proc: fix timerslack_ns CAP_SYS_NICE check when adjusting self
[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         *per_cpu_ptr(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         struct rchan_buf *buf;
386         unsigned int i;
387
388         if (!chan)
389                 return;
390
391         if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0))) {
392                 __relay_reset(buf, 0);
393                 return;
394         }
395
396         mutex_lock(&relay_channels_mutex);
397         for_each_possible_cpu(i)
398                 if ((buf = *per_cpu_ptr(chan->buf, i)))
399                         __relay_reset(buf, 0);
400         mutex_unlock(&relay_channels_mutex);
401 }
402 EXPORT_SYMBOL_GPL(relay_reset);
403
404 static inline void relay_set_buf_dentry(struct rchan_buf *buf,
405                                         struct dentry *dentry)
406 {
407         buf->dentry = dentry;
408         d_inode(buf->dentry)->i_size = buf->early_bytes;
409 }
410
411 static struct dentry *relay_create_buf_file(struct rchan *chan,
412                                             struct rchan_buf *buf,
413                                             unsigned int cpu)
414 {
415         struct dentry *dentry;
416         char *tmpname;
417
418         tmpname = kzalloc(NAME_MAX + 1, GFP_KERNEL);
419         if (!tmpname)
420                 return NULL;
421         snprintf(tmpname, NAME_MAX, "%s%d", chan->base_filename, cpu);
422
423         /* Create file in fs */
424         dentry = chan->cb->create_buf_file(tmpname, chan->parent,
425                                            S_IRUSR, buf,
426                                            &chan->is_global);
427
428         kfree(tmpname);
429
430         return dentry;
431 }
432
433 /*
434  *      relay_open_buf - create a new relay channel buffer
435  *
436  *      used by relay_open() and CPU hotplug.
437  */
438 static struct rchan_buf *relay_open_buf(struct rchan *chan, unsigned int cpu)
439 {
440         struct rchan_buf *buf = NULL;
441         struct dentry *dentry;
442
443         if (chan->is_global)
444                 return *per_cpu_ptr(chan->buf, 0);
445
446         buf = relay_create_buf(chan);
447         if (!buf)
448                 return NULL;
449
450         if (chan->has_base_filename) {
451                 dentry = relay_create_buf_file(chan, buf, cpu);
452                 if (!dentry)
453                         goto free_buf;
454                 relay_set_buf_dentry(buf, dentry);
455         } else {
456                 /* Only retrieve global info, nothing more, nothing less */
457                 dentry = chan->cb->create_buf_file(NULL, NULL,
458                                                    S_IRUSR, buf,
459                                                    &chan->is_global);
460                 if (WARN_ON(dentry))
461                         goto free_buf;
462         }
463
464         buf->cpu = cpu;
465         __relay_reset(buf, 1);
466
467         if(chan->is_global) {
468                 *per_cpu_ptr(chan->buf, 0) = buf;
469                 buf->cpu = 0;
470         }
471
472         return buf;
473
474 free_buf:
475         relay_destroy_buf(buf);
476         return NULL;
477 }
478
479 /**
480  *      relay_close_buf - close a channel buffer
481  *      @buf: channel buffer
482  *
483  *      Marks the buffer finalized and restores the default callbacks.
484  *      The channel buffer and channel buffer data structure are then freed
485  *      automatically when the last reference is given up.
