Merge branch 'for-linus-4.8' of git://git.kernel.org/pub/scm/linux/kernel/git/mason...
[cascardo/linux.git] / 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/export.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  * fault() vm_op implementation for relay file mapping.
41  */
42 static int relay_buf_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
43 {
44         struct page *page;
45         struct rchan_buf *buf = vma->vm_private_data;
46         pgoff_t pgoff = vmf->pgoff;
47
48         if (!buf)
49                 return VM_FAULT_OOM;
50
51         page = vmalloc_to_page(buf->start + (pgoff << PAGE_SHIFT));
52         if (!page)
53                 return VM_FAULT_SIGBUS;
54         get_page(page);
55         vmf->page = page;
56
57         return 0;
58 }
59
60 /*
61  * vm_ops for relay file mappings.
62  */
63 static const struct vm_operations_struct relay_file_mmap_ops = {
64         .fault = relay_buf_fault,
65         .close = relay_file_mmap_close,
66 };
67
68 /*
69  * allocate an array of pointers of struct page
70  */
71 static struct page **relay_alloc_page_array(unsigned int n_pages)
72 {
73         const size_t pa_size = n_pages * sizeof(struct page *);
74         if (pa_size > PAGE_SIZE)
75                 return vzalloc(pa_size);
76         return kzalloc(pa_size, GFP_KERNEL);
77 }
78
79 /*
80  * free an array of pointers of struct page
81  */
82 static void relay_free_page_array(struct page **array)
83 {
84         kvfree(array);
85 }
86
87 /**
88  *      relay_mmap_buf: - mmap channel buffer to process address space
89  *      @buf: relay channel buffer
90  *      @vma: vm_area_struct describing memory to be mapped
91  *
92  *      Returns 0 if ok, negative on error
93  *
94  *      Caller should already have grabbed mmap_sem.
95  */
96 static int relay_mmap_buf(struct rchan_buf *buf, struct vm_area_struct *vma)
97 {
98         unsigned long length = vma->vm_end - vma->vm_start;
99         struct file *filp = vma->vm_file;
100
101         if (!buf)
102                 return -EBADF;
103
104         if (length != (unsigned long)buf->chan->alloc_size)
105                 return -EINVAL;
106
107         vma->vm_ops = &relay_file_mmap_ops;
108         vma->vm_flags |= VM_DONTEXPAND;
109         vma->vm_private_data = buf;
110         buf->chan->cb->buf_mapped(buf, filp);
111
112         return 0;
113 }
114
115 /**
116  *      relay_alloc_buf - allocate a channel buffer
117  *      @buf: the buffer struct
118  *      @size: total size of the buffer
119  *
120  *      Returns a pointer to the resulting buffer, %NULL if unsuccessful. The
121  *      passed in size will get page aligned, if it isn't already.
122  */
123 static void *relay_alloc_buf(struct rchan_buf *buf, size_t *size)
124 {
125         void *mem;
126         unsigned int i, j, n_pages;
127
128         *size = PAGE_ALIGN(*size);
129         n_pages = *size >> PAGE_SHIFT;
130
131         buf->page_array = relay_alloc_page_array(n_pages);
132         if (!buf->page_array)
133                 return NULL;
134
135         for (i = 0; i < n_pages; i++) {
136                 buf->page_array[i] = alloc_page(GFP_KERNEL);
137                 if (unlikely(!buf->page_array[i]))
138                         goto depopulate;
139                 set_page_private(buf->page_array[i], (unsigned long)buf);
140         }
141         mem = vmap(buf->page_array, n_pages, VM_MAP, PAGE_KERNEL);
142         if (!mem)
143                 goto depopulate;
144
145         memset(mem, 0, *size);
146         buf->page_count = n_pages;
147         return mem;
148
149 depopulate:
150         for (j = 0; j < i; j++)
151                 __free_page(buf->page_array[j]);
152         relay_free_page_array(buf->page_array);
153         return NULL;
154 }
155
156 /**
157  *      relay_create_buf - allocate and initialize a channel buffer
158  *      @chan: the relay channel
159  *
160  *      Returns channel buffer if successful, %NULL otherwise.
161  */
162 static struct rchan_buf *relay_create_buf(struct rchan *chan)
163 {
164         struct rchan_buf *buf;
165
166         if (chan->n_subbufs > UINT_MAX / sizeof(size_t *))
167                 return NULL;
168
169         buf = kzalloc(sizeof(struct rchan_buf), GFP_KERNEL);
170         if (!buf)
171                 return NULL;
172         buf->padding = kmalloc(chan->n_subbufs * sizeof(size_t *), GFP_KERNEL);
173         if (!buf->padding)
174                 goto free_buf;
175
176         buf->start = relay_alloc_buf(buf, &chan->alloc_size);
177         if (!buf->start)
178                 goto free_buf;
179
180         buf->chan = chan;
181         kref_get(&buf->chan->kref);
182         return buf;
183
184 free_buf:
185         kfree(buf->padding);
186         kfree(buf);
187         return NULL;
188 }
189
190 /**
191  *      relay_destroy_channel - free the channel struct
192  *      @kref: target kernel reference that contains the relay channel
193  *
194  *      Should only be called from kref_put().
