Merge branch 'master' of master.kernel.org:/pub/scm/linux/kernel/git/davem/net-2.6
[linux-2.6] / 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  * November 2006 - CPU hotplug support by Mathieu Desnoyers
11  *      (mathieu.desnoyers@polymtl.ca)
12  *
13  * This file is released under the GPL.
14  */
15 #include <linux/errno.h>
16 #include <linux/stddef.h>
17 #include <linux/slab.h>
18 #include <linux/module.h>
19 #include <linux/string.h>
20 #include <linux/relay.h>
21 #include <linux/vmalloc.h>
22 #include <linux/mm.h>
23 #include <linux/cpu.h>
24 #include <linux/splice.h>
25
26 /* list of open channels, for cpu hotplug */
27 static DEFINE_MUTEX(relay_channels_mutex);
28 static LIST_HEAD(relay_channels);
29
30 /*
31  * close() vm_op implementation for relay file mapping.
32  */
33 static void relay_file_mmap_close(struct vm_area_struct *vma)
34 {
35         struct rchan_buf *buf = vma->vm_private_data;
36         buf->chan->cb->buf_unmapped(buf, vma->vm_file);
37 }
38
39 /*
40  * nopage() vm_op implementation for relay file mapping.
41  */
42 static struct page *relay_buf_nopage(struct vm_area_struct *vma,
43                                      unsigned long address,
44                                      int *type)
45 {
46         struct page *page;
47         struct rchan_buf *buf = vma->vm_private_data;
48         unsigned long offset = address - vma->vm_start;
49
50         if (address > vma->vm_end)
51                 return NOPAGE_SIGBUS; /* Disallow mremap */
52         if (!buf)
53                 return NOPAGE_OOM;
54
55         page = vmalloc_to_page(buf->start + offset);
56         if (!page)
57                 return NOPAGE_OOM;
58         get_page(page);
59
60         if (type)
61                 *type = VM_FAULT_MINOR;
62
63         return page;
64 }
65
66 /*
67  * vm_ops for relay file mappings.
68  */
69 static struct vm_operations_struct relay_file_mmap_ops = {
70         .nopage = relay_buf_nopage,
71         .close = relay_file_mmap_close,
72 };
73
74 /**
75  *      relay_mmap_buf: - mmap channel buffer to process address space
76  *      @buf: relay channel buffer
77  *      @vma: vm_area_struct describing memory to be mapped
78  *
79  *      Returns 0 if ok, negative on error
80  *
81  *      Caller should already have grabbed mmap_sem.
82  */
83 static int relay_mmap_buf(struct rchan_buf *buf, struct vm_area_struct *vma)
84 {
85         unsigned long length = vma->vm_end - vma->vm_start;
86         struct file *filp = vma->vm_file;
87
88         if (!buf)
89                 return -EBADF;
90
91         if (length != (unsigned long)buf->chan->alloc_size)
92                 return -EINVAL;
93
94         vma->vm_ops = &relay_file_mmap_ops;
95         vma->vm_private_data = buf;
96         buf->chan->cb->buf_mapped(buf, filp);
97
98         return 0;
99 }
100
101 /**
102  *      relay_alloc_buf - allocate a channel buffer
103  *      @buf: the buffer struct
104  *      @size: total size of the buffer
105  *
106  *      Returns a pointer to the resulting buffer, %NULL if unsuccessful. The
107  *      passed in size will get page aligned, if it isn't already.
108  */
109 static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size)
110 {
111         void *mem;
112         unsigned int i, j, n_pages;
113
114         *size = PAGE_ALIGN(*size);
115         n_pages = *size >> PAGE_SHIFT;
116
117         buf->page_array = kcalloc(n_pages, sizeof(struct page *), GFP_KERNEL);
118         if (!buf->page_array)
119                 return NULL;
120
121         for (i = 0; i < n_pages; i++) {
122                 buf->page_array[i] = alloc_page(GFP_KERNEL);
123                 if (unlikely(!buf->page_array[i]))
124                         goto depopulate;
125                 set_page_private(buf->page_array[i], (unsigned long)buf);
126         }
127         mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL);
128         if (!mem)
129                 goto depopulate;
130
131         memset(mem, 0, *size);
132         buf->page_count = n_pages;
133         return mem;
134
135 depopulate:
136         for (j = 0; j < i; j++)
137                 __free_page(buf->page_array[j]);
138         kfree(buf->page_array);
139         return NULL;
140 }
141
142 /**
143  *      relay_create_buf - allocate and initialize a channel buffer
144  *      @chan: the relay channel
145  *
146  *      Returns channel buffer if successful, %NULL otherwise.
