blob: f04bbdb56ac2a9123003b45d909f9ef6e971fd5e [file] [log] [blame]
Jens Axboeb86ff9812006-03-23 19:56:55 +01001/*
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 *
11 * This file is released under the GPL.
12 */
13#include <linux/errno.h>
14#include <linux/stddef.h>
15#include <linux/slab.h>
16#include <linux/module.h>
17#include <linux/string.h>
18#include <linux/relay.h>
19#include <linux/vmalloc.h>
20#include <linux/mm.h>
21
22/*
23 * close() vm_op implementation for relay file mapping.
24 */
25static void relay_file_mmap_close(struct vm_area_struct *vma)
26{
27 struct rchan_buf *buf = vma->vm_private_data;
28 buf->chan->cb->buf_unmapped(buf, vma->vm_file);
29}
30
31/*
32 * nopage() vm_op implementation for relay file mapping.
33 */
34static struct page *relay_buf_nopage(struct vm_area_struct *vma,
35 unsigned long address,
36 int *type)
37{
38 struct page *page;
39 struct rchan_buf *buf = vma->vm_private_data;
40 unsigned long offset = address - vma->vm_start;
41
42 if (address > vma->vm_end)
43 return NOPAGE_SIGBUS; /* Disallow mremap */
44 if (!buf)
45 return NOPAGE_OOM;
46
47 page = vmalloc_to_page(buf->start + offset);
48 if (!page)
49 return NOPAGE_OOM;
50 get_page(page);
51
52 if (type)
53 *type = VM_FAULT_MINOR;
54
55 return page;
56}
57
58/*
59 * vm_ops for relay file mappings.
60 */
61static struct vm_operations_struct relay_file_mmap_ops = {
62 .nopage = relay_buf_nopage,
63 .close = relay_file_mmap_close,
64};
65
66/**
67 * relay_mmap_buf: - mmap channel buffer to process address space
68 * @buf: relay channel buffer
69 * @vma: vm_area_struct describing memory to be mapped
70 *
71 * Returns 0 if ok, negative on error
72 *
73 * Caller should already have grabbed mmap_sem.
74 */
75int relay_mmap_buf(struct rchan_buf *buf, struct vm_area_struct *vma)
76{
77 unsigned long length = vma->vm_end - vma->vm_start;
78 struct file *filp = vma->vm_file;
79
80 if (!buf)
81 return -EBADF;
82
83 if (length != (unsigned long)buf->chan->alloc_size)
84 return -EINVAL;
85
86 vma->vm_ops = &relay_file_mmap_ops;
87 vma->vm_private_data = buf;
88 buf->chan->cb->buf_mapped(buf, filp);
89
90 return 0;
91}
92
93/**
94 * relay_alloc_buf - allocate a channel buffer
95 * @buf: the buffer struct
96 * @size: total size of the buffer
97 *
Randy Dunlap4c78a662006-09-29 01:59:10 -070098 * Returns a pointer to the resulting buffer, %NULL if unsuccessful. The
Jens Axboe221415d2006-03-23 19:57:55 +010099 * passed in size will get page aligned, if it isn't already.
Jens Axboeb86ff9812006-03-23 19:56:55 +0100100 */
Jens Axboe221415d2006-03-23 19:57:55 +0100101static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100102{
103 void *mem;
104 unsigned int i, j, n_pages;
105
Jens Axboe221415d2006-03-23 19:57:55 +0100106 *size = PAGE_ALIGN(*size);
107 n_pages = *size >> PAGE_SHIFT;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100108
109 buf->page_array = kcalloc(n_pages, sizeof(struct page *), GFP_KERNEL);
110 if (!buf->page_array)
111 return NULL;
112
113 for (i = 0; i < n_pages; i++) {
114 buf->page_array[i] = alloc_page(GFP_KERNEL);
115 if (unlikely(!buf->page_array[i]))
116 goto depopulate;
117 }
118 mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL);
119 if (!mem)
120 goto depopulate;
121
Jens Axboe221415d2006-03-23 19:57:55 +0100122 memset(mem, 0, *size);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100123 buf->page_count = n_pages;
124 return mem;
125
126depopulate:
127 for (j = 0; j < i; j++)
128 __free_page(buf->page_array[j]);
129 kfree(buf->page_array);
130 return NULL;
131}
132
133/**
134 * relay_create_buf - allocate and initialize a channel buffer
Randy Dunlap4c78a662006-09-29 01:59:10 -0700135 * @chan: the relay channel
Jens Axboeb86ff9812006-03-23 19:56:55 +0100136 *
Randy Dunlap4c78a662006-09-29 01:59:10 -0700137 * Returns channel buffer if successful, %NULL otherwise.