195  */
196 static void relay_destroy_channel(struct kref *kref)
197 {
198         struct rchan *chan = container_of(kref, struct rchan, kref);
199         kfree(chan);
200 }
201
202 /**
203  *      relay_destroy_buf - destroy an rchan_buf struct and associated buffer
204  *      @buf: the buffer struct
205  */
206 static void relay_destroy_buf(struct rchan_buf *buf)
207 {
208         struct rchan *chan = buf->chan;
209         unsigned int i;
210
211         if (likely(buf->start)) {
212                 vunmap(buf->start);
213                 for (i = 0; i < buf->page_count; i++)
214                         __free_page(buf->page_array[i]);
215                 relay_free_page_array(buf->page_array);
216         }
217         chan->buf[buf->cpu] = NULL;
218         kfree(buf->padding);
219         kfree(buf);
220         kref_put(&chan->kref, relay_destroy_channel);
221 }
222
223 /**
224  *      relay_remove_buf - remove a channel buffer
225  *      @kref: target kernel reference that contains the relay buffer
226  *
227  *      Removes the file from the filesystem, which also frees the
228  *      rchan_buf_struct and the channel buffer.  Should only be called from
229  *      kref_put().
230  */
231 static void relay_remove_buf(struct kref *kref)
232 {
233         struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
234         relay_destroy_buf(buf);
235 }
236
237 /**
238  *      relay_buf_empty - boolean, is the channel buffer empty?
239  *      @buf: channel buffer
240  *
241  *      Returns 1 if the buffer is empty, 0 otherwise.
242  */
243 static int relay_buf_empty(struct rchan_buf *buf)
244 {
245         return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1;
246 }
247
248 /**
249  *      relay_buf_full - boolean, is the channel buffer full?
250  *      @buf: channel buffer
251  *
252  *      Returns 1 if the buffer is full, 0 otherwise.
253  */
254 int relay_buf_full(struct rchan_buf *buf)
255 {
256         size_t ready = buf->subbufs_produced - buf->subbufs_consumed;
257         return (ready >= buf->chan->n_subbufs) ? 1 : 0;
258 }
259 EXPORT_SYMBOL_GPL(relay_buf_full);
260
261 /*
262  * High-level relay kernel API and associated functions.
263  */
264
265 /*
266  * rchan_callback implementations defining default channel behavior.  Used
267  * in place of corresponding NULL values in client callback struct.
268  */
269
270 /*
271  * subbuf_start() default callback.  Does nothing.
272  */
273 static int subbuf_start_default_callback (struct rchan_buf *buf,
274                                           void *subbuf,
275                                           void *prev_subbuf,
276                                           size_t prev_padding)
277 {
278         if (relay_buf_full(buf))
279                 return 0;
280
281         return 1;
282 }
283
284 /*
285  * buf_mapped() default callback.  Does nothing.
286  */
287 static void buf_mapped_default_callback(struct rchan_buf *buf,
288                                         struct file *filp)
289 {
290 }
291
292 /*
293  * buf_unmapped() default callback.  Does nothing.
294  */
295 static void buf_unmapped_default_callback(struct rchan_buf *buf,
296                                           struct file *filp)
297 {
298 }
299
300 /*
301  * create_buf_file_create() default callback.  Does nothing.
302  */
303 static struct dentry *create_buf_file_default_callback(const char *filename,
304                                                        struct dentry *parent,
305                                                        umode_t mode,
306                                                        struct rchan_buf *buf,
307                                                        int *is_global)
308 {
309         return NULL;
310 }
311
312 /*
313  * remove_buf_file() default callback.  Does nothing.
314  */
315 static int remove_buf_file_default_callback(struct dentry *dentry)
316 {
317         return -EINVAL;
318 }
319
320 /* relay channel default callbacks */
321 static struct rchan_callbacks default_channel_callbacks = {
322         .subbuf_start = subbuf_start_default_callback,
323         .buf_mapped = buf_mapped_default_callback,
324         .buf_unmapped = buf_unmapped_default_callback,
325         .create_buf_file = create_buf_file_default_callback,
326         .remove_buf_file = remove_buf_file_default_callback,
327 };
328
329 /**
330  *      wakeup_readers - wake up readers waiting on a channel
331  *      @data: contains the channel buffer
332  *
333  *      This is the timer function used to defer reader waking.
334  */
335 static void wakeup_readers(unsigned long data)
336 {
337         struct rchan_buf *buf = (struct rchan_buf *)data;
338         wake_up_interruptible(&buf->read_wait);
339 }
340
341 /**
342  *      __relay_reset - reset a channel buffer
343  *      @buf: the channel buffer
344  *      @init: 1 if this is a first-time initialization
345  *
346  *      See relay_reset() for description of effect.
347  */
348 static void __relay_reset(struct rchan_buf *buf, unsigned int init)
349 {
350         size_t i;
351
352         if (init) {
353                 init_waitqueue_head(&buf->read_wait);
354                 kref_init(&buf->kref);
355                 setup_timer(&buf->timer, wakeup_readers, (unsigned long)buf);
356         } else
357                 del_timer_sync(&buf->timer);
358
359         buf->subbufs_produced = 0;
360         buf->subbufs_consumed = 0;
361         buf->bytes_consumed = 0;
362         buf->finalized = 0;
363         buf->data = buf->start;
364         buf->offset = 0;
365
366         for (i = 0; i < buf->chan->n_subbufs; i++)
367                 buf->padding[i] = 0;
368
369         buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0);
370 }
371
372 /**
373  *      relay_reset - reset the channel
374  *      @chan: the channel
375  *
376  *      This has the effect of erasing all data from all channel buffers
377  *      and restarting the channel in its initial state.  The buffers
378  *      are not freed, so any mappings are still in effect.
379  *
380  *      NOTE. Care should be taken that the channel isn't actually
381  *      being used by anything when this call is made.
