GNU Linux-libre 4.9.337-gnu1
[releases.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 > KMALLOC_MAX_SIZE / 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         free_percpu(chan->buf);
200         kfree(chan);
201 }
202
203 /**
204  *      relay_destroy_buf - destroy an rchan_buf struct and associated buffer
205  *      @buf: the buffer struct
206  */
207 static void relay_destroy_buf(struct rchan_buf *buf)
208 {
209         struct rchan *chan = buf->chan;
210         unsigned int i;
211
212         if (likely(buf->start)) {
213                 vunmap(buf->start);
214                 for (i = 0; i < buf->page_count; i++)
215                         __free_page(buf->page_array[i]);
216                 relay_free_page_array(buf->page_array);
217         }
218         *per_cpu_ptr(chan->buf, buf->cpu) = NULL;
219         kfree(buf->padding);
220         kfree(buf);
221         kref_put(&chan->kref, relay_destroy_channel);
222 }
223
224 /**
225  *      relay_remove_buf - remove a channel buffer
226  *      @kref: target kernel reference that contains the relay buffer
227  *
228  *      Removes the file from the filesystem, which also frees the
229  *      rchan_buf_struct and the channel buffer.  Should only be called from
230  *      kref_put().
231  */
232 static void relay_remove_buf(struct kref *kref)
233 {
234         struct rchan_buf *buf = container_of(kref, struct rchan_buf, kref);
235         relay_destroy_buf(buf);
236 }
237
238 /**
239  *      relay_buf_empty - boolean, is the channel buffer empty?
240  *      @buf: channel buffer
241  *
242  *      Returns 1 if the buffer is empty, 0 otherwise.
243  */
244 static int relay_buf_empty(struct rchan_buf *buf)
245 {
246         return (buf->subbufs_produced - buf->subbufs_consumed) ? 0 : 1;
247 }
248
249 /**
250  *      relay_buf_full - boolean, is the channel buffer full?
251  *      @buf: channel buffer
252  *
253  *      Returns 1 if the buffer is full, 0 otherwise.
254  */
255 int relay_buf_full(struct rchan_buf *buf)
256 {
257         size_t ready = buf->subbufs_produced - buf->subbufs_consumed;
258         return (ready >= buf->chan->n_subbufs) ? 1 : 0;
259 }
260 EXPORT_SYMBOL_GPL(relay_buf_full);
261
262 /*
263  * High-level relay kernel API and associated functions.
264  */
265
266 /*
267  * rchan_callback implementations defining default channel behavior.  Used
268  * in place of corresponding NULL values in client callback struct.
269  */
270
271 /*
272  * subbuf_start() default callback.  Does nothing.
273  */
274 static int subbuf_start_default_callback (struct rchan_buf *buf,
275                                           void *subbuf,
276                                           void *prev_subbuf,
277                                           size_t prev_padding)
278 {
279         if (relay_buf_full(buf))
280                 return 0;
281
282         return 1;
283 }
284
285 /*
286  * buf_mapped() default callback.  Does nothing.
287  */
288 static void buf_mapped_default_callback(struct rchan_buf *buf,
289                                         struct file *filp)
290 {
291 }
292
293 /*
294  * buf_unmapped() default callback.  Does nothing.
295  */
296 static void buf_unmapped_default_callback(struct rchan_buf *buf,
297                                           struct file *filp)
298 {
299 }
300
301 /*
302  * create_buf_file_create() default callback.  Does nothing.
303  */
304 static struct dentry *create_buf_file_default_callback(const char *filename,
305                                                        struct dentry *parent,
306                                                        umode_t mode,
307                                                        struct rchan_buf *buf,
308                                                        int *is_global)
309 {
310         return NULL;
311 }
312
313 /*
314  * remove_buf_file() default callback.  Does nothing.
315  */
316 static int remove_buf_file_default_callback(struct dentry *dentry)
317 {
318         return -EINVAL;
319 }
320
321 /* relay channel default callbacks */
322 static struct rchan_callbacks default_channel_callbacks = {
323         .subbuf_start = subbuf_start_default_callback,
324         .buf_mapped = buf_mapped_default_callback,
325         .buf_unmapped = buf_unmapped_default_callback,
326         .create_buf_file = create_buf_file_default_callback,
327         .remove_buf_file = remove_buf_file_default_callback,
328 };
329
330 /**
331  *      wakeup_readers - wake up readers waiting on a channel
332  *      @work: contains the channel buffer
333  *
334  *      This is the function used to defer reader waking
335  */
336 static void wakeup_readers(struct irq_work *work)
337 {
338         struct rchan_buf *buf;
339
340         buf = container_of(work, struct rchan_buf, wakeup_work);
341         wake_up_interruptible(&buf->read_wait);
342 }
343
344 /**
345  *      __relay_reset - reset a channel buffer
346  *      @buf: the channel buffer
347  *      @init: 1 if this is a first-time initialization
348  *
349  *      See relay_reset() for description of effect.
