[PATCH] relay: add sendfile() support
[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. The
99  *      passed in size will get page aligned, if it isn't already.
100  */
101 static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size)
102 {
103         void *mem;
104         unsigned int i, j, n_pages;
105
106         *size = PAGE_ALIGN(*size);
107         n_pages = *size >> PAGE_SHIFT;
108
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
122         memset(mem, 0, *size);
123         buf->page_count = n_pages;
124         return mem;
125
126 depopulate:
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
135  *      @alloc_size: size of the buffer to allocate
136  *      @n_subbufs: number of sub-buffers in the channel
137  *
138  *      Returns channel buffer if successful, NULL otherwise
139  */
140 struct rchan_buf *relay_create_buf(struct rchan *chan)
141 {
142         struct rchan_buf *buf = kcalloc(1, sizeof(struct rchan_buf), GFP_KERNEL);
143         if (!buf)
144                 return NULL;
145
146         buf->padding = kmalloc(chan->n_subbufs * sizeof(size_t *), GFP_KERNEL);
147         if (!buf->padding)
148                 goto free_buf;
149
150         buf->start = relay_alloc_buf(buf, &chan->alloc_size);
151         if (!buf->start)
152                 goto free_buf;
153
154         buf->chan = chan;
155         kref_get(&buf->chan->kref);
156         return buf;
157
158 free_buf:
159         kfree(buf->padding);
160         kfree(buf);
161         return NULL;
162 }
163
164 /**
165  *      relay_destroy_channel - free the channel struct
166  *
167  *      Should only be called from kref_put().
168  */
169 void 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  */
179 void 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
197  *
198  *      Removes the file from the fileystem, which also frees the
199  *      rchan_buf_struct and the channel buffer.  Should only be called from
200  *      kref_put().
201  */
202 void relay_remove_buf(struct kref *kref)
203 {
204         struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
205         buf->chan->cb->remove_buf_file(buf->dentry);
206         relay_destroy_buf(buf);
207 }
208
209 /**
210  *      relay_buf_empty - boolean, is the channel buffer empty?
211  *      @buf: channel buffer
212  *
213  *      Returns 1 if the buffer is empty, 0 otherwise.
214  */
215 int relay_buf_empty(struct rchan_buf *buf)
216 {
217         return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1;
218 }
219 EXPORT_SYMBOL_GPL(relay_buf_empty);
220
221 /**
222  *      relay_buf_full - boolean, is the channel buffer full?
223  *      @buf: channel buffer
224  *
225  *      Returns 1 if the buffer is full, 0 otherwise.
226  */
227 int relay_buf_full(struct rchan_buf *buf)
228 {
229         size_t ready = buf->subbufs_produced - buf->subbufs_consumed;
230         return (ready >= buf->chan->n_subbufs) ? 1 : 0;
231 }
232 EXPORT_SYMBOL_GPL(relay_buf_full);
233
234 /*
235  * High-level relay kernel API and associated functions.
236  */
237
238 /*
239  * rchan_callback implementations defining default channel behavior.  Used
240  * in place of corresponding NULL values in client callback struct.
241  */
242
243 /*
244  * subbuf_start() default callback.  Does nothing.
245  */
246 static int subbuf_start_default_callback (struct rchan_buf *buf,
247                                           void *subbuf,
248                                           void *prev_subbuf,
249                                           size_t prev_padding)
250 {
251         if (relay_buf_full(buf))
252                 return 0;
253
254         return 1;
255 }
256
257 /*
258  * buf_mapped() default callback.  Does nothing.
259  */
260 static void buf_mapped_default_callback(struct rchan_buf *buf,
261                                         struct file *filp)
262 {
263 }
264
265 /*
266  * buf_unmapped() default callback.  Does nothing.
267  */
268 static void buf_unmapped_default_callback(struct rchan_buf *buf,
269                                           struct file *filp)
270 {
271 }
272
273 /*
274  * create_buf_file_create() default callback.  Does nothing.