147  */
148 static struct rchan_buf *relay_create_buf(struct rchan *chan)
149 {
150         struct rchan_buf *buf = kzalloc(sizeof(struct rchan_buf), GFP_KERNEL);
151         if (!buf)
152                 return NULL;
153
154         buf->padding = kmalloc(chan->n_subbufs * sizeof(size_t *), GFP_KERNEL);
155         if (!buf->padding)
156                 goto free_buf;
157
158         buf->start = relay_alloc_buf(buf, &chan->alloc_size);
159         if (!buf->start)
160                 goto free_buf;
161
162         buf->chan = chan;
163         kref_get(&buf->chan->kref);
164         return buf;
165
166 free_buf:
167         kfree(buf->padding);
168         kfree(buf);
169         return NULL;
170 }
171
172 /**
173  *      relay_destroy_channel - free the channel struct
174  *      @kref: target kernel reference that contains the relay channel
175  *
176  *      Should only be called from kref_put().
177  */
178 static void relay_destroy_channel(struct kref *kref)
179 {
180         struct rchan *chan = container_of(kref, struct rchan, kref);
181         kfree(chan);
182 }
183
184 /**
185  *      relay_destroy_buf - destroy an rchan_buf struct and associated buffer
186  *      @buf: the buffer struct
187  */
188 static void relay_destroy_buf(struct rchan_buf *buf)
189 {
190         struct rchan *chan = buf->chan;
191         unsigned int i;
192
193         if (likely(buf->start)) {
194                 vunmap(buf->start);
195                 for (i = 0; i < buf->page_count; i++)
196                         __free_page(buf->page_array[i]);
197                 kfree(buf->page_array);
198         }
199         chan->buf[buf->cpu] = NULL;
200         kfree(buf->padding);
201         kfree(buf);
202         kref_put(&chan->kref, relay_destroy_channel);
203 }
204
205 /**
206  *      relay_remove_buf - remove a channel buffer
207  *      @kref: target kernel reference that contains the relay buffer
208  *
209  *      Removes the file from the fileystem, which also frees the
210  *      rchan_buf_struct and the channel buffer.  Should only be called from
211  *      kref_put().
212  */
213 static void relay_remove_buf(struct kref *kref)
214 {
215         struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
216         buf->chan->cb->remove_buf_file(buf->dentry);
217         relay_destroy_buf(buf);
218 }
219
220 /**
221  *      relay_buf_empty - boolean, is the channel buffer empty?
222  *      @buf: channel buffer
223  *
224  *      Returns 1 if the buffer is empty, 0 otherwise.
225  */
226 static int relay_buf_empty(struct rchan_buf *buf)
227 {
228         return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1;
229 }
230
231 /**
232  *      relay_buf_full - boolean, is the channel buffer full?
233  *      @buf: channel buffer
234  *
235  *      Returns 1 if the buffer is full, 0 otherwise.
236  */
237 int relay_buf_full(struct rchan_buf *buf)
238 {
239         size_t ready = buf->subbufs_produced - buf->subbufs_consumed;
240         return (ready >= buf->chan->n_subbufs) ? 1 : 0;
241 }
242 EXPORT_SYMBOL_GPL(relay_buf_full);
243
244 /*
245  * High-level relay kernel API and associated functions.
246  */
247
248 /*
249  * rchan_callback implementations defining default channel behavior.  Used
250  * in place of corresponding NULL values in client callback struct.
251  */
252
253 /*
254  * subbuf_start() default callback.  Does nothing.
255  */
256 static int subbuf_start_default_callback (struct rchan_buf *buf,
257                                           void *subbuf,
258                                           void *prev_subbuf,
259                                           size_t prev_padding)
260 {
261         if (relay_buf_full(buf))
262                 return 0;
263
264         return 1;
265 }
266
267 /*
268  * buf_mapped() default callback.  Does nothing.
269  */
270 static void buf_mapped_default_callback(struct rchan_buf *buf,
271                                         struct file *filp)
272 {
273 }
274
275 /*
276  * buf_unmapped() default callback.  Does nothing.
277  */
278 static void buf_unmapped_default_callback(struct rchan_buf *buf,
279                                           struct file *filp)
280 {
281 }
282
283 /*
284  * create_buf_file_create() default callback.  Does nothing.
285  */
286 static struct dentry *create_buf_file_default_callback(const char *filename,
287                                                        struct dentry *parent,
288                                                        int mode,
289                                                        struct rchan_buf *buf,
290                                                        int *is_global)
291 {
292         return NULL;
293 }
294
295 /*
296  * remove_buf_file() default callback.  Does nothing.
297  */
298 static int remove_buf_file_default_callback(struct dentry *dentry)
299 {
300         return -EINVAL;
301 }
302
303 /* relay channel default callbacks */
304 static struct rchan_callbacks default_channel_callbacks = {
305         .subbuf_start = subbuf_start_default_callback,
306         .buf_mapped = buf_mapped_default_callback,
307         .buf_unmapped = buf_unmapped_default_callback,
308         .create_buf_file = create_buf_file_default_callback,
309         .remove_buf_file = remove_buf_file_default_callback,
310 };
311
312 /**
313  *      wakeup_readers - wake up readers waiting on a channel
314  *      @data: contains the channel buffer
315  *
316  *      This is the timer function used to defer reader waking.