Jens Axboeb86ff9812006-03-23 19:56:55 +0100138 */
139struct rchan_buf *relay_create_buf(struct rchan *chan)
140{
141 struct rchan_buf *buf = kcalloc(1, sizeof(struct rchan_buf), GFP_KERNEL);
142 if (!buf)
143 return NULL;
144
145 buf->padding = kmalloc(chan->n_subbufs * sizeof(size_t *), GFP_KERNEL);
146 if (!buf->padding)
147 goto free_buf;
148
Jens Axboe221415d2006-03-23 19:57:55 +0100149 buf->start = relay_alloc_buf(buf, &chan->alloc_size);
Jens Axboeb86ff9812006-03-23 19:56:55 +0100150 if (!buf->start)
151 goto free_buf;
152
153 buf->chan = chan;
154 kref_get(&buf->chan->kref);
155 return buf;
156
157free_buf:
158 kfree(buf->padding);
159 kfree(buf);
160 return NULL;
161}
162
163/**
164 * relay_destroy_channel - free the channel struct
Randy Dunlap4c78a662006-09-29 01:59:10 -0700165 * @kref: target kernel reference that contains the relay channel
Jens Axboeb86ff9812006-03-23 19:56:55 +0100166 *
167 * Should only be called from kref_put().
168 */
169void relay_destroy_channel(struct kref *kref)
170{
171 struct rchan *chan = container_of(kref, struct rchan, kref);
172 kfree(chan);
173}
174
175/**
176 * relay_destroy_buf - destroy an rchan_buf struct and associated buffer
177 * @buf: the buffer struct
178 */
179void relay_destroy_buf(struct rchan_buf *buf)
180{
181 struct rchan *chan = buf->chan;
182 unsigned int i;
183
184 if (likely(buf->start)) {
185 vunmap(buf->start);
186 for (i = 0; i < buf->page_count; i++)
187 __free_page(buf->page_array[i]);
188 kfree(buf->page_array);
189 }
190 kfree(buf->padding);
191 kfree(buf);
192 kref_put(&chan->kref, relay_destroy_channel);
193}
194
195/**
196 * relay_remove_buf - remove a channel buffer
Randy Dunlap4c78a662006-09-29 01:59:10 -0700197 * @kref: target kernel reference that contains the relay buffer
Jens Axboeb86ff9812006-03-23 19:56:55 +0100198 *
199 * Removes the file from the fileystem, which also frees the
200 * rchan_buf_struct and the channel buffer. Should only be called from
201 * kref_put().
202 */
203void relay_remove_buf(struct kref *kref)
204{
205 struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
206 buf->chan->cb->remove_buf_file(buf->dentry);
207 relay_destroy_buf(buf);
208}
209
210/**
211 * relay_buf_empty - boolean, is the channel buffer empty?
212 * @buf: channel buffer
213 *
214 * Returns 1 if the buffer is empty, 0 otherwise.
215 */
216int relay_buf_empty(struct rchan_buf *buf)
217{
218 return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1;
219}
220EXPORT_SYMBOL_GPL(relay_buf_empty);
221
222/**
223 * relay_buf_full - boolean, is the channel buffer full?
224 * @buf: channel buffer
225 *
226 * Returns 1 if the buffer is full, 0 otherwise.
227 */
228int relay_buf_full(struct rchan_buf *buf)
229{
230 size_t ready = buf->subbufs_produced - buf->subbufs_consumed;
231 return (ready >= buf->chan->n_subbufs) ? 1 : 0;
232}
233EXPORT_SYMBOL_GPL(relay_buf_full);
234
235/*
236 * High-level relay kernel API and associated functions.
237 */
238
239/*
240 * rchan_callback implementations defining default channel behavior. Used
241 * in place of corresponding NULL values in client callback struct.
242 */
243
244/*
245 * subbuf_start() default callback. Does nothing.
246 */
247static int subbuf_start_default_callback (struct rchan_buf *buf,
248 void *subbuf,
249 void *prev_subbuf,
250 size_t prev_padding)
251{
252 if (relay_buf_full(buf))
253 return 0;
254
255 return 1;
256}
257
258/*
259 * buf_mapped() default callback. Does nothing.
260 */
261static void buf_mapped_default_callback(struct rchan_buf *buf,
262 struct file *filp)
263{
264}
265
266/*
267 * buf_unmapped() default callback. Does nothing.
268 */
269static void buf_unmapped_default_callback(struct rchan_buf *buf,
270 struct file *filp)
271{
272}
273
274/*
275 * create_buf_file_create() default callback. Does nothing.
276 */
277static struct dentry *create_buf_file_default_callback(const char *filename,
278 struct dentry *parent,
279 int mode,
280 struct rchan_buf *buf,
281 int *is_global)
282{
283 return NULL;
284}
285
286/*
287 * remove_buf_file() default callback. Does nothing.