350  */
351 static void __relay_reset(struct rchan_buf *buf, unsigned int init)
352 {
353         size_t i;
354
355         if (init) {
356                 init_waitqueue_head(&buf->read_wait);
357                 kref_init(&buf->kref);
358                 init_irq_work(&buf->wakeup_work, wakeup_readers);
359         } else {
360                 irq_work_sync(&buf->wakeup_work);
361         }
362
363         buf->subbufs_produced = 0;
364         buf->subbufs_consumed = 0;
365         buf->bytes_consumed = 0;
366         buf->finalized = 0;
367         buf->data = buf->start;
368         buf->offset = 0;
369
370         for (i = 0; i < buf->chan->n_subbufs; i++)
371                 buf->padding[i] = 0;
372
373         buf->chan->cb->subbuf_start(buf, buf->data, NULL, 0);
374 }
375
376 /**
377  *      relay_reset - reset the channel
378  *      @chan: the channel
379  *
380  *      This has the effect of erasing all data from all channel buffers
381  *      and restarting the channel in its initial state.  The buffers
382  *      are not freed, so any mappings are still in effect.
383  *
384  *      NOTE. Care should be taken that the channel isn't actually
385  *      being used by anything when this call is made.
386  */
387 void relay_reset(struct rchan *chan)
388 {
389         struct rchan_buf *buf;
390         unsigned int i;
391
392         if (!chan)
393                 return;
394
395         if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0))) {
396                 __relay_reset(buf, 0);
397                 return;
398         }
399
400         mutex_lock(&relay_channels_mutex);
401         for_each_possible_cpu(i)
402                 if ((buf = *per_cpu_ptr(chan->buf, i)))
403                         __relay_reset(buf, 0);
404         mutex_unlock(&relay_channels_mutex);
405 }
406 EXPORT_SYMBOL_GPL(relay_reset);
407
408 static inline void relay_set_buf_dentry(struct rchan_buf *buf,
409                                         struct dentry *dentry)
410 {
411         buf->dentry = dentry;
412         d_inode(buf->dentry)->i_size = buf->early_bytes;
413 }
414
415 static struct dentry *relay_create_buf_file(struct rchan *chan,
416                                             struct rchan_buf *buf,
417                                             unsigned int cpu)
418 {
419         struct dentry *dentry;
420         char *tmpname;
421
422         tmpname = kzalloc(NAME_MAX + 1, GFP_KERNEL);
423         if (!tmpname)
424                 return NULL;
425         snprintf(tmpname, NAME_MAX, "%s%d", chan->base_filename, cpu);
426
427         /* Create file in fs */
428         dentry = chan->cb->create_buf_file(tmpname, chan->parent,
429                                            S_IRUSR, buf,
430                                            &chan->is_global);
431
432         kfree(tmpname);
433
434         return dentry;
435 }
436
437 /*
438  *      relay_open_buf - create a new relay channel buffer
439  *
440  *      used by relay_open() and CPU hotplug.
441  */
442 static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu)
443 {
444         struct rchan_buf *buf = NULL;
445         struct dentry *dentry;
446
447         if (chan->is_global)
448                 return *per_cpu_ptr(chan->buf, 0);
449
450         buf = relay_create_buf(chan);
451         if (!buf)
452                 return NULL;
453
454         if (chan->has_base_filename) {
455                 dentry = relay_create_buf_file(chan, buf, cpu);
456                 if (!dentry)
457                         goto free_buf;
458                 relay_set_buf_dentry(buf, dentry);
459         } else {
460                 /* Only retrieve global info, nothing more, nothing less */
461                 dentry = chan->cb->create_buf_file(NULL, NULL,
462                                                    S_IRUSR, buf,
463                                                    &chan->is_global);
464                 if (WARN_ON(dentry))
465                         goto free_buf;
466         }
467
468         buf->cpu = cpu;
469         __relay_reset(buf, 1);
470
471         if(chan->is_global) {
472                 *per_cpu_ptr(chan->buf, 0) = buf;
473                 buf->cpu = 0;
474         }
475
476         return buf;
477
478 free_buf:
479         relay_destroy_buf(buf);
480         return NULL;
481 }
482
483 /**
484  *      relay_close_buf - close a channel buffer
485  *      @buf: channel buffer
486  *
487  *      Marks the buffer finalized and restores the default callbacks.
488  *      The channel buffer and channel buffer data structure are then freed
489  *      automatically when the last reference is given up.
