_cpu_ptr(chan->buf, i)))
402 __relay_reset(buf, 0);
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800403 mutex_unlock(&relay_channels_mutex);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100404}
405EXPORT_SYMBOL_GPL(relay_reset);
406
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700407static inline void relay_set_buf_dentry(struct rchan_buf *buf,
408 struct dentry *dentry)
409{
410 buf->dentry = dentry;
David Howells7682c912015-03-17 22:26:16 +0000411 d_inode(buf->dentry)->i_size = buf->early_bytes;
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700412}
413
414static struct dentry *relay_create_buf_file(struct rchan *chan,
415 struct rchan_buf *buf,
416 unsigned int cpu)
417{
418 struct dentry *dentry;
419 char *tmpname;
420
421 tmpname = kzalloc(NAME_MAX + 1, GFP_KERNEL);
422 if (!tmpname)
423 return NULL;
424 snprintf(tmpname, NAME_MAX, "%s%d", chan->base_filename, cpu);
425
426 /* Create file in fs */
427 dentry = chan->cb->create_buf_file(tmpname, chan->parent,
428 S_IRUSR, buf,
429 &chan->is_global);
430
431 kfree(tmpname);
432
433 return dentry;
434}
435
Randy Dunlap4c78a662006-09-29 01:59:10 -0700436/*
Jens Axboeb86ff9812006-03-23 19:56:55 +0100437 * relay_open_buf - create a new relay channel buffer
438 *
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800439 * used by relay_open() and CPU hotplug.
Jens Axboeb86ff9812006-03-23 19:56:55 +0100440 */
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800441static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100442{
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800443 struct rchan_buf *buf = NULL;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100444 struct dentry *dentry;
445
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800446 if (chan->is_global)
Akash Goel017c59c2016-09-02 21:47:38 +0200447 return *per_cpu_ptr(chan->buf, 0);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100448
449 buf = relay_create_buf(chan);
450 if (!buf)
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700451 return NULL;
452
453 if (chan->has_base_filename) {
454 dentry = relay_create_buf_file(chan, buf, cpu);
455 if (!dentry)
456 goto free_buf;
457 relay_set_buf_dentry(buf, dentry);
Akash Goel59dbb2a2016-08-02 14:07:18 -0700458 } else {
459 /* Only retrieve global info, nothing more, nothing less */
460 dentry = chan->cb->create_buf_file(NULL, NULL,
461 S_IRUSR, buf,
462 &chan->is_global);
463 if (WARN_ON(dentry))
464 goto free_buf;
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700465 }
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800466
467 buf->cpu = cpu;
468 __relay_reset(buf, 1);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100469
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800470 if(chan->is_global) {
Akash Goel017c59c2016-09-02 21:47:38 +0200471 *per_cpu_ptr(chan->buf, 0) = buf;
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800472 buf->cpu = 0;
473 }
474
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700475 return buf;
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800476
477free_buf:
478 relay_destroy_buf(buf);
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700479 return NULL;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100480}
481
482/**
483 * relay_close_buf - close a channel buffer
484 * @buf: channel buffer
485 *
486 * Marks the buffer finalized and restores the default callbacks.
487 * The channel buffer and channel buffer data structure are then freed
488 * automatically when the last reference is given up.
489 */
Andrew Morton192636a2006-12-22 01:11:30 -0800490static void relay_close_buf(struct rchan_buf *buf)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100491{
492 buf->finalized = 1;
Peter Zijlstra26b56792016-10-11 13:54:33 -0700493 irq_work_sync(&buf->wakeup_work);
Dmitry Monakhovb8d4a5b2013-04-22 11:41:41 +0400494 buf->chan->cb->remove_buf_file(buf->dentry);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100495 kref_put(&buf->kref, relay_remove_buf);
496}
497
Andrew Morton192636a2006-12-22 01:11:30 -0800498static void setup_callbacks(struct rchan *chan,
Jens Axboeb86ff9812006-03-23 19:56:55 +0100499 struct rchan_callbacks *cb)
500{
501 if (!cb) {
502 chan->cb = &default_channel_callbacks;
503 return;
504 }
505
506 if (!cb->subbuf_start)
507 cb->subbuf_start = subbuf_start_default_callback;
508 if (!cb->buf_mapped)
509 cb->buf_mapped = buf_mapped_default_callback;
510 if (!cb->buf_unmapped)
511 cb->buf_unmapped = buf_unmapped_default_callback;
512 if (!cb->create_buf_file)
513 cb->create_buf_file = create_buf_file_default_callback;
514 if (!cb->remove_buf_file)
515 cb->remove_buf_file = remove_buf_file_default_callback;
516 chan->cb = cb;
517}
518
Richard Weinbergere6d49892016-08-18 14:57:17 +0200519int relay_prepare_cpu(unsigned int cpu)
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800520{
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800521 struct rchan *chan;
Akash Goel017c59c2016-09-02 21:47:38 +0200522 struct rchan_buf *buf;
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800523
Richard Weinbergere6d49892016-08-18 14:57:17 +0200524 mutex_lock(&relay_channels_mutex);
525 list_for_each_entry(chan, &relay_channels, list) {
526 if ((buf = *per_cpu_ptr(chan->buf, cpu)))
527 continue;
528 buf = relay_open_buf(chan, cpu);
529 if (!buf) {
530 pr_err("relay: cpu %d buffer creation failed\n", cpu);
531 mutex_unlock(&relay_channels_mutex);
532 return -ENOMEM;
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800533 }
Richard Weinbergere6d49892016-08-18 14:57:17 +0200534 *per_cpu_ptr(chan->buf, cpu) = buf;
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800535 }
Richard Weinbergere6d49892016-08-18 14:57:17 +0200536 mutex_unlock(&relay_channels_mutex);
537 return 0;
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800538}
539
540/**
Jens Axboeb86ff9812006-03-23 19:56:55 +0100541 * relay_open - create a new relay channel
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700542 * @base_filename: base name of files to create, %NULL for buffering only
543 * @parent: dentry of parent directory, %NULL for root directory or buffer
Jens Axboeb86ff9812006-03-23 19:56:55 +0100544 * @subbuf_size: size of sub-buffers
545 * @n_subbufs: number of sub-buffers
546 * @cb: client callback functions
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800547 * @private_data: user-defined data
Jens Axboeb86ff9812006-03-23 19:56:55 +0100548 *
Randy Dunlap4c78a662006-09-29 01:59:10 -0700549 * Returns channel pointer if successful, %NULL otherwise.
