Merge branch 'upstream-davem' of master.kernel.org:/pub/scm/linux/kernel/git/jgarzik...
[linux-2.6] / kernel / relay.c
1 /*
2  * Public API and common code for kernel->userspace relay file support.
3  *
4  * See Documentation/filesystems/relay.txt for an overview.
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_flags |= VM_DONTEXPAND;
96         vma->vm_private_data = buf;
97         buf->chan->cb->buf_mapped(buf, filp);
98
99         return 0;
100 }
101
102 /**
103  *      relay_alloc_buf - allocate a channel buffer
104  *      @buf: the buffer struct
105  *      @size: total size of the buffer
106  *
107  *      Returns a pointer to the resulting buffer, %NULL if unsuccessful. The
108  *      passed in size will get page aligned, if it isn't already.
109  */
110 static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size)
111 {
112         void *mem;
113         unsigned int i, j, n_pages;
114
115         *size = PAGE_ALIGN(*size);
116         n_pages = *size >> PAGE_SHIFT;
117
118         buf->page_array = kcalloc(n_pages, sizeof(struct page *), GFP_KERNEL);
119         if (!buf->page_array)
120                 return NULL;
121
122         for (i = 0; i < n_pages; i++) {
123                 buf->page_array[i] = alloc_page(GFP_KERNEL);
124                 if (unlikely(!buf->page_array[i]))
125                         goto depopulate;
126                 set_page_private(buf->page_array[i], (unsigned long)buf);
127         }
128         mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL);
129         if (!mem)
130                 goto depopulate;
131
132         memset(mem, 0, *size);
133         buf->page_count = n_pages;
134         return mem;
135
136 depopulate:
137         for (j = 0; j < i; j++)
138                 __free_page(buf->page_array[j]);
139         kfree(buf->page_array);
140         return NULL;
141 }
142
143 /**
144  *      relay_create_buf - allocate and initialize a channel buffer
145  *      @chan: the relay channel
146  *
147  *      Returns channel buffer if successful, %NULL otherwise.
148  */
149 static struct rchan_buf *relay_create_buf(struct rchan *chan)
150 {
151         struct rchan_buf *buf = kzalloc(sizeof(struct rchan_buf), GFP_KERNEL);
152         if (!buf)
153                 return NULL;
154
155         buf->padding = kmalloc(chan->n_subbufs * sizeof(size_t *), GFP_KERNEL);
156         if (!buf->padding)
157                 goto free_buf;
158
159         buf->start = relay_alloc_buf(buf, &chan->alloc_size);
160         if (!buf->start)
161                 goto free_buf;
162
163         buf->chan = chan;
164         kref_get(&buf->chan->kref);
165         return buf;
166
167 free_buf:
168         kfree(buf->padding);
169         kfree(buf);
170         return NULL;
171 }
172
173 /**
174  *      relay_destroy_channel - free the channel struct
175  *      @kref: target kernel reference that contains the relay channel
176  *
177  *      Should only be called from kref_put().
178  */
179 static void relay_destroy_channel(struct kref *kref)
180 {
181         struct rchan *chan = container_of(kref, struct rchan, kref);
182         kfree(chan);
183 }
184
185 /**
186  *      relay_destroy_buf - destroy an rchan_buf struct and associated buffer
187  *      @buf: the buffer struct
188  */
189 static void relay_destroy_buf(struct rchan_buf *buf)
190 {
191         struct rchan *chan = buf->chan;
192         unsigned int i;
193
194         if (likely(buf->start)) {
195                 vunmap(buf->start);
196                 for (i = 0; i < buf->page_count; i++)
197                         __free_page(buf->page_array[i]);
198                 kfree(buf->page_array);
199         }
200         chan->buf[buf->cpu] = NULL;
201         kfree(buf->padding);
202         kfree(buf);
203         kref_put(&chan->kref, relay_destroy_channel);
204 }
205
206 /**
207  *      relay_remove_buf - remove a channel buffer
208  *      @kref: target kernel reference that contains the relay buffer
209  *
210  *      Removes the file from the fileystem, which also frees the
211  *      rchan_buf_struct and the channel buffer.  Should only be called from
212  *      kref_put().
213  */
214 static void relay_remove_buf(struct kref *kref)
215 {
216         struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
217         buf->chan->cb->remove_buf_file(buf->dentry);
218         relay_destroy_buf(buf);
219 }
220
221 /**
222  *      relay_buf_empty - boolean, is the channel buffer empty?
223  *      @buf: channel buffer
224  *
225  *      Returns 1 if the buffer is empty, 0 otherwise.
226  */
227 static int relay_buf_empty(struct rchan_buf *buf)
228 {
229         return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1;
230 }
231
232 /**
233  *      relay_buf_full - boolean, is the channel buffer full?
