xref: /openbmc/linux/kernel/relay.c (revision 4f205687)
1 /*
2  * Public API and common code for kernel->userspace relay file support.
3  *
4  * See Documentation/filesystems/relay.txt for an overview.
5  *
6  * Copyright (C) 2002-2005 - Tom Zanussi (zanussi@us.ibm.com), IBM Corp
7  * Copyright (C) 1999-2005 - Karim Yaghmour (karim@opersys.com)
8  *
9  * Moved to kernel/relay.c by Paul Mundt, 2006.
10  * November 2006 - CPU hotplug support by Mathieu Desnoyers
11  * 	(mathieu.desnoyers@polymtl.ca)
12  *
13  * This file is released under the GPL.
14  */
15 #include <linux/errno.h>
16 #include <linux/stddef.h>
17 #include <linux/slab.h>
18 #include <linux/export.h>
19 #include <linux/string.h>
20 #include <linux/relay.h>
21 #include <linux/vmalloc.h>
22 #include <linux/mm.h>
23 #include <linux/cpu.h>
24 #include <linux/splice.h>
25 
26 /* list of open channels, for cpu hotplug */
27 static DEFINE_MUTEX(relay_channels_mutex);
28 static LIST_HEAD(relay_channels);
29 
30 /*
31  * close() vm_op implementation for relay file mapping.
32  */
33 static void relay_file_mmap_close(struct vm_area_struct *vma)
34 {
35 	struct rchan_buf *buf = vma->vm_private_data;
36 	buf->chan->cb->buf_unmapped(buf, vma->vm_file);
37 }
38 
39 /*
40  * fault() vm_op implementation for relay file mapping.
41  */
42 static int relay_buf_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
43 {
44 	struct page *page;
45 	struct rchan_buf *buf = vma->vm_private_data;
46 	pgoff_t pgoff = vmf->pgoff;
47 
48 	if (!buf)
49 		return VM_FAULT_OOM;
50 
51 	page = vmalloc_to_page(buf->start + (pgoff << PAGE_SHIFT));
52 	if (!page)
53 		return VM_FAULT_SIGBUS;
54 	get_page(page);
55 	vmf->page = page;
56 
57 	return 0;
58 }
59 
60 /*
61  * vm_ops for relay file mappings.
62  */
63 static const struct vm_operations_struct relay_file_mmap_ops = {
64 	.fault = relay_buf_fault,
65 	.close = relay_file_mmap_close,
66 };
67 
68 /*
69  * allocate an array of pointers of struct page
70  */
71 static struct page **relay_alloc_page_array(unsigned int n_pages)
72 {
73 	const size_t pa_size = n_pages * sizeof(struct page *);
74 	if (pa_size > PAGE_SIZE)
75 		return vzalloc(pa_size);
76 	return kzalloc(pa_size, GFP_KERNEL);
77 }
78 
79 /*
80  * free an array of pointers of struct page
81  */
82 static void relay_free_page_array(struct page **array)
83 {
84 	kvfree(array);
85 }
86 
87 /**
88  *	relay_mmap_buf: - mmap channel buffer to process address space
89  *	@buf: relay channel buffer
90  *	@vma: vm_area_struct describing memory to be mapped
91  *
92  *	Returns 0 if ok, negative on error
93  *
94  *	Caller should already have grabbed mmap_sem.
95  */
96 static int relay_mmap_buf(struct rchan_buf *buf, struct vm_area_struct *vma)
97 {
98 	unsigned long length = vma->vm_end - vma->vm_start;
99 	struct file *filp = vma->vm_file;
100 
101 	if (!buf)
102 		return -EBADF;
103 
104 	if (length != (unsigned long)buf->chan->alloc_size)
105 		return -EINVAL;
106 
107 	vma->vm_ops = &relay_file_mmap_ops;
108 	vma->vm_flags |= VM_DONTEXPAND;
109 	vma->vm_private_data = buf;
110 	buf->chan->cb->buf_mapped(buf, filp);
111 
112 	return 0;
113 }
114 
115 /**
116  *	relay_alloc_buf - allocate a channel buffer
117  *	@buf: the buffer struct
118  *	@size: total size of the buffer
119  *
120  *	Returns a pointer to the resulting buffer, %NULL if unsuccessful. The
121  *	passed in size will get page aligned, if it isn't already.
122  */
123 static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size)
124 {
125 	void *mem;
126 	unsigned int i, j, n_pages;
127 
128 	*size = PAGE_ALIGN(*size);
129 	n_pages = *size >> PAGE_SHIFT;
130 
131 	buf->page_array = relay_alloc_page_array(n_pages);
132 	if (!buf->page_array)
133 		return NULL;
134 
135 	for (i = 0; i < n_pages; i++) {
136 		buf->page_array[i] = alloc_page(GFP_KERNEL);
137 		if (unlikely(!buf->page_array[i]))
138 			goto depopulate;
139 		set_page_private(buf->page_array[i], (unsigned long)buf);
140 	}
141 	mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL);
142 	if (!mem)
143 		goto depopulate;
144 
145 	memset(mem, 0, *size);
146 	buf->page_count = n_pages;
147 	return mem;
148 
149 depopulate:
150 	for (j = 0; j < i; j++)
151 		__free_page(buf->page_array[j]);
152 	relay_free_page_array(buf->page_array);
153 	return NULL;
154 }
155 
156 /**
157  *	relay_create_buf - allocate and initialize a channel buffer
158  *	@chan: the relay channel
159  *
160  *	Returns channel buffer if successful, %NULL otherwise.