486  */
487 static void relay_close_buf(struct rchan_buf *buf)
488 {
489         buf->finalized = 1;
490         del_timer_sync(&buf->timer);
491         buf->chan->cb->remove_buf_file(buf->dentry);
492         kref_put(&buf->kref, relay_remove_buf);
493 }
494
495 static void setup_callbacks(struct rchan *chan,
496                                    struct rchan_callbacks *cb)
497 {
498         if (!cb) {
499                 chan->cb = &default_channel_callbacks;
500                 return;
501         }
502
503         if (!cb->subbuf_start)
504                 cb->subbuf_start = subbuf_start_default_callback;
505         if (!cb->buf_mapped)
506                 cb->buf_mapped = buf_mapped_default_callback;
507         if (!cb->buf_unmapped)
508                 cb->buf_unmapped = buf_unmapped_default_callback;
509         if (!cb->create_buf_file)
510                 cb->create_buf_file = create_buf_file_default_callback;
511         if (!cb->remove_buf_file)
512                 cb->remove_buf_file = remove_buf_file_default_callback;
513         chan->cb = cb;
514 }
515
516 int relay_prepare_cpu(unsigned int cpu)
517 {
518         struct rchan *chan;
519         struct rchan_buf *buf;
520
521         mutex_lock(&relay_channels_mutex);
522         list_for_each_entry(chan, &relay_channels, list) {
523                 if ((buf = *per_cpu_ptr(chan->buf, cpu)))
524                         continue;
525                 buf = relay_open_buf(chan, cpu);
526                 if (!buf) {
527                         pr_err("relay: cpu %d buffer creation failed\n", cpu);
528                         mutex_unlock(&relay_channels_mutex);
529                         return -ENOMEM;
530                 }
531                 *per_cpu_ptr(chan->buf, cpu) = buf;
532         }
533         mutex_unlock(&relay_channels_mutex);
534         return 0;
535 }
536
537 /**
538  *      relay_open - create a new relay channel
539  *      @base_filename: base name of files to create, %NULL for buffering only
540  *      @parent: dentry of parent directory, %NULL for root directory or buffer
541  *      @subbuf_size: size of sub-buffers
542  *      @n_subbufs: number of sub-buffers
543  *      @cb: client callback functions
544  *      @private_data: user-defined data
545  *
546  *      Returns channel pointer if successful, %NULL otherwise.
547  *
548  *      Creates a channel buffer for each cpu using the sizes and
549  *      attributes specified.  The created channel buffer files
550  *      will be named base_filename0...base_filenameN-1.  File
551  *      permissions will be %S_IRUSR.
552  *
553  *      If opening a buffer (@parent = NULL) that you later wish to register
554  *      in a filesystem, call relay_late_setup_files() once the @parent dentry
555  *      is available.
556  */
557 struct rchan *relay_open(const char *base_filename,
558                          struct dentry *parent,
559                          size_t subbuf_size,
560                          size_t n_subbufs,
561                          struct rchan_callbacks *cb,
562                          void *private_data)
563 {
564         unsigned int i;
565         struct rchan *chan;
566         struct rchan_buf *buf;
567
568         if (!(subbuf_size && n_subbufs))
569                 return NULL;
570         if (subbuf_size > UINT_MAX / n_subbufs)
571                 return NULL;
572
573         chan = kzalloc(sizeof(struct rchan), GFP_KERNEL);
574         if (!chan)
575                 return NULL;
576
577         chan->buf = alloc_percpu(struct rchan_buf *);
578         chan->version = RELAYFS_CHANNEL_VERSION;
579         chan->n_subbufs = n_subbufs;
580         chan->subbuf_size = subbuf_size;
581         chan->alloc_size = PAGE_ALIGN(subbuf_size * n_subbufs);
582         chan->parent = parent;
583         chan->private_data = private_data;
584         if (base_filename) {
585                 chan->has_base_filename = 1;
586                 strlcpy(chan->base_filename, base_filename, NAME_MAX);
587         }
588         setup_callbacks(chan, cb);
589         kref_init(&chan->kref);
590
591         mutex_lock(&relay_channels_mutex);
592         for_each_online_cpu(i) {
593                 buf = relay_open_buf(chan, i);
594                 if (!buf)
595                         goto free_bufs;
596                 *per_cpu_ptr(chan->buf, i) = buf;
597         }
598         list_add(&chan->list, &relay_channels);
599         mutex_unlock(&relay_channels_mutex);
600
601         return chan;
602
603 free_bufs:
604         for_each_possible_cpu(i) {
605                 if ((buf = *per_cpu_ptr(chan->buf, i)))
606                         relay_close_buf(buf);
607         }
608
609         kref_put(&chan->kref, relay_destroy_channel);
610         mutex_unlock(&relay_channels_mutex);
611         kfree(chan);
612         return NULL;
613 }
614 EXPORT_SYMBOL_GPL(relay_open);
615
616 struct rchan_percpu_buf_dispatcher {
617         struct rchan_buf *buf;
618         struct dentry *dentry;
619 };
620
621 /* Called in atomic context. */
622 static void __relay_set_buf_dentry(void *info)
623 {
624         struct rchan_percpu_buf_dispatcher *p = info;
625
626         relay_set_buf_dentry(p->buf, p->dentry);
627 }
628
629 /**
630  *      relay_late_setup_files - triggers file creation
631  *      @chan: channel to operate on
632  *      @base_filename: base name of files to create
633  *      @parent: dentry of parent directory, %NULL for root directory
634  *
635  *      Returns 0 if successful, non-zero otherwise.