234  *      @buf: channel buffer
235  *
236  *      Returns 1 if the buffer is full, 0 otherwise.
237  */
238 int relay_buf_full(struct rchan_buf *buf)
239 {
240         size_t ready = buf->subbufs_produced - buf->subbufs_consumed;
241         return (ready >= buf->chan->n_subbufs) ? 1 : 0;
242 }
243 EXPORT_SYMBOL_GPL(relay_buf_full);
244
245 /*
246  * High-level relay kernel API and associated functions.
247  */
248
249 /*
250  * rchan_callback implementations defining default channel behavior.  Used
251  * in place of corresponding NULL values in client callback struct.
252  */
253
254 /*
255  * subbuf_start() default callback.  Does nothing.
256  */
257 static int subbuf_start_default_callback (struct rchan_buf *buf,
258                                           void *subbuf,
259                                           void *prev_subbuf,
260                                           size_t prev_padding)
261 {
262         if (relay_buf_full(buf))
263                 return 0;
264
265         return 1;
266 }
267
268 /*
269  * buf_mapped() default callback.  Does nothing.
270  */
271 static void buf_mapped_default_callback(struct rchan_buf *buf,
272                                         struct file *filp)
273 {
274 }
275
276 /*
277  * buf_unmapped() default callback.  Does nothing.
278  */
279 static void buf_unmapped_default_callback(struct rchan_buf *buf,
280                                           struct file *filp)
281 {
282 }
283
284 /*
285  * create_buf_file_create() default callback.  Does nothing.
286  */
287 static struct dentry *create_buf_file_default_callback(const char *filename,
288                                                        struct dentry *parent,
289                                                        int mode,
290                                                        struct rchan_buf *buf,
291                                                        int *is_global)
292 {
293         return NULL;
294 }
295
296 /*
297  * remove_buf_file() default callback.  Does nothing.
298  */
299 static int remove_buf_file_default_callback(struct dentry *dentry)
300 {
301         return -EINVAL;
302 }
303
304 /* relay channel default callbacks */
305 static struct rchan_callbacks default_channel_callbacks = {
306         .subbuf_start = subbuf_start_default_callback,
307         .buf_mapped = buf_mapped_default_callback,
308         .buf_unmapped = buf_unmapped_default_callback,
309         .create_buf_file = create_buf_file_default_callback,
310         .remove_buf_file = remove_buf_file_default_callback,
311 };
312
313 /**
314  *      wakeup_readers - wake up readers waiting on a channel
315  *      @data: contains the channel buffer
316  *
317  *      This is the timer function used to defer reader waking.
318  */
319 static void wakeup_readers(unsigned long data)
320 {
321         struct rchan_buf *buf = (struct rchan_buf *)data;
322         wake_up_interruptible(&buf->read_wait);
323 }
324
325 /**
326  *      __relay_reset - reset a channel buffer
327  *      @buf: the channel buffer
328  *      @init: 1 if this is a first-time initialization
329  *
330  *      See relay_reset() for description of effect.
331  */
332 static void __relay_reset(struct rchan_buf *buf, unsigned int init)
333 {
334         size_t i;
335
336         if (init) {
337                 init_waitqueue_head(&buf->read_wait);
338                 kref_init(&buf->kref);
339                 setup_timer(&buf->timer, wakeup_readers, (unsigned long)buf);
340         } else
341                 del_timer_sync(&buf->timer);
342
343         buf->subbufs_produced = 0;
344         buf->subbufs_consumed = 0;
345         buf->bytes_consumed = 0;
346         buf->finalized = 0;
347         buf->data = buf->start;
348         buf->offset = 0;
349
350         for (i = 0; i < buf->chan->n_subbufs; i++)
351                 buf->padding[i] = 0;
352
353         buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0);
354 }
355
356 /**
357  *      relay_reset - reset the channel
358  *      @chan: the channel
359  *
360  *      This has the effect of erasing all data from all channel buffers
361  *      and restarting the channel in its initial state.  The buffers
362  *      are not freed, so any mappings are still in effect.
363  *
364  *      NOTE. Care should be taken that the channel isn't actually
365  *      being used by anything when this call is made.
366  */
367 void relay_reset(struct rchan *chan)
368 {
369         unsigned int i;
370
371         if (!chan)
372                 return;
373
374         if (chan->is_global && chan->buf[0]) {
375                 __relay_reset(chan->buf[0], 0);
376                 return;
377         }
378
379         mutex_lock(&relay_channels_mutex);
380         for_each_online_cpu(i)
381                 if (chan->buf[i])
382                         __relay_reset(chan->buf[i], 0);
383         mutex_unlock(&relay_channels_mutex);
384 }
385 EXPORT_SYMBOL_GPL(relay_reset);
386
387 /*
388  *      relay_open_buf - create a new relay channel buffer
389  *
390  *      used by relay_open() and CPU hotplug.