288 */
289static int remove_buf_file_default_callback(struct dentry *dentry)
290{
291 return -EINVAL;
292}
293
294/* relay channel default callbacks */
295static struct rchan_callbacks default_channel_callbacks = {
296 .subbuf_start = subbuf_start_default_callback,
297 .buf_mapped = buf_mapped_default_callback,
298 .buf_unmapped = buf_unmapped_default_callback,
299 .create_buf_file = create_buf_file_default_callback,
300 .remove_buf_file = remove_buf_file_default_callback,
301};
302
303/**
304 * wakeup_readers - wake up readers waiting on a channel
305 * @private: the channel buffer
306 *
307 * This is the work function used to defer reader waking. The
308 * reason waking is deferred is that calling directly from write
309 * causes problems if you're writing from say the scheduler.
310 */
311static void wakeup_readers(void *private)
312{
313 struct rchan_buf *buf = private;
314 wake_up_interruptible(&buf->read_wait);
315}
316
317/**
318 * __relay_reset - reset a channel buffer
319 * @buf: the channel buffer
320 * @init: 1 if this is a first-time initialization
321 *
322 * See relay_reset for description of effect.
323 */
324static inline void __relay_reset(struct rchan_buf *buf, unsigned int init)
325{
326 size_t i;
327
328 if (init) {
329 init_waitqueue_head(&buf->read_wait);
330 kref_init(&buf->kref);
331 INIT_WORK(&buf->wake_readers, NULL, NULL);
332 } else {
333 cancel_delayed_work(&buf->wake_readers);
334 flush_scheduled_work();
335 }
336
337 buf->subbufs_produced = 0;
338 buf->subbufs_consumed = 0;
339 buf->bytes_consumed = 0;
340 buf->finalized = 0;
341 buf->data = buf->start;
342 buf->offset = 0;
343
344 for (i = 0; i < buf->chan->n_subbufs; i++)
345 buf->padding[i] = 0;
346
347 buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0);
348}
349
350/**
351 * relay_reset - reset the channel
352 * @chan: the channel
353 *
354 * This has the effect of erasing all data from all channel buffers
355 * and restarting the channel in its initial state. The buffers
356 * are not freed, so any mappings are still in effect.
357 *
358 * NOTE: Care should be taken that the channel isn't actually
359 * being used by anything when this call is made.
360 */
361void relay_reset(struct rchan *chan)
362{
363 unsigned int i;
364 struct rchan_buf *prev = NULL;
365
366 if (!chan)
367 return;
368
369 for (i = 0; i < NR_CPUS; i++) {
370 if (!chan->buf[i] || chan->buf[i] == prev)
371 break;
372 __relay_reset(chan->buf[i], 0);
373 prev = chan->buf[i];
374 }
375}
376EXPORT_SYMBOL_GPL(relay_reset);
377
Randy Dunlap4c78a662006-09-29 01:59:10 -0700378/*
Jens Axboeb86ff9812006-03-23 19:56:55 +0100379 * relay_open_buf - create a new relay channel buffer
380 *
381 * Internal - used by relay_open().
382 */
383static struct rchan_buf *relay_open_buf(struct rchan *chan,
384 const char *filename,
385 struct dentry *parent,
386 int *is_global)
387{
388 struct rchan_buf *buf;
389 struct dentry *dentry;
390
391 if (*is_global)
392 return chan->buf[0];
393
394 buf = relay_create_buf(chan);
395 if (!buf)
396 return NULL;
397
398 /* Create file in fs */
399 dentry = chan->cb->create_buf_file(filename, parent, S_IRUSR,
400 buf, is_global);
401 if (!dentry) {
402 relay_destroy_buf(buf);
403 return NULL;
404 }
405
406 buf->dentry = dentry;
407 __relay_reset(buf, 1);
408
409 return buf;
410}
411
412/**
413 * relay_close_buf - close a channel buffer
414 * @buf: channel buffer
415 *
416 * Marks the buffer finalized and restores the default callbacks.
417 * The channel buffer and channel buffer data structure are then freed
418 * automatically when the last reference is given up.