Jens Axboeb86ff9812006-03-23 19:56:55 +0100550 *
551 * Creates a channel buffer for each cpu using the sizes and
552 * attributes specified. The created channel buffer files
553 * will be named base_filename0...base_filenameN-1. File
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800554 * permissions will be %S_IRUSR.
Akash Goel59dbb2a2016-08-02 14:07:18 -0700555 *
556 * If opening a buffer (@parent = NULL) that you later wish to register
557 * in a filesystem, call relay_late_setup_files() once the @parent dentry
558 * is available.
Jens Axboeb86ff9812006-03-23 19:56:55 +0100559 */
560struct rchan *relay_open(const char *base_filename,
561 struct dentry *parent,
562 size_t subbuf_size,
563 size_t n_subbufs,
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800564 struct rchan_callbacks *cb,
565 void *private_data)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100566{
567 unsigned int i;
568 struct rchan *chan;
Akash Goel017c59c2016-09-02 21:47:38 +0200569 struct rchan_buf *buf;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100570
571 if (!(subbuf_size && n_subbufs))
572 return NULL;
Dan Carpenterf6302f12012-02-10 09:03:58 +0100573 if (subbuf_size > UINT_MAX / n_subbufs)
574 return NULL;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100575
Robert P. J. Daycd861282006-12-13 00:34:52 -0800576 chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100577 if (!chan)
578 return NULL;
579
Akash Goel017c59c2016-09-02 21:47:38 +0200580 chan->buf = alloc_percpu(struct rchan_buf *);
Daniel Axtensd1774b02020-06-04 16:51:27 -0700581 if (!chan->buf) {
582 kfree(chan);
583 return NULL;
584 }
585
Jens Axboeb86ff9812006-03-23 19:56:55 +0100586 chan->version = RELAYFS_CHANNEL_VERSION;
587 chan->n_subbufs = n_subbufs;
588 chan->subbuf_size = subbuf_size;
zhangwei(Jovi)a05342cb2013-04-30 15:28:41 -0700589 chan->alloc_size = PAGE_ALIGN(subbuf_size * n_subbufs);
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800590 chan->parent = parent;
591 chan->private_data = private_data;
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700592 if (base_filename) {
593 chan->has_base_filename = 1;
594 strlcpy(chan->base_filename, base_filename, NAME_MAX);
595 }
Jens Axboeb86ff9812006-03-23 19:56:55 +0100596 setup_callbacks(chan, cb);
597 kref_init(&chan->kref);
598
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800599 mutex_lock(&relay_channels_mutex);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100600 for_each_online_cpu(i) {
Akash Goel017c59c2016-09-02 21:47:38 +0200601 buf = relay_open_buf(chan, i);
602 if (!buf)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100603 goto free_bufs;
Akash Goel017c59c2016-09-02 21:47:38 +0200604 *per_cpu_ptr(chan->buf, i) = buf;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100605 }
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800606 list_add(&chan->list, &relay_channels);
607 mutex_unlock(&relay_channels_mutex);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100608
Jens Axboeb86ff9812006-03-23 19:56:55 +0100609 return chan;
610
611free_bufs:
Lai Jiangshan98ba4032008-11-14 10:44:59 +0100612 for_each_possible_cpu(i) {
Akash Goel017c59c2016-09-02 21:47:38 +0200613 if ((buf = *per_cpu_ptr(chan->buf, i)))
614 relay_close_buf(buf);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100615 }
Jens Axboeb86ff9812006-03-23 19:56:55 +0100616
Jens Axboeb86ff9812006-03-23 19:56:55 +0100617 kref_put(&chan->kref, relay_destroy_channel);
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800618 mutex_unlock(&relay_channels_mutex);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100619 return NULL;
620}
621EXPORT_SYMBOL_GPL(relay_open);
622
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700623struct rchan_percpu_buf_dispatcher {
624 struct rchan_buf *buf;
625 struct dentry *dentry;
626};
627
628/* Called in atomic context. */
629static void __relay_set_buf_dentry(void *info)
630{
631 struct rchan_percpu_buf_dispatcher *p = info;
632
633 relay_set_buf_dentry(p->buf, p->dentry);
634}
635
636/**
637 * relay_late_setup_files - triggers file creation
638 * @chan: channel to operate on
639 * @base_filename: base name of files to create
640 * @parent: dentry of parent directory, %NULL for root directory
641 *
642 * Returns 0 if successful, non-zero otherwise.
643 *
Akash Goel59dbb2a2016-08-02 14:07:18 -0700644 * Use to setup files for a previously buffer-only channel created
645 * by relay_open() with a NULL parent dentry.