636  *
637  *      Use to setup files for a previously buffer-only channel created
638  *      by relay_open() with a NULL parent dentry.
639  *
640  *      For example, this is useful for perfomring early tracing in kernel,
641  *      before VFS is up and then exposing the early results once the dentry
642  *      is available.
643  */
644 int relay_late_setup_files(struct rchan *chan,
645                            const char *base_filename,
646                            struct dentry *parent)
647 {
648         int err = 0;
649         unsigned int i, curr_cpu;
650         unsigned long flags;
651         struct dentry *dentry;
652         struct rchan_buf *buf;
653         struct rchan_percpu_buf_dispatcher disp;
654
655         if (!chan || !base_filename)
656                 return -EINVAL;
657
658         strlcpy(chan->base_filename, base_filename, NAME_MAX);
659
660         mutex_lock(&relay_channels_mutex);
661         /* Is chan already set up? */
662         if (unlikely(chan->has_base_filename)) {
663                 mutex_unlock(&relay_channels_mutex);
664                 return -EEXIST;
665         }
666         chan->has_base_filename = 1;
667         chan->parent = parent;
668
669         if (chan->is_global) {
670                 err = -EINVAL;
671                 buf = *per_cpu_ptr(chan->buf, 0);
672                 if (!WARN_ON_ONCE(!buf)) {
673                         dentry = relay_create_buf_file(chan, buf, 0);
674                         if (dentry && !WARN_ON_ONCE(!chan->is_global)) {
675                                 relay_set_buf_dentry(buf, dentry);
676                                 err = 0;
677                         }
678                 }
679                 mutex_unlock(&relay_channels_mutex);
680                 return err;
681         }
682
683         curr_cpu = get_cpu();
684         /*
685          * The CPU hotplug notifier ran before us and created buffers with
686          * no files associated. So it's safe to call relay_setup_buf_file()
687          * on all currently online CPUs.
688          */
689         for_each_online_cpu(i) {
690                 buf = *per_cpu_ptr(chan->buf, i);
691                 if (unlikely(!buf)) {
692                         WARN_ONCE(1, KERN_ERR "CPU has no buffer!\n");
693                         err = -EINVAL;
694                         break;
695                 }
696
697                 dentry = relay_create_buf_file(chan, buf, i);
698                 if (unlikely(!dentry)) {
699                         err = -EINVAL;
700                         break;
701                 }
702
703                 if (curr_cpu == i) {
704                         local_irq_save(flags);
705                         relay_set_buf_dentry(buf, dentry);
706                         local_irq_restore(flags);
707                 } else {
708                         disp.buf = buf;
709                         disp.dentry = dentry;
710                         smp_mb();
711                         /* relay_channels_mutex must be held, so wait. */
712                         err = smp_call_function_single(i,
713                                                        __relay_set_buf_dentry,
714                                                        &disp, 1);
715                 }
716                 if (unlikely(err))
717                         break;
718         }
719         put_cpu();
720         mutex_unlock(&relay_channels_mutex);
721
722         return err;
723 }
724 EXPORT_SYMBOL_GPL(relay_late_setup_files);
725
726 /**
727  *      relay_switch_subbuf - switch to a new sub-buffer
728  *      @buf: channel buffer
729  *      @length: size of current event
730  *
731  *      Returns either the length passed in or 0 if full.
732  *
733  *      Performs sub-buffer-switch tasks such as invoking callbacks,
734  *      updating padding counts, waking up readers, etc.
735  */
736 size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
737 {
738         void *old, *new;
739         size_t old_subbuf, new_subbuf;
740
741         if (unlikely(length > buf->chan->subbuf_size))
742                 goto toobig;
743
744         if (buf->offset != buf->chan->subbuf_size + 1) {
745                 buf->prev_padding = buf->chan->subbuf_size - buf->offset;
746                 old_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
747                 buf->padding[old_subbuf] = buf->prev_padding;
748                 buf->subbufs_produced++;
749                 if (buf->dentry)
750                         d_inode(buf->dentry)->i_size +=
751                                 buf->chan->subbuf_size -
752                                 buf->padding[old_subbuf];
753                 else
754                         buf->early_bytes += buf->chan->subbuf_size -
755                                             buf->padding[old_subbuf];
756                 smp_mb();
757                 if (waitqueue_active(&buf->read_wait))
758                         /*
759                          * Calling wake_up_interruptible() from here
760                          * will deadlock if we happen to be logging
761                          * from the scheduler (trying to re-grab
762                          * rq->lock), so defer it.