490  */
491 static void relay_close_buf(struct rchan_buf *buf)
492 {
493         buf->finalized = 1;
494         irq_work_sync(&buf->wakeup_work);
495         buf->chan->cb->remove_buf_file(buf->dentry);
496         kref_put(&buf->kref, relay_remove_buf);
497 }
498
499 static void setup_callbacks(struct rchan *chan,
500                                    struct rchan_callbacks *cb)
501 {
502         if (!cb) {
503                 chan->cb = &default_channel_callbacks;
504                 return;
505         }
506
507         if (!cb->subbuf_start)
508                 cb->subbuf_start = subbuf_start_default_callback;
509         if (!cb->buf_mapped)
510                 cb->buf_mapped = buf_mapped_default_callback;
511         if (!cb->buf_unmapped)
512                 cb->buf_unmapped = buf_unmapped_default_callback;
513         if (!cb->create_buf_file)
514                 cb->create_buf_file = create_buf_file_default_callback;
515         if (!cb->remove_buf_file)
516                 cb->remove_buf_file = remove_buf_file_default_callback;
517         chan->cb = cb;
518 }
519
520 int relay_prepare_cpu(unsigned int cpu)
521 {
522         struct rchan *chan;
523         struct rchan_buf *buf;
524
525         mutex_lock(&relay_channels_mutex);
526         list_for_each_entry(chan, &relay_channels, list) {
527                 if ((buf = *per_cpu_ptr(chan->buf, cpu)))
528                         continue;
529                 buf = relay_open_buf(chan, cpu);
530                 if (!buf) {
531                         pr_err("relay: cpu %d buffer creation failed\n", cpu);
532                         mutex_unlock(&relay_channels_mutex);
533                         return -ENOMEM;
534                 }
535                 *per_cpu_ptr(chan->buf, cpu) = buf;
536         }
537         mutex_unlock(&relay_channels_mutex);
538         return 0;
539 }
540
541 /**
542  *      relay_open - create a new relay channel
543  *      @base_filename: base name of files to create, %NULL for buffering only
544  *      @parent: dentry of parent directory, %NULL for root directory or buffer
545  *      @subbuf_size: size of sub-buffers
546  *      @n_subbufs: number of sub-buffers
547  *      @cb: client callback functions
548  *      @private_data: user-defined data
549  *
550  *      Returns channel pointer if successful, %NULL otherwise.
551  *
552  *      Creates a channel buffer for each cpu using the sizes and
553  *      attributes specified.  The created channel buffer files
554  *      will be named base_filename0...base_filenameN-1.  File
555  *      permissions will be %S_IRUSR.
556  *
557  *      If opening a buffer (@parent = NULL) that you later wish to register
558  *      in a filesystem, call relay_late_setup_files() once the @parent dentry
559  *      is available.
560  */
561 struct rchan *relay_open(const char *base_filename,
562                          struct dentry *parent,
563                          size_t subbuf_size,
564                          size_t n_subbufs,
565                          struct rchan_callbacks *cb,
566                          void *private_data)
567 {
568         unsigned int i;
569         struct rchan *chan;
570         struct rchan_buf *buf;
571
572         if (!(subbuf_size && n_subbufs))
573                 return NULL;
574         if (subbuf_size > UINT_MAX / n_subbufs)
575                 return NULL;
576
577         chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
578         if (!chan)
579                 return NULL;
580
581         chan->buf = alloc_percpu(struct rchan_buf *);
582         if (!chan->buf) {
583                 kfree(chan);
584                 return NULL;
585         }
586
587         chan->version = RELAYFS_CHANNEL_VERSION;
588         chan->n_subbufs = n_subbufs;
589         chan->subbuf_size = subbuf_size;
590         chan->alloc_size = PAGE_ALIGN(subbuf_size * n_subbufs);
591         chan->parent = parent;
592         chan->private_data = private_data;
593         if (base_filename) {
594                 chan->has_base_filename = 1;
595                 strlcpy(chan->base_filename, base_filename, NAME_MAX);
596         }
597         setup_callbacks(chan, cb);
598         kref_init(&chan->kref);
599
600         mutex_lock(&relay_channels_mutex);
601         for_each_online_cpu(i) {
602                 buf = relay_open_buf(chan, i);
603                 if (!buf)
604                         goto free_bufs;
605                 *per_cpu_ptr(chan->buf, i) = buf;
606         }
607         list_add(&chan->list, &relay_channels);
608         mutex_unlock(&relay_channels_mutex);
609
610         return chan;
611
612 free_bufs:
613         for_each_possible_cpu(i) {
614                 if ((buf = *per_cpu_ptr(chan->buf, i)))
615                         relay_close_buf(buf);
616         }
617
618         kref_put(&chan->kref, relay_destroy_channel);
619         mutex_unlock(&relay_channels_mutex);
620         return NULL;
621 }
622 EXPORT_SYMBOL_GPL(relay_open);
623
624 struct rchan_percpu_buf_dispatcher {
625         struct rchan_buf *buf;
626         struct dentry *dentry;
627 };
628
629 /* Called in atomic context. */
630 static void __relay_set_buf_dentry(void *info)
631 {
632         struct rchan_percpu_buf_dispatcher *p = info;
633
634         relay_set_buf_dentry(p->buf, p->dentry);
635 }
636
637 /**
638  *      relay_late_setup_files - triggers file creation
639  *      @chan: channel to operate on
640  *      @base_filename: base name of files to create
641  *      @parent: dentry of parent directory, %NULL for root directory
642  *
643  *      Returns 0 if successful, non-zero otherwise.