646 *
647 * For example, this is useful for perfomring early tracing in kernel,
648 * before VFS is up and then exposing the early results once the dentry
649 * is available.
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700650 */
651int relay_late_setup_files(struct rchan *chan,
652 const char *base_filename,
653 struct dentry *parent)
654{
655 int err = 0;
656 unsigned int i, curr_cpu;
657 unsigned long flags;
658 struct dentry *dentry;
Akash Goel017c59c2016-09-02 21:47:38 +0200659 struct rchan_buf *buf;
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700660 struct rchan_percpu_buf_dispatcher disp;
661
662 if (!chan || !base_filename)
663 return -EINVAL;
664
665 strlcpy(chan->base_filename, base_filename, NAME_MAX);
666
667 mutex_lock(&relay_channels_mutex);
668 /* Is chan already set up? */
Jiri Slabyb786c6a2009-01-17 12:04:36 +0100669 if (unlikely(chan->has_base_filename)) {
670 mutex_unlock(&relay_channels_mutex);
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700671 return -EEXIST;
Jiri Slabyb786c6a2009-01-17 12:04:36 +0100672 }
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700673 chan->has_base_filename = 1;
674 chan->parent = parent;
Akash Goel59dbb2a2016-08-02 14:07:18 -0700675
676 if (chan->is_global) {
677 err = -EINVAL;
Akash Goel017c59c2016-09-02 21:47:38 +0200678 buf = *per_cpu_ptr(chan->buf, 0);
679 if (!WARN_ON_ONCE(!buf)) {
680 dentry = relay_create_buf_file(chan, buf, 0);
Akash Goel59dbb2a2016-08-02 14:07:18 -0700681 if (dentry && !WARN_ON_ONCE(!chan->is_global)) {
Akash Goel017c59c2016-09-02 21:47:38 +0200682 relay_set_buf_dentry(buf, dentry);
Akash Goel59dbb2a2016-08-02 14:07:18 -0700683 err = 0;
684 }
685 }
686 mutex_unlock(&relay_channels_mutex);
687 return err;
688 }
689
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700690 curr_cpu = get_cpu();
691 /*
692 * The CPU hotplug notifier ran before us and created buffers with
693 * no files associated. So it's safe to call relay_setup_buf_file()
694 * on all currently online CPUs.
695 */
696 for_each_online_cpu(i) {
Akash Goel017c59c2016-09-02 21:47:38 +0200697 buf = *per_cpu_ptr(chan->buf, i);
698 if (unlikely(!buf)) {
Ingo Molnar7a51cff2008-12-29 16:03:40 +0100699 WARN_ONCE(1, KERN_ERR "CPU has no buffer!\n");
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700700 err = -EINVAL;
701 break;
702 }
703
Akash Goel017c59c2016-09-02 21:47:38 +0200704 dentry = relay_create_buf_file(chan, buf, i);
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700705 if (unlikely(!dentry)) {
706 err = -EINVAL;
707 break;
708 }
709
710 if (curr_cpu == i) {
711 local_irq_save(flags);
Akash Goel017c59c2016-09-02 21:47:38 +0200712 relay_set_buf_dentry(buf, dentry);
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700713 local_irq_restore(flags);
714 } else {
Akash Goel017c59c2016-09-02 21:47:38 +0200715 disp.buf = buf;
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700716 disp.dentry = dentry;
717 smp_mb();
718 /* relay_channels_mutex must be held, so wait. */
719 err = smp_call_function_single(i,
720 __relay_set_buf_dentry,
721 &disp, 1);
722 }
723 if (unlikely(err))
724 break;
725 }
726 put_cpu();
727 mutex_unlock(&relay_channels_mutex);
728
729 return err;
730}
Akash Goel59dbb2a2016-08-02 14:07:18 -0700731EXPORT_SYMBOL_GPL(relay_late_setup_files);
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700732
Jens Axboeb86ff9812006-03-23 19:56:55 +0100733/**
734 * relay_switch_subbuf - switch to a new sub-buffer
735 * @buf: channel buffer
736 * @length: size of current event
737 *
738 * Returns either the length passed in or 0 if full.
739 *
740 * Performs sub-buffer-switch tasks such as invoking callbacks,
741 * updating padding counts, waking up readers, etc.
742 */
743size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
744{
745 void *old, *new;
746 size_t old_subbuf, new_subbuf;
747
748 if (unlikely(length > buf->chan->subbuf_size))
749 goto toobig;
750
751 if (buf->offset != buf->chan->subbuf_size + 1) {
752 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
753 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
754 buf->padding[old_subbuf] = buf->prev_padding;
755 buf->subbufs_produced++;
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700756 if (buf->dentry)
David Howells7682c912015-03-17 22:26:16 +0000757 d_inode(buf->dentry)->i_size +=
Eduard - Gabriel Munteanu20d8b672008-07-25 19:45:12 -0700758 buf->chan->subbuf_size -
759 buf->padding[old_subbuf];
760 else
761 buf->early_bytes += buf->chan->subbuf_size -
762 buf->padding[old_subbuf];
Jens Axboe221415d2006-03-23 19:57:55 +0100763 smp_mb();
Peter Zijlstra26b56792016-10-11 13:54:33 -0700764 if (waitqueue_active(&buf->read_wait)) {
Tom Zanussi7c9cb382007-05-09 02:34:01 -0700765 /*
766 * Calling wake_up_interruptible() from here
767 * will deadlock if we happen to be logging
768 * from the scheduler (trying to re-grab
769 * rq->lock), so defer it.