382  */
383 void relay_reset(struct rchan *chan)
384 {
385         unsigned int i;
386
387         if (!chan)
388                 return;
389
390         if (chan->is_global && chan->buf[0]) {
391                 __relay_reset(chan->buf[0], 0);
392                 return;
393         }
394
395         mutex_lock(&relay_channels_mutex);
396         for_each_possible_cpu(i)
397                 if (chan->buf[i])
398                         __relay_reset(chan->buf[i], 0);
399         mutex_unlock(&relay_channels_mutex);
400 }
401 EXPORT_SYMBOL_GPL(relay_reset);
402
403 static inline void relay_set_buf_dentry(struct rchan_buf *buf,
404                                         struct dentry *dentry)
405 {
406         buf->dentry = dentry;
407         d_inode(buf->dentry)->i_size = buf->early_bytes;
408 }
409
410 static struct dentry *relay_create_buf_file(struct rchan *chan,
411                                             struct rchan_buf *buf,
412                                             unsigned int cpu)
413 {
414         struct dentry *dentry;
415         char *tmpname;
416
417         tmpname = kzalloc(NAME_MAX + 1, GFP_KERNEL);
418         if (!tmpname)
419                 return NULL;
420         snprintf(tmpname, NAME_MAX, "%s%d", chan->base_filename, cpu);
421
422         /* Create file in fs */
423         dentry = chan->cb->create_buf_file(tmpname, chan->parent,
424                                            S_IRUSR, buf,
425                                            &chan->is_global);
426
427         kfree(tmpname);
428
429         return dentry;
430 }
431
432 /*
433  *      relay_open_buf - create a new relay channel buffer
434  *
435  *      used by relay_open() and CPU hotplug.
436  */
437 static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu)
438 {
439         struct rchan_buf *buf = NULL;
440         struct dentry *dentry;
441
442         if (chan->is_global)
443                 return chan->buf[0];
444
445         buf = relay_create_buf(chan);
446         if (!buf)
447                 return NULL;
448
449         if (chan->has_base_filename) {
450                 dentry = relay_create_buf_file(chan, buf, cpu);
451                 if (!dentry)
452                         goto free_buf;
453                 relay_set_buf_dentry(buf, dentry);
454         } else {
455                 /* Only retrieve global info, nothing more, nothing less */
456                 dentry = chan->cb->create_buf_file(NULL, NULL,
457                                                    S_IRUSR, buf,
458                                                    &chan->is_global);
459                 if (WARN_ON(dentry))
460                         goto free_buf;
461         }
462
463         buf->cpu = cpu;
464         __relay_reset(buf, 1);
465
466         if(chan->is_global) {
467                 chan->buf[0] = buf;
468                 buf->cpu = 0;
469         }
470
471         return buf;
472
473 free_buf:
474         relay_destroy_buf(buf);
475         return NULL;
476 }
477
478 /**
479  *      relay_close_buf - close a channel buffer
480  *      @buf: channel buffer
481  *
482  *      Marks the buffer finalized and restores the default callbacks.
483  *      The channel buffer and channel buffer data structure are then freed
484  *      automatically when the last reference is given up.
485  */
486 static void relay_close_buf(struct rchan_buf *buf)
487 {
488         buf->finalized = 1;
489         del_timer_sync(&buf->timer);
490         buf->chan->cb->remove_buf_file(buf->dentry);
491         kref_put(&buf->kref, relay_remove_buf);
492 }
493
494 static void setup_callbacks(struct rchan *chan,
495                                    struct rchan_callbacks *cb)
496 {
497         if (!cb) {
498                 chan->cb = &default_channel_callbacks;
499                 return;
500         }
501
502         if (!cb->subbuf_start)
503                 cb->subbuf_start = subbuf_start_default_callback;
504         if (!cb->buf_mapped)
505                 cb->buf_mapped = buf_mapped_default_callback;
506         if (!cb->buf_unmapped)
507                 cb->buf_unmapped = buf_unmapped_default_callback;
508         if (!cb->create_buf_file)
509                 cb->create_buf_file = create_buf_file_default_callback;
510         if (!cb->remove_buf_file)
511                 cb->remove_buf_file = remove_buf_file_default_callback;
512         chan->cb = cb;
513 }
514
515 /**
516  *      relay_hotcpu_callback - CPU hotplug callback
517  *      @nb: notifier block
518  *      @action: hotplug action to take
519  *      @hcpu: CPU number
520  *
521  *      Returns the success/failure of the operation. (%NOTIFY_OK, %NOTIFY_BAD)
522  */
523 static int relay_hotcpu_callback(struct notifier_block *nb,
524                                 unsigned long action,
525                                 void *hcpu)
526 {
527         unsigned int hotcpu = (unsigned long)hcpu;
528         struct rchan *chan;
529
530         switch(action) {
531         case CPU_UP_PREPARE:
532         case CPU_UP_PREPARE_FROZEN:
533                 mutex_lock(&relay_channels_mutex);
534                 list_for_each_entry(chan, &relay_channels, list) {
535                         if (chan->buf[hotcpu])
536                                 continue;
537                         chan->buf[hotcpu] = relay_open_buf(chan, hotcpu);
538                         if(!chan->buf[hotcpu]) {
539                                 printk(KERN_ERR
540                                         "relay_hotcpu_callback: cpu %d buffer "
541                                         "creation failed\n", hotcpu);
542                                 mutex_unlock(&relay_channels_mutex);
543                                 return notifier_from_errno(-ENOMEM);
544                         }
545                 }
546                 mutex_unlock(&relay_channels_mutex);
547                 break;
548         case CPU_DEAD:
549         case CPU_DEAD_FROZEN:
550                 /* No need to flush the cpu : will be flushed upon
551                  * final relay_flush() call. */
552                 break;
553         }
554         return NOTIFY_OK;
555 }
556
557 /**
558  *      relay_open - create a new relay channel
559  *      @base_filename: base name of files to create, %NULL for buffering only
560  *      @parent: dentry of parent directory, %NULL for root directory or buffer
561  *      @subbuf_size: size of sub-buffers
562  *      @n_subbufs: number of sub-buffers
563  *      @cb: client callback functions
564  *      @private_data: user-defined data
565  *
566  *      Returns channel pointer if successful, %NULL otherwise.