419 */
420static inline void relay_close_buf(struct rchan_buf *buf)
421{
422 buf->finalized = 1;
423 cancel_delayed_work(&buf->wake_readers);
424 flush_scheduled_work();
425 kref_put(&buf->kref, relay_remove_buf);
426}
427
428static inline void setup_callbacks(struct rchan *chan,
429 struct rchan_callbacks *cb)
430{
431 if (!cb) {
432 chan->cb = &default_channel_callbacks;
433 return;
434 }
435
436 if (!cb->subbuf_start)
437 cb->subbuf_start = subbuf_start_default_callback;
438 if (!cb->buf_mapped)
439 cb->buf_mapped = buf_mapped_default_callback;
440 if (!cb->buf_unmapped)
441 cb->buf_unmapped = buf_unmapped_default_callback;
442 if (!cb->create_buf_file)
443 cb->create_buf_file = create_buf_file_default_callback;
444 if (!cb->remove_buf_file)
445 cb->remove_buf_file = remove_buf_file_default_callback;
446 chan->cb = cb;
447}
448
449/**
450 * relay_open - create a new relay channel
451 * @base_filename: base name of files to create
Randy Dunlap4c78a662006-09-29 01:59:10 -0700452 * @parent: dentry of parent directory, %NULL for root directory
Jens Axboeb86ff9812006-03-23 19:56:55 +0100453 * @subbuf_size: size of sub-buffers
454 * @n_subbufs: number of sub-buffers
455 * @cb: client callback functions
456 *
Randy Dunlap4c78a662006-09-29 01:59:10 -0700457 * Returns channel pointer if successful, %NULL otherwise.
Jens Axboeb86ff9812006-03-23 19:56:55 +0100458 *
459 * Creates a channel buffer for each cpu using the sizes and
460 * attributes specified. The created channel buffer files
461 * will be named base_filename0...base_filenameN-1. File
462 * permissions will be S_IRUSR.
463 */
464struct rchan *relay_open(const char *base_filename,
465 struct dentry *parent,
466 size_t subbuf_size,
467 size_t n_subbufs,
468 struct rchan_callbacks *cb)
469{
470 unsigned int i;
471 struct rchan *chan;
472 char *tmpname;
473 int is_global = 0;
474
475 if (!base_filename)
476 return NULL;
477
478 if (!(subbuf_size && n_subbufs))
479 return NULL;
480
481 chan = kcalloc(1, sizeof(struct rchan), GFP_KERNEL);
482 if (!chan)
483 return NULL;
484
485 chan->version = RELAYFS_CHANNEL_VERSION;
486 chan->n_subbufs = n_subbufs;
487 chan->subbuf_size = subbuf_size;
488 chan->alloc_size = FIX_SIZE(subbuf_size * n_subbufs);
489 setup_callbacks(chan, cb);
490 kref_init(&chan->kref);
491
492 tmpname = kmalloc(NAME_MAX + 1, GFP_KERNEL);
493 if (!tmpname)
494 goto free_chan;
495
496 for_each_online_cpu(i) {
497 sprintf(tmpname, "%s%d", base_filename, i);
498 chan->buf[i] = relay_open_buf(chan, tmpname, parent,
499 &is_global);
500 if (!chan->buf[i])
501 goto free_bufs;
502
503 chan->buf[i]->cpu = i;
504 }
505
506 kfree(tmpname);
507 return chan;
508
509free_bufs:
510 for (i = 0; i < NR_CPUS; i++) {
511 if (!chan->buf[i])
512 break;
513 relay_close_buf(chan->buf[i]);
514 if (is_global)
515 break;
516 }
517 kfree(tmpname);
518
519free_chan:
520 kref_put(&chan->kref, relay_destroy_channel);
521 return NULL;
522}
523EXPORT_SYMBOL_GPL(relay_open);
524
525/**
526 * relay_switch_subbuf - switch to a new sub-buffer
527 * @buf: channel buffer
528 * @length: size of current event
529 *
530 * Returns either the length passed in or 0 if full.
531 *
532 * Performs sub-buffer-switch tasks such as invoking callbacks,
533 * updating padding counts, waking up readers, etc.
534 */
535size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
536{
537 void *old, *new;
538 size_t old_subbuf, new_subbuf;
539
540 if (unlikely(length > buf->chan->subbuf_size))
541 goto toobig;
542
543 if (buf->offset != buf->chan->subbuf_size + 1) {
544 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
545 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
546 buf->padding[old_subbuf] = buf->prev_padding;
547 buf->subbufs_produced++;
Jens Axboe221415d2006-03-23 19:57:55 +0100548 buf->dentry->d_inode->i_size += buf->chan->subbuf_size -
549 buf->padding[old_subbuf];
550 smp_mb();
Jens Axboeb86ff9812006-03-23 19:56:55 +0100551 if (waitqueue_active(&buf->read_wait)) {
552 PREPARE_WORK(&buf->wake_readers, wakeup_readers, buf);
553 schedule_delayed_work(&buf->wake_readers, 1);
554 }
555 }
556
557 old = buf->data;
558 new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
559 new = buf->start + new_subbuf * buf->chan->subbuf_size;
560 buf->offset = 0;
561 if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
562 buf->offset = buf->chan->subbuf_size + 1;
563 return 0;
564 }
565 buf->data = new;
566 buf->padding[new_subbuf] = 0;
567
568 if (unlikely(length + buf->offset > buf->chan->subbuf_size))
569 goto toobig;
570
571 return length;
572
573toobig:
574 buf->chan->last_toobig = length;
575 return 0;
576}
577EXPORT_SYMBOL_GPL(relay_switch_subbuf);
578
579/**
580 * relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
581 * @chan: the channel
582 * @cpu: the cpu associated with the channel buffer to update
583 * @subbufs_consumed: number of sub-buffers to add to current buf's count
584 *
585 * Adds to the channel buffer's consumed sub-buffer count.
