[PATCH] relay: migrate from relayfs to a generic relay API
[linux-2.6.git] / kernel / relay.c
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  *
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  */
25 static 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  */
34 static 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  */
61 static 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  */
75 int 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  *
98  *      Returns a pointer to the resulting buffer, NULL if unsuccessful
99  */
100 static void *relay_alloc_buf(struct rchan_buf *buf, unsigned long size)
101 {
102         void *mem;
103         unsigned int i, j, n_pages;
104
105         size = PAGE_ALIGN(size);
106         n_pages = size >> PAGE_SHIFT;
107
108         buf->page_array = kcalloc(n_pages, sizeof(struct page *), GFP_KERNEL);
109         if (!buf->page_array)
110                 return NULL;
111
112         for (i = 0; i < n_pages; i++) {
113                 buf->page_array[i] = alloc_page(GFP_KERNEL);
114                 if (unlikely(!buf->page_array[i]))
115                         goto depopulate;
116         }
117         mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL);
118         if (!mem)
119                 goto depopulate;
120
121         memset(mem, 0, size);
122         buf->page_count = n_pages;
123         return mem;
124
125 depopulate:
126         for (j = 0; j < i; j++)
127                 __free_page(buf->page_array[j]);
128         kfree(buf->page_array);
129         return NULL;
130 }
131
132 /**
133  *      relay_create_buf - allocate and initialize a channel buffer
134  *      @alloc_size: size of the buffer to allocate
135  *      @n_subbufs: number of sub-buffers in the channel
136  *
137  *      Returns channel buffer if successful, NULL otherwise
138  */
139 struct 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
149         buf->start = relay_alloc_buf(buf, chan->alloc_size);
150         if (!buf->start)
151                 goto free_buf;
152
153         buf->chan = chan;
154         kref_get(&buf->chan->kref);
155         return buf;
156
157 free_buf:
158         kfree(buf->padding);
159         kfree(buf);
160         return NULL;
161 }
162
163 /**
164  *      relay_destroy_channel - free the channel struct
165  *
166  *      Should only be called from kref_put().
167  */
168 void relay_destroy_channel(struct kref *kref)
169 {
170         struct rchan *chan = container_of(kref, struct rchan, kref);
171         kfree(chan);
172 }
173
174 /**
175  *      relay_destroy_buf - destroy an rchan_buf struct and associated buffer
176  *      @buf: the buffer struct
177  */
178 void relay_destroy_buf(struct rchan_buf *buf)
179 {
180         struct rchan *chan = buf->chan;
181         unsigned int i;
182
183         if (likely(buf->start)) {
184                 vunmap(buf->start);
185                 for (i = 0; i < buf->page_count; i++)
186                         __free_page(buf->page_array[i]);
187                 kfree(buf->page_array);
188         }
189         kfree(buf->padding);
190         kfree(buf);
191         kref_put(&chan->kref, relay_destroy_channel);
192 }
193
194 /**
195  *      relay_remove_buf - remove a channel buffer
196  *
197  *      Removes the file from the fileystem, which also frees the
198  *      rchan_buf_struct and the channel buffer.  Should only be called from
199  *      kref_put().
200  */
201 void relay_remove_buf(struct kref *kref)
202 {
203         struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
204         buf->chan->cb->remove_buf_file(buf->dentry);
205         relay_destroy_buf(buf);
206 }
207
208 /**
209  *      relay_buf_empty - boolean, is the channel buffer empty?
210  *      @buf: channel buffer
211  *
212  *      Returns 1 if the buffer is empty, 0 otherwise.
213  */
214 int relay_buf_empty(struct rchan_buf *buf)
215 {
216         return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1;
217 }
218 EXPORT_SYMBOL_GPL(relay_buf_empty);
219
220 /**
221  *      relay_buf_full - boolean, is the channel buffer full?
222  *      @buf: channel buffer
223  *
224  *      Returns 1 if the buffer is full, 0 otherwise.
225  */
226 int relay_buf_full(struct rchan_buf *buf)
227 {
228         size_t ready = buf->subbufs_produced - buf->subbufs_consumed;
229         return (ready >= buf->chan->n_subbufs) ? 1 : 0;
230 }
231 EXPORT_SYMBOL_GPL(relay_buf_full);
232
233 /*
234  * High-level relay kernel API and associated functions.
235  */
236
237 /*
238  * rchan_callback implementations defining default channel behavior.  Used
239  * in place of corresponding NULL values in client callback struct.