770 */
Peter Zijlstra26b56792016-10-11 13:54:33 -0700771 irq_work_queue(&buf->wakeup_work);
772 }
Jens Axboeb86ff9812006-03-23 19:56:55 +0100773 }
774
775 old = buf->data;
776 new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
777 new = buf->start + new_subbuf * buf->chan->subbuf_size;
778 buf->offset = 0;
779 if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
780 buf->offset = buf->chan->subbuf_size + 1;
781 return 0;
782 }
783 buf->data = new;
784 buf->padding[new_subbuf] = 0;
785
786 if (unlikely(length + buf->offset > buf->chan->subbuf_size))
787 goto toobig;
788
789 return length;
790
791toobig:
792 buf->chan->last_toobig = length;
793 return 0;
794}
795EXPORT_SYMBOL_GPL(relay_switch_subbuf);
796
797/**
798 * relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
799 * @chan: the channel
800 * @cpu: the cpu associated with the channel buffer to update
801 * @subbufs_consumed: number of sub-buffers to add to current buf's count
802 *
803 * Adds to the channel buffer's consumed sub-buffer count.
804 * subbufs_consumed should be the number of sub-buffers newly consumed,
805 * not the total consumed.
806 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800807 * NOTE. Kernel clients don't need to call this function if the channel
Jens Axboeb86ff9812006-03-23 19:56:55 +0100808 * mode is 'overwrite'.
809 */
810void relay_subbufs_consumed(struct rchan *chan,
811 unsigned int cpu,
812 size_t subbufs_consumed)
813{
814 struct rchan_buf *buf;
815
Dan Carpenterc4c11b42016-12-14 15:05:38 -0800816 if (!chan || cpu >= NR_CPUS)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100817 return;
818
Akash Goel017c59c2016-09-02 21:47:38 +0200819 buf = *per_cpu_ptr(chan->buf, cpu);
Dan Carpenterc4c11b42016-12-14 15:05:38 -0800820 if (!buf || subbufs_consumed > chan->n_subbufs)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100821 return;
822
Aravind Srinivasan2c53d912009-04-02 16:58:59 -0700823 if (subbufs_consumed > buf->subbufs_produced - buf->subbufs_consumed)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100824 buf->subbufs_consumed = buf->subbufs_produced;
Aravind Srinivasan2c53d912009-04-02 16:58:59 -0700825 else
826 buf->subbufs_consumed += subbufs_consumed;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100827}
828EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
829
830/**
831 * relay_close - close the channel
832 * @chan: the channel
833 *
834 * Closes all channel buffers and frees the channel.
835 */
836void relay_close(struct rchan *chan)
837{
Akash Goel017c59c2016-09-02 21:47:38 +0200838 struct rchan_buf *buf;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100839 unsigned int i;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100840
841 if (!chan)
842 return;
843
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800844 mutex_lock(&relay_channels_mutex);
Akash Goel017c59c2016-09-02 21:47:38 +0200845 if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0)))
846 relay_close_buf(buf);
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800847 else
848 for_each_possible_cpu(i)
Akash Goel017c59c2016-09-02 21:47:38 +0200849 if ((buf = *per_cpu_ptr(chan->buf, i)))
850 relay_close_buf(buf);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100851
852 if (chan->last_toobig)
853 printk(KERN_WARNING "relay: one or more items not logged "
854 "[item size (%Zd) > sub-buffer size (%Zd)]\n",
855 chan->last_toobig, chan->subbuf_size);
856
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800857 list_del(&chan->list);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100858 kref_put(&chan->kref, relay_destroy_channel);
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800859 mutex_unlock(&relay_channels_mutex);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100860}
861EXPORT_SYMBOL_GPL(relay_close);
862
863/**
864 * relay_flush - close the channel
865 * @chan: the channel
866 *
Randy Dunlap4c78a662006-09-29 01:59:10 -0700867 * Flushes all channel buffers, i.e. forces buffer switch.
Jens Axboeb86ff9812006-03-23 19:56:55 +0100868 */
869void relay_flush(struct rchan *chan)
870{
Akash Goel017c59c2016-09-02 21:47:38 +0200871 struct rchan_buf *buf;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100872 unsigned int i;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100873
874 if (!chan)
875 return;
876
Akash Goel017c59c2016-09-02 21:47:38 +0200877 if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0))) {
878 relay_switch_subbuf(buf, 0);
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800879 return;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100880 }
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800881
882 mutex_lock(&relay_channels_mutex);
883 for_each_possible_cpu(i)
Akash Goel017c59c2016-09-02 21:47:38 +0200884 if ((buf = *per_cpu_ptr(chan->buf, i)))
885 relay_switch_subbuf(buf, 0);
Mathieu Desnoyers23c88752007-02-10 01:45:05 -0800886 mutex_unlock(&relay_channels_mutex);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100887}
888EXPORT_SYMBOL_GPL(relay_flush);
889
890/**
891 * relay_file_open - open file op for relay files
892 * @inode: the inode
893 * @filp: the file
894 *
895 * Increments the channel buffer refcount.
896 */
897static int relay_file_open(struct inode *inode, struct file *filp)
898{
Theodore Ts'o8e18e292006-09-27 01:50:46 -0700899 struct rchan_buf *buf = inode->i_private;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100900 kref_get(&buf->kref);
901 filp->private_data = buf;
902
Lai Jiangshan37529fe2008-03-26 12:01:28 +0100903 return nonseekable_open(inode, filp);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100904}
905
906/**
907 * relay_file_mmap - mmap file op for relay files
908 * @filp: the file
909 * @vma: the vma describing what to map
910 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -0800911 * Calls upon relay_mmap_buf() to map the file into user space.