586 * subbufs_consumed should be the number of sub-buffers newly consumed,
587 * not the total consumed.
588 *
Randy Dunlap4c78a662006-09-29 01:59:10 -0700589 * NOTE: Kernel clients don't need to call this function if the channel
Jens Axboeb86ff9812006-03-23 19:56:55 +0100590 * mode is 'overwrite'.
591 */
592void relay_subbufs_consumed(struct rchan *chan,
593 unsigned int cpu,
594 size_t subbufs_consumed)
595{
596 struct rchan_buf *buf;
597
598 if (!chan)
599 return;
600
601 if (cpu >= NR_CPUS || !chan->buf[cpu])
602 return;
603
604 buf = chan->buf[cpu];
605 buf->subbufs_consumed += subbufs_consumed;
606 if (buf->subbufs_consumed > buf->subbufs_produced)
607 buf->subbufs_consumed = buf->subbufs_produced;
608}
609EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
610
611/**
612 * relay_close - close the channel
613 * @chan: the channel
614 *
615 * Closes all channel buffers and frees the channel.
616 */
617void relay_close(struct rchan *chan)
618{
619 unsigned int i;
620 struct rchan_buf *prev = NULL;
621
622 if (!chan)
623 return;
624
625 for (i = 0; i < NR_CPUS; i++) {
626 if (!chan->buf[i] || chan->buf[i] == prev)
627 break;
628 relay_close_buf(chan->buf[i]);
629 prev = chan->buf[i];
630 }
631
632 if (chan->last_toobig)
633 printk(KERN_WARNING "relay: one or more items not logged "
634 "[item size (%Zd) > sub-buffer size (%Zd)]\n",
635 chan->last_toobig, chan->subbuf_size);
636
637 kref_put(&chan->kref, relay_destroy_channel);
638}
639EXPORT_SYMBOL_GPL(relay_close);
640
641/**
642 * relay_flush - close the channel
643 * @chan: the channel
644 *
Randy Dunlap4c78a662006-09-29 01:59:10 -0700645 * Flushes all channel buffers, i.e. forces buffer switch.
Jens Axboeb86ff9812006-03-23 19:56:55 +0100646 */
647void relay_flush(struct rchan *chan)
648{
649 unsigned int i;
650 struct rchan_buf *prev = NULL;
651
652 if (!chan)
653 return;
654
655 for (i = 0; i < NR_CPUS; i++) {
656 if (!chan->buf[i] || chan->buf[i] == prev)
657 break;
658 relay_switch_subbuf(chan->buf[i], 0);
659 prev = chan->buf[i];
660 }
661}
662EXPORT_SYMBOL_GPL(relay_flush);
663
664/**
665 * relay_file_open - open file op for relay files
666 * @inode: the inode
667 * @filp: the file
668 *
669 * Increments the channel buffer refcount.
670 */
671static int relay_file_open(struct inode *inode, struct file *filp)
672{
Theodore Ts'o8e18e292006-09-27 01:50:46 -0700673 struct rchan_buf *buf = inode->i_private;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100674 kref_get(&buf->kref);
675 filp->private_data = buf;
676
677 return 0;
678}
679
680/**
681 * relay_file_mmap - mmap file op for relay files
682 * @filp: the file
683 * @vma: the vma describing what to map
684 *
685 * Calls upon relay_mmap_buf to map the file into user space.
686 */
687static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
688{
689 struct rchan_buf *buf = filp->private_data;
690 return relay_mmap_buf(buf, vma);
691}
692
693/**
694 * relay_file_poll - poll file op for relay files
695 * @filp: the file
696 * @wait: poll table
697 *
698 * Poll implemention.
699 */
700static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
701{
702 unsigned int mask = 0;
703 struct rchan_buf *buf = filp->private_data;
704
705 if (buf->finalized)
706 return POLLERR;
707
708 if (filp->f_mode & FMODE_READ) {
709 poll_wait(filp, &buf->read_wait, wait);
710 if (!relay_buf_empty(buf))
711 mask |= POLLIN | POLLRDNORM;
712 }
713
714 return mask;
715}
716
717/**
718 * relay_file_release - release file op for relay files
719 * @inode: the inode
720 * @filp: the file
721 *
722 * Decrements the channel refcount, as the filesystem is
723 * no longer using it.