317  */
318 static void wakeup_readers(unsigned long data)
319 {
320         struct rchan_buf *buf = (struct rchan_buf *)data;
321         wake_up_interruptible(&buf->read_wait);
322 }
323
324 /**
325  *      __relay_reset - reset a channel buffer
326  *      @buf: the channel buffer
327  *      @init: 1 if this is a first-time initialization
328  *
329  *      See relay_reset() for description of effect.
330  */
331 static void __relay_reset(struct rchan_buf *buf, unsigned int init)
332 {
333         size_t i;
334
335         if (init) {
336                 init_waitqueue_head(&buf->read_wait);
337                 kref_init(&buf->kref);
338                 setup_timer(&buf->timer, wakeup_readers, (unsigned long)buf);
339         } else
340                 del_timer_sync(&buf->timer);
341
342         buf->subbufs_produced = 0;
343         buf->subbufs_consumed = 0;
344         buf->bytes_consumed = 0;
345         buf->finalized = 0;
346         buf->data = buf->start;
347         buf->offset = 0;
348
349         for (i = 0; i < buf->chan->n_subbufs; i++)
350                 buf->padding[i] = 0;
351
352         buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0);
353 }
354
355 /**
356  *      relay_reset - reset the channel
357  *      @chan: the channel
358  *
359  *      This has the effect of erasing all data from all channel buffers
360  *      and restarting the channel in its initial state.  The buffers
361  *      are not freed, so any mappings are still in effect.
362  *
363  *      NOTE. Care should be taken that the channel isn't actually
364  *      being used by anything when this call is made.
365  */
366 void relay_reset(struct rchan *chan)
367 {
368         unsigned int i;
369
370         if (!chan)
371                 return;
372
373         if (chan->is_global && chan->buf[0]) {
374                 __relay_reset(chan->buf[0], 0);
375                 return;
376         }
377
378         mutex_lock(&relay_channels_mutex);
379         for_each_online_cpu(i)
380                 if (chan->buf[i])
381                         __relay_reset(chan->buf[i], 0);
382         mutex_unlock(&relay_channels_mutex);
383 }
384 EXPORT_SYMBOL_GPL(relay_reset);
385
386 /*
387  *      relay_open_buf - create a new relay channel buffer
388  *
389  *      used by relay_open() and CPU hotplug.
390  */
391 static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu)
392 {
393         struct rchan_buf *buf = NULL;
394         struct dentry *dentry;
395         char *tmpname;
396
397         if (chan->is_global)
398                 return chan->buf[0];
399
400         tmpname = kzalloc(NAME_MAX + 1, GFP_KERNEL);
401         if (!tmpname)
402                 goto end;
403         snprintf(tmpname, NAME_MAX, "%s%d", chan->base_filename, cpu);
404
405         buf = relay_create_buf(chan);
406         if (!buf)
407                 goto free_name;
408
409         buf->cpu = cpu;
410         __relay_reset(buf, 1);
411
412         /* Create file in fs */
413         dentry = chan->cb->create_buf_file(tmpname, chan->parent, S_IRUSR,
414                                            buf, &chan->is_global);
415         if (!dentry)
416                 goto free_buf;
417
418         buf->dentry = dentry;
419
420         if(chan->is_global) {
421                 chan->buf[0] = buf;
422                 buf->cpu = 0;
423         }
424
425         goto free_name;
426
427 free_buf:
428         relay_destroy_buf(buf);
429 free_name:
430         kfree(tmpname);
431 end:
432         return buf;
433 }
434
435 /**
436  *      relay_close_buf - close a channel buffer
437  *      @buf: channel buffer
438  *
439  *      Marks the buffer finalized and restores the default callbacks.
440  *      The channel buffer and channel buffer data structure are then freed
441  *      automatically when the last reference is given up.