763                          */
764                         mod_timer(&buf->timer, jiffies + 1);
765         }
766
767         old = buf->data;
768         new_subbuf = buf->subbufs_produced % buf->chan->n_subbufs;
769         new = buf->start + new_subbuf * buf->chan->subbuf_size;
770         buf->offset = 0;
771         if (!buf->chan->cb->subbuf_start(buf, new, old, buf->prev_padding)) {
772                 buf->offset = buf->chan->subbuf_size + 1;
773                 return 0;
774         }
775         buf->data = new;
776         buf->padding[new_subbuf] = 0;
777
778         if (unlikely(length + buf->offset > buf->chan->subbuf_size))
779                 goto toobig;
780
781         return length;
782
783 toobig:
784         buf->chan->last_toobig = length;
785         return 0;
786 }
787 EXPORT_SYMBOL_GPL(relay_switch_subbuf);
788
789 /**
790  *      relay_subbufs_consumed - update the buffer's sub-buffers-consumed count
791  *      @chan: the channel
792  *      @cpu: the cpu associated with the channel buffer to update
793  *      @subbufs_consumed: number of sub-buffers to add to current buf's count
794  *
795  *      Adds to the channel buffer's consumed sub-buffer count.
796  *      subbufs_consumed should be the number of sub-buffers newly consumed,
797  *      not the total consumed.
798  *
799  *      NOTE. Kernel clients don't need to call this function if the channel
800  *      mode is 'overwrite'.
801  */
802 void relay_subbufs_consumed(struct rchan *chan,
803                             unsigned int cpu,
804                             size_t subbufs_consumed)
805 {
806         struct rchan_buf *buf;
807
808         if (!chan)
809                 return;
810
811         buf = *per_cpu_ptr(chan->buf, cpu);
812         if (cpu >= NR_CPUS || !buf || subbufs_consumed > chan->n_subbufs)
813                 return;
814
815         if (subbufs_consumed > buf->subbufs_produced - buf->subbufs_consumed)
816                 buf->subbufs_consumed = buf->subbufs_produced;
817         else
818                 buf->subbufs_consumed += subbufs_consumed;
819 }
820 EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
821
822 /**
823  *      relay_close - close the channel
824  *      @chan: the channel
825  *
826  *      Closes all channel buffers and frees the channel.
827  */
828 void relay_close(struct rchan *chan)
829 {
830         struct rchan_buf *buf;
831         unsigned int i;
832
833         if (!chan)
834                 return;
835
836         mutex_lock(&relay_channels_mutex);
837         if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0)))
838                 relay_close_buf(buf);
839         else
840                 for_each_possible_cpu(i)
841                         if ((buf = *per_cpu_ptr(chan->buf, i)))
842                                 relay_close_buf(buf);
843
844         if (chan->last_toobig)
845                 printk(KERN_WARNING "relay: one or more items not logged "
846                        "[item size (%Zd) > sub-buffer size (%Zd)]\n",
847                        chan->last_toobig, chan->subbuf_size);
848
849         list_del(&chan->list);
850         kref_put(&chan->kref, relay_destroy_channel);
851         mutex_unlock(&relay_channels_mutex);
852 }
853 EXPORT_SYMBOL_GPL(relay_close);
854
855 /**
856  *      relay_flush - close the channel
857  *      @chan: the channel
858  *
859  *      Flushes all channel buffers, i.e. forces buffer switch.
860  */
861 void relay_flush(struct rchan *chan)
862 {
863         struct rchan_buf *buf;
864         unsigned int i;
865
866         if (!chan)
867                 return;
868
869         if (chan->is_global && (buf = *per_cpu_ptr(chan->buf, 0))) {
870                 relay_switch_subbuf(buf, 0);
871                 return;
872         }
873
874         mutex_lock(&relay_channels_mutex);
875         for_each_possible_cpu(i)
876                 if ((buf = *per_cpu_ptr(chan->buf, i)))
877                         relay_switch_subbuf(buf, 0);
878         mutex_unlock(&relay_channels_mutex);
879 }
880 EXPORT_SYMBOL_GPL(relay_flush);
881
882 /**
883  *      relay_file_open - open file op for relay files
884  *      @inode: the inode
885  *      @filp: the file
886  *
887  *      Increments the channel buffer refcount.