Jens Axboeb86ff9812006-03-23 19:56:55 +0100912 */
913static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
914{
915 struct rchan_buf *buf = filp->private_data;
916 return relay_mmap_buf(buf, vma);
917}
918
919/**
920 * relay_file_poll - poll file op for relay files
921 * @filp: the file
922 * @wait: poll table
923 *
924 * Poll implemention.
925 */
926static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
927{
928 unsigned int mask = 0;
929 struct rchan_buf *buf = filp->private_data;
930
931 if (buf->finalized)
932 return POLLERR;
933
934 if (filp->f_mode & FMODE_READ) {
935 poll_wait(filp, &buf->read_wait, wait);
936 if (!relay_buf_empty(buf))
937 mask |= POLLIN | POLLRDNORM;
938 }
939
940 return mask;
941}
942
943/**
944 * relay_file_release - release file op for relay files
945 * @inode: the inode
946 * @filp: the file
947 *
948 * Decrements the channel refcount, as the filesystem is
949 * no longer using it.
950 */
951static int relay_file_release(struct inode *inode, struct file *filp)
952{
953 struct rchan_buf *buf = filp->private_data;
954 kref_put(&buf->kref, relay_remove_buf);
955
956 return 0;
957}
958
Randy Dunlap4c78a662006-09-29 01:59:10 -0700959/*
Jens Axboeb86ff9812006-03-23 19:56:55 +0100960 * relay_file_read_consume - update the consumed count for the buffer
961 */
962static void relay_file_read_consume(struct rchan_buf *buf,
963 size_t read_pos,
964 size_t bytes_consumed)
965{
966 size_t subbuf_size = buf->chan->subbuf_size;
967 size_t n_subbufs = buf->chan->n_subbufs;
968 size_t read_subbuf;
969
Tom Zanussi32194452008-08-05 13:01:10 -0700970 if (buf->subbufs_produced == buf->subbufs_consumed &&
971 buf->offset == buf->bytes_consumed)
972 return;
973
Jens Axboeb86ff9812006-03-23 19:56:55 +0100974 if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
975 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
976 buf->bytes_consumed = 0;
977 }
978
979 buf->bytes_consumed += bytes_consumed;
Masami Hiramatsua66e3562007-06-27 14:10:04 -0700980 if (!read_pos)
981 read_subbuf = buf->subbufs_consumed % n_subbufs;
982 else
983 read_subbuf = read_pos / buf->chan->subbuf_size;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100984 if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
985 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
986 (buf->offset == subbuf_size))
987 return;
988 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
989 buf->bytes_consumed = 0;
990 }
991}
992
Randy Dunlap4c78a662006-09-29 01:59:10 -0700993/*
Jens Axboeb86ff9812006-03-23 19:56:55 +0100994 * relay_file_read_avail - boolean, are there unconsumed bytes available?
995 */
996static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
997{
Jens Axboeb86ff9812006-03-23 19:56:55 +0100998 size_t subbuf_size = buf->chan->subbuf_size;
999 size_t n_subbufs = buf->chan->n_subbufs;
Jens Axboe221415d2006-03-23 19:57:55 +01001000 size_t produced = buf->subbufs_produced;
1001 size_t consumed = buf->subbufs_consumed;
Jens Axboeb86ff9812006-03-23 19:56:55 +01001002
1003 relay_file_read_consume(buf, read_pos, 0);
1004
Tom Zanussi32194452008-08-05 13:01:10 -07001005 consumed = buf->subbufs_consumed;
1006
Jens Axboe221415d2006-03-23 19:57:55 +01001007 if (unlikely(buf->offset > subbuf_size)) {
1008 if (produced == consumed)
1009 return 0;
1010 return 1;
1011 }
1012
1013 if (unlikely(produced - consumed >= n_subbufs)) {
Masami Hiramatsua66e3562007-06-27 14:10:04 -07001014 consumed = produced - n_subbufs + 1;
Jens Axboe221415d2006-03-23 19:57:55 +01001015 buf->subbufs_consumed = consumed;
Masami Hiramatsua66e3562007-06-27 14:10:04 -07001016 buf->bytes_consumed = 0;
Jens Axboe221415d2006-03-23 19:57:55 +01001017 }
Daniel Walker1bfbc602007-10-18 03:06:05 -07001018
Jens Axboe221415d2006-03-23 19:57:55 +01001019 produced = (produced % n_subbufs) * subbuf_size + buf->offset;
1020 consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
1021
1022 if (consumed > produced)
1023 produced += n_subbufs * subbuf_size;
Daniel Walker1bfbc602007-10-18 03:06:05 -07001024
Tom Zanussi32194452008-08-05 13:01:10 -07001025 if (consumed == produced) {
1026 if (buf->offset == subbuf_size &&
1027 buf->subbufs_produced > buf->subbufs_consumed)
1028 return 1;
Jens Axboe221415d2006-03-23 19:57:55 +01001029 return 0;
Tom Zanussi32194452008-08-05 13:01:10 -07001030 }
Jens Axboe221415d2006-03-23 19:57:55 +01001031
Jens Axboeb86ff9812006-03-23 19:56:55 +01001032 return 1;
1033}
1034
1035/**
1036 * relay_file_read_subbuf_avail - return bytes available in sub-buffer
Randy Dunlap4c78a662006-09-29 01:59:10 -07001037 * @read_pos: file read position
1038 * @buf: relay channel buffer
Jens Axboeb86ff9812006-03-23 19:56:55 +01001039 */
1040static size_t relay_file_read_subbuf_avail(size_t read_pos,
1041 struct rchan_buf *buf)
1042{
1043 size_t padding, avail = 0;
1044 size_t read_subbuf, read_offset, write_subbuf, write_offset;
1045 size_t subbuf_size = buf->chan->subbuf_size;
1046
1047 write_subbuf = (buf->data - buf->start) / subbuf_size;
1048 write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
1049 read_subbuf = read_pos / subbuf_size;
1050 read_offset = read_pos % subbuf_size;
1051 padding = buf->padding[read_subbuf];
1052
1053 if (read_subbuf == write_subbuf) {
1054 if (read_offset + padding < write_offset)
1055 avail = write_offset - (read_offset + padding);
1056 } else
1057 avail = (subbuf_size - padding) - read_offset;
1058
1059 return avail;
1060}
1061
1062/**
1063 * relay_file_read_start_pos - find the first available byte to read
Randy Dunlap4c78a662006-09-29 01:59:10 -07001064 * @read_pos: file read position
1065 * @buf: relay channel buffer
Jens Axboeb86ff9812006-03-23 19:56:55 +01001066 *
Robert P. J. Day72fd4a32007-02-10 01:45:59 -08001067 * If the @read_pos is in the middle of padding, return the
Jens Axboeb86ff9812006-03-23 19:56:55 +01001068 * position of the first actually available byte, otherwise
1069 * return the original value.