275  */
276 static struct dentry *create_buf_file_default_callback(const char *filename,
277                                                        struct dentry *parent,
278                                                        int mode,
279                                                        struct rchan_buf *buf,
280                                                        int *is_global)
281 {
282         return NULL;
283 }
284
285 /*
286  * remove_buf_file() default callback.  Does nothing.
287  */
288 static int remove_buf_file_default_callback(struct dentry *dentry)
289 {
290         return -EINVAL;
291 }
292
293 /* relay channel default callbacks */
294 static struct rchan_callbacks default_channel_callbacks = {
295         .subbuf_start = subbuf_start_default_callback,
296         .buf_mapped = buf_mapped_default_callback,
297         .buf_unmapped = buf_unmapped_default_callback,
298         .create_buf_file = create_buf_file_default_callback,
299         .remove_buf_file = remove_buf_file_default_callback,
300 };
301
302 /**
303  *      wakeup_readers - wake up readers waiting on a channel
304  *      @private: the channel buffer
305  *
306  *      This is the work function used to defer reader waking.  The
307  *      reason waking is deferred is that calling directly from write
308  *      causes problems if you're writing from say the scheduler.
309  */
310 static void wakeup_readers(void *private)
311 {
312         struct rchan_buf *buf = private;
313         wake_up_interruptible(&buf->read_wait);
314 }
315
316 /**
317  *      __relay_reset - reset a channel buffer
318  *      @buf: the channel buffer
319  *      @init: 1 if this is a first-time initialization
320  *
321  *      See relay_reset for description of effect.
322  */
323 static inline void __relay_reset(struct rchan_buf *buf, unsigned int init)
324 {
325         size_t i;
326
327         if (init) {
328                 init_waitqueue_head(&buf->read_wait);
329                 kref_init(&buf->kref);
330                 INIT_WORK(&buf->wake_readers, NULL, NULL);
331         } else {
332                 cancel_delayed_work(&buf->wake_readers);
333                 flush_scheduled_work();
334         }
335
336         buf->subbufs_produced = 0;
337         buf->subbufs_consumed = 0;
338         buf->bytes_consumed = 0;
339         buf->finalized = 0;
340         buf->data = buf->start;
341         buf->offset = 0;
342
343         for (i = 0; i < buf->chan->n_subbufs; i++)
344                 buf->padding[i] = 0;
345
346         buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0);
347 }
348
349 /**
350  *      relay_reset - reset the channel
351  *      @chan: the channel
352  *
353  *      This has the effect of erasing all data from all channel buffers
354  *      and restarting the channel in its initial state.  The buffers
355  *      are not freed, so any mappings are still in effect.
356  *
357  *      NOTE: Care should be taken that the channel isn't actually
358  *      being used by anything when this call is made.
359  */
360 void relay_reset(struct rchan *chan)
361 {
362         unsigned int i;
363         struct rchan_buf *prev = NULL;
364
365         if (!chan)
366                 return;
367
368         for (i = 0; i < NR_CPUS; i++) {
369                 if (!chan->buf[i] || chan->buf[i] == prev)
370                         break;
371                 __relay_reset(chan->buf[i], 0);
372                 prev = chan->buf[i];
373         }
374 }
375 EXPORT_SYMBOL_GPL(relay_reset);
376
377 /**
378  *      relay_open_buf - create a new relay channel buffer
379  *
380  *      Internal - used by relay_open().