888  */
889 static int relay_file_open(struct inode *inode, struct file *filp)
890 {
891         struct rchan_buf *buf = inode->i_private;
892         kref_get(&buf->kref);
893         filp->private_data = buf;
894
895         return nonseekable_open(inode, filp);
896 }
897
898 /**
899  *      relay_file_mmap - mmap file op for relay files
900  *      @filp: the file
901  *      @vma: the vma describing what to map
902  *
903  *      Calls upon relay_mmap_buf() to map the file into user space.
904  */
905 static int relay_file_mmap(struct file *filp, struct vm_area_struct *vma)
906 {
907         struct rchan_buf *buf = filp->private_data;
908         return relay_mmap_buf(buf, vma);
909 }
910
911 /**
912  *      relay_file_poll - poll file op for relay files
913  *      @filp: the file
914  *      @wait: poll table
915  *
916  *      Poll implemention.
917  */
918 static unsigned int relay_file_poll(struct file *filp, poll_table *wait)
919 {
920         unsigned int mask = 0;
921         struct rchan_buf *buf = filp->private_data;
922
923         if (buf->finalized)
924                 return POLLERR;
925
926         if (filp->f_mode & FMODE_READ) {
927                 poll_wait(filp, &buf->read_wait, wait);
928                 if (!relay_buf_empty(buf))
929                         mask |= POLLIN | POLLRDNORM;
930         }
931
932         return mask;
933 }
934
935 /**
936  *      relay_file_release - release file op for relay files
937  *      @inode: the inode
938  *      @filp: the file
939  *
940  *      Decrements the channel refcount, as the filesystem is
941  *      no longer using it.
942  */
943 static int relay_file_release(struct inode *inode, struct file *filp)
944 {
945         struct rchan_buf *buf = filp->private_data;
946         kref_put(&buf->kref, relay_remove_buf);
947
948         return 0;
949 }
950
951 /*
952  *      relay_file_read_consume - update the consumed count for the buffer
953  */
954 static void relay_file_read_consume(struct rchan_buf *buf,
955                                     size_t read_pos,
956                                     size_t bytes_consumed)
957 {
958         size_t subbuf_size = buf->chan->subbuf_size;
959         size_t n_subbufs = buf->chan->n_subbufs;
960         size_t read_subbuf;
961
962         if (buf->subbufs_produced == buf->subbufs_consumed &&
963             buf->offset == buf->bytes_consumed)
964                 return;
965
966         if (buf->bytes_consumed + bytes_consumed > subbuf_size) {
967                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
968                 buf->bytes_consumed = 0;
969         }
970
971         buf->bytes_consumed += bytes_consumed;
972         if (!read_pos)
973                 read_subbuf = buf->subbufs_consumed % n_subbufs;
974         else
975                 read_subbuf = read_pos / buf->chan->subbuf_size;
976         if (buf->bytes_consumed + buf->padding[read_subbuf] == subbuf_size) {
977                 if ((read_subbuf == buf->subbufs_produced % n_subbufs) &&
978                     (buf->offset == subbuf_size))
979                         return;
980                 relay_subbufs_consumed(buf->chan, buf->cpu, 1);
981                 buf->bytes_consumed = 0;
982         }
983 }
984
985 /*
986  *      relay_file_read_avail - boolean, are there unconsumed bytes available?