391  */
392 static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu)
393 {
394         struct rchan_buf *buf = NULL;
395         struct dentry *dentry;
396         char *tmpname;
397
398         if (chan->is_global)
399                 return chan->buf[0];
400
401         tmpname = kzalloc(NAME_MAX + 1, GFP_KERNEL);
402         if (!tmpname)
403                 goto end;
404         snprintf(tmpname, NAME_MAX, "%s%d", chan->base_filename, cpu);
405
406         buf = relay_create_buf(chan);
407         if (!buf)
408                 goto free_name;
409
410         buf->cpu = cpu;
411         __relay_reset(buf, 1);
412
413         /* Create file in fs */
414         dentry = chan->cb->create_buf_file(tmpname, chan->parent, S_IRUSR,
415                                            buf, &chan->is_global);
416         if (!dentry)
417                 goto free_buf;
418
419         buf->dentry = dentry;
420
421         if(chan->is_global) {
422                 chan->buf[0] = buf;
423                 buf->cpu = 0;
424         }
425
426         goto free_name;
427
428 free_buf:
429         relay_destroy_buf(buf);
430         buf = NULL;
431 free_name:
432         kfree(tmpname);
433 end:
434         return buf;
435 }
436
437 /**
438  *      relay_close_buf - close a channel buffer
439  *      @buf: channel buffer
440  *
441  *      Marks the buffer finalized and restores the default callbacks.
442  *      The channel buffer and channel buffer data structure are then freed
443  *      automatically when the last reference is given up.
444  */
445 static void relay_close_buf(struct rchan_buf *buf)
446 {
447         buf->finalized = 1;
448         del_timer_sync(&buf->timer);
449         kref_put(&buf->kref, relay_remove_buf);
450 }
451
452 static void setup_callbacks(struct rchan *chan,
453                                    struct rchan_callbacks *cb)
454 {
455         if (!cb) {
456                 chan->cb = &default_channel_callbacks;
457                 return;
458         }
459
460         if (!cb->subbuf_start)
461                 cb->subbuf_start = subbuf_start_default_callback;
462         if (!cb->buf_mapped)
463                 cb->buf_mapped = buf_mapped_default_callback;
464         if (!cb->buf_unmapped)
465                 cb->buf_unmapped = buf_unmapped_default_callback;
466         if (!cb->create_buf_file)
467                 cb->create_buf_file = create_buf_file_default_callback;
468         if (!cb->remove_buf_file)
469                 cb->remove_buf_file = remove_buf_file_default_callback;
470         chan->cb = cb;
471 }
472
473 /**
474  *      relay_hotcpu_callback - CPU hotplug callback
475  *      @nb: notifier block
476  *      @action: hotplug action to take
477  *      @hcpu: CPU number
478  *
479  *      Returns the success/failure of the operation. (%NOTIFY_OK, %NOTIFY_BAD)
480  */
481 static int __cpuinit relay_hotcpu_callback(struct notifier_block *nb,
482                                 unsigned long action,
483                                 void *hcpu)
484 {
485         unsigned int hotcpu = (unsigned long)hcpu;
486         struct rchan *chan;
487
488         switch(action) {
489         case CPU_UP_PREPARE:
490         case CPU_UP_PREPARE_FROZEN:
491                 mutex_lock(&relay_channels_mutex);
492                 list_for_each_entry(chan, &relay_channels, list) {
493                         if (chan->buf[hotcpu])
494                                 continue;
495                         chan->buf[hotcpu] = relay_open_buf(chan, hotcpu);
496                         if(!chan->buf[hotcpu]) {
497                                 printk(KERN_ERR
498                                         "relay_hotcpu_callback: cpu %d buffer "
499                                         "creation failed\n", hotcpu);
500                                 mutex_unlock(&relay_channels_mutex);
501                                 return NOTIFY_BAD;
502                         }
503                 }
504                 mutex_unlock(&relay_channels_mutex);
505                 break;
506         case CPU_DEAD:
507         case CPU_DEAD_FROZEN:
508                 /* No need to flush the cpu : will be flushed upon
509                  * final relay_flush() call. */
510                 break;
511         }
512         return NOTIFY_OK;
513 }
514
515 /**
516  *      relay_open - create a new relay channel
517  *      @base_filename: base name of files to create
518  *      @parent: dentry of parent directory, %NULL for root directory
519  *      @subbuf_size: size of sub-buffers
520  *      @n_subbufs: number of sub-buffers
521  *      @cb: client callback functions
522  *      @private_data: user-defined data
523  *
524  *      Returns channel pointer if successful, %NULL otherwise.
