xref: /openbmc/linux/kernel/relay.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
1 /*
2  * Public API and common code for kernel->userspace relay file support.
3  *
4  * See Documentation/filesystems/relayfs.txt for an overview of relayfs.
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/module.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 
25 /* list of open channels, for cpu hotplug */
26 static DEFINE_MUTEX(relay_channels_mutex);
27 static LIST_HEAD(relay_channels);
28 
29 /*
30  * close() vm_op implementation for relay file mapping.
31  */
32 static void relay_file_mmap_close(struct vm_area_struct *vma)
33 {
34 	struct rchan_buf *buf = vma->vm_private_data;
35 	buf->chan->cb->buf_unmapped(buf, vma->vm_file);
36 }
37 
38 /*
39  * nopage() vm_op implementation for relay file mapping.
40  */
41 static struct page *relay_buf_nopage(struct vm_area_struct *vma,
42 				     unsigned long address,
43 				     int *type)
44 {
45 	struct page *page;
46 	struct rchan_buf *buf = vma->vm_private_data;
47 	unsigned long offset = address - vma->vm_start;
48 
49 	if (address > vma->vm_end)
50 		return NOPAGE_SIGBUS; /* Disallow mremap */
51 	if (!buf)
52 		return NOPAGE_OOM;
53 
54 	page = vmalloc_to_page(buf->start + offset);
55 	if (!page)
56 		return NOPAGE_OOM;
57 	get_page(page);
58 
59 	if (type)
60 		*type = VM_FAULT_MINOR;
61 
62 	return page;
63 }
64 
65 /*
66  * vm_ops for relay file mappings.
67  */
68 static struct vm_operations_struct relay_file_mmap_ops = {
69 	.nopage = relay_buf_nopage,
70 	.close = relay_file_mmap_close,
71 };
72 
73 /**
74  *	relay_mmap_buf: - mmap channel buffer to process address space
75  *	@buf: relay channel buffer
76  *	@vma: vm_area_struct describing memory to be mapped
77  *
78  *	Returns 0 if ok, negative on error
79  *
80  *	Caller should already have grabbed mmap_sem.
81  */
82 int relay_mmap_buf(struct rchan_buf *buf, struct vm_area_struct *vma)
83 {
84 	unsigned long length = vma->vm_end - vma->vm_start;
85 	struct file *filp = vma->vm_file;
86 
87 	if (!buf)
88 		return -EBADF;
89 
90 	if (length != (unsigned long)buf->chan->alloc_size)
91 		return -EINVAL;
92 
93 	vma->vm_ops = &relay_file_mmap_ops;
94 	vma->vm_private_data = buf;
95 	buf->chan->cb->buf_mapped(buf, filp);
96 
97 	return 0;
98 }
99 
100 /**
101  *	relay_alloc_buf - allocate a channel buffer
102  *	@buf: the buffer struct
103  *	@size: total size of the buffer
104  *
105  *	Returns a pointer to the resulting buffer, %NULL if unsuccessful. The
106  *	passed in size will get page aligned, if it isn't already.
107  */
108 static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size)
109 {
110 	void *mem;
111 	unsigned int i, j, n_pages;
112 
113 	*size = PAGE_ALIGN(*size);
114 	n_pages = *size >> PAGE_SHIFT;
115 
116 	buf->page_array = kcalloc(n_pages, sizeof(struct page *), GFP_KERNEL);
117 	if (!buf->page_array)
118 		return NULL;
119 
120 	for (i = 0; i < n_pages; i++) {
121 		buf->page_array[i] = alloc_page(GFP_KERNEL);
122 		if (unlikely(!buf->page_array[i]))
123 			goto depopulate;
124 	}
125 	mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL);
126 	if (!mem)
127 		goto depopulate;
128 
129 	memset(mem, 0, *size);
130 	buf->page_count = n_pages;
131 	return mem;
132 
133 depopulate:
134 	for (j = 0; j < i; j++)
135 		__free_page(buf->page_array[j]);
136 	kfree(buf->page_array);
137 	return NULL;
138 }
139 
140 /**
141  *	relay_create_buf - allocate and initialize a channel buffer
142  *	@chan: the relay channel
143  *
144  *	Returns channel buffer if successful, %NULL otherwise.