161  */
162 static struct rchan_buf *relay_create_buf(struct rchan *chan)
163 {
164 	struct rchan_buf *buf;
165 
166 	if (chan->n_subbufs > UINT_MAX / sizeof(size_t *))
167 		return NULL;
168 
169 	buf = kzalloc(sizeof(struct rchan_buf), GFP_KERNEL);
170 	if (!buf)
171 		return NULL;
172 	buf->padding = kmalloc(chan->n_subbufs * sizeof(size_t *), GFP_KERNEL);
173 	if (!buf->padding)
174 		goto free_buf;
175 
176 	buf->start = relay_alloc_buf(buf, &chan->alloc_size);
177 	if (!buf->start)
178 		goto free_buf;
179 
180 	buf->chan = chan;
181 	kref_get(&buf->chan->kref);
182 	return buf;
183 
184 free_buf:
185 	kfree(buf->padding);
186 	kfree(buf);
187 	return NULL;
188 }
189 
190 /**
191  *	relay_destroy_channel - free the channel struct
192  *	@kref: target kernel reference that contains the relay channel
193  *
194  *	Should only be called from kref_put().
195  */
196 static void relay_destroy_channel(struct kref *kref)
197 {
198 	struct rchan *chan = container_of(kref, struct rchan, kref);
199 	kfree(chan);
200 }
201 
202 /**
203  *	relay_destroy_buf - destroy an rchan_buf struct and associated buffer
204  *	@buf: the buffer struct
205  */
206 static void relay_destroy_buf(struct rchan_buf *buf)
207 {
208 	struct rchan *chan = buf->chan;
209 	unsigned int i;
210 
211 	if (likely(buf->start)) {
212 		vunmap(buf->start);
213 		for (i = 0; i < buf->page_count; i++)
214 			__free_page(buf->page_array[i]);
215 		relay_free_page_array(buf->page_array);
216 	}
217 	chan->buf[buf->cpu] = NULL;
218 	kfree(buf->padding);
219 	kfree(buf);
220 	kref_put(&chan->kref, relay_destroy_channel);
221 }
222 
223 /**
224  *	relay_remove_buf - remove a channel buffer
225  *	@kref: target kernel reference that contains the relay buffer
226  *
227  *	Removes the file from the filesystem, which also frees the
228  *	rchan_buf_struct and the channel buffer.  Should only be called from
229  *	kref_put().
230  */
231 static void relay_remove_buf(struct kref *kref)
232 {
233 	struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
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  */
243 static int relay_buf_empty(struct rchan_buf *buf)
244 {
245 	return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1;
246 }
247 
248 /**
249  *	relay_buf_full - boolean, is the channel buffer full?
250  *	@buf: channel buffer
251  *
252  *	Returns 1 if the buffer is full, 0 otherwise.
253  */
254 int relay_buf_full(struct rchan_buf *buf)
255 {
256 	size_t ready = buf->subbufs_produced - buf->subbufs_consumed;
257 	return (ready >= buf->chan->n_subbufs) ? 1 : 0;
258 }
259 EXPORT_SYMBOL_GPL(relay_buf_full);
260 
261 /*
262  * High-level relay kernel API and associated functions.
263  */
264 
265 /*
266  * rchan_callback implementations defining default channel behavior.  Used
267  * in place of corresponding NULL values in client callback struct.
268  */
269 
270 /*
271  * subbuf_start() default callback.  Does nothing.
272  */
273 static int subbuf_start_default_callback (struct rchan_buf *buf,
274 					  void *subbuf,
275 					  void *prev_subbuf,
276 					  size_t prev_padding)
277 {
278 	if (relay_buf_full(buf))
279 		return 0;
280 
281 	return 1;
282 }
283 
284 /*
285  * buf_mapped() default callback.  Does nothing.
286  */
287 static void buf_mapped_default_callback(struct rchan_buf *buf,
288 					struct file *filp)
289 {
290 }
291 
292 /*
293  * buf_unmapped() default callback.  Does nothing.
294  */
295 static void buf_unmapped_default_callback(struct rchan_buf *buf,
296 					  struct file *filp)
297 {
298 }
299 
300 /*
301  * create_buf_file_create() default callback.  Does nothing.
302  */
303 static struct dentry *create_buf_file_default_callback(const char *filename,
304 						       struct dentry *parent,
305 						       umode_t 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  */
315 static int remove_buf_file_default_callback(struct dentry *dentry)
316 {
317 	return -EINVAL;
318 }
319 
320 /* relay channel default callbacks */
321 static struct rchan_callbacks default_channel_callbacks = {
322 	.subbuf_start = subbuf_start_default_callback,
323 	.buf_mapped = buf_mapped_default_callback,
324 	.buf_unmapped = buf_unmapped_default_callback,
325 	.create_buf_file = create_buf_file_default_callback,
326 	.remove_buf_file = remove_buf_file_default_callback,
327 };
328 
329 /**
330  *	wakeup_readers - wake up readers waiting on a channel
331  *	@data: contains the channel buffer
332  *
333  *	This is the timer function used to defer reader waking.
334  */
335 static void wakeup_readers(unsigned long data)
336 {
337 	struct rchan_buf *buf = (struct rchan_buf *)data;
338 	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  *
346  *	See relay_reset() for description of effect.
347  */
348 static void __relay_reset(struct rchan_buf *buf, unsigned int init)
349 {
350 	size_t i;
351 
352 	if (init) {
353 		init_waitqueue_head(&buf->read_wait);
354 		kref_init(&buf->kref);
355 		setup_timer(&buf->timer, wakeup_readers, (unsigned long)buf);
356 	} else
357 		del_timer_sync(&buf->timer);
358 
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  *
380  *	NOTE. Care should be taken that the channel isn't actually
381  *	being used by anything when this call is made.