525  *
526  *      Creates a channel buffer for each cpu using the sizes and
527  *      attributes specified.  The created channel buffer files
528  *      will be named base_filename0...base_filenameN-1.  File
529  *      permissions will be %S_IRUSR.
530  */
531 struct rchan *relay_open(const char *base_filename,
532                          struct dentry *parent,
533                          size_t subbuf_size,
534                          size_t n_subbufs,
535                          struct rchan_callbacks *cb,
536                          void *private_data)
537 {
538         unsigned int i;
539         struct rchan *chan;
540         if (!base_filename)
541                 return NULL;
542
543         if (!(subbuf_size && n_subbufs))
544                 return NULL;
545
546         chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
547         if (!chan)
548                 return NULL;
549
550         chan->version = RELAYFS_CHANNEL_VERSION;
551         chan->n_subbufs = n_subbufs;
552         chan->subbuf_size = subbuf_size;
553         chan->alloc_size = FIX_SIZE(subbuf_size * n_subbufs);
554         chan->parent = parent;
555         chan->private_data = private_data;
556         strlcpy(chan->base_filename, base_filename, NAME_MAX);
557         setup_callbacks(chan, cb);
558         kref_init(&chan->kref);
559
560         mutex_lock(&relay_channels_mutex);
561         for_each_online_cpu(i) {
562                 chan->buf[i] = relay_open_buf(chan, i);
563                 if (!chan->buf[i])
564                         goto free_bufs;
565         }
566         list_add(&chan->list, &relay_channels);
567         mutex_unlock(&relay_channels_mutex);
568
569         return chan;
570
571 free_bufs:
572         for_each_online_cpu(i) {
573                 if (!chan->buf[i])
574                         break;
575                 relay_close_buf(chan->buf[i]);
576         }
577
578         kref_put(&chan->kref, relay_destroy_channel);
579         mutex_unlock(&relay_channels_mutex);
580         return NULL;
581 }
582 EXPORT_SYMBOL_GPL(relay_open);
583
584 /**
585  *      relay_switch_subbuf - switch to a new sub-buffer
586  *      @buf: channel buffer
587  *      @length: size of current event
588  *
589  *      Returns either the length passed in or 0 if full.
590  *
591  *      Performs sub-buffer-switch tasks such as invoking callbacks,
592  *      updating padding counts, waking up readers, etc.
593  */
594 size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
595 {
596         void *old, *new;
597         size_t old_subbuf, new_subbuf;
598
599         if (unlikely(length > buf->chan->subbuf_size))
600                 goto toobig;
601
602         if (buf->offset != buf->chan->subbuf_size + 1) {
603                 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
604                 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
605                 buf->padding[old_subbuf] = buf->prev_padding;
606                 buf->subbufs_produced++;
607                 buf->dentry->d_inode->i_size += buf->chan->subbuf_size -
608                         buf->padding[old_subbuf];
609                 smp_mb();
610                 if (waitqueue_active(&buf->read_wait))
611                         /*
612                          * Calling wake_up_interruptible() from here
613                          * will deadlock if we happen to be logging
614                          * from the scheduler (trying to re-grab
615                          * rq->lock), so defer it.
616                          */
617                         __mod_timer(&buf->timer, jiffies + 1);
618         }
619
620         old = buf->data;
621         new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
622         new = buf->start + new_subbuf * buf->chan->subbuf_size;
623         buf->offset = 0;
624         if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
625                 buf->offset = buf->chan->subbuf_size + 1;
626                 return 0;
627         }
628         buf->data = new;
629         buf->padding[new_subbuf] = 0;
630
631         if (unlikely(length + buf->offset > buf->chan->subbuf_size))
632                 goto toobig;
633
634         return length;
635
636 toobig:
637         buf->chan->last_toobig = length;
638         return 0;
639 }
640 EXPORT_SYMBOL_GPL(relay_switch_subbuf);
641
642 /**
643  *      relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
644  *      @chan: the channel
645  *      @cpu: the cpu associated with the channel buffer to update
646  *      @subbufs_consumed: number of sub-buffers to add to current buf's count
647  *
648  *      Adds to the channel buffer's consumed sub-buffer count.
649  *      subbufs_consumed should be the number of sub-buffers newly consumed,
650  *      not the total consumed.
651  *
652  *      NOTE. Kernel clients don't need to call this function if the channel
653  *      mode is 'overwrite'.