644  *
645  *      Use to setup files for a previously buffer-only channel created
646  *      by relay_open() with a NULL parent dentry.
647  *
648  *      For example, this is useful for perfomring early tracing in kernel,
649  *      before VFS is up and then exposing the early results once the dentry
650  *      is available.
651  */
652 int relay_late_setup_files(struct rchan *chan,
653                            const char *base_filename,
654                            struct dentry *parent)
655 {
656         int err = 0;
657         unsigned int i, curr_cpu;
658         unsigned long flags;
659         struct dentry *dentry;
660         struct rchan_buf *buf;
661         struct rchan_percpu_buf_dispatcher disp;
662
663         if (!chan || !base_filename)
664                 return -EINVAL;
665
666         strlcpy(chan->base_filename, base_filename, NAME_MAX);
667
668         mutex_lock(&relay_channels_mutex);
669         /* Is chan already set up? */
670         if (unlikely(chan->has_base_filename)) {
671                 mutex_unlock(&relay_channels_mutex);
672                 return -EEXIST;
673         }
674         chan->has_base_filename = 1;
675         chan->parent = parent;
676
677         if (chan->is_global) {
678                 err = -EINVAL;
679                 buf = *per_cpu_ptr(chan->buf, 0);
680                 if (!WARN_ON_ONCE(!buf)) {
681                         dentry = relay_create_buf_file(chan, buf, 0);
682                         if (dentry && !WARN_ON_ONCE(!chan->is_global)) {
683                                 relay_set_buf_dentry(buf, dentry);
684                                 err = 0;
685                         }
686                 }
687                 mutex_unlock(&relay_channels_mutex);
688                 return err;
689         }
690
691         curr_cpu = get_cpu();
692         /*
693          * The CPU hotplug notifier ran before us and created buffers with
694          * no files associated. So it's safe to call relay_setup_buf_file()
695          * on all currently online CPUs.
696          */
697         for_each_online_cpu(i) {
698                 buf = *per_cpu_ptr(chan->buf, i);
699                 if (unlikely(!buf)) {
700                         WARN_ONCE(1, KERN_ERR "CPU has no buffer!\n");
701                         err = -EINVAL;
702                         break;
703                 }
704
705                 dentry = relay_create_buf_file(chan, buf, i);
706                 if (unlikely(!dentry)) {
707                         err = -EINVAL;
708                         break;
709                 }
710
711                 if (curr_cpu == i) {
712                         local_irq_save(flags);
713                         relay_set_buf_dentry(buf, dentry);
714                         local_irq_restore(flags);
715                 } else {
716                         disp.buf = buf;
717                         disp.dentry = dentry;
718                         smp_mb();
719                         /* relay_channels_mutex must be held, so wait. */
720                         err = smp_call_function_single(i,
721                                                        __relay_set_buf_dentry,
722                                                        &disp, 1);
723                 }
724                 if (unlikely(err))
725                         break;
726         }
727         put_cpu();
728         mutex_unlock(&relay_channels_mutex);
729
730         return err;
731 }
732 EXPORT_SYMBOL_GPL(relay_late_setup_files);
733
734 /**
735  *      relay_switch_subbuf - switch to a new sub-buffer
736  *      @buf: channel buffer
737  *      @length: size of current event
738  *
739  *      Returns either the length passed in or 0 if full.
740  *
741  *      Performs sub-buffer-switch tasks such as invoking callbacks,
742  *      updating padding counts, waking up readers, etc.
743  */
744 size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
745 {
746         void *old, *new;
747         size_t old_subbuf, new_subbuf;
748
749         if (unlikely(length > buf->chan->subbuf_size))
750                 goto toobig;
751
752         if (buf->offset != buf->chan->subbuf_size + 1) {
753                 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
754                 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
755                 buf->padding[old_subbuf] = buf->prev_padding;
756                 buf->subbufs_produced++;
757                 if (buf->dentry)
758                         d_inode(buf->dentry)->i_size +=
759                                 buf->chan->subbuf_size -
760                                 buf->padding[old_subbuf];
761                 else
762                         buf->early_bytes += buf->chan->subbuf_size -
763                                             buf->padding[old_subbuf];
764                 smp_mb();
765                 if (waitqueue_active(&buf->read_wait)) {
766                         /*
767                          * Calling wake_up_interruptible() from here
768                          * will deadlock if we happen to be logging
769                          * from the scheduler (trying to re-grab
770                          * rq->lock), so defer it.