145  */
146 struct rchan_buf *relay_create_buf(struct rchan *chan)
147 {
148 	struct rchan_buf *buf = kzalloc(sizeof(struct rchan_buf), GFP_KERNEL);
149 	if (!buf)
150 		return NULL;
151 
152 	buf->padding = kmalloc(chan->n_subbufs * sizeof(size_t *), GFP_KERNEL);
153 	if (!buf->padding)
154 		goto free_buf;
155 
156 	buf->start = relay_alloc_buf(buf, &chan->alloc_size);
157 	if (!buf->start)
158 		goto free_buf;
159 
160 	buf->chan = chan;
161 	kref_get(&buf->chan->kref);
162 	return buf;
163 
164 free_buf:
165 	kfree(buf->padding);
166 	kfree(buf);
167 	return NULL;
168 }
169 
170 /**
171  *	relay_destroy_channel - free the channel struct
172  *	@kref: target kernel reference that contains the relay channel
173  *
174  *	Should only be called from kref_put().
175  */
176 void relay_destroy_channel(struct kref *kref)
177 {
178 	struct rchan *chan = container_of(kref, struct rchan, kref);
179 	kfree(chan);
180 }
181 
182 /**
183  *	relay_destroy_buf - destroy an rchan_buf struct and associated buffer
184  *	@buf: the buffer struct
185  */
186 void relay_destroy_buf(struct rchan_buf *buf)
187 {
188 	struct rchan *chan = buf->chan;
189 	unsigned int i;
190 
191 	if (likely(buf->start)) {
192 		vunmap(buf->start);
193 		for (i = 0; i < buf->page_count; i++)
194 			__free_page(buf->page_array[i]);
195 		kfree(buf->page_array);
196 	}
197 	chan->buf[buf->cpu] = NULL;
198 	kfree(buf->padding);
199 	kfree(buf);
200 	kref_put(&chan->kref, relay_destroy_channel);
201 }
202 
203 /**
204  *	relay_remove_buf - remove a channel buffer
205  *	@kref: target kernel reference that contains the relay buffer
206  *
207  *	Removes the file from the fileystem, which also frees the
208  *	rchan_buf_struct and the channel buffer.  Should only be called from
209  *	kref_put().
210  */
211 void relay_remove_buf(struct kref *kref)
212 {
213 	struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
214 	buf->chan->cb->remove_buf_file(buf->dentry);
215 	relay_destroy_buf(buf);
216 }
217 
218 /**
219  *	relay_buf_empty - boolean, is the channel buffer empty?
220  *	@buf: channel buffer
221  *
222  *	Returns 1 if the buffer is empty, 0 otherwise.
223  */
224 int relay_buf_empty(struct rchan_buf *buf)
225 {
226 	return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1;
227 }
228 EXPORT_SYMBOL_GPL(relay_buf_empty);
229 
230 /**
231  *	relay_buf_full - boolean, is the channel buffer full?
232  *	@buf: channel buffer
233  *
234  *	Returns 1 if the buffer is full, 0 otherwise.
235  */
236 int relay_buf_full(struct rchan_buf *buf)
237 {
238 	size_t ready = buf->subbufs_produced - buf->subbufs_consumed;
239 	return (ready >= buf->chan->n_subbufs) ? 1 : 0;
240 }
241 EXPORT_SYMBOL_GPL(relay_buf_full);
242 
243 /*
244  * High-level relay kernel API and associated functions.
245  */
246 
247 /*
248  * rchan_callback implementations defining default channel behavior.  Used
249  * in place of corresponding NULL values in client callback struct.
250  */
251 
252 /*
253  * subbuf_start() default callback.  Does nothing.
254  */
255 static int subbuf_start_default_callback (struct rchan_buf *buf,
256 					  void *subbuf,
257 					  void *prev_subbuf,
258 					  size_t prev_padding)
259 {
260 	if (relay_buf_full(buf))
261 		return 0;
262 
263 	return 1;
264 }
265 
266 /*
267  * buf_mapped() default callback.  Does nothing.
268  */
269 static void buf_mapped_default_callback(struct rchan_buf *buf,
270 					struct file *filp)
271 {
272 }
273 
274 /*
275  * buf_unmapped() default callback.  Does nothing.
276  */
277 static void buf_unmapped_default_callback(struct rchan_buf *buf,
278 					  struct file *filp)
279 {
280 }
281 
282 /*
283  * create_buf_file_create() default callback.  Does nothing.
284  */
285 static struct dentry *create_buf_file_default_callback(const char *filename,
286 						       struct dentry *parent,
287 						       int mode,
288 						       struct rchan_buf *buf,
289 						       int *is_global)
290 {
291 	return NULL;
292 }
293 
294 /*
295  * remove_buf_file() default callback.  Does nothing.