381  */
382 static struct rchan_buf *relay_open_buf(struct rchan *chan,
383                                         const char *filename,
384                                         struct dentry *parent,
385                                         int *is_global)
386 {
387         struct rchan_buf *buf;
388         struct dentry *dentry;
389
390         if (*is_global)
391                 return chan->buf[0];
392
393         buf = relay_create_buf(chan);
394         if (!buf)
395                 return NULL;
396
397         /* Create file in fs */
398         dentry = chan->cb->create_buf_file(filename, parent, S_IRUSR,
399                                            buf, is_global);
400         if (!dentry) {
401                 relay_destroy_buf(buf);
402                 return NULL;
403         }
404
405         buf->dentry = dentry;
406         __relay_reset(buf, 1);
407
408         return buf;
409 }
410
411 /**
412  *      relay_close_buf - close a channel buffer
413  *      @buf: channel buffer
414  *
415  *      Marks the buffer finalized and restores the default callbacks.
416  *      The channel buffer and channel buffer data structure are then freed
417  *      automatically when the last reference is given up.
418  */
419 static inline void relay_close_buf(struct rchan_buf *buf)
420 {
421         buf->finalized = 1;
422         cancel_delayed_work(&buf->wake_readers);
423         flush_scheduled_work();
424         kref_put(&buf->kref, relay_remove_buf);
425 }
426
427 static inline void setup_callbacks(struct rchan *chan,
428                                    struct rchan_callbacks *cb)
429 {
430         if (!cb) {
431                 chan->cb = &default_channel_callbacks;
432                 return;
433         }
434
435         if (!cb->subbuf_start)
436                 cb->subbuf_start = subbuf_start_default_callback;
437         if (!cb->buf_mapped)
438                 cb->buf_mapped = buf_mapped_default_callback;
439         if (!cb->buf_unmapped)
440                 cb->buf_unmapped = buf_unmapped_default_callback;
441         if (!cb->create_buf_file)
442                 cb->create_buf_file = create_buf_file_default_callback;
443         if (!cb->remove_buf_file)
444                 cb->remove_buf_file = remove_buf_file_default_callback;
445         chan->cb = cb;
446 }
447
448 /**
449  *      relay_open - create a new relay channel
450  *      @base_filename: base name of files to create
451  *      @parent: dentry of parent directory, NULL for root directory
452  *      @subbuf_size: size of sub-buffers
453  *      @n_subbufs: number of sub-buffers
454  *      @cb: client callback functions
455  *
456  *      Returns channel pointer if successful, NULL otherwise.
457  *
458  *      Creates a channel buffer for each cpu using the sizes and
459  *      attributes specified.  The created channel buffer files
460  *      will be named base_filename0...base_filenameN-1.  File
461  *      permissions will be S_IRUSR.
462  */
463 struct rchan *relay_open(const char *base_filename,
464                          struct dentry *parent,
465                          size_t subbuf_size,
466                          size_t n_subbufs,
467                          struct rchan_callbacks *cb)
468 {
469         unsigned int i;
470         struct rchan *chan;
471         char *tmpname;
472         int is_global = 0;
473
474         if (!base_filename)
475                 return NULL;
476
477         if (!(subbuf_size && n_subbufs))
478                 return NULL;
479
480         chan = kcalloc(1, sizeof(struct rchan), GFP_KERNEL);
481         if (!chan)
482                 return NULL;
483
484         chan->version = RELAYFS_CHANNEL_VERSION;
485         chan->n_subbufs = n_subbufs;
486         chan->subbuf_size = subbuf_size;
487         chan->alloc_size = FIX_SIZE(subbuf_size * n_subbufs);
488         setup_callbacks(chan, cb);
489         kref_init(&chan->kref);
490
491         tmpname = kmalloc(NAME_MAX + 1, GFP_KERNEL);
492         if (!tmpname)
493                 goto free_chan;
494
495         for_each_online_cpu(i) {
496                 sprintf(tmpname, "%s%d", base_filename, i);
497                 chan->buf[i] = relay_open_buf(chan, tmpname, parent,
498                                               &is_global);
499                 if (!chan->buf[i])
500                         goto free_bufs;
501
502                 chan->buf[i]->cpu = i;
503         }
504
505         kfree(tmpname);
506         return chan;
507
508 free_bufs:
509         for (i = 0; i < NR_CPUS; i++) {
510                 if (!chan->buf[i])
511                         break;
512                 relay_close_buf(chan->buf[i]);
513                 if (is_global)
514                         break;
515         }
516         kfree(tmpname);
517
518 free_chan:
519         kref_put(&chan->kref, relay_destroy_channel);
520         return NULL;
521 }
522 EXPORT_SYMBOL_GPL(relay_open);
523
524 /**
525  *      relay_switch_subbuf - switch to a new sub-buffer
526  *      @buf: channel buffer
527  *      @length: size of current event
528  *
529  *      Returns either the length passed in or 0 if full.