567  *
568  *      Creates a channel buffer for each cpu using the sizes and
569  *      attributes specified.  The created channel buffer files
570  *      will be named base_filename0...base_filenameN-1.  File
571  *      permissions will be %S_IRUSR.
572  *
573  *      If opening a buffer (@parent = NULL) that you later wish to register
574  *      in a filesystem, call relay_late_setup_files() once the @parent dentry
575  *      is available.
576  */
577 struct rchan *relay_open(const char *base_filename,
578                          struct dentry *parent,
579                          size_t subbuf_size,
580                          size_t n_subbufs,
581                          struct rchan_callbacks *cb,
582                          void *private_data)
583 {
584         unsigned int i;
585         struct rchan *chan;
586
587         if (!(subbuf_size && n_subbufs))
588                 return NULL;
589         if (subbuf_size > UINT_MAX / n_subbufs)
590                 return NULL;
591
592         chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
593         if (!chan)
594                 return NULL;
595
596         chan->version = RELAYFS_CHANNEL_VERSION;
597         chan->n_subbufs = n_subbufs;
598         chan->subbuf_size = subbuf_size;
599         chan->alloc_size = PAGE_ALIGN(subbuf_size * n_subbufs);
600         chan->parent = parent;
601         chan->private_data = private_data;
602         if (base_filename) {
603                 chan->has_base_filename = 1;
604                 strlcpy(chan->base_filename, base_filename, NAME_MAX);
605         }
606         setup_callbacks(chan, cb);
607         kref_init(&chan->kref);
608
609         mutex_lock(&relay_channels_mutex);
610         for_each_online_cpu(i) {
611                 chan->buf[i] = relay_open_buf(chan, i);
612                 if (!chan->buf[i])
613                         goto free_bufs;
614         }
615         list_add(&chan->list, &relay_channels);
616         mutex_unlock(&relay_channels_mutex);
617
618         return chan;
619
620 free_bufs:
621         for_each_possible_cpu(i) {
622                 if (chan->buf[i])
623                         relay_close_buf(chan->buf[i]);
624         }
625
626         kref_put(&chan->kref, relay_destroy_channel);
627         mutex_unlock(&relay_channels_mutex);
628         kfree(chan);
629         return NULL;
630 }
631 EXPORT_SYMBOL_GPL(relay_open);
632
633 struct rchan_percpu_buf_dispatcher {
634         struct rchan_buf *buf;
635         struct dentry *dentry;
636 };
637
638 /* Called in atomic context. */
639 static void __relay_set_buf_dentry(void *info)
640 {
641         struct rchan_percpu_buf_dispatcher *p = info;
642
643         relay_set_buf_dentry(p->buf, p->dentry);
644 }
645
646 /**
647  *      relay_late_setup_files - triggers file creation
648  *      @chan: channel to operate on
649  *      @base_filename: base name of files to create
650  *      @parent: dentry of parent directory, %NULL for root directory
651  *
652  *      Returns 0 if successful, non-zero otherwise.
653  *
654  *      Use to setup files for a previously buffer-only channel created
655  *      by relay_open() with a NULL parent dentry.
656  *
657  *      For example, this is useful for perfomring early tracing in kernel,
658  *      before VFS is up and then exposing the early results once the dentry
659  *      is available.
660  */
661 int relay_late_setup_files(struct rchan *chan,
662                            const char *base_filename,
663                            struct dentry *parent)
664 {
665         int err = 0;
666         unsigned int i, curr_cpu;
667         unsigned long flags;
668         struct dentry *dentry;
669         struct rchan_percpu_buf_dispatcher disp;
670
671         if (!chan || !base_filename)
672                 return -EINVAL;
673
674         strlcpy(chan->base_filename, base_filename, NAME_MAX);
675
676         mutex_lock(&relay_channels_mutex);
677         /* Is chan already set up? */
678         if (unlikely(chan->has_base_filename)) {
679                 mutex_unlock(&relay_channels_mutex);
680                 return -EEXIST;
681         }
682         chan->has_base_filename = 1;
683         chan->parent = parent;
684
685         if (chan->is_global) {
686                 err = -EINVAL;
687                 if (!WARN_ON_ONCE(!chan->buf[0])) {
688                         dentry = relay_create_buf_file(chan, chan->buf[0], 0);
689                         if (dentry && !WARN_ON_ONCE(!chan->is_global)) {
690                                 relay_set_buf_dentry(chan->buf[0], dentry);
691                                 err = 0;
692                         }
693                 }
694                 mutex_unlock(&relay_channels_mutex);
695                 return err;
696         }
697
698         curr_cpu = get_cpu();
699         /*
700          * The CPU hotplug notifier ran before us and created buffers with
701          * no files associated. So it's safe to call relay_setup_buf_file()
702          * on all currently online CPUs.
703          */
704         for_each_online_cpu(i) {
705                 if (unlikely(!chan->buf[i])) {
706                         WARN_ONCE(1, KERN_ERR "CPU has no buffer!\n");
707                         err = -EINVAL;
708                         break;
709                 }
710
711                 dentry = relay_create_buf_file(chan, chan->buf[i], i);
712                 if (unlikely(!dentry)) {
713                         err = -EINVAL;
714                         break;
715                 }
716
717                 if (curr_cpu == i) {
718                         local_irq_save(flags);
719                         relay_set_buf_dentry(chan->buf[i], dentry);
720                         local_irq_restore(flags);
721                 } else {
722                         disp.buf = chan->buf[i];
723                         disp.dentry = dentry;
724                         smp_mb();
725                         /* relay_channels_mutex must be held, so wait. */
726                         err = smp_call_function_single(i,
727                                                        __relay_set_buf_dentry,
728                                                        &disp, 1);
729                 }
730                 if (unlikely(err))
731                         break;
732         }
733         put_cpu();
734         mutex_unlock(&relay_channels_mutex);
735
736         return err;
737 }
738 EXPORT_SYMBOL_GPL(relay_late_setup_files);
739
740 /**
741  *      relay_switch_subbuf - switch to a new sub-buffer
742  *      @buf: channel buffer
743  *      @length: size of current event
744  *
745  *      Returns either the length passed in or 0 if full.