724 */
725static int relay_file_release(struct inode *inode, struct file *filp)
726{
727 struct rchan_buf *buf = filp->private_data;
728 kref_put(&buf->kref, relay_remove_buf);
729
730 return 0;
731}
732
Randy Dunlap4c78a662006-09-29 01:59:10 -0700733/*
Jens Axboeb86ff9812006-03-23 19:56:55 +0100734 * relay_file_read_consume - update the consumed count for the buffer
735 */
736static void relay_file_read_consume(struct rchan_buf *buf,
737 size_t read_pos,
738 size_t bytes_consumed)
739{
740 size_t subbuf_size = buf->chan->subbuf_size;
741 size_t n_subbufs = buf->chan->n_subbufs;
742 size_t read_subbuf;
743
744 if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
745 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
746 buf->bytes_consumed = 0;
747 }
748
749 buf->bytes_consumed += bytes_consumed;
750 read_subbuf = read_pos / buf->chan->subbuf_size;
751 if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
752 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
753 (buf->offset == subbuf_size))
754 return;
755 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
756 buf->bytes_consumed = 0;
757 }
758}
759
Randy Dunlap4c78a662006-09-29 01:59:10 -0700760/*
Jens Axboeb86ff9812006-03-23 19:56:55 +0100761 * relay_file_read_avail - boolean, are there unconsumed bytes available?
762 */
763static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
764{
Jens Axboeb86ff9812006-03-23 19:56:55 +0100765 size_t subbuf_size = buf->chan->subbuf_size;
766 size_t n_subbufs = buf->chan->n_subbufs;
Jens Axboe221415d2006-03-23 19:57:55 +0100767 size_t produced = buf->subbufs_produced;
768 size_t consumed = buf->subbufs_consumed;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100769
770 relay_file_read_consume(buf, read_pos, 0);
771
Jens Axboe221415d2006-03-23 19:57:55 +0100772 if (unlikely(buf->offset > subbuf_size)) {
773 if (produced == consumed)
774 return 0;
775 return 1;
776 }
777
778 if (unlikely(produced - consumed >= n_subbufs)) {
779 consumed = (produced / n_subbufs) * n_subbufs;
780 buf->subbufs_consumed = consumed;
781 }
782
783 produced = (produced % n_subbufs) * subbuf_size + buf->offset;
784 consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
785
786 if (consumed > produced)
787 produced += n_subbufs * subbuf_size;
788
789 if (consumed == produced)
790 return 0;
791
Jens Axboeb86ff9812006-03-23 19:56:55 +0100792 return 1;
793}
794
795/**
796 * relay_file_read_subbuf_avail - return bytes available in sub-buffer
Randy Dunlap4c78a662006-09-29 01:59:10 -0700797 * @read_pos: file read position
798 * @buf: relay channel buffer
Jens Axboeb86ff9812006-03-23 19:56:55 +0100799 */
800static size_t relay_file_read_subbuf_avail(size_t read_pos,
801 struct rchan_buf *buf)
802{
803 size_t padding, avail = 0;
804 size_t read_subbuf, read_offset, write_subbuf, write_offset;
805 size_t subbuf_size = buf->chan->subbuf_size;
806
807 write_subbuf = (buf->data - buf->start) / subbuf_size;
808 write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
809 read_subbuf = read_pos / subbuf_size;
810 read_offset = read_pos % subbuf_size;
811 padding = buf->padding[read_subbuf];
812
813 if (read_subbuf == write_subbuf) {
814 if (read_offset + padding < write_offset)
815 avail = write_offset - (read_offset + padding);
816 } else
817 avail = (subbuf_size - padding) - read_offset;
818
819 return avail;
820}
821
822/**
823 * relay_file_read_start_pos - find the first available byte to read
Randy Dunlap4c78a662006-09-29 01:59:10 -0700824 * @read_pos: file read position
825 * @buf: relay channel buffer
Jens Axboeb86ff9812006-03-23 19:56:55 +0100826 *
827 * If the read_pos is in the middle of padding, return the
828 * position of the first actually available byte, otherwise
829 * return the original value.