530  *
531  *      Performs sub-buffer-switch tasks such as invoking callbacks,
532  *      updating padding counts, waking up readers, etc.
533  */
534 size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
535 {
536         void *old, *new;
537         size_t old_subbuf, new_subbuf;
538
539         if (unlikely(length > buf->chan->subbuf_size))
540                 goto toobig;
541
542         if (buf->offset != buf->chan->subbuf_size + 1) {
543                 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
544                 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
545                 buf->padding[old_subbuf] = buf->prev_padding;
546                 buf->subbufs_produced++;
547                 buf->dentry->d_inode->i_size += buf->chan->subbuf_size -
548                         buf->padding[old_subbuf];
549                 smp_mb();
550                 if (waitqueue_active(&buf->read_wait)) {
551                         PREPARE_WORK(&buf->wake_readers, wakeup_readers, buf);
552                         schedule_delayed_work(&buf->wake_readers, 1);
553                 }
554         }
555
556         old = buf->data;
557         new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
558         new = buf->start + new_subbuf * buf->chan->subbuf_size;
559         buf->offset = 0;
560         if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
561                 buf->offset = buf->chan->subbuf_size + 1;
562                 return 0;
563         }
564         buf->data = new;
565         buf->padding[new_subbuf] = 0;
566
567         if (unlikely(length + buf->offset > buf->chan->subbuf_size))
568                 goto toobig;
569
570         return length;
571
572 toobig:
573         buf->chan->last_toobig = length;
574         return 0;
575 }
576 EXPORT_SYMBOL_GPL(relay_switch_subbuf);
577
578 /**
579  *      relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
580  *      @chan: the channel
581  *      @cpu: the cpu associated with the channel buffer to update
582  *      @subbufs_consumed: number of sub-buffers to add to current buf's count
583  *
584  *      Adds to the channel buffer's consumed sub-buffer count.
585  *      subbufs_consumed should be the number of sub-buffers newly consumed,
586  *      not the total consumed.
587  *
588  *      NOTE: kernel clients don't need to call this function if the channel
589  *      mode is 'overwrite'.
590  */
591 void relay_subbufs_consumed(struct rchan *chan,
592                             unsigned int cpu,
593                             size_t subbufs_consumed)
594 {
595         struct rchan_buf *buf;
596
597         if (!chan)
598                 return;
599
600         if (cpu >= NR_CPUS || !chan->buf[cpu])
601                 return;
602
603         buf = chan->buf[cpu];
604         buf->subbufs_consumed += subbufs_consumed;
605         if (buf->subbufs_consumed > buf->subbufs_produced)
606                 buf->subbufs_consumed = buf->subbufs_produced;
607 }
608 EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
609
610 /**
611  *      relay_close - close the channel
612  *      @chan: the channel
613  *
614  *      Closes all channel buffers and frees the channel.