296  */
297 static int remove_buf_file_default_callback(struct dentry *dentry)
298 {
299 	return -EINVAL;
300 }
301 
302 /* relay channel default callbacks */
303 static struct rchan_callbacks default_channel_callbacks = {
304 	.subbuf_start = subbuf_start_default_callback,
305 	.buf_mapped = buf_mapped_default_callback,
306 	.buf_unmapped = buf_unmapped_default_callback,
307 	.create_buf_file = create_buf_file_default_callback,
308 	.remove_buf_file = remove_buf_file_default_callback,
309 };
310 
311 /**
312  *	wakeup_readers - wake up readers waiting on a channel
313  *	@data: contains the channel buffer
314  *
315  *	This is the timer function used to defer reader waking.
316  */
317 static void wakeup_readers(unsigned long data)
318 {
319 	struct rchan_buf *buf = (struct rchan_buf *)data;
320 	wake_up_interruptible(&buf->read_wait);
321 }
322 
323 /**
324  *	__relay_reset - reset a channel buffer
325  *	@buf: the channel buffer
326  *	@init: 1 if this is a first-time initialization
327  *
328  *	See relay_reset() for description of effect.
329  */
330 static void __relay_reset(struct rchan_buf *buf, unsigned int init)
331 {
332 	size_t i;
333 
334 	if (init) {
335 		init_waitqueue_head(&buf->read_wait);
336 		kref_init(&buf->kref);
337 		setup_timer(&buf->timer, wakeup_readers, (unsigned long)buf);
338 	} else
339 		del_timer_sync(&buf->timer);
340 
341 	buf->subbufs_produced = 0;
342 	buf->subbufs_consumed = 0;
343 	buf->bytes_consumed = 0;
344 	buf->finalized = 0;
345 	buf->data = buf->start;
346 	buf->offset = 0;
347 
348 	for (i = 0; i < buf->chan->n_subbufs; i++)
349 		buf->padding[i] = 0;
350 
351 	buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0);
352 }
353 
354 /**
355  *	relay_reset - reset the channel
356  *	@chan: the channel
357  *
358  *	This has the effect of erasing all data from all channel buffers
359  *	and restarting the channel in its initial state.  The buffers
360  *	are not freed, so any mappings are still in effect.
361  *
362  *	NOTE. Care should be taken that the channel isn't actually
363  *	being used by anything when this call is made.
364  */
365 void relay_reset(struct rchan *chan)
366 {
367 	unsigned int i;
368 
369 	if (!chan)
370 		return;
371 
372  	if (chan->is_global && chan->buf[0]) {
373 		__relay_reset(chan->buf[0], 0);
374 		return;
375 	}
376 
377 	mutex_lock(&relay_channels_mutex);
378 	for_each_online_cpu(i)
379 		if (chan->buf[i])
380 			__relay_reset(chan->buf[i], 0);
381 	mutex_unlock(&relay_channels_mutex);
382 }
383 EXPORT_SYMBOL_GPL(relay_reset);
384 
385 /*
386  *	relay_open_buf - create a new relay channel buffer
387  *
388  *	used by relay_open() and CPU hotplug.
389  */
390 static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu)
391 {
392  	struct rchan_buf *buf = NULL;
393 	struct dentry *dentry;
394  	char *tmpname;
395 
396  	if (chan->is_global)
397 		return chan->buf[0];
398 
399 	tmpname = kzalloc(NAME_MAX + 1, GFP_KERNEL);
400  	if (!tmpname)
401  		goto end;
402  	snprintf(tmpname, NAME_MAX, "%s%d", chan->base_filename, cpu);
403 
404 	buf = relay_create_buf(chan);
405 	if (!buf)
406  		goto free_name;
407 
408  	buf->cpu = cpu;
409  	__relay_reset(buf, 1);
410 
411 	/* Create file in fs */
412  	dentry = chan->cb->create_buf_file(tmpname, chan->parent, S_IRUSR,
413  					   buf, &chan->is_global);
414  	if (!dentry)
415  		goto free_buf;
416 
417 	buf->dentry = dentry;
418 
419  	if(chan->is_global) {
420  		chan->buf[0] = buf;
421  		buf->cpu = 0;
422   	}
423 
424  	goto free_name;
425 
426 free_buf:
427  	relay_destroy_buf(buf);
428 free_name:
429  	kfree(tmpname);
430 end:
431 	return buf;
432 }
433 
434 /**
435  *	relay_close_buf - close a channel buffer
436  *	@buf: channel buffer
437  *
438  *	Marks the buffer finalized and restores the default callbacks.
439  *	The channel buffer and channel buffer data structure are then freed
440  *	automatically when the last reference is given up.