830 */
831static size_t relay_file_read_start_pos(size_t read_pos,
832 struct rchan_buf *buf)
833{
834 size_t read_subbuf, padding, padding_start, padding_end;
835 size_t subbuf_size = buf->chan->subbuf_size;
836 size_t n_subbufs = buf->chan->n_subbufs;
837
838 read_subbuf = read_pos / subbuf_size;
839 padding = buf->padding[read_subbuf];
840 padding_start = (read_subbuf + 1) * subbuf_size - padding;
841 padding_end = (read_subbuf + 1) * subbuf_size;
842 if (read_pos >= padding_start && read_pos < padding_end) {
843 read_subbuf = (read_subbuf + 1) % n_subbufs;
844 read_pos = read_subbuf * subbuf_size;
845 }
846
847 return read_pos;
848}
849
850/**
851 * relay_file_read_end_pos - return the new read position
Randy Dunlap4c78a662006-09-29 01:59:10 -0700852 * @read_pos: file read position
853 * @buf: relay channel buffer
854 * @count: number of bytes to be read
Jens Axboeb86ff9812006-03-23 19:56:55 +0100855 */
856static size_t relay_file_read_end_pos(struct rchan_buf *buf,
857 size_t read_pos,
858 size_t count)
859{
860 size_t read_subbuf, padding, end_pos;
861 size_t subbuf_size = buf->chan->subbuf_size;
862 size_t n_subbufs = buf->chan->n_subbufs;
863
864 read_subbuf = read_pos / subbuf_size;
865 padding = buf->padding[read_subbuf];
866 if (read_pos % subbuf_size + count + padding == subbuf_size)
867 end_pos = (read_subbuf + 1) * subbuf_size;
868 else
869 end_pos = read_pos + count;
870 if (end_pos >= subbuf_size * n_subbufs)
871 end_pos = 0;
872
873 return end_pos;
874}
875
Randy Dunlap4c78a662006-09-29 01:59:10 -0700876/*
Tom Zanussi6dac40a2006-03-23 19:58:45 +0100877 * subbuf_read_actor - read up to one subbuf's worth of data
Jens Axboeb86ff9812006-03-23 19:56:55 +0100878 */
Tom Zanussi6dac40a2006-03-23 19:58:45 +0100879static int subbuf_read_actor(size_t read_start,
880 struct rchan_buf *buf,
881 size_t avail,
882 read_descriptor_t *desc,
883 read_actor_t actor)
Jens Axboeb86ff9812006-03-23 19:56:55 +0100884{
Jens Axboeb86ff9812006-03-23 19:56:55 +0100885 void *from;
Tom Zanussi6dac40a2006-03-23 19:58:45 +0100886 int ret = 0;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100887
888 from = buf->start + read_start;
Tom Zanussi6dac40a2006-03-23 19:58:45 +0100889 ret = avail;
Al Viroba2397e2006-10-10 22:46:57 +0100890 if (copy_to_user(desc->arg.buf, from, avail)) {
Tom Zanussi6dac40a2006-03-23 19:58:45 +0100891 desc->error = -EFAULT;
892 ret = 0;
Jens Axboeb86ff9812006-03-23 19:56:55 +0100893 }
Tom Zanussi6dac40a2006-03-23 19:58:45 +0100894 desc->arg.data += ret;
895 desc->written += ret;
896 desc->count -= ret;
897
Jens Axboeb86ff9812006-03-23 19:56:55 +0100898 return ret;
899}
900
Randy Dunlap4c78a662006-09-29 01:59:10 -0700901/*
Tom Zanussi6dac40a2006-03-23 19:58:45 +0100902 * subbuf_send_actor - send up to one subbuf's worth of data
903 */
904static int subbuf_send_actor(size_t read_start,
905 struct rchan_buf *buf,
906 size_t avail,
907 read_descriptor_t *desc,
908 read_actor_t actor)
Jens Axboe221415d2006-03-23 19:57:55 +0100909{
Jens Axboe221415d2006-03-23 19:57:55 +0100910 unsigned long pidx, poff;
911 unsigned int subbuf_pages;
Tom Zanussi6dac40a2006-03-23 19:58:45 +0100912 int ret = 0;
Jens Axboe221415d2006-03-23 19:57:55 +0100913
Tom Zanussi6dac40a2006-03-23 19:58:45 +0100914 subbuf_pages = buf->chan->alloc_size >> PAGE_SHIFT;
915 pidx = (read_start / PAGE_SIZE) % subbuf_pages;
916 poff = read_start & ~PAGE_MASK;
917 while (avail) {
918 struct page *p = buf->page_array[pidx];
919 unsigned int len;
920
921 len = PAGE_SIZE - poff;
922 if (len > avail)
923 len = avail;
924
925 len = actor(desc, p, poff, len);