240  */
241
242 /*
243  * subbuf_start() default callback.  Does nothing.
244  */
245 static int subbuf_start_default_callback (struct rchan_buf *buf,
246                                           void *subbuf,
247                                           void *prev_subbuf,
248                                           size_t prev_padding)
249 {
250         if (relay_buf_full(buf))
251                 return 0;
252
253         return 1;
254 }
255
256 /*
257  * buf_mapped() default callback.  Does nothing.
258  */
259 static void buf_mapped_default_callback(struct rchan_buf *buf,
260                                         struct file *filp)
261 {
262 }
263
264 /*
265  * buf_unmapped() default callback.  Does nothing.
266  */
267 static void buf_unmapped_default_callback(struct rchan_buf *buf,
268                                           struct file *filp)
269 {
270 }
271
272 /*
273  * create_buf_file_create() default callback.  Does nothing.
274  */
275 static struct dentry *create_buf_file_default_callback(const char *filename,
276                                                        struct dentry *parent,
277                                                        int mode,
278                                                        struct rchan_buf *buf,
279                                                        int *is_global)
280 {
281         return NULL;
282 }
283
284 /*
285  * remove_buf_file() default callback.  Does nothing.
286  */
287 static int remove_buf_file_default_callback(struct dentry *dentry)
288 {
289         return -EINVAL;
290 }
291
292 /* relay channel default callbacks */
293 static struct rchan_callbacks default_channel_callbacks = {
294         .subbuf_start = subbuf_start_default_callback,
295         .buf_mapped = buf_mapped_default_callback,
296         .buf_unmapped = buf_unmapped_default_callback,
297         .create_buf_file = create_buf_file_default_callback,
298         .remove_buf_file = remove_buf_file_default_callback,
299 };
300
301 /**
302  *      wakeup_readers - wake up readers waiting on a channel
303  *      @private: the channel buffer
304  *
305  *      This is the work function used to defer reader waking.  The
306  *      reason waking is deferred is that calling directly from write
307  *      causes problems if you're writing from say the scheduler.
308  */
309 static void wakeup_readers(void *private)
310 {
311         struct rchan_buf *buf = private;
312         wake_up_interruptible(&buf->read_wait);
313 }
314
315 /**
316  *      __relay_reset - reset a channel buffer
317  *      @buf: the channel buffer
318  *      @init: 1 if this is a first-time initialization
319  *
320  *      See relay_reset for description of effect.
321  */
322 static inline void __relay_reset(struct rchan_buf *buf, unsigned int init)
323 {
324         size_t i;
325
326         if (init) {
327                 init_waitqueue_head(&buf->read_wait);
328                 kref_init(&buf->kref);
329                 INIT_WORK(&buf->wake_readers, NULL, NULL);
330         } else {
331                 cancel_delayed_work(&buf->wake_readers);
332                 flush_scheduled_work();
333         }
334
335         buf->subbufs_produced = 0;
336         buf->subbufs_consumed = 0;
337         buf->bytes_consumed = 0;
338         buf->finalized = 0;
339         buf->data = buf->start;
340         buf->offset = 0;
341
342         for (i = 0; i < buf->chan->n_subbufs; i++)
343                 buf->padding[i] = 0;
344
345         buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0);
346 }
347
348 /**
349  *      relay_reset - reset the channel
350  *      @chan: the channel
351  *
352  *      This has the effect of erasing all data from all channel buffers
353  *      and restarting the channel in its initial state.  The buffers
354  *      are not freed, so any mappings are still in effect.
355  *
356  *      NOTE: Care should be taken that the channel isn't actually
357  *      being used by anything when this call is made.
358  */
359 void relay_reset(struct rchan *chan)
360 {
361         unsigned int i;
362         struct rchan_buf *prev = NULL;
363
364         if (!chan)
365                 return;
366
367         for (i = 0; i < NR_CPUS; i++) {
368                 if (!chan->buf[i] || chan->buf[i] == prev)
369                         break;
370                 __relay_reset(chan->buf[i], 0);
371                 prev = chan->buf[i];
372         }
373 }
374 EXPORT_SYMBOL_GPL(relay_reset);
375
376 /**
377  *      relay_open_buf - create a new relay channel buffer
378  *
379  *      Internal - used by relay_open().