442  */
443 static void relay_close_buf(struct rchan_buf *buf)
444 {
445         buf->finalized = 1;
446         del_timer_sync(&buf->timer);
447         kref_put(&buf->kref, relay_remove_buf);
448 }
449
450 static void setup_callbacks(struct rchan *chan,
451                                    struct rchan_callbacks *cb)
452 {
453         if (!cb) {
454                 chan->cb = &default_channel_callbacks;
455                 return;
456         }
457
458         if (!cb->subbuf_start)
459                 cb->subbuf_start = subbuf_start_default_callback;
460         if (!cb->buf_mapped)
461                 cb->buf_mapped = buf_mapped_default_callback;
462         if (!cb->buf_unmapped)
463                 cb->buf_unmapped = buf_unmapped_default_callback;
464         if (!cb->create_buf_file)
465                 cb->create_buf_file = create_buf_file_default_callback;
466         if (!cb->remove_buf_file)
467                 cb->remove_buf_file = remove_buf_file_default_callback;
468         chan->cb = cb;
469 }
470
471 /**
472  *      relay_hotcpu_callback - CPU hotplug callback
473  *      @nb: notifier block
474  *      @action: hotplug action to take
475  *      @hcpu: CPU number
476  *
477  *      Returns the success/failure of the operation. (%NOTIFY_OK, %NOTIFY_BAD)
478  */
479 static int __cpuinit relay_hotcpu_callback(struct notifier_block *nb,
480                                 unsigned long action,
481                                 void *hcpu)
482 {
483         unsigned int hotcpu = (unsigned long)hcpu;
484         struct rchan *chan;
485
486         switch(action) {
487         case CPU_UP_PREPARE:
488         case CPU_UP_PREPARE_FROZEN:
489                 mutex_lock(&relay_channels_mutex);
490                 list_for_each_entry(chan, &relay_channels, list) {
491                         if (chan->buf[hotcpu])
492                                 continue;
493                         chan->buf[hotcpu] = relay_open_buf(chan, hotcpu);
494                         if(!chan->buf[hotcpu]) {
495                                 printk(KERN_ERR
496                                         "relay_hotcpu_callback: cpu %d buffer "
497                                         "creation failed\n", hotcpu);
498                                 mutex_unlock(&relay_channels_mutex);
499                                 return NOTIFY_BAD;
500                         }
501                 }
502                 mutex_unlock(&relay_channels_mutex);
503                 break;
504         case CPU_DEAD:
505         case CPU_DEAD_FROZEN:
506                 /* No need to flush the cpu : will be flushed upon
507                  * final relay_flush() call. */
508                 break;
509         }
510         return NOTIFY_OK;
511 }
512
513 /**
514  *      relay_open - create a new relay channel
515  *      @base_filename: base name of files to create
516  *      @parent: dentry of parent directory, %NULL for root directory
517  *      @subbuf_size: size of sub-buffers
518  *      @n_subbufs: number of sub-buffers
519  *      @cb: client callback functions
520  *      @private_data: user-defined data
521  *
522  *      Returns channel pointer if successful, %NULL otherwise.
523  *
524  *      Creates a channel buffer for each cpu using the sizes and
525  *      attributes specified.  The created channel buffer files
526  *      will be named base_filename0...base_filenameN-1.  File
527  *      permissions will be %S_IRUSR.
528  */
529 struct rchan *relay_open(const char *base_filename,
530                          struct dentry *parent,
531                          size_t subbuf_size,
532                          size_t n_subbufs,
533                          struct rchan_callbacks *cb,
534                          void *private_data)
535 {
536         unsigned int i;
537         struct rchan *chan;
538         if (!base_filename)
539                 return NULL;
540
541         if (!(subbuf_size && n_subbufs))
542                 return NULL;
543
544         chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
545         if (!chan)
546                 return NULL;
547
548         chan->version = RELAYFS_CHANNEL_VERSION;
549         chan->n_subbufs = n_subbufs;
550         chan->subbuf_size = subbuf_size;
551         chan->alloc_size = FIX_SIZE(subbuf_size * n_subbufs);
552         chan->parent = parent;
553         chan->private_data = private_data;
554         strlcpy(chan->base_filename, base_filename, NAME_MAX);
555         setup_callbacks(chan, cb);
556         kref_init(&chan->kref);
557
558         mutex_lock(&relay_channels_mutex);
559         for_each_online_cpu(i) {
560                 chan->buf[i] = relay_open_buf(chan, i);
561                 if (!chan->buf[i])
562                         goto free_bufs;
563         }
564         list_add(&chan->list, &relay_channels);
565         mutex_unlock(&relay_channels_mutex);
566
567         return chan;
568
569 free_bufs:
570         for_each_online_cpu(i) {
571                 if (!chan->buf[i])
572                         break;
573                 relay_close_buf(chan->buf[i]);
574         }
575
576         kref_put(&chan->kref, relay_destroy_channel);
577         mutex_unlock(&relay_channels_mutex);
578         return NULL;
579 }
580 EXPORT_SYMBOL_GPL(relay_open);
581
582 /**
583  *      relay_switch_subbuf - switch to a new sub-buffer
584  *      @buf: channel buffer
585  *      @length: size of current event
586  *
587  *      Returns either the length passed in or 0 if full.
588  *
589  *      Performs sub-buffer-switch tasks such as invoking callbacks,
590  *      updating padding counts, waking up readers, etc.
591  */
592 size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
593 {
594         void *old, *new;
595         size_t old_subbuf, new_subbuf;
596
597         if (unlikely(length > buf->chan->subbuf_size))
598                 goto toobig;
599
600         if (buf->offset != buf->chan->subbuf_size + 1) {
601                 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
602                 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
603                 buf->padding[old_subbuf] = buf->prev_padding;
604                 buf->subbufs_produced++;
605                 buf->dentry->d_inode->i_size += buf->chan->subbuf_size -
606                         buf->padding[old_subbuf];
607                 smp_mb();
608                 if (waitqueue_active(&buf->read_wait))
609                         /*
610                          * Calling wake_up_interruptible() from here
611                          * will deadlock if we happen to be logging
612                          * from the scheduler (trying to re-grab
613                          * rq->lock), so defer it.