382  */
383 void relay_reset(struct rchan *chan)
384 {
385 	unsigned int i;
386 
387 	if (!chan)
388 		return;
389 
390 	if (chan->is_global && chan->buf[0]) {
391 		__relay_reset(chan->buf[0], 0);
392 		return;
393 	}
394 
395 	mutex_lock(&relay_channels_mutex);
396 	for_each_possible_cpu(i)
397 		if (chan->buf[i])
398 			__relay_reset(chan->buf[i], 0);
399 	mutex_unlock(&relay_channels_mutex);
400 }
401 EXPORT_SYMBOL_GPL(relay_reset);
402 
403 static inline void relay_set_buf_dentry(struct rchan_buf *buf,
404 					struct dentry *dentry)
405 {
406 	buf->dentry = dentry;
407 	d_inode(buf->dentry)->i_size = buf->early_bytes;
408 }
409 
410 static 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 
432 /*
433  *	relay_open_buf - create a new relay channel buffer
434  *
435  *	used by relay_open() and CPU hotplug.
436  */
437 static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu)
438 {
439  	struct rchan_buf *buf = NULL;
440 	struct dentry *dentry;
441 
442  	if (chan->is_global)
443 		return chan->buf[0];
444 
445 	buf = relay_create_buf(chan);
446 	if (!buf)
447 		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 	}
455 
456  	buf->cpu = cpu;
457  	__relay_reset(buf, 1);
458 
459  	if(chan->is_global) {
460  		chan->buf[0] = buf;
461  		buf->cpu = 0;
462   	}
463 
464 	return buf;
465 
466 free_buf:
467  	relay_destroy_buf(buf);
468 	return NULL;
469 }
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  */
479 static void relay_close_buf(struct rchan_buf *buf)
480 {
481 	buf->finalized = 1;
482 	del_timer_sync(&buf->timer);
483 	buf->chan->cb->remove_buf_file(buf->dentry);
484 	kref_put(&buf->kref, relay_remove_buf);
485 }
486 
487 static void setup_callbacks(struct rchan *chan,
488 				   struct rchan_callbacks *cb)
489 {
490 	if (!cb) {
491 		chan->cb = &default_channel_callbacks;
492 		return;
493 	}
494 
495 	if (!cb->subbuf_start)
496 		cb->subbuf_start = subbuf_start_default_callback;
497 	if (!cb->buf_mapped)
498 		cb->buf_mapped = buf_mapped_default_callback;
499 	if (!cb->buf_unmapped)
500 		cb->buf_unmapped = buf_unmapped_default_callback;
501 	if (!cb->create_buf_file)
502 		cb->create_buf_file = create_buf_file_default_callback;
503 	if (!cb->remove_buf_file)
504 		cb->remove_buf_file = remove_buf_file_default_callback;
505 	chan->cb = cb;
506 }
507 
508 /**
509  * 	relay_hotcpu_callback - CPU hotplug callback
510  * 	@nb: notifier block
511  * 	@action: hotplug action to take
512  * 	@hcpu: CPU number
513  *
514  * 	Returns the success/failure of the operation. (%NOTIFY_OK, %NOTIFY_BAD)
515  */
516 static int relay_hotcpu_callback(struct notifier_block *nb,
517 				unsigned long action,
518 				void *hcpu)
519 {
520 	unsigned int hotcpu = (unsigned long)hcpu;
521 	struct rchan *chan;
522 
523 	switch(action) {
524 	case CPU_UP_PREPARE:
525 	case CPU_UP_PREPARE_FROZEN:
526 		mutex_lock(&relay_channels_mutex);
527 		list_for_each_entry(chan, &relay_channels, list) {
528 			if (chan->buf[hotcpu])
529 				continue;
530 			chan->buf[hotcpu] = relay_open_buf(chan, hotcpu);
531 			if(!chan->buf[hotcpu]) {
532 				printk(KERN_ERR
533 					"relay_hotcpu_callback: cpu %d buffer "
534 					"creation failed\n", hotcpu);
535 				mutex_unlock(&relay_channels_mutex);
536 				return notifier_from_errno(-ENOMEM);
537 			}
538 		}
539 		mutex_unlock(&relay_channels_mutex);
540 		break;
541 	case CPU_DEAD:
542 	case CPU_DEAD_FROZEN:
543 		/* No need to flush the cpu : will be flushed upon
544 		 * final relay_flush() call. */
545 		break;
546 	}
547 	return NOTIFY_OK;
548 }
549 
550 /**
551  *	relay_open - create a new relay channel
552  *	@base_filename: base name of files to create, %NULL for buffering only
553  *	@parent: dentry of parent directory, %NULL for root directory or buffer
554  *	@subbuf_size: size of sub-buffers
555  *	@n_subbufs: number of sub-buffers
556  *	@cb: client callback functions
557  *	@private_data: user-defined data
558  *
559  *	Returns channel pointer if successful, %NULL otherwise.
560  *
561  *	Creates a channel buffer for each cpu using the sizes and
562  *	attributes specified.  The created channel buffer files
563  *	will be named base_filename0...base_filenameN-1.  File
564  *	permissions will be %S_IRUSR.