380  */
381 static struct rchan_buf *relay_open_buf(struct rchan *chan,
382                                         const char *filename,
383                                         struct dentry *parent,
384                                         int *is_global)
385 {
386         struct rchan_buf *buf;
387         struct dentry *dentry;
388
389         if (*is_global)
390                 return chan->buf[0];
391
392         buf = relay_create_buf(chan);
393         if (!buf)
394                 return NULL;
395
396         /* Create file in fs */
397         dentry = chan->cb->create_buf_file(filename, parent, S_IRUSR,
398                                            buf, is_global);
399         if (!dentry) {
400                 relay_destroy_buf(buf);
401                 return NULL;
402         }
403
404         buf->dentry = dentry;
405         __relay_reset(buf, 1);
406
407         return buf;
408 }
409
410 /**
411  *      relay_close_buf - close a channel buffer
412  *      @buf: channel buffer
413  *
414  *      Marks the buffer finalized and restores the default callbacks.
415  *      The channel buffer and channel buffer data structure are then freed
416  *      automatically when the last reference is given up.
417  */
418 static inline void relay_close_buf(struct rchan_buf *buf)
419 {
420         buf->finalized = 1;
421         cancel_delayed_work(&buf->wake_readers);
422         flush_scheduled_work();
423         kref_put(&buf->kref, relay_remove_buf);
424 }
425
426 static inline void setup_callbacks(struct rchan *chan,
427                                    struct rchan_callbacks *cb)
428 {
429         if (!cb) {
430                 chan->cb = &default_channel_callbacks;
431                 return;
432         }
433
434         if (!cb->subbuf_start)
435                 cb->subbuf_start = subbuf_start_default_callback;
436         if (!cb->buf_mapped)
437                 cb->buf_mapped = buf_mapped_default_callback;
438         if (!cb->buf_unmapped)
439                 cb->buf_unmapped = buf_unmapped_default_callback;
440         if (!cb->create_buf_file)
441                 cb->create_buf_file = create_buf_file_default_callback;
442         if (!cb->remove_buf_file)
443                 cb->remove_buf_file = remove_buf_file_default_callback;
444         chan->cb = cb;
445 }
446
447 /**
448  *      relay_open - create a new relay channel
449  *      @base_filename: base name of files to create
450  *      @parent: dentry of parent directory, NULL for root directory
451  *      @subbuf_size: size of sub-buffers
452  *      @n_subbufs: number of sub-buffers
453  *      @cb: client callback functions
454  *
455  *      Returns channel pointer if successful, NULL otherwise.
456  *
457  *      Creates a channel buffer for each cpu using the sizes and
458  *      attributes specified.  The created channel buffer files
459  *      will be named base_filename0...base_filenameN-1.  File
460  *      permissions will be S_IRUSR.
461  */
462 struct rchan *relay_open(const char *base_filename,
463                          struct dentry *parent,
464                          size_t subbuf_size,
465                          size_t n_subbufs,
466                          struct rchan_callbacks *cb)
467 {
468         unsigned int i;
469         struct rchan *chan;
470         char *tmpname;
471         int is_global = 0;
472
473         if (!base_filename)
474                 return NULL;
475
476         if (!(subbuf_size && n_subbufs))
477                 return NULL;
478
479         chan = kcalloc(1, sizeof(struct rchan), GFP_KERNEL);
480         if (!chan)
481                 return NULL;
482
483         chan->version = RELAYFS_CHANNEL_VERSION;
484         chan->n_subbufs = n_subbufs;
485         chan->subbuf_size = subbuf_size;
486         chan->alloc_size = FIX_SIZE(subbuf_size * n_subbufs);
487         setup_callbacks(chan, cb);
488         kref_init(&chan->kref);
489
490         tmpname = kmalloc(NAME_MAX + 1, GFP_KERNEL);
491         if (!tmpname)
492                 goto free_chan;
493
494         for_each_online_cpu(i) {
495                 sprintf(tmpname, "%s%d", base_filename, i);
496                 chan->buf[i] = relay_open_buf(chan, tmpname, parent,
497                                               &is_global);
498                 if (!chan->buf[i])
499                         goto free_bufs;
500
501                 chan->buf[i]->cpu = i;
502         }
503
504         kfree(tmpname);
505         return chan;
506
507 free_bufs:
508         for (i = 0; i < NR_CPUS; i++) {
509                 if (!chan->buf[i])
510                         break;
511                 relay_close_buf(chan->buf[i]);
512                 if (is_global)
513                         break;
514         }
515         kfree(tmpname);
516
517 free_chan:
518         kref_put(&chan->kref, relay_destroy_channel);
519         return NULL;
520 }
521 EXPORT_SYMBOL_GPL(relay_open);
522
523 /**
524  *      relay_switch_subbuf - switch to a new sub-buffer
525  *      @buf: channel buffer
526  *      @length: size of current event
527  *
528  *      Returns either the length passed in or 0 if full.