926 if (desc->error)
927 break;
928
929 avail -= len;
930 ret += len;
931 poff = 0;
932 pidx = (pidx + 1) % subbuf_pages;
933 }
934
935 return ret;
936}
937
938typedef int (*subbuf_actor_t) (size_t read_start,
939 struct rchan_buf *buf,
940 size_t avail,
941 read_descriptor_t *desc,
942 read_actor_t actor);
943
Randy Dunlap4c78a662006-09-29 01:59:10 -0700944/*
Tom Zanussi6dac40a2006-03-23 19:58:45 +0100945 * relay_file_read_subbufs - read count bytes, bridging subbuf boundaries
946 */
947static inline ssize_t relay_file_read_subbufs(struct file *filp,
948 loff_t *ppos,
Tom Zanussi6dac40a2006-03-23 19:58:45 +0100949 subbuf_actor_t subbuf_actor,
950 read_actor_t actor,
Al Viroba2397e2006-10-10 22:46:57 +0100951 read_descriptor_t *desc)
Tom Zanussi6dac40a2006-03-23 19:58:45 +0100952{
953 struct rchan_buf *buf = filp->private_data;
954 size_t read_start, avail;
Tom Zanussi6dac40a2006-03-23 19:58:45 +0100955 int ret;
956
Al Viroba2397e2006-10-10 22:46:57 +0100957 if (!desc->count)
Jens Axboe221415d2006-03-23 19:57:55 +0100958 return 0;
959
Tom Zanussi6dac40a2006-03-23 19:58:45 +0100960 mutex_lock(&filp->f_dentry->d_inode->i_mutex);
961 do {
962 if (!relay_file_read_avail(buf, *ppos))
Jens Axboe221415d2006-03-23 19:57:55 +0100963 break;
Tom Zanussi6dac40a2006-03-23 19:58:45 +0100964
965 read_start = relay_file_read_start_pos(*ppos, buf);
966 avail = relay_file_read_subbuf_avail(read_start, buf);
967 if (!avail)
968 break;
969
Al Viroba2397e2006-10-10 22:46:57 +0100970 avail = min(desc->count, avail);
971 ret = subbuf_actor(read_start, buf, avail, desc, actor);
972 if (desc->error < 0)
Tom Zanussi6dac40a2006-03-23 19:58:45 +0100973 break;
974
975 if (ret) {
976 relay_file_read_consume(buf, read_start, ret);
977 *ppos = relay_file_read_end_pos(buf, read_start, ret);
Jens Axboe221415d2006-03-23 19:57:55 +0100978 }
Al Viroba2397e2006-10-10 22:46:57 +0100979 } while (desc->count && ret);
Tom Zanussi6dac40a2006-03-23 19:58:45 +0100980 mutex_unlock(&filp->f_dentry->d_inode->i_mutex);
Jens Axboe221415d2006-03-23 19:57:55 +0100981
Al Viroba2397e2006-10-10 22:46:57 +0100982 return desc->written;
Jens Axboe221415d2006-03-23 19:57:55 +0100983}
984
Tom Zanussi6dac40a2006-03-23 19:58:45 +0100985static ssize_t relay_file_read(struct file *filp,
986 char __user *buffer,
987 size_t count,
988 loff_t *ppos)
989{
Al Viroba2397e2006-10-10 22:46:57 +0100990 read_descriptor_t desc;
991 desc.written = 0;
992 desc.count = count;
993 desc.arg.buf = buffer;
994 desc.error = 0;
995 return relay_file_read_subbufs(filp, ppos, subbuf_read_actor,
996 NULL, &desc);
Tom Zanussi6dac40a2006-03-23 19:58:45 +0100997}
998
999static ssize_t relay_file_sendfile(struct file *filp,
1000 loff_t *ppos,
1001 size_t count,
1002 read_actor_t actor,
Jens Axboe221415d2006-03-23 19:57:55 +01001003 void *target)
1004{
Al Viroba2397e2006-10-10 22:46:57 +01001005 read_descriptor_t desc;
1006 desc.written = 0;
1007 desc.count = count;
1008 desc.arg.data = target;
1009 desc.error = 0;
1010 return relay_file_read_subbufs(filp, ppos, subbuf_send_actor,
1011 actor, &desc);
Jens Axboe221415d2006-03-23 19:57:55 +01001012}
1013
Jens Axboeb86ff9812006-03-23 19:56:55 +01001014struct file_operations relay_file_operations = {
1015 .open = relay_file_open,
1016 .poll = relay_file_poll,
1017 .mmap = relay_file_mmap,
1018 .read = relay_file_read,
1019 .llseek = no_llseek,
1020 .release = relay_file_release,
Jens Axboe221415d2006-03-23 19:57:55 +01001021 .sendfile = relay_file_sendfile,
Jens Axboeb86ff9812006-03-23 19:56:55 +01001022};
1023EXPORT_SYMBOL_GPL(relay_file_operations);