565  */
566 struct rchan *relay_open(const char *base_filename,
567 			 struct dentry *parent,
568 			 size_t subbuf_size,
569 			 size_t n_subbufs,
570 			 struct rchan_callbacks *cb,
571 			 void *private_data)
572 {
573 	unsigned int i;
574 	struct rchan *chan;
575 
576 	if (!(subbuf_size && n_subbufs))
577 		return NULL;
578 	if (subbuf_size > UINT_MAX / n_subbufs)
579 		return NULL;
580 
581 	chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
582 	if (!chan)
583 		return NULL;
584 
585 	chan->version = RELAYFS_CHANNEL_VERSION;
586 	chan->n_subbufs = n_subbufs;
587 	chan->subbuf_size = subbuf_size;
588 	chan->alloc_size = PAGE_ALIGN(subbuf_size * n_subbufs);
589 	chan->parent = parent;
590 	chan->private_data = private_data;
591 	if (base_filename) {
592 		chan->has_base_filename = 1;
593 		strlcpy(chan->base_filename, base_filename, NAME_MAX);
594 	}
595 	setup_callbacks(chan, cb);
596 	kref_init(&chan->kref);
597 
598 	mutex_lock(&relay_channels_mutex);
599 	for_each_online_cpu(i) {
600 		chan->buf[i] = relay_open_buf(chan, i);
601 		if (!chan->buf[i])
602 			goto free_bufs;
603 	}
604 	list_add(&chan->list, &relay_channels);
605 	mutex_unlock(&relay_channels_mutex);
606 
607 	return chan;
608 
609 free_bufs:
610 	for_each_possible_cpu(i) {
611 		if (chan->buf[i])
612 			relay_close_buf(chan->buf[i]);
613 	}
614 
615 	kref_put(&chan->kref, relay_destroy_channel);
616 	mutex_unlock(&relay_channels_mutex);
617 	kfree(chan);
618 	return NULL;
619 }
620 EXPORT_SYMBOL_GPL(relay_open);
621 
622 struct rchan_percpu_buf_dispatcher {
623 	struct rchan_buf *buf;
624 	struct dentry *dentry;
625 };
626 
627 /* Called in atomic context. */
628 static void __relay_set_buf_dentry(void *info)
629 {
630 	struct rchan_percpu_buf_dispatcher *p = info;
631 
632 	relay_set_buf_dentry(p->buf, p->dentry);
633 }
634 
635 /**
636  *	relay_late_setup_files - triggers file creation
637  *	@chan: channel to operate on
638  *	@base_filename: base name of files to create
639  *	@parent: dentry of parent directory, %NULL for root directory
640  *
641  *	Returns 0 if successful, non-zero otherwise.
642  *
643  *	Use to setup files for a previously buffer-only channel.
644  *	Useful to do early tracing in kernel, before VFS is up, for example.
645  */
646 int relay_late_setup_files(struct rchan *chan,
647 			   const char *base_filename,
648 			   struct dentry *parent)
649 {
650 	int err = 0;
651 	unsigned int i, curr_cpu;
652 	unsigned long flags;
653 	struct dentry *dentry;
654 	struct rchan_percpu_buf_dispatcher disp;
655 
656 	if (!chan || !base_filename)
657 		return -EINVAL;
658 
659 	strlcpy(chan->base_filename, base_filename, NAME_MAX);
660 
661 	mutex_lock(&relay_channels_mutex);
662 	/* Is chan already set up? */
663 	if (unlikely(chan->has_base_filename)) {
664 		mutex_unlock(&relay_channels_mutex);
665 		return -EEXIST;
666 	}
667 	chan->has_base_filename = 1;
668 	chan->parent = parent;
669 	curr_cpu = get_cpu();
670 	/*
671 	 * The CPU hotplug notifier ran before us and created buffers with
672 	 * no files associated. So it's safe to call relay_setup_buf_file()
673 	 * on all currently online CPUs.
674 	 */
675 	for_each_online_cpu(i) {
676 		if (unlikely(!chan->buf[i])) {
677 			WARN_ONCE(1, KERN_ERR "CPU has no buffer!\n");
678 			err = -EINVAL;
679 			break;
680 		}
681 
682 		dentry = relay_create_buf_file(chan, chan->buf[i], i);
683 		if (unlikely(!dentry)) {
684 			err = -EINVAL;
685 			break;
686 		}
687 
688 		if (curr_cpu == i) {
689 			local_irq_save(flags);
690 			relay_set_buf_dentry(chan->buf[i], dentry);
691 			local_irq_restore(flags);
692 		} else {
693 			disp.buf = chan->buf[i];
694 			disp.dentry = dentry;
695 			smp_mb();
696 			/* relay_channels_mutex must be held, so wait. */
697 			err = smp_call_function_single(i,
698 						       __relay_set_buf_dentry,
699 						       &disp, 1);
700 		}
701 		if (unlikely(err))
702 			break;
703 	}
704 	put_cpu();
705 	mutex_unlock(&relay_channels_mutex);
706 
707 	return err;
708 }
709 
710 /**
711  *	relay_switch_subbuf - switch to a new sub-buffer
712  *	@buf: channel buffer
713  *	@length: size of current event
714  *
715  *	Returns either the length passed in or 0 if full.
716  *
717  *	Performs sub-buffer-switch tasks such as invoking callbacks,
718  *	updating padding counts, waking up readers, etc.