529  *
530  *      Performs sub-buffer-switch tasks such as invoking callbacks,
531  *      updating padding counts, waking up readers, etc.
532  */
533 size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
534 {
535         void *old, *new;
536         size_t old_subbuf, new_subbuf;
537
538         if (unlikely(length > buf->chan->subbuf_size))
539                 goto toobig;
540
541         if (buf->offset != buf->chan->subbuf_size + 1) {
542                 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
543                 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
544                 buf->padding[old_subbuf] = buf->prev_padding;
545                 buf->subbufs_produced++;
546                 if (waitqueue_active(&buf->read_wait)) {
547                         PREPARE_WORK(&buf->wake_readers, wakeup_readers, buf);
548                         schedule_delayed_work(&buf->wake_readers, 1);
549                 }
550         }
551
552         old = buf->data;
553         new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
554         new = buf->start + new_subbuf * buf->chan->subbuf_size;
555         buf->offset = 0;
556         if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
557                 buf->offset = buf->chan->subbuf_size + 1;
558                 return 0;
559         }
560         buf->data = new;
561         buf->padding[new_subbuf] = 0;
562
563         if (unlikely(length + buf->offset > buf->chan->subbuf_size))
564                 goto toobig;
565
566         return length;
567
568 toobig:
569         buf->chan->last_toobig = length;
570         return 0;
571 }
572 EXPORT_SYMBOL_GPL(relay_switch_subbuf);
573
574 /**
575  *      relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
576  *      @chan: the channel
577  *      @cpu: the cpu associated with the channel buffer to update
578  *      @subbufs_consumed: number of sub-buffers to add to current buf's count
579  *
580  *      Adds to the channel buffer's consumed sub-buffer count.
581  *      subbufs_consumed should be the number of sub-buffers newly consumed,
582  *      not the total consumed.
583  *
584  *      NOTE: kernel clients don't need to call this function if the channel
585  *      mode is 'overwrite'.
586  */
587 void relay_subbufs_consumed(struct rchan *chan,
588                             unsigned int cpu,
589                             size_t subbufs_consumed)
590 {
591         struct rchan_buf *buf;
592
593         if (!chan)
594                 return;
595
596         if (cpu >= NR_CPUS || !chan->buf[cpu])
597                 return;
598
599         buf = chan->buf[cpu];
600         buf->subbufs_consumed += subbufs_consumed;
601         if (buf->subbufs_consumed > buf->subbufs_produced)
602                 buf->subbufs_consumed = buf->subbufs_produced;
603 }
604 EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
605
606 /**
607  *      relay_close - close the channel
608  *      @chan: the channel
609  *
610  *      Closes all channel buffers and frees the channel.
611  */
612 void relay_close(struct rchan *chan)
613 {
614         unsigned int i;
615         struct rchan_buf *prev = NULL;
616
617         if (!chan)
618                 return;
619
620         for (i = 0; i < NR_CPUS; i++) {
621                 if (!chan->buf[i] || chan->buf[i] == prev)
622                         break;
623                 relay_close_buf(chan->buf[i]);
624                 prev = chan->buf[i];
625         }
626
627         if (chan->last_toobig)
628                 printk(KERN_WARNING "relay: one or more items not logged "
629                        "[item size (%Zd) > sub-buffer size (%Zd)]\n",
630                        chan->last_toobig, chan->subbuf_size);
631
632         kref_put(&chan->kref, relay_destroy_channel);
633 }
634 EXPORT_SYMBOL_GPL(relay_close);
635
636 /**
637  *      relay_flush - close the channel
638  *      @chan: the channel
639  *
640  *      Flushes all channel buffers i.e. forces buffer switch.
641  */
642 void relay_flush(struct rchan *chan)
643 {
644         unsigned int i;
645         struct rchan_buf *prev = NULL;
646
647         if (!chan)
648                 return;
649
650         for (i = 0; i < NR_CPUS; i++) {
651                 if (!chan->buf[i] || chan->buf[i] == prev)
652                         break;
653                 relay_switch_subbuf(chan->buf[i], 0);
654                 prev = chan->buf[i];
655         }
656 }
657 EXPORT_SYMBOL_GPL(relay_flush);
658
659 /**
660  *      relay_file_open - open file op for relay files
661  *      @inode: the inode
662  *      @filp: the file
663  *
664  *      Increments the channel buffer refcount.