1070 */
1071static size_t relay_file_read_start_pos(size_t read_pos,
1072 struct rchan_buf *buf)
1073{
1074 size_t read_subbuf, padding, padding_start, padding_end;
1075 size_t subbuf_size = buf->chan->subbuf_size;
1076 size_t n_subbufs = buf->chan->n_subbufs;
David Wilder8d62fde2007-06-27 14:10:03 -07001077 size_t consumed = buf->subbufs_consumed % n_subbufs;
Jens Axboeb86ff9812006-03-23 19:56:55 +01001078
David Wilder8d62fde2007-06-27 14:10:03 -07001079 if (!read_pos)
1080 read_pos = consumed * subbuf_size + buf->bytes_consumed;
Jens Axboeb86ff9812006-03-23 19:56:55 +01001081 read_subbuf = read_pos / subbuf_size;
1082 padding = buf->padding[read_subbuf];
1083 padding_start = (read_subbuf + 1) * subbuf_size - padding;
1084 padding_end = (read_subbuf + 1) * subbuf_size;
1085 if (read_pos >= padding_start && read_pos < padding_end) {
1086 read_subbuf = (read_subbuf + 1) % n_subbufs;
1087 read_pos = read_subbuf * subbuf_size;
1088 }
1089
1090 return read_pos;
1091}
1092
1093/**
1094 * relay_file_read_end_pos - return the new read position
Randy Dunlap4c78a662006-09-29 01:59:10 -07001095 * @read_pos: file read position
1096 * @buf: relay channel buffer
1097 * @count: number of bytes to be read
Jens Axboeb86ff9812006-03-23 19:56:55 +01001098 */
1099static size_t relay_file_read_end_pos(struct rchan_buf *buf,
1100 size_t read_pos,
1101 size_t count)
1102{
1103 size_t read_subbuf, padding, end_pos;
1104 size_t subbuf_size = buf->chan->subbuf_size;
1105 size_t n_subbufs = buf->chan->n_subbufs;
1106
1107 read_subbuf = read_pos / subbuf_size;
1108 padding = buf->padding[read_subbuf];
1109 if (read_pos % subbuf_size + count + padding == subbuf_size)
1110 end_pos = (read_subbuf + 1) * subbuf_size;
1111 else
1112 end_pos = read_pos + count;
1113 if (end_pos >= subbuf_size * n_subbufs)
1114 end_pos = 0;
1115
1116 return end_pos;
1117}
1118
Al Viroa7c22422016-09-25 22:52:02 -04001119static ssize_t relay_file_read(struct file *filp,
1120 char __user *buffer,
1121 size_t count,
1122 loff_t *ppos)
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001123{
1124 struct rchan_buf *buf = filp->private_data;
1125 size_t read_start, avail;
Al Viroa7c22422016-09-25 22:52:02 -04001126 size_t written = 0;
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001127 int ret;
1128
Al Viroa7c22422016-09-25 22:52:02 -04001129 if (!count)
Jens Axboe221415d2006-03-23 19:57:55 +01001130 return 0;
1131
Al Viro59551022016-01-22 15:40:57 -05001132 inode_lock(file_inode(filp));
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001133 do {
Al Viroa7c22422016-09-25 22:52:02 -04001134 void *from;
1135
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001136 if (!relay_file_read_avail(buf, *ppos))
Jens Axboe221415d2006-03-23 19:57:55 +01001137 break;
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001138
1139 read_start = relay_file_read_start_pos(*ppos, buf);
1140 avail = relay_file_read_subbuf_avail(read_start, buf);
1141 if (!avail)
1142 break;
1143
Al Viroa7c22422016-09-25 22:52:02 -04001144 avail = min(count, avail);
1145 from = buf->start + read_start;
1146 ret = avail;
1147 if (copy_to_user(buffer, from, avail))
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001148 break;
1149
Al Viroa7c22422016-09-25 22:52:02 -04001150 buffer += ret;
1151 written += ret;
1152 count -= ret;
1153
1154 relay_file_read_consume(buf, read_start, ret);
1155 *ppos = relay_file_read_end_pos(buf, read_start, ret);
1156 } while (count);
Al Viro59551022016-01-22 15:40:57 -05001157 inode_unlock(file_inode(filp));
Jens Axboe221415d2006-03-23 19:57:55 +01001158
Al Viroa7c22422016-09-25 22:52:02 -04001159 return written;
Tom Zanussi6dac40a2006-03-23 19:58:45 +01001160}
1161