719  */
720 size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
721 {
722 	void *old, *new;
723 	size_t old_subbuf, new_subbuf;
724 
725 	if (unlikely(length > buf->chan->subbuf_size))
726 		goto toobig;
727 
728 	if (buf->offset != buf->chan->subbuf_size + 1) {
729 		buf->prev_padding = buf->chan->subbuf_size - buf->offset;
730 		old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
731 		buf->padding[old_subbuf] = buf->prev_padding;
732 		buf->subbufs_produced++;
733 		if (buf->dentry)
734 			d_inode(buf->dentry)->i_size +=
735 				buf->chan->subbuf_size -
736 				buf->padding[old_subbuf];
737 		else
738 			buf->early_bytes += buf->chan->subbuf_size -
739 					    buf->padding[old_subbuf];
740 		smp_mb();
741 		if (waitqueue_active(&buf->read_wait))
742 			/*
743 			 * Calling wake_up_interruptible() from here
744 			 * will deadlock if we happen to be logging
745 			 * from the scheduler (trying to re-grab
746 			 * rq->lock), so defer it.
747 			 */
748 			mod_timer(&buf->timer, jiffies + 1);
749 	}
750 
751 	old = buf->data;
752 	new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
753 	new = buf->start + new_subbuf * buf->chan->subbuf_size;
754 	buf->offset = 0;
755 	if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
756 		buf->offset = buf->chan->subbuf_size + 1;
757 		return 0;
758 	}
759 	buf->data = new;
760 	buf->padding[new_subbuf] = 0;
761 
762 	if (unlikely(length + buf->offset > buf->chan->subbuf_size))
763 		goto toobig;
764 
765 	return length;
766 
767 toobig:
768 	buf->chan->last_toobig = length;
769 	return 0;
770 }
771 EXPORT_SYMBOL_GPL(relay_switch_subbuf);
772 
773 /**
774  *	relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
775  *	@chan: the channel
776  *	@cpu: the cpu associated with the channel buffer to update
777  *	@subbufs_consumed: number of sub-buffers to add to current buf's count
778  *
779  *	Adds to the channel buffer's consumed sub-buffer count.
780  *	subbufs_consumed should be the number of sub-buffers newly consumed,
781  *	not the total consumed.
782  *
783  *	NOTE. Kernel clients don't need to call this function if the channel
784  *	mode is 'overwrite'.
785  */
786 void relay_subbufs_consumed(struct rchan *chan,
787 			    unsigned int cpu,
788 			    size_t subbufs_consumed)
789 {
790 	struct rchan_buf *buf;
791 
792 	if (!chan)
793 		return;
794 
795 	if (cpu >= NR_CPUS || !chan->buf[cpu] ||
796 					subbufs_consumed > chan->n_subbufs)
797 		return;
798 
799 	buf = chan->buf[cpu];
800 	if (subbufs_consumed > buf->subbufs_produced - buf->subbufs_consumed)
801 		buf->subbufs_consumed = buf->subbufs_produced;
802 	else
803 		buf->subbufs_consumed += subbufs_consumed;
804 }
805 EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
806 
807 /**
808  *	relay_close - close the channel
809  *	@chan: the channel
810  *
811  *	Closes all channel buffers and frees the channel.
812  */
813 void relay_close(struct rchan *chan)
814 {
815 	unsigned int i;
816 
817 	if (!chan)
818 		return;
819 
820 	mutex_lock(&relay_channels_mutex);
821 	if (chan->is_global && chan->buf[0])
822 		relay_close_buf(chan->buf[0]);
823 	else
824 		for_each_possible_cpu(i)
825 			if (chan->buf[i])
826 				relay_close_buf(chan->buf[i]);
827 
828 	if (chan->last_toobig)
829 		printk(KERN_WARNING "relay: one or more items not logged "
830 		       "[item size (%Zd) > sub-buffer size (%Zd)]\n",
831 		       chan->last_toobig, chan->subbuf_size);
832 
833 	list_del(&chan->list);
834 	kref_put(&chan->kref, relay_destroy_channel);
835 	mutex_unlock(&relay_channels_mutex);
836 }
837 EXPORT_SYMBOL_GPL(relay_close);
838 
839 /**
840  *	relay_flush - close the channel
841  *	@chan: the channel
842  *
843  *	Flushes all channel buffers, i.e. forces buffer switch.
844  */
845 void relay_flush(struct rchan *chan)
846 {
847 	unsigned int i;
848 
849 	if (!chan)
850 		return;
851 
852 	if (chan->is_global && chan->buf[0]) {
853 		relay_switch_subbuf(chan->buf[0], 0);
854 		return;
855 	}
856 
857 	mutex_lock(&relay_channels_mutex);
858 	for_each_possible_cpu(i)
859 		if (chan->buf[i])
860 			relay_switch_subbuf(chan->buf[i], 0);
861 	mutex_unlock(&relay_channels_mutex);
862 }
863 EXPORT_SYMBOL_GPL(relay_flush);
864 
865 /**
866  *	relay_file_open - open file op for relay files
867  *	@inode: the inode
868  *	@filp: the file
869  *
870  *	Increments the channel buffer refcount.
871  */
872 static int relay_file_open(struct inode *inode, struct file *filp)
873 {
874 	struct rchan_buf *buf = inode->i_private;
875 	kref_get(&buf->kref);
876 	filp->private_data = buf;
877 
878 	return nonseekable_open(inode, filp);
879 }
880 
881 /**
882  *	relay_file_mmap - mmap file op for relay files
883  *	@filp: the file
884  *	@vma: the vma describing what to map
885  *
886  *	Calls upon relay_mmap_buf() to map the file into user space.
887  */
888 static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
889 {
890 	struct rchan_buf *buf = filp->private_data;
891 	return relay_mmap_buf(buf, vma);
892 }
893 
894 /**
895  *	relay_file_poll - poll file op for relay files
896  *	@filp: the file
897  *	@wait: poll table
898  *
899  *	Poll implemention.