615  */
616 void relay_close(struct rchan *chan)
617 {
618         unsigned int i;
619         struct rchan_buf *prev = NULL;
620
621         if (!chan)
622                 return;
623
624         for (i = 0; i < NR_CPUS; i++) {
625                 if (!chan->buf[i] || chan->buf[i] == prev)
626                         break;
627                 relay_close_buf(chan->buf[i]);
628                 prev = chan->buf[i];
629         }
630
631         if (chan->last_toobig)
632                 printk(KERN_WARNING "relay: one or more items not logged "
633                        "[item size (%Zd) > sub-buffer size (%Zd)]\n",
634                        chan->last_toobig, chan->subbuf_size);
635
636         kref_put(&chan->kref, relay_destroy_channel);
637 }
638 EXPORT_SYMBOL_GPL(relay_close);
639
640 /**
641  *      relay_flush - close the channel
642  *      @chan: the channel
643  *
644  *      Flushes all channel buffers i.e. forces buffer switch.
645  */
646 void relay_flush(struct rchan *chan)
647 {
648         unsigned int i;
649         struct rchan_buf *prev = NULL;
650
651         if (!chan)
652                 return;
653
654         for (i = 0; i < NR_CPUS; i++) {
655                 if (!chan->buf[i] || chan->buf[i] == prev)
656                         break;
657                 relay_switch_subbuf(chan->buf[i], 0);
658                 prev = chan->buf[i];
659         }
660 }
661 EXPORT_SYMBOL_GPL(relay_flush);
662
663 /**
664  *      relay_file_open - open file op for relay files
665  *      @inode: the inode
666  *      @filp: the file
667  *
668  *      Increments the channel buffer refcount.
669  */
670 static int relay_file_open(struct inode *inode, struct file *filp)
671 {
672         struct rchan_buf *buf = inode->u.generic_ip;
673         kref_get(&buf->kref);
674         filp->private_data = buf;
675
676         return 0;
677 }
678
679 /**
680  *      relay_file_mmap - mmap file op for relay files
681  *      @filp: the file
682  *      @vma: the vma describing what to map
683  *
684  *      Calls upon relay_mmap_buf to map the file into user space.
685  */
686 static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
687 {
688         struct rchan_buf *buf = filp->private_data;
689         return relay_mmap_buf(buf, vma);
690 }
691
692 /**
693  *      relay_file_poll - poll file op for relay files
694  *      @filp: the file
695  *      @wait: poll table
696  *
697  *      Poll implemention.
698  */
699 static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
700 {
701         unsigned int mask = 0;
702         struct rchan_buf *buf = filp->private_data;
703
704         if (buf->finalized)
705                 return POLLERR;
706
707         if (filp->f_mode & FMODE_READ) {
708                 poll_wait(filp, &buf->read_wait, wait);
709                 if (!relay_buf_empty(buf))
710                         mask |= POLLIN | POLLRDNORM;
711         }
712
713         return mask;
714 }
715
716 /**
717  *      relay_file_release - release file op for relay files
718  *      @inode: the inode
719  *      @filp: the file
720  *
721  *      Decrements the channel refcount, as the filesystem is
722  *      no longer using it.
723  */
724 static int relay_file_release(struct inode *inode, struct file *filp)
725 {
726         struct rchan_buf *buf = filp->private_data;
727         kref_put(&buf->kref, relay_remove_buf);
728
729         return 0;
730 }
731
732 /**
733  *      relay_file_read_consume - update the consumed count for the buffer
734  */
735 static void relay_file_read_consume(struct rchan_buf *buf,
736                                     size_t read_pos,
737                                     size_t bytes_consumed)
738 {
739         size_t subbuf_size = buf->chan->subbuf_size;
740         size_t n_subbufs = buf->chan->n_subbufs;
741         size_t read_subbuf;
742
743         if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
744                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
745                 buf->bytes_consumed = 0;
746         }
747
748         buf->bytes_consumed += bytes_consumed;
749         read_subbuf = read_pos / buf->chan->subbuf_size;
750         if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
751                 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
752                     (buf->offset == subbuf_size))
753                         return;
754                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
755                 buf->bytes_consumed = 0;
756         }
757 }
758
759 /**
760  *      relay_file_read_avail - boolean, are there unconsumed bytes available?