614                          */
615                         __mod_timer(&buf->timer, jiffies + 1);
616         }
617
618         old = buf->data;
619         new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
620         new = buf->start + new_subbuf * buf->chan->subbuf_size;
621         buf->offset = 0;
622         if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
623                 buf->offset = buf->chan->subbuf_size + 1;
624                 return 0;
625         }
626         buf->data = new;
627         buf->padding[new_subbuf] = 0;
628
629         if (unlikely(length + buf->offset > buf->chan->subbuf_size))
630                 goto toobig;
631
632         return length;
633
634 toobig:
635         buf->chan->last_toobig = length;
636         return 0;
637 }
638 EXPORT_SYMBOL_GPL(relay_switch_subbuf);
639
640 /**
641  *      relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
642  *      @chan: the channel
643  *      @cpu: the cpu associated with the channel buffer to update
644  *      @subbufs_consumed: number of sub-buffers to add to current buf's count
645  *
646  *      Adds to the channel buffer's consumed sub-buffer count.
647  *      subbufs_consumed should be the number of sub-buffers newly consumed,
648  *      not the total consumed.
649  *
650  *      NOTE. Kernel clients don't need to call this function if the channel
651  *      mode is 'overwrite'.
652  */
653 void relay_subbufs_consumed(struct rchan *chan,
654                             unsigned int cpu,
655                             size_t subbufs_consumed)
656 {
657         struct rchan_buf *buf;
658
659         if (!chan)
660                 return;
661
662         if (cpu >= NR_CPUS || !chan->buf[cpu])
663                 return;
664
665         buf = chan->buf[cpu];
666         buf->subbufs_consumed += subbufs_consumed;
667         if (buf->subbufs_consumed > buf->subbufs_produced)
668                 buf->subbufs_consumed = buf->subbufs_produced;
669 }
670 EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
671
672 /**
673  *      relay_close - close the channel
674  *      @chan: the channel
675  *
676  *      Closes all channel buffers and frees the channel.
677  */
678 void relay_close(struct rchan *chan)
679 {
680         unsigned int i;
681
682         if (!chan)
683                 return;
684
685         mutex_lock(&relay_channels_mutex);
686         if (chan->is_global && chan->buf[0])
687                 relay_close_buf(chan->buf[0]);
688         else
689                 for_each_possible_cpu(i)
690                         if (chan->buf[i])
691                                 relay_close_buf(chan->buf[i]);
692
693         if (chan->last_toobig)
694                 printk(KERN_WARNING "relay: one or more items not logged "
695                        "[item size (%Zd) > sub-buffer size (%Zd)]\n",
696                        chan->last_toobig, chan->subbuf_size);
697
698         list_del(&chan->list);
699         kref_put(&chan->kref, relay_destroy_channel);
700         mutex_unlock(&relay_channels_mutex);
701 }
702 EXPORT_SYMBOL_GPL(relay_close);
703
704 /**
705  *      relay_flush - close the channel
706  *      @chan: the channel
707  *
708  *      Flushes all channel buffers, i.e. forces buffer switch.
709  */
710 void relay_flush(struct rchan *chan)
711 {
712         unsigned int i;
713
714         if (!chan)
715                 return;
716
717         if (chan->is_global && chan->buf[0]) {
718                 relay_switch_subbuf(chan->buf[0], 0);
719                 return;
720         }
721
722         mutex_lock(&relay_channels_mutex);
723         for_each_possible_cpu(i)
724                 if (chan->buf[i])
725                         relay_switch_subbuf(chan->buf[i], 0);
726         mutex_unlock(&relay_channels_mutex);
727 }
728 EXPORT_SYMBOL_GPL(relay_flush);
729
730 /**
731  *      relay_file_open - open file op for relay files
732  *      @inode: the inode
733  *      @filp: the file
734  *
735  *      Increments the channel buffer refcount.
736  */
737 static int relay_file_open(struct inode *inode, struct file *filp)
738 {
739         struct rchan_buf *buf = inode->i_private;
740         kref_get(&buf->kref);
741         filp->private_data = buf;
742
743         return 0;
744 }
745
746 /**
747  *      relay_file_mmap - mmap file op for relay files
748  *      @filp: the file
749  *      @vma: the vma describing what to map
750  *
751  *      Calls upon relay_mmap_buf() to map the file into user space.
752  */
753 static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
754 {
755         struct rchan_buf *buf = filp->private_data;
756         return relay_mmap_buf(buf, vma);
757 }
758
759 /**
760  *      relay_file_poll - poll file op for relay files
761  *      @filp: the file
762  *      @wait: poll table
763  *
764  *      Poll implemention.
765  */
766 static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
767 {
768         unsigned int mask = 0;
769         struct rchan_buf *buf = filp->private_data;
770
771         if (buf->finalized)
772                 return POLLERR;
773
774         if (filp->f_mode & FMODE_READ) {
775                 poll_wait(filp, &buf->read_wait, wait);
776                 if (!relay_buf_empty(buf))
777                         mask |= POLLIN | POLLRDNORM;
778         }
779
780         return mask;
781 }
782
783 /**
784  *      relay_file_release - release file op for relay files
785  *      @inode: the inode
786  *      @filp: the file
787  *
788  *      Decrements the channel refcount, as the filesystem is
789  *      no longer using it.