900  */
901 static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
902 {
903 	unsigned int mask = 0;
904 	struct rchan_buf *buf = filp->private_data;
905 
906 	if (buf->finalized)
907 		return POLLERR;
908 
909 	if (filp->f_mode & FMODE_READ) {
910 		poll_wait(filp, &buf->read_wait, wait);
911 		if (!relay_buf_empty(buf))
912 			mask |= POLLIN | POLLRDNORM;
913 	}
914 
915 	return mask;
916 }
917 
918 /**
919  *	relay_file_release - release file op for relay files
920  *	@inode: the inode
921  *	@filp: the file
922  *
923  *	Decrements the channel refcount, as the filesystem is
924  *	no longer using it.
925  */
926 static int relay_file_release(struct inode *inode, struct file *filp)
927 {
928 	struct rchan_buf *buf = filp->private_data;
929 	kref_put(&buf->kref, relay_remove_buf);
930 
931 	return 0;
932 }
933 
934 /*
935  *	relay_file_read_consume - update the consumed count for the buffer
936  */
937 static void relay_file_read_consume(struct rchan_buf *buf,
938 				    size_t read_pos,
939 				    size_t bytes_consumed)
940 {
941 	size_t subbuf_size = buf->chan->subbuf_size;
942 	size_t n_subbufs = buf->chan->n_subbufs;
943 	size_t read_subbuf;
944 
945 	if (buf->subbufs_produced == buf->subbufs_consumed &&
946 	    buf->offset == buf->bytes_consumed)
947 		return;
948 
949 	if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
950 		relay_subbufs_consumed(buf->chan, buf->cpu, 1);
951 		buf->bytes_consumed = 0;
952 	}
953 
954 	buf->bytes_consumed += bytes_consumed;
955 	if (!read_pos)
956 		read_subbuf = buf->subbufs_consumed % n_subbufs;
957 	else
958 		read_subbuf = read_pos / buf->chan->subbuf_size;
959 	if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
960 		if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
961 		    (buf->offset == subbuf_size))
962 			return;
963 		relay_subbufs_consumed(buf->chan, buf->cpu, 1);
964 		buf->bytes_consumed = 0;
965 	}
966 }
967 
968 /*
969  *	relay_file_read_avail - boolean, are there unconsumed bytes available?
970  */
971 static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
972 {
973 	size_t subbuf_size = buf->chan->subbuf_size;
974 	size_t n_subbufs = buf->chan->n_subbufs;
975 	size_t produced = buf->subbufs_produced;
976 	size_t consumed = buf->subbufs_consumed;
977 
978 	relay_file_read_consume(buf, read_pos, 0);
979 
980 	consumed = buf->subbufs_consumed;
981 
982 	if (unlikely(buf->offset > subbuf_size)) {
983 		if (produced == consumed)
984 			return 0;
985 		return 1;
986 	}
987 
988 	if (unlikely(produced - consumed >= n_subbufs)) {
989 		consumed = produced - n_subbufs + 1;
990 		buf->subbufs_consumed = consumed;
991 		buf->bytes_consumed = 0;
992 	}
993 
994 	produced = (produced % n_subbufs) * subbuf_size + buf->offset;
995 	consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
996 
997 	if (consumed > produced)
998 		produced += n_subbufs * subbuf_size;
999 
1000 	if (consumed == produced) {
1001 		if (buf->offset == subbuf_size &&
1002 		    buf->subbufs_produced > buf->subbufs_consumed)
1003 			return 1;
1004 		return 0;
1005 	}
1006 
1007 	return 1;
1008 }
1009 
1010 /**
1011  *	relay_file_read_subbuf_avail - return bytes available in sub-buffer
1012  *	@read_pos: file read position
1013  *	@buf: relay channel buffer
1014  */
1015 static size_t relay_file_read_subbuf_avail(size_t read_pos,
1016 					   struct rchan_buf *buf)
1017 {
1018 	size_t padding, avail = 0;
1019 	size_t read_subbuf, read_offset, write_subbuf, write_offset;
1020 	size_t subbuf_size = buf->chan->subbuf_size;
1021 
1022 	write_subbuf = (buf->data - buf->start) / subbuf_size;
1023 	write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
1024 	read_subbuf = read_pos / subbuf_size;
1025 	read_offset = read_pos % subbuf_size;
1026 	padding = buf->padding[read_subbuf];
1027 
1028 	if (read_subbuf == write_subbuf) {
1029 		if (read_offset + padding < write_offset)
1030 			avail = write_offset - (read_offset + padding);
1031 	} else
1032 		avail = (subbuf_size - padding) - read_offset;
1033 
1034 	return avail;
1035 }
1036 
1037 /**
1038  *	relay_file_read_start_pos - find the first available byte to read
1039  *	@read_pos: file read position
1040  *	@buf: relay channel buffer
1041  *
1042  *	If the @read_pos is in the middle of padding, return the
1043  *	position of the first actually available byte, otherwise
1044  *	return the original value.