987  */
988 static int relay_file_read_avail(struct rchan_buf *buf, size_t read_pos)
989 {
990         size_t subbuf_size = buf->chan->subbuf_size;
991         size_t n_subbufs = buf->chan->n_subbufs;
992         size_t produced = buf->subbufs_produced;
993         size_t consumed = buf->subbufs_consumed;
994
995         relay_file_read_consume(buf, read_pos, 0);
996
997         consumed = buf->subbufs_consumed;
998
999         if (unlikely(buf->offset > subbuf_size)) {
1000                 if (produced == consumed)
1001                         return 0;
1002                 return 1;
1003         }
1004
1005         if (unlikely(produced - consumed >= n_subbufs)) {
1006                 consumed = produced - n_subbufs + 1;
1007                 buf->subbufs_consumed = consumed;
1008                 buf->bytes_consumed = 0;
1009         }
1010
1011         produced = (produced % n_subbufs) * subbuf_size + buf->offset;
1012         consumed = (consumed % n_subbufs) * subbuf_size + buf->bytes_consumed;
1013
1014         if (consumed > produced)
1015                 produced += n_subbufs * subbuf_size;
1016
1017         if (consumed == produced) {
1018                 if (buf->offset == subbuf_size &&
1019                     buf->subbufs_produced > buf->subbufs_consumed)
1020                         return 1;
1021                 return 0;
1022         }
1023
1024         return 1;
1025 }
1026
1027 /**
1028  *      relay_file_read_subbuf_avail - return bytes available in sub-buffer
1029  *      @read_pos: file read position
1030  *      @buf: relay channel buffer
1031  */
1032 static size_t relay_file_read_subbuf_avail(size_t read_pos,
1033                                            struct rchan_buf *buf)
1034 {
1035         size_t padding, avail = 0;
1036         size_t read_subbuf, read_offset, write_subbuf, write_offset;
1037         size_t subbuf_size = buf->chan->subbuf_size;
1038
1039         write_subbuf = (buf->data - buf->start) / subbuf_size;
1040         write_offset = buf->offset > subbuf_size ? subbuf_size : buf->offset;
1041         read_subbuf = read_pos / subbuf_size;
1042         read_offset = read_pos % subbuf_size;
1043         padding = buf->padding[read_subbuf];
1044
1045         if (read_subbuf == write_subbuf) {
1046                 if (read_offset + padding < write_offset)
1047                         avail = write_offset - (read_offset + padding);
1048         } else
1049                 avail = (subbuf_size - padding) - read_offset;
1050
1051         return avail;
1052 }
1053
1054 /**
1055  *      relay_file_read_start_pos - find the first available byte to read
1056  *      @read_pos: file read position
1057  *      @buf: relay channel buffer
1058  *
1059  *      If the @read_pos is in the middle of padding, return the
1060  *      position of the first actually available byte, otherwise
1061  *      return the original value.
1062  */
1063 static size_t relay_file_read_start_pos(size_t read_pos,
1064                                         struct rchan_buf *buf)
1065 {
1066         size_t read_subbuf, padding, padding_start, padding_end;
1067         size_t subbuf_size = buf->chan->subbuf_size;
1068         size_t n_subbufs = buf->chan->n_subbufs;
1069         size_t consumed = buf->subbufs_consumed % n_subbufs;
1070
1071         if (!read_pos)
1072                 read_pos = consumed * subbuf_size + buf->bytes_consumed;
1073         read_subbuf = read_pos / subbuf_size;
1074         padding = buf->padding[read_subbuf];
1075         padding_start = (read_subbuf + 1) * subbuf_size - padding;
1076         padding_end = (read_subbuf + 1) * subbuf_size;
1077         if (read_pos >= padding_start && read_pos < padding_end) {
1078                 read_subbuf = (read_subbuf + 1) % n_subbufs;
1079                 read_pos = read_subbuf * subbuf_size;
1080         }
1081
1082         return read_pos;
1083 }
1084
1085 /**
1086  *      relay_file_read_end_pos - return the new read position
1087  *      @read_pos: file read position
1088  *      @buf: relay channel buffer
1089  *      @count: number of bytes to be read
1090  */
1091 static size_t relay_file_read_end_pos(struct rchan_buf *buf,
1092                                       size_t read_pos,
1093                                       size_t count)
1094 {
1095         size_t read_subbuf, padding, end_pos;
1096         size_t subbuf_size = buf->chan->subbuf_size;
1097         size_t n_subbufs = buf->chan->n_subbufs;
1098
1099         read_subbuf = read_pos / subbuf_size;
1100         padding = buf->padding[read_subbuf];
1101         if (read_pos % subbuf_size + count + padding == subbuf_size)
1102                 end_pos = (read_subbuf + 1) * subbuf_size;
1103         else
1104                 end_pos = read_pos + count;
1105         if (end_pos >= subbuf_size * n_subbufs)