790  */
791 static int relay_file_release(struct inode *inode, struct file *filp)
792 {
793         struct rchan_buf *buf = filp->private_data;
794         kref_put(&buf->kref, relay_remove_buf);
795
796         return 0;
797 }
798
799 /*
800  *      relay_file_read_consume - update the consumed count for the buffer
801  */
802 static void relay_file_read_consume(struct rchan_buf *buf,
803                                     size_t read_pos,
804                                     size_t bytes_consumed)
805 {
806         size_t subbuf_size = buf->chan->subbuf_size;
807         size_t n_subbufs = buf->chan->n_subbufs;
808         size_t read_subbuf;
809
810         if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
811                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
812                 buf->bytes_consumed = 0;
813         }
814
815         buf->bytes_consumed += bytes_consumed;
816         if (!read_pos)
817                 read_subbuf = buf->subbufs_consumed % n_subbufs;
818         else
819                 read_subbuf = read_pos / buf->chan->subbuf_size;
820         if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
821                 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
822                     (buf->offset == subbuf_size))
823                         return;
824                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
825                 buf->bytes_consumed = 0;
826         }
827 }
828
829 /*
830  *      relay_file_read_avail - boolean, are there unconsumed bytes available?
831  */
832 static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
833 {
834         size_t subbuf_size = buf->chan->subbuf_size;
835         size_t n_subbufs = buf->chan->n_subbufs;
836         size_t produced = buf->subbufs_produced;
837         size_t consumed = buf->subbufs_consumed;
838
839         relay_file_read_consume(buf, read_pos, 0);
840
841         if (unlikely(buf->offset > subbuf_size)) {
842                 if (produced == consumed)
843                         return 0;
844                 return 1;
845         }
846
847         if (unlikely(produced - consumed >= n_subbufs)) {
848                 consumed = produced - n_subbufs + 1;
849                 buf->subbufs_consumed = consumed;
850                 buf->bytes_consumed = 0;
851         }
852         
853         produced = (produced % n_subbufs) * subbuf_size + buf->offset;
854         consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
855
856         if (consumed > produced)
857                 produced += n_subbufs * subbuf_size;
858         
859         if (consumed == produced)
860                 return 0;
861
862         return 1;
863 }
864
865 /**
866  *      relay_file_read_subbuf_avail - return bytes available in sub-buffer
867  *      @read_pos: file read position
868  *      @buf: relay channel buffer
869  */
870 static size_t relay_file_read_subbuf_avail(size_t read_pos,
871                                            struct rchan_buf *buf)
872 {
873         size_t padding, avail = 0;
874         size_t read_subbuf, read_offset, write_subbuf, write_offset;
875         size_t subbuf_size = buf->chan->subbuf_size;
876
877         write_subbuf = (buf->data - buf->start) / subbuf_size;
878         write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
879         read_subbuf = read_pos / subbuf_size;
880         read_offset = read_pos % subbuf_size;
881         padding = buf->padding[read_subbuf];
882
883         if (read_subbuf == write_subbuf) {
884                 if (read_offset + padding < write_offset)
885                         avail = write_offset - (read_offset + padding);
886         } else
887                 avail = (subbuf_size - padding) - read_offset;
888
889         return avail;
890 }
891
892 /**
893  *      relay_file_read_start_pos - find the first available byte to read
894  *      @read_pos: file read position
895  *      @buf: relay channel buffer
896  *
897  *      If the @read_pos is in the middle of padding, return the
898  *      position of the first actually available byte, otherwise
899  *      return the original value.