654  */
655 void relay_subbufs_consumed(struct rchan *chan,
656                             unsigned int cpu,
657                             size_t subbufs_consumed)
658 {
659         struct rchan_buf *buf;
660
661         if (!chan)
662                 return;
663
664         if (cpu >= NR_CPUS || !chan->buf[cpu])
665                 return;
666
667         buf = chan->buf[cpu];
668         buf->subbufs_consumed += subbufs_consumed;
669         if (buf->subbufs_consumed > buf->subbufs_produced)
670                 buf->subbufs_consumed = buf->subbufs_produced;
671 }
672 EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
673
674 /**
675  *      relay_close - close the channel
676  *      @chan: the channel
677  *
678  *      Closes all channel buffers and frees the channel.
679  */
680 void relay_close(struct rchan *chan)
681 {
682         unsigned int i;
683
684         if (!chan)
685                 return;
686
687         mutex_lock(&relay_channels_mutex);
688         if (chan->is_global && chan->buf[0])
689                 relay_close_buf(chan->buf[0]);
690         else
691                 for_each_possible_cpu(i)
692                         if (chan->buf[i])
693                                 relay_close_buf(chan->buf[i]);
694
695         if (chan->last_toobig)
696                 printk(KERN_WARNING "relay: one or more items not logged "
697                        "[item size (%Zd) > sub-buffer size (%Zd)]\n",
698                        chan->last_toobig, chan->subbuf_size);
699
700         list_del(&chan->list);
701         kref_put(&chan->kref, relay_destroy_channel);
702         mutex_unlock(&relay_channels_mutex);
703 }
704 EXPORT_SYMBOL_GPL(relay_close);
705
706 /**
707  *      relay_flush - close the channel
708  *      @chan: the channel
709  *
710  *      Flushes all channel buffers, i.e. forces buffer switch.
711  */
712 void relay_flush(struct rchan *chan)
713 {
714         unsigned int i;
715
716         if (!chan)
717                 return;
718
719         if (chan->is_global && chan->buf[0]) {
720                 relay_switch_subbuf(chan->buf[0], 0);
721                 return;
722         }
723
724         mutex_lock(&relay_channels_mutex);
725         for_each_possible_cpu(i)
726                 if (chan->buf[i])
727                         relay_switch_subbuf(chan->buf[i], 0);
728         mutex_unlock(&relay_channels_mutex);
729 }
730 EXPORT_SYMBOL_GPL(relay_flush);
731
732 /**
733  *      relay_file_open - open file op for relay files
734  *      @inode: the inode
735  *      @filp: the file
736  *
737  *      Increments the channel buffer refcount.
738  */
739 static int relay_file_open(struct inode *inode, struct file *filp)
740 {
741         struct rchan_buf *buf = inode->i_private;
742         kref_get(&buf->kref);
743         filp->private_data = buf;
744
745         return 0;
746 }
747
748 /**
749  *      relay_file_mmap - mmap file op for relay files
750  *      @filp: the file
751  *      @vma: the vma describing what to map
752  *
753  *      Calls upon relay_mmap_buf() to map the file into user space.
754  */
755 static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
756 {
757         struct rchan_buf *buf = filp->private_data;
758         return relay_mmap_buf(buf, vma);
759 }
760
761 /**
762  *      relay_file_poll - poll file op for relay files
763  *      @filp: the file
764  *      @wait: poll table
765  *
766  *      Poll implemention.
767  */
768 static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
769 {
770         unsigned int mask = 0;
771         struct rchan_buf *buf = filp->private_data;
772
773         if (buf->finalized)
774                 return POLLERR;
775
776         if (filp->f_mode & FMODE_READ) {
777                 poll_wait(filp, &buf->read_wait, wait);
778                 if (!relay_buf_empty(buf))
779                         mask |= POLLIN | POLLRDNORM;
780         }
781
782         return mask;
783 }
784
785 /**
786  *      relay_file_release - release file op for relay files
787  *      @inode: the inode
788  *      @filp: the file
789  *
790  *      Decrements the channel refcount, as the filesystem is
791  *      no longer using it.
792  */
793 static int relay_file_release(struct inode *inode, struct file *filp)
794 {
795         struct rchan_buf *buf = filp->private_data;
796         kref_put(&buf->kref, relay_remove_buf);
797
798         return 0;
799 }
800
801 /*
802  *      relay_file_read_consume - update the consumed count for the buffer
803  */
804 static void relay_file_read_consume(struct rchan_buf *buf,
805                                     size_t read_pos,
806                                     size_t bytes_consumed)
807 {
808         size_t subbuf_size = buf->chan->subbuf_size;
809         size_t n_subbufs = buf->chan->n_subbufs;
810         size_t read_subbuf;
811
812         if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
813                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
814                 buf->bytes_consumed = 0;
815         }
816
817         buf->bytes_consumed += bytes_consumed;
818         if (!read_pos)
819                 read_subbuf = buf->subbufs_consumed % n_subbufs;
820         else
821                 read_subbuf = read_pos / buf->chan->subbuf_size;
822         if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
823                 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
824                     (buf->offset == subbuf_size))
825                         return;
826                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
827                 buf->bytes_consumed = 0;
828         }
829 }
830
831 /*
832  *      relay_file_read_avail - boolean, are there unconsumed bytes available?