441  */
442 static void relay_close_buf(struct rchan_buf *buf)
443 {
444 	buf->finalized = 1;
445 	del_timer_sync(&buf->timer);
446 	kref_put(&buf->kref, relay_remove_buf);
447 }
448 
449 static void setup_callbacks(struct rchan *chan,
450 				   struct rchan_callbacks *cb)
451 {
452 	if (!cb) {
453 		chan->cb = &default_channel_callbacks;
454 		return;
455 	}
456 
457 	if (!cb->subbuf_start)
458 		cb->subbuf_start = subbuf_start_default_callback;
459 	if (!cb->buf_mapped)
460 		cb->buf_mapped = buf_mapped_default_callback;
461 	if (!cb->buf_unmapped)
462 		cb->buf_unmapped = buf_unmapped_default_callback;
463 	if (!cb->create_buf_file)
464 		cb->create_buf_file = create_buf_file_default_callback;
465 	if (!cb->remove_buf_file)
466 		cb->remove_buf_file = remove_buf_file_default_callback;
467 	chan->cb = cb;
468 }
469 
470 /**
471  * 	relay_hotcpu_callback - CPU hotplug callback
472  * 	@nb: notifier block
473  * 	@action: hotplug action to take
474  * 	@hcpu: CPU number
475  *
476  * 	Returns the success/failure of the operation. (%NOTIFY_OK, %NOTIFY_BAD)
477  */
478 static int __cpuinit relay_hotcpu_callback(struct notifier_block *nb,
479 				unsigned long action,
480 				void *hcpu)
481 {
482 	unsigned int hotcpu = (unsigned long)hcpu;
483 	struct rchan *chan;
484 
485 	switch(action) {
486 	case CPU_UP_PREPARE:
487 	case CPU_UP_PREPARE_FROZEN:
488 		mutex_lock(&relay_channels_mutex);
489 		list_for_each_entry(chan, &relay_channels, list) {
490 			if (chan->buf[hotcpu])
491 				continue;
492 			chan->buf[hotcpu] = relay_open_buf(chan, hotcpu);
493 			if(!chan->buf[hotcpu]) {
494 				printk(KERN_ERR
495 					"relay_hotcpu_callback: cpu %d buffer "
496 					"creation failed\n", hotcpu);
497 				mutex_unlock(&relay_channels_mutex);
498 				return NOTIFY_BAD;
499 			}
500 		}
501 		mutex_unlock(&relay_channels_mutex);
502 		break;
503 	case CPU_DEAD:
504 	case CPU_DEAD_FROZEN:
505 		/* No need to flush the cpu : will be flushed upon
506 		 * final relay_flush() call. */
507 		break;
508 	}
509 	return NOTIFY_OK;
510 }
511 
512 /**
513  *	relay_open - create a new relay channel
514  *	@base_filename: base name of files to create
515  *	@parent: dentry of parent directory, %NULL for root directory
516  *	@subbuf_size: size of sub-buffers
517  *	@n_subbufs: number of sub-buffers
518  *	@cb: client callback functions
519  *	@private_data: user-defined data
520  *
521  *	Returns channel pointer if successful, %NULL otherwise.
522  *
523  *	Creates a channel buffer for each cpu using the sizes and
524  *	attributes specified.  The created channel buffer files
525  *	will be named base_filename0...base_filenameN-1.  File
526  *	permissions will be %S_IRUSR.
527  */
528 struct rchan *relay_open(const char *base_filename,
529 			 struct dentry *parent,
530 			 size_t subbuf_size,
531 			 size_t n_subbufs,
532 			 struct rchan_callbacks *cb,
533 			 void *private_data)
534 {
535 	unsigned int i;
536 	struct rchan *chan;
537 	if (!base_filename)
538 		return NULL;
539 
540 	if (!(subbuf_size && n_subbufs))
541 		return NULL;
542 
543 	chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
544 	if (!chan)
545 		return NULL;
546 
547 	chan->version = RELAYFS_CHANNEL_VERSION;
548 	chan->n_subbufs = n_subbufs;
549 	chan->subbuf_size = subbuf_size;
550 	chan->alloc_size = FIX_SIZE(subbuf_size * n_subbufs);
551 	chan->parent = parent;
552 	chan->private_data = private_data;
553 	strlcpy(chan->base_filename, base_filename, NAME_MAX);
554 	setup_callbacks(chan, cb);
555 	kref_init(&chan->kref);
556 
557 	mutex_lock(&relay_channels_mutex);
558 	for_each_online_cpu(i) {
559 		chan->buf[i] = relay_open_buf(chan, i);
560 		if (!chan->buf[i])
561 			goto free_bufs;
562 	}
563 	list_add(&chan->list, &relay_channels);
564 	mutex_unlock(&relay_channels_mutex);
565 
566 	return chan;
567 
568 free_bufs:
569 	for_each_online_cpu(i) {
570 		if (!chan->buf[i])
571 			break;
572 		relay_close_buf(chan->buf[i]);
573 	}
574 
575 	kref_put(&chan->kref, relay_destroy_channel);
576 	mutex_unlock(&relay_channels_mutex);
577 	return NULL;
578 }
579 EXPORT_SYMBOL_GPL(relay_open);
580 
581 /**
582  *	relay_switch_subbuf - switch to a new sub-buffer
583  *	@buf: channel buffer
584  *	@length: size of current event
585  *
586  *	Returns either the length passed in or 0 if full.