771                          */
772                         irq_work_queue(&buf->wakeup_work);
773                 }
774         }
775
776         old = buf->data;
777         new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
778         new = buf->start + new_subbuf * buf->chan->subbuf_size;
779         buf->offset = 0;
780         if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
781                 buf->offset = buf->chan->subbuf_size + 1;
782                 return 0;
783         }
784         buf->data = new;
785         buf->padding[new_subbuf] = 0;
786
787         if (unlikely(length + buf->offset > buf->chan->subbuf_size))
788                 goto toobig;
789
790         return length;
791
792 toobig:
793         buf->chan->last_toobig = length;
794         return 0;
795 }
796 EXPORT_SYMBOL_GPL(relay_switch_subbuf);
797
798 /**
799  *      relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
800  *      @chan: the channel
801  *      @cpu: the cpu associated with the channel buffer to update
802  *      @subbufs_consumed: number of sub-buffers to add to current buf's count
803  *
804  *      Adds to the channel buffer's consumed sub-buffer count.
805  *      subbufs_consumed should be the number of sub-buffers newly consumed,
806  *      not the total consumed.
807  *
808  *      NOTE. Kernel clients don't need to call this function if the channel
809  *      mode is 'overwrite'.
810  */
811 void relay_subbufs_consumed(struct rchan *chan,
812                             unsigned int cpu,
813                             size_t subbufs_consumed)
814 {
815         struct rchan_buf *buf;
816
817         if (!chan || cpu >= NR_CPUS)
818                 return;
819
820         buf = *per_cpu_ptr(chan->buf, cpu);
821         if (!buf || subbufs_consumed > chan->n_subbufs)
822                 return;
823
824         if (subbufs_consumed > buf->subbufs_produced - buf->subbufs_consumed)
825                 buf->subbufs_consumed = buf->subbufs_produced;
826         else
827                 buf->subbufs_consumed += subbufs_consumed;
828 }
829 EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
830
831 /**
832  *      relay_close - close the channel
833  *      @chan: the channel
834  *
835  *      Closes all channel buffers and frees the channel.
836  */
837 void relay_close(struct rchan *chan)
838 {
839         struct rchan_buf *buf;
840         unsigned int i;
841
842         if (!chan)
843                 return;
844
845         mutex_lock(&relay_channels_mutex);
846         if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0)))
847                 relay_close_buf(buf);
848         else
849                 for_each_possible_cpu(i)
850                         if ((buf = *per_cpu_ptr(chan->buf, i)))
851                                 relay_close_buf(buf);
852
853         if (chan->last_toobig)
854                 printk(KERN_WARNING "relay: one or more items not logged "
855                        "[item size (%Zd) > sub-buffer size (%Zd)]\n",
856                        chan->last_toobig, chan->subbuf_size);
857
858         list_del(&chan->list);
859         kref_put(&chan->kref, relay_destroy_channel);
860         mutex_unlock(&relay_channels_mutex);
861 }
862 EXPORT_SYMBOL_GPL(relay_close);
863
864 /**
865  *      relay_flush - close the channel
866  *      @chan: the channel
867  *
868  *      Flushes all channel buffers, i.e. forces buffer switch.
869  */
870 void relay_flush(struct rchan *chan)
871 {
872         struct rchan_buf *buf;
873         unsigned int i;
874
875         if (!chan)
876                 return;
877
878         if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0))) {
879                 relay_switch_subbuf(buf, 0);
880                 return;
881         }
882
883         mutex_lock(&relay_channels_mutex);
884         for_each_possible_cpu(i)
885                 if ((buf = *per_cpu_ptr(chan->buf, i)))
886                         relay_switch_subbuf(buf, 0);
887         mutex_unlock(&relay_channels_mutex);
888 }
889 EXPORT_SYMBOL_GPL(relay_flush);
890
891 /**
892  *      relay_file_open - open file op for relay files
893  *      @inode: the inode
894  *      @filp: the file
895  *
896  *      Increments the channel buffer refcount.
897  */
898 static int relay_file_open(struct inode *inode, struct file *filp)
899 {
900         struct rchan_buf *buf = inode->i_private;
901         kref_get(&buf->kref);
902         filp->private_data = buf;
903
904         return nonseekable_open(inode, filp);
905 }
906
907 /**
908  *      relay_file_mmap - mmap file op for relay files
909  *      @filp: the file
910  *      @vma: the vma describing what to map
911  *
912  *      Calls upon relay_mmap_buf() to map the file into user space.
913  */
914 static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
915 {
916         struct rchan_buf *buf = filp->private_data;
917         return relay_mmap_buf(buf, vma);
918 }
919
920 /**
921  *      relay_file_poll - poll file op for relay files
922  *      @filp: the file
923  *      @wait: poll table
924  *
925  *      Poll implemention.