833  */
834 static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
835 {
836         size_t subbuf_size = buf->chan->subbuf_size;
837         size_t n_subbufs = buf->chan->n_subbufs;
838         size_t produced = buf->subbufs_produced;
839         size_t consumed = buf->subbufs_consumed;
840
841         relay_file_read_consume(buf, read_pos, 0);
842
843         if (unlikely(buf->offset > subbuf_size)) {
844                 if (produced == consumed)
845                         return 0;
846                 return 1;
847         }
848
849         if (unlikely(produced - consumed >= n_subbufs)) {
850                 consumed = produced - n_subbufs + 1;
851                 buf->subbufs_consumed = consumed;
852                 buf->bytes_consumed = 0;
853         }
854
855         produced = (produced % n_subbufs) * subbuf_size + buf->offset;
856         consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
857
858         if (consumed > produced)
859                 produced += n_subbufs * subbuf_size;
860
861         if (consumed == produced)
862                 return 0;
863
864         return 1;
865 }
866
867 /**
868  *      relay_file_read_subbuf_avail - return bytes available in sub-buffer
869  *      @read_pos: file read position
870  *      @buf: relay channel buffer
871  */
872 static size_t relay_file_read_subbuf_avail(size_t read_pos,
873                                            struct rchan_buf *buf)
874 {
875         size_t padding, avail = 0;
876         size_t read_subbuf, read_offset, write_subbuf, write_offset;
877         size_t subbuf_size = buf->chan->subbuf_size;
878
879         write_subbuf = (buf->data - buf->start) / subbuf_size;
880         write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
881         read_subbuf = read_pos / subbuf_size;
882         read_offset = read_pos % subbuf_size;
883         padding = buf->padding[read_subbuf];
884
885         if (read_subbuf == write_subbuf) {
886                 if (read_offset + padding < write_offset)
887                         avail = write_offset - (read_offset + padding);
888         } else
889                 avail = (subbuf_size - padding) - read_offset;
890
891         return avail;
892 }
893
894 /**
895  *      relay_file_read_start_pos - find the first available byte to read
896  *      @read_pos: file read position
897  *      @buf: relay channel buffer
898  *
899  *      If the @read_pos is in the middle of padding, return the
900  *      position of the first actually available byte, otherwise
901  *      return the original value.
902  */
903 static size_t relay_file_read_start_pos(size_t read_pos,
904                                         struct rchan_buf *buf)
905 {
906         size_t read_subbuf, padding, padding_start, padding_end;
907         size_t subbuf_size = buf->chan->subbuf_size;
908         size_t n_subbufs = buf->chan->n_subbufs;
909         size_t consumed = buf->subbufs_consumed % n_subbufs;
910
911         if (!read_pos)
912                 read_pos = consumed * subbuf_size + buf->bytes_consumed;
913         read_subbuf = read_pos / subbuf_size;
914         padding = buf->padding[read_subbuf];
915         padding_start = (read_subbuf + 1) * subbuf_size - padding;
916         padding_end = (read_subbuf + 1) * subbuf_size;
917         if (read_pos >= padding_start && read_pos < padding_end) {
918                 read_subbuf = (read_subbuf + 1) % n_subbufs;
919                 read_pos = read_subbuf * subbuf_size;
920         }
921
922         return read_pos;
923 }
924
925 /**
926  *      relay_file_read_end_pos - return the new read position
927  *      @read_pos: file read position
928  *      @buf: relay channel buffer
929  *      @count: number of bytes to be read
930  */
931 static size_t relay_file_read_end_pos(struct rchan_buf *buf,
932                                       size_t read_pos,
933                                       size_t count)
934 {
935         size_t read_subbuf, padding, end_pos;
936         size_t subbuf_size = buf->chan->subbuf_size;
937         size_t n_subbufs = buf->chan->n_subbufs;
938
939         read_subbuf = read_pos / subbuf_size;
940         padding = buf->padding[read_subbuf];
941         if (read_pos % subbuf_size + count + padding == subbuf_size)
942                 end_pos = (read_subbuf + 1) * subbuf_size;
943         else
944                 end_pos = read_pos + count;
945         if (end_pos >= subbuf_size * n_subbufs)
946                 end_pos = 0;
947
948         return end_pos;
949 }
950
951 /*
952  *      subbuf_read_actor - read up to one subbuf's worth of data
953  */
954 static int subbuf_read_actor(size_t read_start,
955                              struct rchan_buf *buf,