746  *
747  *      Performs sub-buffer-switch tasks such as invoking callbacks,
748  *      updating padding counts, waking up readers, etc.
749  */
750 size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
751 {
752         void *old, *new;
753         size_t old_subbuf, new_subbuf;
754
755         if (unlikely(length > buf->chan->subbuf_size))
756                 goto toobig;
757
758         if (buf->offset != buf->chan->subbuf_size + 1) {
759                 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
760                 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
761                 buf->padding[old_subbuf] = buf->prev_padding;
762                 buf->subbufs_produced++;
763                 if (buf->dentry)
764                         d_inode(buf->dentry)->i_size +=
765                                 buf->chan->subbuf_size -
766                                 buf->padding[old_subbuf];
767                 else
768                         buf->early_bytes += buf->chan->subbuf_size -
769                                             buf->padding[old_subbuf];
770                 smp_mb();
771                 if (waitqueue_active(&buf->read_wait))
772                         /*
773                          * Calling wake_up_interruptible() from here
774                          * will deadlock if we happen to be logging
775                          * from the scheduler (trying to re-grab
776                          * rq->lock), so defer it.
777                          */
778                         mod_timer(&buf->timer, jiffies + 1);
779         }
780
781         old = buf->data;
782         new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
783         new = buf->start + new_subbuf * buf->chan->subbuf_size;
784         buf->offset = 0;
785         if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
786                 buf->offset = buf->chan->subbuf_size + 1;
787                 return 0;
788         }
789         buf->data = new;
790         buf->padding[new_subbuf] = 0;
791
792         if (unlikely(length + buf->offset > buf->chan->subbuf_size))
793                 goto toobig;
794
795         return length;
796
797 toobig:
798         buf->chan->last_toobig = length;
799         return 0;
800 }
801 EXPORT_SYMBOL_GPL(relay_switch_subbuf);
802
803 /**
804  *      relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
805  *      @chan: the channel
806  *      @cpu: the cpu associated with the channel buffer to update
807  *      @subbufs_consumed: number of sub-buffers to add to current buf's count
808  *
809  *      Adds to the channel buffer's consumed sub-buffer count.
810  *      subbufs_consumed should be the number of sub-buffers newly consumed,
811  *      not the total consumed.
812  *
813  *      NOTE. Kernel clients don't need to call this function if the channel
814  *      mode is 'overwrite'.
815  */
816 void relay_subbufs_consumed(struct rchan *chan,
817                             unsigned int cpu,
818                             size_t subbufs_consumed)
819 {
820         struct rchan_buf *buf;
821
822         if (!chan)
823                 return;
824
825         if (cpu >= NR_CPUS || !chan->buf[cpu] ||
826                                         subbufs_consumed > chan->n_subbufs)
827                 return;
828
829         buf = chan->buf[cpu];
830         if (subbufs_consumed > buf->subbufs_produced - buf->subbufs_consumed)
831                 buf->subbufs_consumed = buf->subbufs_produced;
832         else
833                 buf->subbufs_consumed += subbufs_consumed;
834 }
835 EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
836
837 /**
838  *      relay_close - close the channel
839  *      @chan: the channel
840  *
841  *      Closes all channel buffers and frees the channel.
842  */
843 void relay_close(struct rchan *chan)
844 {
845         unsigned int i;
846
847         if (!chan)
848                 return;
849
850         mutex_lock(&relay_channels_mutex);
851         if (chan->is_global && chan->buf[0])
852                 relay_close_buf(chan->buf[0]);
853         else
854                 for_each_possible_cpu(i)
855                         if (chan->buf[i])
856                                 relay_close_buf(chan->buf[i]);
857
858         if (chan->last_toobig)
859                 printk(KERN_WARNING "relay: one or more items not logged "
860                        "[item size (%Zd) > sub-buffer size (%Zd)]\n",
861                        chan->last_toobig, chan->subbuf_size);
862
863         list_del(&chan->list);
864         kref_put(&chan->kref, relay_destroy_channel);
865         mutex_unlock(&relay_channels_mutex);
866 }
867 EXPORT_SYMBOL_GPL(relay_close);
868
869 /**
870  *      relay_flush - close the channel
871  *      @chan: the channel
872  *
873  *      Flushes all channel buffers, i.e. forces buffer switch.
874  */
875 void relay_flush(struct rchan *chan)
876 {
877         unsigned int i;
878
879         if (!chan)
880                 return;
881
882         if (chan->is_global && chan->buf[0]) {
883                 relay_switch_subbuf(chan->buf[0], 0);
884                 return;
885         }
886
887         mutex_lock(&relay_channels_mutex);
888         for_each_possible_cpu(i)
889                 if (chan->buf[i])
890                         relay_switch_subbuf(chan->buf[i], 0);
891         mutex_unlock(&relay_channels_mutex);
892 }
893 EXPORT_SYMBOL_GPL(relay_flush);
894
895 /**
896  *      relay_file_open - open file op for relay files
897  *      @inode: the inode
898  *      @filp: the file
899  *
900  *      Increments the channel buffer refcount.