587  *
588  *	Performs sub-buffer-switch tasks such as invoking callbacks,
589  *	updating padding counts, waking up readers, etc.
590  */
591 size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
592 {
593 	void *old, *new;
594 	size_t old_subbuf, new_subbuf;
595 
596 	if (unlikely(length > buf->chan->subbuf_size))
597 		goto toobig;
598 
599 	if (buf->offset != buf->chan->subbuf_size + 1) {
600 		buf->prev_padding = buf->chan->subbuf_size - buf->offset;
601 		old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
602 		buf->padding[old_subbuf] = buf->prev_padding;
603 		buf->subbufs_produced++;
604 		buf->dentry->d_inode->i_size += buf->chan->subbuf_size -
605 			buf->padding[old_subbuf];
606 		smp_mb();
607 		if (waitqueue_active(&buf->read_wait))
608 			/*
609 			 * Calling wake_up_interruptible() from here
610 			 * will deadlock if we happen to be logging
611 			 * from the scheduler (trying to re-grab
612 			 * rq->lock), so defer it.
613 			 */
614 			__mod_timer(&buf->timer, jiffies + 1);
615 	}
616 
617 	old = buf->data;
618 	new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
619 	new = buf->start + new_subbuf * buf->chan->subbuf_size;
620 	buf->offset = 0;
621 	if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
622 		buf->offset = buf->chan->subbuf_size + 1;
623 		return 0;
624 	}
625 	buf->data = new;
626 	buf->padding[new_subbuf] = 0;
627 
628 	if (unlikely(length + buf->offset > buf->chan->subbuf_size))
629 		goto toobig;
630 
631 	return length;
632 
633 toobig:
634 	buf->chan->last_toobig = length;
635 	return 0;
636 }
637 EXPORT_SYMBOL_GPL(relay_switch_subbuf);
638 
639 /**
640  *	relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
641  *	@chan: the channel
642  *	@cpu: the cpu associated with the channel buffer to update
643  *	@subbufs_consumed: number of sub-buffers to add to current buf's count
644  *
645  *	Adds to the channel buffer's consumed sub-buffer count.
646  *	subbufs_consumed should be the number of sub-buffers newly consumed,
647  *	not the total consumed.
648  *
649  *	NOTE. Kernel clients don't need to call this function if the channel
650  *	mode is 'overwrite'.
651  */
652 void relay_subbufs_consumed(struct rchan *chan,
653 			    unsigned int cpu,
654 			    size_t subbufs_consumed)
655 {
656 	struct rchan_buf *buf;
657 
658 	if (!chan)
659 		return;
660 
661 	if (cpu >= NR_CPUS || !chan->buf[cpu])
662 		return;
663 
664 	buf = chan->buf[cpu];
665 	buf->subbufs_consumed += subbufs_consumed;
666 	if (buf->subbufs_consumed > buf->subbufs_produced)
667 		buf->subbufs_consumed = buf->subbufs_produced;
668 }
669 EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
670 
671 /**
672  *	relay_close - close the channel
673  *	@chan: the channel
674  *
675  *	Closes all channel buffers and frees the channel.
676  */
677 void relay_close(struct rchan *chan)
678 {
679 	unsigned int i;
680 
681 	if (!chan)
682 		return;
683 
684 	mutex_lock(&relay_channels_mutex);
685 	if (chan->is_global && chan->buf[0])
686 		relay_close_buf(chan->buf[0]);
687 	else
688 		for_each_possible_cpu(i)
689 			if (chan->buf[i])
690 				relay_close_buf(chan->buf[i]);
691 
692 	if (chan->last_toobig)
693 		printk(KERN_WARNING "relay: one or more items not logged "
694 		       "[item size (%Zd) > sub-buffer size (%Zd)]\n",
695 		       chan->last_toobig, chan->subbuf_size);
696 
697 	list_del(&chan->list);
698 	kref_put(&chan->kref, relay_destroy_channel);
699 	mutex_unlock(&relay_channels_mutex);
700 }
701 EXPORT_SYMBOL_GPL(relay_close);
702 
703 /**
704  *	relay_flush - close the channel
705  *	@chan: the channel
706  *
707  *	Flushes all channel buffers, i.e. forces buffer switch.