900  */
901 static size_t relay_file_read_start_pos(size_t read_pos,
902                                         struct rchan_buf *buf)
903 {
904         size_t read_subbuf, padding, padding_start, padding_end;
905         size_t subbuf_size = buf->chan->subbuf_size;
906         size_t n_subbufs = buf->chan->n_subbufs;
907         size_t consumed = buf->subbufs_consumed % n_subbufs;
908
909         if (!read_pos)
910                 read_pos = consumed * subbuf_size + buf->bytes_consumed;
911         read_subbuf = read_pos / subbuf_size;
912         padding = buf->padding[read_subbuf];
913         padding_start = (read_subbuf + 1) * subbuf_size - padding;
914         padding_end = (read_subbuf + 1) * subbuf_size;
915         if (read_pos >= padding_start && read_pos < padding_end) {
916                 read_subbuf = (read_subbuf + 1) % n_subbufs;
917                 read_pos = read_subbuf * subbuf_size;
918         }
919
920         return read_pos;
921 }
922
923 /**
924  *      relay_file_read_end_pos - return the new read position
925  *      @read_pos: file read position
926  *      @buf: relay channel buffer
927  *      @count: number of bytes to be read
928  */
929 static size_t relay_file_read_end_pos(struct rchan_buf *buf,
930                                       size_t read_pos,
931                                       size_t count)
932 {
933         size_t read_subbuf, padding, end_pos;
934         size_t subbuf_size = buf->chan->subbuf_size;
935         size_t n_subbufs = buf->chan->n_subbufs;
936
937         read_subbuf = read_pos / subbuf_size;
938         padding = buf->padding[read_subbuf];
939         if (read_pos % subbuf_size + count + padding == subbuf_size)
940                 end_pos = (read_subbuf + 1) * subbuf_size;
941         else
942                 end_pos = read_pos + count;
943         if (end_pos >= subbuf_size * n_subbufs)
944                 end_pos = 0;
945
946         return end_pos;
947 }
948
949 /*
950  *      subbuf_read_actor - read up to one subbuf's worth of data
951  */
952 static int subbuf_read_actor(size_t read_start,
953                              struct rchan_buf *buf,
954                              size_t avail,
955                              read_descriptor_t *desc,
956                              read_actor_t actor)
957 {
958         void *from;
959         int ret = 0;
960
961         from = buf->start + read_start;
962         ret = avail;
963         if (copy_to_user(desc->arg.buf, from, avail)) {
964                 desc->error = -EFAULT;
965                 ret = 0;
966         }
967         desc->arg.data += ret;
968         desc->written += ret;
969         desc->count -= ret;
970
971         return ret;
972 }
973
974 typedef int (*subbuf_actor_t) (size_t read_start,
975                                struct rchan_buf *buf,
976                                size_t avail,
977                                read_descriptor_t *desc,
978                                read_actor_t actor);
979
980 /*
981  *      relay_file_read_subbufs - read count bytes, bridging subbuf boundaries
982  */
983 static ssize_t relay_file_read_subbufs(struct file *filp, loff_t *ppos,
984                                         subbuf_actor_t subbuf_actor,
985                                         read_actor_t actor,
986                                         read_descriptor_t *desc)
987 {
988         struct rchan_buf *buf = filp->private_data;
989         size_t read_start, avail;
990         int ret;
991
992         if (!desc->count)
993                 return 0;
994
995         mutex_lock(&filp->f_path.dentry->d_inode->i_mutex);
996         do {
997                 if (!relay_file_read_avail(buf, *ppos))
998                         break;
999
1000                 read_start = relay_file_read_start_pos(*ppos, buf);
1001                 avail = relay_file_read_subbuf_avail(read_start, buf);
1002                 if (!avail)
1003                         break;
1004
1005                 avail = min(desc->count, avail);
1006                 ret = subbuf_actor(read_start, buf, avail, desc, actor);
1007                 if (desc->error < 0)
1008                         break;
1009
1010                 if (ret) {
1011                         relay_file_read_consume(buf, read_start, ret);
1012                         *ppos = relay_file_read_end_pos(buf, read_start, ret);
1013                 }
1014         } while (desc->count && ret);
1015         mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex);
1016
1017         return desc->written;
1018 }
1019
1020 static ssize_t relay_file_read(struct file *filp,
1021                                char __user *buffer,
1022                                size_t count,
1023                                loff_t *ppos)
1024 {
1025         read_descriptor_t desc;
1026         desc.written = 0;
1027         desc.count = count;
1028         desc.arg.buf = buffer;
1029         desc.error = 0;
1030         return relay_file_read_subbufs(filp, ppos, subbuf_read_actor,
1031                                        NULL, &desc);
1032 }
1033
1034 static void relay_consume_bytes(struct rchan_buf *rbuf, int bytes_consumed)
1035 {
1036         rbuf->bytes_consumed += bytes_consumed;
1037
1038         if (rbuf->bytes_consumed >= rbuf->chan->subbuf_size) {
1039                 relay_subbufs_consumed(rbuf->chan, rbuf->cpu, 1);
1040                 rbuf->bytes_consumed %= rbuf->chan->subbuf_size;
1041         }
1042 }
1043
1044 static void relay_pipe_buf_release(struct pipe_inode_info *pipe,
1045                                    struct pipe_buffer *buf)
1046 {
1047         struct rchan_buf *rbuf;
1048
1049         rbuf = (struct rchan_buf *)page_private(buf->page);
1050         relay_consume_bytes(rbuf, buf->private);
1051 }
1052
1053 static struct pipe_buf_operations relay_pipe_buf_ops = {