1106                 end_pos = 0;
1107
1108         return end_pos;
1109 }
1110
1111 /*
1112  *      subbuf_read_actor - read up to one subbuf's worth of data
1113  */
1114 static int subbuf_read_actor(size_t read_start,
1115                              struct rchan_buf *buf,
1116                              size_t avail,
1117                              read_descriptor_t *desc)
1118 {
1119         void *from;
1120         int ret = 0;
1121
1122         from = buf->start + read_start;
1123         ret = avail;
1124         if (copy_to_user(desc->arg.buf, from, avail)) {
1125                 desc->error = -EFAULT;
1126                 ret = 0;
1127         }
1128         desc->arg.data += ret;
1129         desc->written += ret;
1130         desc->count -= ret;
1131
1132         return ret;
1133 }
1134
1135 typedef int (*subbuf_actor_t) (size_t read_start,
1136                                struct rchan_buf *buf,
1137                                size_t avail,
1138                                read_descriptor_t *desc);
1139
1140 /*
1141  *      relay_file_read_subbufs - read count bytes, bridging subbuf boundaries
1142  */
1143 static ssize_t relay_file_read_subbufs(struct file *filp, loff_t *ppos,
1144                                         subbuf_actor_t subbuf_actor,
1145                                         read_descriptor_t *desc)
1146 {
1147         struct rchan_buf *buf = filp->private_data;
1148         size_t read_start, avail;
1149         int ret;
1150
1151         if (!desc->count)
1152                 return 0;
1153
1154         inode_lock(file_inode(filp));
1155         do {
1156                 if (!relay_file_read_avail(buf, *ppos))
1157                         break;
1158
1159                 read_start = relay_file_read_start_pos(*ppos, buf);
1160                 avail = relay_file_read_subbuf_avail(read_start, buf);
1161                 if (!avail)
1162                         break;
1163
1164                 avail = min(desc->count, avail);
1165                 ret = subbuf_actor(read_start, buf, avail, desc);
1166                 if (desc->error < 0)
1167                         break;
1168
1169                 if (ret) {
1170                         relay_file_read_consume(buf, read_start, ret);
1171                         *ppos = relay_file_read_end_pos(buf, read_start, ret);
1172                 }
1173         } while (desc->count && ret);
1174         inode_unlock(file_inode(filp));
1175
1176         return desc->written;
1177 }
1178
1179 static ssize_t relay_file_read(struct file *filp,
1180                                char __user *buffer,
1181                                size_t count,
1182                                loff_t *ppos)
1183 {
1184         read_descriptor_t desc;
1185         desc.written = 0;
1186         desc.count = count;
1187         desc.arg.buf = buffer;
1188         desc.error = 0;
1189         return relay_file_read_subbufs(filp, ppos, subbuf_read_actor, &desc);
1190 }
1191
1192 static void relay_consume_bytes(struct rchan_buf *rbuf, int bytes_consumed)
1193 {
1194         rbuf->bytes_consumed += bytes_consumed;
1195
1196         if (rbuf->bytes_consumed >= rbuf->chan->subbuf_size) {
1197                 relay_subbufs_consumed(rbuf->chan, rbuf->cpu, 1);
1198                 rbuf->bytes_consumed %= rbuf->chan->subbuf_size;
1199         }
1200 }
1201
1202 static void relay_pipe_buf_release(struct pipe_inode_info *pipe,
1203                                    struct pipe_buffer *buf)
1204 {
1205         struct rchan_buf *rbuf;
1206
1207         rbuf = (struct rchan_buf *)page_private(buf->page);
1208         relay_consume_bytes(rbuf, buf->private);
1209 }
1210
1211 static const struct pipe_buf_operations relay_pipe_buf_ops = {
1212         .can_merge = 0,
1213         .confirm = generic_pipe_buf_confirm,
1214         .release = relay_pipe_buf_release,
1215         .steal = generic_pipe_buf_steal,
1216         .get = generic_pipe_buf_get,
1217 };
1218
1219 static void relay_page_release(struct splice_pipe_desc *spd, unsigned int i)
1220 {
1221 }
1222
1223 /*
1224  *      subbuf_splice_actor - splice up to one subbuf's worth of data
1225  */
1226 static ssize_t subbuf_splice_actor(struct file *in,
1227                                loff_t *ppos,
1228                                struct pipe_inode_info *pipe,
1229                                size_t len,
1230                                unsigned int flags,
1231                                int *nonpad_ret)
1232 {
1233         unsigned int pidx, poff, total_len, subbuf_pages, nr_pages;
1234         struct rchan_buf *rbuf = in->private_data;
1235         unsigned int subbuf_size = rbuf->chan->subbuf_size;
1236         uint64_t pos = (uint64_t) *ppos;
1237         uint32_t alloc_size = (uint32_t) rbuf->chan->alloc_size;