901  */
902 static int relay_file_open(struct inode *inode, struct file *filp)
903 {
904         struct rchan_buf *buf = inode->i_private;
905         kref_get(&buf->kref);
906         filp->private_data = buf;
907
908         return nonseekable_open(inode, filp);
909 }
910
911 /**
912  *      relay_file_mmap - mmap file op for relay files
913  *      @filp: the file
914  *      @vma: the vma describing what to map
915  *
916  *      Calls upon relay_mmap_buf() to map the file into user space.
917  */
918 static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
919 {
920         struct rchan_buf *buf = filp->private_data;
921         return relay_mmap_buf(buf, vma);
922 }
923
924 /**
925  *      relay_file_poll - poll file op for relay files
926  *      @filp: the file
927  *      @wait: poll table
928  *
929  *      Poll implemention.
930  */
931 static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
932 {
933         unsigned int mask = 0;
934         struct rchan_buf *buf = filp->private_data;
935
936         if (buf->finalized)
937                 return POLLERR;
938
939         if (filp->f_mode & FMODE_READ) {
940                 poll_wait(filp, &buf->read_wait, wait);
941                 if (!relay_buf_empty(buf))
942                         mask |= POLLIN | POLLRDNORM;
943         }
944
945         return mask;
946 }
947
948 /**
949  *      relay_file_release - release file op for relay files
950  *      @inode: the inode
951  *      @filp: the file
952  *
953  *      Decrements the channel refcount, as the filesystem is
954  *      no longer using it.
955  */
956 static int relay_file_release(struct inode *inode, struct file *filp)
957 {
958         struct rchan_buf *buf = filp->private_data;
959         kref_put(&buf->kref, relay_remove_buf);
960
961         return 0;
962 }
963
964 /*
965  *      relay_file_read_consume - update the consumed count for the buffer
966  */
967 static void relay_file_read_consume(struct rchan_buf *buf,
968                                     size_t read_pos,
969                                     size_t bytes_consumed)
970 {
971         size_t subbuf_size = buf->chan->subbuf_size;
972         size_t n_subbufs = buf->chan->n_subbufs;
973         size_t read_subbuf;
974
975         if (buf->subbufs_produced == buf->subbufs_consumed &&
976             buf->offset == buf->bytes_consumed)
977                 return;
978
979         if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
980                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
981                 buf->bytes_consumed = 0;
982         }
983
984         buf->bytes_consumed += bytes_consumed;
985         if (!read_pos)
986                 read_subbuf = buf->subbufs_consumed % n_subbufs;
987         else
988                 read_subbuf = read_pos / buf->chan->subbuf_size;
989         if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
990                 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
991                     (buf->offset == subbuf_size))
992                         return;
993                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
994                 buf->bytes_consumed = 0;
995         }
996 }
997
998 /*
999  *      relay_file_read_avail - boolean, are there unconsumed bytes available?
1000  */
1001 static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
1002 {
1003         size_t subbuf_size = buf->chan->subbuf_size;
1004         size_t n_subbufs = buf->chan->n_subbufs;
1005         size_t produced = buf->subbufs_produced;
1006         size_t consumed = buf->subbufs_consumed;
1007
1008         relay_file_read_consume(buf, read_pos, 0);
1009
1010         consumed = buf->subbufs_consumed;
1011
1012         if (unlikely(buf->offset > subbuf_size)) {
1013                 if (produced == consumed)
1014                         return 0;
1015                 return 1;
1016         }
1017
1018         if (unlikely(produced - consumed >= n_subbufs)) {
1019                 consumed = produced - n_subbufs + 1;
1020                 buf->subbufs_consumed = consumed;
1021                 buf->bytes_consumed = 0;
1022         }
1023
1024         produced = (produced % n_subbufs) * subbuf_size + buf->offset;
1025         consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
1026
1027         if (consumed > produced)
1028                 produced += n_subbufs * subbuf_size;
1029
1030         if (consumed == produced) {
1031                 if (buf->offset == subbuf_size &&
1032                     buf->subbufs_produced > buf->subbufs_consumed)
1033                         return 1;
1034                 return 0;
1035         }
1036
1037         return 1;
1038 }
1039
1040 /**
1041  *      relay_file_read_subbuf_avail - return bytes available in sub-buffer
1042  *      @read_pos: file read position
1043  *      @buf: relay channel buffer
1044  */
1045 static size_t relay_file_read_subbuf_avail(size_t read_pos,
1046                                            struct rchan_buf *buf)
1047 {
1048         size_t padding, avail = 0;
1049         size_t read_subbuf, read_offset, write_subbuf, write_offset;
1050         size_t subbuf_size = buf->chan->subbuf_size;
1051
1052         write_subbuf = (buf->data - buf->start) / subbuf_size;
1053         write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
1054         read_subbuf = read_pos / subbuf_size;
1055         read_offset = read_pos % subbuf_size;
1056         padding = buf->padding[read_subbuf];
1057
1058         if (read_subbuf == write_subbuf) {
1059                 if (read_offset + padding < write_offset)
1060                         avail = write_offset - (read_offset + padding);
1061         } else
1062                 avail = (subbuf_size - padding) - read_offset;
1063
1064         return avail;
1065 }
1066
1067 /**
1068  *      relay_file_read_start_pos - find the first available byte to read
1069  *      @read_pos: file read position
1070  *      @buf: relay channel buffer
1071  *
1072  *      If the @read_pos is in the middle of padding, return the
1073  *      position of the first actually available byte, otherwise
1074  *      return the original value.