761  */
762 static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
763 {
764         size_t subbuf_size = buf->chan->subbuf_size;
765         size_t n_subbufs = buf->chan->n_subbufs;
766         size_t produced = buf->subbufs_produced;
767         size_t consumed = buf->subbufs_consumed;
768
769         relay_file_read_consume(buf, read_pos, 0);
770
771         if (unlikely(buf->offset > subbuf_size)) {
772                 if (produced == consumed)
773                         return 0;
774                 return 1;
775         }
776
777         if (unlikely(produced - consumed >= n_subbufs)) {
778                 consumed = (produced / n_subbufs) * n_subbufs;
779                 buf->subbufs_consumed = consumed;
780         }
781         
782         produced = (produced % n_subbufs) * subbuf_size + buf->offset;
783         consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
784
785         if (consumed > produced)
786                 produced += n_subbufs * subbuf_size;
787         
788         if (consumed == produced)
789                 return 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 static ssize_t relay_file_sendsubbuf(struct file *filp, loff_t *ppos,
912                                      size_t count, read_actor_t actor,
913                                      void *target)
914 {
915         struct rchan_buf *buf = filp->private_data;
916         read_descriptor_t desc;
917         size_t read_start, avail;
918         unsigned long pidx, poff;
919         unsigned int subbuf_pages;
920         ssize_t ret = 0;
921
922         if (!relay_file_read_avail(buf, *ppos))
923                 return 0;
924
925         read_start = relay_file_read_start_pos(*ppos, buf);
926         avail = relay_file_read_subbuf_avail(read_start, buf);
927         if (!avail)
928                 return 0;
929
930         count = min(count, avail);
931
932         desc.written = 0;
933         desc.count = count;
934         desc.arg.data = target;
935         desc.error = 0;
936
937         subbuf_pages = buf->chan->alloc_size >> PAGE_SHIFT;
938         pidx = (read_start / PAGE_SIZE) % subbuf_pages;
939         poff = read_start & ~PAGE_MASK;
940         while (count) {
941                 struct page *p = buf->page_array[pidx];
942                 unsigned int len;
943
944                 len = PAGE_SIZE - poff;
945                 if (len > count)
946                         len = count;
947
948                 len = actor(&desc, p, poff, len);
949
950                 if (desc.error) {
951                         if (!ret)
952                                 ret = desc.error;
953                         break;
954                 }
955
956                 count -= len;
957                 ret += len;
958                 poff = 0;
959                 pidx = (pidx + 1) % subbuf_pages;
960         }
961
962         if (ret > 0) {
963                 relay_file_read_consume(buf, read_start, ret);
964                 *ppos = relay_file_read_end_pos(buf, read_start, ret);
965         }
966
967         return ret;
968 }
969
970 static ssize_t relay_file_sendfile(struct file *filp, loff_t *ppos,
971                                    size_t count, read_actor_t actor,
972                                    void *target)
973 {
974         ssize_t sent = 0, ret = 0;
975
976         if (!count)
977                 return 0;
978
979         mutex_lock(&filp->f_dentry->d_inode->i_mutex);
980
981         do {
982                 ret = relay_file_sendsubbuf(filp, ppos, count, actor, target);
983                 if (ret < 0) {
984                         if (!sent)
985                                 sent = ret;
986                         break;
987                 }
988                 count -= ret;
989                 sent += ret;
990         } while (count && ret);
991
992         mutex_unlock(&filp->f_dentry->d_inode->i_mutex);
993         return sent;
994 }
995
996 struct file_operations relay_file_operations = {
997         .open           = relay_file_open,
998         .poll           = relay_file_poll,
999         .mmap           = relay_file_mmap,
1000         .read           = relay_file_read,
1001         .llseek         = no_llseek,
1002         .release        = relay_file_release,
1003         .sendfile       = relay_file_sendfile,
1004 };
1005 EXPORT_SYMBOL_GPL(relay_file_operations);