665  */
666 static int relay_file_open(struct inode *inode, struct file *filp)
667 {
668         struct rchan_buf *buf = inode->u.generic_ip;
669         kref_get(&buf->kref);
670         filp->private_data = buf;
671
672         return 0;
673 }
674
675 /**
676  *      relay_file_mmap - mmap file op for relay files
677  *      @filp: the file
678  *      @vma: the vma describing what to map
679  *
680  *      Calls upon relay_mmap_buf to map the file into user space.
681  */
682 static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
683 {
684         struct rchan_buf *buf = filp->private_data;
685         return relay_mmap_buf(buf, vma);
686 }
687
688 /**
689  *      relay_file_poll - poll file op for relay files
690  *      @filp: the file
691  *      @wait: poll table
692  *
693  *      Poll implemention.
694  */
695 static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
696 {
697         unsigned int mask = 0;
698         struct rchan_buf *buf = filp->private_data;
699
700         if (buf->finalized)
701                 return POLLERR;
702
703         if (filp->f_mode & FMODE_READ) {
704                 poll_wait(filp, &buf->read_wait, wait);
705                 if (!relay_buf_empty(buf))
706                         mask |= POLLIN | POLLRDNORM;
707         }
708
709         return mask;
710 }
711
712 /**
713  *      relay_file_release - release file op for relay files
714  *      @inode: the inode
715  *      @filp: the file
716  *
717  *      Decrements the channel refcount, as the filesystem is
718  *      no longer using it.
719  */
720 static int relay_file_release(struct inode *inode, struct file *filp)
721 {
722         struct rchan_buf *buf = filp->private_data;
723         kref_put(&buf->kref, relay_remove_buf);
724
725         return 0;
726 }
727
728 /**
729  *      relay_file_read_consume - update the consumed count for the buffer
730  */
731 static void relay_file_read_consume(struct rchan_buf *buf,
732                                     size_t read_pos,
733                                     size_t bytes_consumed)
734 {
735         size_t subbuf_size = buf->chan->subbuf_size;
736         size_t n_subbufs = buf->chan->n_subbufs;
737         size_t read_subbuf;
738
739         if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
740                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
741                 buf->bytes_consumed = 0;
742         }
743
744         buf->bytes_consumed += bytes_consumed;
745         read_subbuf = read_pos / buf->chan->subbuf_size;
746         if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
747                 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
748                     (buf->offset == subbuf_size))
749                         return;
750                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
751                 buf->bytes_consumed = 0;
752         }
753 }
754
755 /**
756  *      relay_file_read_avail - boolean, are there unconsumed bytes available?
757  */
758 static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
759 {
760         size_t bytes_produced, bytes_consumed, write_offset;
761         size_t subbuf_size = buf->chan->subbuf_size;
762         size_t n_subbufs = buf->chan->n_subbufs;
763         size_t produced = buf->subbufs_produced % n_subbufs;
764         size_t consumed = buf->subbufs_consumed % n_subbufs;
765
766         write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
767
768         if (consumed > produced) {
769                 if ((produced > n_subbufs) &&
770                     (produced + n_subbufs - consumed <= n_subbufs))
771                         produced += n_subbufs;
772         } else if (consumed == produced) {
773                 if (buf->offset > subbuf_size) {
774                         produced += n_subbufs;
775                         if (buf->subbufs_produced == buf->subbufs_consumed)
776                                 consumed += n_subbufs;
777                 }
778         }
779
780         if (buf->offset > subbuf_size)
781                 bytes_produced = (produced - 1) * subbuf_size + write_offset;
782         else
783                 bytes_produced = produced * subbuf_size + write_offset;
784         bytes_consumed = consumed * subbuf_size + buf->bytes_consumed;
785
786         if (bytes_produced == bytes_consumed)
787                 return 0;
788
789         relay_file_read_consume(buf, read_pos, 0);
790
791         return 1;
792 }
793
794 /**
795  *      relay_file_read_subbuf_avail - return bytes available in sub-buffer
796  */
797 static size_t relay_file_read_subbuf_avail(size_t read_pos,
798                                            struct rchan_buf *buf)
799 {
800         size_t padding, avail = 0;
801         size_t read_subbuf, read_offset, write_subbuf, write_offset;
802         size_t subbuf_size = buf->chan->subbuf_size;
803
804         write_subbuf = (buf->data - buf->start) / subbuf_size;
805         write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
806         read_subbuf = read_pos / subbuf_size;
807         read_offset = read_pos % subbuf_size;
808         padding = buf->padding[read_subbuf];
809
810         if (read_subbuf == write_subbuf) {
811                 if (read_offset + padding < write_offset)
812                         avail = write_offset - (read_offset + padding);
813         } else
814                 avail = (subbuf_size - padding) - read_offset;
815
816         return avail;
817 }
818
819 /**
820  *      relay_file_read_start_pos - find the first available byte to read
821  *
822  *      If the read_pos is in the middle of padding, return the
823  *      position of the first actually available byte, otherwise
824  *      return the original value.