926  */
927 static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
928 {
929         unsigned int mask = 0;
930         struct rchan_buf *buf = filp->private_data;
931
932         if (buf->finalized)
933                 return POLLERR;
934
935         if (filp->f_mode & FMODE_READ) {
936                 poll_wait(filp, &buf->read_wait, wait);
937                 if (!relay_buf_empty(buf))
938                         mask |= POLLIN | POLLRDNORM;
939         }
940
941         return mask;
942 }
943
944 /**
945  *      relay_file_release - release file op for relay files
946  *      @inode: the inode
947  *      @filp: the file
948  *
949  *      Decrements the channel refcount, as the filesystem is
950  *      no longer using it.
951  */
952 static int relay_file_release(struct inode *inode, struct file *filp)
953 {
954         struct rchan_buf *buf = filp->private_data;
955         kref_put(&buf->kref, relay_remove_buf);
956
957         return 0;
958 }
959
960 /*
961  *      relay_file_read_consume - update the consumed count for the buffer
962  */
963 static void relay_file_read_consume(struct rchan_buf *buf,
964                                     size_t read_pos,
965                                     size_t bytes_consumed)
966 {
967         size_t subbuf_size = buf->chan->subbuf_size;
968         size_t n_subbufs = buf->chan->n_subbufs;
969         size_t read_subbuf;
970
971         if (buf->subbufs_produced == buf->subbufs_consumed &&
972             buf->offset == buf->bytes_consumed)
973                 return;
974
975         if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
976                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
977                 buf->bytes_consumed = 0;
978         }
979
980         buf->bytes_consumed += bytes_consumed;
981         if (!read_pos)
982                 read_subbuf = buf->subbufs_consumed % n_subbufs;
983         else
984                 read_subbuf = read_pos / buf->chan->subbuf_size;
985         if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
986                 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
987                     (buf->offset == subbuf_size))
988                         return;
989                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
990                 buf->bytes_consumed = 0;
991         }
992 }
993
994 /*
995  *      relay_file_read_avail - boolean, are there unconsumed bytes available?
996  */
997 static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
998 {
999         size_t subbuf_size = buf->chan->subbuf_size;
1000         size_t n_subbufs = buf->chan->n_subbufs;
1001         size_t produced = buf->subbufs_produced;
1002         size_t consumed = buf->subbufs_consumed;
1003
1004         relay_file_read_consume(buf, read_pos, 0);
1005
1006         consumed = buf->subbufs_consumed;
1007
1008         if (unlikely(buf->offset > subbuf_size)) {
1009                 if (produced == consumed)
1010                         return 0;
1011                 return 1;
1012         }
1013
1014         if (unlikely(produced - consumed >= n_subbufs)) {
1015                 consumed = produced - n_subbufs + 1;
1016                 buf->subbufs_consumed = consumed;
1017                 buf->bytes_consumed = 0;
1018         }
1019
1020         produced = (produced % n_subbufs) * subbuf_size + buf->offset;
1021         consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
1022
1023         if (consumed > produced)
1024                 produced += n_subbufs * subbuf_size;
1025
1026         if (consumed == produced) {
1027                 if (buf->offset == subbuf_size &&
1028                     buf->subbufs_produced > buf->subbufs_consumed)
1029                         return 1;
1030                 return 0;
1031         }
1032
1033         return 1;
1034 }
1035
1036 /**
1037  *      relay_file_read_subbuf_avail - return bytes available in sub-buffer
1038  *      @read_pos: file read position
1039  *      @buf: relay channel buffer
1040  */
1041 static size_t relay_file_read_subbuf_avail(size_t read_pos,
1042                                            struct rchan_buf *buf)
1043 {
1044         size_t padding, avail = 0;
1045         size_t read_subbuf, read_offset, write_subbuf, write_offset;
1046         size_t subbuf_size = buf->chan->subbuf_size;
1047
1048         write_subbuf = (buf->data - buf->start) / subbuf_size;
1049         write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
1050         read_subbuf = read_pos / subbuf_size;
1051         read_offset = read_pos % subbuf_size;
1052         padding = buf->padding[read_subbuf];
1053
1054         if (read_subbuf == write_subbuf) {
1055                 if (read_offset + padding < write_offset)
1056                         avail = write_offset - (read_offset + padding);
1057         } else
1058                 avail = (subbuf_size - padding) - read_offset;
1059
1060         return avail;
1061 }
1062
1063 /**
1064  *      relay_file_read_start_pos - find the first available byte to read
1065  *      @read_pos: file read position
1066  *      @buf: relay channel buffer
1067  *
1068  *      If the @read_pos is in the middle of padding, return the
1069  *      position of the first actually available byte, otherwise
1070  *      return the original value.