708  */
709 void relay_flush(struct rchan *chan)
710 {
711 	unsigned int i;
712 
713 	if (!chan)
714 		return;
715 
716 	if (chan->is_global && chan->buf[0]) {
717 		relay_switch_subbuf(chan->buf[0], 0);
718 		return;
719 	}
720 
721 	mutex_lock(&relay_channels_mutex);
722 	for_each_possible_cpu(i)
723 		if (chan->buf[i])
724 			relay_switch_subbuf(chan->buf[i], 0);
725 	mutex_unlock(&relay_channels_mutex);
726 }
727 EXPORT_SYMBOL_GPL(relay_flush);
728 
729 /**
730  *	relay_file_open - open file op for relay files
731  *	@inode: the inode
732  *	@filp: the file
733  *
734  *	Increments the channel buffer refcount.
735  */
736 static int relay_file_open(struct inode *inode, struct file *filp)
737 {
738 	struct rchan_buf *buf = inode->i_private;
739 	kref_get(&buf->kref);
740 	filp->private_data = buf;
741 
742 	return 0;
743 }
744 
745 /**
746  *	relay_file_mmap - mmap file op for relay files
747  *	@filp: the file
748  *	@vma: the vma describing what to map
749  *
750  *	Calls upon relay_mmap_buf() to map the file into user space.
751  */
752 static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
753 {
754 	struct rchan_buf *buf = filp->private_data;
755 	return relay_mmap_buf(buf, vma);
756 }
757 
758 /**
759  *	relay_file_poll - poll file op for relay files
760  *	@filp: the file
761  *	@wait: poll table
762  *
763  *	Poll implemention.
764  */
765 static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
766 {
767 	unsigned int mask = 0;
768 	struct rchan_buf *buf = filp->private_data;
769 
770 	if (buf->finalized)
771 		return POLLERR;
772 
773 	if (filp->f_mode & FMODE_READ) {
774 		poll_wait(filp, &buf->read_wait, wait);
775 		if (!relay_buf_empty(buf))
776 			mask |= POLLIN | POLLRDNORM;
777 	}
778 
779 	return mask;
780 }
781 
782 /**
783  *	relay_file_release - release file op for relay files
784  *	@inode: the inode
785  *	@filp: the file
786  *
787  *	Decrements the channel refcount, as the filesystem is
788  *	no longer using it.
789  */
790 static int relay_file_release(struct inode *inode, struct file *filp)
791 {
792 	struct rchan_buf *buf = filp->private_data;
793 	kref_put(&buf->kref, relay_remove_buf);
794 
795 	return 0;
796 }
797 
798 /*
799  *	relay_file_read_consume - update the consumed count for the buffer
800  */
801 static void relay_file_read_consume(struct rchan_buf *buf,
802 				    size_t read_pos,
803 				    size_t bytes_consumed)
804 {
805 	size_t subbuf_size = buf->chan->subbuf_size;
806 	size_t n_subbufs = buf->chan->n_subbufs;
807 	size_t read_subbuf;
808 
809 	if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
810 		relay_subbufs_consumed(buf->chan, buf->cpu, 1);
811 		buf->bytes_consumed = 0;
812 	}
813 
814 	buf->bytes_consumed += bytes_consumed;
815 	read_subbuf = read_pos / buf->chan->subbuf_size;
816 	if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
817 		if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
818 		    (buf->offset == subbuf_size))
819 			return;
820 		relay_subbufs_consumed(buf->chan, buf->cpu, 1);
821 		buf->bytes_consumed = 0;
822 	}
823 }
824 
825 /*
826  *	relay_file_read_avail - boolean, are there unconsumed bytes available?