Jens Axboe1db60cf2007-06-12 08:43:46 +02001162static void relay_consume_bytes(struct rchan_buf *rbuf, int bytes_consumed)
1163{
1164 rbuf->bytes_consumed += bytes_consumed;
1165
1166 if (rbuf->bytes_consumed >= rbuf->chan->subbuf_size) {
1167 relay_subbufs_consumed(rbuf->chan, rbuf->cpu, 1);
1168 rbuf->bytes_consumed %= rbuf->chan->subbuf_size;
1169 }
1170}
1171
Tom Zanussiebf99092007-06-04 09:12:05 +02001172static void relay_pipe_buf_release(struct pipe_inode_info *pipe,
1173 struct pipe_buffer *buf)
Jens Axboe221415d2006-03-23 19:57:55 +01001174{
Tom Zanussiebf99092007-06-04 09:12:05 +02001175 struct rchan_buf *rbuf;
1176
1177 rbuf = (struct rchan_buf *)page_private(buf->page);
Jens Axboe1db60cf2007-06-12 08:43:46 +02001178 relay_consume_bytes(rbuf, buf->private);
Tom Zanussiebf99092007-06-04 09:12:05 +02001179}
1180
Alexey Dobriyan28dfef82009-12-15 16:46:48 -08001181static const struct pipe_buf_operations relay_pipe_buf_ops = {
Tom Zanussiebf99092007-06-04 09:12:05 +02001182 .can_merge = 0,
Jens Axboecac36bb02007-06-14 13:10:48 +02001183 .confirm = generic_pipe_buf_confirm,
Tom Zanussiebf99092007-06-04 09:12:05 +02001184 .release = relay_pipe_buf_release,
1185 .steal = generic_pipe_buf_steal,
1186 .get = generic_pipe_buf_get,
1187};
1188
Jens Axboe5eb7f9f2008-03-26 12:04:09 +01001189static void relay_page_release(struct splice_pipe_desc *spd, unsigned int i)
1190{
1191}
1192
Tom Zanussid3f35d92007-07-12 08:12:05 +02001193/*
Tom Zanussiebf99092007-06-04 09:12:05 +02001194 * subbuf_splice_actor - splice up to one subbuf's worth of data
1195 */
Dan Carpenter5f1664f2010-03-05 13:42:51 -08001196static ssize_t subbuf_splice_actor(struct file *in,
Tom Zanussiebf99092007-06-04 09:12:05 +02001197 loff_t *ppos,
1198 struct pipe_inode_info *pipe,
1199 size_t len,
1200 unsigned int flags,
1201 int *nonpad_ret)
1202{
Dan Carpenter5f1664f2010-03-05 13:42:51 -08001203 unsigned int pidx, poff, total_len, subbuf_pages, nr_pages;
Tom Zanussiebf99092007-06-04 09:12:05 +02001204 struct rchan_buf *rbuf = in->private_data;
1205 unsigned int subbuf_size = rbuf->chan->subbuf_size;
Tom Zanussi24da24d2007-07-12 08:12:04 +02001206 uint64_t pos = (uint64_t) *ppos;
1207 uint32_t alloc_size = (uint32_t) rbuf->chan->alloc_size;
1208 size_t read_start = (size_t) do_div(pos, alloc_size);
Tom Zanussiebf99092007-06-04 09:12:05 +02001209 size_t read_subbuf = read_start / subbuf_size;
1210 size_t padding = rbuf->padding[read_subbuf];
1211 size_t nonpad_end = read_subbuf * subbuf_size + subbuf_size - padding;
Jens Axboe35f3d142010-05-20 10:43:18 +02001212 struct page *pages[PIPE_DEF_BUFFERS];
1213 struct partial_page partial[PIPE_DEF_BUFFERS];
Jens Axboe1db60cf2007-06-12 08:43:46 +02001214 struct splice_pipe_desc spd = {
1215 .pages = pages,
1216 .nr_pages = 0,
Eric Dumazet047fe362012-06-12 15:24:40 +02001217 .nr_pages_max = PIPE_DEF_BUFFERS,
Jens Axboe1db60cf2007-06-12 08:43:46 +02001218 .partial = partial,
1219 .flags = flags,
1220 .ops = &relay_pipe_buf_ops,
Jens Axboe5eb7f9f2008-03-26 12:04:09 +01001221 .spd_release = relay_page_release,
Jens Axboe1db60cf2007-06-12 08:43:46 +02001222 };
Dan Carpenter5f1664f2010-03-05 13:42:51 -08001223 ssize_t ret;
Tom Zanussiebf99092007-06-04 09:12:05 +02001224
1225 if (rbuf->subbufs_produced == rbuf->subbufs_consumed)
1226 return 0;
Jens Axboe35f3d142010-05-20 10:43:18 +02001227 if (splice_grow_spd(pipe, &spd))
1228 return -ENOMEM;
Tom Zanussiebf99092007-06-04 09:12:05 +02001229
Jens Axboe1db60cf2007-06-12 08:43:46 +02001230 /*
1231 * Adjust read len, if longer than what is available
1232 */
1233 if (len > (subbuf_size - read_start % subbuf_size))
1234 len = subbuf_size - read_start % subbuf_size;
Tom Zanussiebf99092007-06-04 09:12:05 +02001235