1071  */
1072 static size_t relay_file_read_start_pos(size_t read_pos,
1073                                         struct rchan_buf *buf)
1074 {
1075         size_t read_subbuf, padding, padding_start, padding_end;
1076         size_t subbuf_size = buf->chan->subbuf_size;
1077         size_t n_subbufs = buf->chan->n_subbufs;
1078         size_t consumed = buf->subbufs_consumed % n_subbufs;
1079
1080         if (!read_pos)
1081                 read_pos = consumed * subbuf_size + buf->bytes_consumed;
1082         read_subbuf = read_pos / subbuf_size;
1083         padding = buf->padding[read_subbuf];
1084         padding_start = (read_subbuf + 1) * subbuf_size - padding;
1085         padding_end = (read_subbuf + 1) * subbuf_size;
1086         if (read_pos >= padding_start && read_pos < padding_end) {
1087                 read_subbuf = (read_subbuf + 1) % n_subbufs;
1088                 read_pos = read_subbuf * subbuf_size;
1089         }
1090
1091         return read_pos;
1092 }
1093
1094 /**
1095  *      relay_file_read_end_pos - return the new read position
1096  *      @read_pos: file read position
1097  *      @buf: relay channel buffer
1098  *      @count: number of bytes to be read
1099  */
1100 static size_t relay_file_read_end_pos(struct rchan_buf *buf,
1101                                       size_t read_pos,
1102                                       size_t count)
1103 {
1104         size_t read_subbuf, padding, end_pos;
1105         size_t subbuf_size = buf->chan->subbuf_size;
1106         size_t n_subbufs = buf->chan->n_subbufs;
1107
1108         read_subbuf = read_pos / subbuf_size;
1109         padding = buf->padding[read_subbuf];
1110         if (read_pos % subbuf_size + count + padding == subbuf_size)
1111                 end_pos = (read_subbuf + 1) * subbuf_size;
1112         else
1113                 end_pos = read_pos + count;
1114         if (end_pos >= subbuf_size * n_subbufs)
1115                 end_pos = 0;
1116
1117         return end_pos;
1118 }
1119
1120 static ssize_t relay_file_read(struct file *filp,
1121                                char __user *buffer,
1122                                size_t count,
1123                                loff_t *ppos)
1124 {
1125         struct rchan_buf *buf = filp->private_data;
1126         size_t read_start, avail;
1127         size_t written = 0;
1128         int ret;
1129
1130         if (!count)
1131                 return 0;
1132
1133         inode_lock(file_inode(filp));
1134         do {
1135                 void *from;
1136
1137                 if (!relay_file_read_avail(buf, *ppos))
1138                         break;
1139
1140                 read_start = relay_file_read_start_pos(*ppos, buf);
1141                 avail = relay_file_read_subbuf_avail(read_start, buf);
1142                 if (!avail)
1143                         break;
1144
1145                 avail = min(count, avail);
1146                 from = buf->start + read_start;
1147                 ret = avail;
1148                 if (copy_to_user(buffer, from, avail))
1149                         break;
1150
1151                 buffer += ret;
1152                 written += ret;
1153                 count -= ret;
1154
1155                 relay_file_read_consume(buf, read_start, ret);
1156                 *ppos = relay_file_read_end_pos(buf, read_start, ret);
1157         } while (count);
1158         inode_unlock(file_inode(filp));
1159
1160         return written;
1161 }
1162
1163 static void relay_consume_bytes(struct rchan_buf *rbuf, int bytes_consumed)
1164 {
1165         rbuf->bytes_consumed += bytes_consumed;
1166
1167         if (rbuf->bytes_consumed >= rbuf->chan->subbuf_size) {
1168                 relay_subbufs_consumed(rbuf->chan, rbuf->cpu, 1);
1169                 rbuf->bytes_consumed %= rbuf->chan->subbuf_size;
1170         }
1171 }
1172
1173 static void relay_pipe_buf_release(struct pipe_inode_info *pipe,
1174                                    struct pipe_buffer *buf)
1175 {
1176         struct rchan_buf *rbuf;
1177
1178         rbuf = (struct rchan_buf *)page_private(buf->page);
1179         relay_consume_bytes(rbuf, buf->private);
1180 }
1181
1182 static const struct pipe_buf_operations relay_pipe_buf_ops = {
1183         .can_merge = 0,
1184         .confirm = generic_pipe_buf_confirm,
1185         .release = relay_pipe_buf_release,
1186         .steal = generic_pipe_buf_steal,
1187         .get = generic_pipe_buf_get,
1188 };
1189
1190 static void relay_page_release(struct splice_pipe_desc *spd, unsigned int i)
1191 {
1192 }
1193
1194 /*
1195  *      subbuf_splice_actor - splice up to one subbuf's worth of data
1196  */
1197 static ssize_t subbuf_splice_actor(struct file *in,
1198                                loff_t *ppos,
1199                                struct pipe_inode_info *pipe,
1200                                size_t len,
1201                                unsigned int flags,
1202                                int *nonpad_ret)
1203 {
1204         unsigned int pidx, poff, total_len, subbuf_pages, nr_pages;