1054         .can_merge = 0,
1055         .map = generic_pipe_buf_map,
1056         .unmap = generic_pipe_buf_unmap,
1057         .confirm = generic_pipe_buf_confirm,
1058         .release = relay_pipe_buf_release,
1059         .steal = generic_pipe_buf_steal,
1060         .get = generic_pipe_buf_get,
1061 };
1062
1063 /*
1064  *      subbuf_splice_actor - splice up to one subbuf's worth of data
1065  */
1066 static int subbuf_splice_actor(struct file *in,
1067                                loff_t *ppos,
1068                                struct pipe_inode_info *pipe,
1069                                size_t len,
1070                                unsigned int flags,
1071                                int *nonpad_ret)
1072 {
1073         unsigned int pidx, poff, total_len, subbuf_pages, ret;
1074         struct rchan_buf *rbuf = in->private_data;
1075         unsigned int subbuf_size = rbuf->chan->subbuf_size;
1076         uint64_t pos = (uint64_t) *ppos;
1077         uint32_t alloc_size = (uint32_t) rbuf->chan->alloc_size;
1078         size_t read_start = (size_t) do_div(pos, alloc_size);
1079         size_t read_subbuf = read_start / subbuf_size;
1080         size_t padding = rbuf->padding[read_subbuf];
1081         size_t nonpad_end = read_subbuf * subbuf_size + subbuf_size - padding;
1082         struct page *pages[PIPE_BUFFERS];
1083         struct partial_page partial[PIPE_BUFFERS];
1084         struct splice_pipe_desc spd = {
1085                 .pages = pages,
1086                 .nr_pages = 0,
1087                 .partial = partial,
1088                 .flags = flags,
1089                 .ops = &relay_pipe_buf_ops,
1090         };
1091
1092         if (rbuf->subbufs_produced == rbuf->subbufs_consumed)
1093                 return 0;
1094
1095         /*
1096          * Adjust read len, if longer than what is available
1097          */
1098         if (len > (subbuf_size - read_start % subbuf_size))
1099                 len = subbuf_size - read_start % subbuf_size;
1100
1101         subbuf_pages = rbuf->chan->alloc_size >> PAGE_SHIFT;
1102         pidx = (read_start / PAGE_SIZE) % subbuf_pages;
1103         poff = read_start & ~PAGE_MASK;
1104
1105         for (total_len = 0; spd.nr_pages < subbuf_pages; spd.nr_pages++) {
1106                 unsigned int this_len, this_end, private;
1107                 unsigned int cur_pos = read_start + total_len;
1108
1109                 if (!len)
1110                         break;
1111
1112                 this_len = min_t(unsigned long, len, PAGE_SIZE - poff);
1113                 private = this_len;
1114
1115                 spd.pages[spd.nr_pages] = rbuf->page_array[pidx];
1116                 spd.partial[spd.nr_pages].offset = poff;
1117
1118                 this_end = cur_pos + this_len;
1119                 if (this_end >= nonpad_end) {
1120                         this_len = nonpad_end - cur_pos;
1121                         private = this_len + padding;
1122                 }
1123                 spd.partial[spd.nr_pages].len = this_len;
1124                 spd.partial[spd.nr_pages].private = private;
1125
1126                 len -= this_len;
1127                 total_len += this_len;
1128                 poff = 0;
1129                 pidx = (pidx + 1) % subbuf_pages;
1130
1131                 if (this_end >= nonpad_end) {
1132                         spd.nr_pages++;
1133                         break;
1134                 }
1135         }
1136
1137         if (!spd.nr_pages)
1138                 return 0;
1139
1140         ret = *nonpad_ret = splice_to_pipe(pipe, &spd);
1141         if (ret < 0 || ret < total_len)
1142                 return ret;
1143
1144         if (read_start + ret == nonpad_end)
1145                 ret += padding;
1146
1147         return ret;
1148 }
1149
1150 static ssize_t relay_file_splice_read(struct file *in,
1151                                       loff_t *ppos,
1152                                       struct pipe_inode_info *pipe,
1153                                       size_t len,
1154                                       unsigned int flags)
1155 {
1156         ssize_t spliced;
1157         int ret;
1158         int nonpad_ret = 0;
1159
1160         ret = 0;
1161         spliced = 0;
1162
1163         while (len) {
1164                 ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret);
1165                 if (ret < 0)
1166                         break;
1167                 else if (!ret) {
1168                         if (spliced)
1169                                 break;
1170                         if (flags & SPLICE_F_NONBLOCK) {
1171                                 ret = -EAGAIN;
1172                                 break;
1173                         }
1174                 }
1175
1176                 *ppos += ret;
1177                 if (ret > len)
1178                         len = 0;
1179                 else
1180                         len -= ret;
1181                 spliced += nonpad_ret;
1182                 nonpad_ret = 0;
1183         }
1184
1185         if (spliced)
1186                 return spliced;
1187
1188         return ret;
1189 }
1190
1191 const struct file_operations relay_file_operations = {
1192         .open           = relay_file_open,
1193         .poll           = relay_file_poll,
1194         .mmap           = relay_file_mmap,
1195         .read           = relay_file_read,
1196         .llseek         = no_llseek,
1197         .release        = relay_file_release,
1198         .splice_read    = relay_file_splice_read,
1199 };
1200 EXPORT_SYMBOL_GPL(relay_file_operations);
1201
1202 static __init int relay_init(void)
1203 {
1204
1205         hotcpu_notifier(relay_hotcpu_callback, 0);
1206         return 0;
1207 }
1208
1209 module_init(relay_init);