1236 subbuf_pages = rbuf->chan->alloc_size >> PAGE_SHIFT;
1237 pidx = (read_start / PAGE_SIZE) % subbuf_pages;
1238 poff = read_start & ~PAGE_MASK;
Al Viroa786c062014-04-11 12:01:03 -04001239 nr_pages = min_t(unsigned int, subbuf_pages, spd.nr_pages_max);
Tom Zanussiebf99092007-06-04 09:12:05 +02001240
Jens Axboe16d54662008-03-17 09:04:59 +01001241 for (total_len = 0; spd.nr_pages < nr_pages; spd.nr_pages++) {
Jens Axboe1db60cf2007-06-12 08:43:46 +02001242 unsigned int this_len, this_end, private;
1243 unsigned int cur_pos = read_start + total_len;
Tom Zanussiebf99092007-06-04 09:12:05 +02001244
Jens Axboe1db60cf2007-06-12 08:43:46 +02001245 if (!len)
1246 break;
1247
1248 this_len = min_t(unsigned long, len, PAGE_SIZE - poff);
1249 private = this_len;
1250
1251 spd.pages[spd.nr_pages] = rbuf->page_array[pidx];
1252 spd.partial[spd.nr_pages].offset = poff;
1253
1254 this_end = cur_pos + this_len;
1255 if (this_end >= nonpad_end) {
1256 this_len = nonpad_end - cur_pos;
1257 private = this_len + padding;
1258 }
1259 spd.partial[spd.nr_pages].len = this_len;
1260 spd.partial[spd.nr_pages].private = private;
1261
1262 len -= this_len;
1263 total_len += this_len;
1264 poff = 0;
1265 pidx = (pidx + 1) % subbuf_pages;
1266
1267 if (this_end >= nonpad_end) {
1268 spd.nr_pages++;
Tom Zanussiebf99092007-06-04 09:12:05 +02001269 break;
1270 }
Tom Zanussiebf99092007-06-04 09:12:05 +02001271 }
1272
Jens Axboe35f3d142010-05-20 10:43:18 +02001273 ret = 0;
Jens Axboe1db60cf2007-06-12 08:43:46 +02001274 if (!spd.nr_pages)
Jens Axboe35f3d142010-05-20 10:43:18 +02001275 goto out;
Tom Zanussiebf99092007-06-04 09:12:05 +02001276
Jens Axboe1db60cf2007-06-12 08:43:46 +02001277 ret = *nonpad_ret = splice_to_pipe(pipe, &spd);
1278 if (ret < 0 || ret < total_len)
Jens Axboe35f3d142010-05-20 10:43:18 +02001279 goto out;
Tom Zanussiebf99092007-06-04 09:12:05 +02001280
Jens Axboe1db60cf2007-06-12 08:43:46 +02001281 if (read_start + ret == nonpad_end)
1282 ret += padding;
1283
Jens Axboe35f3d142010-05-20 10:43:18 +02001284out:
Eric Dumazet047fe362012-06-12 15:24:40 +02001285 splice_shrink_spd(&spd);
1286 return ret;
Tom Zanussiebf99092007-06-04 09:12:05 +02001287}
1288
1289static ssize_t relay_file_splice_read(struct file *in,
1290 loff_t *ppos,
1291 struct pipe_inode_info *pipe,
1292 size_t len,
1293 unsigned int flags)
1294{
1295 ssize_t spliced;
1296 int ret;
1297 int nonpad_ret = 0;
1298
1299 ret = 0;
1300 spliced = 0;
1301
Tom Zanussia82c53a2008-05-09 13:28:36 +02001302 while (len && !spliced) {
Tom Zanussiebf99092007-06-04 09:12:05 +02001303 ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret);
1304 if (ret < 0)
1305 break;
1306 else if (!ret) {
Tom Zanussifbb5b7a2008-12-09 13:14:10 -08001307 if (flags & SPLICE_F_NONBLOCK)
Tom Zanussiebf99092007-06-04 09:12:05 +02001308 ret = -EAGAIN;
Tom Zanussifbb5b7a2008-12-09 13:14:10 -08001309 break;
Tom Zanussiebf99092007-06-04 09:12:05 +02001310 }
1311
1312 *ppos += ret;
1313 if (ret > len)
1314 len = 0;
1315 else
1316 len -= ret;
1317 spliced += nonpad_ret;
1318 nonpad_ret = 0;
1319 }
1320
1321 if (spliced)
1322 return spliced;
1323
1324 return ret;
Jens Axboe221415d2006-03-23 19:57:55 +01001325}
1326
Helge Deller15ad7cd2006-12-06 20:40:36 -08001327const struct file_operations relay_file_operations = {
Jens Axboeb86ff9812006-03-23 19:56:55 +01001328 .open = relay_file_open,
1329 .poll = relay_file_poll,
1330 .mmap = relay_file_mmap,
1331 .read = relay_file_read,
1332 .llseek = no_llseek,
1333 .release = relay_file_release,
Tom Zanussiebf99092007-06-04 09:12:05 +02001334 .splice_read = relay_file_splice_read,
Jens Axboeb86ff9812006-03-23 19:56:55 +01001335};
1336EXPORT_SYMBOL_GPL(relay_file_operations);