827  */
828 static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
829 {
830 	size_t subbuf_size = buf->chan->subbuf_size;
831 	size_t n_subbufs = buf->chan->n_subbufs;
832 	size_t produced = buf->subbufs_produced;
833 	size_t consumed = buf->subbufs_consumed;
834 
835 	relay_file_read_consume(buf, read_pos, 0);
836 
837 	if (unlikely(buf->offset > subbuf_size)) {
838 		if (produced == consumed)
839 			return 0;
840 		return 1;
841 	}
842 
843 	if (unlikely(produced - consumed >= n_subbufs)) {
844 		consumed = (produced / n_subbufs) * n_subbufs;
845 		buf->subbufs_consumed = consumed;
846 	}
847 
848 	produced = (produced % n_subbufs) * subbuf_size + buf->offset;
849 	consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
850 
851 	if (consumed > produced)
852 		produced += n_subbufs * subbuf_size;
853 
854 	if (consumed == produced)
855 		return 0;
856 
857 	return 1;
858 }
859 
860 /**
861  *	relay_file_read_subbuf_avail - return bytes available in sub-buffer
862  *	@read_pos: file read position
863  *	@buf: relay channel buffer
864  */
865 static size_t relay_file_read_subbuf_avail(size_t read_pos,
866 					   struct rchan_buf *buf)
867 {
868 	size_t padding, avail = 0;
869 	size_t read_subbuf, read_offset, write_subbuf, write_offset;
870 	size_t subbuf_size = buf->chan->subbuf_size;
871 
872 	write_subbuf = (buf->data - buf->start) / subbuf_size;
873 	write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
874 	read_subbuf = read_pos / subbuf_size;
875 	read_offset = read_pos % subbuf_size;
876 	padding = buf->padding[read_subbuf];
877 
878 	if (read_subbuf == write_subbuf) {
879 		if (read_offset + padding < write_offset)
880 			avail = write_offset - (read_offset + padding);
881 	} else
882 		avail = (subbuf_size - padding) - read_offset;
883 
884 	return avail;
885 }
886 
887 /**
888  *	relay_file_read_start_pos - find the first available byte to read
889  *	@read_pos: file read position
890  *	@buf: relay channel buffer
891  *
892  *	If the @read_pos is in the middle of padding, return the
893  *	position of the first actually available byte, otherwise
894  *	return the original value.
895  */
896 static size_t relay_file_read_start_pos(size_t read_pos,
897 					struct rchan_buf *buf)
898 {
899 	size_t read_subbuf, padding, padding_start, padding_end;
900 	size_t subbuf_size = buf->chan->subbuf_size;
901 	size_t n_subbufs = buf->chan->n_subbufs;
902 
903 	read_subbuf = read_pos / subbuf_size;
904 	padding = buf->padding[read_subbuf];
905 	padding_start = (read_subbuf + 1) * subbuf_size - padding;
906 	padding_end = (read_subbuf + 1) * subbuf_size;
907 	if (read_pos >= padding_start && read_pos < padding_end) {
908 		read_subbuf = (read_subbuf + 1) % n_subbufs;
909 		read_pos = read_subbuf * subbuf_size;
910 	}
911 
912 	return read_pos;
913 }
914 
915 /**
916  *	relay_file_read_end_pos - return the new read position
917  *	@read_pos: file read position
918  *	@buf: relay channel buffer
919  *	@count: number of bytes to be read
920  */
921 static size_t relay_file_read_end_pos(struct rchan_buf *buf,
922 				      size_t read_pos,
923 				      size_t count)
924 {
925 	size_t read_subbuf, padding, end_pos;
926 	size_t subbuf_size = buf->chan->subbuf_size;
927 	size_t n_subbufs = buf->chan->n_subbufs;
928 
929 	read_subbuf = read_pos / subbuf_size;
930 	padding = buf->padding[read_subbuf];
931 	if (read_pos % subbuf_size + count + padding == subbuf_size)
932 		end_pos = (read_subbuf + 1) * subbuf_size;
933 	else
934 		end_pos = read_pos + count;
935 	if (end_pos >= subbuf_size * n_subbufs)
936 		end_pos = 0;
937 
938 	return end_pos;
939 }
940 
941 /*
942  *	subbuf_read_actor - read up to one subbuf's worth of data
943  */
944 static int subbuf_read_actor(size_t read_start,
945 			     struct rchan_buf *buf,
946 			     size_t avail,
947 			     read_descriptor_t *desc,
948 			     read_actor_t actor)
949 {
950 	void *from;