1075  */
1076 static size_t relay_file_read_start_pos(size_t read_pos,
1077                                         struct rchan_buf *buf)
1078 {
1079         size_t read_subbuf, padding, padding_start, padding_end;
1080         size_t subbuf_size = buf->chan->subbuf_size;
1081         size_t n_subbufs = buf->chan->n_subbufs;
1082         size_t consumed = buf->subbufs_consumed % n_subbufs;
1083
1084         if (!read_pos)
1085                 read_pos = consumed * subbuf_size + buf->bytes_consumed;
1086         read_subbuf = read_pos / subbuf_size;
1087         padding = buf->padding[read_subbuf];
1088         padding_start = (read_subbuf + 1) * subbuf_size - padding;
1089         padding_end = (read_subbuf + 1) * subbuf_size;
1090         if (read_pos >= padding_start && read_pos < padding_end) {
1091                 read_subbuf = (read_subbuf + 1) % n_subbufs;
1092                 read_pos = read_subbuf * subbuf_size;
1093         }
1094
1095         return read_pos;
1096 }
1097
1098 /**
1099  *      relay_file_read_end_pos - return the new read position
1100  *      @read_pos: file read position
1101  *      @buf: relay channel buffer
1102  *      @count: number of bytes to be read
1103  */
1104 static size_t relay_file_read_end_pos(struct rchan_buf *buf,
1105                                       size_t read_pos,
1106                                       size_t count)
1107 {
1108         size_t read_subbuf, padding, end_pos;
1109         size_t subbuf_size = buf->chan->subbuf_size;
1110         size_t n_subbufs = buf->chan->n_subbufs;
1111
1112         read_subbuf = read_pos / subbuf_size;
1113         padding = buf->padding[read_subbuf];
1114         if (read_pos % subbuf_size + count + padding == subbuf_size)
1115                 end_pos = (read_subbuf + 1) * subbuf_size;
1116         else
1117                 end_pos = read_pos + count;
1118         if (end_pos >= subbuf_size * n_subbufs)
1119                 end_pos = 0;
1120
1121         return end_pos;
1122 }
1123
1124 /*
1125  *      subbuf_read_actor - read up to one subbuf's worth of data
1126  */
1127 static int subbuf_read_actor(size_t read_start,
1128                              struct rchan_buf *buf,
1129                              size_t avail,
1130                              read_descriptor_t *desc)
1131 {
1132         void *from;
1133         int ret = 0;
1134
1135         from = buf->start + read_start;
1136         ret = avail;
1137         if (copy_to_user(desc->arg.buf, from, avail)) {
1138                 desc->error = -EFAULT;
1139                 ret = 0;
1140         }
1141         desc->arg.data += ret;
1142         desc->written += ret;
1143         desc->count -= ret;
1144
1145         return ret;
1146 }
1147
1148 typedef int (*subbuf_actor_t) (size_t read_start,
1149                                struct rchan_buf *buf,
1150                                size_t avail,
1151                                read_descriptor_t *desc);
1152
1153 /*
1154  *      relay_file_read_subbufs - read count bytes, bridging subbuf boundaries
1155  */
1156 static ssize_t relay_file_read_subbufs(struct file *filp, loff_t *ppos,
1157                                         subbuf_actor_t subbuf_actor,
1158                                         read_descriptor_t *desc)
1159 {
1160         struct rchan_buf *buf = filp->private_data;
1161         size_t read_start, avail;
1162         int ret;
1163
1164         if (!desc->count)
1165                 return 0;
1166
1167         inode_lock(file_inode(filp));
1168         do {
1169                 if (!relay_file_read_avail(buf, *ppos))
1170                         break;
1171
1172                 read_start = relay_file_read_start_pos(*ppos, buf);
1173                 avail = relay_file_read_subbuf_avail(read_start, buf);
1174                 if (!avail)
1175                         break;
1176
1177                 avail = min(desc->count, avail);
1178                 ret = subbuf_actor(read_start, buf, avail, desc);
1179                 if (desc->error < 0)
1180                         break;
1181
1182                 if (ret) {
1183                         relay_file_read_consume(buf, read_start, ret);
1184                         *ppos = relay_file_read_end_pos(buf, read_start, ret);
1185                 }
1186         } while (desc->count && ret);
1187         inode_unlock(file_inode(filp));
1188
1189         return desc->written;
1190 }
1191
1192 static ssize_t relay_file_read(struct file *filp,
1193                                char __user *buffer,
1194                                size_t count,
1195                                loff_t *ppos)
1196 {
1197         read_descriptor_t desc;
1198         desc.written = 0;
1199         desc.count = count;
1200         desc.arg.buf = buffer;
1201         desc.error = 0;
1202         return relay_file_read_subbufs(filp, ppos, subbuf_read_actor, &desc);
1203 }
1204
1205 static void relay_consume_bytes(struct rchan_buf *rbuf, int bytes_consumed)
1206 {
1207         rbuf->bytes_consumed += bytes_consumed;
1208
1209         if (rbuf->bytes_consumed >= rbuf->chan->subbuf_size) {
1210                 relay_subbufs_consumed(rbuf->chan, rbuf->cpu, 1);
1211                 rbuf->bytes_consumed %= rbuf->chan->subbuf_size;
1212         }
1213 }
1214
1215 static void relay_pipe_buf_release(struct pipe_inode_info *pipe,
1216                                    struct pipe_buffer *buf)
1217 {
1218         struct rchan_buf *rbuf;
1219
1220         rbuf = (struct rchan_buf *)page_private(buf->page);
1221         relay_consume_bytes(rbuf, buf->private);
1222 }
1223
1224 static const struct pipe_buf_operations relay_pipe_buf_ops = {
1225         .can_merge = 0,
1226         .confirm = generic_pipe_buf_confirm,
1227         .release = relay_pipe_buf_release,
1228         .steal = generic_pipe_buf_steal,
1229         .get = generic_pipe_buf_get,