956                              size_t avail,
957                              read_descriptor_t *desc,
958                              read_actor_t actor)
959 {
960         void *from;
961         int ret = 0;
962
963         from = buf->start + read_start;
964         ret = avail;
965         if (copy_to_user(desc->arg.buf, from, avail)) {
966                 desc->error = -EFAULT;
967                 ret = 0;
968         }
969         desc->arg.data += ret;
970         desc->written += ret;
971         desc->count -= ret;
972
973         return ret;
974 }
975
976 typedef int (*subbuf_actor_t) (size_t read_start,
977                                struct rchan_buf *buf,
978                                size_t avail,
979                                read_descriptor_t *desc,
980                                read_actor_t actor);
981
982 /*
983  *      relay_file_read_subbufs - read count bytes, bridging subbuf boundaries
984  */
985 static ssize_t relay_file_read_subbufs(struct file *filp, loff_t *ppos,
986                                         subbuf_actor_t subbuf_actor,
987                                         read_actor_t actor,
988                                         read_descriptor_t *desc)
989 {
990         struct rchan_buf *buf = filp->private_data;
991         size_t read_start, avail;
992         int ret;
993
994         if (!desc->count)
995                 return 0;
996
997         mutex_lock(&filp->f_path.dentry->d_inode->i_mutex);
998         do {
999                 if (!relay_file_read_avail(buf, *ppos))
1000                         break;
1001
1002                 read_start = relay_file_read_start_pos(*ppos, buf);
1003                 avail = relay_file_read_subbuf_avail(read_start, buf);
1004                 if (!avail)
1005                         break;
1006
1007                 avail = min(desc->count, avail);
1008                 ret = subbuf_actor(read_start, buf, avail, desc, actor);
1009                 if (desc->error < 0)
1010                         break;
1011
1012                 if (ret) {
1013                         relay_file_read_consume(buf, read_start, ret);
1014                         *ppos = relay_file_read_end_pos(buf, read_start, ret);
1015                 }
1016         } while (desc->count && ret);
1017         mutex_unlock(&filp->f_path.dentry->d_inode->i_mutex);
1018
1019         return desc->written;
1020 }
1021
1022 static ssize_t relay_file_read(struct file *filp,
1023                                char __user *buffer,
1024                                size_t count,
1025                                loff_t *ppos)
1026 {
1027         read_descriptor_t desc;
1028         desc.written = 0;
1029         desc.count = count;
1030         desc.arg.buf = buffer;
1031         desc.error = 0;
1032         return relay_file_read_subbufs(filp, ppos, subbuf_read_actor,
1033                                        NULL, &desc);
1034 }
1035
1036 static void relay_consume_bytes(struct rchan_buf *rbuf, int bytes_consumed)
1037 {
1038         rbuf->bytes_consumed += bytes_consumed;
1039
1040         if (rbuf->bytes_consumed >= rbuf->chan->subbuf_size) {
1041                 relay_subbufs_consumed(rbuf->chan, rbuf->cpu, 1);
1042                 rbuf->bytes_consumed %= rbuf->chan->subbuf_size;
1043         }
1044 }
1045
1046 static void relay_pipe_buf_release(struct pipe_inode_info *pipe,
1047                                    struct pipe_buffer *buf)
1048 {
1049         struct rchan_buf *rbuf;
1050
1051         rbuf = (struct rchan_buf *)page_private(buf->page);
1052         relay_consume_bytes(rbuf, buf->private);
1053 }
1054
1055 static struct pipe_buf_operations relay_pipe_buf_ops = {
1056         .can_merge = 0,
1057         .map = generic_pipe_buf_map,
1058         .unmap = generic_pipe_buf_unmap,
1059         .confirm = generic_pipe_buf_confirm,
1060         .release = relay_pipe_buf_release,
1061         .steal = generic_pipe_buf_steal,
1062         .get = generic_pipe_buf_get,
1063 };
1064
1065 /*
1066  *      subbuf_splice_actor - splice up to one subbuf's worth of data
1067  */
1068 static int subbuf_splice_actor(struct file *in,
1069                                loff_t *ppos,
1070                                struct pipe_inode_info *pipe,
1071                                size_t len,
1072                                unsigned int flags,
1073                                int *nonpad_ret)
1074 {
1075         unsigned int pidx, poff, total_len, subbuf_pages, ret;
1076         struct rchan_buf *rbuf = in->private_data;
1077         unsigned int subbuf_size = rbuf->chan->subbuf_size;
1078         uint64_t pos = (uint64_t) *ppos;
1079         uint32_t alloc_size = (uint32_t) rbuf->chan->alloc_size;