1205         struct rchan_buf *rbuf = in->private_data;
1206         unsigned int subbuf_size = rbuf->chan->subbuf_size;
1207         uint64_t pos = (uint64_t) *ppos;
1208         uint32_t alloc_size = (uint32_t) rbuf->chan->alloc_size;
1209         size_t read_start = (size_t) do_div(pos, alloc_size);
1210         size_t read_subbuf = read_start / subbuf_size;
1211         size_t padding = rbuf->padding[read_subbuf];
1212         size_t nonpad_end = read_subbuf * subbuf_size + subbuf_size - padding;
1213         struct page *pages[PIPE_DEF_BUFFERS];
1214         struct partial_page partial[PIPE_DEF_BUFFERS];
1215         struct splice_pipe_desc spd = {
1216                 .pages = pages,
1217                 .nr_pages = 0,
1218                 .nr_pages_max = PIPE_DEF_BUFFERS,
1219                 .partial = partial,
1220                 .flags = flags,
1221                 .ops = &relay_pipe_buf_ops,
1222                 .spd_release = relay_page_release,
1223         };
1224         ssize_t ret;
1225
1226         if (rbuf->subbufs_produced == rbuf->subbufs_consumed)
1227                 return 0;
1228         if (splice_grow_spd(pipe, &spd))
1229                 return -ENOMEM;
1230
1231         /*
1232          * Adjust read len, if longer than what is available
1233          */
1234         if (len > (subbuf_size - read_start % subbuf_size))
1235                 len = subbuf_size - read_start % subbuf_size;
1236
1237         subbuf_pages = rbuf->chan->alloc_size >> PAGE_SHIFT;
1238         pidx = (read_start / PAGE_SIZE) % subbuf_pages;
1239         poff = read_start & ~PAGE_MASK;
1240         nr_pages = min_t(unsigned int, subbuf_pages, spd.nr_pages_max);
1241
1242         for (total_len = 0; spd.nr_pages < nr_pages; spd.nr_pages++) {
1243                 unsigned int this_len, this_end, private;
1244                 unsigned int cur_pos = read_start + total_len;
1245
1246                 if (!len)
1247                         break;
1248
1249                 this_len = min_t(unsigned long, len, PAGE_SIZE - poff);
1250                 private = this_len;
1251
1252                 spd.pages[spd.nr_pages] = rbuf->page_array[pidx];
1253                 spd.partial[spd.nr_pages].offset = poff;
1254
1255                 this_end = cur_pos + this_len;
1256                 if (this_end >= nonpad_end) {
1257                         this_len = nonpad_end - cur_pos;
1258                         private = this_len + padding;
1259                 }
1260                 spd.partial[spd.nr_pages].len = this_len;
1261                 spd.partial[spd.nr_pages].private = private;
1262
1263                 len -= this_len;
1264                 total_len += this_len;
1265                 poff = 0;
1266                 pidx = (pidx + 1) % subbuf_pages;
1267
1268                 if (this_end >= nonpad_end) {
1269                         spd.nr_pages++;
1270                         break;
1271                 }
1272         }
1273
1274         ret = 0;
1275         if (!spd.nr_pages)
1276                 goto out;
1277
1278         ret = *nonpad_ret = splice_to_pipe(pipe, &spd);
1279         if (ret < 0 || ret < total_len)
1280                 goto out;
1281
1282         if (read_start + ret == nonpad_end)
1283                 ret += padding;
1284
1285 out:
1286         splice_shrink_spd(&spd);
1287         return ret;
1288 }
1289
1290 static ssize_t relay_file_splice_read(struct file *in,
1291                                       loff_t *ppos,
1292                                       struct pipe_inode_info *pipe,
1293                                       size_t len,
1294                                       unsigned int flags)
1295 {
1296         ssize_t spliced;
1297         int ret;
1298         int nonpad_ret = 0;
1299
1300         ret = 0;
1301         spliced = 0;
1302
1303         while (len && !spliced) {
1304                 ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret);
1305                 if (ret < 0)
1306                         break;
1307                 else if (!ret) {
1308                         if (flags & SPLICE_F_NONBLOCK)
1309                                 ret = -EAGAIN;
1310                         break;
1311                 }
1312
1313                 *ppos += ret;
1314                 if (ret > len)
1315                         len = 0;
1316                 else
1317                         len -= ret;
1318                 spliced += nonpad_ret;
1319                 nonpad_ret = 0;
1320         }
1321
1322         if (spliced)
1323                 return spliced;
1324
1325         return ret;
1326 }
1327
1328 const struct file_operations relay_file_operations = {
1329         .open           = relay_file_open,
1330         .poll           = relay_file_poll,
1331         .mmap           = relay_file_mmap,
1332         .read           = relay_file_read,
1333         .llseek         = no_llseek,
1334         .release        = relay_file_release,
1335         .splice_read    = relay_file_splice_read,
1336 };
1337 EXPORT_SYMBOL_GPL(relay_file_operations);