1230 };
1231
1232 static void relay_page_release(struct splice_pipe_desc *spd, unsigned int i)
1233 {
1234 }
1235
1236 /*
1237  *      subbuf_splice_actor - splice up to one subbuf's worth of data
1238  */
1239 static ssize_t subbuf_splice_actor(struct file *in,
1240                                loff_t *ppos,
1241                                struct pipe_inode_info *pipe,
1242                                size_t len,
1243                                unsigned int flags,
1244                                int *nonpad_ret)
1245 {
1246         unsigned int pidx, poff, total_len, subbuf_pages, nr_pages;
1247         struct rchan_buf *rbuf = in->private_data;
1248         unsigned int subbuf_size = rbuf->chan->subbuf_size;
1249         uint64_t pos = (uint64_t) *ppos;
1250         uint32_t alloc_size = (uint32_t) rbuf->chan->alloc_size;
1251         size_t read_start = (size_t) do_div(pos, alloc_size);
1252         size_t read_subbuf = read_start / subbuf_size;
1253         size_t padding = rbuf->padding[read_subbuf];
1254         size_t nonpad_end = read_subbuf * subbuf_size + subbuf_size - padding;
1255         struct page *pages[PIPE_DEF_BUFFERS];
1256         struct partial_page partial[PIPE_DEF_BUFFERS];
1257         struct splice_pipe_desc spd = {
1258                 .pages = pages,
1259                 .nr_pages = 0,
1260                 .nr_pages_max = PIPE_DEF_BUFFERS,
1261                 .partial = partial,
1262                 .flags = flags,
1263                 .ops = &relay_pipe_buf_ops,
1264                 .spd_release = relay_page_release,
1265         };
1266         ssize_t ret;
1267
1268         if (rbuf->subbufs_produced == rbuf->subbufs_consumed)
1269                 return 0;
1270         if (splice_grow_spd(pipe, &spd))
1271                 return -ENOMEM;
1272
1273         /*
1274          * Adjust read len, if longer than what is available
1275          */
1276         if (len > (subbuf_size - read_start % subbuf_size))
1277                 len = subbuf_size - read_start % subbuf_size;
1278
1279         subbuf_pages = rbuf->chan->alloc_size >> PAGE_SHIFT;
1280         pidx = (read_start / PAGE_SIZE) % subbuf_pages;
1281         poff = read_start & ~PAGE_MASK;
1282         nr_pages = min_t(unsigned int, subbuf_pages, spd.nr_pages_max);
1283
1284         for (total_len = 0; spd.nr_pages < nr_pages; spd.nr_pages++) {
1285                 unsigned int this_len, this_end, private;
1286                 unsigned int cur_pos = read_start + total_len;
1287
1288                 if (!len)
1289                         break;
1290
1291                 this_len = min_t(unsigned long, len, PAGE_SIZE - poff);
1292                 private = this_len;
1293
1294                 spd.pages[spd.nr_pages] = rbuf->page_array[pidx];
1295                 spd.partial[spd.nr_pages].offset = poff;
1296
1297                 this_end = cur_pos + this_len;
1298                 if (this_end >= nonpad_end) {
1299                         this_len = nonpad_end - cur_pos;
1300                         private = this_len + padding;
1301                 }
1302                 spd.partial[spd.nr_pages].len = this_len;
1303                 spd.partial[spd.nr_pages].private = private;
1304
1305                 len -= this_len;
1306                 total_len += this_len;
1307                 poff = 0;
1308                 pidx = (pidx + 1) % subbuf_pages;
1309
1310                 if (this_end >= nonpad_end) {
1311                         spd.nr_pages++;
1312                         break;
1313                 }
1314         }
1315
1316         ret = 0;
1317         if (!spd.nr_pages)
1318                 goto out;
1319
1320         ret = *nonpad_ret = splice_to_pipe(pipe, &spd);
1321         if (ret < 0 || ret < total_len)
1322                 goto out;
1323
1324         if (read_start + ret == nonpad_end)
1325                 ret += padding;
1326
1327 out:
1328         splice_shrink_spd(&spd);
1329         return ret;
1330 }
1331
1332 static ssize_t relay_file_splice_read(struct file *in,
1333                                       loff_t *ppos,
1334                                       struct pipe_inode_info *pipe,
1335                                       size_t len,
1336                                       unsigned int flags)
1337 {
1338         ssize_t spliced;
1339         int ret;
1340         int nonpad_ret = 0;
1341
1342         ret = 0;
1343         spliced = 0;
1344
1345         while (len && !spliced) {
1346                 ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret);
1347                 if (ret < 0)
1348                         break;
1349                 else if (!ret) {
1350                         if (flags & SPLICE_F_NONBLOCK)
1351                                 ret = -EAGAIN;
1352                         break;
1353                 }
1354
1355                 *ppos += ret;
1356                 if (ret > len)
1357                         len = 0;
1358                 else
1359                         len -= ret;
1360                 spliced += nonpad_ret;
1361                 nonpad_ret = 0;
1362         }
1363
1364         if (spliced)
1365                 return spliced;
1366
1367         return ret;
1368 }
1369
1370 const struct file_operations relay_file_operations = {
1371         .open           = relay_file_open,
1372         .poll           = relay_file_poll,
1373         .mmap           = relay_file_mmap,
1374         .read           = relay_file_read,
1375         .llseek         = no_llseek,
1376         .release        = relay_file_release,
1377         .splice_read    = relay_file_splice_read,
1378 };
1379 EXPORT_SYMBOL_GPL(relay_file_operations);
1380
1381 static __init int relay_init(void)
1382 {
1383
1384         hotcpu_notifier(relay_hotcpu_callback, 0);
1385         return 0;
1386 }
1387
1388 early_initcall(relay_init);