1080         size_t read_start = (size_t) do_div(pos, alloc_size);
1081         size_t read_subbuf = read_start / subbuf_size;
1082         size_t padding = rbuf->padding[read_subbuf];
1083         size_t nonpad_end = read_subbuf * subbuf_size + subbuf_size - padding;
1084         struct page *pages[PIPE_BUFFERS];
1085         struct partial_page partial[PIPE_BUFFERS];
1086         struct splice_pipe_desc spd = {
1087                 .pages = pages,
1088                 .nr_pages = 0,
1089                 .partial = partial,
1090                 .flags = flags,
1091                 .ops = &relay_pipe_buf_ops,
1092         };
1093
1094         if (rbuf->subbufs_produced == rbuf->subbufs_consumed)
1095                 return 0;
1096
1097         /*
1098          * Adjust read len, if longer than what is available
1099          */
1100         if (len > (subbuf_size - read_start % subbuf_size))
1101                 len = subbuf_size - read_start % subbuf_size;
1102
1103         subbuf_pages = rbuf->chan->alloc_size >> PAGE_SHIFT;
1104         pidx = (read_start / PAGE_SIZE) % subbuf_pages;
1105         poff = read_start & ~PAGE_MASK;
1106
1107         for (total_len = 0; spd.nr_pages < subbuf_pages; spd.nr_pages++) {
1108                 unsigned int this_len, this_end, private;
1109                 unsigned int cur_pos = read_start + total_len;
1110
1111                 if (!len)
1112                         break;
1113
1114                 this_len = min_t(unsigned long, len, PAGE_SIZE - poff);
1115                 private = this_len;
1116
1117                 spd.pages[spd.nr_pages] = rbuf->page_array[pidx];
1118                 spd.partial[spd.nr_pages].offset = poff;
1119
1120                 this_end = cur_pos + this_len;
1121                 if (this_end >= nonpad_end) {
1122                         this_len = nonpad_end - cur_pos;
1123                         private = this_len + padding;
1124                 }
1125                 spd.partial[spd.nr_pages].len = this_len;
1126                 spd.partial[spd.nr_pages].private = private;
1127
1128                 len -= this_len;
1129                 total_len += this_len;
1130                 poff = 0;
1131                 pidx = (pidx + 1) % subbuf_pages;
1132
1133                 if (this_end >= nonpad_end) {
1134                         spd.nr_pages++;
1135                         break;
1136                 }
1137         }
1138
1139         if (!spd.nr_pages)
1140                 return 0;
1141
1142         ret = *nonpad_ret = splice_to_pipe(pipe, &spd);
1143         if (ret < 0 || ret < total_len)
1144                 return ret;
1145
1146         if (read_start + ret == nonpad_end)
1147                 ret += padding;
1148
1149         return ret;
1150 }
1151
1152 static ssize_t relay_file_splice_read(struct file *in,
1153                                       loff_t *ppos,
1154                                       struct pipe_inode_info *pipe,
1155                                       size_t len,
1156                                       unsigned int flags)
1157 {
1158         ssize_t spliced;
1159         int ret;
1160         int nonpad_ret = 0;
1161
1162         ret = 0;
1163         spliced = 0;
1164
1165         while (len) {
1166                 ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret);
1167                 if (ret < 0)
1168                         break;
1169                 else if (!ret) {
1170                         if (spliced)
1171                                 break;
1172                         if (flags & SPLICE_F_NONBLOCK) {
1173                                 ret = -EAGAIN;
1174                                 break;
1175                         }
1176                 }
1177
1178                 *ppos += ret;
1179                 if (ret > len)
1180                         len = 0;
1181                 else
1182                         len -= ret;
1183                 spliced += nonpad_ret;
1184                 nonpad_ret = 0;
1185         }
1186
1187         if (spliced)
1188                 return spliced;
1189
1190         return ret;
1191 }
1192
1193 const struct file_operations relay_file_operations = {
1194         .open           = relay_file_open,
1195         .poll           = relay_file_poll,
1196         .mmap           = relay_file_mmap,
1197         .read           = relay_file_read,
1198         .llseek         = no_llseek,
1199         .release        = relay_file_release,
1200         .splice_read    = relay_file_splice_read,
1201 };
1202 EXPORT_SYMBOL_GPL(relay_file_operations);
1203
1204 static __init int relay_init(void)
1205 {
1206
1207         hotcpu_notifier(relay_hotcpu_callback, 0);
1208         return 0;
1209 }
1210
1211 module_init(relay_init);