1045  */
1046 static size_t relay_file_read_start_pos(size_t read_pos,
1047 					struct rchan_buf *buf)
1048 {
1049 	size_t read_subbuf, padding, padding_start, padding_end;
1050 	size_t subbuf_size = buf->chan->subbuf_size;
1051 	size_t n_subbufs = buf->chan->n_subbufs;
1052 	size_t consumed = buf->subbufs_consumed % n_subbufs;
1053 
1054 	if (!read_pos)
1055 		read_pos = consumed * subbuf_size + buf->bytes_consumed;
1056 	read_subbuf = read_pos / subbuf_size;
1057 	padding = buf->padding[read_subbuf];
1058 	padding_start = (read_subbuf + 1) * subbuf_size - padding;
1059 	padding_end = (read_subbuf + 1) * subbuf_size;
1060 	if (read_pos >= padding_start && read_pos < padding_end) {
1061 		read_subbuf = (read_subbuf + 1) % n_subbufs;
1062 		read_pos = read_subbuf * subbuf_size;
1063 	}
1064 
1065 	return read_pos;
1066 }
1067 
1068 /**
1069  *	relay_file_read_end_pos - return the new read position
1070  *	@read_pos: file read position
1071  *	@buf: relay channel buffer
1072  *	@count: number of bytes to be read
1073  */
1074 static size_t relay_file_read_end_pos(struct rchan_buf *buf,
1075 				      size_t read_pos,
1076 				      size_t count)
1077 {
1078 	size_t read_subbuf, padding, end_pos;
1079 	size_t subbuf_size = buf->chan->subbuf_size;
1080 	size_t n_subbufs = buf->chan->n_subbufs;
1081 
1082 	read_subbuf = read_pos / subbuf_size;
1083 	padding = buf->padding[read_subbuf];
1084 	if (read_pos % subbuf_size + count + padding == subbuf_size)
1085 		end_pos = (read_subbuf + 1) * subbuf_size;
1086 	else
1087 		end_pos = read_pos + count;
1088 	if (end_pos >= subbuf_size * n_subbufs)
1089 		end_pos = 0;
1090 
1091 	return end_pos;
1092 }
1093 
1094 /*
1095  *	subbuf_read_actor - read up to one subbuf's worth of data
1096  */
1097 static int subbuf_read_actor(size_t read_start,
1098 			     struct rchan_buf *buf,
1099 			     size_t avail,
1100 			     read_descriptor_t *desc)
1101 {
1102 	void *from;
1103 	int ret = 0;
1104 
1105 	from = buf->start + read_start;
1106 	ret = avail;
1107 	if (copy_to_user(desc->arg.buf, from, avail)) {
1108 		desc->error = -EFAULT;
1109 		ret = 0;
1110 	}
1111 	desc->arg.data += ret;
1112 	desc->written += ret;
1113 	desc->count -= ret;
1114 
1115 	return ret;
1116 }
1117 
1118 typedef int (*subbuf_actor_t) (size_t read_start,
1119 			       struct rchan_buf *buf,
1120 			       size_t avail,
1121 			       read_descriptor_t *desc);
1122 
1123 /*
1124  *	relay_file_read_subbufs - read count bytes, bridging subbuf boundaries
1125  */
1126 static ssize_t relay_file_read_subbufs(struct file *filp, loff_t *ppos,
1127 					subbuf_actor_t subbuf_actor,
1128 					read_descriptor_t *desc)
1129 {
1130 	struct rchan_buf *buf = filp->private_data;
1131 	size_t read_start, avail;
1132 	int ret;
1133 
1134 	if (!desc->count)
1135 		return 0;
1136 
1137 	inode_lock(file_inode(filp));
1138 	do {
1139 		if (!relay_file_read_avail(buf, *ppos))
1140 			break;
1141 
1142 		read_start = relay_file_read_start_pos(*ppos, buf);
1143 		avail = relay_file_read_subbuf_avail(read_start, buf);
1144 		if (!avail)
1145 			break;
1146 
1147 		avail = min(desc->count, avail);
1148 		ret = subbuf_actor(read_start, buf, avail, desc);
1149 		if (desc->error < 0)
1150 			break;
1151 
1152 		if (ret) {
1153 			relay_file_read_consume(buf, read_start, ret);
1154 			*ppos = relay_file_read_end_pos(buf, read_start, ret);
1155 		}
1156 	} while (desc->count && ret);
1157 	inode_unlock(file_inode(filp));
1158 
1159 	return desc->written;
1160 }
1161 
1162 static ssize_t relay_file_read(struct file *filp,
1163 			       char __user *buffer,
1164 			       size_t count,
1165 			       loff_t *ppos)
1166 {
1167 	read_descriptor_t desc;
1168 	desc.written = 0;
1169 	desc.count = count;
1170 	desc.arg.buf = buffer;
1171 	desc.error = 0;
1172 	return relay_file_read_subbufs(filp, ppos, subbuf_read_actor, &desc);
1173 }
1174 
1175 static 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 
1185 static void relay_pipe_buf_release(struct pipe_inode_info *pipe,
1186 				   struct pipe_buffer *buf)
1187 {
1188 	struct rchan_buf *rbuf;
1189 
1190 	rbuf = (struct rchan_buf *)page_private(buf->page);
1191 	relay_consume_bytes(rbuf, buf->private);
1192 }
1193 
1194 static const struct pipe_buf_operations relay_pipe_buf_ops = {
1195 	.can_merge = 0,
1196 	.confirm = generic_pipe_buf_confirm,
1197 	.release = relay_pipe_buf_release,
1198 	.steal = generic_pipe_buf_steal,
1199 	.get = generic_pipe_buf_get,
1200 };