951 	int ret = 0;
952 
953 	from = buf->start + read_start;
954 	ret = avail;
955 	if (copy_to_user(desc->arg.buf, from, avail)) {
956 		desc->error = -EFAULT;
957 		ret = 0;
958 	}
959 	desc->arg.data += ret;
960 	desc->written += ret;
961 	desc->count -= ret;
962 
963 	return ret;
964 }
965 
966 /*
967  *	subbuf_send_actor - send up to one subbuf's worth of data
968  */
969 static int subbuf_send_actor(size_t read_start,
970 			     struct rchan_buf *buf,
971 			     size_t avail,
972 			     read_descriptor_t *desc,
973 			     read_actor_t actor)
974 {
975 	unsigned long pidx, poff;
976 	unsigned int subbuf_pages;
977 	int ret = 0;
978 
979 	subbuf_pages = buf->chan->alloc_size >> PAGE_SHIFT;
980 	pidx = (read_start / PAGE_SIZE) % subbuf_pages;
981 	poff = read_start & ~PAGE_MASK;
982 	while (avail) {
983 		struct page *p = buf->page_array[pidx];
984 		unsigned int len;
985 
986 		len = PAGE_SIZE - poff;
987 		if (len > avail)
988 			len = avail;
989 
990 		len = actor(desc, p, poff, len);
991 		if (desc->error)
992 			break;
993 
994 		avail -= len;
995 		ret += len;
996 		poff = 0;
997 		pidx = (pidx + 1) % subbuf_pages;
998 	}
999 
1000 	return ret;
1001 }
1002 
1003 typedef int (*subbuf_actor_t) (size_t read_start,
1004 			       struct rchan_buf *buf,
1005 			       size_t avail,
1006 			       read_descriptor_t *desc,
1007 			       read_actor_t actor);
1008 
1009 /*
1010  *	relay_file_read_subbufs - read count bytes, bridging subbuf boundaries
1011  */
1012 static ssize_t relay_file_read_subbufs(struct file *filp, loff_t *ppos,
1013 					subbuf_actor_t subbuf_actor,
1014 					read_actor_t actor,
1015 					read_descriptor_t *desc)
1016 {
1017 	struct rchan_buf *buf = filp->private_data;
1018 	size_t read_start, avail;
1019 	int ret;
1020 
1021 	if (!desc->count)
1022 		return 0;
1023 
1024 	mutex_lock(&filp->f_path.dentry->d_inode->i_mutex);
1025 	do {
1026 		if (!relay_file_read_avail(buf, *ppos))
1027 			break;
1028 
1029 		read_start = relay_file_read_start_pos(*ppos, buf);
1030 		avail = relay_file_read_subbuf_avail(read_start, buf);
1031 		if (!avail)
1032 			break;
1033 
1034 		avail = min(desc->count, avail);
1035 		ret = subbuf_actor(read_start, buf, avail, desc, actor);
1036 		if (desc->error < 0)
1037 			break;
1038 
1039 		if (ret) {
1040 			relay_file_read_consume(buf, read_start, ret);
1041 			*ppos = relay_file_read_end_pos(buf, read_start, ret);
1042 		}
1043 	} while (desc->count && ret);
1044 	mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex);
1045 
1046 	return desc->written;
1047 }
1048 
1049 static ssize_t relay_file_read(struct file *filp,
1050 			       char __user *buffer,
1051 			       size_t count,
1052 			       loff_t *ppos)
1053 {
1054 	read_descriptor_t desc;
1055 	desc.written = 0;
1056 	desc.count = count;
1057 	desc.arg.buf = buffer;
1058 	desc.error = 0;
1059 	return relay_file_read_subbufs(filp, ppos, subbuf_read_actor,
1060 				       NULL, &desc);
1061 }
1062 
1063 static ssize_t relay_file_sendfile(struct file *filp,
1064 				   loff_t *ppos,
1065 				   size_t count,
1066 				   read_actor_t actor,
1067 				   void *target)
1068 {
1069 	read_descriptor_t desc;
1070 	desc.written = 0;
1071 	desc.count = count;
1072 	desc.arg.data = target;
1073 	desc.error = 0;
1074 	return relay_file_read_subbufs(filp, ppos, subbuf_send_actor,
1075 				       actor, &desc);
1076 }
1077 
1078 const struct file_operations relay_file_operations = {
1079 	.open		= relay_file_open,
1080 	.poll		= relay_file_poll,
1081 	.mmap		= relay_file_mmap,
1082 	.read		= relay_file_read,
1083 	.llseek		= no_llseek,
1084 	.release	= relay_file_release,
1085 	.sendfile       = relay_file_sendfile,
1086 };
1087 EXPORT_SYMBOL_GPL(relay_file_operations);
1088 
1089 static __init int relay_init(void)
1090 {
1091 
1092 	hotcpu_notifier(relay_hotcpu_callback, 0);
1093 	return 0;
1094 }
1095 
1096 module_init(relay_init);
1097