1238         size_t read_start = (size_t) do_div(pos, alloc_size);
1239         size_t read_subbuf = read_start / subbuf_size;
1240         size_t padding = rbuf->padding[read_subbuf];
1241         size_t nonpad_end = read_subbuf * subbuf_size + subbuf_size - padding;
1242         struct page *pages[PIPE_DEF_BUFFERS];
1243         struct partial_page partial[PIPE_DEF_BUFFERS];
1244         struct splice_pipe_desc spd = {
1245                 .pages = pages,
1246                 .nr_pages = 0,
1247                 .nr_pages_max = PIPE_DEF_BUFFERS,
1248                 .partial = partial,
1249                 .flags = flags,
1250                 .ops = &relay_pipe_buf_ops,
1251                 .spd_release = relay_page_release,
1252         };
1253         ssize_t ret;
1254
1255         if (rbuf->subbufs_produced == rbuf->subbufs_consumed)
1256                 return 0;
1257         if (splice_grow_spd(pipe, &spd))
1258                 return -ENOMEM;
1259
1260         /*
1261          * Adjust read len, if longer than what is available
1262          */
1263         if (len > (subbuf_size - read_start % subbuf_size))
1264                 len = subbuf_size - read_start % subbuf_size;
1265
1266         subbuf_pages = rbuf->chan->alloc_size >> PAGE_SHIFT;
1267         pidx = (read_start / PAGE_SIZE) % subbuf_pages;
1268         poff = read_start & ~PAGE_MASK;
1269         nr_pages = min_t(unsigned int, subbuf_pages, spd.nr_pages_max);
1270
1271         for (total_len = 0; spd.nr_pages < nr_pages; spd.nr_pages++) {
1272                 unsigned int this_len, this_end, private;
1273                 unsigned int cur_pos = read_start + total_len;
1274
1275                 if (!len)
1276                         break;
1277
1278                 this_len = min_t(unsigned long, len, PAGE_SIZE - poff);
1279                 private = this_len;
1280
1281                 spd.pages[spd.nr_pages] = rbuf->page_array[pidx];
1282                 spd.partial[spd.nr_pages].offset = poff;
1283
1284                 this_end = cur_pos + this_len;
1285                 if (this_end >= nonpad_end) {
1286                         this_len = nonpad_end - cur_pos;
1287                         private = this_len + padding;
1288                 }
1289                 spd.partial[spd.nr_pages].len = this_len;
1290                 spd.partial[spd.nr_pages].private = private;
1291
1292                 len -= this_len;
1293                 total_len += this_len;
1294                 poff = 0;
1295                 pidx = (pidx + 1) % subbuf_pages;
1296
1297                 if (this_end >= nonpad_end) {
1298                         spd.nr_pages++;
1299                         break;
1300                 }
1301         }
1302
1303         ret = 0;
1304         if (!spd.nr_pages)
1305                 goto out;
1306
1307         ret = *nonpad_ret = splice_to_pipe(pipe, &spd);
1308         if (ret < 0 || ret < total_len)
1309                 goto out;
1310
1311         if (read_start + ret == nonpad_end)
1312                 ret += padding;
1313
1314 out:
1315         splice_shrink_spd(&spd);
1316         return ret;
1317 }
1318
1319 static ssize_t relay_file_splice_read(struct file *in,
1320                                       loff_t *ppos,
1321                                       struct pipe_inode_info *pipe,
1322                                       size_t len,
1323                                       unsigned int flags)
1324 {
1325         ssize_t spliced;
1326         int ret;
1327         int nonpad_ret = 0;
1328
1329         ret = 0;
1330         spliced = 0;
1331
1332         while (len && !spliced) {
1333                 ret = subbuf_splice_actor(in, ppos, pipe, len, flags, &nonpad_ret);
1334                 if (ret < 0)
1335                         break;
1336                 else if (!ret) {
1337                         if (flags & SPLICE_F_NONBLOCK)
1338                                 ret = -EAGAIN;
1339                         break;
1340                 }
1341
1342                 *ppos += ret;
1343                 if (ret > len)
1344                         len = 0;
1345                 else
1346                         len -= ret;
1347                 spliced += nonpad_ret;
1348                 nonpad_ret = 0;
1349         }
1350
1351         if (spliced)
1352                 return spliced;
1353
1354         return ret;
1355 }
1356
1357 const struct file_operations relay_file_operations = {
1358         .open           = relay_file_open,
1359         .poll           = relay_file_poll,
1360         .mmap           = relay_file_mmap,
1361         .read           = relay_file_read,
1362         .llseek         = no_llseek,
1363         .release        = relay_file_release,
1364         .splice_read    = relay_file_splice_read,
1365 };
1366 EXPORT_SYMBOL_GPL(relay_file_operations);