1201 
1202 static void relay_page_release(struct splice_pipe_desc *spd, unsigned int i)
1203 {
1204 }
1205 
1206 /*
1207  *	subbuf_splice_actor - splice up to one subbuf's worth of data
1208  */
1209 static ssize_t subbuf_splice_actor(struct file *in,
1210 			       loff_t *ppos,
1211 			       struct pipe_inode_info *pipe,
1212 			       size_t len,
1213 			       unsigned int flags,
1214 			       int *nonpad_ret)
1215 {
1216 	unsigned int pidx, poff, total_len, subbuf_pages, nr_pages;
1217 	struct rchan_buf *rbuf = in->private_data;
1218 	unsigned int subbuf_size = rbuf->chan->subbuf_size;
1219 	uint64_t pos = (uint64_t) *ppos;
1220 	uint32_t alloc_size = (uint32_t) rbuf->chan->alloc_size;
1221 	size_t read_start = (size_t) do_div(pos, alloc_size);
1222 	size_t read_subbuf = read_start / subbuf_size;
1223 	size_t padding = rbuf->padding[read_subbuf];
1224 	size_t nonpad_end = read_subbuf * subbuf_size + subbuf_size - padding;
1225 	struct page *pages[PIPE_DEF_BUFFERS];
1226 	struct partial_page partial[PIPE_DEF_BUFFERS];
1227 	struct splice_pipe_desc spd = {
1228 		.pages = pages,
1229 		.nr_pages = 0,
1230 		.nr_pages_max = PIPE_DEF_BUFFERS,
1231 		.partial = partial,
1232 		.flags = flags,
1233 		.ops = &relay_pipe_buf_ops,
1234 		.spd_release = relay_page_release,
1235 	};
1236 	ssize_t ret;
1237 
1238 	if (rbuf->subbufs_produced == rbuf->subbufs_consumed)
1239 		return 0;
1240 	if (splice_grow_spd(pipe, &spd))
1241 		return -ENOMEM;
1242 
1243 	/*
1244 	 * Adjust read len, if longer than what is available
1245 	 */
1246 	if (len > (subbuf_size - read_start % subbuf_size))
1247 		len = subbuf_size - read_start % subbuf_size;
1248 
1249 	subbuf_pages = rbuf->chan->alloc_size >> PAGE_SHIFT;
1250 	pidx = (read_start / PAGE_SIZE) % subbuf_pages;
1251 	poff = read_start & ~PAGE_MASK;
1252 	nr_pages = min_t(unsigned int, subbuf_pages, spd.nr_pages_max);
1253 
1254 	for (total_len = 0; spd.nr_pages < nr_pages; spd.nr_pages++) {
1255 		unsigned int this_len, this_end, private;
1256 		unsigned int cur_pos = read_start + total_len;
1257 
1258 		if (!len)
1259 			break;
1260 
1261 		this_len = min_t(unsigned long, len, PAGE_SIZE - poff);
1262 		private = this_len;
1263 
1264 		spd.pages[spd.nr_pages] = rbuf->page_array[pidx];
1265 		spd.partial[spd.nr_pages].offset = poff;
1266 
1267 		this_end = cur_pos + this_len;
1268 		if (this_end >= nonpad_end) {
1269 			this_len = nonpad_end - cur_pos;
1270 			private = this_len + padding;
1271 		}
1272 		spd.partial[spd.nr_pages].len = this_len;
1273 		spd.partial[spd.nr_pages].private = private;
1274 
1275 		len -= this_len;
1276 		total_len += this_len;
1277 		poff = 0;
1278 		pidx = (pidx + 1) % subbuf_pages;
1279 
1280 		if (this_end >= nonpad_end) {
1281 			spd.nr_pages++;
1282 			break;
1283 		}
1284 	}
1285 
1286 	ret = 0;
1287 	if (!spd.nr_pages)
1288 		goto out;
1289 
1290 	ret = *nonpad_ret = splice_to_pipe(pipe, &spd);
1291 	if (ret < 0 || ret < total_len)
1292 		goto out;
1293 
1294         if (read_start + ret == nonpad_end)
1295                 ret += padding;
1296 
1297 out:
1298 	splice_shrink_spd(&spd);
1299 	return ret;
1300 }
1301 
1302 static ssize_t relay_file_splice_read(struct file *in,
1303 				      loff_t *ppos,
1304 				      struct pipe_inode_info *pipe,
1305 				      size_t len,
1306 				      unsigned int flags)
1307 {
1308 	ssize_t spliced;
1309 	int ret;
1310 	int nonpad_ret = 0;
1311 
1312 	ret = 0;
1313 	spliced = 0;
1314 
1315 	while (len && !spliced) {
1316 		ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret);
1317 		if (ret < 0)
1318 			break;
1319 		else if (!ret) {
1320 			if (flags & SPLICE_F_NONBLOCK)
1321 				ret = -EAGAIN;
1322 			break;
1323 		}
1324 
1325 		*ppos += ret;
1326 		if (ret > len)
1327 			len = 0;
1328 		else
1329 			len -= ret;
1330 		spliced += nonpad_ret;
1331 		nonpad_ret = 0;
1332 	}
1333 
1334 	if (spliced)
1335 		return spliced;
1336 
1337 	return ret;
1338 }
1339 
1340 const struct file_operations relay_file_operations = {
1341 	.open		= relay_file_open,
1342 	.poll		= relay_file_poll,
1343 	.mmap		= relay_file_mmap,
1344 	.read		= relay_file_read,
1345 	.llseek		= no_llseek,
1346 	.release	= relay_file_release,
1347 	.splice_read	= relay_file_splice_read,
1348 };
1349 EXPORT_SYMBOL_GPL(relay_file_operations);
1350 
1351 static __init int relay_init(void)
1352 {
1353 
1354 	hotcpu_notifier(relay_hotcpu_callback, 0);
1355 	return 0;
1356 }
1357 
1358 early_initcall(relay_init);
1359