825  */
826 static size_t relay_file_read_start_pos(size_t read_pos,
827                                         struct rchan_buf *buf)
828 {
829         size_t read_subbuf, padding, padding_start, padding_end;
830         size_t subbuf_size = buf->chan->subbuf_size;
831         size_t n_subbufs = buf->chan->n_subbufs;
832
833         read_subbuf = read_pos / subbuf_size;
834         padding = buf->padding[read_subbuf];
835         padding_start = (read_subbuf + 1) * subbuf_size - padding;
836         padding_end = (read_subbuf + 1) * subbuf_size;
837         if (read_pos >= padding_start && read_pos < padding_end) {
838                 read_subbuf = (read_subbuf + 1) % n_subbufs;
839                 read_pos = read_subbuf * subbuf_size;
840         }
841
842         return read_pos;
843 }
844
845 /**
846  *      relay_file_read_end_pos - return the new read position
847  */
848 static size_t relay_file_read_end_pos(struct rchan_buf *buf,
849                                       size_t read_pos,
850                                       size_t count)
851 {
852         size_t read_subbuf, padding, end_pos;
853         size_t subbuf_size = buf->chan->subbuf_size;
854         size_t n_subbufs = buf->chan->n_subbufs;
855
856         read_subbuf = read_pos / subbuf_size;
857         padding = buf->padding[read_subbuf];
858         if (read_pos % subbuf_size + count + padding == subbuf_size)
859                 end_pos = (read_subbuf + 1) * subbuf_size;
860         else
861                 end_pos = read_pos + count;
862         if (end_pos >= subbuf_size * n_subbufs)
863                 end_pos = 0;
864
865         return end_pos;
866 }
867
868 /**
869  *      relay_file_read - read file op for relay files
870  *      @filp: the file
871  *      @buffer: the userspace buffer
872  *      @count: number of bytes to read
873  *      @ppos: position to read from
874  *
875  *      Reads count bytes or the number of bytes available in the
876  *      current sub-buffer being read, whichever is smaller.
877  */
878 static ssize_t relay_file_read(struct file *filp,
879                                char __user *buffer,
880                                size_t count,
881                                loff_t *ppos)
882 {
883         struct rchan_buf *buf = filp->private_data;
884         struct inode *inode = filp->f_dentry->d_inode;
885         size_t read_start, avail;
886         ssize_t ret = 0;
887         void *from;
888
889         mutex_lock(&inode->i_mutex);
890         if(!relay_file_read_avail(buf, *ppos))
891                 goto out;
892
893         read_start = relay_file_read_start_pos(*ppos, buf);
894         avail = relay_file_read_subbuf_avail(read_start, buf);
895         if (!avail)
896                 goto out;
897
898         from = buf->start + read_start;
899         ret = count = min(count, avail);
900         if (copy_to_user(buffer, from, count)) {
901                 ret = -EFAULT;
902                 goto out;
903         }
904         relay_file_read_consume(buf, read_start, count);
905         *ppos = relay_file_read_end_pos(buf, read_start, count);
906 out:
907         mutex_unlock(&inode->i_mutex);
908         return ret;
909 }
910
911 struct file_operations relay_file_operations = {
912         .open           = relay_file_open,
913         .poll           = relay_file_poll,
914         .mmap           = relay_file_mmap,
915         .read           = relay_file_read,
916         .llseek         = no_llseek,
917         .release        = relay_file_release,
918 };
919 EXPORT_SYMBOL_GPL(relay_file_operations);